[0-9a-z_.]*)( .*)?$");
/**
* Returns the set of locations in which the tag may be used.
@@ -157,19 +173,50 @@ public class JSpec implements Taglet {
.trim();
Matcher m = TAG_PATTERN.matcher(tagText);
if (m.find()) {
+ // preview-feature-4.6 is preview-feature-, 4, .6
+ String preview = m.group("preview"); // null if no preview feature
String chapter = m.group("chapter");
String section = m.group("section");
String rootParent = currentPath().replaceAll("[^/]+", "..");
- String url = String.format("%1$s/specs/%2$s/%2$s-%3$s.html#%2$s-%3$s%4$s",
- rootParent, idPrefix, chapter, section);
+ String url = preview == null ?
+ String.format("%1$s/specs/%2$s/%2$s-%3$s.html#%2$s-%3$s%4$s",
+ rootParent, idPrefix, chapter, section) :
+ String.format("%1$s/specs/%5$s%2$s.html#%2$s-%3$s%4$s",
+ rootParent, idPrefix, chapter, section, preview);
+
+ var literal = expand(contents).trim();
+ var prefix = (preview == null ? "" : preview) + chapter + section;
+ if (literal.startsWith(prefix)) {
+ var hasFullTitle = literal.length() > prefix.length();
+ if (hasFullTitle) {
+ // Drop the preview identifier
+ literal = chapter + section + literal.substring(prefix.length());
+ } else {
+ // No section sign if the tag refers to a chapter, like {@jvms 4}
+ String sectionSign = section.isEmpty() ? "" : "§";
+ // Change whole text to "§chapter.x" in inline tags.
+ literal = sectionSign + chapter + section;
+ }
+ }
sb.append("")
- .append(expand(contents))
+ .append(literal)
.append("");
+ if (preview != null) {
+ // Add PREVIEW superscript that links to JLS/JVMS 1.5.1
+ // "Restrictions on the Use of Preview Features"
+ // Similar to how APIs link to the Preview info box warning
+ var sectionLink = String.format("%1$s/specs/%2$s/%2$s-%3$s.html#%2$s-%3$s%4$s",
+ rootParent, idPrefix, "1", ".5.1");
+ sb.append("PREVIEW");
+ }
+
if (tag.getKind() == DocTree.Kind.UNKNOWN_BLOCK_TAG) {
sb.append("
");
}
diff --git a/make/modules/java.desktop/lib/AwtLibraries.gmk b/make/modules/java.desktop/lib/AwtLibraries.gmk
index 463e09e12dc..8b6b50b9e62 100644
--- a/make/modules/java.desktop/lib/AwtLibraries.gmk
+++ b/make/modules/java.desktop/lib/AwtLibraries.gmk
@@ -88,6 +88,10 @@ LIBAWT_EXTRA_HEADER_DIRS := \
LIBAWT_CFLAGS := -D__MEDIALIB_OLD_NAMES -D__USE_J2D_NAMES -DMLIB_NO_LIBSUNMATH
+ifeq ($(ENABLE_HEADLESS_ONLY), true)
+ LIBAWT_CFLAGS += -DHEADLESS
+endif
+
ifeq ($(call isTargetOs, windows), true)
LIBAWT_CFLAGS += -EHsc -DUNICODE -D_UNICODE -DMLIB_OS64BIT
LIBAWT_RCFLAGS ?= -I$(TOPDIR)/src/java.base/windows/native/launcher/icons
@@ -167,11 +171,18 @@ ifeq ($(call isTargetOs, windows macosx), false)
$(TOPDIR)/src/$(MODULE)/$(OPENJDK_TARGET_OS_TYPE)/native/common/awt \
#
+ LIBAWT_HEADLESS_EXCLUDE_FILES := \
+ GLXGraphicsConfig.c \
+ GLXSurfaceData.c \
+ X11PMBlitLoops.c \
+ X11Renderer.c \
+ X11SurfaceData.c \
+ #
+
LIBAWT_HEADLESS_EXTRA_HEADER_DIRS := \
$(LIBAWT_DEFAULT_HEADER_DIRS) \
common/awt/debug \
common/font \
- common/java2d/opengl \
java.base:libjvm \
#
@@ -191,7 +202,8 @@ ifeq ($(call isTargetOs, windows macosx), false)
$(eval $(call SetupJdkLibrary, BUILD_LIBAWT_HEADLESS, \
NAME := awt_headless, \
EXTRA_SRC := $(LIBAWT_HEADLESS_EXTRA_SRC), \
- EXCLUDES := medialib, \
+ EXCLUDES := medialib opengl, \
+ EXCLUDE_FILES := $(LIBAWT_HEADLESS_EXCLUDE_FILES), \
ONLY_EXPORTED := $(LIBAWT_HEADLESS_ONLY_EXPORTED), \
OPTIMIZATION := LOW, \
CFLAGS := -DHEADLESS=true $(CUPS_CFLAGS) $(FONTCONFIG_CFLAGS) \
diff --git a/make/modules/jdk.jpackage/Lib.gmk b/make/modules/jdk.jpackage/Lib.gmk
index 704436bbde6..86b11bdafee 100644
--- a/make/modules/jdk.jpackage/Lib.gmk
+++ b/make/modules/jdk.jpackage/Lib.gmk
@@ -1,5 +1,5 @@
#
-# Copyright (c) 2018, 2025, Oracle and/or its affiliates. All rights reserved.
+# Copyright (c) 2018, 2026, Oracle and/or its affiliates. All rights reserved.
# DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
#
# This code is free software; you can redistribute it and/or modify it
@@ -54,7 +54,7 @@ $(eval $(call SetupJdkExecutable, BUILD_JPACKAGEAPPLAUNCHER, \
SRC := applauncher, \
EXTRA_SRC := common, \
INCLUDE_FILES := $(JPACKAGEAPPLAUNCHER_INCLUDE_FILES), \
- OPTIMIZATION := LOW, \
+ OPTIMIZATION := SIZE, \
DISABLED_WARNINGS_clang_JvmLauncherLib.c := format-nonliteral, \
DISABLED_WARNINGS_clang_LinuxPackage.c := format-nonliteral, \
DISABLED_WARNINGS_clang_Log.cpp := unused-const-variable, \
@@ -91,7 +91,7 @@ ifeq ($(call isTargetOs, linux), true)
common, \
EXCLUDE_FILES := LinuxLauncher.c LinuxPackage.c, \
LINK_TYPE := C++, \
- OPTIMIZATION := LOW, \
+ OPTIMIZATION := SIZE, \
DISABLED_WARNINGS_gcc_Log.cpp := unused-const-variable, \
DISABLED_WARNINGS_clang_JvmLauncherLib.c := format-nonliteral, \
DISABLED_WARNINGS_clang_tstrings.cpp := format-nonliteral, \
diff --git a/src/demo/share/jfc/TableExample/OldJTable.java b/src/demo/share/jfc/TableExample/OldJTable.java
deleted file mode 100644
index 8c77978fe8a..00000000000
--- a/src/demo/share/jfc/TableExample/OldJTable.java
+++ /dev/null
@@ -1,264 +0,0 @@
-/*
- * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *
- * - Redistributions of source code must retain the above copyright
- * notice, this list of conditions and the following disclaimer.
- *
- * - Redistributions in binary form must reproduce the above copyright
- * notice, this list of conditions and the following disclaimer in the
- * documentation and/or other materials provided with the distribution.
- *
- * - Neither the name of Oracle nor the names of its
- * contributors may be used to endorse or promote products derived
- * from this software without specific prior written permission.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS
- * IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
- * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
- * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
- * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
- * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
- * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
- * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
- * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
- * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
- * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- */
-
-/*
- * This source code is provided to illustrate the usage of a given feature
- * or technique and has been deliberately simplified. Additional steps
- * required for a production-quality application, such as security checks,
- * input validation and proper error handling, might not be present in
- * this sample code.
- */
-
-
-
-import java.util.EventObject;
-import java.util.List;
-import javax.swing.JTable;
-import javax.swing.table.DefaultTableModel;
-import javax.swing.table.TableCellEditor;
-import javax.swing.table.TableCellRenderer;
-import javax.swing.table.TableColumn;
-
-
-/**
- * The OldJTable is an unsupported class containing some methods that were
- * deleted from the JTable between releases 0.6 and 0.7
- */
-@SuppressWarnings("serial")
-public class OldJTable extends JTable
-{
- /*
- * A new convenience method returning the index of the column in the
- * co-ordinate space of the view.
- */
- public int getColumnIndex(Object identifier) {
- return getColumnModel().getColumnIndex(identifier);
- }
-
-//
-// Methods deleted from the JTable because they only work with the
-// DefaultTableModel.
-//
-
- public TableColumn addColumn(Object columnIdentifier, int width) {
- return addColumn(columnIdentifier, width, null, null, null);
- }
-
- public TableColumn addColumn(Object columnIdentifier, List> columnData) {
- return addColumn(columnIdentifier, -1, null, null, columnData);
- }
-
- // Override the new JTable implementation - it will not add a column to the
- // DefaultTableModel.
- public TableColumn addColumn(Object columnIdentifier, int width,
- TableCellRenderer renderer,
- TableCellEditor editor) {
- return addColumn(columnIdentifier, width, renderer, editor, null);
- }
-
- public TableColumn addColumn(Object columnIdentifier, int width,
- TableCellRenderer renderer,
- TableCellEditor editor, List> columnData) {
- checkDefaultTableModel();
-
- // Set up the model side first
- DefaultTableModel m = (DefaultTableModel)getModel();
- m.addColumn(columnIdentifier, columnData.toArray());
-
- // The column will have been added to the end, so the index of the
- // column in the model is the last element.
- TableColumn newColumn = new TableColumn(
- m.getColumnCount()-1, width, renderer, editor);
- super.addColumn(newColumn);
- return newColumn;
- }
-
- // Not possilble to make this work the same way ... change it so that
- // it does not delete columns from the model.
- public void removeColumn(Object columnIdentifier) {
- super.removeColumn(getColumn(columnIdentifier));
- }
-
- public void addRow(Object[] rowData) {
- checkDefaultTableModel();
- ((DefaultTableModel)getModel()).addRow(rowData);
- }
-
- public void addRow(List> rowData) {
- checkDefaultTableModel();
- ((DefaultTableModel)getModel()).addRow(rowData.toArray());
- }
-
- public void removeRow(int rowIndex) {
- checkDefaultTableModel();
- ((DefaultTableModel)getModel()).removeRow(rowIndex);
- }
-
- public void moveRow(int startIndex, int endIndex, int toIndex) {
- checkDefaultTableModel();
- ((DefaultTableModel)getModel()).moveRow(startIndex, endIndex, toIndex);
- }
-
- public void insertRow(int rowIndex, Object[] rowData) {
- checkDefaultTableModel();
- ((DefaultTableModel)getModel()).insertRow(rowIndex, rowData);
- }
-
- public void insertRow(int rowIndex, List> rowData) {
- checkDefaultTableModel();
- ((DefaultTableModel)getModel()).insertRow(rowIndex, rowData.toArray());
- }
-
- public void setNumRows(int newSize) {
- checkDefaultTableModel();
- ((DefaultTableModel)getModel()).setNumRows(newSize);
- }
-
- public void setDataVector(Object[][] newData, List> columnIds) {
- checkDefaultTableModel();
- ((DefaultTableModel)getModel()).setDataVector(
- newData, columnIds.toArray());
- }
-
- public void setDataVector(Object[][] newData, Object[] columnIds) {
- checkDefaultTableModel();
- ((DefaultTableModel)getModel()).setDataVector(newData, columnIds);
- }
-
- protected void checkDefaultTableModel() {
- if(!(dataModel instanceof DefaultTableModel))
- throw new InternalError("In order to use this method, the data model must be an instance of DefaultTableModel.");
- }
-
-//
-// Methods removed from JTable in the move from identifiers to ints.
-//
-
- public Object getValueAt(Object columnIdentifier, int rowIndex) {
- return super.getValueAt(rowIndex, getColumnIndex(columnIdentifier));
- }
-
- public boolean isCellEditable(Object columnIdentifier, int rowIndex) {
- return super.isCellEditable(rowIndex, getColumnIndex(columnIdentifier));
- }
-
- public void setValueAt(Object aValue, Object columnIdentifier, int rowIndex) {
- super.setValueAt(aValue, rowIndex, getColumnIndex(columnIdentifier));
- }
-
- public boolean editColumnRow(Object identifier, int row) {
- return super.editCellAt(row, getColumnIndex(identifier));
- }
-
- public void moveColumn(Object columnIdentifier, Object targetColumnIdentifier) {
- moveColumn(getColumnIndex(columnIdentifier),
- getColumnIndex(targetColumnIdentifier));
- }
-
- public boolean isColumnSelected(Object identifier) {
- return isColumnSelected(getColumnIndex(identifier));
- }
-
- public TableColumn addColumn(int modelColumn, int width) {
- return addColumn(modelColumn, width, null, null);
- }
-
- public TableColumn addColumn(int modelColumn) {
- return addColumn(modelColumn, 75, null, null);
- }
-
- /**
- * Creates a new column with modelColumn, width,
- * renderer, and editor and adds it to the end of
- * the JTable's array of columns. This method also retrieves the
- * name of the column using the model's getColumnName(modelColumn)
- * method, and sets the both the header value and the identifier
- * for this TableColumn accordingly.
- *
- * The modelColumn is the index of the column in the model which
- * will supply the data for this column in the table. This, like the
- * columnIdentifier in previous releases, does not change as the
- * columns are moved in the view.
- *
- * For the rest of the JTable API, and all of its associated classes,
- * columns are referred to in the co-ordinate system of the view, the
- * index of the column in the model is kept inside the TableColumn
- * and is used only to retrieve the information from the appropraite
- * column in the model.
- *
- *
- * @param modelColumn The index of the column in the model
- * @param width The new column's width. Or -1 to use
- * the default width
- * @param renderer The renderer used with the new column.
- * Or null to use the default renderer.
- * @param editor The editor used with the new column.
- * Or null to use the default editor.
- */
- public TableColumn addColumn(int modelColumn, int width,
- TableCellRenderer renderer,
- TableCellEditor editor) {
- TableColumn newColumn = new TableColumn(
- modelColumn, width, renderer, editor);
- addColumn(newColumn);
- return newColumn;
- }
-
-//
-// Methods that had their arguments switched.
-//
-
-// These won't work with the new table package.
-
-/*
- public Object getValueAt(int columnIndex, int rowIndex) {
- return super.getValueAt(rowIndex, columnIndex);
- }
-
- public boolean isCellEditable(int columnIndex, int rowIndex) {
- return super.isCellEditable(rowIndex, columnIndex);
- }
-
- public void setValueAt(Object aValue, int columnIndex, int rowIndex) {
- super.setValueAt(aValue, rowIndex, columnIndex);
- }
-*/
-
- public boolean editColumnRow(int columnIndex, int rowIndex) {
- return super.editCellAt(rowIndex, columnIndex);
- }
-
- public boolean editColumnRow(int columnIndex, int rowIndex, EventObject e){
- return super.editCellAt(rowIndex, columnIndex, e);
- }
-
-
-} // End Of Class OldJTable
diff --git a/src/hotspot/cpu/aarch64/aarch64.ad b/src/hotspot/cpu/aarch64/aarch64.ad
index b9252cc56ff..a9ca91d9309 100644
--- a/src/hotspot/cpu/aarch64/aarch64.ad
+++ b/src/hotspot/cpu/aarch64/aarch64.ad
@@ -1229,7 +1229,7 @@ public:
// predicate controlling addressing modes
bool size_fits_all_mem_uses(AddPNode* addp, int shift);
- // Convert BootTest condition to Assembler condition.
+ // Convert BoolTest condition to Assembler condition.
// Replicate the logic of cmpOpOper::ccode() and cmpOpUOper::ccode().
Assembler::Condition to_assembler_cond(BoolTest::mask cond);
%}
@@ -2579,7 +2579,7 @@ bool size_fits_all_mem_uses(AddPNode* addp, int shift) {
return true;
}
-// Convert BootTest condition to Assembler condition.
+// Convert BoolTest condition to Assembler condition.
// Replicate the logic of cmpOpOper::ccode() and cmpOpUOper::ccode().
Assembler::Condition to_assembler_cond(BoolTest::mask cond) {
Assembler::Condition result;
diff --git a/src/hotspot/cpu/aarch64/aarch64_vector.ad b/src/hotspot/cpu/aarch64/aarch64_vector.ad
index 78ef121bd29..19f03d97a72 100644
--- a/src/hotspot/cpu/aarch64/aarch64_vector.ad
+++ b/src/hotspot/cpu/aarch64/aarch64_vector.ad
@@ -1,5 +1,5 @@
//
-// Copyright (c) 2020, 2025, Oracle and/or its affiliates. All rights reserved.
+// Copyright (c) 2020, 2026, Oracle and/or its affiliates. All rights reserved.
// Copyright (c) 2020, 2025, Arm Limited. All rights reserved.
// DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
//
@@ -201,6 +201,8 @@ source %{
case Op_XorReductionV:
case Op_MinReductionV:
case Op_MaxReductionV:
+ case Op_UMinReductionV:
+ case Op_UMaxReductionV:
// Reductions with less than 8 bytes vector length are
// not supported.
if (length_in_bytes < 8) {
@@ -383,6 +385,8 @@ source %{
return !VM_Version::use_neon_for_vector(length_in_bytes);
case Op_MinReductionV:
case Op_MaxReductionV:
+ case Op_UMinReductionV:
+ case Op_UMaxReductionV:
// For BYTE/SHORT/INT/FLOAT/DOUBLE types, we prefer using NEON
// instructions rather than SVE predicated instructions for
// better performance.
@@ -4218,6 +4222,224 @@ instruct reduce_minD_masked(vRegD dst, vRegD dsrc, vReg vsrc, pRegGov pg) %{
ins_pipe(pipe_slow);
%}
+// -------------------- Vector reduction unsigned min/max ----------------------
+
+// reduction uminI
+
+instruct reduce_uminI_neon(iRegINoSp dst, iRegIorL2I isrc, vReg vsrc,
+ vReg tmp, rFlagsReg cr) %{
+ predicate(VM_Version::use_neon_for_vector(Matcher::vector_length_in_bytes(n->in(2))) &&
+ (Matcher::vector_element_basic_type(n->in(2)) == T_BYTE ||
+ Matcher::vector_element_basic_type(n->in(2)) == T_SHORT ||
+ Matcher::vector_element_basic_type(n->in(2)) == T_INT));
+ match(Set dst (UMinReductionV isrc vsrc));
+ effect(TEMP_DEF dst, TEMP tmp, KILL cr);
+ format %{ "reduce_uminI_neon $dst, $isrc, $vsrc\t# KILL $tmp, cr" %}
+ ins_encode %{
+ BasicType bt = Matcher::vector_element_basic_type(this, $vsrc);
+ uint length_in_bytes = Matcher::vector_length_in_bytes(this, $vsrc);
+ __ neon_reduce_minmax_integral(this->ideal_Opcode(), $dst$$Register, bt,
+ $isrc$$Register, $vsrc$$FloatRegister,
+ length_in_bytes, $tmp$$FloatRegister);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
+instruct reduce_uminI_sve(iRegINoSp dst, iRegIorL2I isrc, vReg vsrc,
+ vRegD tmp, rFlagsReg cr) %{
+ predicate(!VM_Version::use_neon_for_vector(Matcher::vector_length_in_bytes(n->in(2))) &&
+ (Matcher::vector_element_basic_type(n->in(2)) == T_BYTE ||
+ Matcher::vector_element_basic_type(n->in(2)) == T_SHORT ||
+ Matcher::vector_element_basic_type(n->in(2)) == T_INT));
+ match(Set dst (UMinReductionV isrc vsrc));
+ effect(TEMP_DEF dst, TEMP tmp, KILL cr);
+ format %{ "reduce_uminI_sve $dst, $isrc, $vsrc\t# KILL $tmp, cr" %}
+ ins_encode %{
+ assert(UseSVE > 0, "must be sve");
+ BasicType bt = Matcher::vector_element_basic_type(this, $vsrc);
+ uint length_in_bytes = Matcher::vector_length_in_bytes(this, $vsrc);
+ assert(length_in_bytes == MaxVectorSize, "invalid vector length");
+ __ sve_reduce_integral(this->ideal_Opcode(), $dst$$Register, bt,
+ $isrc$$Register, $vsrc$$FloatRegister,
+ ptrue, $tmp$$FloatRegister);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
+// reduction uminL
+
+instruct reduce_uminL_neon(iRegLNoSp dst, iRegL isrc, vReg vsrc, rFlagsReg cr) %{
+ predicate(UseSVE == 0 && Matcher::vector_element_basic_type(n->in(2)) == T_LONG);
+ match(Set dst (UMinReductionV isrc vsrc));
+ effect(TEMP_DEF dst, KILL cr);
+ format %{ "reduce_uminL_neon $dst, $isrc, $vsrc\t# 2L. KILL cr" %}
+ ins_encode %{
+ __ neon_reduce_minmax_integral(this->ideal_Opcode(), $dst$$Register, T_LONG,
+ $isrc$$Register, $vsrc$$FloatRegister,
+ /* vector_length_in_bytes */ 16, fnoreg);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
+instruct reduce_uminL_sve(iRegLNoSp dst, iRegL isrc, vReg vsrc,
+ vRegD tmp, rFlagsReg cr) %{
+ predicate(UseSVE > 0 && Matcher::vector_element_basic_type(n->in(2)) == T_LONG);
+ match(Set dst (UMinReductionV isrc vsrc));
+ effect(TEMP_DEF dst, TEMP tmp, KILL cr);
+ format %{ "reduce_uminL_sve $dst, $isrc, $vsrc\t# KILL $tmp, cr" %}
+ ins_encode %{
+ uint length_in_bytes = Matcher::vector_length_in_bytes(this, $vsrc);
+ assert(length_in_bytes == MaxVectorSize, "invalid vector length");
+ __ sve_reduce_integral(this->ideal_Opcode(), $dst$$Register, T_LONG,
+ $isrc$$Register, $vsrc$$FloatRegister,
+ ptrue, $tmp$$FloatRegister);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
+// reduction umin - predicated
+
+instruct reduce_uminI_masked(iRegINoSp dst, iRegIorL2I isrc, vReg vsrc, pRegGov pg,
+ vRegD tmp, rFlagsReg cr) %{
+ predicate(UseSVE > 0 &&
+ (Matcher::vector_element_basic_type(n->in(1)->in(2)) == T_BYTE ||
+ Matcher::vector_element_basic_type(n->in(1)->in(2)) == T_SHORT ||
+ Matcher::vector_element_basic_type(n->in(1)->in(2)) == T_INT));
+ match(Set dst (UMinReductionV (Binary isrc vsrc) pg));
+ effect(TEMP_DEF dst, TEMP tmp, KILL cr);
+ format %{ "reduce_uminI_masked $dst, $isrc, $pg, $vsrc\t# KILL $tmp, cr" %}
+ ins_encode %{
+ BasicType bt = Matcher::vector_element_basic_type(this, $vsrc);
+ __ sve_reduce_integral(this->ideal_Opcode(), $dst$$Register, bt,
+ $isrc$$Register, $vsrc$$FloatRegister,
+ $pg$$PRegister, $tmp$$FloatRegister);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
+instruct reduce_uminL_masked(iRegLNoSp dst, iRegL isrc, vReg vsrc, pRegGov pg,
+ vRegD tmp, rFlagsReg cr) %{
+ predicate(UseSVE > 0 && Matcher::vector_element_basic_type(n->in(1)->in(2)) == T_LONG);
+ match(Set dst (UMinReductionV (Binary isrc vsrc) pg));
+ effect(TEMP_DEF dst, TEMP tmp, KILL cr);
+ format %{ "reduce_uminL_masked $dst, $isrc, $pg, $vsrc\t# KILL $tmp, cr" %}
+ ins_encode %{
+ BasicType bt = Matcher::vector_element_basic_type(this, $vsrc);
+ __ sve_reduce_integral(this->ideal_Opcode(), $dst$$Register, bt,
+ $isrc$$Register, $vsrc$$FloatRegister,
+ $pg$$PRegister, $tmp$$FloatRegister);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
+// reduction umaxI
+
+instruct reduce_umaxI_neon(iRegINoSp dst, iRegIorL2I isrc, vReg vsrc,
+ vReg tmp, rFlagsReg cr) %{
+ predicate(VM_Version::use_neon_for_vector(Matcher::vector_length_in_bytes(n->in(2))) &&
+ (Matcher::vector_element_basic_type(n->in(2)) == T_BYTE ||
+ Matcher::vector_element_basic_type(n->in(2)) == T_SHORT ||
+ Matcher::vector_element_basic_type(n->in(2)) == T_INT));
+ match(Set dst (UMaxReductionV isrc vsrc));
+ effect(TEMP_DEF dst, TEMP tmp, KILL cr);
+ format %{ "reduce_umaxI_neon $dst, $isrc, $vsrc\t# KILL $tmp, cr" %}
+ ins_encode %{
+ BasicType bt = Matcher::vector_element_basic_type(this, $vsrc);
+ uint length_in_bytes = Matcher::vector_length_in_bytes(this, $vsrc);
+ __ neon_reduce_minmax_integral(this->ideal_Opcode(), $dst$$Register, bt,
+ $isrc$$Register, $vsrc$$FloatRegister,
+ length_in_bytes, $tmp$$FloatRegister);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
+instruct reduce_umaxI_sve(iRegINoSp dst, iRegIorL2I isrc, vReg vsrc,
+ vRegD tmp, rFlagsReg cr) %{
+ predicate(!VM_Version::use_neon_for_vector(Matcher::vector_length_in_bytes(n->in(2))) &&
+ (Matcher::vector_element_basic_type(n->in(2)) == T_BYTE ||
+ Matcher::vector_element_basic_type(n->in(2)) == T_SHORT ||
+ Matcher::vector_element_basic_type(n->in(2)) == T_INT));
+ match(Set dst (UMaxReductionV isrc vsrc));
+ effect(TEMP_DEF dst, TEMP tmp, KILL cr);
+ format %{ "reduce_umaxI_sve $dst, $isrc, $vsrc\t# KILL $tmp, cr" %}
+ ins_encode %{
+ assert(UseSVE > 0, "must be sve");
+ BasicType bt = Matcher::vector_element_basic_type(this, $vsrc);
+ uint length_in_bytes = Matcher::vector_length_in_bytes(this, $vsrc);
+ assert(length_in_bytes == MaxVectorSize, "invalid vector length");
+ __ sve_reduce_integral(this->ideal_Opcode(), $dst$$Register, bt,
+ $isrc$$Register, $vsrc$$FloatRegister,
+ ptrue, $tmp$$FloatRegister);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
+// reduction umaxL
+
+instruct reduce_umaxL_neon(iRegLNoSp dst, iRegL isrc, vReg vsrc, rFlagsReg cr) %{
+ predicate(UseSVE == 0 && Matcher::vector_element_basic_type(n->in(2)) == T_LONG);
+ match(Set dst (UMaxReductionV isrc vsrc));
+ effect(TEMP_DEF dst, KILL cr);
+ format %{ "reduce_umaxL_neon $dst, $isrc, $vsrc\t# 2L. KILL cr" %}
+ ins_encode %{
+ __ neon_reduce_minmax_integral(this->ideal_Opcode(), $dst$$Register, T_LONG,
+ $isrc$$Register, $vsrc$$FloatRegister,
+ /* vector_length_in_bytes */ 16, fnoreg);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
+instruct reduce_umaxL_sve(iRegLNoSp dst, iRegL isrc, vReg vsrc,
+ vRegD tmp, rFlagsReg cr) %{
+ predicate(UseSVE > 0 && Matcher::vector_element_basic_type(n->in(2)) == T_LONG);
+ match(Set dst (UMaxReductionV isrc vsrc));
+ effect(TEMP_DEF dst, TEMP tmp, KILL cr);
+ format %{ "reduce_umaxL_sve $dst, $isrc, $vsrc\t# KILL $tmp, cr" %}
+ ins_encode %{
+ uint length_in_bytes = Matcher::vector_length_in_bytes(this, $vsrc);
+ assert(length_in_bytes == MaxVectorSize, "invalid vector length");
+ __ sve_reduce_integral(this->ideal_Opcode(), $dst$$Register, T_LONG,
+ $isrc$$Register, $vsrc$$FloatRegister,
+ ptrue, $tmp$$FloatRegister);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
+// reduction umax - predicated
+
+instruct reduce_umaxI_masked(iRegINoSp dst, iRegIorL2I isrc, vReg vsrc, pRegGov pg,
+ vRegD tmp, rFlagsReg cr) %{
+ predicate(UseSVE > 0 &&
+ (Matcher::vector_element_basic_type(n->in(1)->in(2)) == T_BYTE ||
+ Matcher::vector_element_basic_type(n->in(1)->in(2)) == T_SHORT ||
+ Matcher::vector_element_basic_type(n->in(1)->in(2)) == T_INT));
+ match(Set dst (UMaxReductionV (Binary isrc vsrc) pg));
+ effect(TEMP_DEF dst, TEMP tmp, KILL cr);
+ format %{ "reduce_umaxI_masked $dst, $isrc, $pg, $vsrc\t# KILL $tmp, cr" %}
+ ins_encode %{
+ BasicType bt = Matcher::vector_element_basic_type(this, $vsrc);
+ __ sve_reduce_integral(this->ideal_Opcode(), $dst$$Register, bt,
+ $isrc$$Register, $vsrc$$FloatRegister,
+ $pg$$PRegister, $tmp$$FloatRegister);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
+instruct reduce_umaxL_masked(iRegLNoSp dst, iRegL isrc, vReg vsrc, pRegGov pg,
+ vRegD tmp, rFlagsReg cr) %{
+ predicate(UseSVE > 0 && Matcher::vector_element_basic_type(n->in(1)->in(2)) == T_LONG);
+ match(Set dst (UMaxReductionV (Binary isrc vsrc) pg));
+ effect(TEMP_DEF dst, TEMP tmp, KILL cr);
+ format %{ "reduce_umaxL_masked $dst, $isrc, $pg, $vsrc\t# KILL $tmp, cr" %}
+ ins_encode %{
+ BasicType bt = Matcher::vector_element_basic_type(this, $vsrc);
+ __ sve_reduce_integral(this->ideal_Opcode(), $dst$$Register, bt,
+ $isrc$$Register, $vsrc$$FloatRegister,
+ $pg$$PRegister, $tmp$$FloatRegister);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
// ------------------------------ Vector reinterpret ---------------------------
instruct reinterpret_same_size(vReg dst_src) %{
diff --git a/src/hotspot/cpu/aarch64/aarch64_vector_ad.m4 b/src/hotspot/cpu/aarch64/aarch64_vector_ad.m4
index 66dc22c3758..48bffb3cf35 100644
--- a/src/hotspot/cpu/aarch64/aarch64_vector_ad.m4
+++ b/src/hotspot/cpu/aarch64/aarch64_vector_ad.m4
@@ -1,5 +1,5 @@
//
-// Copyright (c) 2020, 2025, Oracle and/or its affiliates. All rights reserved.
+// Copyright (c) 2020, 2026, Oracle and/or its affiliates. All rights reserved.
// Copyright (c) 2020, 2025, Arm Limited. All rights reserved.
// DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
//
@@ -191,6 +191,8 @@ source %{
case Op_XorReductionV:
case Op_MinReductionV:
case Op_MaxReductionV:
+ case Op_UMinReductionV:
+ case Op_UMaxReductionV:
// Reductions with less than 8 bytes vector length are
// not supported.
if (length_in_bytes < 8) {
@@ -373,6 +375,8 @@ source %{
return !VM_Version::use_neon_for_vector(length_in_bytes);
case Op_MinReductionV:
case Op_MaxReductionV:
+ case Op_UMinReductionV:
+ case Op_UMaxReductionV:
// For BYTE/SHORT/INT/FLOAT/DOUBLE types, we prefer using NEON
// instructions rather than SVE predicated instructions for
// better performance.
@@ -2505,6 +2509,32 @@ REDUCE_MAXMIN_INT_PREDICATE(min, L, iRegL, MinReductionV)
REDUCE_MAXMIN_FP_PREDICATE(min, F, fsrc, MinReductionV, sve_fminv, fmins)
REDUCE_MAXMIN_FP_PREDICATE(min, D, dsrc, MinReductionV, sve_fminv, fmind)
+// -------------------- Vector reduction unsigned min/max ----------------------
+
+// reduction uminI
+REDUCE_MAXMIN_I_NEON(umin, UMinReductionV)
+REDUCE_MAXMIN_I_SVE(umin, UMinReductionV)
+
+// reduction uminL
+REDUCE_MAXMIN_L_NEON(umin, UMinReductionV)
+REDUCE_MAXMIN_L_SVE(umin, UMinReductionV)
+
+// reduction umin - predicated
+REDUCE_MAXMIN_INT_PREDICATE(umin, I, iRegIorL2I, UMinReductionV)
+REDUCE_MAXMIN_INT_PREDICATE(umin, L, iRegL, UMinReductionV)
+
+// reduction umaxI
+REDUCE_MAXMIN_I_NEON(umax, UMaxReductionV)
+REDUCE_MAXMIN_I_SVE(umax, UMaxReductionV)
+
+// reduction umaxL
+REDUCE_MAXMIN_L_NEON(umax, UMaxReductionV)
+REDUCE_MAXMIN_L_SVE(umax, UMaxReductionV)
+
+// reduction umax - predicated
+REDUCE_MAXMIN_INT_PREDICATE(umax, I, iRegIorL2I, UMaxReductionV)
+REDUCE_MAXMIN_INT_PREDICATE(umax, L, iRegL, UMaxReductionV)
+
// ------------------------------ Vector reinterpret ---------------------------
instruct reinterpret_same_size(vReg dst_src) %{
diff --git a/src/hotspot/cpu/aarch64/assembler_aarch64.hpp b/src/hotspot/cpu/aarch64/assembler_aarch64.hpp
index 18807c667e3..19b3bb1a65b 100644
--- a/src/hotspot/cpu/aarch64/assembler_aarch64.hpp
+++ b/src/hotspot/cpu/aarch64/assembler_aarch64.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2014, 2024, Red Hat Inc. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
@@ -2658,6 +2658,8 @@ template
INSN(uminv, 1, 0b011011, false); // accepted arrangements: T8B, T16B, T4H, T8H, T2S, T4S
INSN(smaxp, 0, 0b101001, false); // accepted arrangements: T8B, T16B, T4H, T8H, T2S, T4S
INSN(sminp, 0, 0b101011, false); // accepted arrangements: T8B, T16B, T4H, T8H, T2S, T4S
+ INSN(umaxp, 1, 0b101001, false); // accepted arrangements: T8B, T16B, T4H, T8H, T2S, T4S
+ INSN(uminp, 1, 0b101011, false); // accepted arrangements: T8B, T16B, T4H, T8H, T2S, T4S
INSN(sqdmulh,0, 0b101101, false); // accepted arrangements: T4H, T8H, T2S, T4S
INSN(shsubv, 0, 0b001001, false); // accepted arrangements: T8B, T16B, T4H, T8H, T2S, T4S
@@ -3490,7 +3492,9 @@ public:
INSN(sve_sub, 0b00000100, 0b000001000); // vector sub
INSN(sve_uaddv, 0b00000100, 0b000001001); // unsigned add reduction to scalar
INSN(sve_umax, 0b00000100, 0b001001000); // unsigned maximum vectors
+ INSN(sve_umaxv, 0b00000100, 0b001001001); // unsigned maximum reduction to scalar
INSN(sve_umin, 0b00000100, 0b001011000); // unsigned minimum vectors
+ INSN(sve_uminv, 0b00000100, 0b001011001); // unsigned minimum reduction to scalar
#undef INSN
// SVE floating-point arithmetic - predicate
@@ -4325,6 +4329,7 @@ public:
#undef INSN
Assembler(CodeBuffer* code) : AbstractAssembler(code) {
+ MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
}
// Stack overflow checking
diff --git a/src/hotspot/cpu/aarch64/c1_LIRAssembler_aarch64.cpp b/src/hotspot/cpu/aarch64/c1_LIRAssembler_aarch64.cpp
index 37a6a130e0d..c0621cbd5c2 100644
--- a/src/hotspot/cpu/aarch64/c1_LIRAssembler_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/c1_LIRAssembler_aarch64.cpp
@@ -1218,43 +1218,11 @@ void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
__ bind(*op->stub()->continuation());
}
-void LIR_Assembler::type_profile_helper(Register mdo,
- ciMethodData *md, ciProfileData *data,
- Register recv, Label* update_done) {
+void LIR_Assembler::type_profile_helper(Register mdo, ciMethodData *md,
+ ciProfileData *data, Register recv) {
- // Given a profile data offset, generate an Address which points to
- // the corresponding slot in mdo->data().
- // Clobbers rscratch2.
- auto slot_at = [=](ByteSize offset) -> Address {
- return __ form_address(rscratch2, mdo,
- md->byte_offset_of_slot(data, offset),
- LogBytesPerWord);
- };
-
- for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
- Label next_test;
- // See if the receiver is receiver[n].
- __ ldr(rscratch1, slot_at(ReceiverTypeData::receiver_offset(i)));
- __ cmp(recv, rscratch1);
- __ br(Assembler::NE, next_test);
- __ addptr(slot_at(ReceiverTypeData::receiver_count_offset(i)),
- DataLayout::counter_increment);
- __ b(*update_done);
- __ bind(next_test);
- }
-
- // Didn't find receiver; find next empty slot and fill it in
- for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
- Label next_test;
- Address recv_addr(slot_at(ReceiverTypeData::receiver_offset(i)));
- __ ldr(rscratch1, recv_addr);
- __ cbnz(rscratch1, next_test);
- __ str(recv, recv_addr);
- __ mov(rscratch1, DataLayout::counter_increment);
- __ str(rscratch1, slot_at(ReceiverTypeData::receiver_count_offset(i)));
- __ b(*update_done);
- __ bind(next_test);
- }
+ int mdp_offset = md->byte_offset_of_slot(data, in_ByteSize(0));
+ __ profile_receiver_type(recv, mdo, mdp_offset);
}
void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, Label* failure, Label* obj_is_null) {
@@ -1316,14 +1284,9 @@ void LIR_Assembler::emit_typecheck_helper(LIR_OpTypeCheck *op, Label* success, L
__ b(*obj_is_null);
__ bind(not_null);
- Label update_done;
Register recv = k_RInfo;
__ load_klass(recv, obj);
- type_profile_helper(mdo, md, data, recv, &update_done);
- Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
- __ addptr(counter_addr, DataLayout::counter_increment);
-
- __ bind(update_done);
+ type_profile_helper(mdo, md, data, recv);
} else {
__ cbz(obj, *obj_is_null);
}
@@ -1430,13 +1393,9 @@ void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
__ b(done);
__ bind(not_null);
- Label update_done;
Register recv = k_RInfo;
__ load_klass(recv, value);
- type_profile_helper(mdo, md, data, recv, &update_done);
- Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
- __ addptr(counter_addr, DataLayout::counter_increment);
- __ bind(update_done);
+ type_profile_helper(mdo, md, data, recv);
} else {
__ cbz(value, done);
}
@@ -2540,13 +2499,9 @@ void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
if (C1OptimizeVirtualCallProfiling && known_klass != nullptr) {
// We know the type that will be seen at this call site; we can
// statically update the MethodData* rather than needing to do
- // dynamic tests on the receiver type
-
- // NOTE: we should probably put a lock around this search to
- // avoid collisions by concurrent compilations
+ // dynamic tests on the receiver type.
ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
- uint i;
- for (i = 0; i < VirtualCallData::row_limit(); i++) {
+ for (uint i = 0; i < VirtualCallData::row_limit(); i++) {
ciKlass* receiver = vc_data->receiver(i);
if (known_klass->equals(receiver)) {
Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
@@ -2554,36 +2509,13 @@ void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
return;
}
}
-
- // Receiver type not found in profile data; select an empty slot
-
- // Note that this is less efficient than it should be because it
- // always does a write to the receiver part of the
- // VirtualCallData rather than just the first time
- for (i = 0; i < VirtualCallData::row_limit(); i++) {
- ciKlass* receiver = vc_data->receiver(i);
- if (receiver == nullptr) {
- __ mov_metadata(rscratch1, known_klass->constant_encoding());
- Address recv_addr =
- __ form_address(rscratch2, mdo,
- md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)),
- LogBytesPerWord);
- __ str(rscratch1, recv_addr);
- Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
- __ addptr(data_addr, DataLayout::counter_increment);
- return;
- }
- }
+ // Receiver type is not found in profile data.
+ // Fall back to runtime helper to handle the rest at runtime.
+ __ mov_metadata(recv, known_klass->constant_encoding());
} else {
__ load_klass(recv, recv);
- Label update_done;
- type_profile_helper(mdo, md, data, recv, &update_done);
- // Receiver did not match any saved receiver and there is no empty row for it.
- // Increment total counter to indicate polymorphic case.
- __ addptr(counter_addr, DataLayout::counter_increment);
-
- __ bind(update_done);
}
+ type_profile_helper(mdo, md, data, recv);
} else {
// Static call
__ addptr(counter_addr, DataLayout::counter_increment);
diff --git a/src/hotspot/cpu/aarch64/c1_LIRAssembler_aarch64.hpp b/src/hotspot/cpu/aarch64/c1_LIRAssembler_aarch64.hpp
index 21916a5f7dd..5af06fc6a1c 100644
--- a/src/hotspot/cpu/aarch64/c1_LIRAssembler_aarch64.hpp
+++ b/src/hotspot/cpu/aarch64/c1_LIRAssembler_aarch64.hpp
@@ -50,9 +50,8 @@ friend class ArrayCopyStub;
Address stack_slot_address(int index, uint shift, Register tmp, int adjust = 0);
// Record the type of the receiver in ReceiverTypeData
- void type_profile_helper(Register mdo,
- ciMethodData *md, ciProfileData *data,
- Register recv, Label* update_done);
+ void type_profile_helper(Register mdo, ciMethodData *md,
+ ciProfileData *data, Register recv);
void add_debug_info_for_branch(address adr, CodeEmitInfo* info);
void casw(Register addr, Register newval, Register cmpval);
diff --git a/src/hotspot/cpu/aarch64/c2_MacroAssembler_aarch64.cpp b/src/hotspot/cpu/aarch64/c2_MacroAssembler_aarch64.cpp
index 75897a16fe4..958855c7685 100644
--- a/src/hotspot/cpu/aarch64/c2_MacroAssembler_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/c2_MacroAssembler_aarch64.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2020, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2020, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -1960,50 +1960,76 @@ void C2_MacroAssembler::neon_reduce_logical(int opc, Register dst, BasicType bt,
BLOCK_COMMENT("} neon_reduce_logical");
}
-// Vector reduction min/max for integral type with ASIMD instructions.
+// Helper function to decode min/max reduction operation properties
+void C2_MacroAssembler::decode_minmax_reduction_opc(int opc, bool* is_min,
+ bool* is_unsigned,
+ Condition* cond) {
+ switch(opc) {
+ case Op_MinReductionV:
+ *is_min = true; *is_unsigned = false; *cond = LT; break;
+ case Op_MaxReductionV:
+ *is_min = false; *is_unsigned = false; *cond = GT; break;
+ case Op_UMinReductionV:
+ *is_min = true; *is_unsigned = true; *cond = LO; break;
+ case Op_UMaxReductionV:
+ *is_min = false; *is_unsigned = true; *cond = HI; break;
+ default:
+ ShouldNotReachHere();
+ }
+}
+
+// Vector reduction min/max/umin/umax for integral type with ASIMD instructions.
// Note: vtmp is not used and expected to be fnoreg for T_LONG case.
// Clobbers: rscratch1, rflags
void C2_MacroAssembler::neon_reduce_minmax_integral(int opc, Register dst, BasicType bt,
Register isrc, FloatRegister vsrc,
unsigned vector_length_in_bytes,
FloatRegister vtmp) {
- assert(opc == Op_MinReductionV || opc == Op_MaxReductionV, "unsupported");
+ assert(opc == Op_MinReductionV || opc == Op_MaxReductionV ||
+ opc == Op_UMinReductionV || opc == Op_UMaxReductionV, "unsupported");
assert(vector_length_in_bytes == 8 || vector_length_in_bytes == 16, "unsupported");
assert(bt == T_BYTE || bt == T_SHORT || bt == T_INT || bt == T_LONG, "unsupported");
assert_different_registers(dst, isrc);
bool isQ = vector_length_in_bytes == 16;
- bool is_min = opc == Op_MinReductionV;
-
+ bool is_min;
+ bool is_unsigned;
+ Condition cond;
+ decode_minmax_reduction_opc(opc, &is_min, &is_unsigned, &cond);
BLOCK_COMMENT("neon_reduce_minmax_integral {");
if (bt == T_LONG) {
assert(vtmp == fnoreg, "should be");
assert(isQ, "should be");
umov(rscratch1, vsrc, D, 0);
cmp(isrc, rscratch1);
- csel(dst, isrc, rscratch1, is_min ? LT : GT);
+ csel(dst, isrc, rscratch1, cond);
umov(rscratch1, vsrc, D, 1);
cmp(dst, rscratch1);
- csel(dst, dst, rscratch1, is_min ? LT : GT);
+ csel(dst, dst, rscratch1, cond);
} else {
SIMD_Arrangement size = esize2arrangement((unsigned)type2aelembytes(bt), isQ);
if (size == T2S) {
- is_min ? sminp(vtmp, size, vsrc, vsrc) : smaxp(vtmp, size, vsrc, vsrc);
+ // For T2S (2x32-bit elements), use pairwise instructions because
+ // uminv/umaxv/sminv/smaxv don't support arrangement 2S.
+ neon_minmaxp(is_unsigned, is_min, vtmp, size, vsrc, vsrc);
} else {
- is_min ? sminv(vtmp, size, vsrc) : smaxv(vtmp, size, vsrc);
+ // For other sizes, use reduction to scalar instructions.
+ neon_minmaxv(is_unsigned, is_min, vtmp, size, vsrc);
}
if (bt == T_INT) {
umov(dst, vtmp, S, 0);
+ } else if (is_unsigned) {
+ umov(dst, vtmp, elemType_to_regVariant(bt), 0);
} else {
smov(dst, vtmp, elemType_to_regVariant(bt), 0);
}
cmpw(dst, isrc);
- cselw(dst, dst, isrc, is_min ? LT : GT);
+ cselw(dst, dst, isrc, cond);
}
BLOCK_COMMENT("} neon_reduce_minmax_integral");
}
// Vector reduction for integral type with SVE instruction.
-// Supported operations are Add, And, Or, Xor, Max, Min.
+// Supported operations are Add, And, Or, Xor, Max, Min, UMax, UMin.
// rflags would be clobbered if opc is Op_MaxReductionV or Op_MinReductionV.
void C2_MacroAssembler::sve_reduce_integral(int opc, Register dst, BasicType bt, Register src1,
FloatRegister src2, PRegister pg, FloatRegister tmp) {
@@ -2075,35 +2101,27 @@ void C2_MacroAssembler::sve_reduce_integral(int opc, Register dst, BasicType bt,
}
break;
}
- case Op_MaxReductionV: {
- sve_smaxv(tmp, size, pg, src2);
- if (bt == T_INT || bt == T_LONG) {
+ case Op_MaxReductionV:
+ case Op_MinReductionV:
+ case Op_UMaxReductionV:
+ case Op_UMinReductionV: {
+ bool is_min;
+ bool is_unsigned;
+ Condition cond;
+ decode_minmax_reduction_opc(opc, &is_min, &is_unsigned, &cond);
+ sve_minmaxv(is_unsigned, is_min, tmp, size, pg, src2);
+ // Move result from vector to general register
+ if (is_unsigned || bt == T_INT || bt == T_LONG) {
umov(dst, tmp, size, 0);
} else {
smov(dst, tmp, size, 0);
}
if (bt == T_LONG) {
cmp(dst, src1);
- csel(dst, dst, src1, Assembler::GT);
+ csel(dst, dst, src1, cond);
} else {
cmpw(dst, src1);
- cselw(dst, dst, src1, Assembler::GT);
- }
- break;
- }
- case Op_MinReductionV: {
- sve_sminv(tmp, size, pg, src2);
- if (bt == T_INT || bt == T_LONG) {
- umov(dst, tmp, size, 0);
- } else {
- smov(dst, tmp, size, 0);
- }
- if (bt == T_LONG) {
- cmp(dst, src1);
- csel(dst, dst, src1, Assembler::LT);
- } else {
- cmpw(dst, src1);
- cselw(dst, dst, src1, Assembler::LT);
+ cselw(dst, dst, src1, cond);
}
break;
}
diff --git a/src/hotspot/cpu/aarch64/c2_MacroAssembler_aarch64.hpp b/src/hotspot/cpu/aarch64/c2_MacroAssembler_aarch64.hpp
index 412f0f37e9e..4f3a41da402 100644
--- a/src/hotspot/cpu/aarch64/c2_MacroAssembler_aarch64.hpp
+++ b/src/hotspot/cpu/aarch64/c2_MacroAssembler_aarch64.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2020, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2020, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -34,6 +34,37 @@
void neon_reduce_logical_helper(int opc, bool sf, Register Rd, Register Rn, Register Rm,
enum shift_kind kind = Assembler::LSL, unsigned shift = 0);
+ // Helper functions for min/max reduction operations
+
+ void decode_minmax_reduction_opc(int opc, bool* is_min, bool* is_unsigned, Condition* cond);
+
+ void neon_minmaxp(bool is_unsigned, bool is_min, FloatRegister dst,
+ SIMD_Arrangement size, FloatRegister src1, FloatRegister src2) {
+ auto m = is_unsigned ? (is_min ? &Assembler::uminp : &Assembler::umaxp)
+ : (is_min ? &Assembler::sminp : &Assembler::smaxp);
+ (this->*m)(dst, size, src1, src2);
+ }
+
+ // Typedefs used to disambiguate overloaded member functions.
+ typedef void (Assembler::*neon_reduction2)
+ (FloatRegister, SIMD_Arrangement, FloatRegister);
+
+ void neon_minmaxv(bool is_unsigned, bool is_min, FloatRegister dst,
+ SIMD_Arrangement size, FloatRegister src) {
+ auto m = is_unsigned ? (is_min ? (neon_reduction2)&Assembler::uminv
+ : (neon_reduction2)&Assembler::umaxv)
+ : (is_min ? &Assembler::sminv
+ : &Assembler::smaxv);
+ (this->*m)(dst, size, src);
+ }
+
+ void sve_minmaxv(bool is_unsigned, bool is_min, FloatRegister dst,
+ SIMD_RegVariant size, PRegister pg, FloatRegister src) {
+ auto m = is_unsigned ? (is_min ? &Assembler::sve_uminv : &Assembler::sve_umaxv)
+ : (is_min ? &Assembler::sve_sminv : &Assembler::sve_smaxv);
+ (this->*m)(dst, size, pg, src);
+ }
+
void select_from_two_vectors_neon(FloatRegister dst, FloatRegister src1,
FloatRegister src2, FloatRegister index,
FloatRegister tmp, unsigned vector_length_in_bytes);
diff --git a/src/hotspot/cpu/aarch64/gc/shared/barrierSetNMethod_aarch64.cpp b/src/hotspot/cpu/aarch64/gc/shared/barrierSetNMethod_aarch64.cpp
index 4d5ca01b6b4..3d5261c31d1 100644
--- a/src/hotspot/cpu/aarch64/gc/shared/barrierSetNMethod_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/gc/shared/barrierSetNMethod_aarch64.cpp
@@ -209,6 +209,10 @@ void BarrierSetNMethod::set_guard_value(nmethod* nm, int value, int bit_mask) {
bs_asm->increment_patching_epoch();
}
+ // Enable WXWrite: the function is called directly from nmethod_entry_barrier
+ // stub.
+ MACOS_AARCH64_ONLY(ThreadWXEnable wx(WXWrite, Thread::current()));
+
NativeNMethodBarrier barrier(nm);
barrier.set_value(value, bit_mask);
}
diff --git a/src/hotspot/cpu/aarch64/gc/shenandoah/shenandoahBarrierSetAssembler_aarch64.cpp b/src/hotspot/cpu/aarch64/gc/shenandoah/shenandoahBarrierSetAssembler_aarch64.cpp
index 9a035d9f40e..ad7bac4e067 100644
--- a/src/hotspot/cpu/aarch64/gc/shenandoah/shenandoahBarrierSetAssembler_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/gc/shenandoah/shenandoahBarrierSetAssembler_aarch64.cpp
@@ -85,26 +85,16 @@ void ShenandoahBarrierSetAssembler::arraycopy_epilogue(MacroAssembler* masm, Dec
}
}
-void ShenandoahBarrierSetAssembler::shenandoah_write_barrier_pre(MacroAssembler* masm,
- Register obj,
- Register pre_val,
- Register thread,
- Register tmp,
- bool tosca_live,
- bool expand_call) {
- if (ShenandoahSATBBarrier) {
- satb_write_barrier_pre(masm, obj, pre_val, thread, tmp, rscratch1, tosca_live, expand_call);
- }
-}
+void ShenandoahBarrierSetAssembler::satb_barrier(MacroAssembler* masm,
+ Register obj,
+ Register pre_val,
+ Register thread,
+ Register tmp1,
+ Register tmp2,
+ bool tosca_live,
+ bool expand_call) {
+ assert(ShenandoahSATBBarrier, "Should be checked by caller");
-void ShenandoahBarrierSetAssembler::satb_write_barrier_pre(MacroAssembler* masm,
- Register obj,
- Register pre_val,
- Register thread,
- Register tmp1,
- Register tmp2,
- bool tosca_live,
- bool expand_call) {
// If expand_call is true then we expand the call_VM_leaf macro
// directly to skip generating the check by
// InterpreterMacroAssembler::call_VM_leaf_base that checks _last_sp.
@@ -358,20 +348,20 @@ void ShenandoahBarrierSetAssembler::load_at(MacroAssembler* masm, DecoratorSet d
if (ShenandoahBarrierSet::need_keep_alive_barrier(decorators, type)) {
__ enter(/*strip_ret_addr*/true);
__ push_call_clobbered_registers();
- satb_write_barrier_pre(masm /* masm */,
- noreg /* obj */,
- dst /* pre_val */,
- rthread /* thread */,
- tmp1 /* tmp1 */,
- tmp2 /* tmp2 */,
- true /* tosca_live */,
- true /* expand_call */);
+ satb_barrier(masm /* masm */,
+ noreg /* obj */,
+ dst /* pre_val */,
+ rthread /* thread */,
+ tmp1 /* tmp1 */,
+ tmp2 /* tmp2 */,
+ true /* tosca_live */,
+ true /* expand_call */);
__ pop_call_clobbered_registers();
__ leave();
}
}
-void ShenandoahBarrierSetAssembler::store_check(MacroAssembler* masm, Register obj) {
+void ShenandoahBarrierSetAssembler::card_barrier(MacroAssembler* masm, Register obj) {
assert(ShenandoahCardBarrier, "Should have been checked by caller");
__ lsr(obj, obj, CardTable::card_shift());
@@ -394,13 +384,13 @@ void ShenandoahBarrierSetAssembler::store_check(MacroAssembler* masm, Register o
void ShenandoahBarrierSetAssembler::store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
Address dst, Register val, Register tmp1, Register tmp2, Register tmp3) {
- bool on_oop = is_reference_type(type);
- if (!on_oop) {
+ // 1: non-reference types require no barriers
+ if (!is_reference_type(type)) {
BarrierSetAssembler::store_at(masm, decorators, type, dst, val, tmp1, tmp2, tmp3);
return;
}
- // flatten object address if needed
+ // Flatten object address right away for simplicity: likely needed by barriers
if (dst.index() == noreg && dst.offset() == 0) {
if (dst.base() != tmp3) {
__ mov(tmp3, dst.base());
@@ -409,20 +399,26 @@ void ShenandoahBarrierSetAssembler::store_at(MacroAssembler* masm, DecoratorSet
__ lea(tmp3, dst);
}
- shenandoah_write_barrier_pre(masm,
- tmp3 /* obj */,
- tmp2 /* pre_val */,
- rthread /* thread */,
- tmp1 /* tmp */,
- val != noreg /* tosca_live */,
- false /* expand_call */);
+ bool storing_non_null = (val != noreg);
+ // 2: pre-barrier: SATB needs the previous value
+ if (ShenandoahBarrierSet::need_satb_barrier(decorators, type)) {
+ satb_barrier(masm,
+ tmp3 /* obj */,
+ tmp2 /* pre_val */,
+ rthread /* thread */,
+ tmp1 /* tmp */,
+ rscratch1 /* tmp2 */,
+ storing_non_null /* tosca_live */,
+ false /* expand_call */);
+ }
+
+ // Store!
BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp3, 0), val, noreg, noreg, noreg);
- bool in_heap = (decorators & IN_HEAP) != 0;
- bool needs_post_barrier = (val != noreg) && in_heap && ShenandoahCardBarrier;
- if (needs_post_barrier) {
- store_check(masm, tmp3);
+ // 3: post-barrier: card barrier needs store address
+ if (ShenandoahBarrierSet::need_card_barrier(decorators, type) && storing_non_null) {
+ card_barrier(masm, tmp3);
}
}
diff --git a/src/hotspot/cpu/aarch64/gc/shenandoah/shenandoahBarrierSetAssembler_aarch64.hpp b/src/hotspot/cpu/aarch64/gc/shenandoah/shenandoahBarrierSetAssembler_aarch64.hpp
index c0e708e1292..362fcae1ccd 100644
--- a/src/hotspot/cpu/aarch64/gc/shenandoah/shenandoahBarrierSetAssembler_aarch64.hpp
+++ b/src/hotspot/cpu/aarch64/gc/shenandoah/shenandoahBarrierSetAssembler_aarch64.hpp
@@ -40,23 +40,16 @@ class StubCodeGenerator;
class ShenandoahBarrierSetAssembler: public BarrierSetAssembler {
private:
- void satb_write_barrier_pre(MacroAssembler* masm,
- Register obj,
- Register pre_val,
- Register thread,
- Register tmp1,
- Register tmp2,
- bool tosca_live,
- bool expand_call);
- void shenandoah_write_barrier_pre(MacroAssembler* masm,
- Register obj,
- Register pre_val,
- Register thread,
- Register tmp,
- bool tosca_live,
- bool expand_call);
+ void satb_barrier(MacroAssembler* masm,
+ Register obj,
+ Register pre_val,
+ Register thread,
+ Register tmp1,
+ Register tmp2,
+ bool tosca_live,
+ bool expand_call);
- void store_check(MacroAssembler* masm, Register obj);
+ void card_barrier(MacroAssembler* masm, Register obj);
void resolve_forward_pointer(MacroAssembler* masm, Register dst, Register tmp = noreg);
void resolve_forward_pointer_not_null(MacroAssembler* masm, Register dst, Register tmp = noreg);
diff --git a/src/hotspot/cpu/aarch64/globals_aarch64.hpp b/src/hotspot/cpu/aarch64/globals_aarch64.hpp
index 8e520314c8b..a59e83c4b69 100644
--- a/src/hotspot/cpu/aarch64/globals_aarch64.hpp
+++ b/src/hotspot/cpu/aarch64/globals_aarch64.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2000, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2000, 2026, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2015, 2019, Red Hat Inc. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
@@ -39,7 +39,7 @@ define_pd_global(bool, UncommonNullCast, true); // Uncommon-trap nulls
define_pd_global(bool, DelayCompilerStubsGeneration, COMPILER2_OR_JVMCI);
define_pd_global(size_t, CodeCacheSegmentSize, 64);
-define_pd_global(intx, CodeEntryAlignment, 64);
+define_pd_global(uint, CodeEntryAlignment, 64);
define_pd_global(intx, OptoLoopAlignment, 16);
#define DEFAULT_STACK_YELLOW_PAGES (2)
diff --git a/src/hotspot/cpu/aarch64/interp_masm_aarch64.cpp b/src/hotspot/cpu/aarch64/interp_masm_aarch64.cpp
index 957c2aee1c1..2b506b241e0 100644
--- a/src/hotspot/cpu/aarch64/interp_masm_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/interp_masm_aarch64.cpp
@@ -240,15 +240,14 @@ void InterpreterMacroAssembler::load_resolved_klass_at_offset(
// Rsub_klass: subklass
//
// Kills:
-// r2, r5
+// r2
void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
Label& ok_is_subtype) {
assert(Rsub_klass != r0, "r0 holds superklass");
assert(Rsub_klass != r2, "r2 holds 2ndary super array length");
- assert(Rsub_klass != r5, "r5 holds 2ndary super array scan ptr");
// Profile the not-null value's klass.
- profile_typecheck(r2, Rsub_klass, r5); // blows r2, reloads r5
+ profile_typecheck(r2, Rsub_klass); // blows r2
// Do the check.
check_klass_subtype(Rsub_klass, r0, r2, ok_is_subtype); // blows r2
@@ -991,7 +990,6 @@ void InterpreterMacroAssembler::profile_final_call(Register mdp) {
void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
Register mdp,
- Register reg2,
bool receiver_can_be_null) {
if (ProfileInterpreter) {
Label profile_continue;
@@ -1009,7 +1007,7 @@ void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
}
// Record the receiver type.
- record_klass_in_profile(receiver, mdp, reg2);
+ profile_receiver_type(receiver, mdp, 0);
bind(skip_receiver_profile);
// The method data pointer needs to be updated to reflect the new target.
@@ -1018,131 +1016,6 @@ void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
}
}
-// This routine creates a state machine for updating the multi-row
-// type profile at a virtual call site (or other type-sensitive bytecode).
-// The machine visits each row (of receiver/count) until the receiver type
-// is found, or until it runs out of rows. At the same time, it remembers
-// the location of the first empty row. (An empty row records null for its
-// receiver, and can be allocated for a newly-observed receiver type.)
-// Because there are two degrees of freedom in the state, a simple linear
-// search will not work; it must be a decision tree. Hence this helper
-// function is recursive, to generate the required tree structured code.
-// It's the interpreter, so we are trading off code space for speed.
-// See below for example code.
-void InterpreterMacroAssembler::record_klass_in_profile_helper(
- Register receiver, Register mdp,
- Register reg2, int start_row,
- Label& done) {
- if (TypeProfileWidth == 0) {
- increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
- } else {
- record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
- &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
- }
-}
-
-void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
- Register reg2, int start_row, Label& done, int total_rows,
- OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn) {
- int last_row = total_rows - 1;
- assert(start_row <= last_row, "must be work left to do");
- // Test this row for both the item and for null.
- // Take any of three different outcomes:
- // 1. found item => increment count and goto done
- // 2. found null => keep looking for case 1, maybe allocate this cell
- // 3. found something else => keep looking for cases 1 and 2
- // Case 3 is handled by a recursive call.
- for (int row = start_row; row <= last_row; row++) {
- Label next_test;
- bool test_for_null_also = (row == start_row);
-
- // See if the item is item[n].
- int item_offset = in_bytes(item_offset_fn(row));
- test_mdp_data_at(mdp, item_offset, item,
- (test_for_null_also ? reg2 : noreg),
- next_test);
- // (Reg2 now contains the item from the CallData.)
-
- // The item is item[n]. Increment count[n].
- int count_offset = in_bytes(item_count_offset_fn(row));
- increment_mdp_data_at(mdp, count_offset);
- b(done);
- bind(next_test);
-
- if (test_for_null_also) {
- Label found_null;
- // Failed the equality check on item[n]... Test for null.
- if (start_row == last_row) {
- // The only thing left to do is handle the null case.
- cbz(reg2, found_null);
- // Item did not match any saved item and there is no empty row for it.
- // Increment total counter to indicate polymorphic case.
- increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
- b(done);
- bind(found_null);
- break;
- }
- // Since null is rare, make it be the branch-taken case.
- cbz(reg2, found_null);
-
- // Put all the "Case 3" tests here.
- record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
- item_offset_fn, item_count_offset_fn);
-
- // Found a null. Keep searching for a matching item,
- // but remember that this is an empty (unused) slot.
- bind(found_null);
- }
- }
-
- // In the fall-through case, we found no matching item, but we
- // observed the item[start_row] is null.
-
- // Fill in the item field and increment the count.
- int item_offset = in_bytes(item_offset_fn(start_row));
- set_mdp_data_at(mdp, item_offset, item);
- int count_offset = in_bytes(item_count_offset_fn(start_row));
- mov(reg2, DataLayout::counter_increment);
- set_mdp_data_at(mdp, count_offset, reg2);
- if (start_row > 0) {
- b(done);
- }
-}
-
-// Example state machine code for three profile rows:
-// // main copy of decision tree, rooted at row[1]
-// if (row[0].rec == rec) { row[0].incr(); goto done; }
-// if (row[0].rec != nullptr) {
-// // inner copy of decision tree, rooted at row[1]
-// if (row[1].rec == rec) { row[1].incr(); goto done; }
-// if (row[1].rec != nullptr) {
-// // degenerate decision tree, rooted at row[2]
-// if (row[2].rec == rec) { row[2].incr(); goto done; }
-// if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow
-// row[2].init(rec); goto done;
-// } else {
-// // remember row[1] is empty
-// if (row[2].rec == rec) { row[2].incr(); goto done; }
-// row[1].init(rec); goto done;
-// }
-// } else {
-// // remember row[0] is empty
-// if (row[1].rec == rec) { row[1].incr(); goto done; }
-// if (row[2].rec == rec) { row[2].incr(); goto done; }
-// row[0].init(rec); goto done;
-// }
-// done:
-
-void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
- Register mdp, Register reg2) {
- assert(ProfileInterpreter, "must be profiling");
- Label done;
-
- record_klass_in_profile_helper(receiver, mdp, reg2, 0, done);
-
- bind (done);
-}
-
void InterpreterMacroAssembler::profile_ret(Register return_bci,
Register mdp) {
if (ProfileInterpreter) {
@@ -1200,7 +1073,7 @@ void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
}
}
-void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
+void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass) {
if (ProfileInterpreter) {
Label profile_continue;
@@ -1213,7 +1086,7 @@ void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass,
mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
// Record the object type.
- record_klass_in_profile(klass, mdp, reg2);
+ profile_receiver_type(klass, mdp, 0);
}
update_mdp_by_constant(mdp, mdp_delta);
diff --git a/src/hotspot/cpu/aarch64/interp_masm_aarch64.hpp b/src/hotspot/cpu/aarch64/interp_masm_aarch64.hpp
index 2b230a3b73e..74d4430000d 100644
--- a/src/hotspot/cpu/aarch64/interp_masm_aarch64.hpp
+++ b/src/hotspot/cpu/aarch64/interp_masm_aarch64.hpp
@@ -273,15 +273,6 @@ class InterpreterMacroAssembler: public MacroAssembler {
Register test_value_out,
Label& not_equal_continue);
- void record_klass_in_profile(Register receiver, Register mdp,
- Register reg2);
- void record_klass_in_profile_helper(Register receiver, Register mdp,
- Register reg2, int start_row,
- Label& done);
- void record_item_in_profile_helper(Register item, Register mdp,
- Register reg2, int start_row, Label& done, int total_rows,
- OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn);
-
void update_mdp_by_offset(Register mdp_in, int offset_of_offset);
void update_mdp_by_offset(Register mdp_in, Register reg, int offset_of_disp);
void update_mdp_by_constant(Register mdp_in, int constant);
@@ -295,11 +286,10 @@ class InterpreterMacroAssembler: public MacroAssembler {
void profile_call(Register mdp);
void profile_final_call(Register mdp);
void profile_virtual_call(Register receiver, Register mdp,
- Register scratch2,
bool receiver_can_be_null = false);
void profile_ret(Register return_bci, Register mdp);
void profile_null_seen(Register mdp);
- void profile_typecheck(Register mdp, Register klass, Register scratch);
+ void profile_typecheck(Register mdp, Register klass);
void profile_typecheck_failed(Register mdp);
void profile_switch_default(Register mdp);
void profile_switch_case(Register index_in_scratch, Register mdp,
diff --git a/src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp b/src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp
index 27428a5c558..409343b6b8d 100644
--- a/src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2014, 2024, Red Hat Inc. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
@@ -473,6 +473,7 @@ address MacroAssembler::target_addr_for_insn(address insn_addr) {
// Patch any kind of instruction; there may be several instructions.
// Return the total length (in bytes) of the instructions.
int MacroAssembler::pd_patch_instruction_size(address insn_addr, address target) {
+ MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
return RelocActions::run(insn_addr, target);
}
@@ -481,6 +482,8 @@ int MacroAssembler::patch_oop(address insn_addr, address o) {
unsigned insn = *(unsigned*)insn_addr;
assert(nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
+ MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
+
// OOPs are either narrow (32 bits) or wide (48 bits). We encode
// narrow OOPs by setting the upper 16 bits in the first
// instruction.
@@ -510,6 +513,8 @@ int MacroAssembler::patch_narrow_klass(address insn_addr, narrowKlass n) {
assert(Instruction_aarch64::extract(insn->encoding(), 31, 21) == 0b11010010101 &&
nativeInstruction_at(insn_addr+4)->is_movk(), "wrong insns in patch");
+ MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
+
Instruction_aarch64::patch(insn_addr, 20, 5, n >> 16);
Instruction_aarch64::patch(insn_addr+4, 20, 5, n & 0xffff);
return 2 * NativeInstruction::instruction_size;
@@ -2118,6 +2123,161 @@ Address MacroAssembler::argument_address(RegisterOrConstant arg_slot,
}
}
+// Handle the receiver type profile update given the "recv" klass.
+//
+// Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
+// If there are no matching or claimable receiver entries in RD, updates
+// the polymorphic counter.
+//
+// This code expected to run by either the interpreter or JIT-ed code, without
+// extra synchronization. For safety, receiver cells are claimed atomically, which
+// avoids grossly misrepresenting the profiles under concurrent updates. For speed,
+// counter updates are not atomic.
+//
+void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
+ assert_different_registers(recv, mdp, rscratch1, rscratch2);
+
+ int base_receiver_offset = in_bytes(ReceiverTypeData::receiver_offset(0));
+ int end_receiver_offset = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
+ int poly_count_offset = in_bytes(CounterData::count_offset());
+ int receiver_step = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
+ int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
+
+ // Adjust for MDP offsets.
+ base_receiver_offset += mdp_offset;
+ end_receiver_offset += mdp_offset;
+ poly_count_offset += mdp_offset;
+
+#ifdef ASSERT
+ // We are about to walk the MDO slots without asking for offsets.
+ // Check that our math hits all the right spots.
+ for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
+ int real_recv_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
+ int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
+ int offset = base_receiver_offset + receiver_step*c;
+ int count_offset = offset + receiver_to_count_step;
+ assert(offset == real_recv_offset, "receiver slot math");
+ assert(count_offset == real_count_offset, "receiver count math");
+ }
+ int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
+ assert(poly_count_offset == real_poly_count_offset, "poly counter math");
+#endif
+
+ // Corner case: no profile table. Increment poly counter and exit.
+ if (ReceiverTypeData::row_limit() == 0) {
+ increment(Address(mdp, poly_count_offset), DataLayout::counter_increment);
+ return;
+ }
+
+ Register offset = rscratch2;
+
+ Label L_loop_search_receiver, L_loop_search_empty;
+ Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
+
+ // The code here recognizes three major cases:
+ // A. Fastest: receiver found in the table
+ // B. Fast: no receiver in the table, and the table is full
+ // C. Slow: no receiver in the table, free slots in the table
+ //
+ // The case A performance is most important, as perfectly-behaved code would end up
+ // there, especially with larger TypeProfileWidth. The case B performance is
+ // important as well, this is where bulk of code would land for normally megamorphic
+ // cases. The case C performance is not essential, its job is to deal with installation
+ // races, we optimize for code density instead. Case C needs to make sure that receiver
+ // rows are only claimed once. This makes sure we never overwrite a row for another
+ // receiver and never duplicate the receivers in the list, making profile type-accurate.
+ //
+ // It is very tempting to handle these cases in a single loop, and claim the first slot
+ // without checking the rest of the table. But, profiling code should tolerate free slots
+ // in the table, as class unloading can clear them. After such cleanup, the receiver
+ // we need might be _after_ the free slot. Therefore, we need to let at least full scan
+ // to complete, before trying to install new slots. Splitting the code in several tight
+ // loops also helpfully optimizes for cases A and B.
+ //
+ // This code is effectively:
+ //
+ // restart:
+ // // Fastest: receiver is already installed
+ // for (i = 0; i < receiver_count(); i++) {
+ // if (receiver(i) == recv) goto found_recv(i);
+ // }
+ //
+ // // Fast: no receiver, but profile is full
+ // for (i = 0; i < receiver_count(); i++) {
+ // if (receiver(i) == null) goto found_null(i);
+ // }
+ // goto polymorphic
+ //
+ // // Slow: try to install receiver
+ // found_null(i):
+ // CAS(&receiver(i), null, recv);
+ // goto restart
+ //
+ // polymorphic:
+ // count++;
+ // return
+ //
+ // found_recv(i):
+ // *receiver_count(i)++
+ //
+
+ bind(L_restart);
+
+ // Fastest: receiver is already installed
+ mov(offset, base_receiver_offset);
+ bind(L_loop_search_receiver);
+ ldr(rscratch1, Address(mdp, offset));
+ cmp(rscratch1, recv);
+ br(Assembler::EQ, L_found_recv);
+ add(offset, offset, receiver_step);
+ sub(rscratch1, offset, end_receiver_offset);
+ cbnz(rscratch1, L_loop_search_receiver);
+
+ // Fast: no receiver, but profile is full
+ mov(offset, base_receiver_offset);
+ bind(L_loop_search_empty);
+ ldr(rscratch1, Address(mdp, offset));
+ cbz(rscratch1, L_found_empty);
+ add(offset, offset, receiver_step);
+ sub(rscratch1, offset, end_receiver_offset);
+ cbnz(rscratch1, L_loop_search_empty);
+ b(L_polymorphic);
+
+ // Slow: try to install receiver
+ bind(L_found_empty);
+
+ // Atomically swing receiver slot: null -> recv.
+ //
+ // The update uses CAS, which clobbers rscratch1. Therefore, rscratch2
+ // is used to hold the destination address. This is safe because the
+ // offset is no longer needed after the address is computed.
+
+ lea(rscratch2, Address(mdp, offset));
+ cmpxchg(/*addr*/ rscratch2, /*expected*/ zr, /*new*/ recv, Assembler::xword,
+ /*acquire*/ false, /*release*/ false, /*weak*/ true, noreg);
+
+ // CAS success means the slot now has the receiver we want. CAS failure means
+ // something had claimed the slot concurrently: it can be the same receiver we want,
+ // or something else. Since this is a slow path, we can optimize for code density,
+ // and just restart the search from the beginning.
+ b(L_restart);
+
+ // Counter updates:
+
+ // Increment polymorphic counter instead of receiver slot.
+ bind(L_polymorphic);
+ mov(offset, poly_count_offset);
+ b(L_count_update);
+
+ // Found a receiver, convert its slot offset to corresponding count offset.
+ bind(L_found_recv);
+ add(offset, offset, receiver_to_count_step);
+
+ bind(L_count_update);
+ increment(Address(mdp, offset), DataLayout::counter_increment);
+}
+
+
void MacroAssembler::call_VM_leaf_base(address entry_point,
int number_of_arguments,
Label *retaddr) {
@@ -5606,12 +5766,11 @@ void MacroAssembler::adrp(Register reg1, const Address &dest, uint64_t &byte_off
}
void MacroAssembler::load_byte_map_base(Register reg) {
- CardTable::CardValue* byte_map_base =
- ((CardTableBarrierSet*)(BarrierSet::barrier_set()))->card_table()->byte_map_base();
+ CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
- // Strictly speaking the byte_map_base isn't an address at all, and it might
+ // Strictly speaking the card table base isn't an address at all, and it might
// even be negative. It is thus materialised as a constant.
- mov(reg, (uint64_t)byte_map_base);
+ mov(reg, (uint64_t)ctbs->card_table_base_const());
}
void MacroAssembler::build_frame(int framesize) {
@@ -6265,10 +6424,14 @@ void MacroAssembler::fill_words(Register base, Register cnt, Register value)
// Intrinsic for
//
-// - sun/nio/cs/ISO_8859_1$Encoder.implEncodeISOArray
-// return the number of characters copied.
-// - java/lang/StringUTF16.compress
-// return index of non-latin1 character if copy fails, otherwise 'len'.
+// - sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
+// Encodes char[] to byte[] in ISO-8859-1
+//
+// - java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
+// Encodes byte[] (containing UTF-16) to byte[] in ISO-8859-1
+//
+// - java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
+// Encodes char[] to byte[] in ASCII
//
// This version always returns the number of characters copied, and does not
// clobber the 'len' register. A successful copy will complete with the post-
diff --git a/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp b/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp
index 4baa07d7d49..7eb6cea0614 100644
--- a/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp
+++ b/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp
@@ -1122,6 +1122,8 @@ public:
Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
+ void profile_receiver_type(Register recv, Register mdp, int mdp_offset);
+
void verify_sve_vector_length(Register tmp = rscratch1);
void reinitialize_ptrue() {
if (UseSVE > 0) {
diff --git a/src/hotspot/cpu/aarch64/nativeInst_aarch64.cpp b/src/hotspot/cpu/aarch64/nativeInst_aarch64.cpp
index f2003dd9b55..0cdf36f0bc5 100644
--- a/src/hotspot/cpu/aarch64/nativeInst_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/nativeInst_aarch64.cpp
@@ -133,7 +133,6 @@ void NativeMovConstReg::verify() {
intptr_t NativeMovConstReg::data() const {
- // das(uint64_t(instruction_address()),2);
address addr = MacroAssembler::target_addr_for_insn(instruction_address());
if (maybe_cpool_ref(instruction_address())) {
return *(intptr_t*)addr;
@@ -144,6 +143,7 @@ intptr_t NativeMovConstReg::data() const {
void NativeMovConstReg::set_data(intptr_t x) {
if (maybe_cpool_ref(instruction_address())) {
+ MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
address addr = MacroAssembler::target_addr_for_insn(instruction_address());
*(intptr_t*)addr = x;
} else {
@@ -350,8 +350,6 @@ bool NativeInstruction::is_stop() {
//-------------------------------------------------------------------
-void NativeGeneralJump::verify() { }
-
// MT-safe patching of a long jump instruction.
void NativeGeneralJump::replace_mt_safe(address instr_addr, address code_buffer) {
ShouldNotCallThis();
diff --git a/src/hotspot/cpu/aarch64/nativeInst_aarch64.hpp b/src/hotspot/cpu/aarch64/nativeInst_aarch64.hpp
index c30cb911d96..15b6c9ff215 100644
--- a/src/hotspot/cpu/aarch64/nativeInst_aarch64.hpp
+++ b/src/hotspot/cpu/aarch64/nativeInst_aarch64.hpp
@@ -90,16 +90,18 @@ protected:
s_char sbyte_at(int offset) const { return *(s_char*)addr_at(offset); }
u_char ubyte_at(int offset) const { return *(u_char*)addr_at(offset); }
- jint int_at(int offset) const { return *(jint*)addr_at(offset); }
- juint uint_at(int offset) const { return *(juint*)addr_at(offset); }
- address ptr_at(int offset) const { return *(address*)addr_at(offset); }
- oop oop_at(int offset) const { return *(oop*)addr_at(offset); }
+ jint int_at(int offset) const { return *(jint*)addr_at(offset); }
+ juint uint_at(int offset) const { return *(juint*)addr_at(offset); }
+ address ptr_at(int offset) const { return *(address*)addr_at(offset); }
+ oop oop_at(int offset) const { return *(oop*)addr_at(offset); }
- void set_char_at(int offset, char c) { *addr_at(offset) = (u_char)c; }
- void set_int_at(int offset, jint i) { *(jint*)addr_at(offset) = i; }
- void set_uint_at(int offset, jint i) { *(juint*)addr_at(offset) = i; }
- void set_ptr_at(int offset, address ptr) { *(address*)addr_at(offset) = ptr; }
- void set_oop_at(int offset, oop o) { *(oop*)addr_at(offset) = o; }
+#define MACOS_WX_WRITE MACOS_AARCH64_ONLY(os::thread_wx_enable_write())
+ void set_char_at(int offset, char c) { MACOS_WX_WRITE; *addr_at(offset) = (u_char)c; }
+ void set_int_at(int offset, jint i) { MACOS_WX_WRITE; *(jint*)addr_at(offset) = i; }
+ void set_uint_at(int offset, jint i) { MACOS_WX_WRITE; *(juint*)addr_at(offset) = i; }
+ void set_ptr_at(int offset, address ptr) { MACOS_WX_WRITE; *(address*)addr_at(offset) = ptr; }
+ void set_oop_at(int offset, oop o) { MACOS_WX_WRITE; *(oop*)addr_at(offset) = o; }
+#undef MACOS_WX_WRITE
void wrote(int offset);
@@ -380,7 +382,6 @@ public:
void set_jump_destination(address dest);
static void replace_mt_safe(address instr_addr, address code_buffer);
- static void verify();
};
inline NativeGeneralJump* nativeGeneralJump_at(address address) {
diff --git a/src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp b/src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp
index 7e2f333ba40..a459a28b09e 100644
--- a/src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp
@@ -6081,14 +6081,18 @@ class StubGenerator: public StubCodeGenerator {
// static int implKyber12To16(
// byte[] condensed, int index, short[] parsed, int parsedLength) {}
//
- // (parsedLength or (parsedLength - 48) must be divisible by 64.)
+ // we assume that parsed and condensed are allocated such that for
+ // n = (parsedLength + 63) / 64
+ // n blocks of 96 bytes of input can be processed, i.e.
+ // index + n * 96 <= condensed.length and
+ // n * 64 <= parsed.length
//
// condensed (byte[]) = c_rarg0
// condensedIndex = c_rarg1
- // parsed (short[112 or 256]) = c_rarg2
- // parsedLength (112 or 256) = c_rarg3
+ // parsed (short[]) = c_rarg2
+ // parsedLength = c_rarg3
address generate_kyber12To16() {
- Label L_F00, L_loop, L_end;
+ Label L_F00, L_loop;
__ align(CodeEntryAlignment);
StubId stub_id = StubId::stubgen_kyber12To16_id;
@@ -6209,75 +6213,8 @@ class StubGenerator: public StubCodeGenerator {
vs_st2_post(vs_front(vb), __ T8H, parsed);
__ sub(parsedLength, parsedLength, 64);
- __ cmp(parsedLength, (u1)64);
- __ br(Assembler::GE, L_loop);
- __ cbz(parsedLength, L_end);
-
- // if anything is left it should be a final 72 bytes of input
- // i.e. a final 48 12-bit values. so we handle this by loading
- // 48 bytes into all 16B lanes of front(vin) and only 24
- // bytes into the lower 8B lane of back(vin)
- vs_ld3_post(vs_front(vin), __ T16B, condensed);
- vs_ld3(vs_back(vin), __ T8B, condensed);
-
- // Expand vin[0] into va[0:1], and vin[1] into va[2:3] and va[4:5]
- // n.b. target elements 2 and 3 of va duplicate elements 4 and
- // 5 and target element 2 of vb duplicates element 4.
- __ ushll(va[0], __ T8H, vin[0], __ T8B, 0);
- __ ushll2(va[1], __ T8H, vin[0], __ T16B, 0);
- __ ushll(va[2], __ T8H, vin[1], __ T8B, 0);
- __ ushll2(va[3], __ T8H, vin[1], __ T16B, 0);
- __ ushll(va[4], __ T8H, vin[1], __ T8B, 0);
- __ ushll2(va[5], __ T8H, vin[1], __ T16B, 0);
-
- // This time expand just the lower 8 lanes
- __ ushll(vb[0], __ T8H, vin[3], __ T8B, 0);
- __ ushll(vb[2], __ T8H, vin[4], __ T8B, 0);
- __ ushll(vb[4], __ T8H, vin[4], __ T8B, 0);
-
- // shift lo byte of copy 1 of the middle stripe into the high byte
- __ shl(va[2], __ T8H, va[2], 8);
- __ shl(va[3], __ T8H, va[3], 8);
- __ shl(vb[2], __ T8H, vb[2], 8);
-
- // expand vin[2] into va[6:7] and lower 8 lanes of vin[5] into
- // vb[6] pre-shifted by 4 to ensure top bits of the input 12-bit
- // int are in bit positions [4..11].
- __ ushll(va[6], __ T8H, vin[2], __ T8B, 4);
- __ ushll2(va[7], __ T8H, vin[2], __ T16B, 4);
- __ ushll(vb[6], __ T8H, vin[5], __ T8B, 4);
-
- // mask hi 4 bits of each 1st 12-bit int in pair from copy1 and
- // shift lo 4 bits of each 2nd 12-bit int in pair to bottom of
- // copy2
- __ andr(va[2], __ T16B, va[2], v31);
- __ andr(va[3], __ T16B, va[3], v31);
- __ ushr(va[4], __ T8H, va[4], 4);
- __ ushr(va[5], __ T8H, va[5], 4);
- __ andr(vb[2], __ T16B, vb[2], v31);
- __ ushr(vb[4], __ T8H, vb[4], 4);
-
-
-
- // sum hi 4 bits and lo 8 bits of each 1st 12-bit int in pair and
- // hi 8 bits plus lo 4 bits of each 2nd 12-bit int in pair
-
- // n.b. ordering ensures: i) inputs are consumed before they are
- // overwritten ii) order of 16-bit results across succsessive
- // pairs of vectors in va and then lower half of vb reflects order
- // of corresponding 12-bit inputs
- __ addv(va[0], __ T8H, va[0], va[2]);
- __ addv(va[2], __ T8H, va[1], va[3]);
- __ addv(va[1], __ T8H, va[4], va[6]);
- __ addv(va[3], __ T8H, va[5], va[7]);
- __ addv(vb[0], __ T8H, vb[0], vb[2]);
- __ addv(vb[1], __ T8H, vb[4], vb[6]);
-
- // store 48 results interleaved as shorts
- vs_st2_post(vs_front(va), __ T8H, parsed);
- vs_st2_post(vs_front(vs_front(vb)), __ T8H, parsed);
-
- __ BIND(L_end);
+ __ cmp(parsedLength, (u1)0);
+ __ br(Assembler::GT, L_loop);
__ leave(); // required for proper stackwalking of RuntimeStub frame
__ mov(r0, zr); // return 0
@@ -11805,7 +11742,9 @@ class StubGenerator: public StubCodeGenerator {
}
#endif
- StubRoutines::_unsafe_setmemory = generate_unsafe_setmemory();
+ if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_setMemory)) {
+ StubRoutines::_unsafe_setmemory = generate_unsafe_setmemory();
+ }
StubRoutines::aarch64::set_completed(); // Inidicate that arraycopy and zero_blocks stubs are generated
}
diff --git a/src/hotspot/cpu/aarch64/templateTable_aarch64.cpp b/src/hotspot/cpu/aarch64/templateTable_aarch64.cpp
index 07b469650f0..5d4f7103a84 100644
--- a/src/hotspot/cpu/aarch64/templateTable_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/templateTable_aarch64.cpp
@@ -3370,7 +3370,7 @@ void TemplateTable::invokevirtual_helper(Register index,
__ load_klass(r0, recv);
// profile this call
- __ profile_virtual_call(r0, rlocals, r3);
+ __ profile_virtual_call(r0, rlocals);
// get target Method & entry point
__ lookup_virtual_method(r0, index, method);
@@ -3500,7 +3500,7 @@ void TemplateTable::invokeinterface(int byte_no) {
/*return_method=*/false);
// profile this call
- __ profile_virtual_call(r3, r13, r19);
+ __ profile_virtual_call(r3, r13);
// Get declaring interface class from method, and itable index
diff --git a/src/hotspot/cpu/aarch64/vm_version_aarch64.cpp b/src/hotspot/cpu/aarch64/vm_version_aarch64.cpp
index 659c231464a..0a7bc5a8962 100644
--- a/src/hotspot/cpu/aarch64/vm_version_aarch64.cpp
+++ b/src/hotspot/cpu/aarch64/vm_version_aarch64.cpp
@@ -201,16 +201,14 @@ void VM_Version::initialize() {
}
}
- // Cortex A53
- if (_cpu == CPU_ARM && model_is(0xd03)) {
+ if (_cpu == CPU_ARM && model_is(CPU_MODEL_ARM_CORTEX_A53)) {
set_feature(CPU_A53MAC);
if (FLAG_IS_DEFAULT(UseSIMDForArrayEquals)) {
FLAG_SET_DEFAULT(UseSIMDForArrayEquals, false);
}
}
- // Cortex A73
- if (_cpu == CPU_ARM && model_is(0xd09)) {
+ if (_cpu == CPU_ARM && model_is(CPU_MODEL_ARM_CORTEX_A73)) {
if (FLAG_IS_DEFAULT(SoftwarePrefetchHintDistance)) {
FLAG_SET_DEFAULT(SoftwarePrefetchHintDistance, -1);
}
@@ -220,16 +218,11 @@ void VM_Version::initialize() {
}
}
- // Neoverse
- // N1: 0xd0c
- // N2: 0xd49
- // N3: 0xd8e
- // V1: 0xd40
- // V2: 0xd4f
- // V3: 0xd84
- if (_cpu == CPU_ARM && (model_is(0xd0c) || model_is(0xd49) ||
- model_is(0xd40) || model_is(0xd4f) ||
- model_is(0xd8e) || model_is(0xd84))) {
+ if (_cpu == CPU_ARM &&
+ model_is_in({ CPU_MODEL_ARM_NEOVERSE_N1, CPU_MODEL_ARM_NEOVERSE_V1,
+ CPU_MODEL_ARM_NEOVERSE_N2, CPU_MODEL_ARM_NEOVERSE_V2,
+ CPU_MODEL_ARM_NEOVERSE_N3, CPU_MODEL_ARM_NEOVERSE_V3,
+ CPU_MODEL_ARM_NEOVERSE_V3AE })) {
if (FLAG_IS_DEFAULT(UseSIMDForMemoryOps)) {
FLAG_SET_DEFAULT(UseSIMDForMemoryOps, true);
}
@@ -261,12 +254,9 @@ void VM_Version::initialize() {
FLAG_SET_DEFAULT(UseCRC32, false);
}
- // Neoverse
- // V1: 0xd40
- // V2: 0xd4f
- // V3: 0xd84
if (_cpu == CPU_ARM &&
- (model_is(0xd40) || model_is(0xd4f) || model_is(0xd84))) {
+ model_is_in({ CPU_MODEL_ARM_NEOVERSE_V1, CPU_MODEL_ARM_NEOVERSE_V2,
+ CPU_MODEL_ARM_NEOVERSE_V3, CPU_MODEL_ARM_NEOVERSE_V3AE })) {
if (FLAG_IS_DEFAULT(UseCryptoPmullForCRC32)) {
FLAG_SET_DEFAULT(UseCryptoPmullForCRC32, true);
}
@@ -632,6 +622,22 @@ void VM_Version::initialize() {
check_virtualizations();
+#ifdef __APPLE__
+ DefaultWXWriteMode = UseOldWX ? WXWrite : WXArmedForWrite;
+
+ if (TraceWXHealing) {
+ if (pthread_jit_write_protect_supported_np()) {
+ tty->print_cr("### TraceWXHealing is in use");
+ if (StressWXHealing) {
+ tty->print_cr("### StressWXHealing is in use");
+ }
+ } else {
+ tty->print_cr("WX Healing is not in use because MAP_JIT write protection "
+ "does not work on this system.");
+ }
+ }
+#endif
+
// Sync SVE related CPU features with flags
if (UseSVE < 2) {
clear_feature(CPU_SVE2);
diff --git a/src/hotspot/cpu/aarch64/vm_version_aarch64.hpp b/src/hotspot/cpu/aarch64/vm_version_aarch64.hpp
index 17087d243d3..38b112d9936 100644
--- a/src/hotspot/cpu/aarch64/vm_version_aarch64.hpp
+++ b/src/hotspot/cpu/aarch64/vm_version_aarch64.hpp
@@ -30,6 +30,8 @@
#include "runtime/abstract_vm_version.hpp"
#include "utilities/sizes.hpp"
+#include
+
class stringStream;
#define BIT_MASK(flag) (1ULL<<(flag))
@@ -112,14 +114,26 @@ public:
CPU_APPLE = 'a',
};
-enum Ampere_CPU_Model {
+ enum Ampere_CPU_Model {
CPU_MODEL_EMAG = 0x0, /* CPU implementer is CPU_AMCC */
CPU_MODEL_ALTRA = 0xd0c, /* CPU implementer is CPU_ARM, Neoverse N1 */
CPU_MODEL_ALTRAMAX = 0xd0c, /* CPU implementer is CPU_ARM, Neoverse N1 */
CPU_MODEL_AMPERE_1 = 0xac3, /* CPU implementer is CPU_AMPERE */
CPU_MODEL_AMPERE_1A = 0xac4, /* CPU implementer is CPU_AMPERE */
CPU_MODEL_AMPERE_1B = 0xac5 /* AMPERE_1B core Implements ARMv8.7 with CSSC, MTE, SM3/SM4 extensions */
-};
+ };
+
+ enum ARM_CPU_Model {
+ CPU_MODEL_ARM_CORTEX_A53 = 0xd03,
+ CPU_MODEL_ARM_CORTEX_A73 = 0xd09,
+ CPU_MODEL_ARM_NEOVERSE_N1 = 0xd0c,
+ CPU_MODEL_ARM_NEOVERSE_V1 = 0xd40,
+ CPU_MODEL_ARM_NEOVERSE_N2 = 0xd49,
+ CPU_MODEL_ARM_NEOVERSE_V2 = 0xd4f,
+ CPU_MODEL_ARM_NEOVERSE_V3AE = 0xd83,
+ CPU_MODEL_ARM_NEOVERSE_V3 = 0xd84,
+ CPU_MODEL_ARM_NEOVERSE_N3 = 0xd8e,
+ };
#define CPU_FEATURE_FLAGS(decl) \
decl(FP, fp, 0) \
@@ -181,6 +195,15 @@ enum Ampere_CPU_Model {
return _model == cpu_model || _model2 == cpu_model;
}
+ static bool model_is_in(std::initializer_list cpu_models) {
+ for (const int& cpu_model : cpu_models) {
+ if (_model == cpu_model || _model2 == cpu_model) {
+ return true;
+ }
+ }
+ return false;
+ }
+
static bool is_zva_enabled() { return 0 <= _zva_length; }
static int zva_length() {
assert(is_zva_enabled(), "ZVA not available");
diff --git a/src/hotspot/cpu/arm/gc/shared/cardTableBarrierSetAssembler_arm.cpp b/src/hotspot/cpu/arm/gc/shared/cardTableBarrierSetAssembler_arm.cpp
index 2427d46cafa..5d63035ac69 100644
--- a/src/hotspot/cpu/arm/gc/shared/cardTableBarrierSetAssembler_arm.cpp
+++ b/src/hotspot/cpu/arm/gc/shared/cardTableBarrierSetAssembler_arm.cpp
@@ -67,9 +67,7 @@ void CardTableBarrierSetAssembler::store_at(MacroAssembler* masm, DecoratorSet d
void CardTableBarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators,
Register addr, Register count, Register tmp) {
BLOCK_COMMENT("CardTablePostBarrier");
- BarrierSet* bs = BarrierSet::barrier_set();
- CardTableBarrierSet* ctbs = barrier_set_cast(bs);
- CardTable* ct = ctbs->card_table();
+ CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
Label L_cardtable_loop, L_done;
@@ -83,7 +81,7 @@ void CardTableBarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembl
__ sub(count, count, addr); // nb of cards
// warning: Rthread has not been preserved
- __ mov_address(tmp, (address) ct->byte_map_base());
+ __ mov_address(tmp, (address)ctbs->card_table_base_const());
__ add(addr,tmp, addr);
Register zero = __ zero_register(tmp);
@@ -122,8 +120,7 @@ void CardTableBarrierSetAssembler::store_check_part1(MacroAssembler* masm, Regis
assert(bs->kind() == BarrierSet::CardTableBarrierSet,
"Wrong barrier set kind");
- CardTableBarrierSet* ctbs = barrier_set_cast(bs);
- CardTable* ct = ctbs->card_table();
+ CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
// Load card table base address.
@@ -140,7 +137,7 @@ void CardTableBarrierSetAssembler::store_check_part1(MacroAssembler* masm, Regis
Possible cause is a cache miss (card table base address resides in a
rarely accessed area of thread descriptor).
*/
- __ mov_address(card_table_base, (address)ct->byte_map_base());
+ __ mov_address(card_table_base, (address)ctbs->card_table_base_const());
}
// The 2nd part of the store check.
@@ -170,8 +167,8 @@ void CardTableBarrierSetAssembler::store_check_part2(MacroAssembler* masm, Regis
void CardTableBarrierSetAssembler::set_card(MacroAssembler* masm, Register card_table_base, Address card_table_addr, Register tmp) {
CardTableBarrierSet* ctbs = barrier_set_cast(BarrierSet::barrier_set());
- CardTable* ct = ctbs->card_table();
- if ((((uintptr_t)ct->byte_map_base() & 0xff) == 0)) {
+
+ if ((((uintptr_t)ctbs->card_table_base_const() & 0xff) == 0)) {
// Card table is aligned so the lowest byte of the table address base is zero.
// This works only if the code is not saved for later use, possibly
// in a context where the base would no longer be aligned.
diff --git a/src/hotspot/cpu/arm/globals_arm.hpp b/src/hotspot/cpu/arm/globals_arm.hpp
index 363a9a2c25c..c568ea04122 100644
--- a/src/hotspot/cpu/arm/globals_arm.hpp
+++ b/src/hotspot/cpu/arm/globals_arm.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2008, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2008, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -37,7 +37,7 @@ define_pd_global(bool, TrapBasedNullChecks, false); // Not needed
define_pd_global(bool, DelayCompilerStubsGeneration, false); // No need - only few compiler's stubs
define_pd_global(size_t, CodeCacheSegmentSize, 64);
-define_pd_global(intx, CodeEntryAlignment, 16);
+define_pd_global(uint, CodeEntryAlignment, 16);
define_pd_global(intx, OptoLoopAlignment, 16);
#define DEFAULT_STACK_YELLOW_PAGES (2)
diff --git a/src/hotspot/cpu/ppc/assembler_ppc.hpp b/src/hotspot/cpu/ppc/assembler_ppc.hpp
index 15e38411482..23775a3a52e 100644
--- a/src/hotspot/cpu/ppc/assembler_ppc.hpp
+++ b/src/hotspot/cpu/ppc/assembler_ppc.hpp
@@ -568,6 +568,9 @@ class Assembler : public AbstractAssembler {
XSCVDPHP_OPCODE= (60u << OPCODE_SHIFT | 347u << 2 | 17u << 16), // XX2-FORM
XXPERM_OPCODE = (60u << OPCODE_SHIFT | 26u << 3),
XXSEL_OPCODE = (60u << OPCODE_SHIFT | 3u << 4),
+ XSCMPEQDP_OPCODE=(60u << OPCODE_SHIFT | 3u << 3),
+ XSCMPGEDP_OPCODE=(60u << OPCODE_SHIFT | 19u << 3),
+ XSCMPGTDP_OPCODE=(60u << OPCODE_SHIFT | 11u << 3),
XXSPLTIB_OPCODE= (60u << OPCODE_SHIFT | 360u << 1),
XVDIVDP_OPCODE = (60u << OPCODE_SHIFT | 120u << 3),
XVABSSP_OPCODE = (60u << OPCODE_SHIFT | 409u << 2),
@@ -2424,6 +2427,9 @@ class Assembler : public AbstractAssembler {
inline void xscvdphp( VectorSRegister d, VectorSRegister b);
inline void xxland( VectorSRegister d, VectorSRegister a, VectorSRegister b);
inline void xxsel( VectorSRegister d, VectorSRegister a, VectorSRegister b, VectorSRegister c);
+ inline void xscmpeqdp(VectorSRegister t, VectorSRegister a, VectorSRegister b); // Requires Power9
+ inline void xscmpgedp(VectorSRegister t, VectorSRegister a, VectorSRegister b); // Requires Power9
+ inline void xscmpgtdp(VectorSRegister t, VectorSRegister a, VectorSRegister b); // Requires Power9
inline void xxspltib( VectorSRegister d, int ui8);
inline void xvdivsp( VectorSRegister d, VectorSRegister a, VectorSRegister b);
inline void xvdivdp( VectorSRegister d, VectorSRegister a, VectorSRegister b);
diff --git a/src/hotspot/cpu/ppc/assembler_ppc.inline.hpp b/src/hotspot/cpu/ppc/assembler_ppc.inline.hpp
index 7e49ec7455d..4cda782067e 100644
--- a/src/hotspot/cpu/ppc/assembler_ppc.inline.hpp
+++ b/src/hotspot/cpu/ppc/assembler_ppc.inline.hpp
@@ -923,6 +923,10 @@ inline void Assembler::xxmrghw( VectorSRegister d, VectorSRegister a, VectorSReg
inline void Assembler::xxmrglw( VectorSRegister d, VectorSRegister a, VectorSRegister b) { emit_int32( XXMRGHW_OPCODE | vsrt(d) | vsra(a) | vsrb(b)); }
inline void Assembler::xxsel( VectorSRegister d, VectorSRegister a, VectorSRegister b, VectorSRegister c) { emit_int32( XXSEL_OPCODE | vsrt(d) | vsra(a) | vsrb(b) | vsrc(c)); }
+inline void Assembler::xscmpeqdp(VectorSRegister t, VectorSRegister a, VectorSRegister b) { emit_int32( XSCMPEQDP_OPCODE | vsrt(t) | vsra(a) | vsrb(b) );}
+inline void Assembler::xscmpgedp(VectorSRegister t, VectorSRegister a, VectorSRegister b) { emit_int32( XSCMPGEDP_OPCODE | vsrt(t) | vsra(a) | vsrb(b) );}
+inline void Assembler::xscmpgtdp(VectorSRegister t, VectorSRegister a, VectorSRegister b) { emit_int32( XSCMPGTDP_OPCODE | vsrt(t) | vsra(a) | vsrb(b) );}
+
// VSX Extended Mnemonics
inline void Assembler::xxspltd( VectorSRegister d, VectorSRegister a, int x) { xxpermdi(d, a, a, x ? 3 : 0); }
inline void Assembler::xxmrghd( VectorSRegister d, VectorSRegister a, VectorSRegister b) { xxpermdi(d, a, b, 0); }
diff --git a/src/hotspot/cpu/ppc/c2_MacroAssembler_ppc.cpp b/src/hotspot/cpu/ppc/c2_MacroAssembler_ppc.cpp
index edf348fdc50..73b6b132895 100644
--- a/src/hotspot/cpu/ppc/c2_MacroAssembler_ppc.cpp
+++ b/src/hotspot/cpu/ppc/c2_MacroAssembler_ppc.cpp
@@ -664,3 +664,37 @@ void C2_MacroAssembler::reduceI(int opcode, Register dst, Register iSrc, VectorR
fn_scalar_op(opcode, dst, iSrc, R0); // dst <- op(iSrc, R0)
}
+// Works for single and double precision floats.
+// dst = (op1 cmp(cc) op2) ? src1 : src2;
+// Unordered semantics are the same as for CmpF3Node/CmpD3Node which implement the fcmpl/dcmpl bytecodes.
+// Comparing unordered values has the same result as when src1 is less than src2.
+// So dst = src1 for <, <=, != and dst = src2 for >, >=, ==.
+void C2_MacroAssembler::cmovF(int cc, VectorSRegister dst, VectorSRegister op1, VectorSRegister op2,
+ VectorSRegister src1, VectorSRegister src2, VectorSRegister tmp) {
+ // See operand cmpOp() for details.
+ bool invert_cond = (cc & 8) == 0; // invert reflects bcondCRbiIs0
+ auto cmp = (Assembler::Condition)(cc & 3);
+
+ switch(cmp) {
+ case Assembler::Condition::equal:
+ // Use false_result if "unordered".
+ xscmpeqdp(tmp, op1, op2);
+ break;
+ case Assembler::Condition::greater:
+ // Use false_result if "unordered".
+ xscmpgtdp(tmp, op1, op2);
+ break;
+ case Assembler::Condition::less:
+ // Use true_result if "unordered".
+ xscmpgedp(tmp, op1, op2);
+ invert_cond = !invert_cond;
+ break;
+ default:
+ assert(false, "unsupported compare condition: %d", cc);
+ ShouldNotReachHere();
+ }
+
+ VectorSRegister true_result = invert_cond ? src2 : src1;
+ VectorSRegister false_result = invert_cond ? src1 : src2;
+ xxsel(dst, false_result, true_result, tmp);
+}
diff --git a/src/hotspot/cpu/ppc/c2_MacroAssembler_ppc.hpp b/src/hotspot/cpu/ppc/c2_MacroAssembler_ppc.hpp
index 5a114294c1f..e0dffec8396 100644
--- a/src/hotspot/cpu/ppc/c2_MacroAssembler_ppc.hpp
+++ b/src/hotspot/cpu/ppc/c2_MacroAssembler_ppc.hpp
@@ -74,5 +74,7 @@
void count_positives(Register src, Register cnt, Register result, Register tmp1, Register tmp2);
void reduceI(int opcode, Register dst, Register iSrc, VectorRegister vSrc, VectorRegister vTmp1, VectorRegister vTmp2);
+ void cmovF(int cc, VectorSRegister dst, VectorSRegister op1, VectorSRegister op2,
+ VectorSRegister src1, VectorSRegister src2, VectorSRegister tmp);
#endif // CPU_PPC_C2_MACROASSEMBLER_PPC_HPP
diff --git a/src/hotspot/cpu/ppc/gc/shared/cardTableBarrierSetAssembler_ppc.cpp b/src/hotspot/cpu/ppc/gc/shared/cardTableBarrierSetAssembler_ppc.cpp
index 7404f7e2e5c..297ce57a394 100644
--- a/src/hotspot/cpu/ppc/gc/shared/cardTableBarrierSetAssembler_ppc.cpp
+++ b/src/hotspot/cpu/ppc/gc/shared/cardTableBarrierSetAssembler_ppc.cpp
@@ -103,8 +103,7 @@ void CardTableBarrierSetAssembler::resolve_jobject(MacroAssembler* masm, Registe
void CardTableBarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators, Register addr,
Register count, Register preserve) {
- CardTableBarrierSet* ctbs = barrier_set_cast(BarrierSet::barrier_set());
- CardTable* ct = ctbs->card_table();
+ CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
assert_different_registers(addr, count, R0);
Label Lskip_loop, Lstore_loop;
@@ -117,7 +116,7 @@ void CardTableBarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembl
__ srdi(addr, addr, CardTable::card_shift());
__ srdi(count, count, CardTable::card_shift());
__ subf(count, addr, count);
- __ add_const_optimized(addr, addr, (address)ct->byte_map_base(), R0);
+ __ add_const_optimized(addr, addr, (address)ctbs->card_table_base_const(), R0);
__ addi(count, count, 1);
__ li(R0, 0);
__ mtctr(count);
@@ -140,8 +139,8 @@ void CardTableBarrierSetAssembler::card_table_write(MacroAssembler* masm,
}
void CardTableBarrierSetAssembler::card_write_barrier_post(MacroAssembler* masm, Register store_addr, Register tmp) {
- CardTableBarrierSet* bs = barrier_set_cast(BarrierSet::barrier_set());
- card_table_write(masm, bs->card_table()->byte_map_base(), tmp, store_addr);
+ CardTableBarrierSet* bs = CardTableBarrierSet::barrier_set();
+ card_table_write(masm, bs->card_table_base_const(), tmp, store_addr);
}
void CardTableBarrierSetAssembler::oop_store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
diff --git a/src/hotspot/cpu/ppc/gc/shenandoah/shenandoahBarrierSetAssembler_ppc.cpp b/src/hotspot/cpu/ppc/gc/shenandoah/shenandoahBarrierSetAssembler_ppc.cpp
index 1f1bc7622ed..c3bb1811031 100644
--- a/src/hotspot/cpu/ppc/gc/shenandoah/shenandoahBarrierSetAssembler_ppc.cpp
+++ b/src/hotspot/cpu/ppc/gc/shenandoah/shenandoahBarrierSetAssembler_ppc.cpp
@@ -50,14 +50,14 @@
#define __ masm->
-void ShenandoahBarrierSetAssembler::satb_write_barrier(MacroAssembler *masm,
- Register base, RegisterOrConstant ind_or_offs,
- Register tmp1, Register tmp2, Register tmp3,
- MacroAssembler::PreservationLevel preservation_level) {
+void ShenandoahBarrierSetAssembler::satb_barrier(MacroAssembler *masm,
+ Register base, RegisterOrConstant ind_or_offs,
+ Register tmp1, Register tmp2, Register tmp3,
+ MacroAssembler::PreservationLevel preservation_level) {
if (ShenandoahSATBBarrier) {
- __ block_comment("satb_write_barrier (shenandoahgc) {");
- satb_write_barrier_impl(masm, 0, base, ind_or_offs, tmp1, tmp2, tmp3, preservation_level);
- __ block_comment("} satb_write_barrier (shenandoahgc)");
+ __ block_comment("satb_barrier (shenandoahgc) {");
+ satb_barrier_impl(masm, 0, base, ind_or_offs, tmp1, tmp2, tmp3, preservation_level);
+ __ block_comment("} satb_barrier (shenandoahgc)");
}
}
@@ -198,11 +198,12 @@ void ShenandoahBarrierSetAssembler::arraycopy_epilogue(MacroAssembler* masm, Dec
// In "load mode", this register acts as a temporary register and must
// thus not be 'noreg'. In "preloaded mode", its content will be sustained.
// tmp1/tmp2: Temporary registers, one of which must be non-volatile in "preloaded mode".
-void ShenandoahBarrierSetAssembler::satb_write_barrier_impl(MacroAssembler *masm, DecoratorSet decorators,
- Register base, RegisterOrConstant ind_or_offs,
- Register pre_val,
- Register tmp1, Register tmp2,
- MacroAssembler::PreservationLevel preservation_level) {
+void ShenandoahBarrierSetAssembler::satb_barrier_impl(MacroAssembler *masm, DecoratorSet decorators,
+ Register base, RegisterOrConstant ind_or_offs,
+ Register pre_val,
+ Register tmp1, Register tmp2,
+ MacroAssembler::PreservationLevel preservation_level) {
+ assert(ShenandoahSATBBarrier, "Should be checked by caller");
assert_different_registers(tmp1, tmp2, pre_val, noreg);
Label skip_barrier;
@@ -574,13 +575,13 @@ void ShenandoahBarrierSetAssembler::load_at(
if (ShenandoahBarrierSet::need_keep_alive_barrier(decorators, type)) {
if (ShenandoahSATBBarrier) {
__ block_comment("keep_alive_barrier (shenandoahgc) {");
- satb_write_barrier_impl(masm, 0, noreg, noreg, dst, tmp1, tmp2, preservation_level);
+ satb_barrier_impl(masm, 0, noreg, noreg, dst, tmp1, tmp2, preservation_level);
__ block_comment("} keep_alive_barrier (shenandoahgc)");
}
}
}
-void ShenandoahBarrierSetAssembler::store_check(MacroAssembler* masm, Register base, RegisterOrConstant ind_or_offs, Register tmp) {
+void ShenandoahBarrierSetAssembler::card_barrier(MacroAssembler* masm, Register base, RegisterOrConstant ind_or_offs, Register tmp) {
assert(ShenandoahCardBarrier, "Should have been checked by caller");
assert_different_registers(base, tmp, R0);
@@ -603,21 +604,33 @@ void ShenandoahBarrierSetAssembler::store_at(MacroAssembler *masm, DecoratorSet
Register base, RegisterOrConstant ind_or_offs, Register val,
Register tmp1, Register tmp2, Register tmp3,
MacroAssembler::PreservationLevel preservation_level) {
- if (is_reference_type(type)) {
- if (ShenandoahSATBBarrier) {
- satb_write_barrier(masm, base, ind_or_offs, tmp1, tmp2, tmp3, preservation_level);
- }
+ // 1: non-reference types require no barriers
+ if (!is_reference_type(type)) {
+ BarrierSetAssembler::store_at(masm, decorators, type,
+ base, ind_or_offs,
+ val,
+ tmp1, tmp2, tmp3,
+ preservation_level);
+ return;
}
+ bool storing_non_null = (val != noreg);
+
+ // 2: pre-barrier: SATB needs the previous value
+ if (ShenandoahBarrierSet::need_satb_barrier(decorators, type)) {
+ satb_barrier(masm, base, ind_or_offs, tmp1, tmp2, tmp3, preservation_level);
+ }
+
+ // Store!
BarrierSetAssembler::store_at(masm, decorators, type,
base, ind_or_offs,
val,
tmp1, tmp2, tmp3,
preservation_level);
- // No need for post barrier if storing null
- if (ShenandoahCardBarrier && is_reference_type(type) && val != noreg) {
- store_check(masm, base, ind_or_offs, tmp1);
+ // 3: post-barrier: card barrier needs store address
+ if (ShenandoahBarrierSet::need_card_barrier(decorators, type) && storing_non_null) {
+ card_barrier(masm, base, ind_or_offs, tmp1);
}
}
@@ -771,9 +784,6 @@ void ShenandoahBarrierSetAssembler::cmpxchg_oop(MacroAssembler *masm, Register b
void ShenandoahBarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators,
Register addr, Register count, Register preserve) {
assert(ShenandoahCardBarrier, "Should have been checked by caller");
-
- ShenandoahBarrierSet* bs = ShenandoahBarrierSet::barrier_set();
- CardTable* ct = bs->card_table();
assert_different_registers(addr, count, R0);
Label L_skip_loop, L_store_loop;
diff --git a/src/hotspot/cpu/ppc/gc/shenandoah/shenandoahBarrierSetAssembler_ppc.hpp b/src/hotspot/cpu/ppc/gc/shenandoah/shenandoahBarrierSetAssembler_ppc.hpp
index b058dcf1a2e..52615a740af 100644
--- a/src/hotspot/cpu/ppc/gc/shenandoah/shenandoahBarrierSetAssembler_ppc.hpp
+++ b/src/hotspot/cpu/ppc/gc/shenandoah/shenandoahBarrierSetAssembler_ppc.hpp
@@ -45,15 +45,15 @@ class ShenandoahBarrierSetAssembler: public BarrierSetAssembler {
private:
/* ==== Actual barrier implementations ==== */
- void satb_write_barrier_impl(MacroAssembler* masm, DecoratorSet decorators,
- Register base, RegisterOrConstant ind_or_offs,
- Register pre_val,
- Register tmp1, Register tmp2,
- MacroAssembler::PreservationLevel preservation_level);
+ void satb_barrier_impl(MacroAssembler* masm, DecoratorSet decorators,
+ Register base, RegisterOrConstant ind_or_offs,
+ Register pre_val,
+ Register tmp1, Register tmp2,
+ MacroAssembler::PreservationLevel preservation_level);
- void store_check(MacroAssembler* masm,
- Register base, RegisterOrConstant ind_or_offs,
- Register tmp);
+ void card_barrier(MacroAssembler* masm,
+ Register base, RegisterOrConstant ind_or_offs,
+ Register tmp);
void load_reference_barrier_impl(MacroAssembler* masm, DecoratorSet decorators,
Register base, RegisterOrConstant ind_or_offs,
@@ -85,10 +85,10 @@ public:
#endif
/* ==== Available barriers (facades of the actual implementations) ==== */
- void satb_write_barrier(MacroAssembler* masm,
- Register base, RegisterOrConstant ind_or_offs,
- Register tmp1, Register tmp2, Register tmp3,
- MacroAssembler::PreservationLevel preservation_level);
+ void satb_barrier(MacroAssembler* masm,
+ Register base, RegisterOrConstant ind_or_offs,
+ Register tmp1, Register tmp2, Register tmp3,
+ MacroAssembler::PreservationLevel preservation_level);
void load_reference_barrier(MacroAssembler* masm, DecoratorSet decorators,
Register base, RegisterOrConstant ind_or_offs,
diff --git a/src/hotspot/cpu/ppc/globals_ppc.hpp b/src/hotspot/cpu/ppc/globals_ppc.hpp
index 41a8e821ada..927a8cc2be3 100644
--- a/src/hotspot/cpu/ppc/globals_ppc.hpp
+++ b/src/hotspot/cpu/ppc/globals_ppc.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2002, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2002, 2026, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2012, 2024 SAP SE. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
@@ -60,7 +60,7 @@ define_pd_global(bool, VMContinuations, true);
// Use large code-entry alignment.
define_pd_global(size_t, CodeCacheSegmentSize, 128);
-define_pd_global(intx, CodeEntryAlignment, 64);
+define_pd_global(uint, CodeEntryAlignment, 64);
define_pd_global(intx, OptoLoopAlignment, 16);
define_pd_global(intx, InlineSmallCode, 1500);
diff --git a/src/hotspot/cpu/ppc/matcher_ppc.hpp b/src/hotspot/cpu/ppc/matcher_ppc.hpp
index aad41fb7b1c..b50de6323de 100644
--- a/src/hotspot/cpu/ppc/matcher_ppc.hpp
+++ b/src/hotspot/cpu/ppc/matcher_ppc.hpp
@@ -64,12 +64,10 @@
return true;
}
- // Use conditional move (CMOVL) on Power7.
static constexpr int long_cmove_cost() { return 0; } // this only makes long cmoves more expensive than int cmoves
- // Suppress CMOVF. Conditional move available (sort of) on PPC64 only from P7 onwards. Not exploited yet.
- // fsel doesn't accept a condition register as input, so this would be slightly different.
- static int float_cmove_cost() { return ConditionalMoveLimit; }
+ // Suppress CMOVF for Power8 because there are no fast nodes.
+ static int float_cmove_cost() { return (PowerArchitecturePPC64 >= 9) ? 0 : ConditionalMoveLimit; }
// This affects two different things:
// - how Decode nodes are matched
diff --git a/src/hotspot/cpu/ppc/ppc.ad b/src/hotspot/cpu/ppc/ppc.ad
index 2a0a9149bb3..d926fabd353 100644
--- a/src/hotspot/cpu/ppc/ppc.ad
+++ b/src/hotspot/cpu/ppc/ppc.ad
@@ -3024,7 +3024,6 @@ encode %{
%}
enc_class postalloc_expand_encode_oop(iRegNdst dst, iRegPdst src, flagsReg crx) %{
- // use isel instruction with Power 7
cmpP_reg_imm16Node *n_compare = new cmpP_reg_imm16Node();
encodeP_subNode *n_sub_base = new encodeP_subNode();
encodeP_shiftNode *n_shift = new encodeP_shiftNode();
@@ -3099,7 +3098,6 @@ encode %{
n_shift->_opnds[1] = op_src;
n_shift->_bottom_type = _bottom_type;
- // use isel instruction with Power 7
decodeN_addNode *n_add_base = new decodeN_addNode();
n_add_base->add_req(n_region, n_shift);
n_add_base->_opnds[0] = op_dst;
@@ -6618,7 +6616,6 @@ instruct cond_sub_base(iRegNdst dst, flagsRegSrc crx, iRegPsrc src1) %{
ins_pipe(pipe_class_default);
%}
-// Power 7 can use isel instruction
instruct cond_set_0_oop(iRegNdst dst, flagsRegSrc crx, iRegPsrc src1) %{
// The match rule is needed to make it a 'MachTypeNode'!
match(Set dst (EncodeP (Binary crx src1)));
@@ -7293,7 +7290,6 @@ instruct cmovF_reg(cmpOp cmp, flagsRegSrc crx, regF dst, regF src) %{
ins_variable_size_depending_on_alignment(true);
format %{ "CMOVEF $cmp, $crx, $dst, $src\n\t" %}
- // Worst case is branch + move + stop, no stop without scheduler.
size(8);
ins_encode %{
Label done;
@@ -7313,7 +7309,6 @@ instruct cmovD_reg(cmpOp cmp, flagsRegSrc crx, regD dst, regD src) %{
ins_variable_size_depending_on_alignment(true);
format %{ "CMOVEF $cmp, $crx, $dst, $src\n\t" %}
- // Worst case is branch + move + stop, no stop without scheduler.
size(8);
ins_encode %{
Label done;
@@ -7326,6 +7321,70 @@ instruct cmovD_reg(cmpOp cmp, flagsRegSrc crx, regD dst, regD src) %{
ins_pipe(pipe_class_default);
%}
+instruct cmovF_cmpF(cmpOp cop, regF op1, regF op2, regF dst, regF false_result, regF true_result, regD tmp) %{
+ match(Set dst (CMoveF (Binary cop (CmpF op1 op2)) (Binary false_result true_result)));
+ predicate(PowerArchitecturePPC64 >= 9);
+ effect(TEMP tmp);
+ ins_cost(2*DEFAULT_COST);
+ format %{ "cmovF_cmpF $dst = ($op1 $cop $op2) ? $true_result : $false_result\n\t" %}
+ size(8);
+ ins_encode %{
+ __ cmovF($cop$$cmpcode, $dst$$FloatRegister->to_vsr(),
+ $op1$$FloatRegister->to_vsr(), $op2$$FloatRegister->to_vsr(),
+ $true_result$$FloatRegister->to_vsr(), $false_result$$FloatRegister->to_vsr(),
+ $tmp$$FloatRegister->to_vsr());
+ %}
+ ins_pipe(pipe_class_default);
+%}
+
+instruct cmovF_cmpD(cmpOp cop, regD op1, regD op2, regF dst, regF false_result, regF true_result, regD tmp) %{
+ match(Set dst (CMoveF (Binary cop (CmpD op1 op2)) (Binary false_result true_result)));
+ predicate(PowerArchitecturePPC64 >= 9);
+ effect(TEMP tmp);
+ ins_cost(2*DEFAULT_COST);
+ format %{ "cmovF_cmpD $dst = ($op1 $cop $op2) ? $true_result : $false_result\n\t" %}
+ size(8);
+ ins_encode %{
+ __ cmovF($cop$$cmpcode, $dst$$FloatRegister->to_vsr(),
+ $op1$$FloatRegister->to_vsr(), $op2$$FloatRegister->to_vsr(),
+ $true_result$$FloatRegister->to_vsr(), $false_result$$FloatRegister->to_vsr(),
+ $tmp$$FloatRegister->to_vsr());
+ %}
+ ins_pipe(pipe_class_default);
+%}
+
+instruct cmovD_cmpD(cmpOp cop, regD op1, regD op2, regD dst, regD false_result, regD true_result, regD tmp) %{
+ match(Set dst (CMoveD (Binary cop (CmpD op1 op2)) (Binary false_result true_result)));
+ predicate(PowerArchitecturePPC64 >= 9);
+ effect(TEMP tmp);
+ ins_cost(2*DEFAULT_COST);
+ format %{ "cmovD_cmpD $dst = ($op1 $cop $op2) ? $true_result : $false_result\n\t" %}
+ size(8);
+ ins_encode %{
+ __ cmovF($cop$$cmpcode, $dst$$FloatRegister->to_vsr(),
+ $op1$$FloatRegister->to_vsr(), $op2$$FloatRegister->to_vsr(),
+ $true_result$$FloatRegister->to_vsr(), $false_result$$FloatRegister->to_vsr(),
+ $tmp$$FloatRegister->to_vsr());
+ %}
+ ins_pipe(pipe_class_default);
+%}
+
+instruct cmovD_cmpF(cmpOp cop, regF op1, regF op2, regD dst, regD false_result, regD true_result, regD tmp) %{
+ match(Set dst (CMoveD (Binary cop (CmpF op1 op2)) (Binary false_result true_result)));
+ predicate(PowerArchitecturePPC64 >= 9);
+ effect(TEMP tmp);
+ ins_cost(2*DEFAULT_COST);
+ format %{ "cmovD_cmpF $dst = ($op1 $cop $op2) ? $true_result : $false_result\n\t" %}
+ size(8);
+ ins_encode %{
+ __ cmovF($cop$$cmpcode, $dst$$FloatRegister->to_vsr(),
+ $op1$$FloatRegister->to_vsr(), $op2$$FloatRegister->to_vsr(),
+ $true_result$$FloatRegister->to_vsr(), $false_result$$FloatRegister->to_vsr(),
+ $tmp$$FloatRegister->to_vsr());
+ %}
+ ins_pipe(pipe_class_default);
+%}
+
//----------Compare-And-Swap---------------------------------------------------
// CompareAndSwap{P,I,L} have more than one output, therefore "CmpI
@@ -8492,7 +8551,6 @@ instruct cmovI_bne_negI_reg(iRegIdst dst, flagsRegSrc crx, iRegIsrc src1) %{
ins_variable_size_depending_on_alignment(true);
format %{ "CMOVE $dst, neg($src1), $crx" %}
- // Worst case is branch + move + stop, no stop without scheduler.
size(8);
ins_encode %{
Label done;
@@ -8551,7 +8609,6 @@ instruct cmovL_bne_negL_reg(iRegLdst dst, flagsRegSrc crx, iRegLsrc src1) %{
ins_variable_size_depending_on_alignment(true);
format %{ "CMOVE $dst, neg($src1), $crx" %}
- // Worst case is branch + move + stop, no stop without scheduler.
size(8);
ins_encode %{
Label done;
@@ -10262,7 +10319,6 @@ instruct cmovI_bso_stackSlotL(iRegIdst dst, flagsRegSrc crx, stackSlotL src) %{
ins_variable_size_depending_on_alignment(true);
format %{ "cmovI $crx, $dst, $src" %}
- // Worst case is branch + move + stop, no stop without scheduler.
size(8);
ins_encode( enc_cmove_bso_stackSlotL(dst, crx, src) );
ins_pipe(pipe_class_default);
@@ -10276,7 +10332,6 @@ instruct cmovI_bso_reg(iRegIdst dst, flagsRegSrc crx, regD src) %{
ins_variable_size_depending_on_alignment(true);
format %{ "cmovI $crx, $dst, $src" %}
- // Worst case is branch + move + stop, no stop without scheduler.
size(8);
ins_encode( enc_cmove_bso_reg(dst, crx, src) );
ins_pipe(pipe_class_default);
@@ -10439,7 +10494,6 @@ instruct cmovL_bso_stackSlotL(iRegLdst dst, flagsRegSrc crx, stackSlotL src) %{
ins_variable_size_depending_on_alignment(true);
format %{ "cmovL $crx, $dst, $src" %}
- // Worst case is branch + move + stop, no stop without scheduler.
size(8);
ins_encode( enc_cmove_bso_stackSlotL(dst, crx, src) );
ins_pipe(pipe_class_default);
@@ -10453,7 +10507,6 @@ instruct cmovL_bso_reg(iRegLdst dst, flagsRegSrc crx, regD src) %{
ins_variable_size_depending_on_alignment(true);
format %{ "cmovL $crx, $dst, $src" %}
- // Worst case is branch + move + stop, no stop without scheduler.
size(8);
ins_encode( enc_cmove_bso_reg(dst, crx, src) );
ins_pipe(pipe_class_default);
@@ -11080,7 +11133,6 @@ instruct cmov_bns_less(flagsReg crx) %{
ins_variable_size_depending_on_alignment(true);
format %{ "cmov $crx" %}
- // Worst case is branch + move + stop, no stop without scheduler.
size(12);
ins_encode %{
Label done;
diff --git a/src/hotspot/cpu/ppc/sharedRuntime_ppc.cpp b/src/hotspot/cpu/ppc/sharedRuntime_ppc.cpp
index 4eb2028f529..5260ed978ff 100644
--- a/src/hotspot/cpu/ppc/sharedRuntime_ppc.cpp
+++ b/src/hotspot/cpu/ppc/sharedRuntime_ppc.cpp
@@ -1,6 +1,6 @@
/*
* Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, 2025 SAP SE. All rights reserved.
+ * Copyright (c) 2012, 2026 SAP SE. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -775,7 +775,6 @@ int SharedRuntime::java_calling_convention(const BasicType *sig_bt,
return stk;
}
-#if defined(COMPILER1) || defined(COMPILER2)
// Calling convention for calling C code.
int SharedRuntime::c_calling_convention(const BasicType *sig_bt,
VMRegPair *regs,
@@ -913,7 +912,6 @@ int SharedRuntime::c_calling_convention(const BasicType *sig_bt,
return MAX2(arg, 8) * 2 + additional_frame_header_slots;
#endif
}
-#endif // COMPILER2
int SharedRuntime::vector_calling_convention(VMRegPair *regs,
uint num_bits,
@@ -2874,7 +2872,6 @@ void SharedRuntime::generate_deopt_blob() {
CodeBuffer buffer(name, 2048, 1024);
InterpreterMacroAssembler* masm = new InterpreterMacroAssembler(&buffer);
Label exec_mode_initialized;
- int frame_size_in_words;
OopMap* map = nullptr;
OopMapSet *oop_maps = new OopMapSet();
@@ -2886,6 +2883,9 @@ void SharedRuntime::generate_deopt_blob() {
const Register exec_mode_reg = R21_tmp1;
const address start = __ pc();
+ int exception_offset = 0;
+ int exception_in_tls_offset = 0;
+ int reexecute_offset = 0;
#if defined(COMPILER1) || defined(COMPILER2)
// --------------------------------------------------------------------------
@@ -2925,7 +2925,7 @@ void SharedRuntime::generate_deopt_blob() {
// - R3_ARG1: exception oop
// - R4_ARG2: exception pc
- int exception_offset = __ pc() - start;
+ exception_offset = __ pc() - start;
BLOCK_COMMENT("Prolog for exception case");
@@ -2936,7 +2936,7 @@ void SharedRuntime::generate_deopt_blob() {
__ std(R4_ARG2, _abi0(lr), R1_SP);
// Vanilla deoptimization with an exception pending in exception_oop.
- int exception_in_tls_offset = __ pc() - start;
+ exception_in_tls_offset = __ pc() - start;
// Push the "unpack frame".
// Save everything in sight.
@@ -2949,8 +2949,6 @@ void SharedRuntime::generate_deopt_blob() {
__ li(exec_mode_reg, Deoptimization::Unpack_exception);
// fall through
-
- int reexecute_offset = 0;
#ifdef COMPILER1
__ b(exec_mode_initialized);
@@ -3068,11 +3066,12 @@ void SharedRuntime::generate_deopt_blob() {
// Return to the interpreter entry point.
__ blr();
- __ flush();
-#else // COMPILER2
+#else // !defined(COMPILER1) && !defined(COMPILER2)
__ unimplemented("deopt blob needed only with compiler");
- int exception_offset = __ pc() - start;
-#endif // COMPILER2
+#endif
+
+ // Make sure all code is generated
+ __ flush();
_deopt_blob = DeoptimizationBlob::create(&buffer, oop_maps, 0, exception_offset,
reexecute_offset, first_frame_size_in_bytes / wordSize);
diff --git a/src/hotspot/cpu/riscv/c1_LIRAssembler_riscv.cpp b/src/hotspot/cpu/riscv/c1_LIRAssembler_riscv.cpp
index e77a2067e89..63e2fd015d7 100644
--- a/src/hotspot/cpu/riscv/c1_LIRAssembler_riscv.cpp
+++ b/src/hotspot/cpu/riscv/c1_LIRAssembler_riscv.cpp
@@ -1041,31 +1041,10 @@ void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
__ bind(*op->stub()->continuation());
}
-void LIR_Assembler::type_profile_helper(Register mdo, ciMethodData *md, ciProfileData *data,
- Register recv, Label* update_done) {
- for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
- Label next_test;
- // See if the receiver is receiver[n].
- __ ld(t1, Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i))));
- __ bne(recv, t1, next_test);
- Address data_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i)));
- __ increment(data_addr, DataLayout::counter_increment);
- __ j(*update_done);
- __ bind(next_test);
- }
-
- // Didn't find receiver; find next empty slot and fill it in
- for (uint i = 0; i < ReceiverTypeData::row_limit(); i++) {
- Label next_test;
- Address recv_addr(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_offset(i)));
- __ ld(t1, recv_addr);
- __ bnez(t1, next_test);
- __ sd(recv, recv_addr);
- __ mv(t1, DataLayout::counter_increment);
- __ sd(t1, Address(mdo, md->byte_offset_of_slot(data, ReceiverTypeData::receiver_count_offset(i))));
- __ j(*update_done);
- __ bind(next_test);
- }
+void LIR_Assembler::type_profile_helper(Register mdo, ciMethodData *md,
+ ciProfileData *data, Register recv) {
+ int mdp_offset = md->byte_offset_of_slot(data, in_ByteSize(0));
+ __ profile_receiver_type(recv, mdo, mdp_offset);
}
void LIR_Assembler::data_check(LIR_OpTypeCheck *op, ciMethodData **md, ciProfileData **data) {
@@ -1139,14 +1118,9 @@ void LIR_Assembler::profile_object(ciMethodData* md, ciProfileData* data, Regist
__ j(*obj_is_null);
__ bind(not_null);
- Label update_done;
Register recv = k_RInfo;
__ load_klass(recv, obj);
- type_profile_helper(mdo, md, data, recv, &update_done);
- Address counter_addr(mdo, md->byte_offset_of_slot(data, CounterData::count_offset()));
- __ increment(counter_addr, DataLayout::counter_increment);
-
- __ bind(update_done);
+ type_profile_helper(mdo, md, data, recv);
}
void LIR_Assembler::typecheck_loaded(LIR_OpTypeCheck *op, ciKlass* k, Register k_RInfo) {
@@ -1554,11 +1528,8 @@ void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
// We know the type that will be seen at this call site; we can
// statically update the MethodData* rather than needing to do
// dynamic tests on the receiver type
- // NOTE: we should probably put a lock around this search to
- // avoid collisions by concurrent compilations
ciVirtualCallData* vc_data = (ciVirtualCallData*) data;
- uint i;
- for (i = 0; i < VirtualCallData::row_limit(); i++) {
+ for (uint i = 0; i < VirtualCallData::row_limit(); i++) {
ciKlass* receiver = vc_data->receiver(i);
if (known_klass->equals(receiver)) {
Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
@@ -1566,32 +1537,13 @@ void LIR_Assembler::emit_profile_call(LIR_OpProfileCall* op) {
return;
}
}
-
- // Receiver type not found in profile data; select an empty slot
- // Note that this is less efficient than it should be because it
- // always does a write to the receiver part of the
- // VirtualCallData rather than just the first time
- for (i = 0; i < VirtualCallData::row_limit(); i++) {
- ciKlass* receiver = vc_data->receiver(i);
- if (receiver == nullptr) {
- Address recv_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_offset(i)));
- __ mov_metadata(t1, known_klass->constant_encoding());
- __ sd(t1, recv_addr);
- Address data_addr(mdo, md->byte_offset_of_slot(data, VirtualCallData::receiver_count_offset(i)));
- __ increment(data_addr, DataLayout::counter_increment);
- return;
- }
- }
+ // Receiver type is not found in profile data.
+ // Fall back to runtime helper to handle the rest at runtime.
+ __ mov_metadata(recv, known_klass->constant_encoding());
} else {
__ load_klass(recv, recv);
- Label update_done;
- type_profile_helper(mdo, md, data, recv, &update_done);
- // Receiver did not match any saved receiver and there is no empty row for it.
- // Increment total counter to indicate polymorphic case.
- __ increment(counter_addr, DataLayout::counter_increment);
-
- __ bind(update_done);
}
+ type_profile_helper(mdo, md, data, recv);
} else {
// Static call
__ increment(counter_addr, DataLayout::counter_increment);
diff --git a/src/hotspot/cpu/riscv/c1_LIRAssembler_riscv.hpp b/src/hotspot/cpu/riscv/c1_LIRAssembler_riscv.hpp
index 1e466e90d37..90b6b3ee4f4 100644
--- a/src/hotspot/cpu/riscv/c1_LIRAssembler_riscv.hpp
+++ b/src/hotspot/cpu/riscv/c1_LIRAssembler_riscv.hpp
@@ -54,9 +54,8 @@ private:
Address stack_slot_address(int index, uint shift, int adjust = 0);
// Record the type of the receiver in ReceiverTypeData
- void type_profile_helper(Register mdo,
- ciMethodData *md, ciProfileData *data,
- Register recv, Label* update_done);
+ void type_profile_helper(Register mdo, ciMethodData *md,
+ ciProfileData *data, Register recv);
void casw(Register addr, Register newval, Register cmpval);
void caswu(Register addr, Register newval, Register cmpval);
diff --git a/src/hotspot/cpu/riscv/c2_MacroAssembler_riscv.cpp b/src/hotspot/cpu/riscv/c2_MacroAssembler_riscv.cpp
index 824ea872935..b4e0ba69042 100644
--- a/src/hotspot/cpu/riscv/c2_MacroAssembler_riscv.cpp
+++ b/src/hotspot/cpu/riscv/c2_MacroAssembler_riscv.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2020, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2020, 2026, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2020, 2022, Huawei Technologies Co., Ltd. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
@@ -2813,10 +2813,14 @@ void C2_MacroAssembler::char_array_compress_v(Register src, Register dst, Regist
// Intrinsic for
//
-// - sun/nio/cs/ISO_8859_1$Encoder.implEncodeISOArray
-// return the number of characters copied.
-// - java/lang/StringUTF16.compress
-// return index of non-latin1 character if copy fails, otherwise 'len'.
+// - sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
+// Encodes char[] to byte[] in ISO-8859-1
+//
+// - java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
+// Encodes byte[] (containing UTF-16) to byte[] in ISO-8859-1
+//
+// - java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
+// Encodes char[] to byte[] in ASCII
//
// This version always returns the number of characters copied. A successful
// copy will complete with the post-condition: 'res' == 'len', while an
diff --git a/src/hotspot/cpu/riscv/gc/shenandoah/shenandoahBarrierSetAssembler_riscv.cpp b/src/hotspot/cpu/riscv/gc/shenandoah/shenandoahBarrierSetAssembler_riscv.cpp
index dd6c8556307..3cbbb783258 100644
--- a/src/hotspot/cpu/riscv/gc/shenandoah/shenandoahBarrierSetAssembler_riscv.cpp
+++ b/src/hotspot/cpu/riscv/gc/shenandoah/shenandoahBarrierSetAssembler_riscv.cpp
@@ -88,26 +88,16 @@ void ShenandoahBarrierSetAssembler::arraycopy_epilogue(MacroAssembler* masm, Dec
}
}
-void ShenandoahBarrierSetAssembler::shenandoah_write_barrier_pre(MacroAssembler* masm,
- Register obj,
- Register pre_val,
- Register thread,
- Register tmp,
- bool tosca_live,
- bool expand_call) {
- if (ShenandoahSATBBarrier) {
- satb_write_barrier_pre(masm, obj, pre_val, thread, tmp, t0, tosca_live, expand_call);
- }
-}
+void ShenandoahBarrierSetAssembler::satb_barrier(MacroAssembler* masm,
+ Register obj,
+ Register pre_val,
+ Register thread,
+ Register tmp1,
+ Register tmp2,
+ bool tosca_live,
+ bool expand_call) {
+ assert(ShenandoahSATBBarrier, "Should be checked by caller");
-void ShenandoahBarrierSetAssembler::satb_write_barrier_pre(MacroAssembler* masm,
- Register obj,
- Register pre_val,
- Register thread,
- Register tmp1,
- Register tmp2,
- bool tosca_live,
- bool expand_call) {
// If expand_call is true then we expand the call_VM_leaf macro
// directly to skip generating the check by
// InterpreterMacroAssembler::call_VM_leaf_base that checks _last_sp.
@@ -376,21 +366,21 @@ void ShenandoahBarrierSetAssembler::load_at(MacroAssembler* masm,
if (ShenandoahBarrierSet::need_keep_alive_barrier(decorators, type)) {
__ enter();
__ push_call_clobbered_registers();
- satb_write_barrier_pre(masm /* masm */,
- noreg /* obj */,
- dst /* pre_val */,
- xthread /* thread */,
- tmp1 /* tmp1 */,
- tmp2 /* tmp2 */,
- true /* tosca_live */,
- true /* expand_call */);
+ satb_barrier(masm /* masm */,
+ noreg /* obj */,
+ dst /* pre_val */,
+ xthread /* thread */,
+ tmp1 /* tmp1 */,
+ tmp2 /* tmp2 */,
+ true /* tosca_live */,
+ true /* expand_call */);
__ pop_call_clobbered_registers();
__ leave();
}
}
-void ShenandoahBarrierSetAssembler::store_check(MacroAssembler* masm, Register obj) {
- assert(ShenandoahCardBarrier, "Did you mean to enable ShenandoahCardBarrier?");
+void ShenandoahBarrierSetAssembler::card_barrier(MacroAssembler* masm, Register obj) {
+ assert(ShenandoahCardBarrier, "Should have been checked by caller");
__ srli(obj, obj, CardTable::card_shift());
@@ -413,13 +403,13 @@ void ShenandoahBarrierSetAssembler::store_check(MacroAssembler* masm, Register o
void ShenandoahBarrierSetAssembler::store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
Address dst, Register val, Register tmp1, Register tmp2, Register tmp3) {
- bool on_oop = is_reference_type(type);
- if (!on_oop) {
+ // 1: non-reference types require no barriers
+ if (!is_reference_type(type)) {
BarrierSetAssembler::store_at(masm, decorators, type, dst, val, tmp1, tmp2, tmp3);
return;
}
- // flatten object address if needed
+ // Flatten object address right away for simplicity: likely needed by barriers
if (dst.offset() == 0) {
if (dst.base() != tmp3) {
__ mv(tmp3, dst.base());
@@ -428,20 +418,26 @@ void ShenandoahBarrierSetAssembler::store_at(MacroAssembler* masm, DecoratorSet
__ la(tmp3, dst);
}
- shenandoah_write_barrier_pre(masm,
- tmp3 /* obj */,
- tmp2 /* pre_val */,
- xthread /* thread */,
- tmp1 /* tmp */,
- val != noreg /* tosca_live */,
- false /* expand_call */);
+ bool storing_non_null = (val != noreg);
+ // 2: pre-barrier: SATB needs the previous value
+ if (ShenandoahBarrierSet::need_satb_barrier(decorators, type)) {
+ satb_barrier(masm,
+ tmp3 /* obj */,
+ tmp2 /* pre_val */,
+ xthread /* thread */,
+ tmp1 /* tmp */,
+ t0 /* tmp2 */,
+ storing_non_null /* tosca_live */,
+ false /* expand_call */);
+ }
+
+ // Store!
BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp3, 0), val, noreg, noreg, noreg);
- bool in_heap = (decorators & IN_HEAP) != 0;
- bool needs_post_barrier = (val != noreg) && in_heap && ShenandoahCardBarrier;
- if (needs_post_barrier) {
- store_check(masm, tmp3);
+ // 3: post-barrier: card barrier needs store address
+ if (ShenandoahBarrierSet::need_card_barrier(decorators, type) && storing_non_null) {
+ card_barrier(masm, tmp3);
}
}
diff --git a/src/hotspot/cpu/riscv/gc/shenandoah/shenandoahBarrierSetAssembler_riscv.hpp b/src/hotspot/cpu/riscv/gc/shenandoah/shenandoahBarrierSetAssembler_riscv.hpp
index c8a7c35fb83..5085be26b2e 100644
--- a/src/hotspot/cpu/riscv/gc/shenandoah/shenandoahBarrierSetAssembler_riscv.hpp
+++ b/src/hotspot/cpu/riscv/gc/shenandoah/shenandoahBarrierSetAssembler_riscv.hpp
@@ -41,23 +41,16 @@ class StubCodeGenerator;
class ShenandoahBarrierSetAssembler: public BarrierSetAssembler {
private:
- void satb_write_barrier_pre(MacroAssembler* masm,
- Register obj,
- Register pre_val,
- Register thread,
- Register tmp1,
- Register tmp2,
- bool tosca_live,
- bool expand_call);
- void shenandoah_write_barrier_pre(MacroAssembler* masm,
- Register obj,
- Register pre_val,
- Register thread,
- Register tmp,
- bool tosca_live,
- bool expand_call);
+ void satb_barrier(MacroAssembler* masm,
+ Register obj,
+ Register pre_val,
+ Register thread,
+ Register tmp1,
+ Register tmp2,
+ bool tosca_live,
+ bool expand_call);
- void store_check(MacroAssembler* masm, Register obj);
+ void card_barrier(MacroAssembler* masm, Register obj);
void resolve_forward_pointer(MacroAssembler* masm, Register dst, Register tmp = noreg);
void resolve_forward_pointer_not_null(MacroAssembler* masm, Register dst, Register tmp = noreg);
diff --git a/src/hotspot/cpu/riscv/globals_riscv.hpp b/src/hotspot/cpu/riscv/globals_riscv.hpp
index 390ed2daeb9..21b119266e2 100644
--- a/src/hotspot/cpu/riscv/globals_riscv.hpp
+++ b/src/hotspot/cpu/riscv/globals_riscv.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2000, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2000, 2026, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2020, 2023, Huawei Technologies Co., Ltd. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
@@ -39,7 +39,7 @@ define_pd_global(bool, UncommonNullCast, true); // Uncommon-trap nulls
define_pd_global(bool, DelayCompilerStubsGeneration, COMPILER2_OR_JVMCI);
define_pd_global(size_t, CodeCacheSegmentSize, 64 COMPILER1_AND_COMPILER2_PRESENT(+64)); // Tiered compilation has large code-entry alignment.
-define_pd_global(intx, CodeEntryAlignment, 64);
+define_pd_global(uint, CodeEntryAlignment, 64);
define_pd_global(intx, OptoLoopAlignment, 16);
#define DEFAULT_STACK_YELLOW_PAGES (2)
diff --git a/src/hotspot/cpu/riscv/icache_riscv.cpp b/src/hotspot/cpu/riscv/icache_riscv.cpp
index 258bc665770..20de2dbb2ad 100644
--- a/src/hotspot/cpu/riscv/icache_riscv.cpp
+++ b/src/hotspot/cpu/riscv/icache_riscv.cpp
@@ -39,7 +39,8 @@ static int icache_flush(address addr, int lines, int magic) {
// We need to make sure stores happens before the I/D cache synchronization.
__asm__ volatile("fence rw, rw" : : : "memory");
- RiscvFlushIcache::flush((uintptr_t)addr, ((uintptr_t)lines) << ICache::log2_line_size);
+ uintptr_t end = (uintptr_t)addr + ((uintptr_t)lines << ICache::log2_line_size);
+ RiscvFlushIcache::flush((uintptr_t)addr, end);
return magic;
}
diff --git a/src/hotspot/cpu/riscv/interp_masm_riscv.cpp b/src/hotspot/cpu/riscv/interp_masm_riscv.cpp
index 189c7c93d07..744590bec2b 100644
--- a/src/hotspot/cpu/riscv/interp_masm_riscv.cpp
+++ b/src/hotspot/cpu/riscv/interp_masm_riscv.cpp
@@ -237,15 +237,14 @@ void InterpreterMacroAssembler::load_resolved_klass_at_offset(
// Rsub_klass: subklass
//
// Kills:
-// x12, x15
+// x12
void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass,
Label& ok_is_subtype) {
assert(Rsub_klass != x10, "x10 holds superklass");
assert(Rsub_klass != x12, "x12 holds 2ndary super array length");
- assert(Rsub_klass != x15, "x15 holds 2ndary super array scan ptr");
// Profile the not-null value's klass.
- profile_typecheck(x12, Rsub_klass, x15); // blows x12, reloads x15
+ profile_typecheck(x12, Rsub_klass); // blows x12
// Do the check.
check_klass_subtype(Rsub_klass, x10, x12, ok_is_subtype); // blows x12
@@ -1042,7 +1041,6 @@ void InterpreterMacroAssembler::profile_final_call(Register mdp) {
void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
Register mdp,
- Register reg2,
bool receiver_can_be_null) {
if (ProfileInterpreter) {
Label profile_continue;
@@ -1060,7 +1058,7 @@ void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
}
// Record the receiver type.
- record_klass_in_profile(receiver, mdp, reg2);
+ profile_receiver_type(receiver, mdp, 0);
bind(skip_receiver_profile);
// The method data pointer needs to be updated to reflect the new target.
@@ -1072,153 +1070,6 @@ void InterpreterMacroAssembler::profile_virtual_call(Register receiver,
}
}
-// This routine creates a state machine for updating the multi-row
-// type profile at a virtual call site (or other type-sensitive bytecode).
-// The machine visits each row (of receiver/count) until the receiver type
-// is found, or until it runs out of rows. At the same time, it remembers
-// the location of the first empty row. (An empty row records null for its
-// receiver, and can be allocated for a newly-observed receiver type.)
-// Because there are two degrees of freedom in the state, a simple linear
-// search will not work; it must be a decision tree. Hence this helper
-// function is recursive, to generate the required tree structured code.
-// It's the interpreter, so we are trading off code space for speed.
-// See below for example code.
-void InterpreterMacroAssembler::record_klass_in_profile_helper(
- Register receiver, Register mdp,
- Register reg2, Label& done) {
- if (TypeProfileWidth == 0) {
- increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
- } else {
- record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth,
- &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset);
- }
-}
-
-void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp,
- Register reg2, int start_row, Label& done, int total_rows,
- OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn) {
- int last_row = total_rows - 1;
- assert(start_row <= last_row, "must be work left to do");
- // Test this row for both the item and for null.
- // Take any of three different outcomes:
- // 1. found item => increment count and goto done
- // 2. found null => keep looking for case 1, maybe allocate this cell
- // 3. found something else => keep looking for cases 1 and 2
- // Case 3 is handled by a recursive call.
- for (int row = start_row; row <= last_row; row++) {
- Label next_test;
- bool test_for_null_also = (row == start_row);
-
- // See if the item is item[n].
- int item_offset = in_bytes(item_offset_fn(row));
- test_mdp_data_at(mdp, item_offset, item,
- (test_for_null_also ? reg2 : noreg),
- next_test);
- // (Reg2 now contains the item from the CallData.)
-
- // The item is item[n]. Increment count[n].
- int count_offset = in_bytes(item_count_offset_fn(row));
- increment_mdp_data_at(mdp, count_offset);
- j(done);
- bind(next_test);
-
- if (test_for_null_also) {
- Label found_null;
- // Failed the equality check on item[n]... Test for null.
- if (start_row == last_row) {
- // The only thing left to do is handle the null case.
- beqz(reg2, found_null);
- // Item did not match any saved item and there is no empty row for it.
- // Increment total counter to indicate polymorphic case.
- increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
- j(done);
- bind(found_null);
- break;
- }
- // Since null is rare, make it be the branch-taken case.
- beqz(reg2, found_null);
-
- // Put all the "Case 3" tests here.
- record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows,
- item_offset_fn, item_count_offset_fn);
-
- // Found a null. Keep searching for a matching item,
- // but remember that this is an empty (unused) slot.
- bind(found_null);
- }
- }
-
- // In the fall-through case, we found no matching item, but we
- // observed the item[start_row] is null.
- // Fill in the item field and increment the count.
- int item_offset = in_bytes(item_offset_fn(start_row));
- set_mdp_data_at(mdp, item_offset, item);
- int count_offset = in_bytes(item_count_offset_fn(start_row));
- mv(reg2, DataLayout::counter_increment);
- set_mdp_data_at(mdp, count_offset, reg2);
- if (start_row > 0) {
- j(done);
- }
-}
-
-// Example state machine code for three profile rows:
-// # main copy of decision tree, rooted at row[1]
-// if (row[0].rec == rec) then [
-// row[0].incr()
-// goto done
-// ]
-// if (row[0].rec != nullptr) then [
-// # inner copy of decision tree, rooted at row[1]
-// if (row[1].rec == rec) then [
-// row[1].incr()
-// goto done
-// ]
-// if (row[1].rec != nullptr) then [
-// # degenerate decision tree, rooted at row[2]
-// if (row[2].rec == rec) then [
-// row[2].incr()
-// goto done
-// ]
-// if (row[2].rec != nullptr) then [
-// count.incr()
-// goto done
-// ] # overflow
-// row[2].init(rec)
-// goto done
-// ] else [
-// # remember row[1] is empty
-// if (row[2].rec == rec) then [
-// row[2].incr()
-// goto done
-// ]
-// row[1].init(rec)
-// goto done
-// ]
-// else [
-// # remember row[0] is empty
-// if (row[1].rec == rec) then [
-// row[1].incr()
-// goto done
-// ]
-// if (row[2].rec == rec) then [
-// row[2].incr()
-// goto done
-// ]
-// row[0].init(rec)
-// goto done
-// ]
-// done:
-
-void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
- Register mdp, Register reg2) {
- assert(ProfileInterpreter, "must be profiling");
- Label done;
-
- record_klass_in_profile_helper(receiver, mdp, reg2, done);
-
- bind(done);
-}
-
void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) {
if (ProfileInterpreter) {
Label profile_continue;
@@ -1274,7 +1125,7 @@ void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
}
}
-void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) {
+void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass) {
if (ProfileInterpreter) {
Label profile_continue;
@@ -1287,7 +1138,7 @@ void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass,
mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
// Record the object type.
- record_klass_in_profile(klass, mdp, reg2);
+ profile_receiver_type(klass, mdp, 0);
}
update_mdp_by_constant(mdp, mdp_delta);
diff --git a/src/hotspot/cpu/riscv/interp_masm_riscv.hpp b/src/hotspot/cpu/riscv/interp_masm_riscv.hpp
index a9df09d656a..59cc76b022f 100644
--- a/src/hotspot/cpu/riscv/interp_masm_riscv.hpp
+++ b/src/hotspot/cpu/riscv/interp_masm_riscv.hpp
@@ -262,14 +262,6 @@ class InterpreterMacroAssembler: public MacroAssembler {
Register test_value_out,
Label& not_equal_continue);
- void record_klass_in_profile(Register receiver, Register mdp,
- Register reg2);
- void record_klass_in_profile_helper(Register receiver, Register mdp,
- Register reg2, Label& done);
- void record_item_in_profile_helper(Register item, Register mdp,
- Register reg2, int start_row, Label& done, int total_rows,
- OffsetFunction item_offset_fn, OffsetFunction item_count_offset_fn);
-
void update_mdp_by_offset(Register mdp_in, int offset_of_offset);
void update_mdp_by_offset(Register mdp_in, Register reg, int offset_of_disp);
void update_mdp_by_constant(Register mdp_in, int constant);
@@ -283,11 +275,10 @@ class InterpreterMacroAssembler: public MacroAssembler {
void profile_call(Register mdp);
void profile_final_call(Register mdp);
void profile_virtual_call(Register receiver, Register mdp,
- Register t1,
bool receiver_can_be_null = false);
void profile_ret(Register return_bci, Register mdp);
void profile_null_seen(Register mdp);
- void profile_typecheck(Register mdp, Register klass, Register temp);
+ void profile_typecheck(Register mdp, Register klass);
void profile_typecheck_failed(Register mdp);
void profile_switch_default(Register mdp);
void profile_switch_case(Register index_in_scratch, Register mdp,
diff --git a/src/hotspot/cpu/riscv/macroAssembler_riscv.cpp b/src/hotspot/cpu/riscv/macroAssembler_riscv.cpp
index 43b17a13c20..4f5e7afc166 100644
--- a/src/hotspot/cpu/riscv/macroAssembler_riscv.cpp
+++ b/src/hotspot/cpu/riscv/macroAssembler_riscv.cpp
@@ -543,6 +543,160 @@ void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file,
BLOCK_COMMENT("} verify_oop");
}
+// Handle the receiver type profile update given the "recv" klass.
+//
+// Normally updates the ReceiverData (RD) that starts at "mdp" + "mdp_offset".
+// If there are no matching or claimable receiver entries in RD, updates
+// the polymorphic counter.
+//
+// This code expected to run by either the interpreter or JIT-ed code, without
+// extra synchronization. For safety, receiver cells are claimed atomically, which
+// avoids grossly misrepresenting the profiles under concurrent updates. For speed,
+// counter updates are not atomic.
+//
+void MacroAssembler::profile_receiver_type(Register recv, Register mdp, int mdp_offset) {
+ assert_different_registers(recv, mdp, t0, t1);
+
+ int base_receiver_offset = in_bytes(ReceiverTypeData::receiver_offset(0));
+ int end_receiver_offset = in_bytes(ReceiverTypeData::receiver_offset(ReceiverTypeData::row_limit()));
+ int poly_count_offset = in_bytes(CounterData::count_offset());
+ int receiver_step = in_bytes(ReceiverTypeData::receiver_offset(1)) - base_receiver_offset;
+ int receiver_to_count_step = in_bytes(ReceiverTypeData::receiver_count_offset(0)) - base_receiver_offset;
+
+ // Adjust for MDP offsets. Slots are pointer-sized, so is the global offset.
+ base_receiver_offset += mdp_offset;
+ end_receiver_offset += mdp_offset;
+ poly_count_offset += mdp_offset;
+
+#ifdef ASSERT
+ // We are about to walk the MDO slots without asking for offsets.
+ // Check that our math hits all the right spots.
+ for (uint c = 0; c < ReceiverTypeData::row_limit(); c++) {
+ int real_recv_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_offset(c));
+ int real_count_offset = mdp_offset + in_bytes(ReceiverTypeData::receiver_count_offset(c));
+ int offset = base_receiver_offset + receiver_step*c;
+ int count_offset = offset + receiver_to_count_step;
+ assert(offset == real_recv_offset, "receiver slot math");
+ assert(count_offset == real_count_offset, "receiver count math");
+ }
+ int real_poly_count_offset = mdp_offset + in_bytes(CounterData::count_offset());
+ assert(poly_count_offset == real_poly_count_offset, "poly counter math");
+#endif
+
+ // Corner case: no profile table. Increment poly counter and exit.
+ if (ReceiverTypeData::row_limit() == 0) {
+ increment(Address(mdp, poly_count_offset), DataLayout::counter_increment);
+ return;
+ }
+
+ Register offset = t1;
+
+ Label L_loop_search_receiver, L_loop_search_empty;
+ Label L_restart, L_found_recv, L_found_empty, L_polymorphic, L_count_update;
+
+ // The code here recognizes three major cases:
+ // A. Fastest: receiver found in the table
+ // B. Fast: no receiver in the table, and the table is full
+ // C. Slow: no receiver in the table, free slots in the table
+ //
+ // The case A performance is most important, as perfectly-behaved code would end up
+ // there, especially with larger TypeProfileWidth. The case B performance is
+ // important as well, this is where bulk of code would land for normally megamorphic
+ // cases. The case C performance is not essential, its job is to deal with installation
+ // races, we optimize for code density instead. Case C needs to make sure that receiver
+ // rows are only claimed once. This makes sure we never overwrite a row for another
+ // receiver and never duplicate the receivers in the list, making profile type-accurate.
+ //
+ // It is very tempting to handle these cases in a single loop, and claim the first slot
+ // without checking the rest of the table. But, profiling code should tolerate free slots
+ // in the table, as class unloading can clear them. After such cleanup, the receiver
+ // we need might be _after_ the free slot. Therefore, we need to let at least full scan
+ // to complete, before trying to install new slots. Splitting the code in several tight
+ // loops also helpfully optimizes for cases A and B.
+ //
+ // This code is effectively:
+ //
+ // restart:
+ // // Fastest: receiver is already installed
+ // for (i = 0; i < receiver_count(); i++) {
+ // if (receiver(i) == recv) goto found_recv(i);
+ // }
+ //
+ // // Fast: no receiver, but profile is full
+ // for (i = 0; i < receiver_count(); i++) {
+ // if (receiver(i) == null) goto found_null(i);
+ // }
+ // goto polymorphic
+ //
+ // // Slow: try to install receiver
+ // found_null(i):
+ // CAS(&receiver(i), null, recv);
+ // goto restart
+ //
+ // polymorphic:
+ // count++;
+ // return
+ //
+ // found_recv(i):
+ // *receiver_count(i)++
+ //
+
+ bind(L_restart);
+
+ // Fastest: receiver is already installed
+ mv(offset, base_receiver_offset);
+ bind(L_loop_search_receiver);
+ add(t0, mdp, offset);
+ ld(t0, Address(t0));
+ beq(recv, t0, L_found_recv);
+ add(offset, offset, receiver_step);
+ sub(t0, offset, end_receiver_offset);
+ bnez(t0, L_loop_search_receiver);
+
+ // Fast: no receiver, but profile is full
+ mv(offset, base_receiver_offset);
+ bind(L_loop_search_empty);
+ add(t0, mdp, offset);
+ ld(t0, Address(t0));
+ beqz(t0, L_found_empty);
+ add(offset, offset, receiver_step);
+ sub(t0, offset, end_receiver_offset);
+ bnez(t0, L_loop_search_empty);
+ j(L_polymorphic);
+
+ // Slow: try to install receiver
+ bind(L_found_empty);
+
+ // Atomically swing receiver slot: null -> recv.
+ //
+ // The update uses CAS, which clobbers t0. Therefore, t1
+ // is used to hold the destination address. This is safe because the
+ // offset is no longer needed after the address is computed.
+ add(t1, mdp, offset);
+ weak_cmpxchg(/*addr*/ t1, /*expected*/ zr, /*new*/ recv, Assembler::int64,
+ /*acquire*/ Assembler::relaxed, /*release*/ Assembler::relaxed, /*result*/ t0);
+
+ // CAS success means the slot now has the receiver we want. CAS failure means
+ // something had claimed the slot concurrently: it can be the same receiver we want,
+ // or something else. Since this is a slow path, we can optimize for code density,
+ // and just restart the search from the beginning.
+ j(L_restart);
+
+ // Counter updates:
+ // Increment polymorphic counter instead of receiver slot.
+ bind(L_polymorphic);
+ mv(offset, poly_count_offset);
+ j(L_count_update);
+
+ // Found a receiver, convert its slot offset to corresponding count offset.
+ bind(L_found_recv);
+ add(offset, offset, receiver_to_count_step);
+
+ bind(L_count_update);
+ add(t1, mdp, offset);
+ increment(Address(t1), DataLayout::counter_increment);
+}
+
void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
if (!VerifyOops) {
return;
@@ -5110,9 +5264,8 @@ void MacroAssembler::get_thread(Register thread) {
}
void MacroAssembler::load_byte_map_base(Register reg) {
- CardTable::CardValue* byte_map_base =
- ((CardTableBarrierSet*)(BarrierSet::barrier_set()))->card_table()->byte_map_base();
- mv(reg, (uint64_t)byte_map_base);
+ CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
+ mv(reg, (uint64_t)ctbs->card_table_base_const());
}
void MacroAssembler::build_frame(int framesize) {
diff --git a/src/hotspot/cpu/riscv/macroAssembler_riscv.hpp b/src/hotspot/cpu/riscv/macroAssembler_riscv.hpp
index 3b021388fa5..f5e985c28a2 100644
--- a/src/hotspot/cpu/riscv/macroAssembler_riscv.hpp
+++ b/src/hotspot/cpu/riscv/macroAssembler_riscv.hpp
@@ -390,6 +390,8 @@ class MacroAssembler: public Assembler {
Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
+ void profile_receiver_type(Register recv, Register mdp, int mdp_offset);
+
// only if +VerifyOops
void _verify_oop(Register reg, const char* s, const char* file, int line);
void _verify_oop_addr(Address addr, const char* s, const char* file, int line);
diff --git a/src/hotspot/cpu/riscv/templateTable_riscv.cpp b/src/hotspot/cpu/riscv/templateTable_riscv.cpp
index 5a3644f70bb..5cc725e3af4 100644
--- a/src/hotspot/cpu/riscv/templateTable_riscv.cpp
+++ b/src/hotspot/cpu/riscv/templateTable_riscv.cpp
@@ -708,7 +708,6 @@ void TemplateTable::index_check(Register array, Register index) {
__ mv(x11, index);
}
Label ok;
- __ sext(index, index, 32);
__ bltu(index, length, ok);
__ mv(x13, array);
__ mv(t1, Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
@@ -1052,7 +1051,7 @@ void TemplateTable::aastore() {
transition(vtos, vtos);
// stack: ..., array, index, value
__ ld(x10, at_tos()); // value
- __ ld(x12, at_tos_p1()); // index
+ __ lw(x12, at_tos_p1()); // index
__ ld(x13, at_tos_p2()); // array
index_check(x13, x12); // kills x11
@@ -1462,9 +1461,9 @@ void TemplateTable::iinc() {
transition(vtos, vtos);
__ load_signed_byte(x11, at_bcp(2)); // get constant
locals_index(x12);
- __ ld(x10, iaddress(x12, x10, _masm));
+ __ lw(x10, iaddress(x12, x10, _masm));
__ addw(x10, x10, x11);
- __ sd(x10, iaddress(x12, t0, _masm));
+ __ sw(x10, iaddress(x12, t0, _masm));
}
void TemplateTable::wide_iinc() {
@@ -1477,9 +1476,9 @@ void TemplateTable::wide_iinc() {
__ orr(x11, x11, t1);
locals_index_wide(x12);
- __ ld(x10, iaddress(x12, t0, _masm));
+ __ lw(x10, iaddress(x12, t0, _masm));
__ addw(x10, x10, x11);
- __ sd(x10, iaddress(x12, t0, _masm));
+ __ sw(x10, iaddress(x12, t0, _masm));
}
void TemplateTable::convert() {
@@ -3280,7 +3279,7 @@ void TemplateTable::invokevirtual_helper(Register index,
__ load_klass(x10, recv);
// profile this call
- __ profile_virtual_call(x10, xlocals, x13);
+ __ profile_virtual_call(x10, xlocals);
// get target Method & entry point
__ lookup_virtual_method(x10, index, method);
@@ -3407,7 +3406,7 @@ void TemplateTable::invokeinterface(int byte_no) {
/*return_method=*/false);
// profile this call
- __ profile_virtual_call(x13, x30, x9);
+ __ profile_virtual_call(x13, x30);
// Get declaring interface class from method, and itable index
__ load_method_holder(x10, xmethod);
diff --git a/src/hotspot/cpu/s390/gc/shared/cardTableBarrierSetAssembler_s390.cpp b/src/hotspot/cpu/s390/gc/shared/cardTableBarrierSetAssembler_s390.cpp
index a0da6ebe682..9bb7f63ff31 100644
--- a/src/hotspot/cpu/s390/gc/shared/cardTableBarrierSetAssembler_s390.cpp
+++ b/src/hotspot/cpu/s390/gc/shared/cardTableBarrierSetAssembler_s390.cpp
@@ -83,8 +83,7 @@ void CardTableBarrierSetAssembler::resolve_jobject(MacroAssembler* masm, Registe
void CardTableBarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators, Register addr, Register count,
bool do_return) {
- CardTableBarrierSet* ctbs = barrier_set_cast(BarrierSet::barrier_set());
- CardTable* ct = ctbs->card_table();
+ CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
NearLabel doXC, done;
assert_different_registers(Z_R0, Z_R1, addr, count);
@@ -105,7 +104,7 @@ void CardTableBarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembl
__ add2reg_with_index(count, -BytesPerHeapOop, count, addr);
// Get base address of card table.
- __ load_const_optimized(Z_R1, (address)ct->byte_map_base());
+ __ load_const_optimized(Z_R1, (address)ctbs->card_table_base_const());
// count = (count>>shift) - (addr>>shift)
__ z_srlg(addr, addr, CardTable::card_shift());
@@ -179,13 +178,12 @@ void CardTableBarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembl
void CardTableBarrierSetAssembler::store_check(MacroAssembler* masm, Register store_addr, Register tmp) {
// Does a store check for the oop in register obj. The content of
// register obj is destroyed afterwards.
- CardTableBarrierSet* ctbs = barrier_set_cast(BarrierSet::barrier_set());
- CardTable* ct = ctbs->card_table();
+ CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
assert_different_registers(store_addr, tmp);
__ z_srlg(store_addr, store_addr, CardTable::card_shift());
- __ load_absolute_address(tmp, (address)ct->byte_map_base());
+ __ load_absolute_address(tmp, (address)ctbs->card_table_base_const());
__ z_agr(store_addr, tmp);
__ z_mvi(0, store_addr, CardTable::dirty_card_val());
}
diff --git a/src/hotspot/cpu/s390/globals_s390.hpp b/src/hotspot/cpu/s390/globals_s390.hpp
index 07987ea3469..d110443adf8 100644
--- a/src/hotspot/cpu/s390/globals_s390.hpp
+++ b/src/hotspot/cpu/s390/globals_s390.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2016, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2016, 2026, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2016, 2018 SAP SE. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
@@ -43,7 +43,7 @@ define_pd_global(size_t, CodeCacheSegmentSize, 256);
// Ideally, this is 256 (cache line size). This keeps code end data
// on separate lines. But we reduced it to 64 since 256 increased
// code size significantly by padding nops between IVC and second UEP.
-define_pd_global(intx, CodeEntryAlignment, 64);
+define_pd_global(uint, CodeEntryAlignment, 64);
define_pd_global(intx, OptoLoopAlignment, 2);
define_pd_global(intx, InlineSmallCode, 2000);
diff --git a/src/hotspot/cpu/x86/assembler_x86.cpp b/src/hotspot/cpu/x86/assembler_x86.cpp
index cbc5c6988d4..3c8defe62d9 100644
--- a/src/hotspot/cpu/x86/assembler_x86.cpp
+++ b/src/hotspot/cpu/x86/assembler_x86.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -7320,6 +7320,25 @@ void Assembler::ucomisd(XMMRegister dst, XMMRegister src) {
emit_int16(0x2E, (0xC0 | encode));
}
+void Assembler::vucomxsd(XMMRegister dst, Address src) {
+ assert(VM_Version::supports_avx10_2(), "");
+ InstructionMark im(this);
+ InstructionAttr attributes(AVX_128bit, /* rex_w */ true, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
+ attributes.set_address_attributes(/* tuple_type */ EVEX_T1S, /* input_size_in_bits */ EVEX_64bit);
+ attributes.set_is_evex_instruction();
+ vex_prefix(src, 0, dst->encoding(), VEX_SIMD_F2, VEX_OPCODE_0F, &attributes);
+ emit_int8(0x2E);
+ emit_operand(dst, src, 0);
+}
+
+void Assembler::vucomxsd(XMMRegister dst, XMMRegister src) {
+ assert(VM_Version::supports_avx10_2(), "");
+ InstructionAttr attributes(AVX_128bit, /* rex_w */ true, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
+ attributes.set_is_evex_instruction();
+ int encode = vex_prefix_and_encode(dst->encoding(), 0, src->encoding(), VEX_SIMD_F2, VEX_OPCODE_0F, &attributes);
+ emit_int16(0x2E, (0xC0 | encode));
+}
+
void Assembler::ucomiss(XMMRegister dst, Address src) {
InstructionMark im(this);
InstructionAttr attributes(AVX_128bit, /* rex_w */ false, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
@@ -7335,6 +7354,25 @@ void Assembler::ucomiss(XMMRegister dst, XMMRegister src) {
emit_int16(0x2E, (0xC0 | encode));
}
+void Assembler::vucomxss(XMMRegister dst, Address src) {
+ assert(VM_Version::supports_avx10_2(), "");
+ InstructionMark im(this);
+ InstructionAttr attributes(AVX_128bit, /* rex_w */ false, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
+ attributes.set_address_attributes(/* tuple_type */ EVEX_T1S, /* input_size_in_bits */ EVEX_32bit);
+ attributes.set_is_evex_instruction();
+ vex_prefix(src, 0, dst->encoding(), VEX_SIMD_F3, VEX_OPCODE_0F, &attributes);
+ emit_int8(0x2E);
+ emit_operand(dst, src, 0);
+}
+
+void Assembler::vucomxss(XMMRegister dst, XMMRegister src) {
+ assert(VM_Version::supports_avx10_2(), "");
+ InstructionAttr attributes(AVX_128bit, /* rex_w */ false, /* legacy_mode */ false, /* no_mask_reg */ true, /* uses_vl */ false);
+ attributes.set_is_evex_instruction();
+ int encode = vex_prefix_and_encode(dst->encoding(), 0, src->encoding(), VEX_SIMD_F3, VEX_OPCODE_0F, &attributes);
+ emit_int16(0x2E, (0xC0 | encode));
+}
+
void Assembler::xabort(int8_t imm8) {
emit_int24((unsigned char)0xC6, (unsigned char)0xF8, (imm8 & 0xFF));
}
diff --git a/src/hotspot/cpu/x86/assembler_x86.hpp b/src/hotspot/cpu/x86/assembler_x86.hpp
index 26c57fc2d80..97854f712cf 100644
--- a/src/hotspot/cpu/x86/assembler_x86.hpp
+++ b/src/hotspot/cpu/x86/assembler_x86.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -2331,10 +2331,14 @@ private:
// Unordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS
void ucomisd(XMMRegister dst, Address src);
void ucomisd(XMMRegister dst, XMMRegister src);
+ void vucomxsd(XMMRegister dst, Address src);
+ void vucomxsd(XMMRegister dst, XMMRegister src);
// Unordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS
void ucomiss(XMMRegister dst, Address src);
void ucomiss(XMMRegister dst, XMMRegister src);
+ void vucomxss(XMMRegister dst, Address src);
+ void vucomxss(XMMRegister dst, XMMRegister src);
void xabort(int8_t imm8);
diff --git a/src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp b/src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp
index 8fc3d18abb1..c65b439604b 100644
--- a/src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp
+++ b/src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2020, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2020, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -1046,17 +1046,28 @@ void C2_MacroAssembler::signum_fp(int opcode, XMMRegister dst, XMMRegister zero,
Label DONE_LABEL;
+ // Handle special cases +0.0/-0.0 and NaN, if argument is +0.0/-0.0 or NaN, return argument
+ // If AVX10.2 (or newer) floating point comparison instructions used, SF=1 for equal and unordered cases
+ // If other floating point comparison instructions used, ZF=1 for equal and unordered cases
if (opcode == Op_SignumF) {
- ucomiss(dst, zero);
- jcc(Assembler::equal, DONE_LABEL); // handle special case +0.0/-0.0, if argument is +0.0/-0.0, return argument
- jcc(Assembler::parity, DONE_LABEL); // handle special case NaN, if argument NaN, return NaN
+ if (VM_Version::supports_avx10_2()) {
+ vucomxss(dst, zero);
+ jcc(Assembler::negative, DONE_LABEL);
+ } else {
+ ucomiss(dst, zero);
+ jcc(Assembler::equal, DONE_LABEL);
+ }
movflt(dst, one);
jcc(Assembler::above, DONE_LABEL);
xorps(dst, ExternalAddress(StubRoutines::x86::vector_float_sign_flip()), noreg);
} else if (opcode == Op_SignumD) {
- ucomisd(dst, zero);
- jcc(Assembler::equal, DONE_LABEL); // handle special case +0.0/-0.0, if argument is +0.0/-0.0, return argument
- jcc(Assembler::parity, DONE_LABEL); // handle special case NaN, if argument NaN, return NaN
+ if (VM_Version::supports_avx10_2()) {
+ vucomxsd(dst, zero);
+ jcc(Assembler::negative, DONE_LABEL);
+ } else {
+ ucomisd(dst, zero);
+ jcc(Assembler::equal, DONE_LABEL);
+ }
movdbl(dst, one);
jcc(Assembler::above, DONE_LABEL);
xorpd(dst, ExternalAddress(StubRoutines::x86::vector_double_sign_flip()), noreg);
diff --git a/src/hotspot/cpu/x86/gc/shared/cardTableBarrierSetAssembler_x86.cpp b/src/hotspot/cpu/x86/gc/shared/cardTableBarrierSetAssembler_x86.cpp
index 2b91662ddb5..65e6b4e01fc 100644
--- a/src/hotspot/cpu/x86/gc/shared/cardTableBarrierSetAssembler_x86.cpp
+++ b/src/hotspot/cpu/x86/gc/shared/cardTableBarrierSetAssembler_x86.cpp
@@ -95,11 +95,7 @@ void CardTableBarrierSetAssembler::store_at(MacroAssembler* masm, DecoratorSet d
void CardTableBarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators,
Register addr, Register count, Register tmp) {
- BarrierSet *bs = BarrierSet::barrier_set();
- CardTableBarrierSet* ctbs = barrier_set_cast(bs);
- CardTable* ct = ctbs->card_table();
- intptr_t disp = (intptr_t) ct->byte_map_base();
- SHENANDOAHGC_ONLY(assert(!UseShenandoahGC, "Shenandoah byte_map_base is not constant.");)
+ CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
Label L_loop, L_done;
const Register end = count;
@@ -115,7 +111,7 @@ void CardTableBarrierSetAssembler::gen_write_ref_array_post_barrier(MacroAssembl
__ shrptr(end, CardTable::card_shift());
__ subptr(end, addr); // end --> cards count
- __ mov64(tmp, disp);
+ __ mov64(tmp, (intptr_t)ctbs->card_table_base_const());
__ addptr(addr, tmp);
__ BIND(L_loop);
__ movb(Address(addr, count, Address::times_1), 0);
@@ -128,10 +124,7 @@ __ BIND(L_done);
void CardTableBarrierSetAssembler::store_check(MacroAssembler* masm, Register obj, Address dst) {
// Does a store check for the oop in register obj. The content of
// register obj is destroyed afterwards.
- BarrierSet* bs = BarrierSet::barrier_set();
-
- CardTableBarrierSet* ctbs = barrier_set_cast(bs);
- CardTable* ct = ctbs->card_table();
+ CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
__ shrptr(obj, CardTable::card_shift());
@@ -142,7 +135,7 @@ void CardTableBarrierSetAssembler::store_check(MacroAssembler* masm, Register ob
// So this essentially converts an address to a displacement and it will
// never need to be relocated. On 64bit however the value may be too
// large for a 32bit displacement.
- intptr_t byte_map_base = (intptr_t)ct->byte_map_base();
+ intptr_t byte_map_base = (intptr_t)ctbs->card_table_base_const();
if (__ is_simm32(byte_map_base)) {
card_addr = Address(noreg, obj, Address::times_1, byte_map_base);
} else {
diff --git a/src/hotspot/cpu/x86/gc/shenandoah/shenandoahBarrierSetAssembler_x86.cpp b/src/hotspot/cpu/x86/gc/shenandoah/shenandoahBarrierSetAssembler_x86.cpp
index 9e321391f6c..97829a10a3b 100644
--- a/src/hotspot/cpu/x86/gc/shenandoah/shenandoahBarrierSetAssembler_x86.cpp
+++ b/src/hotspot/cpu/x86/gc/shenandoah/shenandoahBarrierSetAssembler_x86.cpp
@@ -174,24 +174,14 @@ void ShenandoahBarrierSetAssembler::arraycopy_epilogue(MacroAssembler* masm, Dec
}
}
-void ShenandoahBarrierSetAssembler::shenandoah_write_barrier_pre(MacroAssembler* masm,
- Register obj,
- Register pre_val,
- Register tmp,
- bool tosca_live,
- bool expand_call) {
+void ShenandoahBarrierSetAssembler::satb_barrier(MacroAssembler* masm,
+ Register obj,
+ Register pre_val,
+ Register tmp,
+ bool tosca_live,
+ bool expand_call) {
+ assert(ShenandoahSATBBarrier, "Should be checked by caller");
- if (ShenandoahSATBBarrier) {
- satb_write_barrier_pre(masm, obj, pre_val, tmp, tosca_live, expand_call);
- }
-}
-
-void ShenandoahBarrierSetAssembler::satb_write_barrier_pre(MacroAssembler* masm,
- Register obj,
- Register pre_val,
- Register tmp,
- bool tosca_live,
- bool expand_call) {
// If expand_call is true then we expand the call_VM_leaf macro
// directly to skip generating the check by
// InterpreterMacroAssembler::call_VM_leaf_base that checks _last_sp.
@@ -533,18 +523,18 @@ void ShenandoahBarrierSetAssembler::load_at(MacroAssembler* masm, DecoratorSet d
assert_different_registers(dst, tmp1, r15_thread);
// Generate the SATB pre-barrier code to log the value of
// the referent field in an SATB buffer.
- shenandoah_write_barrier_pre(masm /* masm */,
- noreg /* obj */,
- dst /* pre_val */,
- tmp1 /* tmp */,
- true /* tosca_live */,
- true /* expand_call */);
+ satb_barrier(masm /* masm */,
+ noreg /* obj */,
+ dst /* pre_val */,
+ tmp1 /* tmp */,
+ true /* tosca_live */,
+ true /* expand_call */);
restore_machine_state(masm, /* handle_gpr = */ true, /* handle_fp = */ true);
}
}
-void ShenandoahBarrierSetAssembler::store_check(MacroAssembler* masm, Register obj) {
+void ShenandoahBarrierSetAssembler::card_barrier(MacroAssembler* masm, Register obj) {
assert(ShenandoahCardBarrier, "Should have been checked by caller");
// Does a store check for the oop in register obj. The content of
@@ -575,41 +565,40 @@ void ShenandoahBarrierSetAssembler::store_check(MacroAssembler* masm, Register o
void ShenandoahBarrierSetAssembler::store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,
Address dst, Register val, Register tmp1, Register tmp2, Register tmp3) {
- bool on_oop = is_reference_type(type);
- bool in_heap = (decorators & IN_HEAP) != 0;
- bool as_normal = (decorators & AS_NORMAL) != 0;
- if (on_oop && in_heap) {
- bool needs_pre_barrier = as_normal;
+ // 1: non-reference types require no barriers
+ if (!is_reference_type(type)) {
+ BarrierSetAssembler::store_at(masm, decorators, type, dst, val, tmp1, tmp2, tmp3);
+ return;
+ }
- // flatten object address if needed
- // We do it regardless of precise because we need the registers
- if (dst.index() == noreg && dst.disp() == 0) {
- if (dst.base() != tmp1) {
- __ movptr(tmp1, dst.base());
- }
- } else {
- __ lea(tmp1, dst);
- }
-
- assert_different_registers(val, tmp1, tmp2, tmp3, r15_thread);
-
- if (needs_pre_barrier) {
- shenandoah_write_barrier_pre(masm /*masm*/,
- tmp1 /* obj */,
- tmp2 /* pre_val */,
- tmp3 /* tmp */,
- val != noreg /* tosca_live */,
- false /* expand_call */);
- }
-
- BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp1, 0), val, noreg, noreg, noreg);
- if (val != noreg) {
- if (ShenandoahCardBarrier) {
- store_check(masm, tmp1);
- }
+ // Flatten object address right away for simplicity: likely needed by barriers
+ assert_different_registers(val, tmp1, tmp2, tmp3, r15_thread);
+ if (dst.index() == noreg && dst.disp() == 0) {
+ if (dst.base() != tmp1) {
+ __ movptr(tmp1, dst.base());
}
} else {
- BarrierSetAssembler::store_at(masm, decorators, type, dst, val, tmp1, tmp2, tmp3);
+ __ lea(tmp1, dst);
+ }
+
+ bool storing_non_null = (val != noreg);
+
+ // 2: pre-barrier: SATB needs the previous value
+ if (ShenandoahBarrierSet::need_satb_barrier(decorators, type)) {
+ satb_barrier(masm,
+ tmp1 /* obj */,
+ tmp2 /* pre_val */,
+ tmp3 /* tmp */,
+ storing_non_null /* tosca_live */,
+ false /* expand_call */);
+ }
+
+ // Store!
+ BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp1, 0), val, noreg, noreg, noreg);
+
+ // 3: post-barrier: card barrier needs store address
+ if (ShenandoahBarrierSet::need_card_barrier(decorators, type) && storing_non_null) {
+ card_barrier(masm, tmp1);
}
}
diff --git a/src/hotspot/cpu/x86/gc/shenandoah/shenandoahBarrierSetAssembler_x86.hpp b/src/hotspot/cpu/x86/gc/shenandoah/shenandoahBarrierSetAssembler_x86.hpp
index b0185f2dbff..b5cc5c8d834 100644
--- a/src/hotspot/cpu/x86/gc/shenandoah/shenandoahBarrierSetAssembler_x86.hpp
+++ b/src/hotspot/cpu/x86/gc/shenandoah/shenandoahBarrierSetAssembler_x86.hpp
@@ -41,21 +41,14 @@ class StubCodeGenerator;
class ShenandoahBarrierSetAssembler: public BarrierSetAssembler {
private:
- void satb_write_barrier_pre(MacroAssembler* masm,
- Register obj,
- Register pre_val,
- Register tmp,
- bool tosca_live,
- bool expand_call);
+ void satb_barrier(MacroAssembler* masm,
+ Register obj,
+ Register pre_val,
+ Register tmp,
+ bool tosca_live,
+ bool expand_call);
- void shenandoah_write_barrier_pre(MacroAssembler* masm,
- Register obj,
- Register pre_val,
- Register tmp,
- bool tosca_live,
- bool expand_call);
-
- void store_check(MacroAssembler* masm, Register obj);
+ void card_barrier(MacroAssembler* masm, Register obj);
void gen_write_ref_array_post_barrier(MacroAssembler* masm, DecoratorSet decorators,
Register addr, Register count,
diff --git a/src/hotspot/cpu/x86/globals_x86.hpp b/src/hotspot/cpu/x86/globals_x86.hpp
index 103e22d0185..4f5b6d31e75 100644
--- a/src/hotspot/cpu/x86/globals_x86.hpp
+++ b/src/hotspot/cpu/x86/globals_x86.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2000, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2000, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -46,9 +46,9 @@ define_pd_global(size_t, CodeCacheSegmentSize, 64 COMPILER1_AND_COMPILER2_PRES
// the uep and the vep doesn't get real alignment but just slops on by
// only assured that the entry instruction meets the 5 byte size requirement.
#if COMPILER2_OR_JVMCI
-define_pd_global(intx, CodeEntryAlignment, 32);
+define_pd_global(uint, CodeEntryAlignment, 32);
#else
-define_pd_global(intx, CodeEntryAlignment, 16);
+define_pd_global(uint, CodeEntryAlignment, 16);
#endif // COMPILER2_OR_JVMCI
define_pd_global(intx, OptoLoopAlignment, 16);
define_pd_global(intx, InlineSmallCode, 1000);
diff --git a/src/hotspot/cpu/x86/macroAssembler_x86.cpp b/src/hotspot/cpu/x86/macroAssembler_x86.cpp
index 7f7bb2c4c7f..83169df3456 100644
--- a/src/hotspot/cpu/x86/macroAssembler_x86.cpp
+++ b/src/hotspot/cpu/x86/macroAssembler_x86.cpp
@@ -765,7 +765,7 @@ void MacroAssembler::align32() {
void MacroAssembler::align(uint modulus) {
// 8273459: Ensure alignment is possible with current segment alignment
- assert(modulus <= (uintx)CodeEntryAlignment, "Alignment must be <= CodeEntryAlignment");
+ assert(modulus <= CodeEntryAlignment, "Alignment must be <= CodeEntryAlignment");
align(modulus, offset());
}
@@ -2656,6 +2656,17 @@ void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src, Register rscra
}
}
+void MacroAssembler::vucomxsd(XMMRegister dst, AddressLiteral src, Register rscratch) {
+ assert(rscratch != noreg || always_reachable(src), "missing");
+
+ if (reachable(src)) {
+ Assembler::vucomxsd(dst, as_Address(src));
+ } else {
+ lea(rscratch, src);
+ Assembler::vucomxsd(dst, Address(rscratch, 0));
+ }
+}
+
void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src, Register rscratch) {
assert(rscratch != noreg || always_reachable(src), "missing");
@@ -2667,6 +2678,17 @@ void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src, Register rscra
}
}
+void MacroAssembler::vucomxss(XMMRegister dst, AddressLiteral src, Register rscratch) {
+ assert(rscratch != noreg || always_reachable(src), "missing");
+
+ if (reachable(src)) {
+ Assembler::vucomxss(dst, as_Address(src));
+ } else {
+ lea(rscratch, src);
+ Assembler::vucomxss(dst, Address(rscratch, 0));
+ }
+}
+
void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src, Register rscratch) {
assert(rscratch != noreg || always_reachable(src), "missing");
@@ -6251,32 +6273,46 @@ void MacroAssembler::evpbroadcast(BasicType type, XMMRegister dst, Register src,
}
}
-// encode char[] to byte[] in ISO_8859_1 or ASCII
- //@IntrinsicCandidate
- //private static int implEncodeISOArray(byte[] sa, int sp,
- //byte[] da, int dp, int len) {
- // int i = 0;
- // for (; i < len; i++) {
- // char c = StringUTF16.getChar(sa, sp++);
- // if (c > '\u00FF')
- // break;
- // da[dp++] = (byte)c;
- // }
- // return i;
- //}
- //
- //@IntrinsicCandidate
- //private static int implEncodeAsciiArray(char[] sa, int sp,
- // byte[] da, int dp, int len) {
- // int i = 0;
- // for (; i < len; i++) {
- // char c = sa[sp++];
- // if (c >= '\u0080')
- // break;
- // da[dp++] = (byte)c;
- // }
- // return i;
- //}
+// Encode given char[]/byte[] to byte[] in ISO_8859_1 or ASCII
+//
+// @IntrinsicCandidate
+// int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(
+// char[] sa, int sp, byte[] da, int dp, int len) {
+// int i = 0;
+// for (; i < len; i++) {
+// char c = sa[sp++];
+// if (c > '\u00FF')
+// break;
+// da[dp++] = (byte) c;
+// }
+// return i;
+// }
+//
+// @IntrinsicCandidate
+// int java.lang.StringCoding.encodeISOArray0(
+// byte[] sa, int sp, byte[] da, int dp, int len) {
+// int i = 0;
+// for (; i < len; i++) {
+// char c = StringUTF16.getChar(sa, sp++);
+// if (c > '\u00FF')
+// break;
+// da[dp++] = (byte) c;
+// }
+// return i;
+// }
+//
+// @IntrinsicCandidate
+// int java.lang.StringCoding.encodeAsciiArray0(
+// char[] sa, int sp, byte[] da, int dp, int len) {
+// int i = 0;
+// for (; i < len; i++) {
+// char c = sa[sp++];
+// if (c >= '\u0080')
+// break;
+// da[dp++] = (byte) c;
+// }
+// return i;
+// }
void MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
XMMRegister tmp1Reg, XMMRegister tmp2Reg,
XMMRegister tmp3Reg, XMMRegister tmp4Reg,
diff --git a/src/hotspot/cpu/x86/macroAssembler_x86.hpp b/src/hotspot/cpu/x86/macroAssembler_x86.hpp
index 93e3529ac1e..eb23199ca63 100644
--- a/src/hotspot/cpu/x86/macroAssembler_x86.hpp
+++ b/src/hotspot/cpu/x86/macroAssembler_x86.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -1313,10 +1313,18 @@ public:
void ucomiss(XMMRegister dst, Address src) { Assembler::ucomiss(dst, src); }
void ucomiss(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
+ void vucomxss(XMMRegister dst, XMMRegister src) { Assembler::vucomxss(dst, src); }
+ void vucomxss(XMMRegister dst, Address src) { Assembler::vucomxss(dst, src); }
+ void vucomxss(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
+
void ucomisd(XMMRegister dst, XMMRegister src) { Assembler::ucomisd(dst, src); }
void ucomisd(XMMRegister dst, Address src) { Assembler::ucomisd(dst, src); }
void ucomisd(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
+ void vucomxsd(XMMRegister dst, XMMRegister src) { Assembler::vucomxsd(dst, src); }
+ void vucomxsd(XMMRegister dst, Address src) { Assembler::vucomxsd(dst, src); }
+ void vucomxsd(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
+
// Bitwise Logical XOR of Packed Double-Precision Floating-Point Values
void xorpd(XMMRegister dst, XMMRegister src);
void xorpd(XMMRegister dst, Address src) { Assembler::xorpd(dst, src); }
diff --git a/src/hotspot/cpu/x86/vm_version_x86.cpp b/src/hotspot/cpu/x86/vm_version_x86.cpp
index 747daefd51d..c65c1c7d219 100644
--- a/src/hotspot/cpu/x86/vm_version_x86.cpp
+++ b/src/hotspot/cpu/x86/vm_version_x86.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -143,7 +143,7 @@ class VM_Version_StubGenerator: public StubCodeGenerator {
Label detect_486, cpu486, detect_586, std_cpuid1, std_cpuid4, std_cpuid24, std_cpuid29;
Label sef_cpuid, sefsl1_cpuid, ext_cpuid, ext_cpuid1, ext_cpuid5, ext_cpuid7;
- Label ext_cpuid8, done, wrapup, vector_save_restore, apx_save_restore_warning;
+ Label ext_cpuid8, done, wrapup, vector_save_restore, apx_save_restore_warning, apx_xstate;
Label legacy_setup, save_restore_except, legacy_save_restore, start_simd_check;
StubCodeMark mark(this, "VM_Version", "get_cpu_info_stub");
@@ -468,6 +468,20 @@ class VM_Version_StubGenerator: public StubCodeGenerator {
__ movq(Address(rsi, 0), r16);
__ movq(Address(rsi, 8), r31);
+ //
+ // Query CPUID 0xD.19 for APX XSAVE offset
+ // Extended State Enumeration Sub-leaf 19 (APX)
+ // EAX = size of APX state (should be 128)
+ // EBX = offset in standard XSAVE format
+ //
+ __ movl(rax, 0xD);
+ __ movl(rcx, 19);
+ __ cpuid();
+ __ lea(rsi, Address(rbp, in_bytes(VM_Version::apx_xstate_size_offset())));
+ __ movl(Address(rsi, 0), rax);
+ __ lea(rsi, Address(rbp, in_bytes(VM_Version::apx_xstate_offset_offset())));
+ __ movl(Address(rsi, 0), rbx);
+
UseAPX = save_apx;
__ bind(vector_save_restore);
//
@@ -921,8 +935,9 @@ void VM_Version::get_processor_features() {
// Check if processor has Intel Ecore
if (FLAG_IS_DEFAULT(EnableX86ECoreOpts) && is_intel() && is_intel_server_family() &&
- (_model == 0x97 || _model == 0xAA || _model == 0xAC || _model == 0xAF ||
- _model == 0xCC || _model == 0xDD)) {
+ (supports_hybrid() ||
+ _model == 0xAF /* Xeon 6 E-cores (Sierra Forest) */ ||
+ _model == 0xDD /* Xeon 6+ E-cores (Clearwater Forest) */ )) {
FLAG_SET_DEFAULT(EnableX86ECoreOpts, true);
}
@@ -1137,6 +1152,10 @@ void VM_Version::get_processor_features() {
warning("AES intrinsics require UseAES flag to be enabled. Intrinsics will be disabled.");
}
FLAG_SET_DEFAULT(UseAESIntrinsics, false);
+ if (UseAESCTRIntrinsics && !FLAG_IS_DEFAULT(UseAESCTRIntrinsics)) {
+ warning("AES_CTR intrinsics require UseAES flag to be enabled. AES_CTR intrinsics will be disabled.");
+ }
+ FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false);
} else {
if (UseSSE > 2) {
if (FLAG_IS_DEFAULT(UseAESIntrinsics)) {
@@ -1155,8 +1174,8 @@ void VM_Version::get_processor_features() {
if (!UseAESIntrinsics) {
if (UseAESCTRIntrinsics && !FLAG_IS_DEFAULT(UseAESCTRIntrinsics)) {
warning("AES-CTR intrinsics require UseAESIntrinsics flag to be enabled. Intrinsics will be disabled.");
- FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false);
}
+ FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false);
} else {
if (supports_sse4_1()) {
if (FLAG_IS_DEFAULT(UseAESCTRIntrinsics)) {
@@ -1176,16 +1195,16 @@ void VM_Version::get_processor_features() {
} else if (UseAES || UseAESIntrinsics || UseAESCTRIntrinsics) {
if (UseAES && !FLAG_IS_DEFAULT(UseAES)) {
warning("AES instructions are not available on this CPU");
- FLAG_SET_DEFAULT(UseAES, false);
}
+ FLAG_SET_DEFAULT(UseAES, false);
if (UseAESIntrinsics && !FLAG_IS_DEFAULT(UseAESIntrinsics)) {
warning("AES intrinsics are not available on this CPU");
- FLAG_SET_DEFAULT(UseAESIntrinsics, false);
}
+ FLAG_SET_DEFAULT(UseAESIntrinsics, false);
if (UseAESCTRIntrinsics && !FLAG_IS_DEFAULT(UseAESCTRIntrinsics)) {
warning("AES-CTR intrinsics are not available on this CPU");
- FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false);
}
+ FLAG_SET_DEFAULT(UseAESCTRIntrinsics, false);
}
// Use CLMUL instructions if available.
@@ -1340,16 +1359,16 @@ void VM_Version::get_processor_features() {
FLAG_SET_DEFAULT(UseSHA512Intrinsics, false);
}
- if (supports_evex() && supports_avx512bw()) {
- if (FLAG_IS_DEFAULT(UseSHA3Intrinsics)) {
- UseSHA3Intrinsics = true;
- }
+ if (UseSHA && supports_evex() && supports_avx512bw()) {
+ if (FLAG_IS_DEFAULT(UseSHA3Intrinsics)) {
+ FLAG_SET_DEFAULT(UseSHA3Intrinsics, true);
+ }
} else if (UseSHA3Intrinsics) {
- warning("Intrinsics for SHA3-224, SHA3-256, SHA3-384 and SHA3-512 crypto hash functions not available on this CPU.");
- FLAG_SET_DEFAULT(UseSHA3Intrinsics, false);
+ warning("Intrinsics for SHA3-224, SHA3-256, SHA3-384 and SHA3-512 crypto hash functions not available on this CPU.");
+ FLAG_SET_DEFAULT(UseSHA3Intrinsics, false);
}
- if (!(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics)) {
+ if (!(UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics)) {
FLAG_SET_DEFAULT(UseSHA, false);
}
diff --git a/src/hotspot/cpu/x86/vm_version_x86.hpp b/src/hotspot/cpu/x86/vm_version_x86.hpp
index cc93ee3564e..a3f2a801198 100644
--- a/src/hotspot/cpu/x86/vm_version_x86.hpp
+++ b/src/hotspot/cpu/x86/vm_version_x86.hpp
@@ -676,6 +676,10 @@ protected:
// Space to save apx registers after signal handle
jlong apx_save[2]; // Save r16 and r31
+ // cpuid function 0xD, subleaf 19 (APX extended state)
+ uint32_t apx_xstate_size; // EAX: size of APX state (128)
+ uint32_t apx_xstate_offset; // EBX: offset in standard XSAVE area
+
VM_Features feature_flags() const;
// Asserts
@@ -739,6 +743,11 @@ public:
static ByteSize ymm_save_offset() { return byte_offset_of(CpuidInfo, ymm_save); }
static ByteSize zmm_save_offset() { return byte_offset_of(CpuidInfo, zmm_save); }
static ByteSize apx_save_offset() { return byte_offset_of(CpuidInfo, apx_save); }
+ static ByteSize apx_xstate_offset_offset() { return byte_offset_of(CpuidInfo, apx_xstate_offset); }
+ static ByteSize apx_xstate_size_offset() { return byte_offset_of(CpuidInfo, apx_xstate_size); }
+
+ static uint32_t apx_xstate_offset() { return _cpuid_info.apx_xstate_offset; }
+ static uint32_t apx_xstate_size() { return _cpuid_info.apx_xstate_size; }
// The value used to check ymm register after signal handle
static int ymm_test_value() { return 0xCAFEBABE; }
diff --git a/src/hotspot/cpu/x86/x86.ad b/src/hotspot/cpu/x86/x86.ad
index 93b306c37d6..aed54fe93d4 100644
--- a/src/hotspot/cpu/x86/x86.ad
+++ b/src/hotspot/cpu/x86/x86.ad
@@ -1699,9 +1699,10 @@ static void emit_cmpfp_fixup(MacroAssembler* masm) {
}
static void emit_cmpfp3(MacroAssembler* masm, Register dst) {
+ // If any floating point comparison instruction is used, unordered case always triggers jump
+ // for below condition, CF=1 is true when at least one input is NaN
Label done;
__ movl(dst, -1);
- __ jcc(Assembler::parity, done);
__ jcc(Assembler::below, done);
__ setcc(Assembler::notEqual, dst);
__ bind(done);
@@ -5529,12 +5530,21 @@ operand rFlagsRegU()
operand rFlagsRegUCF() %{
constraint(ALLOC_IN_RC(int_flags));
match(RegFlags);
- predicate(false);
+ predicate(!UseAPX || !VM_Version::supports_avx10_2());
format %{ "RFLAGS_U_CF" %}
interface(REG_INTER);
%}
+operand rFlagsRegUCFE() %{
+ constraint(ALLOC_IN_RC(int_flags));
+ match(RegFlags);
+ predicate(UseAPX && VM_Version::supports_avx10_2());
+
+ format %{ "RFLAGS_U_CFE" %}
+ interface(REG_INTER);
+%}
+
// Float register operands
operand regF() %{
constraint(ALLOC_IN_RC(float_reg));
@@ -6027,10 +6037,10 @@ operand cmpOp()
interface(COND_INTER) %{
equal(0x4, "e");
not_equal(0x5, "ne");
- less(0xC, "l");
- greater_equal(0xD, "ge");
- less_equal(0xE, "le");
- greater(0xF, "g");
+ less(0xc, "l");
+ greater_equal(0xd, "ge");
+ less_equal(0xe, "le");
+ greater(0xf, "g");
overflow(0x0, "o");
no_overflow(0x1, "no");
%}
@@ -6062,11 +6072,12 @@ operand cmpOpU()
// don't need to use cmpOpUCF2 for eq/ne
operand cmpOpUCF() %{
match(Bool);
- predicate(n->as_Bool()->_test._test == BoolTest::lt ||
- n->as_Bool()->_test._test == BoolTest::ge ||
- n->as_Bool()->_test._test == BoolTest::le ||
- n->as_Bool()->_test._test == BoolTest::gt ||
- n->in(1)->in(1) == n->in(1)->in(2));
+ predicate((!UseAPX || !VM_Version::supports_avx10_2()) &&
+ (n->as_Bool()->_test._test == BoolTest::lt ||
+ n->as_Bool()->_test._test == BoolTest::ge ||
+ n->as_Bool()->_test._test == BoolTest::le ||
+ n->as_Bool()->_test._test == BoolTest::gt ||
+ n->in(1)->in(1) == n->in(1)->in(2)));
format %{ "" %}
interface(COND_INTER) %{
equal(0xb, "np");
@@ -6084,7 +6095,8 @@ operand cmpOpUCF() %{
// Floating comparisons that can be fixed up with extra conditional jumps
operand cmpOpUCF2() %{
match(Bool);
- predicate((n->as_Bool()->_test._test == BoolTest::ne ||
+ predicate((!UseAPX || !VM_Version::supports_avx10_2()) &&
+ (n->as_Bool()->_test._test == BoolTest::ne ||
n->as_Bool()->_test._test == BoolTest::eq) &&
n->in(1)->in(1) != n->in(1)->in(2));
format %{ "" %}
@@ -6100,6 +6112,37 @@ operand cmpOpUCF2() %{
%}
%}
+
+// Floating point comparisons that set condition flags to test more directly,
+// Unsigned tests are used for G (>) and GE (>=) conditions while signed tests
+// are used for L (<) and LE (<=) conditions. It's important to convert these
+// latter conditions to ones that use unsigned tests before passing into an
+// instruction because the preceding comparison might be based on a three way
+// comparison (CmpF3 or CmpD3) that also assigns unordered outcomes to -1.
+operand cmpOpUCFE()
+%{
+ match(Bool);
+ predicate((UseAPX && VM_Version::supports_avx10_2()) &&
+ (n->as_Bool()->_test._test == BoolTest::ne ||
+ n->as_Bool()->_test._test == BoolTest::eq ||
+ n->as_Bool()->_test._test == BoolTest::lt ||
+ n->as_Bool()->_test._test == BoolTest::ge ||
+ n->as_Bool()->_test._test == BoolTest::le ||
+ n->as_Bool()->_test._test == BoolTest::gt));
+
+ format %{ "" %}
+ interface(COND_INTER) %{
+ equal(0x4, "e");
+ not_equal(0x5, "ne");
+ less(0x2, "b");
+ greater_equal(0x3, "ae");
+ less_equal(0x6, "be");
+ greater(0x7, "a");
+ overflow(0x0, "o");
+ no_overflow(0x1, "no");
+ %}
+%}
+
// Operands for bound floating pointer register arguments
operand rxmm0() %{
constraint(ALLOC_IN_RC(xmm0_reg));
@@ -9116,20 +9159,34 @@ instruct cmovI_imm_01UCF(rRegI dst, immI_1 src, rFlagsRegUCF cr, cmpOpUCF cop)
ins_pipe(ialu_reg);
%}
+instruct cmovI_imm_01UCFE(rRegI dst, immI_1 src, rFlagsRegUCFE cr, cmpOpUCFE cop)
+%{
+ predicate(n->in(2)->in(2)->is_Con() && n->in(2)->in(2)->get_int() == 0);
+ match(Set dst (CMoveI (Binary cop cr) (Binary src dst)));
+
+ ins_cost(100); // XXX
+ format %{ "setbn$cop $dst\t# signed, unsigned, int" %}
+ ins_encode %{
+ Assembler::Condition cond = (Assembler::Condition)($cop$$cmpcode);
+ __ setb(MacroAssembler::negate_condition(cond), $dst$$Register);
+ %}
+ ins_pipe(ialu_reg);
+%}
+
instruct cmovI_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, rRegI src) %{
- predicate(!UseAPX);
match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
+
ins_cost(200);
expand %{
cmovI_regU(cop, cr, dst, src);
%}
%}
-instruct cmovI_regUCF_ndd(rRegI dst, cmpOpUCF cop, rFlagsRegUCF cr, rRegI src1, rRegI src2) %{
- predicate(UseAPX);
+instruct cmovI_regUCFE_ndd(rRegI dst, cmpOpUCFE cop, rFlagsRegUCFE cr, rRegI src1, rRegI src2) %{
match(Set dst (CMoveI (Binary cop cr) (Binary src1 src2)));
+
ins_cost(200);
- format %{ "ecmovl$cop $dst, $src1, $src2\t# unsigned, int ndd" %}
+ format %{ "ecmovl$cop $dst, $src1, $src2\t# signed, unsigned, int ndd" %}
ins_encode %{
__ ecmovl((Assembler::Condition)($cop$$cmpcode), $dst$$Register, $src1$$Register, $src2$$Register);
%}
@@ -9137,7 +9194,7 @@ instruct cmovI_regUCF_ndd(rRegI dst, cmpOpUCF cop, rFlagsRegUCF cr, rRegI src1,
%}
instruct cmovI_regUCF2_ne(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegI dst, rRegI src) %{
- predicate(!UseAPX && n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::ne);
+ predicate(n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::ne);
match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
ins_cost(200); // XXX
@@ -9150,25 +9207,10 @@ instruct cmovI_regUCF2_ne(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegI dst, rRegI src)
ins_pipe(pipe_cmov_reg);
%}
-instruct cmovI_regUCF2_ne_ndd(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegI dst, rRegI src1, rRegI src2) %{
- predicate(UseAPX && n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::ne);
- match(Set dst (CMoveI (Binary cop cr) (Binary src1 src2)));
- effect(TEMP dst);
-
- ins_cost(200);
- format %{ "ecmovpl $dst, $src1, $src2\n\t"
- "cmovnel $dst, $src2" %}
- ins_encode %{
- __ ecmovl(Assembler::parity, $dst$$Register, $src1$$Register, $src2$$Register);
- __ cmovl(Assembler::notEqual, $dst$$Register, $src2$$Register);
- %}
- ins_pipe(pipe_cmov_reg);
-%}
-
// Since (x == y) == !(x != y), we can flip the sense of the test by flipping the
// inputs of the CMove
instruct cmovI_regUCF2_eq(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegI dst, rRegI src) %{
- predicate(!UseAPX && n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::eq);
+ predicate(n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::eq);
match(Set dst (CMoveI (Binary cop cr) (Binary src dst)));
effect(TEMP dst);
@@ -9182,23 +9224,6 @@ instruct cmovI_regUCF2_eq(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegI dst, rRegI src)
ins_pipe(pipe_cmov_reg);
%}
-// We need this special handling for only eq / neq comparison since NaN == NaN is false,
-// and parity flag bit is set if any of the operand is a NaN.
-instruct cmovI_regUCF2_eq_ndd(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegI dst, rRegI src1, rRegI src2) %{
- predicate(UseAPX && n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::eq);
- match(Set dst (CMoveI (Binary cop cr) (Binary src2 src1)));
- effect(TEMP dst);
-
- ins_cost(200);
- format %{ "ecmovpl $dst, $src1, $src2\n\t"
- "cmovnel $dst, $src2" %}
- ins_encode %{
- __ ecmovl(Assembler::parity, $dst$$Register, $src1$$Register, $src2$$Register);
- __ cmovl(Assembler::notEqual, $dst$$Register, $src2$$Register);
- %}
- ins_pipe(pipe_cmov_reg);
-%}
-
// Conditional move
instruct cmovI_mem(cmpOp cop, rFlagsReg cr, rRegI dst, memory src) %{
predicate(!UseAPX);
@@ -9241,8 +9266,8 @@ instruct cmovI_memU(cmpOpU cop, rFlagsRegU cr, rRegI dst, memory src)
%}
instruct cmovI_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, memory src) %{
- predicate(!UseAPX);
match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
+
ins_cost(250);
expand %{
cmovI_memU(cop, cr, dst, src);
@@ -9262,12 +9287,12 @@ instruct cmovI_rReg_rReg_memU_ndd(rRegI dst, cmpOpU cop, rFlagsRegU cr, rRegI sr
ins_pipe(pipe_cmov_mem);
%}
-instruct cmovI_rReg_rReg_memUCF_ndd(rRegI dst, cmpOpUCF cop, rFlagsRegUCF cr, rRegI src1, memory src2)
+instruct cmovI_rReg_rReg_memUCFE_ndd(rRegI dst, cmpOpUCFE cop, rFlagsRegUCFE cr, rRegI src1, memory src2)
%{
- predicate(UseAPX);
match(Set dst (CMoveI (Binary cop cr) (Binary src1 (LoadI src2))));
+
ins_cost(250);
- format %{ "ecmovl$cop $dst, $src1, $src2\t# unsigned, int ndd" %}
+ format %{ "ecmovl$cop $dst, $src1, $src2\t# signed, unsigned, int ndd" %}
ins_encode %{
__ ecmovl((Assembler::Condition)($cop$$cmpcode), $dst$$Register, $src1$$Register, $src2$$Address);
%}
@@ -9317,8 +9342,8 @@ instruct cmovN_regU(cmpOpU cop, rFlagsRegU cr, rRegN dst, rRegN src)
%}
instruct cmovN_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegN dst, rRegN src) %{
- predicate(!UseAPX);
match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
+
ins_cost(200);
expand %{
cmovN_regU(cop, cr, dst, src);
@@ -9339,11 +9364,11 @@ instruct cmovN_regU_ndd(rRegN dst, cmpOpU cop, rFlagsRegU cr, rRegN src1, rRegN
ins_pipe(pipe_cmov_reg);
%}
-instruct cmovN_regUCF_ndd(rRegN dst, cmpOpUCF cop, rFlagsRegUCF cr, rRegN src1, rRegN src2) %{
- predicate(UseAPX);
+instruct cmovN_regUCFE_ndd(rRegN dst, cmpOpUCFE cop, rFlagsRegUCFE cr, rRegN src1, rRegN src2) %{
match(Set dst (CMoveN (Binary cop cr) (Binary src1 src2)));
+
ins_cost(200);
- format %{ "ecmovl$cop $dst, $src1, $src2\t# unsigned, compressed ptr ndd" %}
+ format %{ "ecmovl$cop $dst, $src1, $src2\t# signed, unsigned, compressed ptr ndd" %}
ins_encode %{
__ ecmovl((Assembler::Condition)($cop$$cmpcode), $dst$$Register, $src1$$Register, $src2$$Register);
%}
@@ -9437,19 +9462,19 @@ instruct cmovP_regU_ndd(rRegP dst, cmpOpU cop, rFlagsRegU cr, rRegP src1, rRegP
%}
instruct cmovP_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegP dst, rRegP src) %{
- predicate(!UseAPX);
match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
+
ins_cost(200);
expand %{
cmovP_regU(cop, cr, dst, src);
%}
%}
-instruct cmovP_regUCF_ndd(rRegP dst, cmpOpUCF cop, rFlagsRegUCF cr, rRegP src1, rRegP src2) %{
- predicate(UseAPX);
+instruct cmovP_regUCFE_ndd(rRegP dst, cmpOpUCFE cop, rFlagsRegUCFE cr, rRegP src1, rRegP src2) %{
match(Set dst (CMoveP (Binary cop cr) (Binary src1 src2)));
+
ins_cost(200);
- format %{ "ecmovq$cop $dst, $src1, $src2\t# unsigned, ptr ndd" %}
+ format %{ "ecmovq$cop $dst, $src1, $src2\t# signed, unsigned, ptr ndd" %}
ins_encode %{
__ ecmovq((Assembler::Condition)($cop$$cmpcode), $dst$$Register, $src1$$Register, $src2$$Register);
%}
@@ -9457,7 +9482,7 @@ instruct cmovP_regUCF_ndd(rRegP dst, cmpOpUCF cop, rFlagsRegUCF cr, rRegP src1,
%}
instruct cmovP_regUCF2_ne(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegP dst, rRegP src) %{
- predicate(!UseAPX && n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::ne);
+ predicate(n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::ne);
match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
ins_cost(200); // XXX
@@ -9470,25 +9495,10 @@ instruct cmovP_regUCF2_ne(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegP dst, rRegP src)
ins_pipe(pipe_cmov_reg);
%}
-instruct cmovP_regUCF2_ne_ndd(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegP dst, rRegP src1, rRegP src2) %{
- predicate(UseAPX && n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::ne);
- match(Set dst (CMoveP (Binary cop cr) (Binary src1 src2)));
- effect(TEMP dst);
-
- ins_cost(200);
- format %{ "ecmovpq $dst, $src1, $src2\n\t"
- "cmovneq $dst, $src2" %}
- ins_encode %{
- __ ecmovq(Assembler::parity, $dst$$Register, $src1$$Register, $src2$$Register);
- __ cmovq(Assembler::notEqual, $dst$$Register, $src2$$Register);
- %}
- ins_pipe(pipe_cmov_reg);
-%}
-
// Since (x == y) == !(x != y), we can flip the sense of the test by flipping the
// inputs of the CMove
instruct cmovP_regUCF2_eq(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegP dst, rRegP src) %{
- predicate(!UseAPX && n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::eq);
+ predicate(n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::eq);
match(Set dst (CMoveP (Binary cop cr) (Binary src dst)));
ins_cost(200); // XXX
@@ -9501,21 +9511,6 @@ instruct cmovP_regUCF2_eq(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegP dst, rRegP src)
ins_pipe(pipe_cmov_reg);
%}
-instruct cmovP_regUCF2_eq_ndd(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegP dst, rRegP src1, rRegP src2) %{
- predicate(UseAPX && n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::eq);
- match(Set dst (CMoveP (Binary cop cr) (Binary src2 src1)));
- effect(TEMP dst);
-
- ins_cost(200);
- format %{ "ecmovpq $dst, $src1, $src2\n\t"
- "cmovneq $dst, $src2" %}
- ins_encode %{
- __ ecmovq(Assembler::parity, $dst$$Register, $src1$$Register, $src2$$Register);
- __ cmovq(Assembler::notEqual, $dst$$Register, $src2$$Register);
- %}
- ins_pipe(pipe_cmov_reg);
-%}
-
instruct cmovL_imm_01(rRegL dst, immL1 src, rFlagsReg cr, cmpOp cop)
%{
predicate(n->in(2)->in(2)->is_Con() && n->in(2)->in(2)->get_long() == 0);
@@ -9636,21 +9631,35 @@ instruct cmovL_imm_01UCF(rRegL dst, immL1 src, rFlagsRegUCF cr, cmpOpUCF cop)
ins_pipe(ialu_reg);
%}
+instruct cmovL_imm_01UCFE(rRegL dst, immL1 src, rFlagsRegUCFE cr, cmpOpUCFE cop)
+%{
+ predicate(n->in(2)->in(2)->is_Con() && n->in(2)->in(2)->get_long() == 0);
+ match(Set dst (CMoveL (Binary cop cr) (Binary src dst)));
+
+ ins_cost(100); // XXX
+ format %{ "setbn$cop $dst\t# signed, unsigned, long" %}
+ ins_encode %{
+ Assembler::Condition cond = (Assembler::Condition)($cop$$cmpcode);
+ __ setb(MacroAssembler::negate_condition(cond), $dst$$Register);
+ %}
+ ins_pipe(ialu_reg);
+%}
+
instruct cmovL_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, rRegL src) %{
- predicate(!UseAPX);
match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
+
ins_cost(200);
expand %{
cmovL_regU(cop, cr, dst, src);
%}
%}
-instruct cmovL_regUCF_ndd(rRegL dst, cmpOpUCF cop, rFlagsRegUCF cr, rRegL src1, rRegL src2)
+instruct cmovL_regUCFE_ndd(rRegL dst, cmpOpUCFE cop, rFlagsRegUCFE cr, rRegL src1, rRegL src2)
%{
- predicate(UseAPX);
match(Set dst (CMoveL (Binary cop cr) (Binary src1 src2)));
+
ins_cost(200);
- format %{ "ecmovq$cop $dst, $src1, $src2\t# unsigned, long ndd" %}
+ format %{ "ecmovq$cop $dst, $src1, $src2\t# signed, unsigned, long ndd" %}
ins_encode %{
__ ecmovq((Assembler::Condition)($cop$$cmpcode), $dst$$Register, $src1$$Register, $src2$$Register);
%}
@@ -9658,7 +9667,7 @@ instruct cmovL_regUCF_ndd(rRegL dst, cmpOpUCF cop, rFlagsRegUCF cr, rRegL src1,
%}
instruct cmovL_regUCF2_ne(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegL dst, rRegL src) %{
- predicate(!UseAPX && n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::ne);
+ predicate(n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::ne);
match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
ins_cost(200); // XXX
@@ -9671,25 +9680,10 @@ instruct cmovL_regUCF2_ne(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegL dst, rRegL src)
ins_pipe(pipe_cmov_reg);
%}
-instruct cmovL_regUCF2_ne_ndd(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegL dst, rRegL src1, rRegL src2) %{
- predicate(UseAPX && n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::ne);
- match(Set dst (CMoveL (Binary cop cr) (Binary src1 src2)));
- effect(TEMP dst);
-
- ins_cost(200);
- format %{ "ecmovpq $dst, $src1, $src2\n\t"
- "cmovneq $dst, $src2" %}
- ins_encode %{
- __ ecmovq(Assembler::parity, $dst$$Register, $src1$$Register, $src2$$Register);
- __ cmovq(Assembler::notEqual, $dst$$Register, $src2$$Register);
- %}
- ins_pipe(pipe_cmov_reg);
-%}
-
// Since (x == y) == !(x != y), we can flip the sense of the test by flipping the
// inputs of the CMove
instruct cmovL_regUCF2_eq(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegL dst, rRegL src) %{
- predicate(!UseAPX && n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::eq);
+ predicate(n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::eq);
match(Set dst (CMoveL (Binary cop cr) (Binary src dst)));
ins_cost(200); // XXX
@@ -9702,21 +9696,6 @@ instruct cmovL_regUCF2_eq(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegL dst, rRegL src)
ins_pipe(pipe_cmov_reg);
%}
-instruct cmovL_regUCF2_eq_ndd(cmpOpUCF2 cop, rFlagsRegUCF cr, rRegL dst, rRegL src1, rRegL src2) %{
- predicate(UseAPX && n->in(1)->in(1)->as_Bool()->_test._test == BoolTest::eq);
- match(Set dst (CMoveL (Binary cop cr) (Binary src2 src1)));
- effect(TEMP dst);
-
- ins_cost(200);
- format %{ "ecmovpq $dst, $src1, $src2\n\t"
- "cmovneq $dst, $src2" %}
- ins_encode %{
- __ ecmovq(Assembler::parity, $dst$$Register, $src1$$Register, $src2$$Register);
- __ cmovq(Assembler::notEqual, $dst$$Register, $src2$$Register);
- %}
- ins_pipe(pipe_cmov_reg);
-%}
-
instruct cmovL_memU(cmpOpU cop, rFlagsRegU cr, rRegL dst, memory src)
%{
predicate(!UseAPX);
@@ -9731,8 +9710,8 @@ instruct cmovL_memU(cmpOpU cop, rFlagsRegU cr, rRegL dst, memory src)
%}
instruct cmovL_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, memory src) %{
- predicate(!UseAPX);
match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
+
ins_cost(200);
expand %{
cmovL_memU(cop, cr, dst, src);
@@ -9752,12 +9731,12 @@ instruct cmovL_rReg_rReg_memU_ndd(rRegL dst, cmpOpU cop, rFlagsRegU cr, rRegL sr
ins_pipe(pipe_cmov_mem);
%}
-instruct cmovL_rReg_rReg_memUCF_ndd(rRegL dst, cmpOpUCF cop, rFlagsRegUCF cr, rRegL src1, memory src2)
+instruct cmovL_rReg_rReg_memUCFE_ndd(rRegL dst, cmpOpUCFE cop, rFlagsRegUCFE cr, rRegL src1, memory src2)
%{
- predicate(UseAPX);
match(Set dst (CMoveL (Binary cop cr) (Binary src1 (LoadL src2))));
+
ins_cost(200);
- format %{ "ecmovq$cop $dst, $src1, $src2\t# unsigned, long ndd" %}
+ format %{ "ecmovq$cop $dst, $src1, $src2\t# signed, unsigned, long ndd" %}
ins_encode %{
__ ecmovq((Assembler::Condition)($cop$$cmpcode), $dst$$Register, $src1$$Register, $src2$$Address);
%}
@@ -9802,12 +9781,31 @@ instruct cmovF_regU(cmpOpU cop, rFlagsRegU cr, regF dst, regF src)
instruct cmovF_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regF dst, regF src) %{
match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
+
ins_cost(200);
expand %{
cmovF_regU(cop, cr, dst, src);
%}
%}
+instruct cmovF_regUCFE(cmpOpUCFE cop, rFlagsRegUCFE cr, regF dst, regF src)
+%{
+ match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
+
+ ins_cost(200); // XXX
+ format %{ "jn$cop skip\t# signed, unsigned cmove float\n\t"
+ "movss $dst, $src\n"
+ "skip:" %}
+ ins_encode %{
+ Label Lskip;
+ // Invert sense of branch from sense of CMOV
+ __ jccb((Assembler::Condition)($cop$$cmpcode^1), Lskip);
+ __ movflt($dst$$XMMRegister, $src$$XMMRegister);
+ __ bind(Lskip);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
instruct cmovD_reg(cmpOp cop, rFlagsReg cr, regD dst, regD src)
%{
match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
@@ -9846,12 +9844,31 @@ instruct cmovD_regU(cmpOpU cop, rFlagsRegU cr, regD dst, regD src)
instruct cmovD_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regD dst, regD src) %{
match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
+
ins_cost(200);
expand %{
cmovD_regU(cop, cr, dst, src);
%}
%}
+instruct cmovD_regUCFE(cmpOpUCFE cop, rFlagsRegUCFE cr, regD dst, regD src)
+%{
+ match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
+
+ ins_cost(200); // XXX
+ format %{ "jn$cop skip\t# signed, unsigned cmove double\n\t"
+ "movsd $dst, $src\n"
+ "skip:" %}
+ ins_encode %{
+ Label Lskip;
+ // Invert sense of branch from sense of CMOV
+ __ jccb((Assembler::Condition)($cop$$cmpcode^1), Lskip);
+ __ movdbl($dst$$XMMRegister, $src$$XMMRegister);
+ __ bind(Lskip);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
//----------Arithmetic Instructions--------------------------------------------
//----------Addition Instructions----------------------------------------------
@@ -14319,7 +14336,7 @@ instruct cmpF_cc_reg(rFlagsRegU cr, regF src1, regF src2)
ins_pipe(pipe_slow);
%}
-instruct cmpF_cc_reg_CF(rFlagsRegUCF cr, regF src1, regF src2) %{
+instruct cmpF_cc_regCF(rFlagsRegUCF cr, regF src1, regF src2) %{
match(Set cr (CmpF src1 src2));
ins_cost(100);
@@ -14330,6 +14347,17 @@ instruct cmpF_cc_reg_CF(rFlagsRegUCF cr, regF src1, regF src2) %{
ins_pipe(pipe_slow);
%}
+instruct cmpF_cc_regCFE(rFlagsRegUCFE cr, regF src1, regF src2) %{
+ match(Set cr (CmpF src1 src2));
+
+ ins_cost(100);
+ format %{ "vucomxss $src1, $src2" %}
+ ins_encode %{
+ __ vucomxss($src1$$XMMRegister, $src2$$XMMRegister);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
instruct cmpF_cc_memCF(rFlagsRegUCF cr, regF src1, memory src2) %{
match(Set cr (CmpF src1 (LoadF src2)));
@@ -14341,8 +14369,20 @@ instruct cmpF_cc_memCF(rFlagsRegUCF cr, regF src1, memory src2) %{
ins_pipe(pipe_slow);
%}
+instruct cmpF_cc_memCFE(rFlagsRegUCFE cr, regF src1, memory src2) %{
+ match(Set cr (CmpF src1 (LoadF src2)));
+
+ ins_cost(100);
+ format %{ "vucomxss $src1, $src2" %}
+ ins_encode %{
+ __ vucomxss($src1$$XMMRegister, $src2$$Address);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
instruct cmpF_cc_immCF(rFlagsRegUCF cr, regF src, immF con) %{
match(Set cr (CmpF src con));
+
ins_cost(100);
format %{ "ucomiss $src, [$constantaddress]\t# load from constant table: float=$con" %}
ins_encode %{
@@ -14351,6 +14391,17 @@ instruct cmpF_cc_immCF(rFlagsRegUCF cr, regF src, immF con) %{
ins_pipe(pipe_slow);
%}
+instruct cmpF_cc_immCFE(rFlagsRegUCFE cr, regF src, immF con) %{
+ match(Set cr (CmpF src con));
+
+ ins_cost(100);
+ format %{ "vucomxss $src, [$constantaddress]\t# load from constant table: float=$con" %}
+ ins_encode %{
+ __ vucomxss($src$$XMMRegister, $constantaddress($con));
+ %}
+ ins_pipe(pipe_slow);
+%}
+
// Really expensive, avoid
instruct cmpD_cc_reg(rFlagsRegU cr, regD src1, regD src2)
%{
@@ -14370,7 +14421,7 @@ instruct cmpD_cc_reg(rFlagsRegU cr, regD src1, regD src2)
ins_pipe(pipe_slow);
%}
-instruct cmpD_cc_reg_CF(rFlagsRegUCF cr, regD src1, regD src2) %{
+instruct cmpD_cc_regCF(rFlagsRegUCF cr, regD src1, regD src2) %{
match(Set cr (CmpD src1 src2));
ins_cost(100);
@@ -14381,6 +14432,17 @@ instruct cmpD_cc_reg_CF(rFlagsRegUCF cr, regD src1, regD src2) %{
ins_pipe(pipe_slow);
%}
+instruct cmpD_cc_regCFE(rFlagsRegUCFE cr, regD src1, regD src2) %{
+ match(Set cr (CmpD src1 src2));
+
+ ins_cost(100);
+ format %{ "vucomxsd $src1, $src2 test" %}
+ ins_encode %{
+ __ vucomxsd($src1$$XMMRegister, $src2$$XMMRegister);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
instruct cmpD_cc_memCF(rFlagsRegUCF cr, regD src1, memory src2) %{
match(Set cr (CmpD src1 (LoadD src2)));
@@ -14392,6 +14454,17 @@ instruct cmpD_cc_memCF(rFlagsRegUCF cr, regD src1, memory src2) %{
ins_pipe(pipe_slow);
%}
+instruct cmpD_cc_memCFE(rFlagsRegUCFE cr, regD src1, memory src2) %{
+ match(Set cr (CmpD src1 (LoadD src2)));
+
+ ins_cost(100);
+ format %{ "vucomxsd $src1, $src2" %}
+ ins_encode %{
+ __ vucomxsd($src1$$XMMRegister, $src2$$Address);
+ %}
+ ins_pipe(pipe_slow);
+%}
+
instruct cmpD_cc_immCF(rFlagsRegUCF cr, regD src, immD con) %{
match(Set cr (CmpD src con));
ins_cost(100);
@@ -14402,6 +14475,17 @@ instruct cmpD_cc_immCF(rFlagsRegUCF cr, regD src, immD con) %{
ins_pipe(pipe_slow);
%}
+instruct cmpD_cc_immCFE(rFlagsRegUCFE cr, regD src, immD con) %{
+ match(Set cr (CmpD src con));
+
+ ins_cost(100);
+ format %{ "vucomxsd $src, [$constantaddress]\t# load from constant table: double=$con" %}
+ ins_encode %{
+ __ vucomxsd($src$$XMMRegister, $constantaddress($con));
+ %}
+ ins_pipe(pipe_slow);
+%}
+
// Compare into -1,0,1
instruct cmpF_reg(rRegI dst, regF src1, regF src2, rFlagsReg cr)
%{
@@ -16808,6 +16892,21 @@ instruct jmpConUCF2(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{
ins_pipe(pipe_jcc);
%}
+// Jump Direct Conditional - using signed and unsigned comparison
+instruct jmpConUCFE(cmpOpUCFE cop, rFlagsRegUCFE cmp, label labl) %{
+ match(If cop cmp);
+ effect(USE labl);
+
+ ins_cost(200);
+ format %{ "j$cop,su $labl" %}
+ size(6);
+ ins_encode %{
+ Label* L = $labl$$label;
+ __ jcc((Assembler::Condition)($cop$$cmpcode), *L, false); // Always long jump
+ %}
+ ins_pipe(pipe_jcc);
+%}
+
// ============================================================================
// The 2nd slow-half of a subtype check. Scan the subklass's 2ndary
// superklass array for an instance of the superklass. Set a hidden
@@ -17026,6 +17125,22 @@ instruct jmpConUCF2_short(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{
ins_short_branch(1);
%}
+// Jump Direct Conditional - using signed and unsigned comparison
+instruct jmpConUCFE_short(cmpOpUCFE cop, rFlagsRegUCFE cmp, label labl) %{
+ match(If cop cmp);
+ effect(USE labl);
+
+ ins_cost(300);
+ format %{ "j$cop,sus $labl" %}
+ size(2);
+ ins_encode %{
+ Label* L = $labl$$label;
+ __ jccb((Assembler::Condition)($cop$$cmpcode), *L);
+ %}
+ ins_pipe(pipe_jcc);
+ ins_short_branch(1);
+%}
+
// ============================================================================
// inlined locking and unlocking
diff --git a/src/hotspot/cpu/zero/globals_zero.hpp b/src/hotspot/cpu/zero/globals_zero.hpp
index 6b6c6ea983c..6dc7d81275c 100644
--- a/src/hotspot/cpu/zero/globals_zero.hpp
+++ b/src/hotspot/cpu/zero/globals_zero.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2000, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2000, 2026, Oracle and/or its affiliates. All rights reserved.
* Copyright 2007, 2008, 2009, 2010, 2011 Red Hat, Inc.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
@@ -39,7 +39,7 @@ define_pd_global(bool, UncommonNullCast, true);
define_pd_global(bool, DelayCompilerStubsGeneration, false); // Don't have compiler's stubs
define_pd_global(size_t, CodeCacheSegmentSize, 64 COMPILER1_AND_COMPILER2_PRESENT(+64)); // Tiered compilation has large code-entry alignment.
-define_pd_global(intx, CodeEntryAlignment, 32);
+define_pd_global(uint, CodeEntryAlignment, 32);
define_pd_global(intx, OptoLoopAlignment, 16);
define_pd_global(intx, InlineSmallCode, 1000);
diff --git a/src/hotspot/os/aix/decoder_aix.hpp b/src/hotspot/os/aix/decoder_aix.hpp
index 2ba3e1c5a3a..632355ccf4e 100644
--- a/src/hotspot/os/aix/decoder_aix.hpp
+++ b/src/hotspot/os/aix/decoder_aix.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2011, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2011, 2026, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2013 SAP SE. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
@@ -38,7 +38,7 @@ class AIXDecoder: public AbstractDecoder {
virtual bool demangle(const char* symbol, char* buf, int buflen) { return false; } // use AixSymbols::get_function_name to demangle
virtual bool decode(address addr, char* buf, int buflen, int* offset, const char* modulepath, bool demangle) {
- return AixSymbols::get_function_name(addr, buf, buflen, offset, 0, demangle);
+ return AixSymbols::get_function_name(addr, buf, buflen, offset, nullptr, demangle);
}
virtual bool decode(address addr, char *buf, int buflen, int* offset, const void *base) {
ShouldNotReachHere();
diff --git a/src/hotspot/os/aix/os_aix.cpp b/src/hotspot/os/aix/os_aix.cpp
index d7c1911a914..0a8efbece8d 100644
--- a/src/hotspot/os/aix/os_aix.cpp
+++ b/src/hotspot/os/aix/os_aix.cpp
@@ -703,7 +703,7 @@ static void *thread_native_entry(Thread *thread) {
log_info(os, thread)("Thread finished (tid: %zu, kernel thread id: %zu).",
os::current_thread_id(), (uintx) kernel_thread_id);
- return 0;
+ return nullptr;
}
bool os::create_thread(Thread* thread, ThreadType thr_type,
diff --git a/src/hotspot/os/aix/porting_aix.cpp b/src/hotspot/os/aix/porting_aix.cpp
index 7311afc197b..b3f878fbfdd 100644
--- a/src/hotspot/os/aix/porting_aix.cpp
+++ b/src/hotspot/os/aix/porting_aix.cpp
@@ -1,6 +1,6 @@
/*
* Copyright (c) 2012, 2024 SAP SE. All rights reserved.
- * Copyright (c) 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2025, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -78,7 +78,7 @@ class fixed_strings {
public:
- fixed_strings() : first(0) {}
+ fixed_strings() : first(nullptr) {}
~fixed_strings() {
node* n = first;
while (n) {
@@ -113,7 +113,7 @@ bool AixSymbols::get_function_name (
// information (null if not available)
bool demangle // [in] whether to demangle the name
) {
- struct tbtable* tb = 0;
+ struct tbtable* tb = nullptr;
unsigned int searchcount = 0;
// initialize output parameters
@@ -653,10 +653,10 @@ void AixNativeCallstack::print_callstack_for_context(outputStream* st, const uco
// To print the first frame, use the current value of iar:
// current entry indicated by iar (the current pc)
- codeptr_t cur_iar = 0;
- stackptr_t cur_sp = 0;
- codeptr_t cur_rtoc = 0;
- codeptr_t cur_lr = 0;
+ codeptr_t cur_iar = nullptr;
+ stackptr_t cur_sp = nullptr;
+ codeptr_t cur_rtoc = nullptr;
+ codeptr_t cur_lr = nullptr;
const ucontext_t* uc = (const ucontext_t*) context;
@@ -926,7 +926,7 @@ static struct handletableentry* p_handletable = nullptr;
static const char* rtv_linkedin_libpath() {
constexpr int bufsize = 4096;
static char buffer[bufsize];
- static const char* libpath = 0;
+ static const char* libpath = nullptr;
// we only try to retrieve the libpath once. After that try we
// let libpath point to buffer, which then contains a valid libpath
diff --git a/src/hotspot/os/bsd/globals_bsd.hpp b/src/hotspot/os/bsd/globals_bsd.hpp
index 850d491a11f..22f587ed789 100644
--- a/src/hotspot/os/bsd/globals_bsd.hpp
+++ b/src/hotspot/os/bsd/globals_bsd.hpp
@@ -28,6 +28,7 @@
//
// Declare Bsd specific flags. They are not available on other platforms.
//
+#ifdef AARCH64
#define RUNTIME_OS_FLAGS(develop, \
develop_pd, \
product, \
@@ -35,9 +36,21 @@
range, \
constraint) \
\
- AARCH64_ONLY(develop(bool, AssertWXAtThreadSync, true, \
- "Conservatively check W^X thread state at possible safepoint" \
- "or handshake"))
+ develop(bool, TraceWXHealing, false, \
+ "track occurrences of W^X mode healing") \
+ develop(bool, UseOldWX, false, \
+ "Choose old W^X implementation.") \
+ product(bool, StressWXHealing, false, DIAGNOSTIC, \
+ "Stress W xor X healing on MacOS")
+
+#else
+#define RUNTIME_OS_FLAGS(develop, \
+ develop_pd, \
+ product, \
+ product_pd, \
+ range, \
+ constraint)
+#endif
// end of RUNTIME_OS_FLAGS
diff --git a/src/hotspot/os/bsd/os_bsd.cpp b/src/hotspot/os/bsd/os_bsd.cpp
index 0e21c2d1785..81320b4f1aa 100644
--- a/src/hotspot/os/bsd/os_bsd.cpp
+++ b/src/hotspot/os/bsd/os_bsd.cpp
@@ -841,6 +841,7 @@ jlong os::javaTimeNanos() {
// We might also condition (c) on the magnitude of the delta between obsv and now.
// Avoiding excessive CAS operations to hot RW locations is critical.
// See https://blogs.oracle.com/dave/entry/cas_and_cache_trivia_invalidate
+ // https://web.archive.org/web/20131214182431/https://blogs.oracle.com/dave/entry/cas_and_cache_trivia_invalidate
return (prev == obsv) ? now : obsv;
}
diff --git a/src/hotspot/os_cpu/aix_ppc/os_aix_ppc.cpp b/src/hotspot/os_cpu/aix_ppc/os_aix_ppc.cpp
index afef21b091a..3ab81697280 100644
--- a/src/hotspot/os_cpu/aix_ppc/os_aix_ppc.cpp
+++ b/src/hotspot/os_cpu/aix_ppc/os_aix_ppc.cpp
@@ -1,6 +1,6 @@
/*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, 2025 SAP SE. All rights reserved.
+ * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012, 2026 SAP SE. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -412,12 +412,8 @@ run_stub:
}
void os::Aix::init_thread_fpu_state(void) {
-#if !defined(USE_XLC_BUILTINS)
// Disable FP exceptions.
__asm__ __volatile__ ("mtfsfi 6,0");
-#else
- __mtfsfi(6, 0);
-#endif
}
////////////////////////////////////////////////////////////////////////////////
diff --git a/src/hotspot/os_cpu/aix_ppc/prefetch_aix_ppc.inline.hpp b/src/hotspot/os_cpu/aix_ppc/prefetch_aix_ppc.inline.hpp
index c741335b5f0..d9dac0e231f 100644
--- a/src/hotspot/os_cpu/aix_ppc/prefetch_aix_ppc.inline.hpp
+++ b/src/hotspot/os_cpu/aix_ppc/prefetch_aix_ppc.inline.hpp
@@ -1,6 +1,6 @@
/*
* Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, 2013 SAP SE. All rights reserved.
+ * Copyright (c) 2012, 2026 SAP SE. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -29,29 +29,21 @@
// Included in runtime/prefetch.inline.hpp
inline void Prefetch::read(const void *loc, intx interval) {
-#if !defined(USE_XLC_BUILTINS)
__asm__ __volatile__ (
" dcbt 0, %0 \n"
:
: /*%0*/"r" ( ((address)loc) +((long)interval) )
//:
);
-#else
- __dcbt(((address)loc) +((long)interval));
-#endif
}
inline void Prefetch::write(void *loc, intx interval) {
-#if !defined(USE_XLC_BUILTINS)
__asm__ __volatile__ (
" dcbtst 0, %0 \n"
:
: /*%0*/"r" ( ((address)loc) +((long)interval) )
//:
);
-#else
- __dcbtst( ((address)loc) +((long)interval) );
-#endif
}
#endif // OS_CPU_AIX_PPC_PREFETCH_AIX_PPC_INLINE_HPP
diff --git a/src/hotspot/os_cpu/bsd_aarch64/os_bsd_aarch64.cpp b/src/hotspot/os_cpu/bsd_aarch64/os_bsd_aarch64.cpp
index 62dba218b2f..36599594842 100644
--- a/src/hotspot/os_cpu/bsd_aarch64/os_bsd_aarch64.cpp
+++ b/src/hotspot/os_cpu/bsd_aarch64/os_bsd_aarch64.cpp
@@ -54,8 +54,11 @@
#include "signals_posix.hpp"
#include "utilities/align.hpp"
#include "utilities/debug.hpp"
+#include "utilities/decoder.hpp"
#include "utilities/events.hpp"
+#include "utilities/nativeStackPrinter.hpp"
#include "utilities/vmError.hpp"
+#include "compiler/disassembler.hpp"
// put OS-includes here
# include
@@ -85,6 +88,8 @@
#define SPELL_REG_SP "sp"
#ifdef __APPLE__
+WXMode DefaultWXWriteMode;
+
// see darwin-xnu/osfmk/mach/arm/_structs.h
// 10.5 UNIX03 member name prefixes
@@ -233,19 +238,56 @@ NOINLINE frame os::current_frame() {
bool PosixSignals::pd_hotspot_signal_handler(int sig, siginfo_t* info,
ucontext_t* uc, JavaThread* thread) {
- // Enable WXWrite: this function is called by the signal handler at arbitrary
- // point of execution.
- ThreadWXEnable wx(WXWrite, thread);
-
// decide if this trap can be handled by a stub
address stub = nullptr;
-
- address pc = nullptr;
+ address pc = nullptr;
//%note os_trap_1
if (info != nullptr && uc != nullptr && thread != nullptr) {
pc = (address) os::Posix::ucontext_get_pc(uc);
+#ifdef MACOS_AARCH64
+ // If we got a SIGBUS because we tried to write into the code
+ // cache, try enabling WXWrite mode.
+ if (sig == SIGBUS
+ && pc != info->si_addr
+ && CodeCache::contains(info->si_addr)
+ && os::address_is_in_vm(pc)) {
+ WXMode *entry_mode = thread->_cur_wx_mode;
+ if (entry_mode != nullptr && *entry_mode == WXArmedForWrite) {
+ if (TraceWXHealing) {
+ static const char *mode_names[3] = {"WXWrite", "WXExec", "WXArmedForWrite"};
+ tty->print("Healing WXMode %s at %p to WXWrite",
+ mode_names[*entry_mode], entry_mode);
+ char name[128];
+ int offset = 0;
+ if (os::dll_address_to_function_name(pc, name, sizeof name, &offset)) {
+ tty->print_cr(" (%s+0x%x)", name, offset);
+ } else {
+ tty->cr();
+ }
+ if (Verbose) {
+ char buf[O_BUFLEN];
+ NativeStackPrinter nsp(thread);
+ nsp.print_stack(tty, buf, sizeof(buf), pc,
+ true /* print_source_info */, -1 /* max stack */);
+ }
+ }
+#ifndef PRODUCT
+ guarantee(StressWXHealing,
+ "We should not reach here unless StressWXHealing");
+#endif
+ *(thread->_cur_wx_mode) = WXWrite;
+ return thread->wx_enable_write();
+ }
+ }
+
+ // There may be cases where code after this point that we call
+ // from the signal handler changes WX state, so we protect against
+ // that by saving and restoring the state.
+ ThreadWXEnable wx(thread->get_wx_state(), thread);
+#endif
+
// Handle ALL stack overflow variations here
if (sig == SIGSEGV || sig == SIGBUS) {
address addr = (address) info->si_addr;
@@ -515,11 +557,42 @@ int os::extra_bang_size_in_bytes() {
return 0;
}
-#ifdef __APPLE__
+#ifdef MACOS_AARCH64
+THREAD_LOCAL bool os::_jit_exec_enabled;
+
+// This is a wrapper around the standard library function
+// pthread_jit_write_protect_np(3). We keep track of the state of
+// per-thread write protection on the MAP_JIT region in the
+// thread-local variable os::_jit_exec_enabled
void os::current_thread_enable_wx(WXMode mode) {
- pthread_jit_write_protect_np(mode == WXExec);
+ bool exec_enabled = mode != WXWrite;
+ if (exec_enabled != _jit_exec_enabled NOT_PRODUCT( || DefaultWXWriteMode == WXWrite)) {
+ permit_forbidden_function::pthread_jit_write_protect_np(exec_enabled);
+ _jit_exec_enabled = exec_enabled;
+ }
}
-#endif
+
+// If the current thread is in the WX state WXArmedForWrite, change
+// the state to WXWrite.
+bool Thread::wx_enable_write() {
+ if (_wx_state == WXArmedForWrite) {
+ _wx_state = WXWrite;
+ os::current_thread_enable_wx(WXWrite);
+ return true;
+ } else {
+ return false;
+ }
+}
+
+// A wrapper around wx_enable_write() for when the current thread is
+// not known.
+void os::thread_wx_enable_write_impl() {
+ if (!StressWXHealing) {
+ Thread::current()->wx_enable_write();
+ }
+}
+
+#endif // MACOS_AARCH64
static inline void atomic_copy64(const volatile void *src, volatile void *dst) {
*(jlong *) dst = *(const jlong *) src;
diff --git a/src/hotspot/os_cpu/linux_arm/javaThread_linux_arm.cpp b/src/hotspot/os_cpu/linux_arm/javaThread_linux_arm.cpp
index 3dc0035ed87..2b96e978980 100644
--- a/src/hotspot/os_cpu/linux_arm/javaThread_linux_arm.cpp
+++ b/src/hotspot/os_cpu/linux_arm/javaThread_linux_arm.cpp
@@ -42,8 +42,19 @@ frame JavaThread::pd_last_frame() {
void JavaThread::cache_global_variables() {
BarrierSet* bs = BarrierSet::barrier_set();
+#if INCLUDE_G1GC
+ if (bs->is_a(BarrierSet::G1BarrierSet)) {
+ _card_table_base = nullptr;
+ } else
+#endif
+#if INCLUDE_SHENANDOAHGC
+ if (bs->is_a(BarrierSet::ShenandoahBarrierSet)) {
+ _card_table_base = nullptr;
+ } else
+#endif
if (bs->is_a(BarrierSet::CardTableBarrierSet)) {
- _card_table_base = (address) (barrier_set_cast(bs)->card_table()->byte_map_base());
+ CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
+ _card_table_base = (address)ctbs->card_table_base_const();
} else {
_card_table_base = nullptr;
}
diff --git a/src/hotspot/os_cpu/linux_x86/os_linux_x86.cpp b/src/hotspot/os_cpu/linux_x86/os_linux_x86.cpp
index 07f53582a76..ee08738c678 100644
--- a/src/hotspot/os_cpu/linux_x86/os_linux_x86.cpp
+++ b/src/hotspot/os_cpu/linux_x86/os_linux_x86.cpp
@@ -52,6 +52,7 @@
#include "utilities/debug.hpp"
#include "utilities/events.hpp"
#include "utilities/vmError.hpp"
+#include "runtime/vm_version.hpp"
// put OS-includes here
# include
@@ -380,6 +381,43 @@ size_t os::Posix::default_stack_size(os::ThreadType thr_type) {
/////////////////////////////////////////////////////////////////////////////
// helper functions for fatal error handler
+// XSAVE constants - from Intel SDM Vol. 1, Chapter 13
+#define XSAVE_HDR_OFFSET 512
+#define XFEATURE_APX (1ULL << 19)
+
+// XSAVE header structure
+// See: Intel SDM Vol. 1, Section 13.4.2 "XSAVE Header"
+// Also: Linux kernel arch/x86/include/asm/fpu/types.h
+struct xstate_header {
+ uint64_t xfeatures;
+ uint64_t xcomp_bv;
+ uint64_t reserved[6];
+};
+
+// APX extended state - R16-R31 (16 x 64-bit registers)
+// See: Intel APX Architecture Specification
+struct apx_state {
+ uint64_t regs[16]; // r16-r31
+};
+
+static apx_state* get_apx_state(const ucontext_t* uc) {
+ uint32_t offset = VM_Version::apx_xstate_offset();
+ if (offset == 0 || uc->uc_mcontext.fpregs == nullptr) {
+ return nullptr;
+ }
+
+ char* xsave = (char*)uc->uc_mcontext.fpregs;
+ xstate_header* hdr = (xstate_header*)(xsave + XSAVE_HDR_OFFSET);
+
+ // Check if APX state is present in this context
+ if (!(hdr->xfeatures & XFEATURE_APX)) {
+ return nullptr;
+ }
+
+ return (apx_state*)(xsave + offset);
+}
+
+
void os::print_context(outputStream *st, const void *context) {
if (context == nullptr) return;
@@ -406,6 +444,14 @@ void os::print_context(outputStream *st, const void *context) {
st->print(", R14=" INTPTR_FORMAT, (intptr_t)uc->uc_mcontext.gregs[REG_R14]);
st->print(", R15=" INTPTR_FORMAT, (intptr_t)uc->uc_mcontext.gregs[REG_R15]);
st->cr();
+ // Dump APX EGPRs (R16-R31)
+ apx_state* apx = UseAPX ? get_apx_state(uc) : nullptr;
+ if (apx != nullptr) {
+ for (int i = 0; i < 16; i++) {
+ st->print("%sR%d=" INTPTR_FORMAT, (i % 4 == 0) ? "" : ", ", 16 + i, (intptr_t)apx->regs[i]);
+ if (i % 4 == 3) st->cr();
+ }
+ }
st->print( "RIP=" INTPTR_FORMAT, (intptr_t)uc->uc_mcontext.gregs[REG_RIP]);
st->print(", EFLAGS=" INTPTR_FORMAT, (intptr_t)uc->uc_mcontext.gregs[REG_EFL]);
st->print(", CSGSFS=" INTPTR_FORMAT, (intptr_t)uc->uc_mcontext.gregs[REG_CSGSFS]);
@@ -432,37 +478,50 @@ void os::print_context(outputStream *st, const void *context) {
}
void os::print_register_info(outputStream *st, const void *context, int& continuation) {
- const int register_count = 16;
+ if (context == nullptr) {
+ return;
+ }
+ const ucontext_t *uc = (const ucontext_t*)context;
+ apx_state* apx = UseAPX ? get_apx_state(uc) : nullptr;
+
+ const int register_count = 16 + (apx != nullptr ? 16 : 0);
int n = continuation;
assert(n >= 0 && n <= register_count, "Invalid continuation value");
- if (context == nullptr || n == register_count) {
+ if (n == register_count) {
return;
}
- const ucontext_t *uc = (const ucontext_t*)context;
while (n < register_count) {
// Update continuation with next index before printing location
continuation = n + 1;
+
+ if (n < 16) {
+ // Standard registers (RAX-R15)
# define CASE_PRINT_REG(n, str, id) case n: st->print(str); print_location(st, uc->uc_mcontext.gregs[REG_##id]);
- switch (n) {
- CASE_PRINT_REG( 0, "RAX=", RAX); break;
- CASE_PRINT_REG( 1, "RBX=", RBX); break;
- CASE_PRINT_REG( 2, "RCX=", RCX); break;
- CASE_PRINT_REG( 3, "RDX=", RDX); break;
- CASE_PRINT_REG( 4, "RSP=", RSP); break;
- CASE_PRINT_REG( 5, "RBP=", RBP); break;
- CASE_PRINT_REG( 6, "RSI=", RSI); break;
- CASE_PRINT_REG( 7, "RDI=", RDI); break;
- CASE_PRINT_REG( 8, "R8 =", R8); break;
- CASE_PRINT_REG( 9, "R9 =", R9); break;
- CASE_PRINT_REG(10, "R10=", R10); break;
- CASE_PRINT_REG(11, "R11=", R11); break;
- CASE_PRINT_REG(12, "R12=", R12); break;
- CASE_PRINT_REG(13, "R13=", R13); break;
- CASE_PRINT_REG(14, "R14=", R14); break;
- CASE_PRINT_REG(15, "R15=", R15); break;
- }
+ switch (n) {
+ CASE_PRINT_REG( 0, "RAX=", RAX); break;
+ CASE_PRINT_REG( 1, "RBX=", RBX); break;
+ CASE_PRINT_REG( 2, "RCX=", RCX); break;
+ CASE_PRINT_REG( 3, "RDX=", RDX); break;
+ CASE_PRINT_REG( 4, "RSP=", RSP); break;
+ CASE_PRINT_REG( 5, "RBP=", RBP); break;
+ CASE_PRINT_REG( 6, "RSI=", RSI); break;
+ CASE_PRINT_REG( 7, "RDI=", RDI); break;
+ CASE_PRINT_REG( 8, "R8 =", R8); break;
+ CASE_PRINT_REG( 9, "R9 =", R9); break;
+ CASE_PRINT_REG(10, "R10=", R10); break;
+ CASE_PRINT_REG(11, "R11=", R11); break;
+ CASE_PRINT_REG(12, "R12=", R12); break;
+ CASE_PRINT_REG(13, "R13=", R13); break;
+ CASE_PRINT_REG(14, "R14=", R14); break;
+ CASE_PRINT_REG(15, "R15=", R15); break;
+ }
# undef CASE_PRINT_REG
+ } else {
+ // APX extended general purpose registers (R16-R31)
+ st->print("R%d=", n);
+ print_location(st, apx->regs[n - 16]);
+ }
++n;
}
}
diff --git a/src/hotspot/share/adlc/adlArena.cpp b/src/hotspot/share/adlc/adlArena.cpp
index ebd1f74911d..e3ae60e91a9 100644
--- a/src/hotspot/share/adlc/adlArena.cpp
+++ b/src/hotspot/share/adlc/adlArena.cpp
@@ -136,9 +136,9 @@ void *AdlArena::Acalloc( size_t items, size_t x ) {
}
//------------------------------realloc----------------------------------------
-static size_t pointer_delta(const void *left, const void *right) {
- assert(left >= right, "pointer delta underflow");
- return (uintptr_t)left - (uintptr_t)right;
+static size_t pointer_delta(const void* high, const void* low) {
+ assert(high >= low, "pointer delta underflow");
+ return (uintptr_t)high - (uintptr_t)low;
}
// Reallocate storage in AdlArena.
diff --git a/src/hotspot/share/asm/codeBuffer.cpp b/src/hotspot/share/asm/codeBuffer.cpp
index 7871134e923..ba525588f32 100644
--- a/src/hotspot/share/asm/codeBuffer.cpp
+++ b/src/hotspot/share/asm/codeBuffer.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -98,6 +98,8 @@ CodeBuffer::CodeBuffer(const CodeBlob* blob) DEBUG_ONLY(: Scrubber(this, sizeof(
}
void CodeBuffer::initialize(csize_t code_size, csize_t locs_size) {
+ MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
+
// Always allow for empty slop around each section.
int slop = (int) CodeSection::end_slop();
@@ -466,9 +468,7 @@ void CodeBuffer::compute_final_layout(CodeBuffer* dest) const {
assert(!_finalize_stubs, "non-finalized stubs");
{
- // not sure why this is here, but why not...
- int alignSize = MAX2((intx) sizeof(jdouble), CodeEntryAlignment);
- assert( (dest->_total_start - _insts.start()) % alignSize == 0, "copy must preserve alignment");
+ assert( (dest->_total_start - _insts.start()) % CodeEntryAlignment == 0, "copy must preserve alignment");
}
const CodeSection* prev_cs = nullptr;
diff --git a/src/hotspot/share/c1/c1_Runtime1.cpp b/src/hotspot/share/c1/c1_Runtime1.cpp
index a4c956ff5be..63764dd113a 100644
--- a/src/hotspot/share/c1/c1_Runtime1.cpp
+++ b/src/hotspot/share/c1/c1_Runtime1.cpp
@@ -541,6 +541,7 @@ extern void vm_exit(int code);
// unpack_with_exception entry instead. This makes life for the exception blob easier
// because making that same check and diverting is painful from assembly language.
JRT_ENTRY_NO_ASYNC(static address, exception_handler_for_pc_helper(JavaThread* current, oopDesc* ex, address pc, nmethod*& nm))
+ MACOS_AARCH64_ONLY(current->wx_enable_write());
Handle exception(current, ex);
// This function is called when we are about to throw an exception. Therefore,
diff --git a/src/hotspot/share/cds/aotMapLogger.cpp b/src/hotspot/share/cds/aotMapLogger.cpp
index 5e4e0956824..fa769aee1bf 100644
--- a/src/hotspot/share/cds/aotMapLogger.cpp
+++ b/src/hotspot/share/cds/aotMapLogger.cpp
@@ -88,7 +88,7 @@ void AOTMapLogger::ergo_initialize() {
}
void AOTMapLogger::dumptime_log(ArchiveBuilder* builder, FileMapInfo* mapinfo,
- ArchiveMappedHeapInfo* mapped_heap_info, ArchiveStreamedHeapInfo* streamed_heap_info,
+ AOTMappedHeapInfo* mapped_heap_info, AOTStreamedHeapInfo* streamed_heap_info,
char* bitmap, size_t bitmap_size_in_bytes) {
_is_runtime_logging = false;
_buffer_to_requested_delta = ArchiveBuilder::current()->buffer_to_requested_delta();
@@ -823,7 +823,7 @@ public:
}
}; // AOTMapLogger::ArchivedFieldPrinter
-void AOTMapLogger::dumptime_log_mapped_heap_region(ArchiveMappedHeapInfo* heap_info) {
+void AOTMapLogger::dumptime_log_mapped_heap_region(AOTMappedHeapInfo* heap_info) {
MemRegion r = heap_info->buffer_region();
address buffer_start = address(r.start()); // start of the current oop inside the buffer
address buffer_end = address(r.end());
@@ -835,7 +835,7 @@ void AOTMapLogger::dumptime_log_mapped_heap_region(ArchiveMappedHeapInfo* heap_i
log_archived_objects(AOTMappedHeapWriter::oop_iterator(heap_info));
}
-void AOTMapLogger::dumptime_log_streamed_heap_region(ArchiveStreamedHeapInfo* heap_info) {
+void AOTMapLogger::dumptime_log_streamed_heap_region(AOTStreamedHeapInfo* heap_info) {
MemRegion r = heap_info->buffer_region();
address buffer_start = address(r.start()); // start of the current oop inside the buffer
address buffer_end = address(r.end());
diff --git a/src/hotspot/share/cds/aotMapLogger.hpp b/src/hotspot/share/cds/aotMapLogger.hpp
index bf7ce0028b9..f495ed97f40 100644
--- a/src/hotspot/share/cds/aotMapLogger.hpp
+++ b/src/hotspot/share/cds/aotMapLogger.hpp
@@ -33,8 +33,8 @@
#include "utilities/globalDefinitions.hpp"
#include "utilities/growableArray.hpp"
-class ArchiveMappedHeapInfo;
-class ArchiveStreamedHeapInfo;
+class AOTMappedHeapInfo;
+class AOTStreamedHeapInfo;
class CompileTrainingData;
class DumpRegion;
class FileMapInfo;
@@ -157,8 +157,8 @@ private:
#if INCLUDE_CDS_JAVA_HEAP
- static void dumptime_log_mapped_heap_region(ArchiveMappedHeapInfo* mapped_heap_info);
- static void dumptime_log_streamed_heap_region(ArchiveStreamedHeapInfo* streamed_heap_info);
+ static void dumptime_log_mapped_heap_region(AOTMappedHeapInfo* mapped_heap_info);
+ static void dumptime_log_streamed_heap_region(AOTStreamedHeapInfo* streamed_heap_info);
static void runtime_log_heap_region(FileMapInfo* mapinfo);
static void print_oop_info_cr(outputStream* st, FakeOop fake_oop, bool print_location = true);
@@ -173,7 +173,7 @@ public:
static bool is_logging_at_bootstrap() { return _is_logging_at_bootstrap; }
static void dumptime_log(ArchiveBuilder* builder, FileMapInfo* mapinfo,
- ArchiveMappedHeapInfo* mapped_heap_info, ArchiveStreamedHeapInfo* streamed_heap_info,
+ AOTMappedHeapInfo* mapped_heap_info, AOTStreamedHeapInfo* streamed_heap_info,
char* bitmap, size_t bitmap_size_in_bytes);
static void runtime_log(FileMapInfo* static_mapinfo, FileMapInfo* dynamic_mapinfo);
};
diff --git a/src/hotspot/share/cds/aotMappedHeap.cpp b/src/hotspot/share/cds/aotMappedHeap.cpp
new file mode 100644
index 00000000000..ba24c43eea1
--- /dev/null
+++ b/src/hotspot/share/cds/aotMappedHeap.cpp
@@ -0,0 +1,49 @@
+/*
+ * Copyright (c) 2026, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+#include "cds/aotMappedHeap.hpp"
+
+// Anything that goes in the header must be thoroughly purged from uninitialized memory
+// as it will be written to disk. Therefore, the constructors memset the memory to 0.
+// This is not the prettiest thing, but we need to know every byte is initialized,
+// including potential padding between fields.
+
+AOTMappedHeapHeader::AOTMappedHeapHeader(size_t ptrmap_start_pos,
+ size_t oopmap_start_pos,
+ HeapRootSegments root_segments) {
+ memset((char*)this, 0, sizeof(*this));
+ _ptrmap_start_pos = ptrmap_start_pos;
+ _oopmap_start_pos = oopmap_start_pos;
+ _root_segments = root_segments;
+}
+
+AOTMappedHeapHeader::AOTMappedHeapHeader() {
+ memset((char*)this, 0, sizeof(*this));
+}
+
+AOTMappedHeapHeader AOTMappedHeapInfo::create_header() {
+ return AOTMappedHeapHeader{_ptrmap_start_pos,
+ _oopmap_start_pos,
+ _root_segments};
+}
diff --git a/src/hotspot/share/cds/aotMappedHeap.hpp b/src/hotspot/share/cds/aotMappedHeap.hpp
new file mode 100644
index 00000000000..307451b24d4
--- /dev/null
+++ b/src/hotspot/share/cds/aotMappedHeap.hpp
@@ -0,0 +1,168 @@
+/*
+ * Copyright (c) 2026, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+#ifndef SHARE_CDS_AOTMAPPEDHEAP_HPP
+#define SHARE_CDS_AOTMAPPEDHEAP_HPP
+
+#include "cds/aotMapLogger.hpp"
+#include "utilities/growableArray.hpp"
+#include "utilities/macros.hpp"
+
+class AOTMappedHeapHeader {
+ size_t _ptrmap_start_pos; // The first bit in the ptrmap corresponds to this position in the heap.
+ size_t _oopmap_start_pos; // The first bit in the oopmap corresponds to this position in the heap.
+ HeapRootSegments _root_segments; // Heap root segments info
+
+public:
+ AOTMappedHeapHeader();
+ AOTMappedHeapHeader(size_t ptrmap_start_pos,
+ size_t oopmap_start_pos,
+ HeapRootSegments root_segments);
+
+ size_t ptrmap_start_pos() const { return _ptrmap_start_pos; }
+ size_t oopmap_start_pos() const { return _oopmap_start_pos; }
+ HeapRootSegments root_segments() const { return _root_segments; }
+
+ // This class is trivially copyable and assignable.
+ AOTMappedHeapHeader(const AOTMappedHeapHeader&) = default;
+ AOTMappedHeapHeader& operator=(const AOTMappedHeapHeader&) = default;
+};
+
+class AOTMappedHeapInfo {
+ MemRegion _buffer_region; // Contains the archived objects to be written into the CDS archive.
+ CHeapBitMap _oopmap;
+ CHeapBitMap _ptrmap;
+ HeapRootSegments _root_segments;
+ size_t _oopmap_start_pos; // How many zeros were removed from the beginning of the bit map?
+ size_t _ptrmap_start_pos; // How many zeros were removed from the beginning of the bit map?
+
+public:
+ AOTMappedHeapInfo() :
+ _buffer_region(),
+ _oopmap(128, mtClassShared),
+ _ptrmap(128, mtClassShared),
+ _root_segments(),
+ _oopmap_start_pos(),
+ _ptrmap_start_pos() {}
+ bool is_used() { return !_buffer_region.is_empty(); }
+
+ MemRegion buffer_region() { return _buffer_region; }
+ void set_buffer_region(MemRegion r) { _buffer_region = r; }
+
+ char* buffer_start() { return (char*)_buffer_region.start(); }
+ size_t buffer_byte_size() { return _buffer_region.byte_size(); }
+
+ CHeapBitMap* oopmap() { return &_oopmap; }
+ CHeapBitMap* ptrmap() { return &_ptrmap; }
+
+ void set_oopmap_start_pos(size_t start_pos) { _oopmap_start_pos = start_pos; }
+ void set_ptrmap_start_pos(size_t start_pos) { _ptrmap_start_pos = start_pos; }
+
+ void set_root_segments(HeapRootSegments segments) { _root_segments = segments; };
+ HeapRootSegments root_segments() { return _root_segments; }
+
+ AOTMappedHeapHeader create_header();
+};
+
+#if INCLUDE_CDS_JAVA_HEAP
+class AOTMappedHeapOopIterator : public AOTMapLogger::OopDataIterator {
+protected:
+ address _current;
+ address _next;
+
+ address _buffer_start;
+ address _buffer_end;
+ uint64_t _buffer_start_narrow_oop;
+ intptr_t _buffer_to_requested_delta;
+ int _requested_shift;
+
+ size_t _num_root_segments;
+ size_t _num_obj_arrays_logged;
+
+public:
+ AOTMappedHeapOopIterator(address buffer_start,
+ address buffer_end,
+ address requested_base,
+ address requested_start,
+ int requested_shift,
+ size_t num_root_segments)
+ : _current(nullptr),
+ _next(buffer_start),
+ _buffer_start(buffer_start),
+ _buffer_end(buffer_end),
+ _requested_shift(requested_shift),
+ _num_root_segments(num_root_segments),
+ _num_obj_arrays_logged(0) {
+ _buffer_to_requested_delta = requested_start - buffer_start;
+ _buffer_start_narrow_oop = 0xdeadbeed;
+ if (UseCompressedOops) {
+ _buffer_start_narrow_oop = (uint64_t)(pointer_delta(requested_start, requested_base, 1)) >> requested_shift;
+ assert(_buffer_start_narrow_oop < 0xffffffff, "sanity");
+ }
+ }
+
+ virtual AOTMapLogger::OopData capture(address buffered_addr) = 0;
+
+ bool has_next() override {
+ return _next < _buffer_end;
+ }
+
+ AOTMapLogger::OopData next() override {
+ _current = _next;
+ AOTMapLogger::OopData result = capture(_current);
+ if (result._klass->is_objArray_klass()) {
+ result._is_root_segment = _num_obj_arrays_logged++ < _num_root_segments;
+ }
+ _next = _current + result._size * BytesPerWord;
+ return result;
+ }
+
+ AOTMapLogger::OopData obj_at(narrowOop* addr) override {
+ uint64_t n = (uint64_t)(*addr);
+ if (n == 0) {
+ return null_data();
+ } else {
+ precond(n >= _buffer_start_narrow_oop);
+ address buffer_addr = _buffer_start + ((n - _buffer_start_narrow_oop) << _requested_shift);
+ return capture(buffer_addr);
+ }
+ }
+
+ AOTMapLogger::OopData obj_at(oop* addr) override {
+ address requested_value = cast_from_oop(*addr);
+ if (requested_value == nullptr) {
+ return null_data();
+ } else {
+ address buffer_addr = requested_value - _buffer_to_requested_delta;
+ return capture(buffer_addr);
+ }
+ }
+
+ GrowableArrayCHeap* roots() override {
+ return new GrowableArrayCHeap();
+ }
+};
+#endif // INCLUDE_CDS_JAVA_HEAP
+
+#endif // SHARE_CDS_AOTMAPPEDHEAP_HPP
diff --git a/src/hotspot/share/cds/aotMappedHeapLoader.cpp b/src/hotspot/share/cds/aotMappedHeapLoader.cpp
index 84051cbd9e5..7a201d8297f 100644
--- a/src/hotspot/share/cds/aotMappedHeapLoader.cpp
+++ b/src/hotspot/share/cds/aotMappedHeapLoader.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2018, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -23,6 +23,7 @@
*/
#include "cds/aotLogging.hpp"
+#include "cds/aotMappedHeap.hpp"
#include "cds/aotMappedHeapLoader.inline.hpp"
#include "cds/aotMappedHeapWriter.hpp"
#include "cds/aotMetaspace.hpp"
@@ -221,7 +222,7 @@ void AOTMappedHeapLoader::patch_embedded_pointers(FileMapInfo* info,
// the heap object may be loaded at a different address at run time. This structure is used
// to translate the dump time addresses for all objects in FileMapInfo::space_at(region_index)
// to their runtime addresses.
-struct LoadedArchiveHeapRegion {
+struct AOTMappedHeapRegion {
int _region_index; // index for FileMapInfo::space_at(index)
size_t _region_size; // number of bytes in this region
uintptr_t _dumptime_base; // The dump-time (decoded) address of the first object in this region
@@ -232,7 +233,7 @@ struct LoadedArchiveHeapRegion {
}
};
-void AOTMappedHeapLoader::init_loaded_heap_relocation(LoadedArchiveHeapRegion* loaded_region) {
+void AOTMappedHeapLoader::init_loaded_heap_relocation(AOTMappedHeapRegion* loaded_region) {
_dumptime_base = loaded_region->_dumptime_base;
_dumptime_top = loaded_region->top();
_runtime_offset = loaded_region->_runtime_offset;
@@ -249,7 +250,7 @@ class AOTMappedHeapLoader::PatchLoadedRegionPointers: public BitMapClosure {
uintptr_t _top;
public:
- PatchLoadedRegionPointers(narrowOop* start, LoadedArchiveHeapRegion* loaded_region)
+ PatchLoadedRegionPointers(narrowOop* start, AOTMappedHeapRegion* loaded_region)
: _start(start),
_offset(loaded_region->_runtime_offset),
_base(loaded_region->_dumptime_base),
@@ -270,7 +271,7 @@ class AOTMappedHeapLoader::PatchLoadedRegionPointers: public BitMapClosure {
}
};
-bool AOTMappedHeapLoader::init_loaded_region(FileMapInfo* mapinfo, LoadedArchiveHeapRegion* loaded_region,
+bool AOTMappedHeapLoader::init_loaded_region(FileMapInfo* mapinfo, AOTMappedHeapRegion* loaded_region,
MemRegion& archive_space) {
size_t total_bytes = 0;
FileMapRegion* r = mapinfo->region_at(AOTMetaspace::hp);
@@ -301,7 +302,7 @@ bool AOTMappedHeapLoader::init_loaded_region(FileMapInfo* mapinfo, LoadedArchive
return true;
}
-bool AOTMappedHeapLoader::load_heap_region_impl(FileMapInfo* mapinfo, LoadedArchiveHeapRegion* loaded_region,
+bool AOTMappedHeapLoader::load_heap_region_impl(FileMapInfo* mapinfo, AOTMappedHeapRegion* loaded_region,
uintptr_t load_address) {
uintptr_t bitmap_base = (uintptr_t)mapinfo->map_bitmap_region();
if (bitmap_base == 0) {
@@ -340,7 +341,7 @@ bool AOTMappedHeapLoader::load_heap_region(FileMapInfo* mapinfo) {
assert(can_load(), "loaded heap for must be supported");
init_narrow_oop_decoding(mapinfo->narrow_oop_base(), mapinfo->narrow_oop_shift());
- LoadedArchiveHeapRegion loaded_region;
+ AOTMappedHeapRegion loaded_region;
memset(&loaded_region, 0, sizeof(loaded_region));
MemRegion archive_space;
@@ -360,10 +361,8 @@ bool AOTMappedHeapLoader::load_heap_region(FileMapInfo* mapinfo) {
}
objArrayOop AOTMappedHeapLoader::root_segment(int segment_idx) {
- if (CDSConfig::is_dumping_heap()) {
- assert(Thread::current() == (Thread*)VMThread::vm_thread(), "should be in vm thread");
- } else {
- assert(CDSConfig::is_using_archive(), "must be");
+ if (!CDSConfig::is_using_archive()) {
+ assert(CDSConfig::is_dumping_heap() && Thread::current() == (Thread*)VMThread::vm_thread(), "sanity");
}
objArrayOop segment = (objArrayOop)_root_segments->at(segment_idx).resolve();
@@ -466,7 +465,9 @@ void AOTMappedHeapLoader::finish_initialization(FileMapInfo* info) {
add_root_segment((objArrayOop)segment_oop);
}
- StringTable::load_shared_strings_array();
+ if (CDSConfig::is_dumping_final_static_archive()) {
+ StringTable::move_shared_strings_into_runtime_table();
+ }
}
}
@@ -619,7 +620,7 @@ bool AOTMappedHeapLoader::map_heap_region_impl(FileMapInfo* info) {
aot_log_info(aot)("Preferred address to map heap data (to avoid relocation) is " INTPTR_FORMAT, p2i(requested_start));
// allocate from java heap
- HeapWord* start = G1CollectedHeap::heap()->alloc_archive_region(word_size, (HeapWord*)requested_start);
+ HeapWord* start = G1CollectedHeap::heap()->alloc_archive_region(word_size);
if (start == nullptr) {
AOTMetaspace::report_loading_error("UseSharedSpaces: Unable to allocate java heap region for archive heap.");
return false;
@@ -733,40 +734,22 @@ void AOTMappedHeapLoader::dealloc_heap_region(FileMapInfo* info) {
}
AOTMapLogger::OopDataIterator* AOTMappedHeapLoader::oop_iterator(FileMapInfo* info, address buffer_start, address buffer_end) {
- class MappedLoaderOopIterator : public AOTMapLogger::OopDataIterator {
- private:
- address _current;
- address _next;
-
- address _buffer_start;
- address _buffer_end;
- uint64_t _buffer_start_narrow_oop;
- intptr_t _buffer_to_requested_delta;
- int _requested_shift;
-
- size_t _num_root_segments;
- size_t _num_obj_arrays_logged;
-
+ class MappedLoaderOopIterator : public AOTMappedHeapOopIterator {
public:
MappedLoaderOopIterator(address buffer_start,
address buffer_end,
- uint64_t buffer_start_narrow_oop,
- intptr_t buffer_to_requested_delta,
+ address requested_base,
+ address requested_start,
int requested_shift,
- size_t num_root_segments)
- : _current(nullptr),
- _next(buffer_start),
- _buffer_start(buffer_start),
- _buffer_end(buffer_end),
- _buffer_start_narrow_oop(buffer_start_narrow_oop),
- _buffer_to_requested_delta(buffer_to_requested_delta),
- _requested_shift(requested_shift),
- _num_root_segments(num_root_segments),
- _num_obj_arrays_logged(0) {
- }
+ size_t num_root_segments) :
+ AOTMappedHeapOopIterator(buffer_start,
+ buffer_end,
+ requested_base,
+ requested_start,
+ requested_shift,
+ num_root_segments) {}
-
- AOTMapLogger::OopData capture(address buffered_addr) {
+ AOTMapLogger::OopData capture(address buffered_addr) override {
oopDesc* raw_oop = (oopDesc*)buffered_addr;
size_t size = raw_oop->size();
address requested_addr = buffered_addr + _buffer_to_requested_delta;
@@ -784,62 +767,17 @@ AOTMapLogger::OopDataIterator* AOTMappedHeapLoader::oop_iterator(FileMapInfo* in
size,
false };
}
-
- bool has_next() override {
- return _next < _buffer_end;
- }
-
- AOTMapLogger::OopData next() override {
- _current = _next;
- AOTMapLogger::OopData result = capture(_current);
- if (result._klass->is_objArray_klass()) {
- result._is_root_segment = _num_obj_arrays_logged++ < _num_root_segments;
- }
- _next = _current + result._size * BytesPerWord;
- return result;
- }
-
- AOTMapLogger::OopData obj_at(narrowOop* addr) override {
- uint64_t n = (uint64_t)(*addr);
- if (n == 0) {
- return null_data();
- } else {
- precond(n >= _buffer_start_narrow_oop);
- address buffer_addr = _buffer_start + ((n - _buffer_start_narrow_oop) << _requested_shift);
- return capture(buffer_addr);
- }
- }
-
- AOTMapLogger::OopData obj_at(oop* addr) override {
- address requested_value = cast_from_oop(*addr);
- if (requested_value == nullptr) {
- return null_data();
- } else {
- address buffer_addr = requested_value - _buffer_to_requested_delta;
- return capture(buffer_addr);
- }
- }
-
- GrowableArrayCHeap* roots() override {
- return new GrowableArrayCHeap();
- }
};
FileMapRegion* r = info->region_at(AOTMetaspace::hp);
address requested_base = UseCompressedOops ? (address)info->narrow_oop_base() : heap_region_requested_address(info);
address requested_start = requested_base + r->mapping_offset();
int requested_shift = info->narrow_oop_shift();
- intptr_t buffer_to_requested_delta = requested_start - buffer_start;
- uint64_t buffer_start_narrow_oop = 0xdeadbeed;
- if (UseCompressedOops) {
- buffer_start_narrow_oop = (uint64_t)(pointer_delta(requested_start, requested_base, 1)) >> requested_shift;
- assert(buffer_start_narrow_oop < 0xffffffff, "sanity");
- }
return new MappedLoaderOopIterator(buffer_start,
buffer_end,
- buffer_start_narrow_oop,
- buffer_to_requested_delta,
+ requested_base,
+ requested_start,
requested_shift,
info->mapped_heap()->root_segments().count());
}
diff --git a/src/hotspot/share/cds/aotMappedHeapLoader.hpp b/src/hotspot/share/cds/aotMappedHeapLoader.hpp
index d344d7b0b0a..7c5ca1b1f9e 100644
--- a/src/hotspot/share/cds/aotMappedHeapLoader.hpp
+++ b/src/hotspot/share/cds/aotMappedHeapLoader.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2018, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -37,8 +37,8 @@
#include "utilities/growableArray.hpp"
#include "utilities/macros.hpp"
+struct AOTMappedHeapRegion;
class FileMapInfo;
-struct LoadedArchiveHeapRegion;
class AOTMappedHeapLoader : AllStatic {
friend class AOTMapLogger;
@@ -93,7 +93,7 @@ public:
// function instead.
inline static oop decode_from_archive(narrowOop v) NOT_CDS_JAVA_HEAP_RETURN_(nullptr);
- // More efficient version, but works only when ArchiveHeap is mapped.
+ // More efficient version, but works only when is_mapped()
inline static oop decode_from_mapped_archive(narrowOop v) NOT_CDS_JAVA_HEAP_RETURN_(nullptr);
static void patch_compressed_embedded_pointers(BitMapView bm,
@@ -113,7 +113,7 @@ private:
static bool _is_loaded;
// Support for loaded archived heap. These are cached values from
- // LoadedArchiveHeapRegion's.
+ // AOTMappedHeapRegion's.
static uintptr_t _dumptime_base;
static uintptr_t _dumptime_top;
static intx _runtime_offset;
@@ -141,10 +141,10 @@ private:
static bool _heap_pointers_need_patching;
static void init_narrow_oop_decoding(address base, int shift);
- static bool init_loaded_region(FileMapInfo* mapinfo, LoadedArchiveHeapRegion* loaded_region,
+ static bool init_loaded_region(FileMapInfo* mapinfo, AOTMappedHeapRegion* loaded_region,
MemRegion& archive_space);
- static bool load_heap_region_impl(FileMapInfo* mapinfo, LoadedArchiveHeapRegion* loaded_region, uintptr_t buffer);
- static void init_loaded_heap_relocation(LoadedArchiveHeapRegion* reloc_info);
+ static bool load_heap_region_impl(FileMapInfo* mapinfo, AOTMappedHeapRegion* loaded_region, uintptr_t buffer);
+ static void init_loaded_heap_relocation(AOTMappedHeapRegion* reloc_info);
static void patch_native_pointers();
static void finish_loaded_heap();
static void verify_loaded_heap();
diff --git a/src/hotspot/share/cds/aotMappedHeapWriter.cpp b/src/hotspot/share/cds/aotMappedHeapWriter.cpp
index e73b980614a..64c0e3c40e8 100644
--- a/src/hotspot/share/cds/aotMappedHeapWriter.cpp
+++ b/src/hotspot/share/cds/aotMappedHeapWriter.cpp
@@ -22,7 +22,7 @@
*
*/
-#include "cds/aotMappedHeapLoader.hpp"
+#include "cds/aotMappedHeap.hpp"
#include "cds/aotMappedHeapWriter.hpp"
#include "cds/aotReferenceObjSupport.hpp"
#include "cds/cdsConfig.hpp"
@@ -151,7 +151,7 @@ void AOTMappedHeapWriter::add_source_obj(oop src_obj) {
}
void AOTMappedHeapWriter::write(GrowableArrayCHeap* roots,
- ArchiveMappedHeapInfo* heap_info) {
+ AOTMappedHeapInfo* heap_info) {
assert(CDSConfig::is_dumping_heap(), "sanity");
allocate_buffer();
copy_source_objs_to_buffer(roots);
@@ -598,7 +598,7 @@ size_t AOTMappedHeapWriter::copy_one_source_obj_to_buffer(oop src_obj) {
//
// So we just hard code it to NOCOOPS_REQUESTED_BASE.
//
-void AOTMappedHeapWriter::set_requested_address_range(ArchiveMappedHeapInfo* info) {
+void AOTMappedHeapWriter::set_requested_address_range(AOTMappedHeapInfo* info) {
assert(!info->is_used(), "only set once");
size_t heap_region_byte_size = _buffer_used;
@@ -792,7 +792,7 @@ static void log_bitmap_usage(const char* which, BitMap* bitmap, size_t total_bit
// Update all oop fields embedded in the buffered objects
void AOTMappedHeapWriter::relocate_embedded_oops(GrowableArrayCHeap* roots,
- ArchiveMappedHeapInfo* heap_info) {
+ AOTMappedHeapInfo* heap_info) {
size_t oopmap_unit = (UseCompressedOops ? sizeof(narrowOop) : sizeof(oop));
size_t heap_region_byte_size = _buffer_used;
heap_info->oopmap()->resize(heap_region_byte_size / oopmap_unit);
@@ -862,7 +862,7 @@ void AOTMappedHeapWriter::mark_native_pointers(oop orig_obj) {
});
}
-void AOTMappedHeapWriter::compute_ptrmap(ArchiveMappedHeapInfo* heap_info) {
+void AOTMappedHeapWriter::compute_ptrmap(AOTMappedHeapInfo* heap_info) {
int num_non_null_ptrs = 0;
Metadata** bottom = (Metadata**) _requested_bottom;
Metadata** top = (Metadata**) _requested_top; // exclusive
@@ -909,40 +909,23 @@ void AOTMappedHeapWriter::compute_ptrmap(ArchiveMappedHeapInfo* heap_info) {
num_non_null_ptrs, size_t(heap_info->ptrmap()->size()));
}
-AOTMapLogger::OopDataIterator* AOTMappedHeapWriter::oop_iterator(ArchiveMappedHeapInfo* heap_info) {
- class MappedWriterOopIterator : public AOTMapLogger::OopDataIterator {
- private:
- address _current;
- address _next;
-
- address _buffer_start;
- address _buffer_end;
- uint64_t _buffer_start_narrow_oop;
- intptr_t _buffer_to_requested_delta;
- int _requested_shift;
-
- size_t _num_root_segments;
- size_t _num_obj_arrays_logged;
-
+AOTMapLogger::OopDataIterator* AOTMappedHeapWriter::oop_iterator(AOTMappedHeapInfo* heap_info) {
+ class MappedWriterOopIterator : public AOTMappedHeapOopIterator {
public:
MappedWriterOopIterator(address buffer_start,
address buffer_end,
- uint64_t buffer_start_narrow_oop,
- intptr_t buffer_to_requested_delta,
+ address requested_base,
+ address requested_start,
int requested_shift,
- size_t num_root_segments)
- : _current(nullptr),
- _next(buffer_start),
- _buffer_start(buffer_start),
- _buffer_end(buffer_end),
- _buffer_start_narrow_oop(buffer_start_narrow_oop),
- _buffer_to_requested_delta(buffer_to_requested_delta),
- _requested_shift(requested_shift),
- _num_root_segments(num_root_segments),
- _num_obj_arrays_logged(0) {
- }
+ size_t num_root_segments) :
+ AOTMappedHeapOopIterator(buffer_start,
+ buffer_end,
+ requested_base,
+ requested_start,
+ requested_shift,
+ num_root_segments) {}
- AOTMapLogger::OopData capture(address buffered_addr) {
+ AOTMapLogger::OopData capture(address buffered_addr) override {
oopDesc* raw_oop = (oopDesc*)buffered_addr;
size_t size = size_of_buffered_oop(buffered_addr);
address requested_addr = buffered_addr_to_requested_addr(buffered_addr);
@@ -960,45 +943,6 @@ AOTMapLogger::OopDataIterator* AOTMappedHeapWriter::oop_iterator(ArchiveMappedHe
size,
false };
}
-
- bool has_next() override {
- return _next < _buffer_end;
- }
-
- AOTMapLogger::OopData next() override {
- _current = _next;
- AOTMapLogger::OopData result = capture(_current);
- if (result._klass->is_objArray_klass()) {
- result._is_root_segment = _num_obj_arrays_logged++ < _num_root_segments;
- }
- _next = _current + result._size * BytesPerWord;
- return result;
- }
-
- AOTMapLogger::OopData obj_at(narrowOop* addr) override {
- uint64_t n = (uint64_t)(*addr);
- if (n == 0) {
- return null_data();
- } else {
- precond(n >= _buffer_start_narrow_oop);
- address buffer_addr = _buffer_start + ((n - _buffer_start_narrow_oop) << _requested_shift);
- return capture(buffer_addr);
- }
- }
-
- AOTMapLogger::OopData obj_at(oop* addr) override {
- address requested_value = cast_from_oop(*addr);
- if (requested_value == nullptr) {
- return null_data();
- } else {
- address buffer_addr = requested_value - _buffer_to_requested_delta;
- return capture(buffer_addr);
- }
- }
-
- GrowableArrayCHeap* roots() override {
- return new GrowableArrayCHeap();
- }
};
MemRegion r = heap_info->buffer_region();
@@ -1008,17 +952,11 @@ AOTMapLogger::OopDataIterator* AOTMappedHeapWriter::oop_iterator(ArchiveMappedHe
address requested_base = UseCompressedOops ? AOTMappedHeapWriter::narrow_oop_base() : (address)AOTMappedHeapWriter::NOCOOPS_REQUESTED_BASE;
address requested_start = UseCompressedOops ? AOTMappedHeapWriter::buffered_addr_to_requested_addr(buffer_start) : requested_base;
int requested_shift = AOTMappedHeapWriter::narrow_oop_shift();
- intptr_t buffer_to_requested_delta = requested_start - buffer_start;
- uint64_t buffer_start_narrow_oop = 0xdeadbeed;
- if (UseCompressedOops) {
- buffer_start_narrow_oop = (uint64_t)(pointer_delta(requested_start, requested_base, 1)) >> requested_shift;
- assert(buffer_start_narrow_oop < 0xffffffff, "sanity");
- }
return new MappedWriterOopIterator(buffer_start,
buffer_end,
- buffer_start_narrow_oop,
- buffer_to_requested_delta,
+ requested_base,
+ requested_start,
requested_shift,
heap_info->root_segments().count());
}
diff --git a/src/hotspot/share/cds/aotMappedHeapWriter.hpp b/src/hotspot/share/cds/aotMappedHeapWriter.hpp
index eafd38ac8bb..7481e7922a0 100644
--- a/src/hotspot/share/cds/aotMappedHeapWriter.hpp
+++ b/src/hotspot/share/cds/aotMappedHeapWriter.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2024, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2024, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -196,10 +196,10 @@ private:
static int filler_array_length(size_t fill_bytes);
static HeapWord* init_filler_array_at_buffer_top(int array_length, size_t fill_bytes);
- static void set_requested_address_range(ArchiveMappedHeapInfo* info);
+ static void set_requested_address_range(AOTMappedHeapInfo* info);
static void mark_native_pointers(oop orig_obj);
- static void relocate_embedded_oops(GrowableArrayCHeap* roots, ArchiveMappedHeapInfo* info);
- static void compute_ptrmap(ArchiveMappedHeapInfo *info);
+ static void relocate_embedded_oops(GrowableArrayCHeap* roots, AOTMappedHeapInfo* info);
+ static void compute_ptrmap(AOTMappedHeapInfo *info);
static bool is_in_requested_range(oop o);
static oop requested_obj_from_buffer_offset(size_t offset);
@@ -229,7 +229,7 @@ public:
static bool is_string_too_large_to_archive(oop string);
static bool is_dumped_interned_string(oop o);
static void add_to_dumped_interned_strings(oop string);
- static void write(GrowableArrayCHeap*, ArchiveMappedHeapInfo* heap_info);
+ static void write(GrowableArrayCHeap*, AOTMappedHeapInfo* heap_info);
static address requested_address(); // requested address of the lowest achived heap object
static size_t get_filler_size_at(address buffered_addr);
@@ -240,7 +240,7 @@ public:
static Klass* real_klass_of_buffered_oop(address buffered_addr);
static size_t size_of_buffered_oop(address buffered_addr);
- static AOTMapLogger::OopDataIterator* oop_iterator(ArchiveMappedHeapInfo* heap_info);
+ static AOTMapLogger::OopDataIterator* oop_iterator(AOTMappedHeapInfo* heap_info);
};
#endif // INCLUDE_CDS_JAVA_HEAP
#endif // SHARE_CDS_AOTMAPPEDHEAPWRITER_HPP
diff --git a/src/hotspot/share/cds/aotMetaspace.cpp b/src/hotspot/share/cds/aotMetaspace.cpp
index 683c897d855..8bb8387f1ab 100644
--- a/src/hotspot/share/cds/aotMetaspace.cpp
+++ b/src/hotspot/share/cds/aotMetaspace.cpp
@@ -661,8 +661,8 @@ void AOTMetaspace::rewrite_bytecodes_and_calculate_fingerprints(Thread* thread,
class VM_PopulateDumpSharedSpace : public VM_Operation {
private:
- ArchiveMappedHeapInfo _mapped_heap_info;
- ArchiveStreamedHeapInfo _streamed_heap_info;
+ AOTMappedHeapInfo _mapped_heap_info;
+ AOTStreamedHeapInfo _streamed_heap_info;
FileMapInfo* _map_info;
StaticArchiveBuilder& _builder;
@@ -682,8 +682,8 @@ public:
bool skip_operation() const { return false; }
VMOp_Type type() const { return VMOp_PopulateDumpSharedSpace; }
- ArchiveMappedHeapInfo* mapped_heap_info() { return &_mapped_heap_info; }
- ArchiveStreamedHeapInfo* streamed_heap_info() { return &_streamed_heap_info; }
+ AOTMappedHeapInfo* mapped_heap_info() { return &_mapped_heap_info; }
+ AOTStreamedHeapInfo* streamed_heap_info() { return &_streamed_heap_info; }
void doit(); // outline because gdb sucks
bool allow_nested_vm_operations() const { return true; }
}; // class VM_PopulateDumpSharedSpace
@@ -1104,7 +1104,12 @@ void AOTMetaspace::dump_static_archive_impl(StaticArchiveBuilder& builder, TRAPS
#if INCLUDE_CDS_JAVA_HEAP
if (CDSConfig::is_dumping_heap()) {
- assert(CDSConfig::allow_only_single_java_thread(), "Required");
+ if (!CDSConfig::is_dumping_preimage_static_archive()) {
+ // A single thread is required for Reference handling and deterministic CDS archive.
+ // Its's not required for dumping preimage, where References won't be archived and
+ // determinism is not needed.
+ assert(CDSConfig::allow_only_single_java_thread(), "Required");
+ }
if (!HeapShared::is_archived_boot_layer_available(THREAD)) {
report_loading_error("archivedBootLayer not available, disabling full module graph");
CDSConfig::stop_dumping_full_module_graph();
@@ -1162,12 +1167,6 @@ void AOTMetaspace::dump_static_archive_impl(StaticArchiveBuilder& builder, TRAPS
// Perhaps there is a way to avoid hard-coding these names here.
// See discussion in JDK-8342481.
}
-
- if (HeapShared::is_writing_mapping_mode()) {
- // Do this at the very end, when no Java code will be executed. Otherwise
- // some new strings may be added to the intern table.
- StringTable::allocate_shared_strings_array(CHECK);
- }
} else {
log_info(aot)("Not dumping heap, reset CDSConfig::_is_using_optimized_module_handling");
CDSConfig::stop_using_optimized_module_handling();
@@ -1213,8 +1212,8 @@ void AOTMetaspace::dump_static_archive_impl(StaticArchiveBuilder& builder, TRAPS
bool AOTMetaspace::write_static_archive(ArchiveBuilder* builder,
FileMapInfo* map_info,
- ArchiveMappedHeapInfo* mapped_heap_info,
- ArchiveStreamedHeapInfo* streamed_heap_info) {
+ AOTMappedHeapInfo* mapped_heap_info,
+ AOTStreamedHeapInfo* streamed_heap_info) {
// relocate the data so that it can be mapped to AOTMetaspace::requested_base_address()
// without runtime relocation.
builder->relocate_to_requested();
diff --git a/src/hotspot/share/cds/aotMetaspace.hpp b/src/hotspot/share/cds/aotMetaspace.hpp
index ab78787288f..4607a936abe 100644
--- a/src/hotspot/share/cds/aotMetaspace.hpp
+++ b/src/hotspot/share/cds/aotMetaspace.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2012, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -33,8 +33,8 @@
#include "utilities/macros.hpp"
class ArchiveBuilder;
-class ArchiveMappedHeapInfo;
-class ArchiveStreamedHeapInfo;
+class AOTMappedHeapInfo;
+class AOTStreamedHeapInfo;
class FileMapInfo;
class Method;
class outputStream;
@@ -192,8 +192,8 @@ private:
static void open_output_mapinfo();
static bool write_static_archive(ArchiveBuilder* builder,
FileMapInfo* map_info,
- ArchiveMappedHeapInfo* mapped_heap_info,
- ArchiveStreamedHeapInfo* streamed_heap_info);
+ AOTMappedHeapInfo* mapped_heap_info,
+ AOTStreamedHeapInfo* streamed_heap_info);
static FileMapInfo* open_static_archive();
static FileMapInfo* open_dynamic_archive();
// use_requested_addr: If true (default), attempt to map at the address the
diff --git a/src/hotspot/share/cds/aotReferenceObjSupport.cpp b/src/hotspot/share/cds/aotReferenceObjSupport.cpp
index aa7cc875533..0c27c8ce5f0 100644
--- a/src/hotspot/share/cds/aotReferenceObjSupport.cpp
+++ b/src/hotspot/share/cds/aotReferenceObjSupport.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2025, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -177,12 +177,17 @@ void AOTReferenceObjSupport::init_keep_alive_objs_table() {
// Returns true IFF obj is an instance of java.lang.ref.Reference. If so, perform extra eligibility checks.
bool AOTReferenceObjSupport::check_if_ref_obj(oop obj) {
- // We have a single Java thread. This means java.lang.ref.Reference$ReferenceHandler thread
- // is not running. Otherwise the checks for next/discovered may not work.
- precond(CDSConfig::allow_only_single_java_thread());
assert_at_safepoint(); // _keep_alive_objs_table uses raw oops
if (obj->klass()->is_subclass_of(vmClasses::Reference_klass())) {
+ // The following check works only if the java.lang.ref.Reference$ReferenceHandler thread
+ // is not running.
+ //
+ // This code is called on every object found by AOTArtifactFinder. When dumping the
+ // preimage archive, AOTArtifactFinder should not find any Reference objects.
+ precond(!CDSConfig::is_dumping_preimage_static_archive());
+ precond(CDSConfig::allow_only_single_java_thread());
+
precond(AOTReferenceObjSupport::is_enabled());
precond(JavaClasses::is_supported_for_archiving(obj));
precond(_keep_alive_objs_table != nullptr);
diff --git a/src/hotspot/share/cds/aotStreamedHeap.cpp b/src/hotspot/share/cds/aotStreamedHeap.cpp
new file mode 100644
index 00000000000..3378924bf32
--- /dev/null
+++ b/src/hotspot/share/cds/aotStreamedHeap.cpp
@@ -0,0 +1,55 @@
+/*
+ * Copyright (c) 2026, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+#include "cds/aotStreamedHeap.hpp"
+
+// Anything that goes in the header must be thoroughly purged from uninitialized memory
+// as it will be written to disk. Therefore, the constructors memset the memory to 0.
+// This is not the prettiest thing, but we need to know every byte is initialized,
+// including potential padding between fields.
+
+AOTStreamedHeapHeader::AOTStreamedHeapHeader(size_t forwarding_offset,
+ size_t roots_offset,
+ size_t num_roots,
+ size_t root_highest_object_index_table_offset,
+ size_t num_archived_objects) {
+ memset((char*)this, 0, sizeof(*this));
+ _forwarding_offset = forwarding_offset;
+ _roots_offset = roots_offset;
+ _num_roots = num_roots;
+ _root_highest_object_index_table_offset = root_highest_object_index_table_offset;
+ _num_archived_objects = num_archived_objects;
+}
+
+AOTStreamedHeapHeader::AOTStreamedHeapHeader() {
+ memset((char*)this, 0, sizeof(*this));
+}
+
+AOTStreamedHeapHeader AOTStreamedHeapInfo::create_header() {
+ return AOTStreamedHeapHeader{_forwarding_offset,
+ _roots_offset,
+ _num_roots,
+ _root_highest_object_index_table_offset,
+ _num_archived_objects};
+}
diff --git a/src/hotspot/share/cds/aotStreamedHeap.hpp b/src/hotspot/share/cds/aotStreamedHeap.hpp
new file mode 100644
index 00000000000..f06b1bcb4c6
--- /dev/null
+++ b/src/hotspot/share/cds/aotStreamedHeap.hpp
@@ -0,0 +1,147 @@
+/*
+ * Copyright (c) 2026, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+#ifndef SHARE_CDS_AOTSTREAMEDHEAP_HPP
+#define SHARE_CDS_AOTSTREAMEDHEAP_HPP
+
+#include "cds/aotMapLogger.hpp"
+#include "utilities/growableArray.hpp"
+#include "utilities/macros.hpp"
+
+class AOTStreamedHeapHeader {
+ size_t _forwarding_offset; // Offset of forwarding information in the heap region.
+ size_t _roots_offset; // Start position for the roots
+ size_t _root_highest_object_index_table_offset; // Offset of root dfs depth information
+ size_t _num_roots; // Number of embedded roots
+ size_t _num_archived_objects; // The number of archived heap objects
+
+public:
+ AOTStreamedHeapHeader();
+ AOTStreamedHeapHeader(size_t forwarding_offset,
+ size_t roots_offset,
+ size_t num_roots,
+ size_t root_highest_object_index_table_offset,
+ size_t num_archived_objects);
+
+ size_t forwarding_offset() const { return _forwarding_offset; }
+ size_t roots_offset() const { return _roots_offset; }
+ size_t num_roots() const { return _num_roots; }
+ size_t root_highest_object_index_table_offset() const { return _root_highest_object_index_table_offset; }
+ size_t num_archived_objects() const { return _num_archived_objects; }
+
+ // This class is trivially copyable and assignable.
+ AOTStreamedHeapHeader(const AOTStreamedHeapHeader&) = default;
+ AOTStreamedHeapHeader& operator=(const AOTStreamedHeapHeader&) = default;
+};
+
+class AOTStreamedHeapInfo {
+ MemRegion _buffer_region; // Contains the archived objects to be written into the CDS archive.
+ CHeapBitMap _oopmap;
+ size_t _roots_offset; // Offset of the HeapShared::roots() object, from the bottom
+ // of the archived heap objects, in bytes.
+ size_t _num_roots;
+
+ size_t _forwarding_offset; // Offset of forwarding information from the bottom
+ size_t _root_highest_object_index_table_offset; // Offset to root dfs depth information
+ size_t _num_archived_objects; // The number of archived objects written into the CDS archive.
+
+public:
+ AOTStreamedHeapInfo()
+ : _buffer_region(),
+ _oopmap(128, mtClassShared),
+ _roots_offset(),
+ _forwarding_offset(),
+ _root_highest_object_index_table_offset(),
+ _num_archived_objects() {}
+
+ bool is_used() { return !_buffer_region.is_empty(); }
+
+ void set_buffer_region(MemRegion r) { _buffer_region = r; }
+ MemRegion buffer_region() { return _buffer_region; }
+ char* buffer_start() { return (char*)_buffer_region.start(); }
+ size_t buffer_byte_size() { return _buffer_region.byte_size(); }
+
+ CHeapBitMap* oopmap() { return &_oopmap; }
+ void set_roots_offset(size_t n) { _roots_offset = n; }
+ size_t roots_offset() { return _roots_offset; }
+ void set_num_roots(size_t n) { _num_roots = n; }
+ size_t num_roots() { return _num_roots; }
+ void set_forwarding_offset(size_t n) { _forwarding_offset = n; }
+ void set_root_highest_object_index_table_offset(size_t n) { _root_highest_object_index_table_offset = n; }
+ void set_num_archived_objects(size_t n) { _num_archived_objects = n; }
+ size_t num_archived_objects() { return _num_archived_objects; }
+
+ AOTStreamedHeapHeader create_header();
+};
+
+#if INCLUDE_CDS_JAVA_HEAP
+class AOTStreamedHeapOopIterator : public AOTMapLogger::OopDataIterator {
+protected:
+ int _current;
+ int _next;
+ address _buffer_start;
+ int _num_archived_objects;
+
+public:
+ AOTStreamedHeapOopIterator(address buffer_start,
+ int num_archived_objects)
+ : _current(0),
+ _next(1),
+ _buffer_start(buffer_start),
+ _num_archived_objects(num_archived_objects) {}
+
+ virtual AOTMapLogger::OopData capture(int dfs_index) = 0;
+
+ bool has_next() override {
+ return _next <= _num_archived_objects;
+ }
+
+ AOTMapLogger::OopData next() override {
+ _current = _next;
+ AOTMapLogger::OopData result = capture(_current);
+ _next = _current + 1;
+ return result;
+ }
+
+ AOTMapLogger::OopData obj_at(narrowOop* addr) override {
+ int dfs_index = (int)(*addr);
+ if (dfs_index == 0) {
+ return null_data();
+ } else {
+ return capture(dfs_index);
+ }
+ }
+
+ AOTMapLogger::OopData obj_at(oop* addr) override {
+ int dfs_index = (int)cast_from_oop(*addr);
+ if (dfs_index == 0) {
+ return null_data();
+ } else {
+ return capture(dfs_index);
+ }
+ }
+};
+#endif // INCLUDE_CDS_JAVA_HEAP
+
+#endif // SHARE_CDS_AOTSTREAMEDHEAP_HPP
diff --git a/src/hotspot/share/cds/aotStreamedHeapLoader.cpp b/src/hotspot/share/cds/aotStreamedHeapLoader.cpp
index 6719f9bf898..39f735543cd 100644
--- a/src/hotspot/share/cds/aotStreamedHeapLoader.cpp
+++ b/src/hotspot/share/cds/aotStreamedHeapLoader.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2018, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -1102,25 +1102,13 @@ void AOTStreamedHeapLoader::finish_initialization(FileMapInfo* static_mapinfo) {
}
AOTMapLogger::OopDataIterator* AOTStreamedHeapLoader::oop_iterator(FileMapInfo* info, address buffer_start, address buffer_end) {
- class StreamedLoaderOopIterator : public AOTMapLogger::OopDataIterator {
- private:
- int _current;
- int _next;
-
- address _buffer_start;
-
- int _num_archived_objects;
-
+ class StreamedLoaderOopIterator : public AOTStreamedHeapOopIterator {
public:
StreamedLoaderOopIterator(address buffer_start,
int num_archived_objects)
- : _current(0),
- _next(1),
- _buffer_start(buffer_start),
- _num_archived_objects(num_archived_objects) {
- }
+ : AOTStreamedHeapOopIterator(buffer_start, num_archived_objects) {}
- AOTMapLogger::OopData capture(int dfs_index) {
+ AOTMapLogger::OopData capture(int dfs_index) override {
size_t buffered_offset = buffer_offset_for_object_index(dfs_index);
address buffered_addr = _buffer_start + buffered_offset;
oopDesc* raw_oop = (oopDesc*)buffered_addr;
@@ -1142,35 +1130,6 @@ AOTMapLogger::OopDataIterator* AOTStreamedHeapLoader::oop_iterator(FileMapInfo*
false };
}
- bool has_next() override {
- return _next <= _num_archived_objects;
- }
-
- AOTMapLogger::OopData next() override {
- _current = _next;
- AOTMapLogger::OopData result = capture(_current);
- _next = _current + 1;
- return result;
- }
-
- AOTMapLogger::OopData obj_at(narrowOop* addr) override {
- int dfs_index = (int)(*addr);
- if (dfs_index == 0) {
- return null_data();
- } else {
- return capture(dfs_index);
- }
- }
-
- AOTMapLogger::OopData obj_at(oop* addr) override {
- int dfs_index = (int)cast_from_oop(*addr);
- if (dfs_index == 0) {
- return null_data();
- } else {
- return capture(dfs_index);
- }
- }
-
GrowableArrayCHeap* roots() override {
GrowableArrayCHeap* result = new GrowableArrayCHeap();
diff --git a/src/hotspot/share/cds/aotStreamedHeapWriter.cpp b/src/hotspot/share/cds/aotStreamedHeapWriter.cpp
index 16acebc7d8d..f52532b2f2a 100644
--- a/src/hotspot/share/cds/aotStreamedHeapWriter.cpp
+++ b/src/hotspot/share/cds/aotStreamedHeapWriter.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2025, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -163,7 +163,7 @@ void AOTStreamedHeapWriter::order_source_objs(GrowableArrayCHeap* roots,
- ArchiveStreamedHeapInfo* heap_info) {
+ AOTStreamedHeapInfo* heap_info) {
assert(CDSConfig::is_dumping_heap(), "sanity");
allocate_buffer();
order_source_objs(roots);
@@ -453,7 +453,7 @@ static void log_bitmap_usage(const char* which, BitMap* bitmap, size_t total_bit
}
// Update all oop fields embedded in the buffered objects
-void AOTStreamedHeapWriter::map_embedded_oops(ArchiveStreamedHeapInfo* heap_info) {
+void AOTStreamedHeapWriter::map_embedded_oops(AOTStreamedHeapInfo* heap_info) {
size_t oopmap_unit = (UseCompressedOops ? sizeof(narrowOop) : sizeof(oop));
size_t heap_region_byte_size = _buffer_used;
heap_info->oopmap()->resize(heap_region_byte_size / oopmap_unit);
@@ -497,7 +497,7 @@ oop AOTStreamedHeapWriter::buffered_addr_to_source_obj(address buffered_addr) {
return buffered_offset_to_source_obj(buffered_address_to_offset(buffered_addr));
}
-void AOTStreamedHeapWriter::populate_archive_heap_info(ArchiveStreamedHeapInfo* info) {
+void AOTStreamedHeapWriter::populate_archive_heap_info(AOTStreamedHeapInfo* info) {
assert(!info->is_used(), "only set once");
size_t heap_region_byte_size = _buffer_used;
@@ -512,15 +512,9 @@ void AOTStreamedHeapWriter::populate_archive_heap_info(ArchiveStreamedHeapInfo*
info->set_num_archived_objects((size_t)_source_objs->length());
}
-AOTMapLogger::OopDataIterator* AOTStreamedHeapWriter::oop_iterator(ArchiveStreamedHeapInfo* heap_info) {
- class StreamedWriterOopIterator : public AOTMapLogger::OopDataIterator {
+AOTMapLogger::OopDataIterator* AOTStreamedHeapWriter::oop_iterator(AOTStreamedHeapInfo* heap_info) {
+ class StreamedWriterOopIterator : public AOTStreamedHeapOopIterator {
private:
- int _current;
- int _next;
-
- address _buffer_start;
-
- int _num_archived_objects;
int _num_archived_roots;
int* _roots;
@@ -529,15 +523,11 @@ AOTMapLogger::OopDataIterator* AOTStreamedHeapWriter::oop_iterator(ArchiveStream
int num_archived_objects,
int num_archived_roots,
int* roots)
- : _current(0),
- _next(1),
- _buffer_start(buffer_start),
- _num_archived_objects(num_archived_objects),
+ : AOTStreamedHeapOopIterator(buffer_start, num_archived_objects),
_num_archived_roots(num_archived_roots),
- _roots(roots) {
- }
+ _roots(roots) {}
- AOTMapLogger::OopData capture(int dfs_index) {
+ AOTMapLogger::OopData capture(int dfs_index) override {
size_t buffered_offset = _dfs_to_archive_object_table[dfs_index];
address buffered_addr = _buffer_start + buffered_offset;
oop src_obj = AOTStreamedHeapWriter::buffered_offset_to_source_obj(buffered_offset);
@@ -561,35 +551,6 @@ AOTMapLogger::OopDataIterator* AOTStreamedHeapWriter::oop_iterator(ArchiveStream
false };
}
- bool has_next() override {
- return _next <= _num_archived_objects;
- }
-
- AOTMapLogger::OopData next() override {
- _current = _next;
- AOTMapLogger::OopData result = capture(_current);
- _next = _current + 1;
- return result;
- }
-
- AOTMapLogger::OopData obj_at(narrowOop* addr) override {
- int dfs_index = (int)(*addr);
- if (dfs_index == 0) {
- return null_data();
- } else {
- return capture(dfs_index);
- }
- }
-
- AOTMapLogger::OopData obj_at(oop* addr) override {
- int dfs_index = (int)cast_from_oop(*addr);
- if (dfs_index == 0) {
- return null_data();
- } else {
- return capture(dfs_index);
- }
- }
-
GrowableArrayCHeap* roots() override {
GrowableArrayCHeap* result = new GrowableArrayCHeap();
diff --git a/src/hotspot/share/cds/aotStreamedHeapWriter.hpp b/src/hotspot/share/cds/aotStreamedHeapWriter.hpp
index bde82f8ce29..ab5aec0327b 100644
--- a/src/hotspot/share/cds/aotStreamedHeapWriter.hpp
+++ b/src/hotspot/share/cds/aotStreamedHeapWriter.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2025, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -117,7 +117,7 @@ class AOTStreamedHeapWriter : AllStatic {
static void copy_forwarding_to_buffer();
static void copy_roots_max_dfs_to_buffer(int roots_length);
- static void map_embedded_oops(ArchiveStreamedHeapInfo* info);
+ static void map_embedded_oops(AOTStreamedHeapInfo* info);
static bool is_in_requested_range(oop o);
static oop requested_obj_from_buffer_offset(size_t offset);
@@ -131,14 +131,14 @@ class AOTStreamedHeapWriter : AllStatic {
static void update_header_for_buffered_addr(address buffered_addr, oop src_obj, Klass* src_klass);
- static void populate_archive_heap_info(ArchiveStreamedHeapInfo* info);
+ static void populate_archive_heap_info(AOTStreamedHeapInfo* info);
public:
static void init() NOT_CDS_JAVA_HEAP_RETURN;
static void delete_tables_with_raw_oops();
static void add_source_obj(oop src_obj);
- static void write(GrowableArrayCHeap*, ArchiveStreamedHeapInfo* heap_info);
+ static void write(GrowableArrayCHeap*, AOTStreamedHeapInfo* heap_info);
static address buffered_heap_roots_addr() {
return offset_to_buffered_address(_roots_offset);
}
@@ -156,7 +156,7 @@ public:
static oop buffered_offset_to_source_obj(size_t buffered_offset);
static oop buffered_addr_to_source_obj(address buffered_addr);
- static AOTMapLogger::OopDataIterator* oop_iterator(ArchiveStreamedHeapInfo* heap_info);
+ static AOTMapLogger::OopDataIterator* oop_iterator(AOTStreamedHeapInfo* heap_info);
};
#endif // INCLUDE_CDS_JAVA_HEAP
#endif // SHARE_CDS_AOTSTREAMEDHEAPWRITER_HPP
diff --git a/src/hotspot/share/cds/archiveBuilder.cpp b/src/hotspot/share/cds/archiveBuilder.cpp
index 9161980c4be..e65bd3985ac 100644
--- a/src/hotspot/share/cds/archiveBuilder.cpp
+++ b/src/hotspot/share/cds/archiveBuilder.cpp
@@ -1154,7 +1154,7 @@ void ArchiveBuilder::print_stats() {
_alloc_stats.print_stats(int(_ro_region.used()), int(_rw_region.used()));
}
-void ArchiveBuilder::write_archive(FileMapInfo* mapinfo, ArchiveMappedHeapInfo* mapped_heap_info, ArchiveStreamedHeapInfo* streamed_heap_info) {
+void ArchiveBuilder::write_archive(FileMapInfo* mapinfo, AOTMappedHeapInfo* mapped_heap_info, AOTStreamedHeapInfo* streamed_heap_info) {
// Make sure NUM_CDS_REGIONS (exported in cds.h) agrees with
// AOTMetaspace::n_regions (internal to hotspot).
assert(NUM_CDS_REGIONS == AOTMetaspace::n_regions, "sanity");
@@ -1213,8 +1213,8 @@ void ArchiveBuilder::count_relocated_pointer(bool tagged, bool nulled) {
}
void ArchiveBuilder::print_region_stats(FileMapInfo *mapinfo,
- ArchiveMappedHeapInfo* mapped_heap_info,
- ArchiveStreamedHeapInfo* streamed_heap_info) {
+ AOTMappedHeapInfo* mapped_heap_info,
+ AOTStreamedHeapInfo* streamed_heap_info) {
// Print statistics of all the regions
const size_t bitmap_used = mapinfo->region_at(AOTMetaspace::bm)->used();
const size_t bitmap_reserved = mapinfo->region_at(AOTMetaspace::bm)->used_aligned();
diff --git a/src/hotspot/share/cds/archiveBuilder.hpp b/src/hotspot/share/cds/archiveBuilder.hpp
index 9de6c02edc5..2284dbf70f8 100644
--- a/src/hotspot/share/cds/archiveBuilder.hpp
+++ b/src/hotspot/share/cds/archiveBuilder.hpp
@@ -39,8 +39,8 @@
#include "utilities/hashTable.hpp"
#include "utilities/resizableHashTable.hpp"
-class ArchiveMappedHeapInfo;
-class ArchiveStreamedHeapInfo;
+class AOTMappedHeapInfo;
+class AOTStreamedHeapInfo;
class CHeapBitMap;
class FileMapInfo;
class Klass;
@@ -247,8 +247,8 @@ private:
} _relocated_ptr_info;
void print_region_stats(FileMapInfo *map_info,
- ArchiveMappedHeapInfo* mapped_heap_info,
- ArchiveStreamedHeapInfo* streamed_heap_info);
+ AOTMappedHeapInfo* mapped_heap_info,
+ AOTStreamedHeapInfo* streamed_heap_info);
void print_bitmap_region_stats(size_t size, size_t total_size);
void print_heap_region_stats(char* start, size_t size, size_t total_size);
@@ -438,8 +438,8 @@ public:
void make_training_data_shareable();
void relocate_to_requested();
void write_archive(FileMapInfo* mapinfo,
- ArchiveMappedHeapInfo* mapped_heap_info,
- ArchiveStreamedHeapInfo* streamed_heap_info);
+ AOTMappedHeapInfo* mapped_heap_info,
+ AOTStreamedHeapInfo* streamed_heap_info);
void write_region(FileMapInfo* mapinfo, int region_idx, DumpRegion* dump_region,
bool read_only, bool allow_exec);
diff --git a/src/hotspot/share/cds/cdsConfig.cpp b/src/hotspot/share/cds/cdsConfig.cpp
index 5f6b568dd6e..f4ef3c66f7a 100644
--- a/src/hotspot/share/cds/cdsConfig.cpp
+++ b/src/hotspot/share/cds/cdsConfig.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2023, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2023, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -556,7 +556,9 @@ void CDSConfig::check_aotmode_record() {
// At VM exit, the module graph may be contaminated with program states.
// We will rebuild the module graph when dumping the CDS final image.
- disable_heap_dumping();
+ _is_using_optimized_module_handling = false;
+ _is_using_full_module_graph = false;
+ _is_dumping_full_module_graph = false;
}
void CDSConfig::check_aotmode_create() {
@@ -582,6 +584,7 @@ void CDSConfig::check_aotmode_create() {
substitute_aot_filename(FLAG_MEMBER_ENUM(AOTCache));
_is_dumping_final_static_archive = true;
+ _is_using_full_module_graph = false;
UseSharedSpaces = true;
RequireSharedSpaces = true;
@@ -954,7 +957,9 @@ bool CDSConfig::are_vm_options_incompatible_with_dumping_heap() {
}
bool CDSConfig::is_dumping_heap() {
- if (!(is_dumping_classic_static_archive() || is_dumping_final_static_archive())
+ // Note: when dumping preimage static archive, only a very limited set of oops
+ // are dumped.
+ if (!is_dumping_static_archive()
|| are_vm_options_incompatible_with_dumping_heap()
|| _disable_heap_dumping) {
return false;
@@ -966,6 +971,26 @@ bool CDSConfig::is_loading_heap() {
return HeapShared::is_archived_heap_in_use();
}
+bool CDSConfig::is_dumping_klass_subgraphs() {
+ if (is_dumping_classic_static_archive() || is_dumping_final_static_archive()) {
+ // KlassSubGraphs (see heapShared.cpp) is a legacy mechanism for archiving oops. It
+ // has been superceded by AOT class linking. This feature is used only when
+ // AOT class linking is disabled.
+ //
+ // KlassSubGraphs are disabled in the preimage static archive, which contains a very
+ // limited set of oops.
+ return is_dumping_heap() && !is_dumping_aot_linked_classes();
+ } else {
+ return false;
+ }
+}
+
+bool CDSConfig::is_using_klass_subgraphs() {
+ return (is_loading_heap() &&
+ !CDSConfig::is_using_aot_linked_classes() &&
+ !CDSConfig::is_dumping_final_static_archive());
+}
+
bool CDSConfig::is_using_full_module_graph() {
if (ClassLoaderDataShared::is_full_module_graph_loaded()) {
return true;
diff --git a/src/hotspot/share/cds/cdsConfig.hpp b/src/hotspot/share/cds/cdsConfig.hpp
index 202904e8231..739dbb4937b 100644
--- a/src/hotspot/share/cds/cdsConfig.hpp
+++ b/src/hotspot/share/cds/cdsConfig.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2023, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2023, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -188,6 +188,9 @@ public:
static bool is_dumping_heap() NOT_CDS_JAVA_HEAP_RETURN_(false);
static bool is_loading_heap() NOT_CDS_JAVA_HEAP_RETURN_(false);
+ static bool is_dumping_klass_subgraphs() NOT_CDS_JAVA_HEAP_RETURN_(false);
+ static bool is_using_klass_subgraphs() NOT_CDS_JAVA_HEAP_RETURN_(false);
+
static bool is_dumping_invokedynamic() NOT_CDS_JAVA_HEAP_RETURN_(false);
static bool is_dumping_method_handles() NOT_CDS_JAVA_HEAP_RETURN_(false);
diff --git a/src/hotspot/share/cds/filemap.cpp b/src/hotspot/share/cds/filemap.cpp
index 0eeb96bb269..7cd736885ad 100644
--- a/src/hotspot/share/cds/filemap.cpp
+++ b/src/hotspot/share/cds/filemap.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2003, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -974,8 +974,8 @@ size_t FileMapInfo::remove_bitmap_zeros(CHeapBitMap* map) {
char* FileMapInfo::write_bitmap_region(CHeapBitMap* rw_ptrmap,
CHeapBitMap* ro_ptrmap,
- ArchiveMappedHeapInfo* mapped_heap_info,
- ArchiveStreamedHeapInfo* streamed_heap_info,
+ AOTMappedHeapInfo* mapped_heap_info,
+ AOTStreamedHeapInfo* streamed_heap_info,
size_t &size_in_bytes) {
size_t removed_rw_leading_zeros = remove_bitmap_zeros(rw_ptrmap);
size_t removed_ro_leading_zeros = remove_bitmap_zeros(ro_ptrmap);
@@ -1035,7 +1035,7 @@ char* FileMapInfo::write_bitmap_region(CHeapBitMap* rw_ptrmap,
}
#if INCLUDE_CDS_JAVA_HEAP
-size_t FileMapInfo::write_mapped_heap_region(ArchiveMappedHeapInfo* heap_info) {
+size_t FileMapInfo::write_mapped_heap_region(AOTMappedHeapInfo* heap_info) {
char* buffer_start = heap_info->buffer_start();
size_t buffer_size = heap_info->buffer_byte_size();
write_region(AOTMetaspace::hp, buffer_start, buffer_size, false, false);
@@ -1043,7 +1043,7 @@ size_t FileMapInfo::write_mapped_heap_region(ArchiveMappedHeapInfo* heap_info) {
return buffer_size;
}
-size_t FileMapInfo::write_streamed_heap_region(ArchiveStreamedHeapInfo* heap_info) {
+size_t FileMapInfo::write_streamed_heap_region(AOTStreamedHeapInfo* heap_info) {
char* buffer_start = heap_info->buffer_start();
size_t buffer_size = heap_info->buffer_byte_size();
write_region(AOTMetaspace::hp, buffer_start, buffer_size, true, false);
diff --git a/src/hotspot/share/cds/filemap.hpp b/src/hotspot/share/cds/filemap.hpp
index fbd3c8e1681..ec7b58a6d19 100644
--- a/src/hotspot/share/cds/filemap.hpp
+++ b/src/hotspot/share/cds/filemap.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2003, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -25,7 +25,9 @@
#ifndef SHARE_CDS_FILEMAP_HPP
#define SHARE_CDS_FILEMAP_HPP
+#include "cds/aotMappedHeap.hpp"
#include "cds/aotMetaspace.hpp"
+#include "cds/aotStreamedHeap.hpp"
#include "cds/archiveUtils.hpp"
#include "cds/heapShared.hpp"
#include "include/cds.h"
@@ -144,8 +146,8 @@ private:
size_t _rw_ptrmap_start_pos; // The first bit in the ptrmap corresponds to this position in the rw region
size_t _ro_ptrmap_start_pos; // The first bit in the ptrmap corresponds to this position in the ro region
- ArchiveMappedHeapHeader _mapped_heap_header;
- ArchiveStreamedHeapHeader _streamed_heap_header;
+ AOTMappedHeapHeader _mapped_heap_header;
+ AOTStreamedHeapHeader _streamed_heap_header;
// The following are parameters that affect MethodData layout.
u1 _compiler_type;
@@ -209,11 +211,11 @@ public:
size_t ro_ptrmap_start_pos() const { return _ro_ptrmap_start_pos; }
// Heap archiving
- const ArchiveMappedHeapHeader* mapped_heap() const { return &_mapped_heap_header; }
- const ArchiveStreamedHeapHeader* streamed_heap() const { return &_streamed_heap_header; }
+ const AOTMappedHeapHeader* mapped_heap() const { return &_mapped_heap_header; }
+ const AOTStreamedHeapHeader* streamed_heap() const { return &_streamed_heap_header; }
- void set_streamed_heap_header(ArchiveStreamedHeapHeader header) { _streamed_heap_header = header; }
- void set_mapped_heap_header(ArchiveMappedHeapHeader header) { _mapped_heap_header = header; }
+ void set_streamed_heap_header(AOTStreamedHeapHeader header) { _streamed_heap_header = header; }
+ void set_mapped_heap_header(AOTMappedHeapHeader header) { _mapped_heap_header = header; }
void set_has_platform_or_app_classes(bool v) { _has_platform_or_app_classes = v; }
void set_cloned_vtables(char* p) { set_as_offset(p, &_cloned_vtables_offset); }
@@ -309,8 +311,8 @@ public:
uintx max_heap_size() const { return header()->max_heap_size(); }
size_t core_region_alignment() const { return header()->core_region_alignment(); }
- const ArchiveMappedHeapHeader* mapped_heap() const { return header()->mapped_heap(); }
- const ArchiveStreamedHeapHeader* streamed_heap() const { return header()->streamed_heap(); }
+ const AOTMappedHeapHeader* mapped_heap() const { return header()->mapped_heap(); }
+ const AOTStreamedHeapHeader* streamed_heap() const { return header()->streamed_heap(); }
bool object_streaming_mode() const { return header()->object_streaming_mode(); }
CompressedOops::Mode narrow_oop_mode() const { return header()->narrow_oop_mode(); }
@@ -372,11 +374,11 @@ public:
size_t remove_bitmap_zeros(CHeapBitMap* map);
char* write_bitmap_region(CHeapBitMap* rw_ptrmap,
CHeapBitMap* ro_ptrmap,
- ArchiveMappedHeapInfo* mapped_heap_info,
- ArchiveStreamedHeapInfo* streamed_heap_info,
+ AOTMappedHeapInfo* mapped_heap_info,
+ AOTStreamedHeapInfo* streamed_heap_info,
size_t &size_in_bytes);
- size_t write_mapped_heap_region(ArchiveMappedHeapInfo* heap_info) NOT_CDS_JAVA_HEAP_RETURN_(0);
- size_t write_streamed_heap_region(ArchiveStreamedHeapInfo* heap_info) NOT_CDS_JAVA_HEAP_RETURN_(0);
+ size_t write_mapped_heap_region(AOTMappedHeapInfo* heap_info) NOT_CDS_JAVA_HEAP_RETURN_(0);
+ size_t write_streamed_heap_region(AOTStreamedHeapInfo* heap_info) NOT_CDS_JAVA_HEAP_RETURN_(0);
void write_bytes(const void* buffer, size_t count);
void write_bytes_aligned(const void* buffer, size_t count);
size_t read_bytes(void* buffer, size_t count);
diff --git a/src/hotspot/share/cds/heapShared.cpp b/src/hotspot/share/cds/heapShared.cpp
index 89694c6780e..69a098c67b3 100644
--- a/src/hotspot/share/cds/heapShared.cpp
+++ b/src/hotspot/share/cds/heapShared.cpp
@@ -95,55 +95,6 @@ struct ArchivableStaticFieldInfo {
}
};
-// Anything that goes in the header must be thoroughly purged from uninitialized memory
-// as it will be written to disk. Therefore, the constructors memset the memory to 0.
-// This is not the prettiest thing, but we need to know every byte is initialized,
-// including potential padding between fields.
-
-ArchiveMappedHeapHeader::ArchiveMappedHeapHeader(size_t ptrmap_start_pos,
- size_t oopmap_start_pos,
- HeapRootSegments root_segments) {
- memset((char*)this, 0, sizeof(*this));
- _ptrmap_start_pos = ptrmap_start_pos;
- _oopmap_start_pos = oopmap_start_pos;
- _root_segments = root_segments;
-}
-
-ArchiveMappedHeapHeader::ArchiveMappedHeapHeader() {
- memset((char*)this, 0, sizeof(*this));
-}
-
-ArchiveMappedHeapHeader ArchiveMappedHeapInfo::create_header() {
- return ArchiveMappedHeapHeader{_ptrmap_start_pos,
- _oopmap_start_pos,
- _root_segments};
-}
-
-ArchiveStreamedHeapHeader::ArchiveStreamedHeapHeader(size_t forwarding_offset,
- size_t roots_offset,
- size_t num_roots,
- size_t root_highest_object_index_table_offset,
- size_t num_archived_objects) {
- memset((char*)this, 0, sizeof(*this));
- _forwarding_offset = forwarding_offset;
- _roots_offset = roots_offset;
- _num_roots = num_roots;
- _root_highest_object_index_table_offset = root_highest_object_index_table_offset;
- _num_archived_objects = num_archived_objects;
-}
-
-ArchiveStreamedHeapHeader::ArchiveStreamedHeapHeader() {
- memset((char*)this, 0, sizeof(*this));
-}
-
-ArchiveStreamedHeapHeader ArchiveStreamedHeapInfo::create_header() {
- return ArchiveStreamedHeapHeader{_forwarding_offset,
- _roots_offset,
- _num_roots,
- _root_highest_object_index_table_offset,
- _num_archived_objects};
-}
-
HeapArchiveMode HeapShared::_heap_load_mode = HeapArchiveMode::_uninitialized;
HeapArchiveMode HeapShared::_heap_write_mode = HeapArchiveMode::_uninitialized;
@@ -210,7 +161,7 @@ static bool is_subgraph_root_class_of(ArchivableStaticFieldInfo fields[], Instan
bool HeapShared::is_subgraph_root_class(InstanceKlass* ik) {
assert(CDSConfig::is_dumping_heap(), "dump-time only");
- if (!CDSConfig::is_dumping_aot_linked_classes()) {
+ if (CDSConfig::is_dumping_klass_subgraphs()) {
// Legacy CDS archive support (to be deprecated)
return is_subgraph_root_class_of(archive_subgraph_entry_fields, ik) ||
is_subgraph_root_class_of(fmg_archive_subgraph_entry_fields, ik);
@@ -413,6 +364,8 @@ void HeapShared::materialize_thread_object() {
void HeapShared::add_to_dumped_interned_strings(oop string) {
assert(HeapShared::is_writing_mapping_mode(), "Only used by this mode");
AOTMappedHeapWriter::add_to_dumped_interned_strings(string);
+ bool success = archive_reachable_objects_from(1, _dump_time_special_subgraph, string);
+ assert(success, "shared strings array must not point to arrays or strings that are too large to archive");
}
void HeapShared::finalize_initialization(FileMapInfo* static_mapinfo) {
@@ -453,7 +406,6 @@ int HeapShared::append_root(oop obj) {
oop HeapShared::get_root(int index, bool clear) {
assert(index >= 0, "sanity");
- assert(!CDSConfig::is_dumping_heap() && CDSConfig::is_using_archive(), "runtime only");
assert(is_archived_heap_in_use(), "getting roots into heap that is not used");
oop result;
@@ -598,8 +550,7 @@ public:
void set_oop(MetaspaceObj* ptr, oop o) {
MutexLocker ml(ScratchObjects_lock, Mutex::_no_safepoint_check_flag);
OopHandle handle(Universe::vm_global(), o);
- bool is_new = put(ptr, handle);
- assert(is_new, "cannot set twice");
+ put_when_absent(ptr, handle);
}
void remove_oop(MetaspaceObj* ptr) {
MutexLocker ml(ScratchObjects_lock, Mutex::_no_safepoint_check_flag);
@@ -612,6 +563,11 @@ public:
};
void HeapShared::add_scratch_resolved_references(ConstantPool* src, objArrayOop dest) {
+ if (CDSConfig::is_dumping_preimage_static_archive() && scratch_resolved_references(src) != nullptr) {
+ // We are in AOT training run. The class has been redefined and we are giving it a new resolved_reference.
+ // Ignore it, as this class will be excluded from the AOT config.
+ return;
+ }
if (SystemDictionaryShared::is_builtin_loader(src->pool_holder()->class_loader_data())) {
_scratch_objects_table->set_oop(src, dest);
}
@@ -831,14 +787,6 @@ static objArrayOop get_archived_resolved_references(InstanceKlass* src_ik) {
return nullptr;
}
-void HeapShared::archive_strings() {
- assert(HeapShared::is_writing_mapping_mode(), "should not reach here");
- oop shared_strings_array = StringTable::init_shared_strings_array();
- bool success = archive_reachable_objects_from(1, _dump_time_special_subgraph, shared_strings_array);
- assert(success, "shared strings array must not point to arrays or strings that are too large to archive");
- StringTable::set_shared_strings_array_index(append_root(shared_strings_array));
-}
-
int HeapShared::archive_exception_instance(oop exception) {
bool success = archive_reachable_objects_from(1, _dump_time_special_subgraph, exception);
assert(success, "sanity");
@@ -890,12 +838,12 @@ void HeapShared::start_scanning_for_oops() {
void HeapShared::end_scanning_for_oops() {
if (is_writing_mapping_mode()) {
- archive_strings();
+ StringTable::init_shared_table();
}
delete_seen_objects_table();
}
-void HeapShared::write_heap(ArchiveMappedHeapInfo* mapped_heap_info, ArchiveStreamedHeapInfo* streamed_heap_info) {
+void HeapShared::write_heap(AOTMappedHeapInfo* mapped_heap_info, AOTStreamedHeapInfo* streamed_heap_info) {
{
NoSafepointVerifier nsv;
CDSHeapVerifier::verify();
@@ -940,7 +888,7 @@ void HeapShared::scan_java_class(Klass* orig_k) {
void HeapShared::archive_subgraphs() {
assert(CDSConfig::is_dumping_heap(), "must be");
- if (!CDSConfig::is_dumping_aot_linked_classes()) {
+ if (CDSConfig::is_dumping_klass_subgraphs()) {
archive_object_subgraphs(archive_subgraph_entry_fields,
false /* is_full_module_graph */);
if (CDSConfig::is_dumping_full_module_graph()) {
@@ -1298,10 +1246,7 @@ static void verify_the_heap(Klass* k, const char* which) {
// this case, we will not load the ArchivedKlassSubGraphInfoRecord and will clear its roots.
void HeapShared::resolve_classes(JavaThread* current) {
assert(CDSConfig::is_using_archive(), "runtime only!");
- if (!is_archived_heap_in_use()) {
- return; // nothing to do
- }
- if (!CDSConfig::is_using_aot_linked_classes()) {
+ if (CDSConfig::is_using_klass_subgraphs()) {
resolve_classes_for_subgraphs(current, archive_subgraph_entry_fields);
resolve_classes_for_subgraphs(current, fmg_archive_subgraph_entry_fields);
}
@@ -1391,7 +1336,7 @@ void HeapShared::init_classes_for_special_subgraph(Handle class_loader, TRAPS) {
void HeapShared::initialize_from_archived_subgraph(JavaThread* current, Klass* k) {
JavaThread* THREAD = current;
- if (!is_archived_heap_in_use()) {
+ if (!CDSConfig::is_using_klass_subgraphs()) {
return; // nothing to do
}
@@ -1867,7 +1812,7 @@ void HeapShared::archive_reachable_objects_from_static_field(InstanceKlass *k,
const char* klass_name,
int field_offset,
const char* field_name) {
- assert(CDSConfig::is_dumping_heap(), "dump time only");
+ precond(CDSConfig::is_dumping_klass_subgraphs());
assert(k->defined_by_boot_loader(), "must be boot class");
oop m = k->java_mirror();
@@ -1918,7 +1863,7 @@ class VerifySharedOopClosure: public BasicOopIterateClosure {
};
void HeapShared::verify_subgraph_from_static_field(InstanceKlass* k, int field_offset) {
- assert(CDSConfig::is_dumping_heap(), "dump time only");
+ precond(CDSConfig::is_dumping_klass_subgraphs());
assert(k->defined_by_boot_loader(), "must be boot class");
oop m = k->java_mirror();
@@ -2144,7 +2089,7 @@ void HeapShared::init_subgraph_entry_fields(ArchivableStaticFieldInfo fields[],
void HeapShared::init_subgraph_entry_fields(TRAPS) {
assert(CDSConfig::is_dumping_heap(), "must be");
_dump_time_subgraph_info_table = new (mtClass)DumpTimeKlassSubGraphInfoTable();
- if (!CDSConfig::is_dumping_aot_linked_classes()) {
+ if (CDSConfig::is_dumping_klass_subgraphs()) {
init_subgraph_entry_fields(archive_subgraph_entry_fields, CHECK);
if (CDSConfig::is_dumping_full_module_graph()) {
init_subgraph_entry_fields(fmg_archive_subgraph_entry_fields, CHECK);
diff --git a/src/hotspot/share/cds/heapShared.hpp b/src/hotspot/share/cds/heapShared.hpp
index 118c60faa60..ba17ddda267 100644
--- a/src/hotspot/share/cds/heapShared.hpp
+++ b/src/hotspot/share/cds/heapShared.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2018, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -142,129 +142,6 @@ enum class HeapArchiveMode {
_streaming
};
-class ArchiveMappedHeapHeader {
- size_t _ptrmap_start_pos; // The first bit in the ptrmap corresponds to this position in the heap.
- size_t _oopmap_start_pos; // The first bit in the oopmap corresponds to this position in the heap.
- HeapRootSegments _root_segments; // Heap root segments info
-
-public:
- ArchiveMappedHeapHeader();
- ArchiveMappedHeapHeader(size_t ptrmap_start_pos,
- size_t oopmap_start_pos,
- HeapRootSegments root_segments);
-
- size_t ptrmap_start_pos() const { return _ptrmap_start_pos; }
- size_t oopmap_start_pos() const { return _oopmap_start_pos; }
- HeapRootSegments root_segments() const { return _root_segments; }
-
- // This class is trivially copyable and assignable.
- ArchiveMappedHeapHeader(const ArchiveMappedHeapHeader&) = default;
- ArchiveMappedHeapHeader& operator=(const ArchiveMappedHeapHeader&) = default;
-};
-
-
-class ArchiveStreamedHeapHeader {
- size_t _forwarding_offset; // Offset of forwarding information in the heap region.
- size_t _roots_offset; // Start position for the roots
- size_t _root_highest_object_index_table_offset; // Offset of root dfs depth information
- size_t _num_roots; // Number of embedded roots
- size_t _num_archived_objects; // The number of archived heap objects
-
-public:
- ArchiveStreamedHeapHeader();
- ArchiveStreamedHeapHeader(size_t forwarding_offset,
- size_t roots_offset,
- size_t num_roots,
- size_t root_highest_object_index_table_offset,
- size_t num_archived_objects);
-
- size_t forwarding_offset() const { return _forwarding_offset; }
- size_t roots_offset() const { return _roots_offset; }
- size_t num_roots() const { return _num_roots; }
- size_t root_highest_object_index_table_offset() const { return _root_highest_object_index_table_offset; }
- size_t num_archived_objects() const { return _num_archived_objects; }
-
- // This class is trivially copyable and assignable.
- ArchiveStreamedHeapHeader(const ArchiveStreamedHeapHeader&) = default;
- ArchiveStreamedHeapHeader& operator=(const ArchiveStreamedHeapHeader&) = default;
-};
-
-class ArchiveMappedHeapInfo {
- MemRegion _buffer_region; // Contains the archived objects to be written into the CDS archive.
- CHeapBitMap _oopmap;
- CHeapBitMap _ptrmap;
- HeapRootSegments _root_segments;
- size_t _oopmap_start_pos; // How many zeros were removed from the beginning of the bit map?
- size_t _ptrmap_start_pos; // How many zeros were removed from the beginning of the bit map?
-
-public:
- ArchiveMappedHeapInfo() :
- _buffer_region(),
- _oopmap(128, mtClassShared),
- _ptrmap(128, mtClassShared),
- _root_segments(),
- _oopmap_start_pos(),
- _ptrmap_start_pos() {}
- bool is_used() { return !_buffer_region.is_empty(); }
-
- MemRegion buffer_region() { return _buffer_region; }
- void set_buffer_region(MemRegion r) { _buffer_region = r; }
-
- char* buffer_start() { return (char*)_buffer_region.start(); }
- size_t buffer_byte_size() { return _buffer_region.byte_size(); }
-
- CHeapBitMap* oopmap() { return &_oopmap; }
- CHeapBitMap* ptrmap() { return &_ptrmap; }
-
- void set_oopmap_start_pos(size_t start_pos) { _oopmap_start_pos = start_pos; }
- void set_ptrmap_start_pos(size_t start_pos) { _ptrmap_start_pos = start_pos; }
-
- void set_root_segments(HeapRootSegments segments) { _root_segments = segments; };
- HeapRootSegments root_segments() { return _root_segments; }
-
- ArchiveMappedHeapHeader create_header();
-};
-
-class ArchiveStreamedHeapInfo {
- MemRegion _buffer_region; // Contains the archived objects to be written into the CDS archive.
- CHeapBitMap _oopmap;
- size_t _roots_offset; // Offset of the HeapShared::roots() object, from the bottom
- // of the archived heap objects, in bytes.
- size_t _num_roots;
-
- size_t _forwarding_offset; // Offset of forwarding information from the bottom
- size_t _root_highest_object_index_table_offset; // Offset to root dfs depth information
- size_t _num_archived_objects; // The number of archived objects written into the CDS archive.
-
-public:
- ArchiveStreamedHeapInfo()
- : _buffer_region(),
- _oopmap(128, mtClassShared),
- _roots_offset(),
- _forwarding_offset(),
- _root_highest_object_index_table_offset(),
- _num_archived_objects() {}
-
- bool is_used() { return !_buffer_region.is_empty(); }
-
- void set_buffer_region(MemRegion r) { _buffer_region = r; }
- MemRegion buffer_region() { return _buffer_region; }
- char* buffer_start() { return (char*)_buffer_region.start(); }
- size_t buffer_byte_size() { return _buffer_region.byte_size(); }
-
- CHeapBitMap* oopmap() { return &_oopmap; }
- void set_roots_offset(size_t n) { _roots_offset = n; }
- size_t roots_offset() { return _roots_offset; }
- void set_num_roots(size_t n) { _num_roots = n; }
- size_t num_roots() { return _num_roots; }
- void set_forwarding_offset(size_t n) { _forwarding_offset = n; }
- void set_root_highest_object_index_table_offset(size_t n) { _root_highest_object_index_table_offset = n; }
- void set_num_archived_objects(size_t n) { _num_archived_objects = n; }
- size_t num_archived_objects() { return _num_archived_objects; }
-
- ArchiveStreamedHeapHeader create_header();
-};
-
class HeapShared: AllStatic {
friend class VerifySharedOopClosure;
@@ -478,7 +355,6 @@ private:
static bool has_been_archived(oop orig_obj);
static void prepare_resolved_references();
- static void archive_strings();
static void archive_subgraphs();
static void copy_java_mirror(oop orig_mirror, oop scratch_m);
@@ -576,7 +452,7 @@ private:
public:
static void finish_materialize_objects() NOT_CDS_JAVA_HEAP_RETURN;
- static void write_heap(ArchiveMappedHeapInfo* mapped_heap_info, ArchiveStreamedHeapInfo* streamed_heap_info) NOT_CDS_JAVA_HEAP_RETURN;
+ static void write_heap(AOTMappedHeapInfo* mapped_heap_info, AOTStreamedHeapInfo* streamed_heap_info) NOT_CDS_JAVA_HEAP_RETURN;
static objArrayOop scratch_resolved_references(ConstantPool* src);
static void add_scratch_resolved_references(ConstantPool* src, objArrayOop dest) NOT_CDS_JAVA_HEAP_RETURN;
static void init_dumping() NOT_CDS_JAVA_HEAP_RETURN;
diff --git a/src/hotspot/share/ci/ciField.cpp b/src/hotspot/share/ci/ciField.cpp
index e0c818f02fc..946fea5346f 100644
--- a/src/hotspot/share/ci/ciField.cpp
+++ b/src/hotspot/share/ci/ciField.cpp
@@ -213,7 +213,7 @@ ciField::ciField(fieldDescriptor *fd) :
"bootstrap classes must not create & cache unshared fields");
}
-static bool trust_final_non_static_fields(ciInstanceKlass* holder) {
+static bool trust_final_nonstatic_fields(ciInstanceKlass* holder) {
if (holder == nullptr)
return false;
if (holder->trust_final_fields()) {
@@ -259,7 +259,7 @@ void ciField::initialize_from(fieldDescriptor* fd) {
// An instance field can be constant if it's a final static field or if
// it's a final non-static field of a trusted class (classes in
// java.lang.invoke and sun.invoke packages and subpackages).
- _is_constant = is_stable_field || trust_final_non_static_fields(_holder);
+ _is_constant = is_stable_field || trust_final_nonstatic_fields(_holder);
}
} else {
// For CallSite objects treat the target field as a compile time constant.
diff --git a/src/hotspot/share/ci/ciInstanceKlass.cpp b/src/hotspot/share/ci/ciInstanceKlass.cpp
index 33bcabc4566..6243258acd9 100644
--- a/src/hotspot/share/ci/ciInstanceKlass.cpp
+++ b/src/hotspot/share/ci/ciInstanceKlass.cpp
@@ -392,7 +392,7 @@ bool ciInstanceKlass::contains_field_offset(int offset) {
return get_instanceKlass()->contains_field_offset(offset);
}
-ciField* ciInstanceKlass::get_non_static_field_by_offset(const int field_offset) {
+ciField* ciInstanceKlass::get_nonstatic_field_by_offset(const int field_offset) {
for (int i = 0, len = nof_nonstatic_fields(); i < len; i++) {
ciField* field = _nonstatic_fields->at(i);
int field_off = field->offset_in_bytes();
@@ -406,7 +406,7 @@ ciField* ciInstanceKlass::get_non_static_field_by_offset(const int field_offset)
// ciInstanceKlass::get_field_by_offset
ciField* ciInstanceKlass::get_field_by_offset(int field_offset, bool is_static) {
if (!is_static) {
- return get_non_static_field_by_offset(field_offset);
+ return get_nonstatic_field_by_offset(field_offset);
}
VM_ENTRY_MARK;
InstanceKlass* k = get_instanceKlass();
@@ -437,7 +437,7 @@ ciField* ciInstanceKlass::get_field_by_name(ciSymbol* name, ciSymbol* signature,
// except this does not require allocating memory for a new ciField
BasicType ciInstanceKlass::get_field_type_by_offset(const int field_offset, const bool is_static) {
if (!is_static) {
- ciField* field = get_non_static_field_by_offset(field_offset);
+ ciField* field = get_nonstatic_field_by_offset(field_offset);
return field != nullptr ? field->layout_type() : T_ILLEGAL;
}
diff --git a/src/hotspot/share/ci/ciInstanceKlass.hpp b/src/hotspot/share/ci/ciInstanceKlass.hpp
index 8ccf1fadfb7..a84c63981c9 100644
--- a/src/hotspot/share/ci/ciInstanceKlass.hpp
+++ b/src/hotspot/share/ci/ciInstanceKlass.hpp
@@ -83,7 +83,7 @@ private:
bool compute_injected_fields_helper();
void compute_transitive_interfaces();
- ciField* get_non_static_field_by_offset(int field_offset);
+ ciField* get_nonstatic_field_by_offset(int field_offset);
protected:
ciInstanceKlass(Klass* k);
diff --git a/src/hotspot/share/ci/ciUtilities.cpp b/src/hotspot/share/ci/ciUtilities.cpp
index 0c5b4d9824f..011fa049275 100644
--- a/src/hotspot/share/ci/ciUtilities.cpp
+++ b/src/hotspot/share/ci/ciUtilities.cpp
@@ -42,10 +42,7 @@ const char* basictype_to_str(BasicType t) {
// ------------------------------------------------------------------
// card_table_base
-CardTable::CardValue* ci_card_table_address() {
- BarrierSet* bs = BarrierSet::barrier_set();
- CardTableBarrierSet* ctbs = barrier_set_cast(bs);
- CardTable* ct = ctbs->card_table();
- assert(!UseShenandoahGC, "Shenandoah byte_map_base is not constant.");
- return ct->byte_map_base();
+CardTable::CardValue* ci_card_table_address_const() {
+ CardTableBarrierSet* ctbs = CardTableBarrierSet::barrier_set();
+ return ctbs->card_table_base_const();
}
diff --git a/src/hotspot/share/ci/ciUtilities.hpp b/src/hotspot/share/ci/ciUtilities.hpp
index 75dbb03adf4..3333972d57d 100644
--- a/src/hotspot/share/ci/ciUtilities.hpp
+++ b/src/hotspot/share/ci/ciUtilities.hpp
@@ -51,9 +51,9 @@ inline const char* bool_to_str(bool b) {
const char* basictype_to_str(BasicType t);
-CardTable::CardValue* ci_card_table_address();
+CardTable::CardValue* ci_card_table_address_const();
template T ci_card_table_address_as() {
- return reinterpret_cast(ci_card_table_address());
+ return reinterpret_cast(ci_card_table_address_const());
}
#endif // SHARE_CI_CIUTILITIES_HPP
diff --git a/src/hotspot/share/classfile/classLoader.cpp b/src/hotspot/share/classfile/classLoader.cpp
index d9a63cd154b..eced83577cb 100644
--- a/src/hotspot/share/classfile/classLoader.cpp
+++ b/src/hotspot/share/classfile/classLoader.cpp
@@ -127,6 +127,7 @@ PerfCounter* ClassLoader::_perf_ik_link_methods_count = nullptr;
PerfCounter* ClassLoader::_perf_method_adapters_count = nullptr;
PerfCounter* ClassLoader::_unsafe_defineClassCallCounter = nullptr;
PerfCounter* ClassLoader::_perf_secondary_hash_time = nullptr;
+PerfCounter* ClassLoader::_perf_change_wx_time = nullptr;
PerfCounter* ClassLoader::_perf_resolve_indy_time = nullptr;
PerfCounter* ClassLoader::_perf_resolve_invokehandle_time = nullptr;
@@ -1370,6 +1371,7 @@ void ClassLoader::initialize(TRAPS) {
NEWPERFBYTECOUNTER(_perf_sys_classfile_bytes_read, SUN_CLS, "sysClassBytes");
NEWPERFEVENTCOUNTER(_unsafe_defineClassCallCounter, SUN_CLS, "unsafeDefineClassCalls");
NEWPERFTICKCOUNTER(_perf_secondary_hash_time, SUN_CLS, "secondarySuperHashTime");
+ NEWPERFTICKCOUNTER(_perf_change_wx_time, SUN_CLS, "changeWXTime");
if (log_is_enabled(Info, perf, class, link)) {
NEWPERFTICKCOUNTER(_perf_ik_link_methods_time, SUN_CLS, "linkMethodsTime");
@@ -1418,6 +1420,10 @@ char* ClassLoader::lookup_vm_options() {
jio_snprintf(modules_path, JVM_MAXPATHLEN, "%s%slib%smodules", Arguments::get_java_home(), fileSep, fileSep);
JImage_file =(*JImageOpen)(modules_path, &error);
if (JImage_file == nullptr) {
+ if (Arguments::has_jimage()) {
+ // The modules file exists but is unreadable or corrupt
+ vm_exit_during_initialization(err_msg("Unable to load %s", modules_path));
+ }
return nullptr;
}
diff --git a/src/hotspot/share/classfile/classLoader.hpp b/src/hotspot/share/classfile/classLoader.hpp
index afb0a581dcc..a935d3027ac 100644
--- a/src/hotspot/share/classfile/classLoader.hpp
+++ b/src/hotspot/share/classfile/classLoader.hpp
@@ -184,6 +184,7 @@ class ClassLoader: AllStatic {
// Count the time taken to hash the scondary superclass arrays.
static PerfCounter* _perf_secondary_hash_time;
+ static PerfCounter* _perf_change_wx_time;
// The boot class path consists of 3 ordered pieces:
// 1. the module/path pairs specified to --patch-module
@@ -268,6 +269,9 @@ class ClassLoader: AllStatic {
static PerfCounter* perf_secondary_hash_time() {
return _perf_secondary_hash_time;
}
+ static PerfCounter* perf_change_wx_time() {
+ return _perf_change_wx_time;
+ }
static PerfCounter* perf_sys_classload_time() { return _perf_sys_classload_time; }
static PerfCounter* perf_app_classload_time() { return _perf_app_classload_time; }
static PerfCounter* perf_app_classload_selftime() { return _perf_app_classload_selftime; }
diff --git a/src/hotspot/share/classfile/javaClasses.cpp b/src/hotspot/share/classfile/javaClasses.cpp
index dd70d7b49ab..c6b0fcb90e0 100644
--- a/src/hotspot/share/classfile/javaClasses.cpp
+++ b/src/hotspot/share/classfile/javaClasses.cpp
@@ -1263,6 +1263,10 @@ bool java_lang_Class::restore_archived_mirror(Klass *k,
"Restored %s archived mirror " PTR_FORMAT, k->external_name(), p2i(mirror()));
}
+ if (CDSConfig::is_dumping_heap()) {
+ create_scratch_mirror(k, CHECK_(false));
+ }
+
return true;
}
#endif // INCLUDE_CDS_JAVA_HEAP
@@ -1904,16 +1908,16 @@ oop java_lang_Thread::park_blocker(oop java_thread) {
return java_thread->obj_field_access(_park_blocker_offset);
}
-// Obtain stack trace for platform or mounted virtual thread.
-// If jthread is a virtual thread and it has been unmounted (or remounted to different carrier) the method returns null.
-// The caller (java.lang.VirtualThread) handles returned nulls via retry.
+// Obtain stack trace for a platform or virtual thread.
oop java_lang_Thread::async_get_stack_trace(jobject jthread, TRAPS) {
ThreadsListHandle tlh(THREAD);
JavaThread* java_thread = nullptr;
- oop thread_oop;
+ oop thread_oop = nullptr;
bool has_java_thread = tlh.cv_internal_thread_to_JavaThread(jthread, &java_thread, &thread_oop);
- if (!has_java_thread) {
+ assert(thread_oop != nullptr, "Missing Thread oop");
+ bool is_virtual = java_lang_VirtualThread::is_instance(thread_oop);
+ if (!has_java_thread && !is_virtual) {
return nullptr;
}
@@ -1921,12 +1925,11 @@ oop java_lang_Thread::async_get_stack_trace(jobject jthread, TRAPS) {
public:
const Handle _thread_h;
int _depth;
- bool _retry_handshake;
GrowableArray* _methods;
GrowableArray* _bcis;
GetStackTraceHandshakeClosure(Handle thread_h) :
- HandshakeClosure("GetStackTraceHandshakeClosure"), _thread_h(thread_h), _depth(0), _retry_handshake(false),
+ HandshakeClosure("GetStackTraceHandshakeClosure"), _thread_h(thread_h), _depth(0),
_methods(nullptr), _bcis(nullptr) {
}
~GetStackTraceHandshakeClosure() {
@@ -1934,37 +1937,15 @@ oop java_lang_Thread::async_get_stack_trace(jobject jthread, TRAPS) {
delete _bcis;
}
- bool read_reset_retry() {
- bool ret = _retry_handshake;
- // If we re-execute the handshake this method need to return false
- // when the handshake cannot be performed. (E.g. thread terminating)
- _retry_handshake = false;
- return ret;
- }
-
void do_thread(Thread* th) {
- if (!Thread::current()->is_Java_thread()) {
- _retry_handshake = true;
+ JavaThread* java_thread = th != nullptr ? JavaThread::cast(th) : nullptr;
+ if (java_thread != nullptr && !java_thread->has_last_Java_frame()) {
+ // stack trace is empty
return;
}
- JavaThread* java_thread = JavaThread::cast(th);
-
- if (!java_thread->has_last_Java_frame()) {
- return;
- }
-
- bool carrier = false;
- if (java_lang_VirtualThread::is_instance(_thread_h())) {
- // Ensure _thread_h is still mounted to java_thread.
- const ContinuationEntry* ce = java_thread->vthread_continuation();
- if (ce == nullptr || ce->cont_oop(java_thread) != java_lang_VirtualThread::continuation(_thread_h())) {
- // Target thread has been unmounted.
- return;
- }
- } else {
- carrier = (java_thread->vthread_continuation() != nullptr);
- }
+ bool is_virtual = java_lang_VirtualThread::is_instance(_thread_h());
+ bool vthread_carrier = !is_virtual && (java_thread->vthread_continuation() != nullptr);
const int max_depth = MaxJavaStackTraceDepth;
const bool skip_hidden = !ShowHiddenFrames;
@@ -1975,7 +1956,10 @@ oop java_lang_Thread::async_get_stack_trace(jobject jthread, TRAPS) {
_bcis = new (mtInternal) GrowableArray(init_length, mtInternal);
int total_count = 0;
- for (vframeStream vfst(java_thread, false, false, carrier); // we don't process frames as we don't care about oops
+ vframeStream vfst(java_thread != nullptr
+ ? vframeStream(java_thread, false, false, vthread_carrier) // we don't process frames as we don't care about oops
+ : vframeStream(java_lang_VirtualThread::continuation(_thread_h())));
+ for (;
!vfst.at_end() && (max_depth == 0 || max_depth != total_count);
vfst.next()) {
@@ -1997,9 +1981,11 @@ oop java_lang_Thread::async_get_stack_trace(jobject jthread, TRAPS) {
ResourceMark rm(THREAD);
HandleMark hm(THREAD);
GetStackTraceHandshakeClosure gsthc(Handle(THREAD, thread_oop));
- do {
- Handshake::execute(&gsthc, &tlh, java_thread);
- } while (gsthc.read_reset_retry());
+ if (is_virtual) {
+ Handshake::execute(&gsthc, thread_oop);
+ } else {
+ Handshake::execute(&gsthc, &tlh, java_thread);
+ }
// Stop if no stack trace is found.
if (gsthc._depth == 0) {
@@ -2196,7 +2182,7 @@ void java_lang_VirtualThread::set_timeout(oop vthread, jlong value) {
JavaThreadStatus java_lang_VirtualThread::map_state_to_thread_status(int state) {
JavaThreadStatus status = JavaThreadStatus::NEW;
- switch (state & ~SUSPENDED) {
+ switch (state) {
case NEW:
status = JavaThreadStatus::NEW;
break;
diff --git a/src/hotspot/share/classfile/javaClasses.hpp b/src/hotspot/share/classfile/javaClasses.hpp
index a8562a345c8..3276d398faf 100644
--- a/src/hotspot/share/classfile/javaClasses.hpp
+++ b/src/hotspot/share/classfile/javaClasses.hpp
@@ -592,9 +592,6 @@ class java_lang_VirtualThread : AllStatic {
TIMED_WAITING = 17,
TIMED_WAIT = 18, // waiting in timed-Object.wait
TERMINATED = 99,
-
- // additional state bits
- SUSPENDED = 1 << 8, // suspended when unmounted
};
static void compute_offsets();
diff --git a/src/hotspot/share/classfile/stringTable.cpp b/src/hotspot/share/classfile/stringTable.cpp
index 20dfad0d980..2b8b7780a41 100644
--- a/src/hotspot/share/classfile/stringTable.cpp
+++ b/src/hotspot/share/classfile/stringTable.cpp
@@ -74,24 +74,9 @@ const size_t REHASH_LEN = 100;
const double CLEAN_DEAD_HIGH_WATER_MARK = 0.5;
#if INCLUDE_CDS_JAVA_HEAP
-bool StringTable::_is_two_dimensional_shared_strings_array = false;
-OopHandle StringTable::_shared_strings_array;
-int StringTable::_shared_strings_array_root_index;
-
inline oop StringTable::read_string_from_compact_hashtable(address base_address, u4 index) {
assert(AOTMappedHeapLoader::is_in_use(), "sanity");
- objArrayOop array = (objArrayOop)(_shared_strings_array.resolve());
- oop s;
-
- if (!_is_two_dimensional_shared_strings_array) {
- s = array->obj_at((int)index);
- } else {
- int primary_index = index >> _secondary_array_index_bits;
- int secondary_index = index & _secondary_array_index_mask;
- objArrayOop secondary = (objArrayOop)array->obj_at(primary_index);
- s = secondary->obj_at(secondary_index);
- }
-
+ oop s = HeapShared::get_root((int)index, false);
assert(java_lang_String::is_instance(s), "must be");
return s;
}
@@ -115,7 +100,6 @@ OopStorage* StringTable::_oop_storage;
static size_t _current_size = 0;
static volatile size_t _items_count = 0;
-DEBUG_ONLY(static bool _disable_interning_during_cds_dump = false);
volatile bool _alt_hash = false;
@@ -317,12 +301,6 @@ void StringTable::create_table() {
_oop_storage->register_num_dead_callback(&gc_notification);
}
-#if INCLUDE_CDS_JAVA_HEAP
-void StringTable::load_shared_strings_array() {
- _shared_strings_array = OopHandle(Universe::vm_global(), HeapShared::get_root(_shared_strings_array_root_index));
-}
-#endif
-
void StringTable::item_added() {
AtomicAccess::inc(&_items_count);
}
@@ -509,9 +487,6 @@ oop StringTable::intern(const char* utf8_string, TRAPS) {
}
oop StringTable::intern(const StringWrapper& name, TRAPS) {
- assert(!AtomicAccess::load_acquire(&_disable_interning_during_cds_dump),
- "All threads that may intern strings should have been stopped before CDS starts copying the interned string table");
-
// shared table always uses java_lang_String::hash_code
unsigned int hash = hash_wrapped_string(name);
oop found_string = lookup_shared(name, hash);
@@ -957,118 +932,13 @@ oop StringTable::lookup_shared(const jchar* name, int len) {
return _shared_table.lookup(wrapped_name, java_lang_String::hash_code(name, len), 0);
}
-// This is called BEFORE we enter the CDS safepoint. We can still allocate Java object arrays to
-// be used by the shared strings table.
-void StringTable::allocate_shared_strings_array(TRAPS) {
- if (!CDSConfig::is_dumping_heap()) {
- return;
- }
+void StringTable::init_shared_table() {
+ assert(SafepointSynchronize::is_at_safepoint(), "inside AOT safepoint");
+ precond(CDSConfig::is_dumping_heap());
+ assert(HeapShared::is_writing_mapping_mode(), "not used for streamed oops");
- assert(HeapShared::is_writing_mapping_mode(), "should not reach here");
-
- CompileBroker::wait_for_no_active_tasks();
-
- precond(CDSConfig::allow_only_single_java_thread());
-
- // At this point, no more strings will be added:
- // - There's only a single Java thread (this thread). It no longer executes Java bytecodes
- // so JIT compilation will eventually stop.
- // - CompileBroker has no more active tasks, so all JIT requests have been processed.
-
- // This flag will be cleared after intern table dumping has completed, so we can run the
- // compiler again (for future AOT method compilation, etc).
- DEBUG_ONLY(AtomicAccess::release_store(&_disable_interning_during_cds_dump, true));
-
- if (items_count_acquire() > (size_t)max_jint) {
- fatal("Too many strings to be archived: %zu", items_count_acquire());
- }
-
- int total = (int)items_count_acquire();
- size_t single_array_size = objArrayOopDesc::object_size(total);
-
- log_info(aot)("allocated string table for %d strings", total);
-
- if (!HeapShared::is_too_large_to_archive(single_array_size)) {
- // The entire table can fit in a single array
- objArrayOop array = oopFactory::new_objArray(vmClasses::Object_klass(), total, CHECK);
- _shared_strings_array = OopHandle(Universe::vm_global(), array);
- log_info(aot)("string table array (single level) length = %d", total);
- } else {
- // Split the table in two levels of arrays.
- int primary_array_length = (total + _secondary_array_max_length - 1) / _secondary_array_max_length;
- size_t primary_array_size = objArrayOopDesc::object_size(primary_array_length);
- size_t secondary_array_size = objArrayOopDesc::object_size(_secondary_array_max_length);
-
- if (HeapShared::is_too_large_to_archive(secondary_array_size)) {
- // This can only happen if you have an extremely large number of classes that
- // refer to more than 16384 * 16384 = 26M interned strings! Not a practical concern
- // but bail out for safety.
- log_error(aot)("Too many strings to be archived: %zu", items_count_acquire());
- AOTMetaspace::unrecoverable_writing_error();
- }
-
- objArrayOop primary = oopFactory::new_objArray(vmClasses::Object_klass(), primary_array_length, CHECK);
- objArrayHandle primaryHandle(THREAD, primary);
- _shared_strings_array = OopHandle(Universe::vm_global(), primary);
-
- log_info(aot)("string table array (primary) length = %d", primary_array_length);
- for (int i = 0; i < primary_array_length; i++) {
- int len;
- if (total > _secondary_array_max_length) {
- len = _secondary_array_max_length;
- } else {
- len = total;
- }
- total -= len;
-
- objArrayOop secondary = oopFactory::new_objArray(vmClasses::Object_klass(), len, CHECK);
- primaryHandle()->obj_at_put(i, secondary);
-
- log_info(aot)("string table array (secondary)[%d] length = %d", i, len);
- assert(!HeapShared::is_too_large_to_archive(secondary), "sanity");
- }
-
- assert(total == 0, "must be");
- _is_two_dimensional_shared_strings_array = true;
- }
-}
-
-#ifndef PRODUCT
-void StringTable::verify_secondary_array_index_bits() {
- assert(HeapShared::is_writing_mapping_mode(), "should not reach here");
- int max;
- for (max = 1; ; max++) {
- size_t next_size = objArrayOopDesc::object_size(1 << (max + 1));
- if (HeapShared::is_too_large_to_archive(next_size)) {
- break;
- }
- }
- // Currently max is 17 for +UseCompressedOops, 16 for -UseCompressedOops.
- // When we add support for Shenandoah (which has a smaller mininum region size than G1),
- // max will become 15/14.
- //
- // We use _secondary_array_index_bits==14 as that will be the eventual value, and will
- // make testing easier.
- assert(_secondary_array_index_bits <= max,
- "_secondary_array_index_bits (%d) must be smaller than max possible value (%d)",
- _secondary_array_index_bits, max);
-}
-#endif // PRODUCT
-
-// This is called AFTER we enter the CDS safepoint.
-//
-// For each shared string:
-// [1] Store it into _shared_strings_array. Encode its position as a 32-bit index.
-// [2] Store the index and hashcode into _shared_table.
-oop StringTable::init_shared_strings_array() {
- assert(CDSConfig::is_dumping_heap(), "must be");
- assert(HeapShared::is_writing_mapping_mode(), "should not reach here");
- objArrayOop array = (objArrayOop)(_shared_strings_array.resolve());
-
- verify_secondary_array_index_bits();
-
- int index = 0;
- auto copy_into_array = [&] (WeakHandle* val) {
+ int n = 0;
+ auto copy_into_aot_heap = [&] (WeakHandle* val) {
oop string = val->peek();
if (string != nullptr && !HeapShared::is_string_too_large_to_archive(string)) {
// If string is too large, don't put it into the string table.
@@ -1077,53 +947,34 @@ oop StringTable::init_shared_strings_array() {
// - If there's a reference to it, we will report an error inside HeapShared.cpp and
// dumping will fail.
HeapShared::add_to_dumped_interned_strings(string);
- if (!_is_two_dimensional_shared_strings_array) {
- assert(index < array->length(), "no strings should have been added");
- array->obj_at_put(index, string);
- } else {
- int primary_index = index >> _secondary_array_index_bits;
- int secondary_index = index & _secondary_array_index_mask;
-
- assert(primary_index < array->length(), "no strings should have been added");
- objArrayOop secondary = (objArrayOop)array->obj_at(primary_index);
-
- assert(secondary != nullptr && secondary->is_objArray(), "must be");
- assert(secondary_index < secondary->length(), "no strings should have been added");
- secondary->obj_at_put(secondary_index, string);
- }
- index ++;
}
+ n++;
return true;
};
- _local_table->do_safepoint_scan(copy_into_array);
- log_info(aot)("Archived %d interned strings", index);
- return array;
+ _local_table->do_safepoint_scan(copy_into_aot_heap);
+ log_info(aot)("Archived %d interned strings", n);
};
void StringTable::write_shared_table() {
+ assert(SafepointSynchronize::is_at_safepoint(), "inside AOT safepoint");
+ precond(CDSConfig::is_dumping_heap());
+ assert(HeapShared::is_writing_mapping_mode(), "not used for streamed oops");
+
_shared_table.reset();
CompactHashtableWriter writer((int)items_count_acquire(), ArchiveBuilder::string_stats());
- int index = 0;
auto copy_into_shared_table = [&] (WeakHandle* val) {
oop string = val->peek();
if (string != nullptr && !HeapShared::is_string_too_large_to_archive(string)) {
unsigned int hash = java_lang_String::hash_code(string);
- writer.add(hash, index);
- index ++;
+ int root_id = HeapShared::append_root(string);
+ writer.add(hash, root_id);
}
return true;
};
_local_table->do_safepoint_scan(copy_into_shared_table);
writer.dump(&_shared_table, "string");
-
- DEBUG_ONLY(AtomicAccess::release_store(&_disable_interning_during_cds_dump, false));
-}
-
-void StringTable::set_shared_strings_array_index(int root_index) {
- assert(HeapShared::is_writing_mapping_mode(), "should not reach here");
- _shared_strings_array_root_index = root_index;
}
void StringTable::serialize_shared_table_header(SerializeClosure* soc) {
@@ -1135,8 +986,27 @@ void StringTable::serialize_shared_table_header(SerializeClosure* soc) {
} else if (!AOTMappedHeapLoader::is_in_use()) {
_shared_table.reset();
}
+}
- soc->do_bool(&_is_two_dimensional_shared_strings_array);
- soc->do_int(&_shared_strings_array_root_index);
+void StringTable::move_shared_strings_into_runtime_table() {
+ precond(CDSConfig::is_dumping_final_static_archive());
+ JavaThread* THREAD = JavaThread::current();
+ HandleMark hm(THREAD);
+
+ int n = 0;
+ _shared_table.iterate_all([&](oop string) {
+ int length = java_lang_String::length(string);
+ Handle h_string (THREAD, string);
+ StringWrapper name(h_string, length);
+ unsigned int hash = hash_wrapped_string(name);
+
+ assert(!_alt_hash, "too early");
+ oop interned = do_intern(name, hash, THREAD);
+ assert(string == interned, "must be");
+ n++;
+ });
+
+ _shared_table.reset();
+ log_info(aot)("Moved %d interned strings to runtime table", n);
}
#endif //INCLUDE_CDS_JAVA_HEAP
diff --git a/src/hotspot/share/classfile/stringTable.hpp b/src/hotspot/share/classfile/stringTable.hpp
index 839e9d9053d..0024a45a2f2 100644
--- a/src/hotspot/share/classfile/stringTable.hpp
+++ b/src/hotspot/share/classfile/stringTable.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -109,49 +109,17 @@ public:
static bool needs_rehashing() { return _needs_rehashing; }
static inline void update_needs_rehash(bool rehash);
- // Sharing
-#if INCLUDE_CDS_JAVA_HEAP
- static inline oop read_string_from_compact_hashtable(address base_address, u4 index);
-
+ // AOT support
+ static inline oop read_string_from_compact_hashtable(address base_address, u4 index) NOT_CDS_JAVA_HEAP_RETURN_(nullptr);
private:
- static bool _is_two_dimensional_shared_strings_array;
- static OopHandle _shared_strings_array;
- static int _shared_strings_array_root_index;
-
- // All the shared strings are referenced through _shared_strings_array to keep them alive.
- // Each shared string is stored as a 32-bit index in ::_shared_table. The index
- // is interpreted in two ways:
- //
- // [1] _is_two_dimensional_shared_strings_array = false: _shared_strings_array is an Object[].
- // Each shared string is stored as _shared_strings_array[index]
- //
- // [2] _is_two_dimensional_shared_strings_array = true: _shared_strings_array is an Object[][]
- // This happens when there are too many elements in the shared table. We store them
- // using two levels of objArrays, such that none of the arrays are too big for
- // AOTMappedHeapWriter::is_too_large_to_archive(). In this case, the index is splited into two
- // parts. Each shared string is stored as _shared_strings_array[primary_index][secondary_index]:
- //
- // [bits 31 .. 14][ bits 13 .. 0 ]
- // primary_index secondary_index
- const static int _secondary_array_index_bits = 14;
- const static int _secondary_array_max_length = 1 << _secondary_array_index_bits;
- const static int _secondary_array_index_mask = _secondary_array_max_length - 1;
-
- // make sure _secondary_array_index_bits is not too big
- static void verify_secondary_array_index_bits() PRODUCT_RETURN;
-#endif // INCLUDE_CDS_JAVA_HEAP
-
- private:
static oop lookup_shared(const StringWrapper& name, unsigned int hash) NOT_CDS_JAVA_HEAP_RETURN_(nullptr);
- public:
+public:
static oop lookup_shared(const jchar* name, int len) NOT_CDS_JAVA_HEAP_RETURN_(nullptr);
static size_t shared_entry_count() NOT_CDS_JAVA_HEAP_RETURN_(0);
- static void allocate_shared_strings_array(TRAPS) NOT_CDS_JAVA_HEAP_RETURN;
- static void load_shared_strings_array() NOT_CDS_JAVA_HEAP_RETURN;
- static oop init_shared_strings_array() NOT_CDS_JAVA_HEAP_RETURN_(nullptr);
+ static void init_shared_table() NOT_CDS_JAVA_HEAP_RETURN;
static void write_shared_table() NOT_CDS_JAVA_HEAP_RETURN;
- static void set_shared_strings_array_index(int root_index) NOT_CDS_JAVA_HEAP_RETURN;
static void serialize_shared_table_header(SerializeClosure* soc) NOT_CDS_JAVA_HEAP_RETURN;
+ static void move_shared_strings_into_runtime_table();
// Jcmd
static void dump(outputStream* st, bool verbose=false);
diff --git a/src/hotspot/share/classfile/verifier.cpp b/src/hotspot/share/classfile/verifier.cpp
index 38dba1d3d5f..30f147b9ae7 100644
--- a/src/hotspot/share/classfile/verifier.cpp
+++ b/src/hotspot/share/classfile/verifier.cpp
@@ -190,9 +190,8 @@ bool Verifier::verify(InstanceKlass* klass, bool should_verify_class, TRAPS) {
// effect (sic!) for external_name(), but instead of doing that, we opt to
// explicitly push the hashcode in here. This is signify the following block
// is IMPORTANT:
- if (klass->java_mirror() != nullptr) {
- klass->java_mirror()->identity_hash();
- }
+ assert(klass->java_mirror() != nullptr, "must be");
+ klass->java_mirror()->identity_hash();
if (!is_eligible_for_verification(klass, should_verify_class)) {
return true;
diff --git a/src/hotspot/share/classfile/vmIntrinsics.hpp b/src/hotspot/share/classfile/vmIntrinsics.hpp
index 07fa294e8e1..75592fd61c8 100644
--- a/src/hotspot/share/classfile/vmIntrinsics.hpp
+++ b/src/hotspot/share/classfile/vmIntrinsics.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2020, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2020, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -415,18 +415,18 @@ class methodHandle;
\
do_class(java_lang_StringCoding, "java/lang/StringCoding") \
do_intrinsic(_countPositives, java_lang_StringCoding, countPositives_name, countPositives_signature, F_S) \
- do_name( countPositives_name, "countPositives") \
+ do_name( countPositives_name, "countPositives0") \
do_signature(countPositives_signature, "([BII)I") \
\
do_class(sun_nio_cs_iso8859_1_Encoder, "sun/nio/cs/ISO_8859_1$Encoder") \
do_intrinsic(_encodeISOArray, sun_nio_cs_iso8859_1_Encoder, encodeISOArray_name, encodeISOArray_signature, F_S) \
- do_name( encodeISOArray_name, "implEncodeISOArray") \
+ do_name( encodeISOArray_name, "encodeISOArray0") \
do_signature(encodeISOArray_signature, "([CI[BII)I") \
\
do_intrinsic(_encodeByteISOArray, java_lang_StringCoding, encodeISOArray_name, indexOfI_signature, F_S) \
\
do_intrinsic(_encodeAsciiArray, java_lang_StringCoding, encodeAsciiArray_name, encodeISOArray_signature, F_S) \
- do_name( encodeAsciiArray_name, "implEncodeAsciiArray") \
+ do_name( encodeAsciiArray_name, "encodeAsciiArray0") \
\
do_class(java_math_BigInteger, "java/math/BigInteger") \
do_intrinsic(_multiplyToLen, java_math_BigInteger, multiplyToLen_name, multiplyToLen_signature, F_S) \
@@ -1003,7 +1003,7 @@ class methodHandle;
do_signature(vector_unary_op_sig, "(I" \
"Ljava/lang/Class;" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;" \
"Ljdk/internal/vm/vector/VectorSupport$VectorMask;" \
@@ -1015,7 +1015,7 @@ class methodHandle;
do_signature(vector_binary_op_sig, "(I" \
"Ljava/lang/Class;" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljdk/internal/vm/vector/VectorSupport$VectorPayload;" \
"Ljdk/internal/vm/vector/VectorSupport$VectorPayload;" \
@@ -1051,7 +1051,7 @@ class methodHandle;
do_signature(vector_ternary_op_sig, "(I" \
"Ljava/lang/Class;" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;" \
@@ -1063,7 +1063,7 @@ class methodHandle;
\
do_intrinsic(_VectorSelectFromTwoVectorOp, jdk_internal_vm_vector_VectorSupport, vector_select_from_op_name, vector_select_from_op_sig, F_S) \
do_signature(vector_select_from_op_sig, "(Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;" \
@@ -1074,7 +1074,7 @@ class methodHandle;
\
do_intrinsic(_VectorFromBitsCoerced, jdk_internal_vm_vector_VectorSupport, vector_frombits_coerced_name, vector_frombits_coerced_sig, F_S) \
do_signature(vector_frombits_coerced_sig, "(Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"J" \
"I" \
@@ -1085,7 +1085,7 @@ class methodHandle;
\
do_intrinsic(_VectorLoadOp, jdk_internal_vm_vector_VectorSupport, vector_load_op_name, vector_load_op_sig, F_S) \
do_signature(vector_load_op_sig, "(Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljava/lang/Object;" \
"J" \
@@ -1100,7 +1100,7 @@ class methodHandle;
do_intrinsic(_VectorLoadMaskedOp, jdk_internal_vm_vector_VectorSupport, vector_load_masked_op_name, vector_load_masked_op_sig, F_S) \
do_signature(vector_load_masked_op_sig, "(Ljava/lang/Class;" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljava/lang/Object;" \
"J" \
@@ -1116,7 +1116,7 @@ class methodHandle;
\
do_intrinsic(_VectorStoreOp, jdk_internal_vm_vector_VectorSupport, vector_store_op_name, vector_store_op_sig, F_S) \
do_signature(vector_store_op_sig, "(Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljava/lang/Object;" \
"J" \
@@ -1131,7 +1131,7 @@ class methodHandle;
do_intrinsic(_VectorStoreMaskedOp, jdk_internal_vm_vector_VectorSupport, vector_store_masked_op_name, vector_store_masked_op_sig, F_S) \
do_signature(vector_store_masked_op_sig, "(Ljava/lang/Class;" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljava/lang/Object;" \
"J" \
@@ -1148,7 +1148,7 @@ class methodHandle;
do_signature(vector_reduction_coerced_sig, "(I" \
"Ljava/lang/Class;" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;" \
"Ljdk/internal/vm/vector/VectorSupport$VectorMask;" \
@@ -1159,7 +1159,7 @@ class methodHandle;
do_intrinsic(_VectorTest, jdk_internal_vm_vector_VectorSupport, vector_test_name, vector_test_sig, F_S) \
do_signature(vector_test_sig, "(I" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljdk/internal/vm/vector/VectorSupport$VectorMask;" \
"Ljdk/internal/vm/vector/VectorSupport$VectorMask;" \
@@ -1170,7 +1170,7 @@ class methodHandle;
do_intrinsic(_VectorBlend, jdk_internal_vm_vector_VectorSupport, vector_blend_name, vector_blend_sig, F_S) \
do_signature(vector_blend_sig, "(Ljava/lang/Class;" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;" \
@@ -1182,7 +1182,8 @@ class methodHandle;
do_intrinsic(_VectorCompare, jdk_internal_vm_vector_VectorSupport, vector_compare_name, vector_compare_sig, F_S) \
do_signature(vector_compare_sig, "(I" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;Ljava/lang/Class;" \
+ "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;" \
@@ -1195,7 +1196,7 @@ class methodHandle;
do_signature(vector_rearrange_sig, "(Ljava/lang/Class;" \
"Ljava/lang/Class;" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;" \
"Ljdk/internal/vm/vector/VectorSupport$VectorShuffle;" \
@@ -1207,18 +1208,18 @@ class methodHandle;
do_intrinsic(_VectorSelectFrom, jdk_internal_vm_vector_VectorSupport, vector_select_from_name, vector_select_from_sig, F_S) \
do_signature(vector_select_from_sig, "(Ljava/lang/Class;" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;" \
"Ljdk/internal/vm/vector/VectorSupport$VectorMask;" \
"Ljdk/internal/vm/vector/VectorSupport$VectorSelectFromOp;)" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;") \
- do_name(vector_select_from_name, "selectFromOp") \
+ do_name(vector_select_from_name, "selectFromOp") \
\
do_intrinsic(_VectorExtract, jdk_internal_vm_vector_VectorSupport, vector_extract_name, vector_extract_sig, F_S) \
do_signature(vector_extract_sig, "(Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljdk/internal/vm/vector/VectorSupport$VectorPayload;" \
"I" \
@@ -1228,7 +1229,7 @@ class methodHandle;
\
do_intrinsic(_VectorInsert, jdk_internal_vm_vector_VectorSupport, vector_insert_name, vector_insert_sig, F_S) \
do_signature(vector_insert_sig, "(Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;" \
"IJ" \
@@ -1240,7 +1241,7 @@ class methodHandle;
do_signature(vector_broadcast_int_sig, "(I" \
"Ljava/lang/Class;" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;" \
"I" \
@@ -1252,10 +1253,10 @@ class methodHandle;
do_intrinsic(_VectorConvert, jdk_internal_vm_vector_VectorSupport, vector_convert_name, vector_convert_sig, F_S) \
do_signature(vector_convert_sig, "(I" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljdk/internal/vm/vector/VectorSupport$VectorPayload;" \
"Ljdk/internal/vm/vector/VectorSupport$VectorSpecies;" \
@@ -1266,7 +1267,7 @@ class methodHandle;
do_intrinsic(_VectorGatherOp, jdk_internal_vm_vector_VectorSupport, vector_gather_name, vector_gather_sig, F_S) \
do_signature(vector_gather_sig, "(Ljava/lang/Class;" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljava/lang/Class;" \
"I" \
@@ -1287,7 +1288,7 @@ class methodHandle;
do_intrinsic(_VectorScatterOp, jdk_internal_vm_vector_VectorSupport, vector_scatter_name, vector_scatter_sig, F_S) \
do_signature(vector_scatter_sig, "(Ljava/lang/Class;" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljava/lang/Class;" \
"I" \
@@ -1309,7 +1310,7 @@ class methodHandle;
do_intrinsic(_VectorMaskOp, jdk_internal_vm_vector_VectorSupport, vector_mask_oper_name, vector_mask_oper_sig, F_S) \
do_signature(vector_mask_oper_sig, "(I" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljdk/internal/vm/vector/VectorSupport$VectorMask;" \
"Ljdk/internal/vm/vector/VectorSupport$VectorMaskOp;)" \
@@ -1320,7 +1321,7 @@ class methodHandle;
do_signature(vector_compress_expand_op_sig, "(I" \
"Ljava/lang/Class;" \
"Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;" \
"Ljdk/internal/vm/vector/VectorSupport$VectorMask;" \
@@ -1330,7 +1331,7 @@ class methodHandle;
\
do_intrinsic(_IndexVector, jdk_internal_vm_vector_VectorSupport, index_vector_op_name, index_vector_op_sig, F_S) \
do_signature(index_vector_op_sig, "(Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"Ljdk/internal/vm/vector/VectorSupport$Vector;" \
"I" \
@@ -1341,7 +1342,7 @@ class methodHandle;
\
do_intrinsic(_IndexPartiallyInUpperRange, jdk_internal_vm_vector_VectorSupport, index_partially_in_upper_range_name, index_partially_in_upper_range_sig, F_S)\
do_signature(index_partially_in_upper_range_sig, "(Ljava/lang/Class;" \
- "Ljava/lang/Class;" \
+ "I" \
"I" \
"J" \
"J" \
diff --git a/src/hotspot/share/classfile/vmSymbols.hpp b/src/hotspot/share/classfile/vmSymbols.hpp
index 79646f24d0e..0054b7ba3f2 100644
--- a/src/hotspot/share/classfile/vmSymbols.hpp
+++ b/src/hotspot/share/classfile/vmSymbols.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -748,8 +748,6 @@ class SerializeClosure;
template(jdk_internal_vm_ThreadLock, "jdk/internal/vm/ThreadSnapshot$ThreadLock") \
template(jdk_internal_vm_ThreadLock_signature, "Ljdk/internal/vm/ThreadSnapshot$ThreadLock;") \
template(jdk_internal_vm_ThreadLock_array, "[Ljdk/internal/vm/ThreadSnapshot$ThreadLock;") \
- template(java_lang_StackTraceElement_of_name, "of") \
- template(java_lang_StackTraceElement_of_signature, "([Ljava/lang/StackTraceElement;)[Ljava/lang/StackTraceElement;") \
\
/* jcmd Thread.vthread_scheduler and Thread.vthread_pollers */ \
template(jdk_internal_vm_JcmdVThreadCommands, "jdk/internal/vm/JcmdVThreadCommands") \
diff --git a/src/hotspot/share/code/codeBlob.cpp b/src/hotspot/share/code/codeBlob.cpp
index a0a34ec23fa..094b4f82cf0 100644
--- a/src/hotspot/share/code/codeBlob.cpp
+++ b/src/hotspot/share/code/codeBlob.cpp
@@ -520,6 +520,8 @@ VtableBlob* VtableBlob::create(const char* name, int buffer_size) {
// eventually.
return nullptr;
}
+
+ MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
blob = new (size) VtableBlob(name, size);
CodeCache_lock->unlock();
}
diff --git a/src/hotspot/share/code/codeCache.cpp b/src/hotspot/share/code/codeCache.cpp
index 95a2fb908de..2a0256cc316 100644
--- a/src/hotspot/share/code/codeCache.cpp
+++ b/src/hotspot/share/code/codeCache.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -248,13 +248,61 @@ void CodeCache::initialize_heaps() {
set_size_of_unset_code_heap(&non_nmethod, cache_size, profiled.size + non_profiled.size, non_nmethod_min_size);
}
- size_t total = non_nmethod.size + profiled.size + non_profiled.size;
- if (total != cache_size && !cache_size_set) {
- log_info(codecache)("ReservedCodeCache size %zuK changed to total segments size NonNMethod "
- "%zuK NonProfiled %zuK Profiled %zuK = %zuK",
- cache_size/K, non_nmethod.size/K, non_profiled.size/K, profiled.size/K, total/K);
- // Adjust ReservedCodeCacheSize as necessary because it was not set explicitly
- cache_size = total;
+ // Note: if large page support is enabled, min_size is at least the large
+ // page size. This ensures that the code cache is covered by large pages.
+ non_nmethod.size = align_up(non_nmethod.size, min_size);
+ profiled.size = align_up(profiled.size, min_size);
+ non_profiled.size = align_up(non_profiled.size, min_size);
+
+ size_t aligned_total = non_nmethod.size + profiled.size + non_profiled.size;
+ if (!cache_size_set) {
+ // If ReservedCodeCacheSize is explicitly set and exceeds CODE_CACHE_SIZE_LIMIT,
+ // it is rejected by flag validation elsewhere. Here we only handle the case
+ // where ReservedCodeCacheSize is not set explicitly, but the computed segmented
+ // sizes (after alignment) exceed the platform limit.
+ if (aligned_total > CODE_CACHE_SIZE_LIMIT) {
+ err_msg message("ReservedCodeCacheSize (%zuK), Max (%zuK)."
+ "Segments: NonNMethod (%zuK), NonProfiled (%zuK), Profiled (%zuK).",
+ aligned_total/K, CODE_CACHE_SIZE_LIMIT/K,
+ non_nmethod.size/K, non_profiled.size/K, profiled.size/K);
+ vm_exit_during_initialization("Code cache size exceeds platform limit", message);
+ }
+ if (aligned_total != cache_size) {
+ log_info(codecache)("ReservedCodeCache size %zuK changed to total segments size NonNMethod "
+ "%zuK NonProfiled %zuK Profiled %zuK = %zuK",
+ cache_size/K, non_nmethod.size/K, non_profiled.size/K, profiled.size/K, aligned_total/K);
+ // Adjust ReservedCodeCacheSize as necessary because it was not set explicitly
+ cache_size = aligned_total;
+ }
+ } else {
+ check_min_size("reserved code cache", cache_size, min_cache_size);
+ // ReservedCodeCacheSize was set explicitly, so treat it as a hard cap.
+ // If alignment causes the total to exceed the cap, shrink unset heaps
+ // in min_size steps, never below their minimum sizes.
+ //
+ // A total smaller than cache_size typically happens when all segment sizes
+ // are explicitly set. In that case there is nothing to adjust, so we
+ // only validate the sizes.
+ if (aligned_total > cache_size) {
+ size_t delta = (aligned_total - cache_size) / min_size;
+ while (delta > 0) {
+ size_t start_delta = delta;
+ // Do not shrink the non-nmethod heap here: running out of non-nmethod space
+ // is more critical and may lead to unrecoverable VM errors.
+ if (non_profiled.enabled && !non_profiled.set && non_profiled.size > min_size) {
+ non_profiled.size -= min_size;
+ if (--delta == 0) break;
+ }
+ if (profiled.enabled && !profiled.set && profiled.size > min_size) {
+ profiled.size -= min_size;
+ delta--;
+ }
+ if (delta == start_delta) {
+ break;
+ }
+ }
+ aligned_total = non_nmethod.size + profiled.size + non_profiled.size;
+ }
}
log_debug(codecache)("Initializing code heaps ReservedCodeCache %zuK NonNMethod %zuK"
@@ -270,12 +318,9 @@ void CodeCache::initialize_heaps() {
if (non_profiled.enabled) { // non_profiled.enabled is always ON for segmented code heap, leave it checked for clarity
check_min_size("non-profiled code heap", non_profiled.size, min_size);
}
- if (cache_size_set) {
- check_min_size("reserved code cache", cache_size, min_cache_size);
- }
// ReservedCodeCacheSize was set explicitly, so report an error and abort if it doesn't match the segment sizes
- if (total != cache_size && cache_size_set) {
+ if (aligned_total != cache_size && cache_size_set) {
err_msg message("NonNMethodCodeHeapSize (%zuK)", non_nmethod.size/K);
if (profiled.enabled) {
message.append(" + ProfiledCodeHeapSize (%zuK)", profiled.size/K);
@@ -283,8 +328,8 @@ void CodeCache::initialize_heaps() {
if (non_profiled.enabled) {
message.append(" + NonProfiledCodeHeapSize (%zuK)", non_profiled.size/K);
}
- message.append(" = %zuK", total/K);
- message.append((total > cache_size) ? " is greater than " : " is less than ");
+ message.append(" = %zuK", aligned_total/K);
+ message.append((aligned_total > cache_size) ? " is greater than " : " is less than ");
message.append("ReservedCodeCacheSize (%zuK).", cache_size/K);
vm_exit_during_initialization("Invalid code heap sizes", message);
@@ -300,13 +345,6 @@ void CodeCache::initialize_heaps() {
}
}
- // Note: if large page support is enabled, min_size is at least the large
- // page size. This ensures that the code cache is covered by large pages.
- non_nmethod.size = align_up(non_nmethod.size, min_size);
- profiled.size = align_up(profiled.size, min_size);
- non_profiled.size = align_up(non_profiled.size, min_size);
- cache_size = non_nmethod.size + profiled.size + non_profiled.size;
-
FLAG_SET_ERGO(NonNMethodCodeHeapSize, non_nmethod.size);
FLAG_SET_ERGO(ProfiledCodeHeapSize, profiled.size);
FLAG_SET_ERGO(NonProfiledCodeHeapSize, non_profiled.size);
@@ -1101,7 +1139,7 @@ size_t CodeCache::freelists_length() {
void icache_init();
void CodeCache::initialize() {
- assert(CodeCacheSegmentSize >= (size_t)CodeEntryAlignment, "CodeCacheSegmentSize must be large enough to align entry points");
+ assert(CodeCacheSegmentSize >= CodeEntryAlignment, "CodeCacheSegmentSize must be large enough to align entry points");
#ifdef COMPILER2
assert(CodeCacheSegmentSize >= (size_t)OptoLoopAlignment, "CodeCacheSegmentSize must be large enough to align inner loops");
#endif
diff --git a/src/hotspot/share/code/dependencies.cpp b/src/hotspot/share/code/dependencies.cpp
index d90695739a1..dbfe1cd884e 100644
--- a/src/hotspot/share/code/dependencies.cpp
+++ b/src/hotspot/share/code/dependencies.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2005, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2005, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -1124,7 +1124,7 @@ class AbstractClassHierarchyWalker {
Klass* find_witness(InstanceKlass* context_type, KlassDepChange* changes = nullptr);
static void init();
- static void print_statistics();
+ NOT_PRODUCT(static void print_statistics();)
};
PerfCounter* AbstractClassHierarchyWalker::_perf_find_witness_anywhere_calls_count = nullptr;
@@ -2277,6 +2277,7 @@ bool KlassDepChange::involves_context(Klass* k) {
return is_contained;
}
+#ifndef PRODUCT
void Dependencies::print_statistics() {
AbstractClassHierarchyWalker::print_statistics();
}
@@ -2302,6 +2303,7 @@ void AbstractClassHierarchyWalker::print_statistics() {
}
}
}
+#endif
CallSiteDepChange::CallSiteDepChange(Handle call_site, Handle method_handle) :
_call_site(call_site),
diff --git a/src/hotspot/share/code/dependencies.hpp b/src/hotspot/share/code/dependencies.hpp
index d11c51a66dc..582a08183f9 100644
--- a/src/hotspot/share/code/dependencies.hpp
+++ b/src/hotspot/share/code/dependencies.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2005, 2024, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2005, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -649,7 +649,7 @@ class Dependencies: public ResourceObj {
};
friend class Dependencies::DepStream;
- static void print_statistics();
+ NOT_PRODUCT(static void print_statistics();)
};
diff --git a/src/hotspot/share/code/nmethod.cpp b/src/hotspot/share/code/nmethod.cpp
index edfca5c98ee..13eb1ff1604 100644
--- a/src/hotspot/share/code/nmethod.cpp
+++ b/src/hotspot/share/code/nmethod.cpp
@@ -2137,6 +2137,9 @@ void nmethod::make_deoptimized() {
ResourceMark rm;
RelocIterator iter(this, oops_reloc_begin());
+ // Assume there will be some calls to make deoptimized.
+ MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
+
while (iter.next()) {
switch (iter.type()) {
@@ -2213,6 +2216,7 @@ void nmethod::verify_clean_inline_caches() {
}
void nmethod::mark_as_maybe_on_stack() {
+ MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
AtomicAccess::store(&_gc_epoch, CodeCache::gc_epoch());
}
@@ -2305,6 +2309,8 @@ bool nmethod::make_not_entrant(InvalidationReason invalidation_reason) {
return false;
}
+ MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
+
{
// Enter critical section. Does not block for safepoint.
ConditionalMutexLocker ml(NMethodState_lock, !NMethodState_lock->owned_by_self(), Mutex::_no_safepoint_check_flag);
@@ -2740,6 +2746,8 @@ bool nmethod::is_unloading() {
state_is_unloading = IsUnloadingBehaviour::is_unloading(this);
uint8_t new_state = IsUnloadingState::create(state_is_unloading, state_unloading_cycle);
+ MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
+
// Note that if an nmethod has dead oops, everyone will agree that the
// nmethod is_unloading. However, the is_cold heuristics can yield
// different outcomes, so we guard the computed result with a CAS
diff --git a/src/hotspot/share/code/vtableStubs.cpp b/src/hotspot/share/code/vtableStubs.cpp
index b926888595d..35b226a8798 100644
--- a/src/hotspot/share/code/vtableStubs.cpp
+++ b/src/hotspot/share/code/vtableStubs.cpp
@@ -51,6 +51,9 @@ VMReg VtableStub::_receiver_location = VMRegImpl::Bad();
void* VtableStub::operator new(size_t size, int code_size) throw() {
assert_lock_strong(VtableStubs_lock);
assert(size == sizeof(VtableStub), "mismatched size");
+
+ MACOS_AARCH64_ONLY(os::thread_wx_enable_write());
+
// compute real VtableStub size (rounded to nearest word)
const int real_size = align_up(code_size + (int)sizeof(VtableStub), wordSize);
// malloc them in chunks to minimize header overhead
diff --git a/src/hotspot/share/compiler/compileBroker.cpp b/src/hotspot/share/compiler/compileBroker.cpp
index 574f4d6543b..7b236ed3589 100644
--- a/src/hotspot/share/compiler/compileBroker.cpp
+++ b/src/hotspot/share/compiler/compileBroker.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -2346,12 +2346,18 @@ void CompileBroker::invoke_compiler_on_method(CompileTask* task) {
/* Repeat compilation without installing code for profiling purposes */
int repeat_compilation_count = directive->RepeatCompilationOption;
- while (repeat_compilation_count > 0) {
- ResourceMark rm(thread);
- task->print_ul("NO CODE INSTALLED");
- thread->timeout()->reset();
- comp->compile_method(&ci_env, target, osr_bci, false, directive);
- repeat_compilation_count--;
+ if (repeat_compilation_count > 0) {
+ CHeapStringHolder failure_reason;
+ failure_reason.set(ci_env._failure_reason.get());
+ while (repeat_compilation_count > 0) {
+ ResourceMark rm(thread);
+ task->print_ul("NO CODE INSTALLED");
+ thread->timeout()->reset();
+ ci_env._failure_reason.clear();
+ comp->compile_method(&ci_env, target, osr_bci, false, directive);
+ repeat_compilation_count--;
+ }
+ ci_env._failure_reason.set(failure_reason.get());
}
}
diff --git a/src/hotspot/share/compiler/disassembler.cpp b/src/hotspot/share/compiler/disassembler.cpp
index c79c15e0f32..2c1ef235e07 100644
--- a/src/hotspot/share/compiler/disassembler.cpp
+++ b/src/hotspot/share/compiler/disassembler.cpp
@@ -607,10 +607,27 @@ void decode_env::print_address(address adr) {
return;
}
+ address card_table_base = nullptr;
BarrierSet* bs = BarrierSet::barrier_set();
- if (bs->is_a(BarrierSet::CardTableBarrierSet) &&
- adr == ci_card_table_address_as()) {
- st->print("word_map_base");
+#if INCLUDE_G1GC
+ if (bs->is_a(BarrierSet::G1BarrierSet)) {
+ G1BarrierSet* g1bs = barrier_set_cast(bs);
+ card_table_base = g1bs->card_table()->byte_map_base();
+ } else
+#endif
+#if INCLUDE_SHENANDOAHGC
+ if (bs->is_a(BarrierSet::ShenandoahBarrierSet)) {
+ ShenandoahBarrierSet* sbs = barrier_set_cast(bs);
+ if (sbs->card_table() != nullptr) {
+ card_table_base = sbs->card_table()->byte_map_base();
+ }
+ } else
+#endif
+ if (bs->is_a(BarrierSet::CardTableBarrierSet)) {
+ card_table_base = ci_card_table_address_as();
+ }
+ if (card_table_base != nullptr && adr == card_table_base) {
+ st->print("card_table_base");
if (WizardMode) st->print(" " INTPTR_FORMAT, p2i(adr));
return;
}
diff --git a/src/hotspot/share/gc/g1/c2/g1BarrierSetC2.cpp b/src/hotspot/share/gc/g1/c2/g1BarrierSetC2.cpp
index 61402301eb1..34d31702e80 100644
--- a/src/hotspot/share/gc/g1/c2/g1BarrierSetC2.cpp
+++ b/src/hotspot/share/gc/g1/c2/g1BarrierSetC2.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2018, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -351,7 +351,6 @@ Node* G1BarrierSetC2::store_at_resolved(C2Access& access, C2AccessValue& val) co
Node* G1BarrierSetC2::atomic_cmpxchg_val_at_resolved(C2AtomicParseAccess& access, Node* expected_val,
Node* new_val, const Type* value_type) const {
- GraphKit* kit = access.kit();
if (!access.is_oop()) {
return BarrierSetC2::atomic_cmpxchg_val_at_resolved(access, expected_val, new_val, value_type);
}
@@ -361,7 +360,6 @@ Node* G1BarrierSetC2::atomic_cmpxchg_val_at_resolved(C2AtomicParseAccess& access
Node* G1BarrierSetC2::atomic_cmpxchg_bool_at_resolved(C2AtomicParseAccess& access, Node* expected_val,
Node* new_val, const Type* value_type) const {
- GraphKit* kit = access.kit();
if (!access.is_oop()) {
return BarrierSetC2::atomic_cmpxchg_bool_at_resolved(access, expected_val, new_val, value_type);
}
@@ -370,7 +368,6 @@ Node* G1BarrierSetC2::atomic_cmpxchg_bool_at_resolved(C2AtomicParseAccess& acces
}
Node* G1BarrierSetC2::atomic_xchg_at_resolved(C2AtomicParseAccess& access, Node* new_val, const Type* value_type) const {
- GraphKit* kit = access.kit();
if (!access.is_oop()) {
return BarrierSetC2::atomic_xchg_at_resolved(access, new_val, value_type);
}
diff --git a/src/hotspot/share/gc/g1/g1BarrierSet.cpp b/src/hotspot/share/gc/g1/g1BarrierSet.cpp
index 622651ce0d8..dee50500e07 100644
--- a/src/hotspot/share/gc/g1/g1BarrierSet.cpp
+++ b/src/hotspot/share/gc/g1/g1BarrierSet.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -64,13 +64,13 @@ G1BarrierSet::G1BarrierSet(G1CardTable* card_table,
{}
G1BarrierSet::~G1BarrierSet() {
- delete _refinement_table;
+ delete refinement_table();
}
void G1BarrierSet::swap_global_card_table() {
- G1CardTable* temp = static_cast(_card_table);
- _card_table = _refinement_table;
- _refinement_table = temp;
+ G1CardTable* temp = static_cast(card_table());
+ _card_table.store_relaxed(refinement_table());
+ _refinement_table.store_relaxed(temp);
}
void G1BarrierSet::update_card_table_base(Thread* thread) {
@@ -80,7 +80,7 @@ void G1BarrierSet::update_card_table_base(Thread* thread) {
assert(thread->is_Java_thread(), "may only update card table base of JavaThreads, not %s", thread->name());
}
#endif
- G1ThreadLocalData::set_byte_map_base(thread, _card_table->byte_map_base());
+ G1ThreadLocalData::set_byte_map_base(thread, card_table()->byte_map_base());
}
template void
@@ -135,10 +135,10 @@ void G1BarrierSet::write_region(MemRegion mr) {
// marks next time.
// If we write to the old card table (after the switching, then the refinement
// table) the oncoming handshake will do the memory synchronization.
- CardTable* card_table = AtomicAccess::load(&_card_table);
+ CardTable* local_card_table = card_table();
- volatile CardValue* byte = card_table->byte_for(mr.start());
- CardValue* last_byte = card_table->byte_for(mr.last());
+ volatile CardValue* byte = local_card_table->byte_for(mr.start());
+ CardValue* last_byte = local_card_table->byte_for(mr.last());
// Dirty cards only if necessary.
for (; byte <= last_byte; byte++) {
@@ -190,6 +190,6 @@ void G1BarrierSet::on_thread_detach(Thread* thread) {
}
void G1BarrierSet::print_on(outputStream* st) const {
- _card_table->print_on(st, "Card");
- _refinement_table->print_on(st, "Refinement");
+ card_table()->print_on(st, "Card");
+ refinement_table()->print_on(st, "Refinement");
}
diff --git a/src/hotspot/share/gc/g1/g1BarrierSet.hpp b/src/hotspot/share/gc/g1/g1BarrierSet.hpp
index bf595973a32..406096acf10 100644
--- a/src/hotspot/share/gc/g1/g1BarrierSet.hpp
+++ b/src/hotspot/share/gc/g1/g1BarrierSet.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -29,6 +29,7 @@
#include "gc/shared/bufferNode.hpp"
#include "gc/shared/cardTable.hpp"
#include "gc/shared/cardTableBarrierSet.hpp"
+#include "runtime/atomic.hpp"
class G1CardTable;
class Thread;
@@ -66,7 +67,7 @@ class G1BarrierSet: public CardTableBarrierSet {
BufferNode::Allocator _satb_mark_queue_buffer_allocator;
G1SATBMarkQueueSet _satb_mark_queue_set;
- G1CardTable* _refinement_table;
+ Atomic _refinement_table;
public:
G1BarrierSet(G1CardTable* card_table, G1CardTable* refinement_table);
@@ -76,7 +77,7 @@ class G1BarrierSet: public CardTableBarrierSet {
return barrier_set_cast(BarrierSet::barrier_set());
}
- G1CardTable* refinement_table() const { return _refinement_table; }
+ G1CardTable* refinement_table() const { return _refinement_table.load_relaxed(); }
// Swap the global card table references, without synchronization.
void swap_global_card_table();
diff --git a/src/hotspot/share/gc/g1/g1BarrierSet.inline.hpp b/src/hotspot/share/gc/g1/g1BarrierSet.inline.hpp
index 794e5db0634..54892c9191d 100644
--- a/src/hotspot/share/gc/g1/g1BarrierSet.inline.hpp
+++ b/src/hotspot/share/gc/g1/g1BarrierSet.inline.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2016, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2016, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -73,8 +73,8 @@ inline void G1BarrierSet::write_ref_field_post(T* field) {
// Make sure that the card table reference is read only once. Otherwise the compiler
// might reload that value in the two accesses below, that could cause writes to
// the wrong card table.
- CardTable* card_table = AtomicAccess::load(&_card_table);
- CardValue* byte = card_table->byte_for(field);
+ CardTable* local_card_table = card_table();
+ CardValue* byte = local_card_table->byte_for(field);
if (*byte == G1CardTable::clean_card_val()) {
*byte = G1CardTable::dirty_card_val();
}
diff --git a/src/hotspot/share/gc/g1/g1BlockOffsetTable.cpp b/src/hotspot/share/gc/g1/g1BlockOffsetTable.cpp
index 4653f96980d..c695ad977fe 100644
--- a/src/hotspot/share/gc/g1/g1BlockOffsetTable.cpp
+++ b/src/hotspot/share/gc/g1/g1BlockOffsetTable.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -40,26 +40,26 @@ G1BlockOffsetTable::G1BlockOffsetTable(MemRegion heap, G1RegionToSpaceMapper* st
MemRegion bot_reserved = storage->reserved();
- _offset_base = ((uint8_t*)bot_reserved.start() - (uintptr_t(_reserved.start()) >> CardTable::card_shift()));
+ _offset_base = ((Atomic*)bot_reserved.start() - (uintptr_t(_reserved.start()) >> CardTable::card_shift()));
log_trace(gc, bot)("G1BlockOffsetTable::G1BlockOffsetTable: ");
log_trace(gc, bot)(" rs.base(): " PTR_FORMAT " rs.size(): %zu rs end(): " PTR_FORMAT,
p2i(bot_reserved.start()), bot_reserved.byte_size(), p2i(bot_reserved.end()));
}
-void G1BlockOffsetTable::set_offset_array(uint8_t* addr, uint8_t offset) {
+void G1BlockOffsetTable::set_offset_array(Atomic* addr, uint8_t offset) {
check_address(addr, "Block offset table address out of range");
- AtomicAccess::store(addr, offset);
+ addr->store_relaxed(offset);
}
-void G1BlockOffsetTable::set_offset_array(uint8_t* addr, HeapWord* high, HeapWord* low) {
+void G1BlockOffsetTable::set_offset_array(Atomic* addr, HeapWord* high, HeapWord* low) {
assert(high >= low, "addresses out of order");
size_t offset = pointer_delta(high, low);
check_offset(offset, "offset too large");
set_offset_array(addr, (uint8_t)offset);
}
-void G1BlockOffsetTable::set_offset_array(uint8_t* left, uint8_t* right, uint8_t offset) {
+void G1BlockOffsetTable::set_offset_array(Atomic* left, Atomic* right, uint8_t offset) {
check_address(right, "Right block offset table address out of range");
assert(left <= right, "indexes out of order");
size_t num_cards = right - left + 1;
@@ -67,9 +67,9 @@ void G1BlockOffsetTable::set_offset_array(uint8_t* left, uint8_t* right, uint8_t
}
#ifdef ASSERT
-void G1BlockOffsetTable::check_address(uint8_t* addr, const char* msg) const {
- uint8_t* start_addr = const_cast(_offset_base + (uintptr_t(_reserved.start()) >> CardTable::card_shift()));
- uint8_t* end_addr = const_cast(_offset_base + (uintptr_t(_reserved.end()) >> CardTable::card_shift()));
+void G1BlockOffsetTable::check_address(Atomic* addr, const char* msg) const {
+ Atomic* start_addr = const_cast*>(_offset_base + (uintptr_t(_reserved.start()) >> CardTable::card_shift()));
+ Atomic* end_addr = const_cast*>(_offset_base + (uintptr_t(_reserved.end()) >> CardTable::card_shift()));
assert(addr >= start_addr && addr <= end_addr,
"%s - offset address: " PTR_FORMAT ", start address: " PTR_FORMAT ", end address: " PTR_FORMAT,
msg, (p2i(addr)), (p2i(start_addr)), (p2i(end_addr)));
@@ -113,17 +113,17 @@ void G1BlockOffsetTable::check_address(uint8_t* addr, const char* msg) const {
// Move back N (e.g., 8) entries and repeat with the
// value of the new entry
//
-void G1BlockOffsetTable::set_remainder_to_point_to_start_incl(uint8_t* start_card, uint8_t* end_card) {
+void G1BlockOffsetTable::set_remainder_to_point_to_start_incl(Atomic* start_card, Atomic* end_card) {
assert(start_card <= end_card, "precondition");
assert(offset_array(start_card-1) < CardTable::card_size_in_words(),
"Offset card has an unexpected value");
- uint8_t* start_card_for_region = start_card;
+ Atomic* start_card_for_region = start_card;
uint8_t offset = UINT8_MAX;
for (uint i = 0; i < BOTConstants::N_powers; i++) {
// -1 so that the card with the actual offset is counted. Another -1
// so that the reach ends in this region and not at the start
// of the next.
- uint8_t* reach = start_card - 1 + (BOTConstants::power_to_cards_back(i+1) - 1);
+ Atomic* reach = start_card - 1 + (BOTConstants::power_to_cards_back(i+1) - 1);
offset = CardTable::card_size_in_words() + i;
if (reach >= end_card) {
set_offset_array(start_card_for_region, end_card, offset);
@@ -141,12 +141,12 @@ void G1BlockOffsetTable::set_remainder_to_point_to_start_incl(uint8_t* start_car
// The card-interval [start_card, end_card] is a closed interval; this
// is an expensive check -- use with care and only under protection of
// suitable flag.
-void G1BlockOffsetTable::check_all_cards(uint8_t* start_card, uint8_t* end_card) const {
+void G1BlockOffsetTable::check_all_cards(Atomic* start_card, Atomic* end_card) const {
if (end_card < start_card) {
return;
}
guarantee(offset_array(start_card) == CardTable::card_size_in_words(), "Wrong value in second card");
- for (uint8_t* c = start_card + 1; c <= end_card; c++ /* yeah! */) {
+ for (Atomic* c = start_card + 1; c <= end_card; c++ /* yeah! */) {
uint8_t entry = offset_array(c);
if ((unsigned)(c - start_card) > BOTConstants::power_to_cards_back(1)) {
guarantee(entry > CardTable::card_size_in_words(),
@@ -157,7 +157,7 @@ void G1BlockOffsetTable::check_all_cards(uint8_t* start_card, uint8_t* end_card)
(uint)entry, (uint)offset_array(c), CardTable::card_size_in_words());
}
size_t backskip = BOTConstants::entry_to_cards_back(entry);
- uint8_t* landing_card = c - backskip;
+ Atomic* landing_card = c - backskip;
guarantee(landing_card >= (start_card - 1), "Inv");
if (landing_card >= start_card) {
guarantee(offset_array(landing_card) <= entry,
@@ -188,7 +188,7 @@ void G1BlockOffsetTable::check_all_cards(uint8_t* start_card, uint8_t* end_card)
//
void G1BlockOffsetTable::update_for_block_work(HeapWord* blk_start, HeapWord* blk_end) {
HeapWord* const cur_card_boundary = align_up_by_card_size(blk_start);
- uint8_t* const offset_card = entry_for_addr(cur_card_boundary);
+ Atomic* const offset_card = entry_for_addr(cur_card_boundary);
assert(blk_start != nullptr && blk_end > blk_start,
"phantom block");
@@ -209,7 +209,7 @@ void G1BlockOffsetTable::update_for_block_work(HeapWord* blk_start, HeapWord* bl
// We need to now mark the subsequent cards that this block spans.
// Index of card on which the block ends.
- uint8_t* end_card = entry_for_addr(blk_end - 1);
+ Atomic* end_card = entry_for_addr(blk_end - 1);
// Are there more cards left to be updated?
if (offset_card + 1 <= end_card) {
@@ -224,7 +224,7 @@ void G1BlockOffsetTable::update_for_block_work(HeapWord* blk_start, HeapWord* bl
// The offset can be 0 if the block starts on a boundary. That
// is checked by an assertion above.
- uint8_t* previous_card = entry_for_addr(blk_start);
+ Atomic* previous_card = entry_for_addr(blk_start);
HeapWord* boundary = addr_for_entry(previous_card);
assert((offset_array(offset_card) == 0 && blk_start == boundary) ||
(offset_array(offset_card) > 0 && offset_array(offset_card) < CardTable::card_size_in_words()),
@@ -240,7 +240,7 @@ void G1BlockOffsetTable::update_for_block_work(HeapWord* blk_start, HeapWord* bl
}
#ifdef ASSERT
-void G1BlockOffsetTable::verify_offset(uint8_t* card_index, uint8_t upper_boundary) const {
+void G1BlockOffsetTable::verify_offset(Atomic* card_index, uint8_t upper_boundary) const {
assert(offset_array(card_index) <= upper_boundary,
"Offset %u should not be larger than upper boundary %u.",
(uint) offset_array(card_index),
@@ -250,19 +250,19 @@ void G1BlockOffsetTable::verify_offset(uint8_t* card_index, uint8_t upper_bounda
void G1BlockOffsetTable::verify_for_block(HeapWord* blk_start, HeapWord* blk_end) const {
assert(is_crossing_card_boundary(blk_start, blk_end), "precondition");
- uint8_t* start_card = entry_for_addr(align_up_by_card_size(blk_start));
- uint8_t* end_card = entry_for_addr(blk_end - 1);
+ Atomic* start_card = entry_for_addr(align_up_by_card_size(blk_start));
+ Atomic* end_card = entry_for_addr(blk_end - 1);
// Check cards in [start_card, end_card]
verify_offset(start_card, CardTable::card_size_in_words());
- for (uint8_t* current_card = start_card + 1; current_card <= end_card; ++current_card) {
+ for (Atomic* current_card = start_card + 1; current_card <= end_card; ++current_card) {
assert(offset_array(current_card) > 0,
"Offset %u is not larger than 0.",
(uint) offset_array(current_card));
verify_offset(current_card, (uint8_t) (CardTable::card_size_in_words() + BOTConstants::N_powers - 1));
- uint8_t* prev = current_card - 1;
- uint8_t* value = current_card;
+ Atomic* prev = current_card - 1;
+ Atomic* value = current_card;
if (offset_array(prev) != offset_array(value)) {
assert(offset_array(value) >= offset_array(prev), "monotonic");
size_t n_cards_back = BOTConstants::entry_to_cards_back(offset_array(value));
diff --git a/src/hotspot/share/gc/g1/g1BlockOffsetTable.hpp b/src/hotspot/share/gc/g1/g1BlockOffsetTable.hpp
index 3b97efc4f0f..89c68ce96d2 100644
--- a/src/hotspot/share/gc/g1/g1BlockOffsetTable.hpp
+++ b/src/hotspot/share/gc/g1/g1BlockOffsetTable.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -29,6 +29,7 @@
#include "gc/shared/blockOffsetTable.hpp"
#include "gc/shared/cardTable.hpp"
#include "memory/memRegion.hpp"
+#include "runtime/atomic.hpp"
#include "utilities/globalDefinitions.hpp"
// This implementation of "G1BlockOffsetTable" divides the covered region
@@ -41,7 +42,7 @@ private:
MemRegion _reserved;
// Biased array-start of BOT array for fast BOT entry translation
- volatile uint8_t* _offset_base;
+ Atomic* _offset_base;
void check_offset(size_t offset, const char* msg) const {
assert(offset < CardTable::card_size_in_words(),
@@ -51,32 +52,32 @@ private:
// Bounds checking accessors:
// For performance these have to devolve to array accesses in product builds.
- inline uint8_t offset_array(uint8_t* addr) const;
+ inline uint8_t offset_array(Atomic* addr) const;
- inline void set_offset_array(uint8_t* addr, uint8_t offset);
+ inline void set_offset_array(Atomic* addr, uint8_t offset);
- inline void set_offset_array(uint8_t* addr, HeapWord* high, HeapWord* low);
+ inline void set_offset_array(Atomic* addr, HeapWord* high, HeapWord* low);
- inline void set_offset_array(uint8_t* left, uint8_t* right, uint8_t offset);
+ inline void set_offset_array(Atomic* left, Atomic* right, uint8_t offset);
// Mapping from address to object start array entry
- inline uint8_t* entry_for_addr(const void* const p) const;
+ inline Atomic* entry_for_addr(const void* const p) const;
// Mapping from object start array entry to address of first word
- inline HeapWord* addr_for_entry(const uint8_t* const p) const;
+ inline HeapWord* addr_for_entry(const Atomic* const p) const;
- void check_address(uint8_t* addr, const char* msg) const NOT_DEBUG_RETURN;
+ void check_address(Atomic* addr, const char* msg) const NOT_DEBUG_RETURN;
// Sets the entries corresponding to the cards starting at "start" and ending
// at "end" to point back to the card before "start"; [start, end]
- void set_remainder_to_point_to_start_incl(uint8_t* start, uint8_t* end);
+ void set_remainder_to_point_to_start_incl(Atomic* start, Atomic* end);
// Update BOT entries corresponding to the mem range [blk_start, blk_end).
void update_for_block_work(HeapWord* blk_start, HeapWord* blk_end);
- void check_all_cards(uint8_t* left_card, uint8_t* right_card) const NOT_DEBUG_RETURN;
+ void check_all_cards(Atomic* left_card, Atomic* right_card) const NOT_DEBUG_RETURN;
- void verify_offset(uint8_t* card_index, uint8_t upper) const NOT_DEBUG_RETURN;
+ void verify_offset(Atomic* card_index, uint8_t upper) const NOT_DEBUG_RETURN;
void verify_for_block(HeapWord* blk_start, HeapWord* blk_end) const NOT_DEBUG_RETURN;
static HeapWord* align_up_by_card_size(HeapWord* const addr) {
diff --git a/src/hotspot/share/gc/g1/g1BlockOffsetTable.inline.hpp b/src/hotspot/share/gc/g1/g1BlockOffsetTable.inline.hpp
index 900e9516c1a..0d809b65526 100644
--- a/src/hotspot/share/gc/g1/g1BlockOffsetTable.inline.hpp
+++ b/src/hotspot/share/gc/g1/g1BlockOffsetTable.inline.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -31,12 +31,11 @@
#include "gc/shared/cardTable.hpp"
#include "gc/shared/memset_with_concurrent_readers.hpp"
#include "oops/oop.inline.hpp"
-#include "runtime/atomicAccess.hpp"
inline HeapWord* G1BlockOffsetTable::block_start_reaching_into_card(const void* addr) const {
assert(_reserved.contains(addr), "invalid address");
- uint8_t* entry = entry_for_addr(addr);
+ Atomic* entry = entry_for_addr(addr);
uint8_t offset = offset_array(entry);
while (offset >= CardTable::card_size_in_words()) {
// The excess of the offset from N_words indicates a power of Base
@@ -50,19 +49,19 @@ inline HeapWord* G1BlockOffsetTable::block_start_reaching_into_card(const void*
return q - offset;
}
-uint8_t G1BlockOffsetTable::offset_array(uint8_t* addr) const {
+uint8_t G1BlockOffsetTable::offset_array(Atomic* addr) const {
check_address(addr, "Block offset table address out of range");
- return AtomicAccess::load(addr);
+ return addr->load_relaxed();
}
-inline uint8_t* G1BlockOffsetTable::entry_for_addr(const void* const p) const {
+inline Atomic* G1BlockOffsetTable::entry_for_addr(const void* const p) const {
assert(_reserved.contains(p),
"out of bounds access to block offset table");
- uint8_t* result = const_cast(&_offset_base[uintptr_t(p) >> CardTable::card_shift()]);
+ Atomic* result = const_cast*>(&_offset_base[uintptr_t(p) >> CardTable::card_shift()]);
return result;
}
-inline HeapWord* G1BlockOffsetTable::addr_for_entry(const uint8_t* const p) const {
+inline HeapWord* G1BlockOffsetTable::addr_for_entry(const Atomic* const p) const {
// _offset_base can be "negative", so can't use pointer_delta().
size_t delta = p - _offset_base;
HeapWord* result = (HeapWord*) (delta << CardTable::card_shift());
diff --git a/src/hotspot/share/gc/g1/g1CollectedHeap.cpp b/src/hotspot/share/gc/g1/g1CollectedHeap.cpp
index b6c3c0b0907..8fc2c7c9941 100644
--- a/src/hotspot/share/gc/g1/g1CollectedHeap.cpp
+++ b/src/hotspot/share/gc/g1/g1CollectedHeap.cpp
@@ -138,6 +138,26 @@ void G1RegionMappingChangedListener::on_commit(uint start_idx, size_t num_region
reset_from_card_cache(start_idx, num_regions);
}
+// Collects commonly used scoped objects that are related to initial setup.
+class G1GCMark : StackObj {
+ ResourceMark _rm;
+ IsSTWGCActiveMark _active_gc_mark;
+ GCIdMark _gc_id_mark;
+ SvcGCMarker _sgcm;
+ GCTraceCPUTime _tcpu;
+
+public:
+ G1GCMark(GCTracer* tracer, bool is_full_gc) :
+ _rm(),
+ _active_gc_mark(),
+ _gc_id_mark(),
+ _sgcm(is_full_gc ? SvcGCMarker::FULL : SvcGCMarker::MINOR),
+ _tcpu(tracer) {
+
+ assert_at_safepoint_on_vm_thread();
+ }
+};
+
void G1CollectedHeap::run_batch_task(G1BatchedTask* cl) {
uint num_workers = MAX2(1u, MIN2(cl->num_workers_estimate(), workers()->active_workers()));
cl->set_max_workers(num_workers);
@@ -525,7 +545,7 @@ void G1CollectedHeap::iterate_regions_in_range(MemRegion range, const Func& func
}
}
-HeapWord* G1CollectedHeap::alloc_archive_region(size_t word_size, HeapWord* preferred_addr) {
+HeapWord* G1CollectedHeap::alloc_archive_region(size_t word_size) {
assert(!is_init_completed(), "Expect to be called at JVM init time");
MutexLocker x(Heap_lock);
@@ -914,12 +934,11 @@ void G1CollectedHeap::verify_after_full_collection() {
void G1CollectedHeap::do_full_collection(size_t allocation_word_size,
bool clear_all_soft_refs,
bool do_maximal_compaction) {
- assert_at_safepoint_on_vm_thread();
-
- G1FullGCMark gc_mark;
+ G1FullGCTracer tracer;
+ G1GCMark gc_mark(&tracer, true /* is_full_gc */);
GCTraceTime(Info, gc) tm("Pause Full", nullptr, gc_cause(), true);
- G1FullCollector collector(this, clear_all_soft_refs, do_maximal_compaction, gc_mark.tracer());
+ G1FullCollector collector(this, clear_all_soft_refs, do_maximal_compaction, &tracer);
collector.prepare_collection();
collector.collect();
collector.complete_collection(allocation_word_size);
@@ -1320,7 +1339,6 @@ G1CollectedHeap::G1CollectedHeap() :
_card_set_freelist_pool(G1CardSetConfiguration::num_mem_object_types()),
_young_regions_cset_group(card_set_config(), &_card_set_freelist_pool, G1CSetCandidateGroup::YoungRegionId),
_cm(nullptr),
- _cm_thread(nullptr),
_cr(nullptr),
_task_queues(nullptr),
_partial_array_state_manager(nullptr),
@@ -1564,7 +1582,6 @@ jint G1CollectedHeap::initialize() {
// Create the G1ConcurrentMark data structure and thread.
// (Must do this late, so that "max_[reserved_]regions" is defined.)
_cm = new G1ConcurrentMark(this, bitmap_storage);
- _cm_thread = _cm->cm_thread();
// Now expand into the initial heap size.
if (!expand(init_byte_size, _workers)) {
@@ -1636,7 +1653,9 @@ jint G1CollectedHeap::initialize() {
}
bool G1CollectedHeap::concurrent_mark_is_terminating() const {
- return _cm_thread->should_terminate();
+ assert(_cm != nullptr, "_cm must have been created");
+ assert(_cm->is_fully_initialized(), "thread must exist in order to check if mark is terminating");
+ return _cm->cm_thread()->should_terminate();
}
void G1CollectedHeap::stop() {
@@ -1645,7 +1664,9 @@ void G1CollectedHeap::stop() {
// that are destroyed during shutdown.
_cr->stop();
_service_thread->stop();
- _cm_thread->stop();
+ if (_cm->is_fully_initialized()) {
+ _cm->cm_thread()->stop();
+ }
}
void G1CollectedHeap::safepoint_synchronize_begin() {
@@ -1842,7 +1863,7 @@ void G1CollectedHeap::increment_old_marking_cycles_completed(bool concurrent,
// is set) so that if a waiter requests another System.gc() it doesn't
// incorrectly see that a marking cycle is still in progress.
if (concurrent) {
- _cm_thread->set_idle();
+ _cm->cm_thread()->set_idle();
}
// Notify threads waiting in System.gc() (with ExplicitGCInvokesConcurrent)
@@ -2421,7 +2442,6 @@ void G1CollectedHeap::print_gc_on(outputStream* st) const {
void G1CollectedHeap::gc_threads_do(ThreadClosure* tc) const {
workers()->threads_do(tc);
- tc->do_thread(_cm_thread);
_cm->threads_do(tc);
_cr->threads_do(tc);
tc->do_thread(_service_thread);
@@ -2542,15 +2562,15 @@ HeapWord* G1CollectedHeap::do_collection_pause(size_t word_size,
}
void G1CollectedHeap::start_concurrent_cycle(bool concurrent_operation_is_full_mark) {
- assert(!_cm_thread->in_progress(), "Can not start concurrent operation while in progress");
-
+ assert(_cm->is_fully_initialized(), "sanity");
+ assert(!_cm->in_progress(), "Can not start concurrent operation while in progress");
MutexLocker x(G1CGC_lock, Mutex::_no_safepoint_check_flag);
if (concurrent_operation_is_full_mark) {
_cm->post_concurrent_mark_start();
- _cm_thread->start_full_mark();
+ _cm->cm_thread()->start_full_mark();
} else {
_cm->post_concurrent_undo_start();
- _cm_thread->start_undo_mark();
+ _cm->cm_thread()->start_undo_mark();
}
G1CGC_lock->notify();
}
@@ -2713,19 +2733,12 @@ void G1CollectedHeap::flush_region_pin_cache() {
}
void G1CollectedHeap::do_collection_pause_at_safepoint(size_t allocation_word_size) {
- assert_at_safepoint_on_vm_thread();
- assert(!is_stw_gc_active(), "collection is not reentrant");
-
- ResourceMark rm;
-
- IsSTWGCActiveMark active_gc_mark;
- GCIdMark gc_id_mark;
- SvcGCMarker sgcm(SvcGCMarker::MINOR);
-
- GCTraceCPUTime tcpu(_gc_tracer_stw);
+ G1GCMark gcm(_gc_tracer_stw, false /* is_full_gc */);
_bytes_used_during_gc = 0;
+ _cm->fully_initialize();
+
policy()->decide_on_concurrent_start_pause();
// Record whether this pause may need to trigger a concurrent operation. Later,
// when we signal the G1ConcurrentMarkThread, the collector state has already
diff --git a/src/hotspot/share/gc/g1/g1CollectedHeap.hpp b/src/hotspot/share/gc/g1/g1CollectedHeap.hpp
index 8009df1fa6a..8ff9d481000 100644
--- a/src/hotspot/share/gc/g1/g1CollectedHeap.hpp
+++ b/src/hotspot/share/gc/g1/g1CollectedHeap.hpp
@@ -736,12 +736,10 @@ public:
void iterate_regions_in_range(MemRegion range, const Func& func);
// Commit the required number of G1 region(s) according to the size requested
- // and mark them as 'old' region(s). Preferred address is treated as a hint for
- // the location of the archive space in the heap. The returned address may or may
- // not be same as the preferred address.
+ // and mark them as 'old' region(s).
// This API is only used for allocating heap space for the archived heap objects
// in the CDS archive.
- HeapWord* alloc_archive_region(size_t word_size, HeapWord* preferred_addr);
+ HeapWord* alloc_archive_region(size_t word_size);
// Populate the G1BlockOffsetTable for archived regions with the given
// memory range.
@@ -825,7 +823,6 @@ public:
// The concurrent marker (and the thread it runs in.)
G1ConcurrentMark* _cm;
- G1ConcurrentMarkThread* _cm_thread;
// The concurrent refiner.
G1ConcurrentRefine* _cr;
diff --git a/src/hotspot/share/gc/g1/g1ConcurrentMark.cpp b/src/hotspot/share/gc/g1/g1ConcurrentMark.cpp
index 2bbfb5032b3..8f3cafe1f5b 100644
--- a/src/hotspot/share/gc/g1/g1ConcurrentMark.cpp
+++ b/src/hotspot/share/gc/g1/g1ConcurrentMark.cpp
@@ -382,12 +382,12 @@ G1CMRootMemRegions::~G1CMRootMemRegions() {
}
void G1CMRootMemRegions::reset() {
- _num_root_regions = 0;
+ _num_root_regions.store_relaxed(0);
}
void G1CMRootMemRegions::add(HeapWord* start, HeapWord* end) {
assert_at_safepoint();
- size_t idx = AtomicAccess::fetch_then_add(&_num_root_regions, 1u);
+ size_t idx = _num_root_regions.fetch_then_add(1u);
assert(idx < _max_regions, "Trying to add more root MemRegions than there is space %zu", _max_regions);
assert(start != nullptr && end != nullptr && start <= end, "Start (" PTR_FORMAT ") should be less or equal to "
"end (" PTR_FORMAT ")", p2i(start), p2i(end));
@@ -398,36 +398,38 @@ void G1CMRootMemRegions::add(HeapWord* start, HeapWord* end) {
void G1CMRootMemRegions::prepare_for_scan() {
assert(!scan_in_progress(), "pre-condition");
- _scan_in_progress = _num_root_regions > 0;
+ _scan_in_progress.store_relaxed(num_root_regions() > 0);
- _claimed_root_regions = 0;
- _should_abort = false;
+ _claimed_root_regions.store_relaxed(0);
+ _should_abort.store_relaxed(false);
}
const MemRegion* G1CMRootMemRegions::claim_next() {
- if (_should_abort) {
+ if (_should_abort.load_relaxed()) {
// If someone has set the should_abort flag, we return null to
// force the caller to bail out of their loop.
return nullptr;
}
- if (_claimed_root_regions >= _num_root_regions) {
+ uint local_num_root_regions = num_root_regions();
+ if (_claimed_root_regions.load_relaxed() >= local_num_root_regions) {
return nullptr;
}
- size_t claimed_index = AtomicAccess::fetch_then_add(&_claimed_root_regions, 1u);
- if (claimed_index < _num_root_regions) {
+ size_t claimed_index = _claimed_root_regions.fetch_then_add(1u);
+ if (claimed_index < local_num_root_regions) {
return &_root_regions[claimed_index];
}
return nullptr;
}
uint G1CMRootMemRegions::num_root_regions() const {
- return (uint)_num_root_regions;
+ return (uint)_num_root_regions.load_relaxed();
}
bool G1CMRootMemRegions::contains(const MemRegion mr) const {
- for (uint i = 0; i < _num_root_regions; i++) {
+ uint local_num_root_regions = num_root_regions();
+ for (uint i = 0; i < local_num_root_regions; i++) {
if (_root_regions[i].equals(mr)) {
return true;
}
@@ -437,7 +439,7 @@ bool G1CMRootMemRegions::contains(const MemRegion mr) const {
void G1CMRootMemRegions::notify_scan_done() {
MutexLocker x(G1RootRegionScan_lock, Mutex::_no_safepoint_check_flag);
- _scan_in_progress = false;
+ _scan_in_progress.store_relaxed(false);
G1RootRegionScan_lock->notify_all();
}
@@ -448,10 +450,10 @@ void G1CMRootMemRegions::cancel_scan() {
void G1CMRootMemRegions::scan_finished() {
assert(scan_in_progress(), "pre-condition");
- if (!_should_abort) {
- assert(_claimed_root_regions >= num_root_regions(),
+ if (!_should_abort.load_relaxed()) {
+ assert(_claimed_root_regions.load_relaxed() >= num_root_regions(),
"we should have claimed all root regions, claimed %zu, length = %u",
- _claimed_root_regions, num_root_regions());
+ _claimed_root_regions.load_relaxed(), num_root_regions());
}
notify_scan_done();
@@ -473,7 +475,7 @@ bool G1CMRootMemRegions::wait_until_scan_finished() {
G1ConcurrentMark::G1ConcurrentMark(G1CollectedHeap* g1h,
G1RegionToSpaceMapper* bitmap_storage) :
- // _cm_thread set inside the constructor
+ _cm_thread(nullptr),
_g1h(g1h),
_mark_bitmap(),
@@ -484,13 +486,12 @@ G1ConcurrentMark::G1ConcurrentMark(G1CollectedHeap* g1h,
_global_mark_stack(),
- // _finger set in set_non_marking_state
+ _finger(nullptr), // _finger set in set_non_marking_state
_worker_id_offset(G1ConcRefinementThreads), // The refinement control thread does not refine cards, so it's just the worker threads.
_max_num_tasks(MAX2(ConcGCThreads, ParallelGCThreads)),
- // _num_active_tasks set in set_non_marking_state()
- // _tasks set inside the constructor
-
+ _num_active_tasks(0), // _num_active_tasks set in set_non_marking_state()
+ _tasks(nullptr), // _tasks set inside late_init()
_task_queues(new G1CMTaskQueueSet(_max_num_tasks)),
_terminator(_max_num_tasks, _task_queues),
_partial_array_state_manager(new PartialArrayStateManager(_max_num_tasks)),
@@ -525,6 +526,12 @@ G1ConcurrentMark::G1ConcurrentMark(G1CollectedHeap* g1h,
assert(G1CGC_lock != nullptr, "CGC_lock must be initialized");
_mark_bitmap.initialize(g1h->reserved(), bitmap_storage);
+}
+
+void G1ConcurrentMark::fully_initialize() {
+ if (is_fully_initialized()) {
+ return;
+ }
// Create & start ConcurrentMark thread.
_cm_thread = new G1ConcurrentMarkThread(this);
@@ -560,6 +567,10 @@ G1ConcurrentMark::G1ConcurrentMark(G1CollectedHeap* g1h,
reset_at_marking_complete();
}
+bool G1ConcurrentMark::in_progress() const {
+ return is_fully_initialized() ? _cm_thread->in_progress() : false;
+}
+
PartialArrayStateManager* G1ConcurrentMark::partial_array_state_manager() const {
return _partial_array_state_manager;
}
@@ -628,8 +639,7 @@ void G1ConcurrentMark::reset_marking_for_restart() {
_finger = _heap.start();
for (uint i = 0; i < _max_num_tasks; ++i) {
- G1CMTaskQueue* queue = _task_queues->queue(i);
- queue->set_empty();
+ _tasks[i]->reset_for_restart();
}
}
@@ -765,7 +775,7 @@ private:
// as asserts here to minimize their overhead on the product. However, we
// will have them as guarantees at the beginning / end of the bitmap
// clearing to get some checking in the product.
- assert(!suspendible() || _cm->cm_thread()->in_progress(), "invariant");
+ assert(!suspendible() || _cm->in_progress(), "invariant");
assert(!suspendible() || !G1CollectedHeap::heap()->collector_state()->mark_or_rebuild_in_progress(), "invariant");
// Abort iteration if necessary.
@@ -821,7 +831,8 @@ void G1ConcurrentMark::clear_bitmap(WorkerThreads* workers, bool may_yield) {
void G1ConcurrentMark::cleanup_for_next_mark() {
// Make sure that the concurrent mark thread looks to still be in
// the current cycle.
- guarantee(cm_thread()->in_progress(), "invariant");
+ guarantee(is_fully_initialized(), "should be initializd");
+ guarantee(in_progress(), "invariant");
// We are finishing up the current cycle by clearing the next
// marking bitmap and getting it ready for the next cycle. During
@@ -834,7 +845,8 @@ void G1ConcurrentMark::cleanup_for_next_mark() {
reset_partial_array_state_manager();
// Repeat the asserts from above.
- guarantee(cm_thread()->in_progress(), "invariant");
+ guarantee(is_fully_initialized(), "should be initializd");
+ guarantee(in_progress(), "invariant");
guarantee(!_g1h->collector_state()->mark_or_rebuild_in_progress(), "invariant");
}
@@ -1925,15 +1937,12 @@ bool G1ConcurrentMark::concurrent_cycle_abort() {
// nothing, but this situation should be extremely rare (a full gc after shutdown
// has been signalled is already rare), and this work should be negligible compared
// to actual full gc work.
- if (!cm_thread()->in_progress() && !_g1h->concurrent_mark_is_terminating()) {
+
+ if (!is_fully_initialized() || (!cm_thread()->in_progress() && !_g1h->concurrent_mark_is_terminating())) {
return false;
}
- // Empty mark stack
reset_marking_for_restart();
- for (uint i = 0; i < _max_num_tasks; ++i) {
- _tasks[i]->clear_region_fields();
- }
abort_marking_threads();
@@ -1987,6 +1996,10 @@ void G1ConcurrentMark::print_summary_info() {
}
log.trace(" Concurrent marking:");
+ if (!is_fully_initialized()) {
+ log.trace(" has not been initialized yet");
+ return;
+ }
print_ms_time_info(" ", "remarks", _remark_times);
{
print_ms_time_info(" ", "final marks", _remark_mark_times);
@@ -2003,7 +2016,10 @@ void G1ConcurrentMark::print_summary_info() {
}
void G1ConcurrentMark::threads_do(ThreadClosure* tc) const {
- _concurrent_workers->threads_do(tc);
+ if (is_fully_initialized()) { // they are initialized late
+ tc->do_thread(_cm_thread);
+ _concurrent_workers->threads_do(tc);
+ }
}
void G1ConcurrentMark::print_on(outputStream* st) const {
@@ -2097,6 +2113,13 @@ void G1CMTask::reset(G1CMBitMap* mark_bitmap) {
_mark_stats_cache.reset();
}
+void G1CMTask::reset_for_restart() {
+ clear_region_fields();
+ _task_queue->set_empty();
+ TASKQUEUE_STATS_ONLY(_partial_array_splitter.stats()->reset());
+ TASKQUEUE_STATS_ONLY(_task_queue->stats.reset());
+}
+
void G1CMTask::register_partial_array_splitter() {
::new (&_partial_array_splitter) PartialArraySplitter(_cm->partial_array_state_manager(),
diff --git a/src/hotspot/share/gc/g1/g1ConcurrentMark.hpp b/src/hotspot/share/gc/g1/g1ConcurrentMark.hpp
index 836d7793f81..0271e6a4208 100644
--- a/src/hotspot/share/gc/g1/g1ConcurrentMark.hpp
+++ b/src/hotspot/share/gc/g1/g1ConcurrentMark.hpp
@@ -290,12 +290,12 @@ class G1CMRootMemRegions {
MemRegion* _root_regions;
size_t const _max_regions;
- volatile size_t _num_root_regions; // Actual number of root regions.
+ Atomic _num_root_regions; // Actual number of root regions.
- volatile size_t _claimed_root_regions; // Number of root regions currently claimed.
+ Atomic _claimed_root_regions; // Number of root regions currently claimed.
- volatile bool _scan_in_progress;
- volatile bool _should_abort;
+ Atomic _scan_in_progress;
+ Atomic _should_abort;
void notify_scan_done();
@@ -312,11 +312,11 @@ public:
void prepare_for_scan();
// Forces get_next() to return null so that the iteration aborts early.
- void abort() { _should_abort = true; }
+ void abort() { _should_abort.store_relaxed(true); }
// Return true if the CM thread are actively scanning root regions,
// false otherwise.
- bool scan_in_progress() { return _scan_in_progress; }
+ bool scan_in_progress() { return _scan_in_progress.load_relaxed(); }
// Claim the next root MemRegion to scan atomically, or return null if
// all have been claimed.
@@ -555,6 +555,9 @@ public:
uint worker_id_offset() const { return _worker_id_offset; }
+ void fully_initialize();
+ bool is_fully_initialized() const { return _cm_thread != nullptr; }
+ bool in_progress() const;
uint max_num_tasks() const {return _max_num_tasks; }
// Clear statistics gathered during the concurrent cycle for the given region after
@@ -841,8 +844,10 @@ private:
// Apply the closure to the given range of elements in the objArray.
inline void process_array_chunk(objArrayOop obj, size_t start, size_t end);
public:
- // Resets the task; should be called right at the beginning of a marking phase.
+ // Resets the task completely for a new marking; should be called right at the beginning of a marking phase.
void reset(G1CMBitMap* mark_bitmap);
+ // Minimal reset of the task, making it ready for continuing to mark.
+ void reset_for_restart();
// Register/unregister Partial Array Splitter Allocator with the PartialArrayStateManager.
// This allows us to discard memory arenas used for partial object array states at the end
// of a concurrent mark cycle.
diff --git a/src/hotspot/share/gc/g1/g1EvacStats.cpp b/src/hotspot/share/gc/g1/g1EvacStats.cpp
index 1d54b184e64..d93f63383c4 100644
--- a/src/hotspot/share/gc/g1/g1EvacStats.cpp
+++ b/src/hotspot/share/gc/g1/g1EvacStats.cpp
@@ -48,11 +48,11 @@ void G1EvacStats::log_plab_allocation() {
"used: %zuB, "
"undo waste: %zuB, ",
_description,
- _allocated * HeapWordSize,
- _wasted * HeapWordSize,
- _unused * HeapWordSize,
+ allocated() * HeapWordSize,
+ wasted() * HeapWordSize,
+ unused() * HeapWordSize,
used() * HeapWordSize,
- _undo_wasted * HeapWordSize);
+ undo_wasted() * HeapWordSize);
log_debug(gc, plab)("%s other allocation: "
"region end waste: %zuB, "
"regions filled: %u, "
@@ -157,13 +157,13 @@ void G1EvacStats::adjust_desired_plab_size() {
assert(is_object_aligned(max_size()) && min_size() <= max_size(),
"PLAB clipping computation may be incorrect");
- assert(_allocated != 0 || _unused == 0,
+ assert(allocated() != 0 || unused() == 0,
"Inconsistency in PLAB stats: "
"_allocated: %zu, "
"_wasted: %zu, "
"_unused: %zu, "
"_undo_wasted: %zu",
- _allocated, _wasted, _unused, _undo_wasted);
+ allocated(), wasted(), unused(), undo_wasted());
size_t plab_size = compute_desired_plab_size();
// Take historical weighted average
diff --git a/src/hotspot/share/gc/g1/g1FullCollector.cpp b/src/hotspot/share/gc/g1/g1FullCollector.cpp
index 6c8cc7028cc..b6388c2f722 100644
--- a/src/hotspot/share/gc/g1/g1FullCollector.cpp
+++ b/src/hotspot/share/gc/g1/g1FullCollector.cpp
@@ -110,7 +110,7 @@ uint G1FullCollector::calc_active_workers() {
G1FullCollector::G1FullCollector(G1CollectedHeap* heap,
bool clear_soft_refs,
bool do_maximal_compaction,
- G1FullGCTracer* tracer) :
+ GCTracer* tracer) :
_heap(heap),
_scope(heap->monitoring_support(), clear_soft_refs, do_maximal_compaction, tracer),
_num_workers(calc_active_workers()),
@@ -276,6 +276,21 @@ void G1FullCollector::before_marking_update_attribute_table(G1HeapRegion* hr) {
class G1FullGCRefProcProxyTask : public RefProcProxyTask {
G1FullCollector& _collector;
+ // G1 Full GC specific closure for handling discovered fields. Do NOT need any
+ // barriers as Full GC discards all this information anyway.
+ class G1FullGCDiscoveredFieldClosure : public EnqueueDiscoveredFieldClosure {
+ G1CollectedHeap* _g1h;
+
+ public:
+ G1FullGCDiscoveredFieldClosure() : _g1h(G1CollectedHeap::heap()) { }
+
+ void enqueue(HeapWord* discovered_field_addr, oop value) override {
+ assert(_g1h->is_in(discovered_field_addr), PTR_FORMAT " is not in heap ", p2i(discovered_field_addr));
+ // Store the value and done.
+ RawAccess<>::oop_store(discovered_field_addr, value);
+ }
+ };
+
public:
G1FullGCRefProcProxyTask(G1FullCollector &collector, uint max_workers)
: RefProcProxyTask("G1FullGCRefProcProxyTask", max_workers),
@@ -286,7 +301,7 @@ public:
G1IsAliveClosure is_alive(&_collector);
uint index = (_tm == RefProcThreadModel::Single) ? 0 : worker_id;
G1FullKeepAliveClosure keep_alive(_collector.marker(index));
- BarrierEnqueueDiscoveredFieldClosure enqueue;
+ G1FullGCDiscoveredFieldClosure enqueue;
G1MarkStackClosure* complete_marking = _collector.marker(index)->stack_closure();
_rp_task->rp_work(worker_id, &is_alive, &keep_alive, &enqueue, complete_marking);
}
diff --git a/src/hotspot/share/gc/g1/g1FullCollector.hpp b/src/hotspot/share/gc/g1/g1FullCollector.hpp
index 7e455b07013..605556a2ba6 100644
--- a/src/hotspot/share/gc/g1/g1FullCollector.hpp
+++ b/src/hotspot/share/gc/g1/g1FullCollector.hpp
@@ -58,18 +58,6 @@ public:
}
};
-// Full GC Mark that holds GC id and CPU time trace. Needs to be separate
-// from the G1FullCollector and G1FullGCScope to allow the Full GC logging
-// to have the same structure as the Young GC logging.
-class G1FullGCMark : StackObj {
- GCIdMark _gc_id;
- G1FullGCTracer _tracer;
- GCTraceCPUTime _cpu_time;
-public:
- G1FullGCMark() : _gc_id(), _tracer(), _cpu_time(&_tracer) { }
- G1FullGCTracer* tracer() { return &_tracer; }
-};
-
// The G1FullCollector holds data associated with the current Full GC.
class G1FullCollector : StackObj {
G1CollectedHeap* _heap;
@@ -102,7 +90,7 @@ public:
G1FullCollector(G1CollectedHeap* heap,
bool clear_soft_refs,
bool do_maximal_compaction,
- G1FullGCTracer* tracer);
+ GCTracer* tracer);
~G1FullCollector();
void prepare_collection();
diff --git a/src/hotspot/share/gc/g1/g1FullGCScope.cpp b/src/hotspot/share/gc/g1/g1FullGCScope.cpp
index 8b92d51a8a3..cb4ebe423ff 100644
--- a/src/hotspot/share/gc/g1/g1FullGCScope.cpp
+++ b/src/hotspot/share/gc/g1/g1FullGCScope.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2017, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -38,18 +38,14 @@ G1FullGCJFRTracerMark::~G1FullGCJFRTracerMark() {
G1FullGCScope::G1FullGCScope(G1MonitoringSupport* monitoring_support,
bool clear_soft,
bool do_maximal_compaction,
- G1FullGCTracer* tracer) :
- _rm(),
+ GCTracer* tracer) :
_should_clear_soft_refs(clear_soft),
_do_maximal_compaction(do_maximal_compaction),
- _g1h(G1CollectedHeap::heap()),
- _svc_marker(SvcGCMarker::FULL),
_timer(),
_tracer(tracer),
- _active(),
_tracer_mark(&_timer, _tracer),
_monitoring_scope(monitoring_support),
- _heap_printer(_g1h),
+ _heap_printer(G1CollectedHeap::heap()),
_region_compaction_threshold(do_maximal_compaction ?
G1HeapRegion::GrainWords :
(1 - MarkSweepDeadRatio / 100.0) * G1HeapRegion::GrainWords) { }
@@ -58,7 +54,7 @@ STWGCTimer* G1FullGCScope::timer() {
return &_timer;
}
-G1FullGCTracer* G1FullGCScope::tracer() {
+GCTracer* G1FullGCScope::tracer() {
return _tracer;
}
diff --git a/src/hotspot/share/gc/g1/g1FullGCScope.hpp b/src/hotspot/share/gc/g1/g1FullGCScope.hpp
index 7a0d0a5395e..fc9d5a71f92 100644
--- a/src/hotspot/share/gc/g1/g1FullGCScope.hpp
+++ b/src/hotspot/share/gc/g1/g1FullGCScope.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2017, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -46,14 +46,10 @@ public:
// Class used to group scoped objects used in the Full GC together.
class G1FullGCScope : public StackObj {
- ResourceMark _rm;
bool _should_clear_soft_refs;
bool _do_maximal_compaction;
- G1CollectedHeap* _g1h;
- SvcGCMarker _svc_marker;
STWGCTimer _timer;
- G1FullGCTracer* _tracer;
- IsSTWGCActiveMark _active;
+ GCTracer* _tracer;
G1FullGCJFRTracerMark _tracer_mark;
G1FullGCMonitoringScope _monitoring_scope;
G1HeapPrinterMark _heap_printer;
@@ -63,13 +59,13 @@ public:
G1FullGCScope(G1MonitoringSupport* monitoring_support,
bool clear_soft,
bool do_maximal_compaction,
- G1FullGCTracer* tracer);
+ GCTracer* tracer);
bool should_clear_soft_refs() const { return _should_clear_soft_refs; }
bool do_maximal_compaction() { return _do_maximal_compaction; }
STWGCTimer* timer();
- G1FullGCTracer* tracer();
+ GCTracer* tracer();
size_t region_compaction_threshold() const;
};
diff --git a/src/hotspot/share/gc/g1/g1HeapRegion.cpp b/src/hotspot/share/gc/g1/g1HeapRegion.cpp
index 361e19d4be5..2052a3ce156 100644
--- a/src/hotspot/share/gc/g1/g1HeapRegion.cpp
+++ b/src/hotspot/share/gc/g1/g1HeapRegion.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -44,7 +44,7 @@
#include "oops/access.inline.hpp"
#include "oops/compressedOops.inline.hpp"
#include "oops/oop.inline.hpp"
-#include "runtime/atomicAccess.hpp"
+#include "runtime/atomic.hpp"
#include "runtime/globals_extension.hpp"
#include "utilities/powerOfTwo.hpp"
@@ -131,8 +131,8 @@ void G1HeapRegion::hr_clear(bool clear_space) {
G1CollectedHeap::heap()->concurrent_mark()->reset_top_at_mark_start(this);
- _parsable_bottom = bottom();
- _garbage_bytes = 0;
+ _parsable_bottom.store_relaxed(bottom());
+ _garbage_bytes.store_relaxed(0);
_incoming_refs = 0;
if (clear_space) clear(SpaceDecorator::Mangle);
@@ -294,12 +294,12 @@ void G1HeapRegion::report_region_type_change(G1HeapRegionTraceType::Type to) {
// young gen regions never have their PB set to anything other than bottom.
assert(parsable_bottom_acquire() == bottom(), "must be");
- _garbage_bytes = 0;
+ _garbage_bytes.store_relaxed(0);
_incoming_refs = 0;
}
void G1HeapRegion::note_self_forward_chunk_done(size_t garbage_bytes) {
- AtomicAccess::add(&_garbage_bytes, garbage_bytes, memory_order_relaxed);
+ _garbage_bytes.add_then_fetch(garbage_bytes, memory_order_relaxed);
}
// Code roots support
@@ -448,7 +448,7 @@ void G1HeapRegion::print_on(outputStream* st) const {
st->print("|-");
}
}
- st->print("|%3zu", AtomicAccess::load(&_pinned_object_count));
+ st->print("|%3zu", _pinned_object_count.load_relaxed());
st->print_cr("");
}
diff --git a/src/hotspot/share/gc/g1/g1HeapRegion.hpp b/src/hotspot/share/gc/g1/g1HeapRegion.hpp
index fe915b0dafe..2b4b640d52b 100644
--- a/src/hotspot/share/gc/g1/g1HeapRegion.hpp
+++ b/src/hotspot/share/gc/g1/g1HeapRegion.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -33,6 +33,7 @@
#include "gc/shared/ageTable.hpp"
#include "gc/shared/spaceDecorator.hpp"
#include "gc/shared/verifyOption.hpp"
+#include "runtime/atomic.hpp"
#include "runtime/mutex.hpp"
#include "utilities/macros.hpp"
@@ -73,7 +74,7 @@ class G1HeapRegion : public CHeapObj {
HeapWord* const _bottom;
HeapWord* const _end;
- HeapWord* volatile _top;
+ Atomic _top;
G1BlockOffsetTable* _bot;
@@ -89,8 +90,8 @@ public:
HeapWord* bottom() const { return _bottom; }
HeapWord* end() const { return _end; }
- void set_top(HeapWord* value) { _top = value; }
- HeapWord* top() const { return _top; }
+ void set_top(HeapWord* value) { _top.store_relaxed(value); }
+ HeapWord* top() const { return _top.load_relaxed(); }
// See the comment above in the declaration of _pre_dummy_top for an
// explanation of what it is.
@@ -231,10 +232,10 @@ private:
//
// Below this limit the marking bitmap must be used to determine size and
// liveness.
- HeapWord* volatile _parsable_bottom;
+ Atomic _parsable_bottom;
// Amount of dead data in the region.
- size_t _garbage_bytes;
+ Atomic _garbage_bytes;
// Approximate number of references to this regions at the end of concurrent
// marking. We we do not mark through all objects, so this is an estimate.
@@ -249,7 +250,7 @@ private:
uint _node_index;
// Number of objects in this region that are currently pinned.
- volatile size_t _pinned_object_count;
+ Atomic _pinned_object_count;
void report_region_type_change(G1HeapRegionTraceType::Type to);
@@ -331,7 +332,7 @@ public:
}
// A lower bound on the amount of garbage bytes in the region.
- size_t garbage_bytes() const { return _garbage_bytes; }
+ size_t garbage_bytes() const { return _garbage_bytes.load_relaxed(); }
// Return the amount of bytes we'll reclaim if we collect this
// region. This includes not only the known garbage bytes in the
@@ -393,8 +394,8 @@ public:
bool is_old_or_humongous() const { return _type.is_old_or_humongous(); }
- size_t pinned_count() const { return AtomicAccess::load(&_pinned_object_count); }
- bool has_pinned_objects() const { return pinned_count() > 0; }
+ inline size_t pinned_count() const;
+ inline bool has_pinned_objects() const;
void set_free();
diff --git a/src/hotspot/share/gc/g1/g1HeapRegion.inline.hpp b/src/hotspot/share/gc/g1/g1HeapRegion.inline.hpp
index f25bf62c9be..4f242b7a537 100644
--- a/src/hotspot/share/gc/g1/g1HeapRegion.inline.hpp
+++ b/src/hotspot/share/gc/g1/g1HeapRegion.inline.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -35,7 +35,6 @@
#include "gc/g1/g1Policy.hpp"
#include "gc/g1/g1Predictions.hpp"
#include "oops/oop.inline.hpp"
-#include "runtime/atomicAccess.hpp"
#include "runtime/init.hpp"
#include "runtime/prefetch.inline.hpp"
#include "runtime/safepoint.hpp"
@@ -131,7 +130,7 @@ inline void G1HeapRegion::prepare_for_full_gc() {
// After marking and class unloading the heap temporarily contains dead objects
// with unloaded klasses. Moving parsable_bottom makes some (debug) code correctly
// skip dead objects.
- _parsable_bottom = top();
+ _parsable_bottom.store_relaxed(top());
}
inline void G1HeapRegion::reset_compacted_after_full_gc(HeapWord* new_top) {
@@ -154,7 +153,7 @@ inline void G1HeapRegion::reset_after_full_gc_common() {
// Everything above bottom() is parsable and live.
reset_parsable_bottom();
- _garbage_bytes = 0;
+ _garbage_bytes.store_relaxed(0);
_incoming_refs = 0;
@@ -188,20 +187,22 @@ inline void G1HeapRegion::apply_to_marked_objects(G1CMBitMap* bitmap, ApplyToMar
inline HeapWord* G1HeapRegion::par_allocate(size_t min_word_size,
size_t desired_word_size,
size_t* actual_word_size) {
+ HeapWord* obj = top();
do {
- HeapWord* obj = top();
size_t available = pointer_delta(end(), obj);
size_t want_to_allocate = MIN2(available, desired_word_size);
if (want_to_allocate >= min_word_size) {
HeapWord* new_top = obj + want_to_allocate;
- HeapWord* result = AtomicAccess::cmpxchg(&_top, obj, new_top);
- // result can be one of two:
- // the old top value: the exchange succeeded
+ HeapWord* result = _top.compare_exchange(obj, new_top);
+ // Result can be one of two:
+ // the old top value: the exchange succeeded, return.
// otherwise: the new value of the top is returned.
if (result == obj) {
assert(is_object_aligned(obj) && is_object_aligned(new_top), "checking alignment");
*actual_word_size = want_to_allocate;
return obj;
+ } else {
+ obj = result;
}
} else {
return nullptr;
@@ -254,27 +255,27 @@ inline void G1HeapRegion::update_bot_for_block(HeapWord* start, HeapWord* end) {
inline HeapWord* G1HeapRegion::parsable_bottom() const {
assert(!is_init_completed() || SafepointSynchronize::is_at_safepoint(), "only during initialization or safepoint");
- return _parsable_bottom;
+ return _parsable_bottom.load_relaxed();
}
inline HeapWord* G1HeapRegion::parsable_bottom_acquire() const {
- return AtomicAccess::load_acquire(&_parsable_bottom);
+ return _parsable_bottom.load_acquire();
}
inline void G1HeapRegion::reset_parsable_bottom() {
- AtomicAccess::release_store(&_parsable_bottom, bottom());
+ _parsable_bottom.release_store(bottom());
}
inline void G1HeapRegion::note_end_of_marking(HeapWord* top_at_mark_start, size_t marked_bytes, size_t incoming_refs) {
assert_at_safepoint();
if (top_at_mark_start != bottom()) {
- _garbage_bytes = byte_size(bottom(), top_at_mark_start) - marked_bytes;
+ _garbage_bytes.store_relaxed(byte_size(bottom(), top_at_mark_start) - marked_bytes);
_incoming_refs = incoming_refs;
}
if (needs_scrubbing()) {
- _parsable_bottom = top_at_mark_start;
+ _parsable_bottom.store_relaxed(top_at_mark_start);
}
}
@@ -286,6 +287,14 @@ inline bool G1HeapRegion::needs_scrubbing() const {
return is_old();
}
+inline size_t G1HeapRegion::pinned_count() const {
+ return _pinned_object_count.load_relaxed();
+}
+
+inline bool G1HeapRegion::has_pinned_objects() const {
+ return pinned_count() > 0;
+}
+
inline bool G1HeapRegion::in_collection_set() const {
return G1CollectedHeap::heap()->is_in_cset(this);
}
@@ -511,7 +520,7 @@ inline void G1HeapRegion::record_surv_words_in_group(size_t words_survived) {
inline void G1HeapRegion::add_pinned_object_count(size_t value) {
assert(value != 0, "wasted effort");
assert(!is_free(), "trying to pin free region %u, adding %zu", hrm_index(), value);
- AtomicAccess::add(&_pinned_object_count, value, memory_order_relaxed);
+ _pinned_object_count.add_then_fetch(value, memory_order_relaxed);
}
inline void G1HeapRegion::install_cset_group(G1CSetCandidateGroup* cset_group) {
diff --git a/src/hotspot/share/gc/g1/g1HeapRegionManager.cpp b/src/hotspot/share/gc/g1/g1HeapRegionManager.cpp
index 795b6543bae..44897c8a277 100644
--- a/src/hotspot/share/gc/g1/g1HeapRegionManager.cpp
+++ b/src/hotspot/share/gc/g1/g1HeapRegionManager.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -713,8 +713,10 @@ void G1HeapRegionManager::verify_optional() {
G1HeapRegionClaimer::G1HeapRegionClaimer(uint n_workers) :
_n_workers(n_workers), _n_regions(G1CollectedHeap::heap()->_hrm._next_highest_used_hrm_index), _claims(nullptr) {
- uint* new_claims = NEW_C_HEAP_ARRAY(uint, _n_regions, mtGC);
- memset(new_claims, Unclaimed, sizeof(*_claims) * _n_regions);
+ Atomic* new_claims = NEW_C_HEAP_ARRAY(Atomic, _n_regions, mtGC);
+ for (uint i = 0; i < _n_regions; i++) {
+ new_claims[i].store_relaxed(Unclaimed);
+ }
_claims = new_claims;
}
@@ -730,13 +732,12 @@ uint G1HeapRegionClaimer::offset_for_worker(uint worker_id) const {
bool G1HeapRegionClaimer::is_region_claimed(uint region_index) const {
assert(region_index < _n_regions, "Invalid index.");
- return _claims[region_index] == Claimed;
+ return _claims[region_index].load_relaxed() == Claimed;
}
bool G1HeapRegionClaimer::claim_region(uint region_index) {
assert(region_index < _n_regions, "Invalid index.");
- uint old_val = AtomicAccess::cmpxchg(&_claims[region_index], Unclaimed, Claimed);
- return old_val == Unclaimed;
+ return _claims[region_index].compare_set(Unclaimed, Claimed);
}
class G1RebuildFreeListTask : public WorkerTask {
diff --git a/src/hotspot/share/gc/g1/g1HeapRegionManager.hpp b/src/hotspot/share/gc/g1/g1HeapRegionManager.hpp
index b4ce3b0a8be..eb593ff408e 100644
--- a/src/hotspot/share/gc/g1/g1HeapRegionManager.hpp
+++ b/src/hotspot/share/gc/g1/g1HeapRegionManager.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -30,6 +30,7 @@
#include "gc/g1/g1HeapRegionSet.hpp"
#include "gc/g1/g1RegionToSpaceMapper.hpp"
#include "memory/allocation.hpp"
+#include "runtime/atomic.hpp"
#include "services/memoryUsage.hpp"
class G1HeapRegion;
@@ -294,7 +295,7 @@ public:
class G1HeapRegionClaimer : public StackObj {
uint _n_workers;
uint _n_regions;
- volatile uint* _claims;
+ Atomic* _claims;
static const uint Unclaimed = 0;
static const uint Claimed = 1;
diff --git a/src/hotspot/share/gc/g1/g1PeriodicGCTask.cpp b/src/hotspot/share/gc/g1/g1PeriodicGCTask.cpp
index 50002ac2bfe..f280d76f3c7 100644
--- a/src/hotspot/share/gc/g1/g1PeriodicGCTask.cpp
+++ b/src/hotspot/share/gc/g1/g1PeriodicGCTask.cpp
@@ -39,7 +39,7 @@ bool G1PeriodicGCTask::should_start_periodic_gc(G1CollectedHeap* g1h,
SuspendibleThreadSetJoiner sts;
// If we are currently in a concurrent mark we are going to uncommit memory soon.
- if (g1h->concurrent_mark()->cm_thread()->in_progress()) {
+ if (g1h->concurrent_mark()->in_progress()) {
log_debug(gc, periodic)("Concurrent cycle in progress. Skipping.");
return false;
}
diff --git a/src/hotspot/share/gc/g1/g1Policy.cpp b/src/hotspot/share/gc/g1/g1Policy.cpp
index 1d0b29c303e..98e6acc1d77 100644
--- a/src/hotspot/share/gc/g1/g1Policy.cpp
+++ b/src/hotspot/share/gc/g1/g1Policy.cpp
@@ -739,7 +739,7 @@ double G1Policy::constant_other_time_ms(double pause_time_ms) const {
}
bool G1Policy::about_to_start_mixed_phase() const {
- return _g1h->concurrent_mark()->cm_thread()->in_progress() || collector_state()->in_young_gc_before_mixed();
+ return _g1h->concurrent_mark()->in_progress() || collector_state()->in_young_gc_before_mixed();
}
bool G1Policy::need_to_start_conc_mark(const char* source, size_t allocation_word_size) {
@@ -1235,7 +1235,7 @@ bool G1Policy::force_concurrent_start_if_outside_cycle(GCCause::Cause gc_cause)
// We actually check whether we are marking here and not if we are in a
// reclamation phase. This means that we will schedule a concurrent mark
// even while we are still in the process of reclaiming memory.
- bool during_cycle = _g1h->concurrent_mark()->cm_thread()->in_progress();
+ bool during_cycle = _g1h->concurrent_mark()->in_progress();
if (!during_cycle) {
log_debug(gc, ergo)("Request concurrent cycle initiation (requested by GC cause). "
"GC cause: %s",
diff --git a/src/hotspot/share/gc/g1/g1RegionMarkStatsCache.cpp b/src/hotspot/share/gc/g1/g1RegionMarkStatsCache.cpp
index d9b7ec294bd..c5f55e1d20c 100644
--- a/src/hotspot/share/gc/g1/g1RegionMarkStatsCache.cpp
+++ b/src/hotspot/share/gc/g1/g1RegionMarkStatsCache.cpp
@@ -29,12 +29,12 @@
G1RegionMarkStatsCache::G1RegionMarkStatsCache(G1RegionMarkStats* target, uint num_cache_entries) :
_target(target),
+ _cache(NEW_C_HEAP_ARRAY(G1RegionMarkStatsCacheEntry, num_cache_entries, mtGC)),
_num_cache_entries(num_cache_entries),
_num_cache_entries_mask(_num_cache_entries - 1) {
guarantee(is_power_of_2(num_cache_entries),
"Number of cache entries must be power of two, but is %u", num_cache_entries);
- _cache = NEW_C_HEAP_ARRAY(G1RegionMarkStatsCacheEntry, _num_cache_entries, mtGC);
}
G1RegionMarkStatsCache::~G1RegionMarkStatsCache() {
diff --git a/src/hotspot/share/gc/g1/g1VMOperations.cpp b/src/hotspot/share/gc/g1/g1VMOperations.cpp
index 1c024f2943b..56ab3a4b0fe 100644
--- a/src/hotspot/share/gc/g1/g1VMOperations.cpp
+++ b/src/hotspot/share/gc/g1/g1VMOperations.cpp
@@ -85,7 +85,7 @@ void VM_G1TryInitiateConcMark::doit() {
GCCauseSetter x(g1h, _gc_cause);
_mark_in_progress = g1h->collector_state()->mark_in_progress();
- _cycle_already_in_progress = g1h->concurrent_mark()->cm_thread()->in_progress();
+ _cycle_already_in_progress = g1h->concurrent_mark()->in_progress();
if (!g1h->policy()->force_concurrent_start_if_outside_cycle(_gc_cause)) {
// Failure to force the next GC pause to be a concurrent start indicates
diff --git a/src/hotspot/share/gc/g1/g1YoungCollector.cpp b/src/hotspot/share/gc/g1/g1YoungCollector.cpp
index 36cc44a8b7c..a9db9a7c269 100644
--- a/src/hotspot/share/gc/g1/g1YoungCollector.cpp
+++ b/src/hotspot/share/gc/g1/g1YoungCollector.cpp
@@ -36,6 +36,7 @@
#include "gc/g1/g1EvacFailureRegions.inline.hpp"
#include "gc/g1/g1EvacInfo.hpp"
#include "gc/g1/g1GCPhaseTimes.hpp"
+#include "gc/g1/g1HeapRegion.inline.hpp"
#include "gc/g1/g1HeapRegionPrinter.hpp"
#include "gc/g1/g1MonitoringSupport.hpp"
#include "gc/g1/g1ParScanThreadState.inline.hpp"
diff --git a/src/hotspot/share/gc/g1/g1YoungGCPostEvacuateTasks.cpp b/src/hotspot/share/gc/g1/g1YoungGCPostEvacuateTasks.cpp
index 46d12df575c..3f47d386015 100644
--- a/src/hotspot/share/gc/g1/g1YoungGCPostEvacuateTasks.cpp
+++ b/src/hotspot/share/gc/g1/g1YoungGCPostEvacuateTasks.cpp
@@ -854,7 +854,7 @@ public:
void do_thread(Thread* thread) {
if (UseTLAB && ResizeTLAB) {
- static_cast(thread)->tlab().resize();
+ thread->tlab().resize();
}
G1BarrierSet::g1_barrier_set()->update_card_table_base(thread);
diff --git a/src/hotspot/share/gc/g1/vmStructs_g1.hpp b/src/hotspot/share/gc/g1/vmStructs_g1.hpp
index 21c86d47a6b..af236ec8581 100644
--- a/src/hotspot/share/gc/g1/vmStructs_g1.hpp
+++ b/src/hotspot/share/gc/g1/vmStructs_g1.hpp
@@ -28,6 +28,7 @@
#include "gc/g1/g1CollectedHeap.hpp"
#include "gc/g1/g1HeapRegion.hpp"
#include "gc/g1/g1HeapRegionManager.hpp"
+#include "runtime/atomic.hpp"
#include "utilities/macros.hpp"
#define VM_STRUCTS_G1GC(nonstatic_field, \
@@ -39,9 +40,9 @@
\
nonstatic_field(G1HeapRegion, _type, G1HeapRegionType) \
nonstatic_field(G1HeapRegion, _bottom, HeapWord* const) \
- nonstatic_field(G1HeapRegion, _top, HeapWord* volatile) \
+ nonstatic_field(G1HeapRegion, _top, Atomic) \
nonstatic_field(G1HeapRegion, _end, HeapWord* const) \
- volatile_nonstatic_field(G1HeapRegion, _pinned_object_count, size_t) \
+ volatile_nonstatic_field(G1HeapRegion, _pinned_object_count, Atomic)\
\
nonstatic_field(G1HeapRegionType, _tag, G1HeapRegionType::Tag volatile) \
\
diff --git a/src/hotspot/share/gc/parallel/mutableNUMASpace.cpp b/src/hotspot/share/gc/parallel/mutableNUMASpace.cpp
index e0b1edf2efc..c5d112ffbc1 100644
--- a/src/hotspot/share/gc/parallel/mutableNUMASpace.cpp
+++ b/src/hotspot/share/gc/parallel/mutableNUMASpace.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2006, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2006, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -31,7 +31,7 @@
#include "memory/allocation.inline.hpp"
#include "oops/oop.inline.hpp"
#include "oops/typeArrayOop.hpp"
-#include "runtime/atomicAccess.hpp"
+#include "runtime/atomic.hpp"
#include "runtime/java.hpp"
#include "runtime/javaThread.hpp"
#include "runtime/os.inline.hpp"
@@ -489,7 +489,7 @@ HeapWord* MutableNUMASpace::cas_allocate(size_t size) {
if (p != nullptr) {
HeapWord* cur_top, *cur_chunk_top = p + size;
while ((cur_top = top()) < cur_chunk_top) { // Keep _top updated.
- if (AtomicAccess::cmpxchg(top_addr(), cur_top, cur_chunk_top) == cur_top) {
+ if (top_addr()->compare_set(cur_top, cur_chunk_top)) {
break;
}
}
diff --git a/src/hotspot/share/gc/parallel/mutableSpace.cpp b/src/hotspot/share/gc/parallel/mutableSpace.cpp
index fc42fc1eab2..d99db493989 100644
--- a/src/hotspot/share/gc/parallel/mutableSpace.cpp
+++ b/src/hotspot/share/gc/parallel/mutableSpace.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -28,7 +28,6 @@
#include "memory/iterator.inline.hpp"
#include "memory/universe.hpp"
#include "oops/oop.inline.hpp"
-#include "runtime/atomicAccess.hpp"
#include "runtime/javaThread.hpp"
#include "runtime/safepoint.hpp"
#include "utilities/align.hpp"
@@ -123,7 +122,7 @@ void MutableSpace::initialize(MemRegion mr,
// makes the new space available for allocation by other threads. So this
// assignment must follow all other configuration and initialization that
// might be done for expansion.
- AtomicAccess::release_store(end_addr(), mr.end());
+ _end.release_store(mr.end());
if (clear_space) {
clear(mangle_space);
@@ -140,7 +139,7 @@ void MutableSpace::clear(bool mangle_space) {
#ifndef PRODUCT
void MutableSpace::mangle_unused_area() {
- mangle_region(MemRegion(_top, _end));
+ mangle_region(MemRegion(top(), end()));
}
void MutableSpace::mangle_region(MemRegion mr) {
@@ -155,14 +154,10 @@ HeapWord* MutableSpace::cas_allocate(size_t size) {
// If end is read first, other threads may advance end and top such that
// current top > old end and current top + size > current end. Then
// pointer_delta underflows, allowing installation of top > current end.
- HeapWord* obj = AtomicAccess::load_acquire(top_addr());
+ HeapWord* obj = _top.load_acquire();
if (pointer_delta(end(), obj) >= size) {
HeapWord* new_top = obj + size;
- HeapWord* result = AtomicAccess::cmpxchg(top_addr(), obj, new_top);
- // result can be one of two:
- // the old top value: the exchange succeeded
- // otherwise: the new value of the top is returned.
- if (result != obj) {
+ if (!_top.compare_set(obj, new_top)) {
continue; // another thread beat us to the allocation, try again
}
assert(is_object_aligned(obj) && is_object_aligned(new_top),
@@ -177,7 +172,7 @@ HeapWord* MutableSpace::cas_allocate(size_t size) {
// Try to deallocate previous allocation. Returns true upon success.
bool MutableSpace::cas_deallocate(HeapWord *obj, size_t size) {
HeapWord* expected_top = obj + size;
- return AtomicAccess::cmpxchg(top_addr(), expected_top, obj) == expected_top;
+ return _top.compare_set(expected_top, obj);
}
void MutableSpace::oop_iterate(OopIterateClosure* cl) {
diff --git a/src/hotspot/share/gc/parallel/mutableSpace.hpp b/src/hotspot/share/gc/parallel/mutableSpace.hpp
index 9d3894e2489..28df19a7c4b 100644
--- a/src/hotspot/share/gc/parallel/mutableSpace.hpp
+++ b/src/hotspot/share/gc/parallel/mutableSpace.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -28,6 +28,7 @@
#include "memory/allocation.hpp"
#include "memory/iterator.hpp"
#include "memory/memRegion.hpp"
+#include "runtime/atomic.hpp"
#include "utilities/copy.hpp"
#include "utilities/globalDefinitions.hpp"
#include "utilities/macros.hpp"
@@ -53,8 +54,8 @@ class MutableSpace: public CHeapObj {
MemRegion _last_setup_region;
size_t _page_size;
HeapWord* _bottom;
- HeapWord* volatile _top;
- HeapWord* _end;
+ Atomic _top;
+ Atomic _end;
void numa_setup_pages(MemRegion mr, bool clear_space);
@@ -64,21 +65,20 @@ class MutableSpace: public CHeapObj {
protected:
size_t page_size() const { return _page_size; }
+ Atomic* top_addr() { return &_top; }
+
public:
virtual ~MutableSpace() = default;
MutableSpace(size_t page_size);
// Accessors
HeapWord* bottom() const { return _bottom; }
- HeapWord* top() const { return _top; }
- HeapWord* end() const { return _end; }
+ HeapWord* top() const { return _top.load_relaxed(); }
+ HeapWord* end() const { return _end.load_relaxed(); }
void set_bottom(HeapWord* value) { _bottom = value; }
- virtual void set_top(HeapWord* value) { _top = value; }
- void set_end(HeapWord* value) { _end = value; }
-
- HeapWord* volatile* top_addr() { return &_top; }
- HeapWord** end_addr() { return &_end; }
+ virtual void set_top(HeapWord* value) { _top.store_relaxed(value); }
+ void set_end(HeapWord* value) { _end.store_relaxed(value); }
MemRegion region() const { return MemRegion(bottom(), end()); }
@@ -110,7 +110,7 @@ public:
// Boolean queries.
bool is_empty() const { return used_in_words() == 0; }
bool not_empty() const { return used_in_words() > 0; }
- bool contains(const void* p) const { return _bottom <= p && p < _end; }
+ bool contains(const void* p) const { return _bottom <= p && p < end(); }
// Size computations. Sizes are in bytes.
size_t used_in_bytes() const { return used_in_words() * HeapWordSize; }
diff --git a/src/hotspot/share/gc/parallel/parallelScavengeHeap.cpp b/src/hotspot/share/gc/parallel/parallelScavengeHeap.cpp
index 0d8a3166f79..f49419595e1 100644
--- a/src/hotspot/share/gc/parallel/parallelScavengeHeap.cpp
+++ b/src/hotspot/share/gc/parallel/parallelScavengeHeap.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -48,6 +48,7 @@
#include "memory/reservedSpace.hpp"
#include "memory/universe.hpp"
#include "oops/oop.inline.hpp"
+#include "runtime/atomic.hpp"
#include "runtime/cpuTimeCounters.hpp"
#include "runtime/globals_extension.hpp"
#include "runtime/handles.inline.hpp"
@@ -438,7 +439,7 @@ bool ParallelScavengeHeap::check_gc_overhead_limit() {
log_debug(gc)("GC Overhead Limit: GC Time %f Free Space Young %f Old %f Counter %zu",
(100 - _size_policy->mutator_time_percent()),
percent_of(_young_gen->free_in_bytes(), _young_gen->capacity_in_bytes()),
- percent_of(_old_gen->free_in_bytes(), _young_gen->capacity_in_bytes()),
+ percent_of(_old_gen->free_in_bytes(), _old_gen->capacity_in_bytes()),
_gc_overhead_counter);
if (little_mutator_time && little_free_space) {
@@ -594,7 +595,7 @@ void ParallelScavengeHeap::object_iterate(ObjectClosure* cl) {
// these spaces.
// The old space is divided into fixed-size blocks.
class HeapBlockClaimer : public StackObj {
- size_t _claimed_index;
+ Atomic _claimed_index;
public:
static const size_t InvalidIndex = SIZE_MAX;
@@ -606,7 +607,7 @@ public:
// Claim the block and get the block index.
size_t claim_and_get_block() {
size_t block_index;
- block_index = AtomicAccess::fetch_then_add(&_claimed_index, 1u);
+ block_index = _claimed_index.fetch_then_add(1u);
PSOldGen* old_gen = ParallelScavengeHeap::heap()->old_gen();
size_t num_claims = old_gen->num_iterable_blocks() + NumNonOldGenClaims;
diff --git a/src/hotspot/share/gc/parallel/psCardTable.cpp b/src/hotspot/share/gc/parallel/psCardTable.cpp
index fcd0dd67a45..6429766309a 100644
--- a/src/hotspot/share/gc/parallel/psCardTable.cpp
+++ b/src/hotspot/share/gc/parallel/psCardTable.cpp
@@ -108,7 +108,7 @@ void PSCardTable::scan_obj_with_limit(PSPromotionManager* pm,
}
void PSCardTable::pre_scavenge(uint active_workers) {
- _preprocessing_active_workers = active_workers;
+ _preprocessing_active_workers.store_relaxed(active_workers);
}
// The "shadow" table is a copy of the card table entries of the current stripe.
@@ -382,9 +382,9 @@ void PSCardTable::scavenge_contents_parallel(ObjectStartArray* start_array,
preprocess_card_table_parallel(object_start, old_gen_bottom, old_gen_top, stripe_index, n_stripes);
// Sync with other workers.
- AtomicAccess::dec(&_preprocessing_active_workers);
+ _preprocessing_active_workers.sub_then_fetch(1);
SpinYield spin_yield;
- while (AtomicAccess::load_acquire(&_preprocessing_active_workers) > 0) {
+ while (_preprocessing_active_workers.load_acquire() > 0) {
spin_yield.wait();
}
diff --git a/src/hotspot/share/gc/parallel/psCardTable.hpp b/src/hotspot/share/gc/parallel/psCardTable.hpp
index 70c32d23b7f..033933bcbf1 100644
--- a/src/hotspot/share/gc/parallel/psCardTable.hpp
+++ b/src/hotspot/share/gc/parallel/psCardTable.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2001, 2019, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -27,6 +27,7 @@
#include "gc/shared/cardTable.hpp"
#include "oops/oop.hpp"
+#include "runtime/atomic.hpp"
class MutableSpace;
class ObjectStartArray;
@@ -37,7 +38,7 @@ class PSCardTable: public CardTable {
static constexpr size_t num_cards_in_stripe = 128;
static_assert(num_cards_in_stripe >= 1, "progress");
- volatile int _preprocessing_active_workers;
+ Atomic _preprocessing_active_workers;
bool is_dirty(CardValue* card) {
return !is_clean(card);
diff --git a/src/hotspot/share/gc/parallel/psParallelCompact.cpp b/src/hotspot/share/gc/parallel/psParallelCompact.cpp
index bab72296d4c..4c6ea01e45f 100644
--- a/src/hotspot/share/gc/parallel/psParallelCompact.cpp
+++ b/src/hotspot/share/gc/parallel/psParallelCompact.cpp
@@ -30,6 +30,7 @@
#include "code/codeCache.hpp"
#include "code/nmethod.hpp"
#include "compiler/oopMap.hpp"
+#include "cppstdlib/new.hpp"
#include "gc/parallel/objectStartArray.inline.hpp"
#include "gc/parallel/parallelArguments.hpp"
#include "gc/parallel/parallelScavengeHeap.inline.hpp"
@@ -135,8 +136,8 @@ bool ParallelCompactData::RegionData::is_clear() {
(_source_region == 0) &&
(_partial_obj_addr == nullptr) &&
(_partial_obj_size == 0) &&
- (_dc_and_los == 0) &&
- (_shadow_state == 0);
+ (dc_and_los() == 0) &&
+ (shadow_state() == 0);
}
#ifdef ASSERT
@@ -145,8 +146,8 @@ void ParallelCompactData::RegionData::verify_clear() {
assert(_source_region == 0, "inv");
assert(_partial_obj_addr == nullptr, "inv");
assert(_partial_obj_size == 0, "inv");
- assert(_dc_and_los == 0, "inv");
- assert(_shadow_state == 0, "inv");
+ assert(dc_and_los() == 0, "inv");
+ assert(shadow_state() == 0, "inv");
}
#endif
@@ -296,7 +297,9 @@ void ParallelCompactData::clear_range(size_t beg_region, size_t end_region) {
assert(end_region <= _region_count, "end_region out of range");
const size_t region_cnt = end_region - beg_region;
- memset(_region_data + beg_region, 0, region_cnt * sizeof(RegionData));
+ for (size_t i = beg_region; i < end_region; i++) {
+ ::new (&_region_data[i]) RegionData{};
+ }
}
// The total live words on src_region would overflow the target space, so find
@@ -1294,7 +1297,7 @@ void PSParallelCompact::marking_phase(ParallelOldTracer *gc_tracer) {
}
template
-void PSParallelCompact::adjust_in_space_helper(SpaceId id, volatile uint* claim_counter, Func&& on_stripe) {
+void PSParallelCompact::adjust_in_space_helper(SpaceId id, Atomic* claim_counter, Func&& on_stripe) {
MutableSpace* sp = PSParallelCompact::space(id);
HeapWord* const bottom = sp->bottom();
HeapWord* const top = sp->top();
@@ -1307,7 +1310,7 @@ void PSParallelCompact::adjust_in_space_helper(SpaceId id, volatile uint* claim_
const size_t stripe_size = num_regions_per_stripe * region_size;
while (true) {
- uint counter = AtomicAccess::fetch_then_add(claim_counter, num_regions_per_stripe);
+ uint counter = claim_counter->fetch_then_add(num_regions_per_stripe);
HeapWord* cur_stripe = bottom + counter * region_size;
if (cur_stripe >= top) {
break;
@@ -1317,7 +1320,7 @@ void PSParallelCompact::adjust_in_space_helper(SpaceId id, volatile uint* claim_
}
}
-void PSParallelCompact::adjust_in_old_space(volatile uint* claim_counter) {
+void PSParallelCompact::adjust_in_old_space(Atomic* claim_counter) {
// Regions in old-space shouldn't be split.
assert(!_space_info[old_space_id].split_info().is_valid(), "inv");
@@ -1348,7 +1351,7 @@ void PSParallelCompact::adjust_in_old_space(volatile uint* claim_counter) {
});
}
-void PSParallelCompact::adjust_in_young_space(SpaceId id, volatile uint* claim_counter) {
+void PSParallelCompact::adjust_in_young_space(SpaceId id, Atomic* claim_counter) {
adjust_in_space_helper(id, claim_counter, [](HeapWord* stripe_start, HeapWord* stripe_end) {
HeapWord* obj_start = stripe_start;
while (obj_start < stripe_end) {
@@ -1362,7 +1365,7 @@ void PSParallelCompact::adjust_in_young_space(SpaceId id, volatile uint* claim_c
});
}
-void PSParallelCompact::adjust_pointers_in_spaces(uint worker_id, volatile uint* claim_counters) {
+void PSParallelCompact::adjust_pointers_in_spaces(uint worker_id, Atomic* claim_counters) {
auto start_time = Ticks::now();
adjust_in_old_space(&claim_counters[0]);
for (uint id = eden_space_id; id < last_space_id; ++id) {
@@ -1376,12 +1379,12 @@ class PSAdjustTask final : public WorkerTask {
WeakProcessor::Task _weak_proc_task;
OopStorageSetStrongParState _oop_storage_iter;
uint _nworkers;
- volatile bool _code_cache_claimed;
- volatile uint _claim_counters[PSParallelCompact::last_space_id] = {};
+ Atomic _code_cache_claimed;
+ Atomic _claim_counters[PSParallelCompact::last_space_id];
bool try_claim_code_cache_task() {
- return AtomicAccess::load(&_code_cache_claimed) == false
- && AtomicAccess::cmpxchg(&_code_cache_claimed, false, true) == false;
+ return _code_cache_claimed.load_relaxed() == false
+ && _code_cache_claimed.compare_set(false, true);
}
public:
@@ -1393,6 +1396,9 @@ public:
_nworkers(nworkers),
_code_cache_claimed(false) {
+ for (unsigned int i = PSParallelCompact::old_space_id; i < PSParallelCompact::last_space_id; ++i) {
+ ::new (&_claim_counters[i]) Atomic{};
+ }
ClassLoaderDataGraph::verify_claimed_marks_cleared(ClassLoaderData::_claim_stw_fullgc_adjust);
}
diff --git a/src/hotspot/share/gc/parallel/psParallelCompact.hpp b/src/hotspot/share/gc/parallel/psParallelCompact.hpp
index 4ac9395d727..f5ab041fa97 100644
--- a/src/hotspot/share/gc/parallel/psParallelCompact.hpp
+++ b/src/hotspot/share/gc/parallel/psParallelCompact.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2005, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2005, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -34,7 +34,7 @@
#include "gc/shared/referenceProcessor.hpp"
#include "gc/shared/taskTerminator.hpp"
#include "oops/oop.hpp"
-#include "runtime/atomicAccess.hpp"
+#include "runtime/atomic.hpp"
#include "runtime/orderAccess.hpp"
class ParallelScavengeHeap;
@@ -236,7 +236,7 @@ public:
// in this region (words). This does not include the partial object
// extending onto the region (if any), or the part of an object that extends
// onto the next region (if any).
- size_t live_obj_size() const { return _dc_and_los & los_mask; }
+ size_t live_obj_size() const { return dc_and_los() & los_mask; }
// Total live data that lies within the region (words).
size_t data_size() const { return partial_obj_size() + live_obj_size(); }
@@ -268,9 +268,9 @@ public:
// Minor subtlety: claimed() returns true if the region is marked
// completed(), which is desirable since a region must be claimed before it
// can be completed.
- bool available() const { return _dc_and_los < dc_one; }
- bool claimed() const { return _dc_and_los >= dc_claimed; }
- bool completed() const { return _dc_and_los >= dc_completed; }
+ bool available() const { return dc_and_los() < dc_one; }
+ bool claimed() const { return dc_and_los() >= dc_claimed; }
+ bool completed() const { return dc_and_los() >= dc_completed; }
// These are not atomic.
void set_destination(HeapWord* addr) { _destination = addr; }
@@ -315,7 +315,7 @@ public:
// Return to the normal path here
inline void shadow_to_normal();
- int shadow_state() { return _shadow_state; }
+ int shadow_state() { return _shadow_state.load_relaxed(); }
bool is_clear();
@@ -339,9 +339,10 @@ public:
size_t _source_region;
HeapWord* _partial_obj_addr;
region_sz_t _partial_obj_size;
- region_sz_t volatile _dc_and_los;
- int volatile _shadow_state;
+ Atomic _dc_and_los;
+ Atomic _shadow_state;
+ region_sz_t dc_and_los() const { return _dc_and_los.load_relaxed(); }
#ifdef ASSERT
public:
uint _pushed; // 0 until region is pushed onto a stack
@@ -411,7 +412,7 @@ private:
inline uint
ParallelCompactData::RegionData::destination_count_raw() const
{
- return _dc_and_los & dc_mask;
+ return dc_and_los() & dc_mask;
}
inline uint
@@ -425,26 +426,26 @@ ParallelCompactData::RegionData::set_destination_count(uint count)
{
assert(count <= (dc_completed >> dc_shift), "count too large");
const region_sz_t live_sz = (region_sz_t) live_obj_size();
- _dc_and_los = (count << dc_shift) | live_sz;
+ _dc_and_los.store_relaxed((count << dc_shift) | live_sz);
}
inline void ParallelCompactData::RegionData::set_live_obj_size(size_t words)
{
assert(words <= los_mask, "would overflow");
- _dc_and_los = destination_count_raw() | (region_sz_t)words;
+ _dc_and_los.store_relaxed(destination_count_raw() | (region_sz_t)words);
}
inline void ParallelCompactData::RegionData::decrement_destination_count()
{
- assert(_dc_and_los < dc_claimed, "already claimed");
- assert(_dc_and_los >= dc_one, "count would go negative");
- AtomicAccess::add(&_dc_and_los, dc_mask);
+ assert(dc_and_los() < dc_claimed, "already claimed");
+ assert(dc_and_los() >= dc_one, "count would go negative");
+ _dc_and_los.add_then_fetch(dc_mask);
}
inline void ParallelCompactData::RegionData::set_completed()
{
assert(claimed(), "must be claimed first");
- _dc_and_los = dc_completed | (region_sz_t) live_obj_size();
+ _dc_and_los.store_relaxed(dc_completed | (region_sz_t) live_obj_size());
}
// MT-unsafe claiming of a region. Should only be used during single threaded
@@ -452,7 +453,7 @@ inline void ParallelCompactData::RegionData::set_completed()
inline bool ParallelCompactData::RegionData::claim_unsafe()
{
if (available()) {
- _dc_and_los |= dc_claimed;
+ _dc_and_los.store_relaxed(dc_and_los() | dc_claimed);
return true;
}
return false;
@@ -461,36 +462,35 @@ inline bool ParallelCompactData::RegionData::claim_unsafe()
inline void ParallelCompactData::RegionData::add_live_obj(size_t words)
{
assert(words <= (size_t)los_mask - live_obj_size(), "overflow");
- AtomicAccess::add(&_dc_and_los, static_cast(words));
+ _dc_and_los.add_then_fetch(static_cast(words));
}
inline bool ParallelCompactData::RegionData::claim()
{
const region_sz_t los = static_cast(live_obj_size());
- const region_sz_t old = AtomicAccess::cmpxchg(&_dc_and_los, los, dc_claimed | los);
- return old == los;
+ return _dc_and_los.compare_set(los, dc_claimed | los);
}
inline bool ParallelCompactData::RegionData::mark_normal() {
- return AtomicAccess::cmpxchg(&_shadow_state, UnusedRegion, NormalRegion) == UnusedRegion;
+ return _shadow_state.compare_set(UnusedRegion, NormalRegion);
}
inline bool ParallelCompactData::RegionData::mark_shadow() {
- if (_shadow_state != UnusedRegion) return false;
- return AtomicAccess::cmpxchg(&_shadow_state, UnusedRegion, ShadowRegion) == UnusedRegion;
+ if (shadow_state() != UnusedRegion) return false;
+ return _shadow_state.compare_set(UnusedRegion, ShadowRegion);
}
inline void ParallelCompactData::RegionData::mark_filled() {
- int old = AtomicAccess::cmpxchg(&_shadow_state, ShadowRegion, FilledShadow);
+ int old = _shadow_state.compare_exchange(ShadowRegion, FilledShadow);
assert(old == ShadowRegion, "Fail to mark the region as filled");
}
inline bool ParallelCompactData::RegionData::mark_copied() {
- return AtomicAccess::cmpxchg(&_shadow_state, FilledShadow, CopiedShadow) == FilledShadow;
+ return _shadow_state.compare_set(FilledShadow, CopiedShadow);
}
void ParallelCompactData::RegionData::shadow_to_normal() {
- int old = AtomicAccess::cmpxchg(&_shadow_state, ShadowRegion, NormalRegion);
+ int old = _shadow_state.compare_exchange(ShadowRegion, NormalRegion);
assert(old == ShadowRegion, "Fail to mark the region as finish");
}
@@ -764,13 +764,13 @@ public:
static bool invoke(bool clear_all_soft_refs, bool should_do_max_compaction);
template
- static void adjust_in_space_helper(SpaceId id, volatile uint* claim_counter, Func&& on_stripe);
+ static void adjust_in_space_helper(SpaceId id, Atomic* claim_counter, Func&& on_stripe);
- static void adjust_in_old_space(volatile uint* claim_counter);
+ static void adjust_in_old_space(Atomic* claim_counter);
- static void adjust_in_young_space(SpaceId id, volatile uint* claim_counter);
+ static void adjust_in_young_space(SpaceId id, Atomic* claim_counter);
- static void adjust_pointers_in_spaces(uint worker_id, volatile uint* claim_counter);
+ static void adjust_pointers_in_spaces(uint worker_id, Atomic* claim_counter);
static void post_initialize();
// Perform initialization for PSParallelCompact that requires
diff --git a/src/hotspot/share/gc/parallel/psPromotionManager.cpp b/src/hotspot/share/gc/parallel/psPromotionManager.cpp
index a41a9403082..d6208755374 100644
--- a/src/hotspot/share/gc/parallel/psPromotionManager.cpp
+++ b/src/hotspot/share/gc/parallel/psPromotionManager.cpp
@@ -43,6 +43,7 @@
#include "memory/resourceArea.hpp"
#include "oops/access.inline.hpp"
#include "oops/compressedOops.inline.hpp"
+#include "oops/oopsHierarchy.hpp"
#include "utilities/checkedCast.hpp"
PaddedEnd* PSPromotionManager::_manager_array = nullptr;
@@ -248,30 +249,19 @@ void PSPromotionManager::flush_labs() {
}
}
-template
-void PSPromotionManager::process_array_chunk_work(oop obj, int start, int end) {
- assert(start <= end, "invariant");
- T* const base = (T*)objArrayOop(obj)->base();
- T* p = base + start;
- T* const chunk_end = base + end;
- while (p < chunk_end) {
- claim_or_forward_depth(p);
- ++p;
- }
+void PSPromotionManager::process_array_chunk(objArrayOop obj, size_t start, size_t end) {
+ PSPushContentsClosure pcc(this);
+ obj->oop_iterate_elements_range(&pcc,
+ checked_cast(start),
+ checked_cast(end));
}
void PSPromotionManager::process_array_chunk(PartialArrayState* state, bool stolen) {
// Access before release by claim().
- oop new_obj = state->destination();
+ objArrayOop to_array = objArrayOop(state->destination());
PartialArraySplitter::Claim claim =
_partial_array_splitter.claim(state, &_claimed_stack_depth, stolen);
- int start = checked_cast(claim._start);
- int end = checked_cast(claim._end);
- if (UseCompressedOops) {
- process_array_chunk_work(new_obj, start, end);
- } else {
- process_array_chunk_work(new_obj, start, end);
- }
+ process_array_chunk(to_array, claim._start, claim._end);
}
void PSPromotionManager::push_objArray(oop old_obj, oop new_obj) {
@@ -284,12 +274,8 @@ void PSPromotionManager::push_objArray(oop old_obj, oop new_obj) {
size_t initial_chunk_size =
// The source array is unused when processing states.
_partial_array_splitter.start(&_claimed_stack_depth, nullptr, to_array, array_length);
- int end = checked_cast(initial_chunk_size);
- if (UseCompressedOops) {
- process_array_chunk_work(to_array, 0, end);
- } else {
- process_array_chunk_work(to_array, 0, end);
- }
+
+ process_array_chunk(to_array, 0, initial_chunk_size);
}
oop PSPromotionManager::oop_promotion_failed(oop obj, markWord obj_mark) {
diff --git a/src/hotspot/share/gc/parallel/psPromotionManager.hpp b/src/hotspot/share/gc/parallel/psPromotionManager.hpp
index 44df708eea4..2b0fc56c0bf 100644
--- a/src/hotspot/share/gc/parallel/psPromotionManager.hpp
+++ b/src/hotspot/share/gc/parallel/psPromotionManager.hpp
@@ -97,9 +97,8 @@ class PSPromotionManager {
inline static PSPromotionManager* manager_array(uint index);
- template void process_array_chunk_work(oop obj,
- int start, int end);
void process_array_chunk(PartialArrayState* state, bool stolen);
+ void process_array_chunk(objArrayOop obj, size_t start, size_t end);
void push_objArray(oop old_obj, oop new_obj);
inline void promotion_trace_event(oop new_obj, Klass* klass, size_t obj_size,
diff --git a/src/hotspot/share/gc/parallel/psPromotionManager.inline.hpp b/src/hotspot/share/gc/parallel/psPromotionManager.inline.hpp
index f1fd49c7dfe..9e904e44b22 100644
--- a/src/hotspot/share/gc/parallel/psPromotionManager.inline.hpp
+++ b/src/hotspot/share/gc/parallel/psPromotionManager.inline.hpp
@@ -51,7 +51,7 @@ inline PSPromotionManager* PSPromotionManager::manager_array(uint index) {
}
template
-inline void PSPromotionManager::claim_or_forward_depth(T* p) {
+ALWAYSINLINE void PSPromotionManager::claim_or_forward_depth(T* p) {
assert(ParallelScavengeHeap::heap()->is_in(p), "pointer outside heap");
T heap_oop = RawAccess<>::oop_load(p);
if (PSScavenge::is_obj_in_young(heap_oop)) {
diff --git a/src/hotspot/share/gc/parallel/vmStructs_parallelgc.hpp b/src/hotspot/share/gc/parallel/vmStructs_parallelgc.hpp
index f69219a1f40..f8dabc4539e 100644
--- a/src/hotspot/share/gc/parallel/vmStructs_parallelgc.hpp
+++ b/src/hotspot/share/gc/parallel/vmStructs_parallelgc.hpp
@@ -47,8 +47,8 @@
nonstatic_field(PSVirtualSpace, _committed_high_addr, char*) \
\
nonstatic_field(MutableSpace, _bottom, HeapWord*) \
- nonstatic_field(MutableSpace, _end, HeapWord*) \
- volatile_nonstatic_field(MutableSpace, _top, HeapWord*) \
+ nonstatic_field(MutableSpace, _end, Atomic) \
+ volatile_nonstatic_field(MutableSpace, _top, Atomic) \
\
nonstatic_field(PSYoungGen, _reserved, MemRegion) \
nonstatic_field(PSYoungGen, _virtual_space, PSVirtualSpace*) \
diff --git a/src/hotspot/share/gc/shared/barrierSetNMethod.cpp b/src/hotspot/share/gc/shared/barrierSetNMethod.cpp
index ab94bae079a..a1f03a4bf50 100644
--- a/src/hotspot/share/gc/shared/barrierSetNMethod.cpp
+++ b/src/hotspot/share/gc/shared/barrierSetNMethod.cpp
@@ -111,6 +111,8 @@ bool BarrierSetNMethod::nmethod_entry_barrier(nmethod* nm) {
return true;
}
+ // Enable WXWrite: the function is called directly from nmethod_entry_barrier
+ // stub.
MACOS_AARCH64_ONLY(ThreadWXEnable wx(WXWrite, Thread::current()));
// If the nmethod is the only thing pointing to the oops, and we are using a
diff --git a/src/hotspot/share/gc/shared/c1/cardTableBarrierSetC1.cpp b/src/hotspot/share/gc/shared/c1/cardTableBarrierSetC1.cpp
index ebc1c1c7fb7..914358760aa 100644
--- a/src/hotspot/share/gc/shared/c1/cardTableBarrierSetC1.cpp
+++ b/src/hotspot/share/gc/shared/c1/cardTableBarrierSetC1.cpp
@@ -104,11 +104,8 @@ void CardTableBarrierSetC1::post_barrier(LIRAccess& access, LIR_Opr addr, LIR_Op
return;
}
- BarrierSet* bs = BarrierSet::barrier_set();
- CardTableBarrierSet* ctbs = barrier_set_cast(bs);
- CardTable* ct = ctbs->card_table();
- LIR_Const* card_table_base = new LIR_Const(ct->byte_map_base());
- SHENANDOAHGC_ONLY(assert(!UseShenandoahGC, "Shenandoah byte_map_base is not constant.");)
+ CardTableBarrierSet* ctbs = barrier_set_cast(BarrierSet::barrier_set());
+ LIR_Const* card_table_base = new LIR_Const(ctbs->card_table_base_const());
if (addr->is_address()) {
LIR_Address* address = addr->as_address_ptr();
diff --git a/src/hotspot/share/gc/shared/c2/cardTableBarrierSetC2.cpp b/src/hotspot/share/gc/shared/c2/cardTableBarrierSetC2.cpp
index fada2672e9f..42af77ebdf4 100644
--- a/src/hotspot/share/gc/shared/c2/cardTableBarrierSetC2.cpp
+++ b/src/hotspot/share/gc/shared/c2/cardTableBarrierSetC2.cpp
@@ -116,7 +116,7 @@ Node* CardTableBarrierSetC2::atomic_xchg_at_resolved(C2AtomicParseAccess& access
Node* CardTableBarrierSetC2::byte_map_base_node(GraphKit* kit) const {
// Get base of card map
- CardTable::CardValue* card_table_base = ci_card_table_address();
+ CardTable::CardValue* card_table_base = ci_card_table_address_const();
if (card_table_base != nullptr) {
return kit->makecon(TypeRawPtr::make((address)card_table_base));
} else {
diff --git a/src/hotspot/share/gc/shared/cardTableBarrierSet.cpp b/src/hotspot/share/gc/shared/cardTableBarrierSet.cpp
index 539e40820a8..d6541198858 100644
--- a/src/hotspot/share/gc/shared/cardTableBarrierSet.cpp
+++ b/src/hotspot/share/gc/shared/cardTableBarrierSet.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2000, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2000, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -73,15 +73,15 @@ CardTableBarrierSet::CardTableBarrierSet(CardTable* card_table) :
{}
CardTableBarrierSet::~CardTableBarrierSet() {
- delete _card_table;
+ delete card_table();
}
void CardTableBarrierSet::write_region(MemRegion mr) {
- _card_table->dirty_MemRegion(mr);
+ card_table()->dirty_MemRegion(mr);
}
void CardTableBarrierSet::print_on(outputStream* st) const {
- _card_table->print_on(st);
+ card_table()->print_on(st);
}
// Helper for ReduceInitialCardMarks. For performance,
@@ -116,7 +116,7 @@ void CardTableBarrierSet::on_slowpath_allocation_exit(JavaThread* thread, oop ne
if (!ReduceInitialCardMarks) {
return;
}
- if (new_obj->is_typeArray() || _card_table->is_in_young(new_obj)) {
+ if (new_obj->is_typeArray() || card_table()->is_in_young(new_obj)) {
// Arrays of non-references don't need a post-barrier.
} else {
MemRegion mr(cast_from_oop(new_obj), new_obj->size());
diff --git a/src/hotspot/share/gc/shared/cardTableBarrierSet.hpp b/src/hotspot/share/gc/shared/cardTableBarrierSet.hpp
index faeb007c77d..3a9b46d9df8 100644
--- a/src/hotspot/share/gc/shared/cardTableBarrierSet.hpp
+++ b/src/hotspot/share/gc/shared/cardTableBarrierSet.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2000, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2000, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -27,7 +27,9 @@
#include "gc/shared/barrierSet.hpp"
#include "gc/shared/cardTable.hpp"
+#include "gc/shared/gc_globals.hpp"
#include "memory/memRegion.hpp"
+#include "runtime/atomic.hpp"
#include "utilities/align.hpp"
// This kind of "BarrierSet" allows a "CollectedHeap" to detect and
@@ -48,7 +50,7 @@ class CardTableBarrierSet: public BarrierSet {
protected:
typedef CardTable::CardValue CardValue;
- CardTable* _card_table;
+ Atomic _card_table;
CardTableBarrierSet(BarrierSetAssembler* barrier_set_assembler,
BarrierSetC1* barrier_set_c1,
@@ -60,6 +62,10 @@ public:
CardTableBarrierSet(CardTable* card_table);
virtual ~CardTableBarrierSet();
+ inline static CardTableBarrierSet* barrier_set() {
+ return barrier_set_cast(BarrierSet::barrier_set());
+ }
+
template
inline void write_ref_field_pre(T* addr) {}
@@ -85,7 +91,13 @@ public:
// at the address "start", which may not necessarily be HeapWord-aligned
inline void write_ref_array(HeapWord* start, size_t count);
- CardTable* card_table() const { return _card_table; }
+ CardTable* card_table() { return _card_table.load_relaxed(); }
+ CardTable* card_table() const { return _card_table.load_relaxed(); }
+
+ CardValue* card_table_base_const() const {
+ assert(UseSerialGC || UseParallelGC, "Only these GCs have constant card table base");
+ return card_table()->byte_map_base();
+ }
virtual void on_slowpath_allocation_exit(JavaThread* thread, oop new_obj);
diff --git a/src/hotspot/share/gc/shared/cardTableBarrierSet.inline.hpp b/src/hotspot/share/gc/shared/cardTableBarrierSet.inline.hpp
index ea539a70be5..f60a7f47a19 100644
--- a/src/hotspot/share/gc/shared/cardTableBarrierSet.inline.hpp
+++ b/src/hotspot/share/gc/shared/cardTableBarrierSet.inline.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2000, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2000, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -35,7 +35,7 @@
template
inline void CardTableBarrierSet::write_ref_field_post(T* field) {
- volatile CardValue* byte = _card_table->byte_for(field);
+ volatile CardValue* byte = card_table()->byte_for(field);
*byte = CardTable::dirty_card_val();
}
diff --git a/src/hotspot/share/gc/shared/gc_globals.hpp b/src/hotspot/share/gc/shared/gc_globals.hpp
index d08e95378f7..6aa1fcf066b 100644
--- a/src/hotspot/share/gc/shared/gc_globals.hpp
+++ b/src/hotspot/share/gc/shared/gc_globals.hpp
@@ -261,6 +261,7 @@
develop(uintx, ObjArrayMarkingStride, 2048, \
"Number of object array elements to push onto the marking stack " \
"before pushing a continuation entry") \
+ range(1, INT_MAX/2) \
\
product_pd(bool, NeverActAsServerClassMachine, \
"(Deprecated) Never act like a server-class machine") \
diff --git a/src/hotspot/share/gc/shared/oopStorageSetParState.inline.hpp b/src/hotspot/share/gc/shared/oopStorageSetParState.inline.hpp
index 8e220e745e5..8341a3b20c4 100644
--- a/src/hotspot/share/gc/shared/oopStorageSetParState.inline.hpp
+++ b/src/hotspot/share/gc/shared/oopStorageSetParState.inline.hpp
@@ -31,7 +31,6 @@
#include "gc/shared/oopStorageSet.hpp"
#include "memory/iterator.hpp"
#include "oops/access.inline.hpp"
-#include "runtime/atomicAccess.hpp"
#include "utilities/debug.hpp"
template
diff --git a/src/hotspot/share/gc/shared/parallelCleaning.cpp b/src/hotspot/share/gc/shared/parallelCleaning.cpp
index e302085d0cc..1a0d536f3b3 100644
--- a/src/hotspot/share/gc/shared/parallelCleaning.cpp
+++ b/src/hotspot/share/gc/shared/parallelCleaning.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2018, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -39,7 +39,7 @@ CodeCacheUnloadingTask::CodeCacheUnloadingTask(bool unloading_occurred) :
if(iter.next()) {
_first_nmethod = iter.method();
}
- _claimed_nmethod = _first_nmethod;
+ _claimed_nmethod.store_relaxed(_first_nmethod);
}
CodeCacheUnloadingTask::~CodeCacheUnloadingTask() {
@@ -53,7 +53,7 @@ void CodeCacheUnloadingTask::claim_nmethods(nmethod** claimed_nmethods, int *num
do {
*num_claimed_nmethods = 0;
- first = _claimed_nmethod;
+ first = _claimed_nmethod.load_relaxed();
last = NMethodIterator(NMethodIterator::all, first);
if (first != nullptr) {
@@ -67,7 +67,7 @@ void CodeCacheUnloadingTask::claim_nmethods(nmethod** claimed_nmethods, int *num
}
}
- } while (AtomicAccess::cmpxchg(&_claimed_nmethod, first, last.method()) != first);
+ } while (!_claimed_nmethod.compare_set(first, last.method()));
}
void CodeCacheUnloadingTask::work(uint worker_id) {
diff --git a/src/hotspot/share/gc/shared/parallelCleaning.hpp b/src/hotspot/share/gc/shared/parallelCleaning.hpp
index ed76c4c9df9..0f5cb78bf55 100644
--- a/src/hotspot/share/gc/shared/parallelCleaning.hpp
+++ b/src/hotspot/share/gc/shared/parallelCleaning.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2018, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -29,6 +29,7 @@
#include "code/codeCache.hpp"
#include "gc/shared/oopStorageParState.hpp"
#include "gc/shared/workerThread.hpp"
+#include "runtime/atomic.hpp"
class CodeCacheUnloadingTask {
@@ -36,7 +37,7 @@ class CodeCacheUnloadingTask {
// Variables used to claim nmethods.
nmethod* _first_nmethod;
- nmethod* volatile _claimed_nmethod;
+ Atomic _claimed_nmethod;
public:
CodeCacheUnloadingTask(bool unloading_occurred);
diff --git a/src/hotspot/share/gc/shared/partialArrayState.cpp b/src/hotspot/share/gc/shared/partialArrayState.cpp
index 6f714d48a35..aadbc46b7c1 100644
--- a/src/hotspot/share/gc/shared/partialArrayState.cpp
+++ b/src/hotspot/share/gc/shared/partialArrayState.cpp
@@ -28,7 +28,6 @@
#include "memory/arena.hpp"
#include "nmt/memTag.hpp"
#include "oops/oopsHierarchy.hpp"
-#include "runtime/atomicAccess.hpp"
#include "runtime/orderAccess.hpp"
#include "utilities/debug.hpp"
#include "utilities/globalDefinitions.hpp"
diff --git a/src/hotspot/share/gc/shared/partialArrayTaskStepper.inline.hpp b/src/hotspot/share/gc/shared/partialArrayTaskStepper.inline.hpp
index aaa86e2de16..6946f7c69ff 100644
--- a/src/hotspot/share/gc/shared/partialArrayTaskStepper.inline.hpp
+++ b/src/hotspot/share/gc/shared/partialArrayTaskStepper.inline.hpp
@@ -28,7 +28,6 @@
#include "gc/shared/partialArrayTaskStepper.hpp"
#include "gc/shared/partialArrayState.hpp"
-#include "runtime/atomicAccess.hpp"
#include "utilities/checkedCast.hpp"
#include "utilities/debug.hpp"
diff --git a/src/hotspot/share/gc/shared/plab.hpp b/src/hotspot/share/gc/shared/plab.hpp
index 2eebdeeadb4..5200f022633 100644
--- a/src/hotspot/share/gc/shared/plab.hpp
+++ b/src/hotspot/share/gc/shared/plab.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2001, 2023, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -27,6 +27,7 @@
#include "gc/shared/collectedHeap.hpp"
#include "memory/allocation.hpp"
+#include "runtime/atomic.hpp"
#include "utilities/globalDefinitions.hpp"
// Forward declarations.
@@ -149,16 +150,16 @@ class PLABStats : public CHeapObj {
protected:
const char* _description; // Identifying string.
- size_t _allocated; // Total allocated
- size_t _wasted; // of which wasted (internal fragmentation)
- size_t _undo_wasted; // of which wasted on undo (is not used for calculation of PLAB size)
- size_t _unused; // Unused in last buffer
+ Atomic _allocated; // Total allocated
+ Atomic _wasted; // of which wasted (internal fragmentation)
+ Atomic _undo_wasted; // of which wasted on undo (is not used for calculation of PLAB size)
+ Atomic _unused; // Unused in last buffer
virtual void reset() {
- _allocated = 0;
- _wasted = 0;
- _undo_wasted = 0;
- _unused = 0;
+ _allocated.store_relaxed(0);
+ _wasted.store_relaxed(0);
+ _undo_wasted.store_relaxed(0);
+ _unused.store_relaxed(0);
}
public:
@@ -172,11 +173,11 @@ public:
virtual ~PLABStats() { }
- size_t allocated() const { return _allocated; }
- size_t wasted() const { return _wasted; }
- size_t unused() const { return _unused; }
+ size_t allocated() const { return _allocated.load_relaxed(); }
+ size_t wasted() const { return _wasted.load_relaxed(); }
+ size_t undo_wasted() const { return _undo_wasted.load_relaxed(); }
+ size_t unused() const { return _unused.load_relaxed(); }
size_t used() const { return allocated() - (wasted() + unused()); }
- size_t undo_wasted() const { return _undo_wasted; }
static size_t min_size() {
return PLAB::min_size();
diff --git a/src/hotspot/share/gc/shared/plab.inline.hpp b/src/hotspot/share/gc/shared/plab.inline.hpp
index 020738352d3..5f3e9c91e26 100644
--- a/src/hotspot/share/gc/shared/plab.inline.hpp
+++ b/src/hotspot/share/gc/shared/plab.inline.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2014, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2014, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -29,22 +29,21 @@
#include "gc/shared/collectedHeap.inline.hpp"
#include "memory/allocation.inline.hpp"
-#include "runtime/atomicAccess.hpp"
void PLABStats::add_allocated(size_t v) {
- AtomicAccess::add(&_allocated, v);
+ _allocated.add_then_fetch(v);
}
void PLABStats::add_unused(size_t v) {
- AtomicAccess::add(&_unused, v);
+ _unused.add_then_fetch(v);
}
void PLABStats::add_wasted(size_t v) {
- AtomicAccess::add(&_wasted, v);
+ _wasted.add_then_fetch(v);
}
void PLABStats::add_undo_wasted(size_t v) {
- AtomicAccess::add(&_undo_wasted, v);
+ _undo_wasted.add_then_fetch(v);
}
#endif // SHARE_GC_SHARED_PLAB_INLINE_HPP
diff --git a/src/hotspot/share/gc/shared/preservedMarks.cpp b/src/hotspot/share/gc/shared/preservedMarks.cpp
index 1c9f1c82e6f..605b7afe072 100644
--- a/src/hotspot/share/gc/shared/preservedMarks.cpp
+++ b/src/hotspot/share/gc/shared/preservedMarks.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2016, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2016, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -29,7 +29,7 @@
#include "memory/allocation.inline.hpp"
#include "memory/resourceArea.hpp"
#include "oops/oop.inline.hpp"
-#include "runtime/atomicAccess.hpp"
+#include "runtime/atomic.hpp"
#include "utilities/macros.hpp"
void PreservedMarks::restore() {
@@ -55,15 +55,6 @@ void PreservedMarks::adjust_during_full_gc() {
}
}
-void PreservedMarks::restore_and_increment(volatile size_t* const total_size_addr) {
- const size_t stack_size = size();
- restore();
- // Only do the atomic add if the size is > 0.
- if (stack_size > 0) {
- AtomicAccess::add(total_size_addr, stack_size);
- }
-}
-
#ifndef PRODUCT
void PreservedMarks::assert_empty() {
assert(_stack.is_empty(), "stack expected to be empty, size = %zu",
@@ -93,7 +84,7 @@ void PreservedMarksSet::init(uint num) {
class RestorePreservedMarksTask : public WorkerTask {
PreservedMarksSet* const _preserved_marks_set;
SequentialSubTasksDone _sub_tasks;
- volatile size_t _total_size;
+ Atomic _total_size;
#ifdef ASSERT
size_t _total_size_before;
#endif // ASSERT
@@ -102,7 +93,12 @@ public:
void work(uint worker_id) override {
uint task_id = 0;
while (_sub_tasks.try_claim_task(task_id)) {
- _preserved_marks_set->get(task_id)->restore_and_increment(&_total_size);
+ PreservedMarks* next = _preserved_marks_set->get(task_id);
+ size_t num_restored = next->size();
+ next->restore();
+ if (num_restored > 0) {
+ _total_size.add_then_fetch(num_restored);
+ }
}
}
@@ -121,9 +117,11 @@ public:
}
~RestorePreservedMarksTask() {
- assert(_total_size == _total_size_before, "total_size = %zu before = %zu", _total_size, _total_size_before);
- size_t mem_size = _total_size * (sizeof(oop) + sizeof(markWord));
- log_trace(gc)("Restored %zu marks, occupying %zu %s", _total_size,
+ size_t local_total_size = _total_size.load_relaxed();
+
+ assert(local_total_size == _total_size_before, "total_size = %zu before = %zu", local_total_size, _total_size_before);
+ size_t mem_size = local_total_size * (sizeof(oop) + sizeof(markWord));
+ log_trace(gc)("Restored %zu marks, occupying %zu %s", local_total_size,
byte_size_in_proper_unit(mem_size),
proper_unit_for_byte_size(mem_size));
}
diff --git a/src/hotspot/share/gc/shared/preservedMarks.hpp b/src/hotspot/share/gc/shared/preservedMarks.hpp
index 10f75116524..3bbbd335011 100644
--- a/src/hotspot/share/gc/shared/preservedMarks.hpp
+++ b/src/hotspot/share/gc/shared/preservedMarks.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2016, 2024, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2016, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -59,8 +59,7 @@ public:
size_t size() const { return _stack.size(); }
inline void push_if_necessary(oop obj, markWord m);
inline void push_always(oop obj, markWord m);
- // Iterate over the stack, restore all preserved marks, and
- // reclaim the memory taken up by the stack segments.
+ // Restore all preserved marks, and reclaim the memory taken up by the stack segments.
void restore();
// Adjust the preserved mark according to its
@@ -71,8 +70,6 @@ public:
// to their forwarding location stored in the mark.
void adjust_during_full_gc();
- void restore_and_increment(volatile size_t* const _total_size_addr);
-
// Assert the stack is empty and has no cached segments.
void assert_empty() PRODUCT_RETURN;
diff --git a/src/hotspot/share/gc/shared/referenceProcessorPhaseTimes.cpp b/src/hotspot/share/gc/shared/referenceProcessorPhaseTimes.cpp
index df7d8f7b38d..0371ed2c73b 100644
--- a/src/hotspot/share/gc/shared/referenceProcessorPhaseTimes.cpp
+++ b/src/hotspot/share/gc/shared/referenceProcessorPhaseTimes.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2017, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -30,7 +30,6 @@
#include "logging/logStream.hpp"
#include "memory/allocation.inline.hpp"
#include "memory/resourceArea.hpp"
-#include "runtime/atomicAccess.hpp"
#define ASSERT_REF_TYPE(ref_type) assert((ref_type) >= REF_SOFT && (ref_type) <= REF_PHANTOM, \
"Invariant (%d)", (int)ref_type)
@@ -196,7 +195,7 @@ void ReferenceProcessorPhaseTimes::reset() {
_soft_weak_final_refs_phase_worker_time_sec->reset();
for (int i = 0; i < number_of_subclasses_of_ref; i++) {
- _ref_dropped[i] = 0;
+ _ref_dropped[i].store_relaxed(0);
_ref_discovered[i] = 0;
}
@@ -214,7 +213,7 @@ ReferenceProcessorPhaseTimes::~ReferenceProcessorPhaseTimes() {
void ReferenceProcessorPhaseTimes::add_ref_dropped(ReferenceType ref_type, size_t count) {
ASSERT_REF_TYPE(ref_type);
- AtomicAccess::add(&_ref_dropped[ref_type_2_index(ref_type)], count, memory_order_relaxed);
+ _ref_dropped[ref_type_2_index(ref_type)].add_then_fetch(count, memory_order_relaxed);
}
void ReferenceProcessorPhaseTimes::set_ref_discovered(ReferenceType ref_type, size_t count) {
@@ -271,7 +270,7 @@ void ReferenceProcessorPhaseTimes::print_reference(ReferenceType ref_type, uint
int const ref_type_index = ref_type_2_index(ref_type);
size_t discovered = _ref_discovered[ref_type_index];
- size_t dropped = _ref_dropped[ref_type_index];
+ size_t dropped = _ref_dropped[ref_type_index].load_relaxed();
assert(discovered >= dropped, "invariant");
size_t processed = discovered - dropped;
diff --git a/src/hotspot/share/gc/shared/referenceProcessorPhaseTimes.hpp b/src/hotspot/share/gc/shared/referenceProcessorPhaseTimes.hpp
index 16691452ef4..82d26902bce 100644
--- a/src/hotspot/share/gc/shared/referenceProcessorPhaseTimes.hpp
+++ b/src/hotspot/share/gc/shared/referenceProcessorPhaseTimes.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017, 2019, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2017, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -30,6 +30,7 @@
#include "gc/shared/workerDataArray.hpp"
#include "memory/allocation.hpp"
#include "memory/referenceType.hpp"
+#include "runtime/atomic.hpp"
#include "utilities/ticks.hpp"
class DiscoveredList;
@@ -52,7 +53,7 @@ class ReferenceProcessorPhaseTimes : public CHeapObj {
// Total spent time for reference processing.
double _total_time_ms;
- size_t _ref_dropped[number_of_subclasses_of_ref];
+ Atomic _ref_dropped[number_of_subclasses_of_ref];
size_t _ref_discovered[number_of_subclasses_of_ref];
bool _processing_is_mt;
diff --git a/src/hotspot/share/gc/shared/space.cpp b/src/hotspot/share/gc/shared/space.cpp
index 011a0f5cfd8..84ba21527fd 100644
--- a/src/hotspot/share/gc/shared/space.cpp
+++ b/src/hotspot/share/gc/shared/space.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -30,7 +30,6 @@
#include "memory/iterator.inline.hpp"
#include "memory/universe.hpp"
#include "oops/oop.inline.hpp"
-#include "runtime/atomicAccess.hpp"
#include "runtime/java.hpp"
#include "runtime/safepoint.hpp"
#include "utilities/align.hpp"
@@ -69,7 +68,7 @@ void ContiguousSpace::clear(bool mangle_space) {
#ifndef PRODUCT
void ContiguousSpace::mangle_unused_area() {
- mangle_unused_area(MemRegion(_top, _end));
+ mangle_unused_area(MemRegion(top(), _end));
}
void ContiguousSpace::mangle_unused_area(MemRegion mr) {
@@ -128,11 +127,8 @@ inline HeapWord* ContiguousSpace::par_allocate_impl(size_t size) {
HeapWord* obj = top();
if (pointer_delta(end(), obj) >= size) {
HeapWord* new_top = obj + size;
- HeapWord* result = AtomicAccess::cmpxchg(top_addr(), obj, new_top);
- // result can be one of two:
- // the old top value: the exchange succeeded
- // otherwise: the new value of the top is returned.
- if (result == obj) {
+ // Retry if we did not successfully updated the top pointers.
+ if (_top.compare_set(obj, new_top)) {
assert(is_object_aligned(obj) && is_object_aligned(new_top), "checking alignment");
return obj;
}
diff --git a/src/hotspot/share/gc/shared/space.hpp b/src/hotspot/share/gc/shared/space.hpp
index 7f2887275b3..05b22f680bf 100644
--- a/src/hotspot/share/gc/shared/space.hpp
+++ b/src/hotspot/share/gc/shared/space.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -32,6 +32,7 @@
#include "memory/iterator.hpp"
#include "memory/memRegion.hpp"
#include "oops/markWord.hpp"
+#include "runtime/atomic.hpp"
#include "runtime/mutexLocker.hpp"
#include "utilities/align.hpp"
#include "utilities/macros.hpp"
@@ -53,7 +54,7 @@ class ContiguousSpace: public CHeapObj {
private:
HeapWord* _bottom;
HeapWord* _end;
- HeapWord* _top;
+ Atomic _top;
// Allocation helpers (return null if full).
inline HeapWord* allocate_impl(size_t word_size);
@@ -64,12 +65,12 @@ public:
// Accessors
HeapWord* bottom() const { return _bottom; }
- HeapWord* end() const { return _end; }
- HeapWord* top() const { return _top; }
+ HeapWord* end() const { return _end; }
+ HeapWord* top() const { return _top.load_relaxed(); }
void set_bottom(HeapWord* value) { _bottom = value; }
void set_end(HeapWord* value) { _end = value; }
- void set_top(HeapWord* value) { _top = value; }
+ void set_top(HeapWord* value) { _top.store_relaxed(value); }
// Testers
bool is_empty() const { return used() == 0; }
@@ -121,9 +122,6 @@ public:
// Iteration
void object_iterate(ObjectClosure* blk);
- // Addresses for inlined allocation
- HeapWord** top_addr() { return &_top; }
-
// Debugging
void verify() const;
};
diff --git a/src/hotspot/share/gc/shared/taskqueue.cpp b/src/hotspot/share/gc/shared/taskqueue.cpp
index f75dc4c2923..58af1793a48 100644
--- a/src/hotspot/share/gc/shared/taskqueue.cpp
+++ b/src/hotspot/share/gc/shared/taskqueue.cpp
@@ -25,7 +25,6 @@
#include "gc/shared/taskqueue.hpp"
#include "logging/log.hpp"
#include "oops/oop.inline.hpp"
-#include "runtime/atomicAccess.hpp"
#include "runtime/javaThread.hpp"
#include "runtime/os.hpp"
#include "utilities/debug.hpp"
diff --git a/src/hotspot/share/gc/shared/taskqueue.inline.hpp b/src/hotspot/share/gc/shared/taskqueue.inline.hpp
index 55851495a5f..e77645f4fcf 100644
--- a/src/hotspot/share/gc/shared/taskqueue.inline.hpp
+++ b/src/hotspot/share/gc/shared/taskqueue.inline.hpp
@@ -32,7 +32,6 @@
#include "memory/allocation.inline.hpp"
#include "memory/resourceArea.hpp"
#include "oops/oop.inline.hpp"
-#include "runtime/atomicAccess.hpp"
#include "runtime/orderAccess.hpp"
#include "utilities/debug.hpp"
#include "utilities/ostream.hpp"
diff --git a/src/hotspot/share/gc/shared/threadLocalAllocBuffer.cpp b/src/hotspot/share/gc/shared/threadLocalAllocBuffer.cpp
index 9635ed4d0cb..d99544c0573 100644
--- a/src/hotspot/share/gc/shared/threadLocalAllocBuffer.cpp
+++ b/src/hotspot/share/gc/shared/threadLocalAllocBuffer.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -458,10 +458,22 @@ size_t ThreadLocalAllocBuffer::end_reserve() {
return MAX2(reserve_size, (size_t)_reserve_for_allocation_prefetch);
}
-const HeapWord* ThreadLocalAllocBuffer::start_relaxed() const {
- return AtomicAccess::load(&_start);
-}
-
-const HeapWord* ThreadLocalAllocBuffer::top_relaxed() const {
- return AtomicAccess::load(&_top);
+size_t ThreadLocalAllocBuffer::estimated_used_bytes() const {
+ // Data races due to unsynchronized access like the following reads to _start
+ // and _top are undefined behavior. Atomic would not provide any additional
+ // guarantees, so use AtomicAccess directly.
+ HeapWord* start = AtomicAccess::load(&_start);
+ HeapWord* top = AtomicAccess::load(&_top);
+ // If there has been a race when retrieving _top and _start, return 0.
+ if (top < start) {
+ return 0;
+ }
+ size_t used_bytes = pointer_delta(top, start, 1);
+ // Comparing diff with the maximum allowed size will ensure that we don't add
+ // the used bytes from a semi-initialized TLAB ending up with implausible values.
+ // In this case also just return 0.
+ if (used_bytes > ThreadLocalAllocBuffer::max_size_in_bytes()) {
+ return 0;
+ }
+ return used_bytes;
}
diff --git a/src/hotspot/share/gc/shared/threadLocalAllocBuffer.hpp b/src/hotspot/share/gc/shared/threadLocalAllocBuffer.hpp
index 8267a103539..61caac7ec51 100644
--- a/src/hotspot/share/gc/shared/threadLocalAllocBuffer.hpp
+++ b/src/hotspot/share/gc/shared/threadLocalAllocBuffer.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -32,17 +32,16 @@
class ThreadLocalAllocStats;
-// ThreadLocalAllocBuffer: a descriptor for thread-local storage used by
-// the threads for allocation.
-// It is thread-private at any time, but maybe multiplexed over
-// time across multiple threads. The park()/unpark() pair is
-// used to make it available for such multiplexing.
+// ThreadLocalAllocBuffer is a descriptor for thread-local storage used by
+// mutator threads for local/private allocation. As a TLAB is thread-private,
+// there is no concurrent/parallel access to its memory or its members,
+// other than by estimated_used_bytes().
//
-// Heap sampling is performed via the end and allocation_end
-// fields.
-// allocation_end contains the real end of the tlab allocation,
-// whereas end can be set to an arbitrary spot in the tlab to
-// trip the return and sample the allocation.
+// Heap sampling is performed via the end and allocation_end
+// fields.
+// allocation_end contains the real end of the tlab allocation,
+// whereas end can be set to an arbitrary spot in the tlab to
+// trip the return and sample the allocation.
class ThreadLocalAllocBuffer: public CHeapObj {
friend class VMStructs;
friend class JVMCIVMStructs;
@@ -116,17 +115,18 @@ public:
HeapWord* end() const { return _end; }
HeapWord* top() const { return _top; }
HeapWord* hard_end();
- HeapWord* pf_top() const { return _pf_top; }
size_t desired_size() const { return _desired_size; }
- size_t used() const { return pointer_delta(top(), start()); }
size_t used_bytes() const { return pointer_delta(top(), start(), 1); }
size_t free() const { return pointer_delta(end(), top()); }
// Don't discard tlab if remaining space is larger than this.
size_t refill_waste_limit() const { return _refill_waste_limit; }
- // For external inspection.
- const HeapWord* start_relaxed() const;
- const HeapWord* top_relaxed() const;
+ // Returns an estimate of the number of bytes currently used in the TLAB.
+ // Due to races with concurrent allocations and/or resetting the TLAB the return
+ // value may be inconsistent with any other metrics (e.g. total allocated
+ // bytes), and may just incorrectly return 0.
+ // Intended for external inspection only where accuracy is not 100% required.
+ size_t estimated_used_bytes() const;
// Allocate size HeapWords. The memory is NOT initialized to zero.
inline HeapWord* allocate(size_t size);
@@ -171,14 +171,6 @@ public:
static size_t refill_waste_limit_increment();
- template void addresses_do(T f) {
- f(&_start);
- f(&_top);
- f(&_pf_top);
- f(&_end);
- f(&_allocation_end);
- }
-
// Code generation support
static ByteSize start_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _start); }
static ByteSize end_offset() { return byte_offset_of(ThreadLocalAllocBuffer, _end); }
diff --git a/src/hotspot/share/gc/shared/vmStructs_gc.hpp b/src/hotspot/share/gc/shared/vmStructs_gc.hpp
index db968e28f67..9348fd980f4 100644
--- a/src/hotspot/share/gc/shared/vmStructs_gc.hpp
+++ b/src/hotspot/share/gc/shared/vmStructs_gc.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2018, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -48,6 +48,7 @@
#if INCLUDE_ZGC
#include "gc/z/vmStructs_z.hpp"
#endif
+#include "runtime/atomic.hpp"
#define VM_STRUCTS_GC(nonstatic_field, \
volatile_static_field, \
@@ -88,7 +89,7 @@
nonstatic_field(CardTable, _byte_map_size, const size_t) \
nonstatic_field(CardTable, _byte_map, CardTable::CardValue*) \
nonstatic_field(CardTable, _byte_map_base, CardTable::CardValue*) \
- nonstatic_field(CardTableBarrierSet, _card_table, CardTable*) \
+ nonstatic_field(CardTableBarrierSet, _card_table, Atomic) \
\
static_field(CollectedHeap, _lab_alignment_reserve, size_t) \
nonstatic_field(CollectedHeap, _reserved, MemRegion) \
@@ -97,7 +98,7 @@
\
nonstatic_field(ContiguousSpace, _bottom, HeapWord*) \
nonstatic_field(ContiguousSpace, _end, HeapWord*) \
- nonstatic_field(ContiguousSpace, _top, HeapWord*) \
+ nonstatic_field(ContiguousSpace, _top, Atomic) \
\
nonstatic_field(MemRegion, _start, HeapWord*) \
nonstatic_field(MemRegion, _word_size, size_t)
@@ -149,6 +150,7 @@
\
declare_toplevel_type(BarrierSet*) \
declare_toplevel_type(CardTable*) \
+ declare_toplevel_type(Atomic) \
declare_toplevel_type(CardTable*const) \
declare_toplevel_type(CardTableBarrierSet*) \
declare_toplevel_type(CardTableBarrierSet**) \
diff --git a/src/hotspot/share/gc/shared/workerThread.cpp b/src/hotspot/share/gc/shared/workerThread.cpp
index e4831d25d26..2f6f003608f 100644
--- a/src/hotspot/share/gc/shared/workerThread.cpp
+++ b/src/hotspot/share/gc/shared/workerThread.cpp
@@ -26,7 +26,6 @@
#include "gc/shared/workerThread.hpp"
#include "logging/log.hpp"
#include "memory/iterator.hpp"
-#include "runtime/atomicAccess.hpp"
#include "runtime/init.hpp"
#include "runtime/java.hpp"
#include "runtime/os.hpp"
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAdaptiveHeuristics.cpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAdaptiveHeuristics.cpp
index 46d9f19d35f..7a8bd55c795 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAdaptiveHeuristics.cpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAdaptiveHeuristics.cpp
@@ -68,9 +68,9 @@ ShenandoahAdaptiveHeuristics::ShenandoahAdaptiveHeuristics(ShenandoahSpaceInfo*
ShenandoahAdaptiveHeuristics::~ShenandoahAdaptiveHeuristics() {}
-size_t ShenandoahAdaptiveHeuristics::choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
- RegionData* data, size_t size,
- size_t actual_free) {
+void ShenandoahAdaptiveHeuristics::choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
+ RegionData* data, size_t size,
+ size_t actual_free) {
size_t garbage_threshold = ShenandoahHeapRegion::region_size_bytes() * ShenandoahGarbageThreshold / 100;
// The logic for cset selection in adaptive is as follows:
@@ -124,7 +124,6 @@ size_t ShenandoahAdaptiveHeuristics::choose_collection_set_from_regiondata(Shena
cur_garbage = new_garbage;
}
}
- return 0;
}
void ShenandoahAdaptiveHeuristics::record_cycle_start() {
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAdaptiveHeuristics.hpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAdaptiveHeuristics.hpp
index c4fdf819391..9b7824a50d7 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAdaptiveHeuristics.hpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAdaptiveHeuristics.hpp
@@ -33,7 +33,7 @@
#include "utilities/numberSeq.hpp"
/**
- * ShenanoahAllocationRate maintains a truncated history of recently sampled allocation rates for the purpose of providing
+ * ShenandoahAllocationRate maintains a truncated history of recently sampled allocation rates for the purpose of providing
* informed estimates of current and future allocation rates based on weighted averages and standard deviations of the
* truncated history. More recently sampled allocations are weighted more heavily than older samples when computing
* averages and standard deviations.
@@ -108,20 +108,20 @@ public:
virtual ~ShenandoahAdaptiveHeuristics();
- virtual size_t choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
+ void choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
RegionData* data, size_t size,
size_t actual_free) override;
- virtual void record_cycle_start() override;
- virtual void record_success_concurrent() override;
- virtual void record_degenerated() override;
- virtual void record_success_full() override;
+ void record_cycle_start() override;
+ void record_success_concurrent() override;
+ void record_degenerated() override;
+ void record_success_full() override;
- virtual bool should_start_gc() override;
+ bool should_start_gc() override;
- virtual const char* name() override { return "Adaptive"; }
- virtual bool is_diagnostic() override { return false; }
- virtual bool is_experimental() override { return false; }
+ const char* name() override { return "Adaptive"; }
+ bool is_diagnostic() override { return false; }
+ bool is_experimental() override { return false; }
private:
// These are used to adjust the margin of error and the spike threshold
@@ -185,7 +185,7 @@ protected:
// in the generational case. Controlled by global flag ShenandoahMinFreeThreshold.
size_t min_free_threshold();
- inline void accept_trigger_with_type(Trigger trigger_type) {
+ void accept_trigger_with_type(Trigger trigger_type) {
_last_trigger = trigger_type;
ShenandoahHeuristics::accept_trigger();
}
@@ -193,7 +193,7 @@ protected:
public:
// Sample the allocation rate at GC trigger time if possible. Return the number of allocated bytes that were
// not accounted for in the sample. This must be called before resetting bytes allocated since gc start.
- virtual size_t force_alloc_rate_sample(size_t bytes_allocated) override {
+ size_t force_alloc_rate_sample(size_t bytes_allocated) override {
size_t unaccounted_bytes;
_allocation_rate.force_sample(bytes_allocated, unaccounted_bytes);
return unaccounted_bytes;
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAggressiveHeuristics.cpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAggressiveHeuristics.cpp
index 990b59ec853..a833e39631c 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAggressiveHeuristics.cpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAggressiveHeuristics.cpp
@@ -39,16 +39,15 @@ ShenandoahAggressiveHeuristics::ShenandoahAggressiveHeuristics(ShenandoahSpaceIn
SHENANDOAH_ERGO_ENABLE_FLAG(ShenandoahEvacReserveOverflow);
}
-size_t ShenandoahAggressiveHeuristics::choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
- RegionData* data, size_t size,
- size_t free) {
+void ShenandoahAggressiveHeuristics::choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
+ RegionData* data, size_t size,
+ size_t free) {
for (size_t idx = 0; idx < size; idx++) {
ShenandoahHeapRegion* r = data[idx].get_region();
if (r->garbage() > 0) {
cset->add_region(r);
}
}
- return 0;
}
bool ShenandoahAggressiveHeuristics::should_start_gc() {
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAggressiveHeuristics.hpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAggressiveHeuristics.hpp
index 25c8635489f..9dc88a61bf5 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAggressiveHeuristics.hpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahAggressiveHeuristics.hpp
@@ -35,17 +35,17 @@ class ShenandoahAggressiveHeuristics : public ShenandoahHeuristics {
public:
ShenandoahAggressiveHeuristics(ShenandoahSpaceInfo* space_info);
- virtual size_t choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
- RegionData* data, size_t size,
- size_t free);
+ void choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
+ RegionData* data, size_t size,
+ size_t free) override;
- virtual bool should_start_gc();
+ bool should_start_gc() override;
- virtual bool should_unload_classes();
+ bool should_unload_classes() override;
- virtual const char* name() { return "Aggressive"; }
- virtual bool is_diagnostic() { return true; }
- virtual bool is_experimental() { return false; }
+ const char* name() override { return "Aggressive"; }
+ bool is_diagnostic() override { return true; }
+ bool is_experimental() override { return false; }
};
#endif // SHARE_GC_SHENANDOAH_HEURISTICS_SHENANDOAHAGGRESSIVEHEURISTICS_HPP
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahCompactHeuristics.cpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahCompactHeuristics.cpp
index 09a8394a4b1..28673b28612 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahCompactHeuristics.cpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahCompactHeuristics.cpp
@@ -76,9 +76,9 @@ bool ShenandoahCompactHeuristics::should_start_gc() {
return ShenandoahHeuristics::should_start_gc();
}
-size_t ShenandoahCompactHeuristics::choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
- RegionData* data, size_t size,
- size_t actual_free) {
+void ShenandoahCompactHeuristics::choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
+ RegionData* data, size_t size,
+ size_t actual_free) {
// Do not select too large CSet that would overflow the available free space
size_t max_cset = actual_free * 3 / 4;
@@ -97,5 +97,4 @@ size_t ShenandoahCompactHeuristics::choose_collection_set_from_regiondata(Shenan
cset->add_region(r);
}
}
- return 0;
}
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahCompactHeuristics.hpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahCompactHeuristics.hpp
index 4988d5d495d..a32c9c88478 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahCompactHeuristics.hpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahCompactHeuristics.hpp
@@ -33,17 +33,17 @@
*/
class ShenandoahCompactHeuristics : public ShenandoahHeuristics {
public:
- ShenandoahCompactHeuristics(ShenandoahSpaceInfo* space_info);
+ explicit ShenandoahCompactHeuristics(ShenandoahSpaceInfo* space_info);
- virtual bool should_start_gc();
+ bool should_start_gc() override;
- virtual size_t choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
- RegionData* data, size_t size,
- size_t actual_free);
+ void choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
+ RegionData* data, size_t size,
+ size_t actual_free) override;
- virtual const char* name() { return "Compact"; }
- virtual bool is_diagnostic() { return false; }
- virtual bool is_experimental() { return false; }
+ const char* name() override { return "Compact"; }
+ bool is_diagnostic() override { return false; }
+ bool is_experimental() override { return false; }
};
#endif // SHARE_GC_SHENANDOAH_HEURISTICS_SHENANDOAHCOMPACTHEURISTICS_HPP
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahGenerationalHeuristics.cpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahGenerationalHeuristics.cpp
index ee315ce5c7e..80e6decf57d 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahGenerationalHeuristics.cpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahGenerationalHeuristics.cpp
@@ -25,19 +25,205 @@
#include "gc/shenandoah/heuristics/shenandoahGenerationalHeuristics.hpp"
#include "gc/shenandoah/shenandoahCollectionSet.hpp"
+#include "gc/shenandoah/shenandoahCollectionSetPreselector.hpp"
#include "gc/shenandoah/shenandoahCollectorPolicy.hpp"
#include "gc/shenandoah/shenandoahGeneration.hpp"
#include "gc/shenandoah/shenandoahGenerationalHeap.inline.hpp"
#include "gc/shenandoah/shenandoahHeapRegion.inline.hpp"
#include "gc/shenandoah/shenandoahOldGeneration.hpp"
#include "gc/shenandoah/shenandoahTrace.hpp"
+#include "gc/shenandoah/shenandoahYoungGeneration.hpp"
#include "logging/log.hpp"
+#include "utilities/quickSort.hpp"
-ShenandoahGenerationalHeuristics::ShenandoahGenerationalHeuristics(ShenandoahGeneration* generation)
- : ShenandoahAdaptiveHeuristics(generation), _generation(generation) {
+using idx_t = ShenandoahSimpleBitMap::idx_t;
+
+typedef struct {
+ ShenandoahHeapRegion* _region;
+ size_t _live_data;
+} AgedRegionData;
+
+static int compare_by_aged_live(AgedRegionData a, AgedRegionData b) {
+ if (a._live_data < b._live_data)
+ return -1;
+ if (a._live_data > b._live_data)
+ return 1;
+ return 0;
}
-size_t ShenandoahGenerationalHeuristics::choose_collection_set(ShenandoahCollectionSet* collection_set) {
+inline void assert_no_in_place_promotions() {
+#ifdef ASSERT
+ class ShenandoahNoInPlacePromotions : public ShenandoahHeapRegionClosure {
+ public:
+ void heap_region_do(ShenandoahHeapRegion *r) override {
+ assert(r->get_top_before_promote() == nullptr,
+ "Region %zu should not be ready for in-place promotion", r->index());
+ }
+ } cl;
+ ShenandoahHeap::heap()->heap_region_iterate(&cl);
+#endif
+}
+
+ShenandoahGenerationalHeuristics::ShenandoahGenerationalHeuristics(ShenandoahGeneration* generation)
+ : ShenandoahAdaptiveHeuristics(generation), _generation(generation), _add_regions_to_old(0) {
+}
+
+void ShenandoahGenerationalHeuristics::choose_collection_set(ShenandoahCollectionSet* collection_set) {
+ ShenandoahHeap* heap = ShenandoahHeap::heap();
+
+ _add_regions_to_old = 0;
+
+ // Seed the collection set with resource area-allocated
+ // preselected regions, which are removed when we exit this scope.
+ ShenandoahCollectionSetPreselector preselector(collection_set, heap->num_regions());
+
+ // Find the amount that will be promoted, regions that will be promoted in
+ // place, and preselected older regions that will be promoted by evacuation.
+ compute_evacuation_budgets(heap);
+
+ // Choose the collection set, including the regions preselected above for promotion into the old generation.
+ filter_regions(collection_set);
+
+ // Even if collection_set->is_empty(), we want to adjust budgets, making reserves available to mutator.
+ adjust_evacuation_budgets(heap, collection_set);
+
+ if (_generation->is_global()) {
+ // We have just chosen a collection set for a global cycle. The mark bitmap covering old regions is complete, so
+ // the remembered set scan can use that to avoid walking into garbage. When the next old mark begins, we will
+ // use the mark bitmap to make the old regions parsable by coalescing and filling any unmarked objects. Thus,
+ // we prepare for old collections by remembering which regions are old at this time. Note that any objects
+ // promoted into old regions will be above TAMS, and so will be considered marked. However, free regions that
+ // become old after this point will not be covered correctly by the mark bitmap, so we must be careful not to
+ // coalesce those regions. Only the old regions which are not part of the collection set at this point are
+ // eligible for coalescing. As implemented now, this has the side effect of possibly initiating mixed-evacuations
+ // after a global cycle for old regions that were not included in this collection set.
+ heap->old_generation()->prepare_for_mixed_collections_after_global_gc();
+ }
+}
+
+void ShenandoahGenerationalHeuristics::compute_evacuation_budgets(ShenandoahHeap* const heap) {
+ shenandoah_assert_generational();
+
+ ShenandoahOldGeneration* const old_generation = heap->old_generation();
+ ShenandoahYoungGeneration* const young_generation = heap->young_generation();
+ const size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
+
+ // During initialization and phase changes, it is more likely that fewer objects die young and old-gen
+ // memory is not yet full (or is in the process of being replaced). During these times especially, it
+ // is beneficial to loan memory from old-gen to young-gen during the evacuation and update-refs phases
+ // of execution.
+
+ // Calculate EvacuationReserve before PromotionReserve. Evacuation is more critical than promotion.
+ // If we cannot evacuate old-gen, we will not be able to reclaim old-gen memory. Promotions are less
+ // critical. If we cannot promote, there may be degradation of young-gen memory because old objects
+ // accumulate there until they can be promoted. This increases the young-gen marking and evacuation work.
+
+ // First priority is to reclaim the easy garbage out of young-gen.
+
+ // maximum_young_evacuation_reserve is upper bound on memory to be evacuated into young Collector Reserve. This is
+ // bounded at the end of previous GC cycle, based on available memory and balancing of evacuation to old and young.
+ size_t maximum_young_evacuation_reserve = young_generation->get_evacuation_reserve();
+
+ // maximum_old_evacuation_reserve is an upper bound on memory evacuated from old and evacuated to old (promoted),
+ // clamped by the old generation space available.
+ //
+ // Here's the algebra.
+ // Let SOEP = ShenandoahOldEvacPercent,
+ // OE = old evac,
+ // YE = young evac, and
+ // TE = total evac = OE + YE
+ // By definition:
+ // SOEP/100 = OE/TE
+ // = OE/(OE+YE)
+ // => SOEP/(100-SOEP) = OE/((OE+YE)-OE) // componendo-dividendo: If a/b = c/d, then a/(b-a) = c/(d-c)
+ // = OE/YE
+ // => OE = YE*SOEP/(100-SOEP)
+
+ // We have to be careful in the event that SOEP is set to 100 by the user.
+ assert(ShenandoahOldEvacPercent <= 100, "Error");
+ const size_t old_available = old_generation->available();
+ const size_t maximum_old_evacuation_reserve = (ShenandoahOldEvacPercent == 100) ?
+ old_available : MIN2((maximum_young_evacuation_reserve * ShenandoahOldEvacPercent) / (100 - ShenandoahOldEvacPercent),
+ old_available);
+
+ // In some cases, maximum_old_reserve < old_available (when limited by ShenandoahOldEvacPercent)
+ // This limit affects mixed evacuations, but does not affect promotions.
+
+ // Second priority is to reclaim garbage out of old-gen if there are old-gen collection candidates. Third priority
+ // is to promote as much as we have room to promote. However, if old-gen memory is in short supply, this means young
+ // GC is operating under "duress" and was unable to transfer the memory that we would normally expect. In this case,
+ // old-gen will refrain from compacting itself in order to allow a quicker young-gen cycle (by avoiding the update-refs
+ // through ALL of old-gen). If there is some memory available in old-gen, we will use this for promotions as promotions
+ // do not add to the update-refs burden of GC.
+
+ size_t old_evacuation_reserve, old_promo_reserve;
+ if (_generation->is_global()) {
+ // Global GC is typically triggered by user invocation of System.gc(), and typically indicates that there is lots
+ // of garbage to be reclaimed because we are starting a new phase of execution. Marking for global GC may take
+ // significantly longer than typical young marking because we must mark through all old objects. To expedite
+ // evacuation and update-refs, we give emphasis to reclaiming garbage first, wherever that garbage is found.
+ // Global GC will adjust generation sizes to accommodate the collection set it chooses.
+
+ // Use remnant of old_available to hold promotions.
+ old_promo_reserve = old_available - maximum_old_evacuation_reserve;
+
+ // Dedicate all available old memory to old_evacuation reserve. This may be small, because old-gen is only
+ // expanded based on an existing mixed evacuation workload at the end of the previous GC cycle. We'll expand
+ // the budget for evacuation of old during GLOBAL cset selection.
+ old_evacuation_reserve = maximum_old_evacuation_reserve;
+ } else if (old_generation->has_unprocessed_collection_candidates()) {
+ // We reserved all old-gen memory at end of previous GC to hold anticipated evacuations to old-gen. If this is
+ // mixed evacuation, reserve all of this memory for compaction of old-gen and do not promote. Prioritize compaction
+ // over promotion in order to defragment OLD so that it will be better prepared to efficiently receive promoted memory.
+ old_evacuation_reserve = maximum_old_evacuation_reserve;
+ old_promo_reserve = old_available - maximum_old_evacuation_reserve;
+ } else {
+ // Make all old-evacuation memory for promotion, but if we can't use it all for promotion, we'll allow some evacuation.
+ old_evacuation_reserve = old_available - maximum_old_evacuation_reserve;
+ old_promo_reserve = maximum_old_evacuation_reserve;
+ }
+ assert(old_evacuation_reserve <= old_available, "Error");
+
+
+ // We see too many old-evacuation failures if we force ourselves to evacuate into regions that are not initially empty.
+ // So we limit the old-evacuation reserve to unfragmented memory. Even so, old-evacuation is free to fill in nooks and
+ // crannies within existing partially used regions and it generally tries to do so.
+ const size_t old_free_unfragmented = old_generation->free_unaffiliated_regions() * region_size_bytes;
+ if (old_evacuation_reserve > old_free_unfragmented) {
+ const size_t delta = old_evacuation_reserve - old_free_unfragmented;
+ old_evacuation_reserve -= delta;
+ // Let promo consume fragments of old-gen memory
+ old_promo_reserve += delta;
+ }
+
+ // If is_global(), we let garbage-first heuristic determine cset membership. Otherwise, we give priority
+ // to tenurable regions by preselecting regions for promotion by evacuation (obtaining the live data to seed promoted_reserve).
+ // This also identifies regions that will be promoted in place. These use the tenuring threshold.
+ const size_t consumed_by_advance_promotion = select_aged_regions(_generation->is_global()? 0: old_promo_reserve);
+ assert(consumed_by_advance_promotion <= old_promo_reserve, "Do not promote more than budgeted");
+
+ // The young evacuation reserve can be no larger than young_unaffiliated. Planning to evacuate into partially consumed
+ // young regions is doomed to failure if any of those partially consumed regions is selected for the collection set.
+ size_t young_unaffiliated = young_generation->free_unaffiliated_regions() * region_size_bytes;
+
+ // If any regions have been selected for promotion in place, this has the effect of decreasing available within mutator
+ // and collector partitions, due to padding of remnant memory within each promoted in place region. This will affect
+ // young_evacuation_reserve but not old_evacuation_reserve or consumed_by_advance_promotion. So recompute.
+ size_t young_evacuation_reserve = MIN2(maximum_young_evacuation_reserve, young_unaffiliated);
+
+ // Note that unused old_promo_reserve might not be entirely consumed_by_advance_promotion. Do not transfer this
+ // to old_evacuation_reserve because this memory is likely very fragmented, and we do not want to increase the likelihood
+ // of old evacuation failure. Leave this memory in the promoted reserve as it may be targeted by opportunistic
+ // promotions (found during evacuation of young regions).
+ young_generation->set_evacuation_reserve(young_evacuation_reserve);
+ old_generation->set_evacuation_reserve(old_evacuation_reserve);
+ old_generation->set_promoted_reserve(old_promo_reserve);
+
+ // There is no need to expand OLD because all memory used here was set aside at end of previous GC, except in the
+ // case of a GLOBAL gc. During choose_collection_set() of GLOBAL, old will be expanded on demand.
+}
+
+void ShenandoahGenerationalHeuristics::filter_regions(ShenandoahCollectionSet* collection_set) {
assert(collection_set->is_empty(), "Must be empty");
auto heap = ShenandoahGenerationalHeap::heap();
@@ -170,10 +356,9 @@ size_t ShenandoahGenerationalHeuristics::choose_collection_set(ShenandoahCollect
size_t immediate_percent = (total_garbage == 0) ? 0 : (immediate_garbage * 100 / total_garbage);
bool doing_promote_in_place = (humongous_regions_promoted + regular_regions_promoted_in_place > 0);
- size_t add_regions_to_old = 0;
if (doing_promote_in_place || (preselected_candidates > 0) || (immediate_percent <= ShenandoahImmediateThreshold)) {
// Call the subclasses to add young-gen regions into the collection set.
- add_regions_to_old = choose_collection_set_from_regiondata(collection_set, candidates, cand_idx, immediate_garbage + free);
+ choose_collection_set_from_regiondata(collection_set, candidates, cand_idx, immediate_garbage + free);
}
if (collection_set->has_old_regions()) {
@@ -190,9 +375,359 @@ size_t ShenandoahGenerationalHeuristics::choose_collection_set(ShenandoahCollect
regular_regions_promoted_free,
immediate_regions,
immediate_garbage);
- return add_regions_to_old;
}
+// Preselect for inclusion into the collection set all regions whose age is at or above tenure age and for which the
+// garbage percentage exceeds a dynamically adjusted threshold (known as the old-garbage threshold percentage). We
+// identify these regions by setting the appropriate entry of the collection set's preselected regions array to true.
+// All entries are initialized to false before calling this function.
+//
+// During the subsequent selection of the collection set, we give priority to these promotion set candidates.
+// Without this prioritization, we found that the aged regions tend to be ignored because they typically have
+// much less garbage and much more live data than the recently allocated "eden" regions. When aged regions are
+// repeatedly excluded from the collection set, the amount of live memory within the young generation tends to
+// accumulate and this has the undesirable side effect of causing young-generation collections to require much more
+// CPU and wall-clock time.
+//
+// A second benefit of treating aged regions differently than other regions during collection set selection is
+// that this allows us to more accurately budget memory to hold the results of evacuation. Memory for evacuation
+// of aged regions must be reserved in the old generation. Memory for evacuation of all other regions must be
+// reserved in the young generation.
+size_t ShenandoahGenerationalHeuristics::select_aged_regions(const size_t old_promotion_reserve) {
+
+ // There should be no regions configured for subsequent in-place-promotions carried over from the previous cycle.
+ assert_no_in_place_promotions();
+
+ auto const heap = ShenandoahGenerationalHeap::heap();
+ ShenandoahFreeSet* free_set = heap->free_set();
+ bool* const candidate_regions_for_promotion_by_copy = heap->collection_set()->preselected_regions();
+ ShenandoahMarkingContext* const ctx = heap->marking_context();
+
+ const size_t old_garbage_threshold =
+ (ShenandoahHeapRegion::region_size_bytes() * heap->old_generation()->heuristics()->get_old_garbage_threshold()) / 100;
+
+ const size_t pip_used_threshold = (ShenandoahHeapRegion::region_size_bytes() * ShenandoahGenerationalMinPIPUsage) / 100;
+
+ size_t promo_potential = 0;
+ size_t candidates = 0;
+
+ // Tracks the padding of space above top in regions eligible for promotion in place
+ size_t promote_in_place_pad = 0;
+
+ // Sort the promotion-eligible regions in order of increasing live-data-bytes so that we can first reclaim regions that require
+ // less evacuation effort. This prioritizes garbage first, expanding the allocation pool early before we reclaim regions that
+ // have more live data.
+ const idx_t num_regions = heap->num_regions();
+
+ ResourceMark rm;
+ AgedRegionData* sorted_regions = NEW_RESOURCE_ARRAY(AgedRegionData, num_regions);
+
+ ShenandoahFreeSet* freeset = heap->free_set();
+
+ // Any region that is to be promoted in place needs to be retired from its Collector or Mutator partition.
+ idx_t pip_low_collector_idx = freeset->max_regions();
+ idx_t pip_high_collector_idx = -1;
+ idx_t pip_low_mutator_idx = freeset->max_regions();
+ idx_t pip_high_mutator_idx = -1;
+ size_t collector_regions_to_pip = 0;
+ size_t mutator_regions_to_pip = 0;
+
+ size_t pip_mutator_regions = 0;
+ size_t pip_collector_regions = 0;
+ size_t pip_mutator_bytes = 0;
+ size_t pip_collector_bytes = 0;
+
+ for (idx_t i = 0; i < num_regions; i++) {
+ ShenandoahHeapRegion* const r = heap->get_region(i);
+ if (r->is_empty() || !r->has_live() || !r->is_young() || !r->is_regular()) {
+ // skip over regions that aren't regular young with some live data
+ continue;
+ }
+ if (heap->is_tenurable(r)) {
+ if ((r->garbage() < old_garbage_threshold) && (r->used() > pip_used_threshold)) {
+ // We prefer to promote this region in place because it has a small amount of garbage and a large usage.
+ HeapWord* tams = ctx->top_at_mark_start(r);
+ HeapWord* original_top = r->top();
+ if (!heap->is_concurrent_old_mark_in_progress() && tams == original_top) {
+ // No allocations from this region have been made during concurrent mark. It meets all the criteria
+ // for in-place-promotion. Though we only need the value of top when we fill the end of the region,
+ // we use this field to indicate that this region should be promoted in place during the evacuation
+ // phase.
+ r->save_top_before_promote();
+ size_t remnant_bytes = r->free();
+ size_t remnant_words = remnant_bytes / HeapWordSize;
+ assert(ShenandoahHeap::min_fill_size() <= PLAB::min_size(), "Implementation makes invalid assumptions");
+ if (remnant_words >= ShenandoahHeap::min_fill_size()) {
+ ShenandoahHeap::fill_with_object(original_top, remnant_words);
+ // Fill the remnant memory within this region to assure no allocations prior to promote in place. Otherwise,
+ // newly allocated objects will not be parsable when promote in place tries to register them. Furthermore, any
+ // new allocations would not necessarily be eligible for promotion. This addresses both issues.
+ r->set_top(r->end());
+ // The region r is either in the Mutator or Collector partition if remnant_words > heap()->plab_min_size.
+ // Otherwise, the region is in the NotFree partition.
+ ShenandoahFreeSetPartitionId p = free_set->membership(i);
+ if (p == ShenandoahFreeSetPartitionId::Mutator) {
+ mutator_regions_to_pip++;
+ if (i < pip_low_mutator_idx) {
+ pip_low_mutator_idx = i;
+ }
+ if (i > pip_high_mutator_idx) {
+ pip_high_mutator_idx = i;
+ }
+ pip_mutator_regions++;
+ pip_mutator_bytes += remnant_bytes;
+ } else if (p == ShenandoahFreeSetPartitionId::Collector) {
+ collector_regions_to_pip++;
+ if (i < pip_low_collector_idx) {
+ pip_low_collector_idx = i;
+ }
+ if (i > pip_high_collector_idx) {
+ pip_high_collector_idx = i;
+ }
+ pip_collector_regions++;
+ pip_collector_bytes += remnant_bytes;
+ } else {
+ assert((p == ShenandoahFreeSetPartitionId::NotFree) && (remnant_words < heap->plab_min_size()),
+ "Should be NotFree if not in Collector or Mutator partitions");
+ // In this case, the memory is already counted as used and the region has already been retired. There is
+ // no need for further adjustments to used. Further, the remnant memory for this region will not be
+ // unallocated or made available to OldCollector after pip.
+ remnant_bytes = 0;
+ }
+ promote_in_place_pad += remnant_bytes;
+ free_set->prepare_to_promote_in_place(i, remnant_bytes);
+ } else {
+ // Since the remnant is so small that this region has already been retired, we don't have to worry about any
+ // accidental allocations occurring within this region before the region is promoted in place.
+
+ // This region was already not in the Collector or Mutator set, so no need to remove it.
+ assert(free_set->membership(i) == ShenandoahFreeSetPartitionId::NotFree, "sanity");
+ }
+ }
+ // Else, we do not promote this region (either in place or by copy) because it has received new allocations.
+
+ // During evacuation, we exclude from promotion regions for which age > tenure threshold, garbage < garbage-threshold,
+ // used > pip_used_threshold, and get_top_before_promote() != tams
+ } else {
+ // Record this promotion-eligible candidate region. After sorting and selecting the best candidates below,
+ // we may still decide to exclude this promotion-eligible region from the current collection set. If this
+ // happens, we will consider this region as part of the anticipated promotion potential for the next GC
+ // pass; see further below.
+ sorted_regions[candidates]._region = r;
+ sorted_regions[candidates++]._live_data = r->get_live_data_bytes();
+ }
+ } else {
+ // We only evacuate & promote objects from regular regions whose garbage() is above old-garbage-threshold.
+ // Objects in tenure-worthy regions with less garbage are promoted in place. These take a different path to
+ // old-gen. Regions excluded from promotion because their garbage content is too low (causing us to anticipate that
+ // the region would be promoted in place) may be eligible for evacuation promotion by the time promotion takes
+ // place during a subsequent GC pass because more garbage is found within the region between now and then. This
+ // should not happen if we are properly adapting the tenure age. The theory behind adaptive tenuring threshold
+ // is to choose the youngest age that demonstrates no "significant" further loss of population since the previous
+ // age. If not this, we expect the tenure age to demonstrate linear population decay for at least two population
+ // samples, whereas we expect to observe exponential population decay for ages younger than the tenure age.
+ //
+ // In the case that certain regions which were anticipated to be promoted in place need to be promoted by
+ // evacuation, it may be the case that there is not sufficient reserve within old-gen to hold evacuation of
+ // these regions. The likely outcome is that these regions will not be selected for evacuation or promotion
+ // in the current cycle and we will anticipate that they will be promoted in the next cycle. This will cause
+ // us to reserve more old-gen memory so that these objects can be promoted in the subsequent cycle.
+ if (heap->is_aging_cycle() && heap->age_census()->is_tenurable(r->age() + 1)) {
+ if (r->garbage() >= old_garbage_threshold) {
+ promo_potential += r->get_live_data_bytes();
+ }
+ }
+ }
+ // Note that we keep going even if one region is excluded from selection.
+ // Subsequent regions may be selected if they have smaller live data.
+ }
+
+ if (pip_mutator_regions + pip_collector_regions > 0) {
+ freeset->account_for_pip_regions(pip_mutator_regions, pip_mutator_bytes, pip_collector_regions, pip_collector_bytes);
+ }
+
+ // Retire any regions that have been selected for promote in place
+ if (collector_regions_to_pip > 0) {
+ freeset->shrink_interval_if_range_modifies_either_boundary(ShenandoahFreeSetPartitionId::Collector,
+ pip_low_collector_idx, pip_high_collector_idx,
+ collector_regions_to_pip);
+ }
+ if (mutator_regions_to_pip > 0) {
+ freeset->shrink_interval_if_range_modifies_either_boundary(ShenandoahFreeSetPartitionId::Mutator,
+ pip_low_mutator_idx, pip_high_mutator_idx,
+ mutator_regions_to_pip);
+ }
+
+ // Sort in increasing order according to live data bytes. Note that candidates represents the number of regions
+ // that qualify to be promoted by evacuation.
+ size_t old_consumed = 0;
+ if (candidates > 0) {
+ size_t selected_regions = 0;
+ size_t selected_live = 0;
+ QuickSort::sort(sorted_regions, candidates, compare_by_aged_live);
+ for (size_t i = 0; i < candidates; i++) {
+ ShenandoahHeapRegion* const region = sorted_regions[i]._region;
+ const size_t region_live_data = sorted_regions[i]._live_data;
+ const size_t promotion_need = (size_t) (region_live_data * ShenandoahPromoEvacWaste);
+ if (old_consumed + promotion_need <= old_promotion_reserve) {
+ old_consumed += promotion_need;
+ candidate_regions_for_promotion_by_copy[region->index()] = true;
+ selected_regions++;
+ selected_live += region_live_data;
+ } else {
+ // We rejected this promotable region from the collection set because we had no room to hold its copy.
+ // Add this region to promo potential for next GC.
+ promo_potential += region_live_data;
+ assert(!candidate_regions_for_promotion_by_copy[region->index()], "Shouldn't be selected");
+ }
+ // We keep going even if one region is excluded from selection because we need to accumulate all eligible
+ // regions that are not preselected into promo_potential
+ }
+ log_debug(gc, ergo)("Preselected %zu regions containing " PROPERFMT " live data,"
+ " consuming: " PROPERFMT " of budgeted: " PROPERFMT,
+ selected_regions, PROPERFMTARGS(selected_live), PROPERFMTARGS(old_consumed), PROPERFMTARGS(old_promotion_reserve));
+ }
+
+ log_info(gc, ergo)("Promotion potential of aged regions with sufficient garbage: " PROPERFMT, PROPERFMTARGS(promo_potential));
+
+ heap->old_generation()->set_pad_for_promote_in_place(promote_in_place_pad);
+ heap->old_generation()->set_promotion_potential(promo_potential);
+ return old_consumed;
+}
+
+// Having chosen the collection set, adjust the budgets for generational mode based on its composition. Note
+// that young_generation->available() now knows about recently discovered immediate garbage.
+void ShenandoahGenerationalHeuristics::adjust_evacuation_budgets(ShenandoahHeap* const heap,
+ ShenandoahCollectionSet* const collection_set) {
+ shenandoah_assert_generational();
+ // We may find that old_evacuation_reserve and/or loaned_for_young_evacuation are not fully consumed, in which case we may
+ // be able to increase regions_available_to_loan
+
+ // The role of adjust_evacuation_budgets() is to compute the correct value of regions_available_to_loan and to make
+ // effective use of this memory, including the remnant memory within these regions that may result from rounding loan to
+ // integral number of regions. Excess memory that is available to be loaned is applied to an allocation supplement,
+ // which allows mutators to allocate memory beyond the current capacity of young-gen on the promise that the loan
+ // will be repaid as soon as we finish updating references for the recently evacuated collection set.
+
+ // We cannot recalculate regions_available_to_loan by simply dividing old_generation->available() by region_size_bytes
+ // because the available memory may be distributed between many partially occupied regions that are already holding old-gen
+ // objects. Memory in partially occupied regions is not "available" to be loaned. Note that an increase in old-gen
+ // available that results from a decrease in memory consumed by old evacuation is not necessarily available to be loaned
+ // to young-gen.
+
+ const size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
+ ShenandoahOldGeneration* const old_generation = heap->old_generation();
+ ShenandoahYoungGeneration* const young_generation = heap->young_generation();
+
+ const size_t old_evacuated = collection_set->get_live_bytes_in_old_regions();
+ size_t old_evacuated_committed = (size_t) (ShenandoahOldEvacWaste * double(old_evacuated));
+ size_t old_evacuation_reserve = old_generation->get_evacuation_reserve();
+
+ if (old_evacuated_committed > old_evacuation_reserve) {
+ // This should only happen due to round-off errors when enforcing ShenandoahOldEvacWaste
+ assert(old_evacuated_committed <= (33 * old_evacuation_reserve) / 32,
+ "Round-off errors should be less than 3.125%%, committed: %zu, reserved: %zu",
+ old_evacuated_committed, old_evacuation_reserve);
+ old_evacuated_committed = old_evacuation_reserve;
+ // Leave old_evac_reserve as previously configured
+ } else if (old_evacuated_committed < old_evacuation_reserve) {
+ // This happens if the old-gen collection consumes less than full budget.
+ log_debug(gc, cset)("Shrinking old evac reserve to match old_evac_commited: " PROPERFMT,
+ PROPERFMTARGS(old_evacuated_committed));
+ old_evacuation_reserve = old_evacuated_committed;
+ old_generation->set_evacuation_reserve(old_evacuation_reserve);
+ }
+
+ size_t young_advance_promoted = collection_set->get_live_bytes_in_tenurable_regions();
+ size_t young_advance_promoted_reserve_used = (size_t) (ShenandoahPromoEvacWaste * double(young_advance_promoted));
+
+ size_t young_evacuated = collection_set->get_live_bytes_in_untenurable_regions();
+ size_t young_evacuated_reserve_used = (size_t) (ShenandoahEvacWaste * double(young_evacuated));
+
+ size_t total_young_available = young_generation->available_with_reserve() - _add_regions_to_old * region_size_bytes;;
+ assert(young_evacuated_reserve_used <= total_young_available, "Cannot evacuate (%zu) more than is available in young (%zu)",
+ young_evacuated_reserve_used, total_young_available);
+ young_generation->set_evacuation_reserve(young_evacuated_reserve_used);
+
+ // We have not yet rebuilt the free set. Some of the memory that is thought to be avaiable within old may no
+ // longer be available if that memory had been free within regions that were selected for the collection set.
+ // Make the necessary adjustments to old_available.
+ size_t old_available =
+ old_generation->available() + _add_regions_to_old * region_size_bytes - collection_set->get_old_available_bytes_collected();
+
+ // Now that we've established the collection set, we know how much memory is really required by old-gen for evacuation
+ // and promotion reserves. Try shrinking OLD now in case that gives us a bit more runway for mutator allocations during
+ // evac and update phases.
+ size_t old_consumed = old_evacuated_committed + young_advance_promoted_reserve_used;
+
+ if (old_available < old_consumed) {
+ // This can happen due to round-off errors when adding the results of truncated integer arithmetic.
+ // We've already truncated old_evacuated_committed. Truncate young_advance_promoted_reserve_used here.
+
+ assert(young_advance_promoted_reserve_used <= (33 * (old_available - old_evacuated_committed)) / 32,
+ "Round-off errors should be less than 3.125%%, committed: %zu, reserved: %zu",
+ young_advance_promoted_reserve_used, old_available - old_evacuated_committed);
+ if (old_available > old_evacuated_committed) {
+ young_advance_promoted_reserve_used = old_available - old_evacuated_committed;
+ } else {
+ young_advance_promoted_reserve_used = 0;
+ old_evacuated_committed = old_available;
+ }
+ // TODO: reserve for full promotion reserve, not just for advance (preselected) promotion
+ old_consumed = old_evacuated_committed + young_advance_promoted_reserve_used;
+ }
+
+ assert(old_available >= old_consumed, "Cannot consume (%zu) more than is available (%zu)",
+ old_consumed, old_available);
+ size_t excess_old = old_available - old_consumed;
+ size_t unaffiliated_old_regions = old_generation->free_unaffiliated_regions() + _add_regions_to_old;
+ size_t unaffiliated_old = unaffiliated_old_regions * region_size_bytes;
+ assert(unaffiliated_old >= old_evacuated_committed, "Do not evacuate (%zu) more than unaffiliated old (%zu)",
+ old_evacuated_committed, unaffiliated_old);
+
+ // Make sure old_evac_committed is unaffiliated
+ if (old_evacuated_committed > 0) {
+ if (unaffiliated_old > old_evacuated_committed) {
+ size_t giveaway = unaffiliated_old - old_evacuated_committed;
+ size_t giveaway_regions = giveaway / region_size_bytes; // round down
+ if (giveaway_regions > 0) {
+ excess_old = MIN2(excess_old, giveaway_regions * region_size_bytes);
+ } else {
+ excess_old = 0;
+ }
+ } else {
+ excess_old = 0;
+ }
+ }
+
+ // If we find that OLD has excess regions, give them back to YOUNG now to reduce likelihood we run out of allocation
+ // runway during evacuation and update-refs. We may make further adjustments to balance.
+ ssize_t add_regions_to_young = 0;
+ if (excess_old > unaffiliated_old) {
+ // we can give back unaffiliated_old (all of unaffiliated is excess)
+ if (unaffiliated_old_regions > 0) {
+ add_regions_to_young = unaffiliated_old_regions;
+ }
+ } else if (unaffiliated_old_regions > 0) {
+ // excess_old < unaffiliated old: we can give back MIN(excess_old/region_size_bytes, unaffiliated_old_regions)
+ size_t excess_regions = excess_old / region_size_bytes;
+ add_regions_to_young = MIN2(excess_regions, unaffiliated_old_regions);
+ }
+
+ if (add_regions_to_young > 0) {
+ assert(excess_old >= add_regions_to_young * region_size_bytes, "Cannot xfer more than excess old");
+ excess_old -= add_regions_to_young * region_size_bytes;
+ log_debug(gc, ergo)("Before start of evacuation, total_promotion reserve is young_advance_promoted_reserve: %zu "
+ "plus excess: old: %zu", young_advance_promoted_reserve_used, excess_old);
+ }
+
+ // Add in the excess_old memory to hold unanticipated promotions, if any. If there are more unanticipated
+ // promotions than fit in reserved memory, they will be deferred until a future GC pass.
+ size_t total_promotion_reserve = young_advance_promoted_reserve_used + excess_old;
+
+ old_generation->set_promoted_reserve(total_promotion_reserve);
+ old_generation->reset_promoted_expended();
+}
size_t ShenandoahGenerationalHeuristics::add_preselected_regions_to_collection_set(ShenandoahCollectionSet* cset,
const RegionData* data,
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahGenerationalHeuristics.hpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahGenerationalHeuristics.hpp
index 9b4c93af9b4..74d657feab7 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahGenerationalHeuristics.hpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahGenerationalHeuristics.hpp
@@ -29,6 +29,9 @@
#include "gc/shenandoah/heuristics/shenandoahAdaptiveHeuristics.hpp"
class ShenandoahGeneration;
+class ShenandoahHeap;
+class ShenandoahCollectionSet;
+class RegionData;
/*
* This class serves as the base class for heuristics used to trigger and
@@ -44,10 +47,42 @@ class ShenandoahGenerationalHeuristics : public ShenandoahAdaptiveHeuristics {
public:
explicit ShenandoahGenerationalHeuristics(ShenandoahGeneration* generation);
- size_t choose_collection_set(ShenandoahCollectionSet* collection_set) override;
+ void choose_collection_set(ShenandoahCollectionSet* collection_set) override;
+
+private:
+ // Compute evacuation budgets prior to choosing collection set.
+ void compute_evacuation_budgets(ShenandoahHeap* const heap);
+
+ // Preselect for possible inclusion into the collection set exactly the most
+ // garbage-dense regions, including those that satisfy criteria 1 & 2 below,
+ // and whose live bytes will fit within old_available budget:
+ // Criterion 1. region age >= tenuring threshold
+ // Criterion 2. region garbage percentage > old garbage threshold
+ //
+ // Identifies regions eligible for promotion in place,
+ // being those of at least tenuring_threshold age that have lower garbage
+ // density.
+ //
+ // Updates promotion_potential and pad_for_promote_in_place fields
+ // of the heap. Returns bytes of live object memory in the preselected
+ // regions, which are marked in the preselected_regions() indicator
+ // array of the heap's collection set, which should be initialized
+ // to false.
+ size_t select_aged_regions(const size_t old_promotion_reserve);
+
+ // Filter and sort remaining regions before adding to collection set.
+ void filter_regions(ShenandoahCollectionSet* collection_set);
+
+ // Adjust evacuation budgets after choosing collection set. The argument regions_to_xfer
+ // represents regions to be transferred to old based on decisions made in top_off_collection_set()
+ void adjust_evacuation_budgets(ShenandoahHeap* const heap,
+ ShenandoahCollectionSet* const collection_set);
+
protected:
ShenandoahGeneration* _generation;
+ size_t _add_regions_to_old;
+
size_t add_preselected_regions_to_collection_set(ShenandoahCollectionSet* cset,
const RegionData* data,
size_t size) const;
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahGlobalHeuristics.cpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahGlobalHeuristics.cpp
index f47371c14d5..dd2ad28aa4b 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahGlobalHeuristics.cpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahGlobalHeuristics.cpp
@@ -36,14 +36,13 @@ ShenandoahGlobalHeuristics::ShenandoahGlobalHeuristics(ShenandoahGlobalGeneratio
}
-size_t ShenandoahGlobalHeuristics::choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
- RegionData* data, size_t size,
- size_t actual_free) {
+void ShenandoahGlobalHeuristics::choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
+ RegionData* data, size_t size,
+ size_t actual_free) {
// Better select garbage-first regions
- QuickSort::sort(data, (int) size, compare_by_garbage);
+ QuickSort::sort(data, size, compare_by_garbage);
choose_global_collection_set(cset, data, size, actual_free, 0 /* cur_young_garbage */);
- return 0;
}
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahGlobalHeuristics.hpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahGlobalHeuristics.hpp
index e0513f60da9..1f95f75c521 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahGlobalHeuristics.hpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahGlobalHeuristics.hpp
@@ -39,9 +39,9 @@ class ShenandoahGlobalHeuristics : public ShenandoahGenerationalHeuristics {
public:
ShenandoahGlobalHeuristics(ShenandoahGlobalGeneration* generation);
- size_t choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
- RegionData* data, size_t size,
- size_t actual_free) override;
+ void choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
+ RegionData* data, size_t size,
+ size_t actual_free) override;
private:
void choose_global_collection_set(ShenandoahCollectionSet* cset,
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahHeuristics.cpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahHeuristics.cpp
index aeb64b6f1df..8fc744112bf 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahHeuristics.cpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahHeuristics.cpp
@@ -72,7 +72,7 @@ ShenandoahHeuristics::~ShenandoahHeuristics() {
FREE_C_HEAP_ARRAY(RegionGarbage, _region_data);
}
-size_t ShenandoahHeuristics::choose_collection_set(ShenandoahCollectionSet* collection_set) {
+void ShenandoahHeuristics::choose_collection_set(ShenandoahCollectionSet* collection_set) {
ShenandoahHeap* heap = ShenandoahHeap::heap();
assert(collection_set->is_empty(), "Must be empty");
@@ -154,7 +154,6 @@ size_t ShenandoahHeuristics::choose_collection_set(ShenandoahCollectionSet* coll
choose_collection_set_from_regiondata(collection_set, candidates, cand_idx, immediate_garbage + free);
}
collection_set->summarize(total_garbage, immediate_garbage, immediate_regions);
- return 0;
}
void ShenandoahHeuristics::record_cycle_start() {
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahHeuristics.hpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahHeuristics.hpp
index ae34a9743a9..633c4e87126 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahHeuristics.hpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahHeuristics.hpp
@@ -183,12 +183,10 @@ protected:
static int compare_by_garbage(RegionData a, RegionData b);
- // This is a helper function to choose_collection_set(), returning the number of regions that need to be transferred to
- // the old reserve from the young reserve in order to effectively evacuate the chosen collection set. In non-generational
- // mode, the return value is 0.
- virtual size_t choose_collection_set_from_regiondata(ShenandoahCollectionSet* set,
- RegionData* data, size_t data_size,
- size_t free) = 0;
+ // This is a helper function to choose_collection_set()
+ virtual void choose_collection_set_from_regiondata(ShenandoahCollectionSet* set,
+ RegionData* data, size_t data_size,
+ size_t free) = 0;
void adjust_penalty(intx step);
@@ -238,7 +236,7 @@ public:
// Choose the collection set, returning the number of regions that need to be transferred to the old reserve from the young
// reserve in order to effectively evacuate the chosen collection set. In non-generational mode, the return value is 0.
- virtual size_t choose_collection_set(ShenandoahCollectionSet* collection_set);
+ virtual void choose_collection_set(ShenandoahCollectionSet* collection_set);
virtual bool can_unload_classes();
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahOldHeuristics.cpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahOldHeuristics.cpp
index f47d0cbe819..e0cab781674 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahOldHeuristics.cpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahOldHeuristics.cpp
@@ -884,9 +884,8 @@ bool ShenandoahOldHeuristics::is_experimental() {
return true;
}
-size_t ShenandoahOldHeuristics::choose_collection_set_from_regiondata(ShenandoahCollectionSet* set,
- ShenandoahHeuristics::RegionData* data,
- size_t data_size, size_t free) {
+void ShenandoahOldHeuristics::choose_collection_set_from_regiondata(ShenandoahCollectionSet* set,
+ ShenandoahHeuristics::RegionData* data,
+ size_t data_size, size_t free) {
ShouldNotReachHere();
- return 0;
}
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahOldHeuristics.hpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahOldHeuristics.hpp
index 97a5b1ebf24..fc7a35aa6c8 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahOldHeuristics.hpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahOldHeuristics.hpp
@@ -155,8 +155,8 @@ private:
void set_trigger_if_old_is_overgrown();
protected:
- size_t
- choose_collection_set_from_regiondata(ShenandoahCollectionSet* set, RegionData* data, size_t data_size, size_t free) override;
+ void choose_collection_set_from_regiondata(ShenandoahCollectionSet* set,
+ RegionData* data, size_t data_size, size_t free) override;
// This internal helper routine adds as many mixed evacuation candidate regions as fit within the old-gen evacuation budget
// to the collection set. This may be called twice to prepare for any given mixed evacuation cycle, the first time with
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahPassiveHeuristics.cpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahPassiveHeuristics.cpp
index d4a38278161..b5e9cc433ea 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahPassiveHeuristics.cpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahPassiveHeuristics.cpp
@@ -50,9 +50,9 @@ bool ShenandoahPassiveHeuristics::should_degenerate_cycle() {
return ShenandoahDegeneratedGC;
}
-size_t ShenandoahPassiveHeuristics::choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
- RegionData* data, size_t size,
- size_t actual_free) {
+void ShenandoahPassiveHeuristics::choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
+ RegionData* data, size_t size,
+ size_t actual_free) {
assert(ShenandoahDegeneratedGC, "This path is only taken for Degenerated GC");
// Do not select too large CSet that would overflow the available free space.
@@ -76,5 +76,4 @@ size_t ShenandoahPassiveHeuristics::choose_collection_set_from_regiondata(Shenan
cset->add_region(r);
}
}
- return 0;
}
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahPassiveHeuristics.hpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahPassiveHeuristics.hpp
index 7a64fad7cc9..3cb85f5d05f 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahPassiveHeuristics.hpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahPassiveHeuristics.hpp
@@ -40,19 +40,19 @@ class ShenandoahPassiveHeuristics : public ShenandoahHeuristics {
public:
ShenandoahPassiveHeuristics(ShenandoahSpaceInfo* space_info);
- virtual bool should_start_gc();
+ bool should_start_gc() override;
- virtual bool should_unload_classes();
+ bool should_unload_classes() override;
- virtual bool should_degenerate_cycle();
+ bool should_degenerate_cycle() override;
- virtual size_t choose_collection_set_from_regiondata(ShenandoahCollectionSet* set,
- RegionData* data, size_t data_size,
- size_t free);
+ void choose_collection_set_from_regiondata(ShenandoahCollectionSet* set,
+ RegionData* data, size_t data_size,
+ size_t free) override;
- virtual const char* name() { return "Passive"; }
- virtual bool is_diagnostic() { return true; }
- virtual bool is_experimental() { return false; }
+ const char* name() override { return "Passive"; }
+ bool is_diagnostic() override { return true; }
+ bool is_experimental() override { return false; }
};
#endif // SHARE_GC_SHENANDOAH_HEURISTICS_SHENANDOAHPASSIVEHEURISTICS_HPP
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahStaticHeuristics.cpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahStaticHeuristics.cpp
index 3843e434781..5f384f3dc73 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahStaticHeuristics.cpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahStaticHeuristics.cpp
@@ -37,8 +37,6 @@ ShenandoahStaticHeuristics::ShenandoahStaticHeuristics(ShenandoahSpaceInfo* spac
SHENANDOAH_ERGO_ENABLE_FLAG(ShenandoahImplicitGCInvokesConcurrent);
}
-ShenandoahStaticHeuristics::~ShenandoahStaticHeuristics() {}
-
bool ShenandoahStaticHeuristics::should_start_gc() {
size_t capacity = ShenandoahHeap::heap()->soft_max_capacity();
size_t available = _space_info->soft_mutator_available();
@@ -59,9 +57,9 @@ bool ShenandoahStaticHeuristics::should_start_gc() {
return ShenandoahHeuristics::should_start_gc();
}
-size_t ShenandoahStaticHeuristics::choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
- RegionData* data, size_t size,
- size_t free) {
+void ShenandoahStaticHeuristics::choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
+ RegionData* data, size_t size,
+ size_t free) {
size_t threshold = ShenandoahHeapRegion::region_size_bytes() * ShenandoahGarbageThreshold / 100;
for (size_t idx = 0; idx < size; idx++) {
@@ -70,5 +68,4 @@ size_t ShenandoahStaticHeuristics::choose_collection_set_from_regiondata(Shenand
cset->add_region(r);
}
}
- return 0;
}
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahStaticHeuristics.hpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahStaticHeuristics.hpp
index 27dc3c8e0ae..b1514b55e5a 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahStaticHeuristics.hpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahStaticHeuristics.hpp
@@ -34,19 +34,17 @@
*/
class ShenandoahStaticHeuristics : public ShenandoahHeuristics {
public:
- ShenandoahStaticHeuristics(ShenandoahSpaceInfo* space_info);
+ explicit ShenandoahStaticHeuristics(ShenandoahSpaceInfo* space_info);
- virtual ~ShenandoahStaticHeuristics();
+ bool should_start_gc() override;
- virtual bool should_start_gc();
+ void choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
+ RegionData* data, size_t size,
+ size_t free) override;
- virtual size_t choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
- RegionData* data, size_t size,
- size_t free);
-
- virtual const char* name() { return "Static"; }
- virtual bool is_diagnostic() { return false; }
- virtual bool is_experimental() { return false; }
+ const char* name() override { return "Static"; }
+ bool is_diagnostic() override { return false; }
+ bool is_experimental() override { return false; }
};
#endif // SHARE_GC_SHENANDOAH_HEURISTICS_SHENANDOAHSTATICHEURISTICS_HPP
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahYoungHeuristics.cpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahYoungHeuristics.cpp
index 01c3873df72..beff2200d90 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahYoungHeuristics.cpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahYoungHeuristics.cpp
@@ -37,7 +37,7 @@ ShenandoahYoungHeuristics::ShenandoahYoungHeuristics(ShenandoahYoungGeneration*
}
-size_t ShenandoahYoungHeuristics::choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
+void ShenandoahYoungHeuristics::choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
RegionData* data, size_t size,
size_t actual_free) {
// See comments in ShenandoahAdaptiveHeuristics::choose_collection_set_from_regiondata():
@@ -52,7 +52,7 @@ size_t ShenandoahYoungHeuristics::choose_collection_set_from_regiondata(Shenando
bool need_to_finalize_mixed = heap->old_generation()->heuristics()->prime_collection_set(cset);
// Better select garbage-first regions
- QuickSort::sort(data, (int) size, compare_by_garbage);
+ QuickSort::sort(data, size, compare_by_garbage);
size_t cur_young_garbage = add_preselected_regions_to_collection_set(cset, data, size);
@@ -62,12 +62,10 @@ size_t ShenandoahYoungHeuristics::choose_collection_set_from_regiondata(Shenando
// enough consolidated garbage to make effective use of young-gen evacuation reserve. If there is still
// young-gen reserve available following selection of the young-gen collection set, see if we can use
// this memory to expand the old-gen evacuation collection set.
- size_t add_regions_to_old;
- need_to_finalize_mixed |= heap->old_generation()->heuristics()->top_off_collection_set(add_regions_to_old);
+ need_to_finalize_mixed |= heap->old_generation()->heuristics()->top_off_collection_set(_add_regions_to_old);
if (need_to_finalize_mixed) {
heap->old_generation()->heuristics()->finalize_mixed_evacs();
}
- return add_regions_to_old;
}
void ShenandoahYoungHeuristics::choose_young_collection_set(ShenandoahCollectionSet* cset,
diff --git a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahYoungHeuristics.hpp b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahYoungHeuristics.hpp
index 85587887663..b9d64059680 100644
--- a/src/hotspot/share/gc/shenandoah/heuristics/shenandoahYoungHeuristics.hpp
+++ b/src/hotspot/share/gc/shenandoah/heuristics/shenandoahYoungHeuristics.hpp
@@ -38,9 +38,9 @@ public:
explicit ShenandoahYoungHeuristics(ShenandoahYoungGeneration* generation);
- size_t choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
- RegionData* data, size_t size,
- size_t actual_free) override;
+ void choose_collection_set_from_regiondata(ShenandoahCollectionSet* cset,
+ RegionData* data, size_t size,
+ size_t actual_free) override;
bool should_start_gc() override;
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahBarrierSet.cpp b/src/hotspot/share/gc/shenandoah/shenandoahBarrierSet.cpp
index 2aa37d7c575..004558a9fa8 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahBarrierSet.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahBarrierSet.cpp
@@ -72,21 +72,33 @@ void ShenandoahBarrierSet::print_on(outputStream* st) const {
bool ShenandoahBarrierSet::need_load_reference_barrier(DecoratorSet decorators, BasicType type) {
if (!ShenandoahLoadRefBarrier) return false;
- // Only needed for references
return is_reference_type(type);
}
bool ShenandoahBarrierSet::need_keep_alive_barrier(DecoratorSet decorators, BasicType type) {
if (!ShenandoahSATBBarrier) return false;
- // Only needed for references
if (!is_reference_type(type)) return false;
-
bool keep_alive = (decorators & AS_NO_KEEPALIVE) == 0;
bool unknown = (decorators & ON_UNKNOWN_OOP_REF) != 0;
bool on_weak_ref = (decorators & (ON_WEAK_OOP_REF | ON_PHANTOM_OOP_REF)) != 0;
return (on_weak_ref || unknown) && keep_alive;
}
+bool ShenandoahBarrierSet::need_satb_barrier(DecoratorSet decorators, BasicType type) {
+ if (!ShenandoahSATBBarrier) return false;
+ if (!is_reference_type(type)) return false;
+ bool as_normal = (decorators & AS_NORMAL) != 0;
+ bool dest_uninitialized = (decorators & IS_DEST_UNINITIALIZED) != 0;
+ return as_normal && !dest_uninitialized;
+}
+
+bool ShenandoahBarrierSet::need_card_barrier(DecoratorSet decorators, BasicType type) {
+ if (!ShenandoahCardBarrier) return false;
+ if (!is_reference_type(type)) return false;
+ bool in_heap = (decorators & IN_HEAP) != 0;
+ return in_heap;
+}
+
void ShenandoahBarrierSet::on_slowpath_allocation_exit(JavaThread* thread, oop new_obj) {
#if COMPILER2_OR_JVMCI
if (ReduceInitialCardMarks && ShenandoahCardBarrier && !ShenandoahHeap::heap()->is_in_young(new_obj)) {
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahBarrierSet.hpp b/src/hotspot/share/gc/shenandoah/shenandoahBarrierSet.hpp
index 28765605267..e7a0ed57740 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahBarrierSet.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahBarrierSet.hpp
@@ -60,6 +60,8 @@ public:
static bool need_load_reference_barrier(DecoratorSet decorators, BasicType type);
static bool need_keep_alive_barrier(DecoratorSet decorators, BasicType type);
+ static bool need_satb_barrier(DecoratorSet decorators, BasicType type);
+ static bool need_card_barrier(DecoratorSet decorators, BasicType type);
static bool is_strong_access(DecoratorSet decorators) {
return (decorators & (ON_WEAK_OOP_REF | ON_PHANTOM_OOP_REF)) == 0;
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahCollectorPolicy.hpp b/src/hotspot/share/gc/shenandoah/shenandoahCollectorPolicy.hpp
index 5fe90f64f98..1166333ae3a 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahCollectorPolicy.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahCollectorPolicy.hpp
@@ -63,17 +63,10 @@ private:
public:
// The most common scenario for lack of good progress following a degenerated GC is an accumulation of floating
- // garbage during the most recently aborted concurrent GC effort. With generational GC, it is far more effective to
+ // garbage during the most recently aborted concurrent GC effort. Usually, it is far more effective to
// reclaim this floating garbage with another degenerated cycle (which focuses on young generation and might require
- // a pause of 200 ms) rather than a full GC cycle (which may require over 2 seconds with a 10 GB old generation).
- //
- // In generational mode, we'll only upgrade to full GC if we've done two degen cycles in a row and both indicated
- // bad progress. In non-generational mode, we'll preserve the original behavior, which is to upgrade to full
- // immediately following a degenerated cycle with bad progress. This preserves original behavior of non-generational
- // Shenandoah to avoid introducing "surprising new behavior." It also makes less sense with non-generational
- // Shenandoah to replace a full GC with a degenerated GC, because both have similar pause times in non-generational
- // mode.
- static constexpr size_t GENERATIONAL_CONSECUTIVE_BAD_DEGEN_PROGRESS_THRESHOLD = 2;
+ // a pause of 200 ms) rather than a full GC cycle (which may require multiple seconds with a 10 GB old generation).
+ static constexpr size_t CONSECUTIVE_BAD_DEGEN_PROGRESS_THRESHOLD = 2;
ShenandoahCollectorPolicy();
@@ -117,9 +110,9 @@ public:
return _consecutive_degenerated_gcs;
}
- // Genshen will only upgrade to a full gc after the configured number of futile degenerated cycles.
- bool generational_should_upgrade_degenerated_gc() const {
- return _consecutive_degenerated_gcs_without_progress >= GENERATIONAL_CONSECUTIVE_BAD_DEGEN_PROGRESS_THRESHOLD;
+ // Only upgrade to a full gc after the configured number of futile degenerated cycles.
+ bool should_upgrade_degenerated_gc() const {
+ return _consecutive_degenerated_gcs_without_progress >= CONSECUTIVE_BAD_DEGEN_PROGRESS_THRESHOLD;
}
static bool is_allocation_failure(GCCause::Cause cause);
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahControlThread.cpp b/src/hotspot/share/gc/shenandoah/shenandoahControlThread.cpp
index 590e7831f07..bc11659c5e5 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahControlThread.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahControlThread.cpp
@@ -44,7 +44,7 @@ ShenandoahControlThread::ShenandoahControlThread() :
ShenandoahController(),
_requested_gc_cause(GCCause::_no_gc),
_degen_point(ShenandoahGC::_degenerated_outside_cycle),
- _control_lock(Mutex::nosafepoint - 2, "ShenandoahGCRequest_lock", true) {
+ _control_lock(CONTROL_LOCK_RANK, "ShenandoahControl_lock", true) {
set_name("Shenandoah Control Thread");
create_and_start();
}
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahController.hpp b/src/hotspot/share/gc/shenandoah/shenandoahController.hpp
index a6a699fac3b..b8ff4df4771 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahController.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahController.hpp
@@ -42,6 +42,9 @@ private:
shenandoah_padding(1);
protected:
+ const Mutex::Rank WAITERS_LOCK_RANK = Mutex::safepoint - 5;
+ const Mutex::Rank CONTROL_LOCK_RANK = Mutex::nosafepoint - 2;
+
// While we could have a single lock for these, it may risk unblocking
// GC waiters when alloc failure GC cycle finishes. We want instead
// to make complete explicit cycle for demanding customers.
@@ -54,8 +57,8 @@ protected:
public:
ShenandoahController():
_gc_id(0),
- _alloc_failure_waiters_lock(Mutex::safepoint-2, "ShenandoahAllocFailureGC_lock", true),
- _gc_waiters_lock(Mutex::safepoint-2, "ShenandoahRequestedGC_lock", true)
+ _alloc_failure_waiters_lock(WAITERS_LOCK_RANK, "ShenandoahAllocFailureWaiters_lock", true),
+ _gc_waiters_lock(WAITERS_LOCK_RANK, "ShenandoahGCWaiters_lock", true)
{ }
// Request a collection cycle. This handles "explicit" gc requests
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahDegeneratedGC.cpp b/src/hotspot/share/gc/shenandoah/shenandoahDegeneratedGC.cpp
index 333bdbc6e72..99776e38bfe 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahDegeneratedGC.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahDegeneratedGC.cpp
@@ -314,7 +314,7 @@ void ShenandoahDegenGC::op_degenerated() {
if (progress) {
heap->notify_gc_progress();
_generation->heuristics()->record_degenerated();
- } else if (!heap->mode()->is_generational() || policy->generational_should_upgrade_degenerated_gc()) {
+ } else if (policy->should_upgrade_degenerated_gc()) {
// Upgrade to full GC, register full-GC impact on heuristics.
op_degenerated_futile();
} else {
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahGeneration.cpp b/src/hotspot/share/gc/shenandoah/shenandoahGeneration.cpp
index cdc7e1a328a..ddb50ee0020 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahGeneration.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahGeneration.cpp
@@ -24,7 +24,6 @@
*/
#include "gc/shenandoah/heuristics/shenandoahHeuristics.hpp"
-#include "gc/shenandoah/shenandoahCollectionSetPreselector.hpp"
#include "gc/shenandoah/shenandoahCollectorPolicy.hpp"
#include "gc/shenandoah/shenandoahFreeSet.hpp"
#include "gc/shenandoah/shenandoahGeneration.hpp"
@@ -245,506 +244,6 @@ void ShenandoahGeneration::parallel_heap_region_iterate_free(ShenandoahHeapRegio
ShenandoahHeap::heap()->parallel_heap_region_iterate(cl);
}
-void ShenandoahGeneration::compute_evacuation_budgets(ShenandoahHeap* const heap) {
- shenandoah_assert_generational();
-
- ShenandoahOldGeneration* const old_generation = heap->old_generation();
- ShenandoahYoungGeneration* const young_generation = heap->young_generation();
- const size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
-
- // During initialization and phase changes, it is more likely that fewer objects die young and old-gen
- // memory is not yet full (or is in the process of being replaced). During these times especially, it
- // is beneficial to loan memory from old-gen to young-gen during the evacuation and update-refs phases
- // of execution.
-
- // Calculate EvacuationReserve before PromotionReserve. Evacuation is more critical than promotion.
- // If we cannot evacuate old-gen, we will not be able to reclaim old-gen memory. Promotions are less
- // critical. If we cannot promote, there may be degradation of young-gen memory because old objects
- // accumulate there until they can be promoted. This increases the young-gen marking and evacuation work.
-
- // First priority is to reclaim the easy garbage out of young-gen.
-
- // maximum_young_evacuation_reserve is upper bound on memory to be evacuated into young Collector Reserve. This is
- // bounded at the end of previous GC cycle, based on available memory and balancing of evacuation to old and young.
- size_t maximum_young_evacuation_reserve = young_generation->get_evacuation_reserve();
-
- // maximum_old_evacuation_reserve is an upper bound on memory evacuated from old and evacuated to old (promoted),
- // clamped by the old generation space available.
- //
- // Here's the algebra.
- // Let SOEP = ShenandoahOldEvacPercent,
- // OE = old evac,
- // YE = young evac, and
- // TE = total evac = OE + YE
- // By definition:
- // SOEP/100 = OE/TE
- // = OE/(OE+YE)
- // => SOEP/(100-SOEP) = OE/((OE+YE)-OE) // componendo-dividendo: If a/b = c/d, then a/(b-a) = c/(d-c)
- // = OE/YE
- // => OE = YE*SOEP/(100-SOEP)
-
- // We have to be careful in the event that SOEP is set to 100 by the user.
- assert(ShenandoahOldEvacPercent <= 100, "Error");
- const size_t old_available = old_generation->available();
- const size_t maximum_old_evacuation_reserve = (ShenandoahOldEvacPercent == 100) ?
- old_available : MIN2((maximum_young_evacuation_reserve * ShenandoahOldEvacPercent) / (100 - ShenandoahOldEvacPercent),
- old_available);
-
- // In some cases, maximum_old_reserve < old_available (when limited by ShenandoahOldEvacPercent)
- // This limit affects mixed evacuations, but does not affect promotions.
-
- // Second priority is to reclaim garbage out of old-gen if there are old-gen collection candidates. Third priority
- // is to promote as much as we have room to promote. However, if old-gen memory is in short supply, this means young
- // GC is operating under "duress" and was unable to transfer the memory that we would normally expect. In this case,
- // old-gen will refrain from compacting itself in order to allow a quicker young-gen cycle (by avoiding the update-refs
- // through ALL of old-gen). If there is some memory available in old-gen, we will use this for promotions as promotions
- // do not add to the update-refs burden of GC.
-
- size_t old_evacuation_reserve, old_promo_reserve;
- if (is_global()) {
- // Global GC is typically triggered by user invocation of System.gc(), and typically indicates that there is lots
- // of garbage to be reclaimed because we are starting a new phase of execution. Marking for global GC may take
- // significantly longer than typical young marking because we must mark through all old objects. To expedite
- // evacuation and update-refs, we give emphasis to reclaiming garbage first, wherever that garbage is found.
- // Global GC will adjust generation sizes to accommodate the collection set it chooses.
-
- // Use remnant of old_available to hold promotions.
- old_promo_reserve = old_available - maximum_old_evacuation_reserve;
-
- // Dedicate all available old memory to old_evacuation reserve. This may be small, because old-gen is only
- // expanded based on an existing mixed evacuation workload at the end of the previous GC cycle. We'll expand
- // the budget for evacuation of old during GLOBAL cset selection.
- old_evacuation_reserve = maximum_old_evacuation_reserve;
- } else if (old_generation->has_unprocessed_collection_candidates()) {
- // We reserved all old-gen memory at end of previous GC to hold anticipated evacuations to old-gen. If this is
- // mixed evacuation, reserve all of this memory for compaction of old-gen and do not promote. Prioritize compaction
- // over promotion in order to defragment OLD so that it will be better prepared to efficiently receive promoted memory.
- old_evacuation_reserve = maximum_old_evacuation_reserve;
- old_promo_reserve = old_available - maximum_old_evacuation_reserve;
- } else {
- // Make all old-evacuation memory for promotion, but if we can't use it all for promotion, we'll allow some evacuation.
- old_evacuation_reserve = old_available - maximum_old_evacuation_reserve;
- old_promo_reserve = maximum_old_evacuation_reserve;
- }
- assert(old_evacuation_reserve <= old_available, "Error");
-
-
- // We see too many old-evacuation failures if we force ourselves to evacuate into regions that are not initially empty.
- // So we limit the old-evacuation reserve to unfragmented memory. Even so, old-evacuation is free to fill in nooks and
- // crannies within existing partially used regions and it generally tries to do so.
- const size_t old_free_unfragmented = old_generation->free_unaffiliated_regions() * region_size_bytes;
- if (old_evacuation_reserve > old_free_unfragmented) {
- const size_t delta = old_evacuation_reserve - old_free_unfragmented;
- old_evacuation_reserve -= delta;
- // Let promo consume fragments of old-gen memory
- old_promo_reserve += delta;
- }
-
- // If is_global(), we let garbage-first heuristic determine cset membership. Otherwise, we give priority
- // to tenurable regions by preselecting regions for promotion by evacuation (obtaining the live data to seed promoted_reserve).
- // This also identifies regions that will be promoted in place. These use the tenuring threshold.
- const size_t consumed_by_advance_promotion = select_aged_regions(is_global()? 0: old_promo_reserve);
- assert(consumed_by_advance_promotion <= old_promo_reserve, "Do not promote more than budgeted");
-
- // The young evacuation reserve can be no larger than young_unaffiliated. Planning to evacuate into partially consumed
- // young regions is doomed to failure if any of those partially consumed regions is selected for the collection set.
- size_t young_unaffiliated = young_generation->free_unaffiliated_regions() * region_size_bytes;
-
- // If any regions have been selected for promotion in place, this has the effect of decreasing available within mutator
- // and collector partitions, due to padding of remnant memory within each promoted in place region. This will affect
- // young_evacuation_reserve but not old_evacuation_reserve or consumed_by_advance_promotion. So recompute.
- size_t young_evacuation_reserve = MIN2(maximum_young_evacuation_reserve, young_unaffiliated);
-
- // Note that unused old_promo_reserve might not be entirely consumed_by_advance_promotion. Do not transfer this
- // to old_evacuation_reserve because this memory is likely very fragmented, and we do not want to increase the likelihood
- // of old evacuation failure. Leave this memory in the promoted reserve as it may be targeted by opportunistic
- // promotions (found during evacuation of young regions).
- young_generation->set_evacuation_reserve(young_evacuation_reserve);
- old_generation->set_evacuation_reserve(old_evacuation_reserve);
- old_generation->set_promoted_reserve(old_promo_reserve);
-
- // There is no need to expand OLD because all memory used here was set aside at end of previous GC, except in the
- // case of a GLOBAL gc. During choose_collection_set() of GLOBAL, old will be expanded on demand.
-}
-
-// Having chosen the collection set, adjust the budgets for generational mode based on its composition. Note
-// that young_generation->available() now knows about recently discovered immediate garbage.
-void ShenandoahGeneration::adjust_evacuation_budgets(ShenandoahHeap* const heap,
- ShenandoahCollectionSet* const collection_set, size_t add_regions_to_old) {
- shenandoah_assert_generational();
- // We may find that old_evacuation_reserve and/or loaned_for_young_evacuation are not fully consumed, in which case we may
- // be able to increase regions_available_to_loan
-
- // The role of adjust_evacuation_budgets() is to compute the correct value of regions_available_to_loan and to make
- // effective use of this memory, including the remnant memory within these regions that may result from rounding loan to
- // integral number of regions. Excess memory that is available to be loaned is applied to an allocation supplement,
- // which allows mutators to allocate memory beyond the current capacity of young-gen on the promise that the loan
- // will be repaid as soon as we finish updating references for the recently evacuated collection set.
-
- // We cannot recalculate regions_available_to_loan by simply dividing old_generation->available() by region_size_bytes
- // because the available memory may be distributed between many partially occupied regions that are already holding old-gen
- // objects. Memory in partially occupied regions is not "available" to be loaned. Note that an increase in old-gen
- // available that results from a decrease in memory consumed by old evacuation is not necessarily available to be loaned
- // to young-gen.
-
- const size_t region_size_bytes = ShenandoahHeapRegion::region_size_bytes();
- ShenandoahOldGeneration* const old_generation = heap->old_generation();
- ShenandoahYoungGeneration* const young_generation = heap->young_generation();
-
- const size_t old_evacuated = collection_set->get_live_bytes_in_old_regions();
- size_t old_evacuated_committed = (size_t) (ShenandoahOldEvacWaste * double(old_evacuated));
- size_t old_evacuation_reserve = old_generation->get_evacuation_reserve();
-
- if (old_evacuated_committed > old_evacuation_reserve) {
- // This should only happen due to round-off errors when enforcing ShenandoahOldEvacWaste
- assert(old_evacuated_committed <= (33 * old_evacuation_reserve) / 32,
- "Round-off errors should be less than 3.125%%, committed: %zu, reserved: %zu",
- old_evacuated_committed, old_evacuation_reserve);
- old_evacuated_committed = old_evacuation_reserve;
- // Leave old_evac_reserve as previously configured
- } else if (old_evacuated_committed < old_evacuation_reserve) {
- // This happens if the old-gen collection consumes less than full budget.
- log_debug(gc, cset)("Shrinking old evac reserve to match old_evac_commited: " PROPERFMT,
- PROPERFMTARGS(old_evacuated_committed));
- old_evacuation_reserve = old_evacuated_committed;
- old_generation->set_evacuation_reserve(old_evacuation_reserve);
- }
-
- size_t young_advance_promoted = collection_set->get_live_bytes_in_tenurable_regions();
- size_t young_advance_promoted_reserve_used = (size_t) (ShenandoahPromoEvacWaste * double(young_advance_promoted));
-
- size_t young_evacuated = collection_set->get_live_bytes_in_untenurable_regions();
- size_t young_evacuated_reserve_used = (size_t) (ShenandoahEvacWaste * double(young_evacuated));
-
- size_t total_young_available = young_generation->available_with_reserve() - add_regions_to_old * region_size_bytes;;
- assert(young_evacuated_reserve_used <= total_young_available, "Cannot evacuate (%zu) more than is available in young (%zu)",
- young_evacuated_reserve_used, total_young_available);
- young_generation->set_evacuation_reserve(young_evacuated_reserve_used);
-
- // We have not yet rebuilt the free set. Some of the memory that is thought to be avaiable within old may no
- // longer be available if that memory had been free within regions that were selected for the collection set.
- // Make the necessary adjustments to old_available.
- size_t old_available =
- old_generation->available() + add_regions_to_old * region_size_bytes - collection_set->get_old_available_bytes_collected();
-
- // Now that we've established the collection set, we know how much memory is really required by old-gen for evacuation
- // and promotion reserves. Try shrinking OLD now in case that gives us a bit more runway for mutator allocations during
- // evac and update phases.
- size_t old_consumed = old_evacuated_committed + young_advance_promoted_reserve_used;
-
- if (old_available < old_consumed) {
- // This can happen due to round-off errors when adding the results of truncated integer arithmetic.
- // We've already truncated old_evacuated_committed. Truncate young_advance_promoted_reserve_used here.
-
- assert(young_advance_promoted_reserve_used <= (33 * (old_available - old_evacuated_committed)) / 32,
- "Round-off errors should be less than 3.125%%, committed: %zu, reserved: %zu",
- young_advance_promoted_reserve_used, old_available - old_evacuated_committed);
- if (old_available > old_evacuated_committed) {
- young_advance_promoted_reserve_used = old_available - old_evacuated_committed;
- } else {
- young_advance_promoted_reserve_used = 0;
- old_evacuated_committed = old_available;
- }
- // TODO: reserve for full promotion reserve, not just for advance (preselected) promotion
- old_consumed = old_evacuated_committed + young_advance_promoted_reserve_used;
- }
-
- assert(old_available >= old_consumed, "Cannot consume (%zu) more than is available (%zu)",
- old_consumed, old_available);
- size_t excess_old = old_available - old_consumed;
- size_t unaffiliated_old_regions = old_generation->free_unaffiliated_regions() + add_regions_to_old;
- size_t unaffiliated_old = unaffiliated_old_regions * region_size_bytes;
- assert(unaffiliated_old >= old_evacuated_committed, "Do not evacuate (%zu) more than unaffiliated old (%zu)",
- old_evacuated_committed, unaffiliated_old);
-
- // Make sure old_evac_committed is unaffiliated
- if (old_evacuated_committed > 0) {
- if (unaffiliated_old > old_evacuated_committed) {
- size_t giveaway = unaffiliated_old - old_evacuated_committed;
- size_t giveaway_regions = giveaway / region_size_bytes; // round down
- if (giveaway_regions > 0) {
- excess_old = MIN2(excess_old, giveaway_regions * region_size_bytes);
- } else {
- excess_old = 0;
- }
- } else {
- excess_old = 0;
- }
- }
-
- // If we find that OLD has excess regions, give them back to YOUNG now to reduce likelihood we run out of allocation
- // runway during evacuation and update-refs. We may make further adjustments to balance.
- ssize_t add_regions_to_young = 0;
- if (excess_old > unaffiliated_old) {
- // we can give back unaffiliated_old (all of unaffiliated is excess)
- if (unaffiliated_old_regions > 0) {
- add_regions_to_young = unaffiliated_old_regions;
- }
- } else if (unaffiliated_old_regions > 0) {
- // excess_old < unaffiliated old: we can give back MIN(excess_old/region_size_bytes, unaffiliated_old_regions)
- size_t excess_regions = excess_old / region_size_bytes;
- add_regions_to_young = MIN2(excess_regions, unaffiliated_old_regions);
- }
-
- if (add_regions_to_young > 0) {
- assert(excess_old >= add_regions_to_young * region_size_bytes, "Cannot xfer more than excess old");
- excess_old -= add_regions_to_young * region_size_bytes;
- log_debug(gc, ergo)("Before start of evacuation, total_promotion reserve is young_advance_promoted_reserve: %zu "
- "plus excess: old: %zu", young_advance_promoted_reserve_used, excess_old);
- }
-
- // Add in the excess_old memory to hold unanticipated promotions, if any. If there are more unanticipated
- // promotions than fit in reserved memory, they will be deferred until a future GC pass.
- size_t total_promotion_reserve = young_advance_promoted_reserve_used + excess_old;
-
- old_generation->set_promoted_reserve(total_promotion_reserve);
- old_generation->reset_promoted_expended();
-}
-
-typedef struct {
- ShenandoahHeapRegion* _region;
- size_t _live_data;
-} AgedRegionData;
-
-static int compare_by_aged_live(AgedRegionData a, AgedRegionData b) {
- if (a._live_data < b._live_data)
- return -1;
- else if (a._live_data > b._live_data)
- return 1;
- else return 0;
-}
-
-inline void assert_no_in_place_promotions() {
-#ifdef ASSERT
- class ShenandoahNoInPlacePromotions : public ShenandoahHeapRegionClosure {
- public:
- void heap_region_do(ShenandoahHeapRegion *r) override {
- assert(r->get_top_before_promote() == nullptr,
- "Region %zu should not be ready for in-place promotion", r->index());
- }
- } cl;
- ShenandoahHeap::heap()->heap_region_iterate(&cl);
-#endif
-}
-
-// Preselect for inclusion into the collection set all regions whose age is at or above tenure age and for which the
-// garbage percentage exceeds a dynamically adjusted threshold (known as the old-garbage threshold percentage). We
-// identify these regions by setting the appropriate entry of the collection set's preselected regions array to true.
-// All entries are initialized to false before calling this function.
-//
-// During the subsequent selection of the collection set, we give priority to these promotion set candidates.
-// Without this prioritization, we found that the aged regions tend to be ignored because they typically have
-// much less garbage and much more live data than the recently allocated "eden" regions. When aged regions are
-// repeatedly excluded from the collection set, the amount of live memory within the young generation tends to
-// accumulate and this has the undesirable side effect of causing young-generation collections to require much more
-// CPU and wall-clock time.
-//
-// A second benefit of treating aged regions differently than other regions during collection set selection is
-// that this allows us to more accurately budget memory to hold the results of evacuation. Memory for evacuation
-// of aged regions must be reserved in the old generation. Memory for evacuation of all other regions must be
-// reserved in the young generation.
-size_t ShenandoahGeneration::select_aged_regions(const size_t old_promotion_reserve) {
-
- // There should be no regions configured for subsequent in-place-promotions carried over from the previous cycle.
- assert_no_in_place_promotions();
-
- auto const heap = ShenandoahGenerationalHeap::heap();
- ShenandoahFreeSet* free_set = heap->free_set();
- bool* const candidate_regions_for_promotion_by_copy = heap->collection_set()->preselected_regions();
- ShenandoahMarkingContext* const ctx = heap->marking_context();
-
- const size_t old_garbage_threshold =
- (ShenandoahHeapRegion::region_size_bytes() * heap->old_generation()->heuristics()->get_old_garbage_threshold()) / 100;
-
- const size_t pip_used_threshold = (ShenandoahHeapRegion::region_size_bytes() * ShenandoahGenerationalMinPIPUsage) / 100;
-
- size_t promo_potential = 0;
- size_t candidates = 0;
-
- // Tracks the padding of space above top in regions eligible for promotion in place
- size_t promote_in_place_pad = 0;
-
- // Sort the promotion-eligible regions in order of increasing live-data-bytes so that we can first reclaim regions that require
- // less evacuation effort. This prioritizes garbage first, expanding the allocation pool early before we reclaim regions that
- // have more live data.
- const idx_t num_regions = heap->num_regions();
-
- ResourceMark rm;
- AgedRegionData* sorted_regions = NEW_RESOURCE_ARRAY(AgedRegionData, num_regions);
-
- ShenandoahFreeSet* freeset = heap->free_set();
-
- // Any region that is to be promoted in place needs to be retired from its Collector or Mutator partition.
- idx_t pip_low_collector_idx = freeset->max_regions();
- idx_t pip_high_collector_idx = -1;
- idx_t pip_low_mutator_idx = freeset->max_regions();
- idx_t pip_high_mutator_idx = -1;
- size_t collector_regions_to_pip = 0;
- size_t mutator_regions_to_pip = 0;
-
- size_t pip_mutator_regions = 0;
- size_t pip_collector_regions = 0;
- size_t pip_mutator_bytes = 0;
- size_t pip_collector_bytes = 0;
-
- for (idx_t i = 0; i < num_regions; i++) {
- ShenandoahHeapRegion* const r = heap->get_region(i);
- if (r->is_empty() || !r->has_live() || !r->is_young() || !r->is_regular()) {
- // skip over regions that aren't regular young with some live data
- continue;
- }
- if (heap->is_tenurable(r)) {
- if ((r->garbage() < old_garbage_threshold) && (r->used() > pip_used_threshold)) {
- // We prefer to promote this region in place because it has a small amount of garbage and a large usage.
- HeapWord* tams = ctx->top_at_mark_start(r);
- HeapWord* original_top = r->top();
- if (!heap->is_concurrent_old_mark_in_progress() && tams == original_top) {
- // No allocations from this region have been made during concurrent mark. It meets all the criteria
- // for in-place-promotion. Though we only need the value of top when we fill the end of the region,
- // we use this field to indicate that this region should be promoted in place during the evacuation
- // phase.
- r->save_top_before_promote();
- size_t remnant_bytes = r->free();
- size_t remnant_words = remnant_bytes / HeapWordSize;
- assert(ShenandoahHeap::min_fill_size() <= PLAB::min_size(), "Implementation makes invalid assumptions");
- if (remnant_words >= ShenandoahHeap::min_fill_size()) {
- ShenandoahHeap::fill_with_object(original_top, remnant_words);
- // Fill the remnant memory within this region to assure no allocations prior to promote in place. Otherwise,
- // newly allocated objects will not be parsable when promote in place tries to register them. Furthermore, any
- // new allocations would not necessarily be eligible for promotion. This addresses both issues.
- r->set_top(r->end());
- // The region r is either in the Mutator or Collector partition if remnant_words > heap()->plab_min_size.
- // Otherwise, the region is in the NotFree partition.
- ShenandoahFreeSetPartitionId p = free_set->membership(i);
- if (p == ShenandoahFreeSetPartitionId::Mutator) {
- mutator_regions_to_pip++;
- if (i < pip_low_mutator_idx) {
- pip_low_mutator_idx = i;
- }
- if (i > pip_high_mutator_idx) {
- pip_high_mutator_idx = i;
- }
- pip_mutator_regions++;
- pip_mutator_bytes += remnant_bytes;
- } else if (p == ShenandoahFreeSetPartitionId::Collector) {
- collector_regions_to_pip++;
- if (i < pip_low_collector_idx) {
- pip_low_collector_idx = i;
- }
- if (i > pip_high_collector_idx) {
- pip_high_collector_idx = i;
- }
- pip_collector_regions++;
- pip_collector_bytes += remnant_bytes;
- } else {
- assert((p == ShenandoahFreeSetPartitionId::NotFree) && (remnant_words < heap->plab_min_size()),
- "Should be NotFree if not in Collector or Mutator partitions");
- // In this case, the memory is already counted as used and the region has already been retired. There is
- // no need for further adjustments to used. Further, the remnant memory for this region will not be
- // unallocated or made available to OldCollector after pip.
- remnant_bytes = 0;
- }
- promote_in_place_pad += remnant_bytes;
- free_set->prepare_to_promote_in_place(i, remnant_bytes);
- } else {
- // Since the remnant is so small that this region has already been retired, we don't have to worry about any
- // accidental allocations occurring within this region before the region is promoted in place.
-
- // This region was already not in the Collector or Mutator set, so no need to remove it.
- assert(free_set->membership(i) == ShenandoahFreeSetPartitionId::NotFree, "sanity");
- }
- }
- // Else, we do not promote this region (either in place or by copy) because it has received new allocations.
-
- // During evacuation, we exclude from promotion regions for which age > tenure threshold, garbage < garbage-threshold,
- // used > pip_used_threshold, and get_top_before_promote() != tams
- } else {
- // Record this promotion-eligible candidate region. After sorting and selecting the best candidates below,
- // we may still decide to exclude this promotion-eligible region from the current collection set. If this
- // happens, we will consider this region as part of the anticipated promotion potential for the next GC
- // pass; see further below.
- sorted_regions[candidates]._region = r;
- sorted_regions[candidates++]._live_data = r->get_live_data_bytes();
- }
- } else {
- // We only evacuate & promote objects from regular regions whose garbage() is above old-garbage-threshold.
- // Objects in tenure-worthy regions with less garbage are promoted in place. These take a different path to
- // old-gen. Regions excluded from promotion because their garbage content is too low (causing us to anticipate that
- // the region would be promoted in place) may be eligible for evacuation promotion by the time promotion takes
- // place during a subsequent GC pass because more garbage is found within the region between now and then. This
- // should not happen if we are properly adapting the tenure age. The theory behind adaptive tenuring threshold
- // is to choose the youngest age that demonstrates no "significant" further loss of population since the previous
- // age. If not this, we expect the tenure age to demonstrate linear population decay for at least two population
- // samples, whereas we expect to observe exponential population decay for ages younger than the tenure age.
- //
- // In the case that certain regions which were anticipated to be promoted in place need to be promoted by
- // evacuation, it may be the case that there is not sufficient reserve within old-gen to hold evacuation of
- // these regions. The likely outcome is that these regions will not be selected for evacuation or promotion
- // in the current cycle and we will anticipate that they will be promoted in the next cycle. This will cause
- // us to reserve more old-gen memory so that these objects can be promoted in the subsequent cycle.
- if (heap->is_aging_cycle() && heap->age_census()->is_tenurable(r->age() + 1)) {
- if (r->garbage() >= old_garbage_threshold) {
- promo_potential += r->get_live_data_bytes();
- }
- }
- }
- // Note that we keep going even if one region is excluded from selection.
- // Subsequent regions may be selected if they have smaller live data.
- }
-
- if (pip_mutator_regions + pip_collector_regions > 0) {
- freeset->account_for_pip_regions(pip_mutator_regions, pip_mutator_bytes, pip_collector_regions, pip_collector_bytes);
- }
-
- // Retire any regions that have been selected for promote in place
- if (collector_regions_to_pip > 0) {
- freeset->shrink_interval_if_range_modifies_either_boundary(ShenandoahFreeSetPartitionId::Collector,
- pip_low_collector_idx, pip_high_collector_idx,
- collector_regions_to_pip);
- }
- if (mutator_regions_to_pip > 0) {
- freeset->shrink_interval_if_range_modifies_either_boundary(ShenandoahFreeSetPartitionId::Mutator,
- pip_low_mutator_idx, pip_high_mutator_idx,
- mutator_regions_to_pip);
- }
-
- // Sort in increasing order according to live data bytes. Note that candidates represents the number of regions
- // that qualify to be promoted by evacuation.
- size_t old_consumed = 0;
- if (candidates > 0) {
- size_t selected_regions = 0;
- size_t selected_live = 0;
- QuickSort::sort(sorted_regions, candidates, compare_by_aged_live);
- for (size_t i = 0; i < candidates; i++) {
- ShenandoahHeapRegion* const region = sorted_regions[i]._region;
- const size_t region_live_data = sorted_regions[i]._live_data;
- const size_t promotion_need = (size_t) (region_live_data * ShenandoahPromoEvacWaste);
- if (old_consumed + promotion_need <= old_promotion_reserve) {
- old_consumed += promotion_need;
- candidate_regions_for_promotion_by_copy[region->index()] = true;
- selected_regions++;
- selected_live += region_live_data;
- } else {
- // We rejected this promotable region from the collection set because we had no room to hold its copy.
- // Add this region to promo potential for next GC.
- promo_potential += region_live_data;
- assert(!candidate_regions_for_promotion_by_copy[region->index()], "Shouldn't be selected");
- }
- // We keep going even if one region is excluded from selection because we need to accumulate all eligible
- // regions that are not preselected into promo_potential
- }
- log_debug(gc, ergo)("Preselected %zu regions containing " PROPERFMT " live data,"
- " consuming: " PROPERFMT " of budgeted: " PROPERFMT,
- selected_regions, PROPERFMTARGS(selected_live), PROPERFMTARGS(old_consumed), PROPERFMTARGS(old_promotion_reserve));
- }
-
- log_info(gc, ergo)("Promotion potential of aged regions with sufficient garbage: " PROPERFMT, PROPERFMTARGS(promo_potential));
-
- heap->old_generation()->set_pad_for_promote_in_place(promote_in_place_pad);
- heap->old_generation()->set_promotion_potential(promo_potential);
- return old_consumed;
-}
-
void ShenandoahGeneration::prepare_regions_and_collection_set(bool concurrent) {
ShenandoahHeap* heap = ShenandoahHeap::heap();
ShenandoahCollectionSet* collection_set = heap->collection_set();
@@ -798,34 +297,7 @@ void ShenandoahGeneration::prepare_regions_and_collection_set(bool concurrent) {
collection_set->clear();
ShenandoahHeapLocker locker(heap->lock());
- if (is_generational) {
- // Seed the collection set with resource area-allocated
- // preselected regions, which are removed when we exit this scope.
- ShenandoahCollectionSetPreselector preselector(collection_set, heap->num_regions());
-
- // Find the amount that will be promoted, regions that will be promoted in
- // place, and preselected older regions that will be promoted by evacuation.
- compute_evacuation_budgets(heap);
-
- // Choose the collection set, including the regions preselected above for promotion into the old generation.
- size_t add_regions_to_old = _heuristics->choose_collection_set(collection_set);
- // Even if collection_set->is_empty(), we want to adjust budgets, making reserves available to mutator.
- adjust_evacuation_budgets(heap, collection_set, add_regions_to_old);
- if (is_global()) {
- // We have just chosen a collection set for a global cycle. The mark bitmap covering old regions is complete, so
- // the remembered set scan can use that to avoid walking into garbage. When the next old mark begins, we will
- // use the mark bitmap to make the old regions parsable by coalescing and filling any unmarked objects. Thus,
- // we prepare for old collections by remembering which regions are old at this time. Note that any objects
- // promoted into old regions will be above TAMS, and so will be considered marked. However, free regions that
- // become old after this point will not be covered correctly by the mark bitmap, so we must be careful not to
- // coalesce those regions. Only the old regions which are not part of the collection set at this point are
- // eligible for coalescing. As implemented now, this has the side effect of possibly initiating mixed-evacuations
- // after a global cycle for old regions that were not included in this collection set.
- heap->old_generation()->prepare_for_mixed_collections_after_global_gc();
- }
- } else {
- _heuristics->choose_collection_set(collection_set);
- }
+ _heuristics->choose_collection_set(collection_set);
}
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahGeneration.hpp b/src/hotspot/share/gc/shenandoah/shenandoahGeneration.hpp
index d49e3bed5f8..946f2b91520 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahGeneration.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahGeneration.hpp
@@ -60,31 +60,6 @@ protected:
ShenandoahHeuristics* _heuristics;
private:
- // Compute evacuation budgets prior to choosing collection set.
- void compute_evacuation_budgets(ShenandoahHeap* heap);
-
- // Adjust evacuation budgets after choosing collection set. The argument regions_to_xfer represents regions to be
- // transfered to old based on decisions made in top_off_collection_set()
- void adjust_evacuation_budgets(ShenandoahHeap* heap,
- ShenandoahCollectionSet* collection_set, size_t regions_to_xfer);
-
- // Preselect for possible inclusion into the collection set exactly the most
- // garbage-dense regions, including those that satisfy criteria 1 & 2 below,
- // and whose live bytes will fit within old_available budget:
- // Criterion 1. region age >= tenuring threshold
- // Criterion 2. region garbage percentage > old garbage threshold
- //
- // Identifies regions eligible for promotion in place,
- // being those of at least tenuring_threshold age that have lower garbage
- // density.
- //
- // Updates promotion_potential and pad_for_promote_in_place fields
- // of the heap. Returns bytes of live object memory in the preselected
- // regions, which are marked in the preselected_regions() indicator
- // array of the heap's collection set, which should be initialized
- // to false.
- size_t select_aged_regions(size_t old_promotion_reserve);
-
// Return available assuming that we can allocate no more than capacity bytes within this generation.
size_t available(size_t capacity) const;
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahGenerationalControlThread.cpp b/src/hotspot/share/gc/shenandoah/shenandoahGenerationalControlThread.cpp
index 018b4898a19..3b57190cc75 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahGenerationalControlThread.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahGenerationalControlThread.cpp
@@ -47,7 +47,7 @@
#include "utilities/events.hpp"
ShenandoahGenerationalControlThread::ShenandoahGenerationalControlThread() :
- _control_lock(Mutex::nosafepoint - 2, "ShenandoahGCRequest_lock", true),
+ _control_lock(CONTROL_LOCK_RANK, "ShenandoahGCRequest_lock", true),
_requested_gc_cause(GCCause::_no_gc),
_requested_generation(nullptr),
_gc_mode(none),
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahHeap.cpp b/src/hotspot/share/gc/shenandoah/shenandoahHeap.cpp
index ef99bd98c93..ccfc1c036c2 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahHeap.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahHeap.cpp
@@ -2719,18 +2719,6 @@ bool ShenandoahRegionIterator::has_next() const {
return _index < _heap->num_regions();
}
-char ShenandoahHeap::gc_state() const {
- return _gc_state.raw_value();
-}
-
-bool ShenandoahHeap::is_gc_state(GCState state) const {
- // If the global gc state has been changed, but hasn't yet been propagated to all threads, then
- // the global gc state is the correct value. Once the gc state has been synchronized with all threads,
- // _gc_state_changed will be toggled to false and we need to use the thread local state.
- return _gc_state_changed ? _gc_state.is_set(state) : ShenandoahThreadLocalData::is_gc_state(state);
-}
-
-
ShenandoahLiveData* ShenandoahHeap::get_liveness_cache(uint worker_id) {
#ifdef ASSERT
assert(_liveness_cache != nullptr, "sanity");
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahHeap.hpp b/src/hotspot/share/gc/shenandoah/shenandoahHeap.hpp
index 174001170f4..9240091070b 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahHeap.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahHeap.hpp
@@ -353,7 +353,7 @@ private:
public:
// This returns the raw value of the singular, global gc state.
- char gc_state() const;
+ inline char gc_state() const;
// Compares the given state against either the global gc state, or the thread local state.
// The global gc state may change on a safepoint and is the correct value to use until
@@ -361,7 +361,7 @@ public:
// compare against the thread local state). The thread local gc state may also be changed
// by a handshake operation, in which case, this function continues using the updated thread
// local value.
- bool is_gc_state(GCState state) const;
+ inline bool is_gc_state(GCState state) const;
// This copies the global gc state into a thread local variable for all threads.
// The thread local gc state is primarily intended to support quick access at barriers.
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahHeap.inline.hpp b/src/hotspot/share/gc/shenandoah/shenandoahHeap.inline.hpp
index 34c279a1495..e35f116b843 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahHeap.inline.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahHeap.inline.hpp
@@ -452,6 +452,17 @@ inline bool ShenandoahHeap::in_collection_set_loc(void* p) const {
return collection_set()->is_in_loc(p);
}
+inline char ShenandoahHeap::gc_state() const {
+ return _gc_state.raw_value();
+}
+
+inline bool ShenandoahHeap::is_gc_state(GCState state) const {
+ // If the global gc state has been changed, but hasn't yet been propagated to all threads, then
+ // the global gc state is the correct value. Once the gc state has been synchronized with all threads,
+ // _gc_state_changed will be toggled to false and we need to use the thread local state.
+ return _gc_state_changed ? _gc_state.is_set(state) : ShenandoahThreadLocalData::is_gc_state(state);
+}
+
inline bool ShenandoahHeap::is_idle() const {
return _gc_state_changed ? _gc_state.is_clear() : ShenandoahThreadLocalData::gc_state(Thread::current()) == 0;
}
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.cpp b/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.cpp
index 6bb8382de0a..b0c13df6c4f 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.cpp
@@ -43,7 +43,6 @@
#include "memory/resourceArea.hpp"
#include "memory/universe.hpp"
#include "oops/oop.inline.hpp"
-#include "runtime/atomicAccess.hpp"
#include "runtime/globals_extension.hpp"
#include "runtime/java.hpp"
#include "runtime/mutexLocker.hpp"
@@ -384,7 +383,7 @@ size_t ShenandoahHeapRegion::get_plab_allocs() const {
void ShenandoahHeapRegion::set_live_data(size_t s) {
assert(Thread::current()->is_VM_thread(), "by VM thread");
- _live_data = (s >> LogHeapWordSize);
+ _live_data.store_relaxed(s >> LogHeapWordSize);
}
void ShenandoahHeapRegion::print_on(outputStream* st) const {
@@ -435,7 +434,7 @@ void ShenandoahHeapRegion::print_on(outputStream* st) const {
st->print("|TAMS " SHR_PTR_FORMAT,
p2i(ShenandoahHeap::heap()->marking_context()->top_at_mark_start(const_cast(this))));
st->print("|UWM " SHR_PTR_FORMAT,
- p2i(_update_watermark));
+ p2i(_update_watermark.load_relaxed()));
st->print("|U %5zu%1s", byte_size_in_proper_unit(used()), proper_unit_for_byte_size(used()));
st->print("|T %5zu%1s", byte_size_in_proper_unit(get_tlab_allocs()), proper_unit_for_byte_size(get_tlab_allocs()));
st->print("|G %5zu%1s", byte_size_in_proper_unit(get_gclab_allocs()), proper_unit_for_byte_size(get_gclab_allocs()));
@@ -839,20 +838,20 @@ void ShenandoahHeapRegion::set_state(RegionState to) {
evt.set_to(to);
evt.commit();
}
- AtomicAccess::store(&_state, to);
+ _state.store_relaxed(to);
}
void ShenandoahHeapRegion::record_pin() {
- AtomicAccess::add(&_critical_pins, (size_t)1);
+ _critical_pins.add_then_fetch((size_t)1);
}
void ShenandoahHeapRegion::record_unpin() {
assert(pin_count() > 0, "Region %zu should have non-zero pins", index());
- AtomicAccess::sub(&_critical_pins, (size_t)1);
+ _critical_pins.sub_then_fetch((size_t)1);
}
size_t ShenandoahHeapRegion::pin_count() const {
- return AtomicAccess::load(&_critical_pins);
+ return _critical_pins.load_relaxed();
}
void ShenandoahHeapRegion::set_affiliation(ShenandoahAffiliation new_affiliation) {
@@ -864,7 +863,7 @@ void ShenandoahHeapRegion::set_affiliation(ShenandoahAffiliation new_affiliation
log_debug(gc)("Setting affiliation of Region %zu from %s to %s, top: " PTR_FORMAT ", TAMS: " PTR_FORMAT
", watermark: " PTR_FORMAT ", top_bitmap: " PTR_FORMAT,
index(), shenandoah_affiliation_name(region_affiliation), shenandoah_affiliation_name(new_affiliation),
- p2i(top()), p2i(ctx->top_at_mark_start(this)), p2i(_update_watermark), p2i(ctx->top_bitmap(this)));
+ p2i(top()), p2i(ctx->top_at_mark_start(this)), p2i(_update_watermark.load_relaxed()), p2i(ctx->top_bitmap(this)));
}
#ifdef ASSERT
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.hpp b/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.hpp
index 9da2816e2c9..3a0ac042f57 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.hpp
@@ -34,6 +34,7 @@
#include "gc/shenandoah/shenandoahAsserts.hpp"
#include "gc/shenandoah/shenandoahHeap.hpp"
#include "gc/shenandoah/shenandoahPadding.hpp"
+#include "runtime/atomic.hpp"
#include "utilities/sizes.hpp"
class VMStructs;
@@ -217,7 +218,7 @@ public:
bool is_alloc_allowed() const { auto cur_state = state(); return is_empty_state(cur_state) || cur_state == _regular || cur_state == _pinned; }
bool is_stw_move_allowed() const { auto cur_state = state(); return cur_state == _regular || cur_state == _cset || (ShenandoahHumongousMoves && cur_state == _humongous_start); }
- RegionState state() const { return AtomicAccess::load(&_state); }
+ RegionState state() const { return _state.load_relaxed(); }
int state_ordinal() const { return region_state_to_ordinal(state()); }
void record_pin();
@@ -247,7 +248,7 @@ private:
HeapWord* _top_before_promoted;
// Seldom updated fields
- volatile RegionState _state;
+ Atomic _state;
HeapWord* _coalesce_and_fill_boundary; // for old regions not selected as collection set candidates.
// Frequently updated fields
@@ -257,12 +258,12 @@ private:
size_t _gclab_allocs;
size_t _plab_allocs;
- volatile size_t _live_data;
- volatile size_t _critical_pins;
+ Atomic _live_data;
+ Atomic _critical_pins;
size_t _mixed_candidate_garbage_words;
- HeapWord* volatile _update_watermark;
+ Atomic _update_watermark;
uint _age;
bool _promoted_in_place;
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.inline.hpp b/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.inline.hpp
index be982433885..39b7c732703 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.inline.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahHeapRegion.inline.hpp
@@ -32,7 +32,6 @@
#include "gc/shenandoah/shenandoahGenerationalHeap.hpp"
#include "gc/shenandoah/shenandoahHeap.inline.hpp"
#include "gc/shenandoah/shenandoahOldGeneration.hpp"
-#include "runtime/atomicAccess.hpp"
HeapWord* ShenandoahHeapRegion::allocate_aligned(size_t size, ShenandoahAllocRequest &req, size_t alignment_in_bytes) {
shenandoah_assert_heaplocked_or_safepoint();
@@ -147,16 +146,16 @@ inline void ShenandoahHeapRegion::increase_live_data_gc_words(size_t s) {
}
inline void ShenandoahHeapRegion::internal_increase_live_data(size_t s) {
- AtomicAccess::add(&_live_data, s, memory_order_relaxed);
+ _live_data.add_then_fetch(s, memory_order_relaxed);
}
inline void ShenandoahHeapRegion::clear_live_data() {
- AtomicAccess::store(&_live_data, (size_t)0);
+ _live_data.store_relaxed((size_t)0);
_promoted_in_place = false;
}
inline size_t ShenandoahHeapRegion::get_live_data_words() const {
- return AtomicAccess::load(&_live_data);
+ return _live_data.load_relaxed();
}
inline size_t ShenandoahHeapRegion::get_live_data_bytes() const {
@@ -205,21 +204,21 @@ inline size_t ShenandoahHeapRegion::garbage_before_padded_for_promote() const {
}
inline HeapWord* ShenandoahHeapRegion::get_update_watermark() const {
- HeapWord* watermark = AtomicAccess::load_acquire(&_update_watermark);
+ HeapWord* watermark = _update_watermark.load_acquire();
assert(bottom() <= watermark && watermark <= top(), "within bounds");
return watermark;
}
inline void ShenandoahHeapRegion::set_update_watermark(HeapWord* w) {
assert(bottom() <= w && w <= top(), "within bounds");
- AtomicAccess::release_store(&_update_watermark, w);
+ _update_watermark.release_store(w);
}
// Fast version that avoids synchronization, only to be used at safepoints.
inline void ShenandoahHeapRegion::set_update_watermark_at_safepoint(HeapWord* w) {
assert(bottom() <= w && w <= top(), "within bounds");
assert(SafepointSynchronize::is_at_safepoint(), "Should be at Shenandoah safepoint");
- _update_watermark = w;
+ _update_watermark.store_relaxed(w);
}
inline ShenandoahAffiliation ShenandoahHeapRegion::affiliation() const {
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahMark.inline.hpp b/src/hotspot/share/gc/shenandoah/shenandoahMark.inline.hpp
index 849459157b5..ba24e890769 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahMark.inline.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahMark.inline.hpp
@@ -226,8 +226,6 @@ inline void ShenandoahMark::do_chunked_array(ShenandoahObjToScanQueue* q, T* cl,
assert(obj->is_objArray(), "expect object array");
objArrayOop array = objArrayOop(obj);
- assert (ObjArrayMarkingStride > 0, "sanity");
-
// Split out tasks, as suggested in ShenandoahMarkTask docs. Avoid pushing tasks that
// are known to start beyond the array.
while ((1 << pow) > (int)ObjArrayMarkingStride && (chunk*2 < ShenandoahMarkTask::chunk_size())) {
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahOldGeneration.cpp b/src/hotspot/share/gc/shenandoah/shenandoahOldGeneration.cpp
index aed768b9db1..d5e34d02b13 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahOldGeneration.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahOldGeneration.cpp
@@ -63,7 +63,7 @@ private:
uint _nworkers;
ShenandoahHeapRegion** _coalesce_and_fill_region_array;
uint _coalesce_and_fill_region_count;
- volatile bool _is_preempted;
+ Atomic _is_preempted;
public:
ShenandoahConcurrentCoalesceAndFillTask(uint nworkers,
@@ -88,7 +88,7 @@ public:
if (!r->oop_coalesce_and_fill(true)) {
// Coalesce and fill has been preempted
- AtomicAccess::store(&_is_preempted, true);
+ _is_preempted.store_relaxed(true);
return;
}
}
@@ -96,7 +96,7 @@ public:
// Value returned from is_completed() is only valid after all worker thread have terminated.
bool is_completed() {
- return !AtomicAccess::load(&_is_preempted);
+ return !_is_preempted.load_relaxed();
}
};
@@ -147,23 +147,23 @@ void ShenandoahOldGeneration::augment_promoted_reserve(size_t increment) {
void ShenandoahOldGeneration::reset_promoted_expended() {
shenandoah_assert_heaplocked_or_safepoint();
- AtomicAccess::store(&_promoted_expended, static_cast(0));
- AtomicAccess::store(&_promotion_failure_count, static_cast(0));
- AtomicAccess::store(&_promotion_failure_words, static_cast(0));
+ _promoted_expended.store_relaxed(0);
+ _promotion_failure_count.store_relaxed(0);
+ _promotion_failure_words.store_relaxed(0);
}
size_t ShenandoahOldGeneration::expend_promoted(size_t increment) {
shenandoah_assert_heaplocked_or_safepoint();
assert(get_promoted_expended() + increment <= get_promoted_reserve(), "Do not expend more promotion than budgeted");
- return AtomicAccess::add(&_promoted_expended, increment);
+ return _promoted_expended.add_then_fetch(increment);
}
size_t ShenandoahOldGeneration::unexpend_promoted(size_t decrement) {
- return AtomicAccess::sub(&_promoted_expended, decrement);
+ return _promoted_expended.sub_then_fetch(decrement);
}
size_t ShenandoahOldGeneration::get_promoted_expended() const {
- return AtomicAccess::load(&_promoted_expended);
+ return _promoted_expended.load_relaxed();
}
bool ShenandoahOldGeneration::can_allocate(const ShenandoahAllocRequest &req) const {
@@ -582,8 +582,8 @@ void ShenandoahOldGeneration::handle_failed_evacuation() {
}
void ShenandoahOldGeneration::handle_failed_promotion(Thread* thread, size_t size) {
- AtomicAccess::inc(&_promotion_failure_count);
- AtomicAccess::add(&_promotion_failure_words, size);
+ _promotion_failure_count.add_then_fetch(1UL);
+ _promotion_failure_words.and_then_fetch(size);
LogTarget(Debug, gc, plab) lt;
LogStream ls(lt);
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahOldGeneration.hpp b/src/hotspot/share/gc/shenandoah/shenandoahOldGeneration.hpp
index 633d2c9f617..5ebad461f3c 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahOldGeneration.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahOldGeneration.hpp
@@ -64,7 +64,7 @@ private:
// is therefore always accessed through atomic operations. This is increased when a
// PLAB is allocated for promotions. The value is decreased by the amount of memory
// remaining in a PLAB when it is retired.
- size_t _promoted_expended;
+ Atomic _promoted_expended;
// Represents the quantity of live bytes we expect to promote during the next GC cycle, either by
// evacuation or by promote-in-place. This value is used by the young heuristic to trigger mixed collections.
@@ -78,8 +78,8 @@ private:
// Keep track of the number and size of promotions that failed. Perhaps we should use this to increase
// the size of the old generation for the next collection cycle.
- size_t _promotion_failure_count;
- size_t _promotion_failure_words;
+ Atomic _promotion_failure_count;
+ Atomic _promotion_failure_words;
// During construction of the collection set, we keep track of regions that are eligible
// for promotion in place. These fields track the count of those humongous and regular regions.
@@ -126,8 +126,8 @@ public:
size_t get_promoted_expended() const;
// Return the count and size (in words) of failed promotions since the last reset
- size_t get_promotion_failed_count() const { return AtomicAccess::load(&_promotion_failure_count); }
- size_t get_promotion_failed_words() const { return AtomicAccess::load(&_promotion_failure_words); }
+ size_t get_promotion_failed_count() const { return _promotion_failure_count.load_relaxed(); }
+ size_t get_promotion_failed_words() const { return _promotion_failure_words.load_relaxed(); }
// Test if there is enough memory reserved for this promotion
bool can_promote(size_t requested_bytes) const {
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahSharedVariables.hpp b/src/hotspot/share/gc/shenandoah/shenandoahSharedVariables.hpp
index 12c01ad5c90..e23187a5d3f 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahSharedVariables.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahSharedVariables.hpp
@@ -27,7 +27,7 @@
#include "gc/shenandoah/shenandoahPadding.hpp"
#include "memory/allocation.hpp"
-#include "runtime/atomicAccess.hpp"
+#include "runtime/atomic.hpp"
typedef int32_t ShenandoahSharedValue;
typedef struct ShenandoahSharedFlag {
@@ -37,7 +37,7 @@ typedef struct ShenandoahSharedFlag {
};
shenandoah_padding(0);
- volatile ShenandoahSharedValue value;
+ Atomic value;
shenandoah_padding(1);
ShenandoahSharedFlag() {
@@ -45,19 +45,19 @@ typedef struct ShenandoahSharedFlag {
}
void set() {
- AtomicAccess::release_store_fence(&value, (ShenandoahSharedValue)SET);
+ value.release_store_fence((ShenandoahSharedValue)SET);
}
void unset() {
- AtomicAccess::release_store_fence(&value, (ShenandoahSharedValue)UNSET);
+ value.release_store_fence((ShenandoahSharedValue)UNSET);
}
bool is_set() const {
- return AtomicAccess::load_acquire(&value) == SET;
+ return value.load_acquire() == SET;
}
bool is_unset() const {
- return AtomicAccess::load_acquire(&value) == UNSET;
+ return value.load_acquire() == UNSET;
}
void set_cond(bool val) {
@@ -72,7 +72,7 @@ typedef struct ShenandoahSharedFlag {
if (is_set()) {
return false;
}
- ShenandoahSharedValue old = AtomicAccess::cmpxchg(&value, (ShenandoahSharedValue)UNSET, (ShenandoahSharedValue)SET);
+ ShenandoahSharedValue old = value.compare_exchange((ShenandoahSharedValue)UNSET, (ShenandoahSharedValue)SET);
return old == UNSET; // success
}
@@ -80,17 +80,13 @@ typedef struct ShenandoahSharedFlag {
if (!is_set()) {
return false;
}
- ShenandoahSharedValue old = AtomicAccess::cmpxchg(&value, (ShenandoahSharedValue)SET, (ShenandoahSharedValue)UNSET);
+ ShenandoahSharedValue old = value.compare_exchange((ShenandoahSharedValue)SET, (ShenandoahSharedValue)UNSET);
return old == SET; // success
}
- volatile ShenandoahSharedValue* addr_of() {
- return &value;
- }
-
private:
volatile ShenandoahSharedValue* operator&() {
- fatal("Use addr_of() instead");
+ fatal("Not supported");
return nullptr;
}
@@ -105,7 +101,7 @@ private:
typedef struct ShenandoahSharedBitmap {
shenandoah_padding(0);
- volatile ShenandoahSharedValue value;
+ Atomic value;
shenandoah_padding(1);
ShenandoahSharedBitmap() {
@@ -116,7 +112,7 @@ typedef struct ShenandoahSharedBitmap {
assert (mask < (sizeof(ShenandoahSharedValue) * CHAR_MAX), "sanity");
ShenandoahSharedValue mask_val = (ShenandoahSharedValue) mask;
while (true) {
- ShenandoahSharedValue ov = AtomicAccess::load_acquire(&value);
+ ShenandoahSharedValue ov = value.load_acquire();
// We require all bits of mask_val to be set
if ((ov & mask_val) == mask_val) {
// already set
@@ -124,7 +120,7 @@ typedef struct ShenandoahSharedBitmap {
}
ShenandoahSharedValue nv = ov | mask_val;
- if (AtomicAccess::cmpxchg(&value, ov, nv) == ov) {
+ if (value.compare_exchange(ov, nv) == ov) {
// successfully set: if value returned from cmpxchg equals ov, then nv has overwritten value.
return;
}
@@ -135,14 +131,14 @@ typedef struct ShenandoahSharedBitmap {
assert (mask < (sizeof(ShenandoahSharedValue) * CHAR_MAX), "sanity");
ShenandoahSharedValue mask_val = (ShenandoahSharedValue) mask;
while (true) {
- ShenandoahSharedValue ov = AtomicAccess::load_acquire(&value);
+ ShenandoahSharedValue ov = value.load_acquire();
if ((ov & mask_val) == 0) {
// already unset
return;
}
ShenandoahSharedValue nv = ov & ~mask_val;
- if (AtomicAccess::cmpxchg(&value, ov, nv) == ov) {
+ if (value.compare_exchange(ov, nv) == ov) {
// successfully unset
return;
}
@@ -150,7 +146,7 @@ typedef struct ShenandoahSharedBitmap {
}
void clear() {
- AtomicAccess::release_store_fence(&value, (ShenandoahSharedValue)0);
+ value.release_store_fence((ShenandoahSharedValue)0);
}
// Returns true iff any bit set in mask is set in this.value.
@@ -161,18 +157,18 @@ typedef struct ShenandoahSharedBitmap {
// Returns true iff all bits set in mask are set in this.value.
bool is_set_exactly(uint mask) const {
assert (mask < (sizeof(ShenandoahSharedValue) * CHAR_MAX), "sanity");
- uint uvalue = AtomicAccess::load_acquire(&value);
+ uint uvalue = value.load_acquire();
return (uvalue & mask) == mask;
}
// Returns true iff all bits set in mask are unset in this.value.
bool is_unset(uint mask) const {
assert (mask < (sizeof(ShenandoahSharedValue) * CHAR_MAX), "sanity");
- return (AtomicAccess::load_acquire(&value) & (ShenandoahSharedValue) mask) == 0;
+ return (value.load_acquire() & (ShenandoahSharedValue) mask) == 0;
}
bool is_clear() const {
- return (AtomicAccess::load_acquire(&value)) == 0;
+ return (value.load_acquire()) == 0;
}
void set_cond(uint mask, bool val) {
@@ -183,17 +179,13 @@ typedef struct ShenandoahSharedBitmap {
}
}
- volatile ShenandoahSharedValue* addr_of() {
- return &value;
- }
-
ShenandoahSharedValue raw_value() const {
- return value;
+ return value.load_relaxed();
}
private:
volatile ShenandoahSharedValue* operator&() {
- fatal("Use addr_of() instead");
+ fatal("Not supported");
return nullptr;
}
@@ -210,42 +202,36 @@ template
struct ShenandoahSharedEnumFlag {
typedef uint32_t EnumValueType;
shenandoah_padding(0);
- volatile EnumValueType value;
+ Atomic value;
shenandoah_padding(1);
- ShenandoahSharedEnumFlag() {
- value = 0;
- }
+ ShenandoahSharedEnumFlag() : value(0) {}
void set(T v) {
assert (v >= 0, "sanity");
assert (v < (sizeof(EnumValueType) * CHAR_MAX), "sanity");
- AtomicAccess::release_store_fence(&value, (EnumValueType)v);
+ value.release_store_fence((EnumValueType)v);
}
T get() const {
- return (T)AtomicAccess::load_acquire(&value);
+ return (T)value.load_acquire();
}
T cmpxchg(T new_value, T expected) {
assert (new_value >= 0, "sanity");
assert (new_value < (sizeof(EnumValueType) * CHAR_MAX), "sanity");
- return (T)AtomicAccess::cmpxchg(&value, (EnumValueType)expected, (EnumValueType)new_value);
+ return (T)value.compare_exchange((EnumValueType)expected, (EnumValueType)new_value);
}
T xchg(T new_value) {
assert (new_value >= 0, "sanity");
assert (new_value < (sizeof(EnumValueType) * CHAR_MAX), "sanity");
- return (T)AtomicAccess::xchg(&value, (EnumValueType)new_value);
- }
-
- volatile EnumValueType* addr_of() {
- return &value;
+ return (T)value.exchange((EnumValueType)new_value);
}
private:
volatile T* operator&() {
- fatal("Use addr_of() instead");
+ fatal("Not supported");
return nullptr;
}
@@ -260,7 +246,7 @@ private:
typedef struct ShenandoahSharedSemaphore {
shenandoah_padding(0);
- volatile ShenandoahSharedValue value;
+ Atomic value;
shenandoah_padding(1);
static uint max_tokens() {
@@ -269,17 +255,17 @@ typedef struct ShenandoahSharedSemaphore {
ShenandoahSharedSemaphore(uint tokens) {
assert(tokens <= max_tokens(), "sanity");
- AtomicAccess::release_store_fence(&value, (ShenandoahSharedValue)tokens);
+ value.release_store_fence((ShenandoahSharedValue)tokens);
}
bool try_acquire() {
while (true) {
- ShenandoahSharedValue ov = AtomicAccess::load_acquire(&value);
+ ShenandoahSharedValue ov = value.load_acquire();
if (ov == 0) {
return false;
}
ShenandoahSharedValue nv = ov - 1;
- if (AtomicAccess::cmpxchg(&value, ov, nv) == ov) {
+ if (value.compare_exchange(ov, nv) == ov) {
// successfully set
return true;
}
@@ -287,7 +273,7 @@ typedef struct ShenandoahSharedSemaphore {
}
void claim_all() {
- AtomicAccess::release_store_fence(&value, (ShenandoahSharedValue)0);
+ value.release_store_fence((ShenandoahSharedValue)0);
}
} ShenandoahSharedSemaphore;
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahVerifier.cpp b/src/hotspot/share/gc/shenandoah/shenandoahVerifier.cpp
index 0cc6d4c6ed4..225339a3219 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahVerifier.cpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahVerifier.cpp
@@ -42,7 +42,7 @@
#include "memory/iterator.inline.hpp"
#include "memory/resourceArea.hpp"
#include "oops/compressedOops.inline.hpp"
-#include "runtime/atomicAccess.hpp"
+#include "runtime/atomic.hpp"
#include "runtime/orderAccess.hpp"
#include "runtime/threads.hpp"
#include "utilities/align.hpp"
@@ -110,15 +110,15 @@ private:
void do_oop_work(T* p) {
T o = RawAccess<>::oop_load(p);
if (!CompressedOops::is_null(o)) {
- oop obj = CompressedOops::decode_not_null(o);
+ // Basic verification should happen before we touch anything else.
+ // For performance reasons, only fully verify non-marked field values.
+ // We are here when the host object for *p is already marked.
+ oop obj = CompressedOops::decode_raw_not_null(o);
+ verify_oop_at_basic(p, obj);
+
if (is_instance_ref_klass(ShenandoahForwarding::klass(obj))) {
obj = ShenandoahForwarding::get_forwardee(obj);
}
- // Single threaded verification can use faster non-atomic stack and bitmap
- // methods.
- //
- // For performance reasons, only fully verify non-marked field values.
- // We are here when the host object for *p is already marked.
if (in_generation(obj) && _map->par_mark(obj)) {
verify_oop_at(p, obj);
_stack->push(ShenandoahVerifierTask(obj));
@@ -131,7 +131,7 @@ private:
return _generation->contains(region);
}
- void verify_oop(oop obj) {
+ void verify_oop(oop obj, bool basic = false) {
// Perform consistency checks with gradually decreasing safety level. This guarantees
// that failure report would not try to touch something that was not yet verified to be
// safe to process.
@@ -174,17 +174,21 @@ private:
}
}
+ check(ShenandoahAsserts::_safe_unknown, obj, obj_reg->is_active(),
+ "Object should be in active region");
+
// ------------ obj is safe at this point --------------
- check(ShenandoahAsserts::_safe_oop, obj, obj_reg->is_active(),
- "Object should be in active region");
+ if (basic) {
+ return;
+ }
switch (_options._verify_liveness) {
case ShenandoahVerifier::_verify_liveness_disable:
// skip
break;
case ShenandoahVerifier::_verify_liveness_complete:
- AtomicAccess::add(&_ld[obj_reg->index()], (uint) ShenandoahForwarding::size(obj), memory_order_relaxed);
+ _ld[obj_reg->index()].add_then_fetch((uint) ShenandoahForwarding::size(obj), memory_order_relaxed);
// fallthrough for fast failure for un-live regions:
case ShenandoahVerifier::_verify_liveness_conservative:
check(ShenandoahAsserts::_safe_oop, obj, obj_reg->has_live() ||
@@ -331,6 +335,18 @@ public:
_interior_loc = nullptr;
}
+ /**
+ * Verify object with known interior reference, with only basic verification.
+ * @param p interior reference where the object is referenced from; can be off-heap
+ * @param obj verified object
+ */
+ template
+ void verify_oop_at_basic(T* p, oop obj) {
+ _interior_loc = p;
+ verify_oop(obj, /* basic = */ true);
+ _interior_loc = nullptr;
+ }
+
/**
* Verify object without known interior reference.
* Useful when picking up the object at known offset in heap,
@@ -593,7 +609,7 @@ private:
ShenandoahHeap* _heap;
ShenandoahLivenessData* _ld;
MarkBitMap* _bitmap;
- volatile size_t _processed;
+ Atomic _processed;
ShenandoahGeneration* _generation;
public:
@@ -612,7 +628,7 @@ public:
_generation(generation) {};
size_t processed() const {
- return _processed;
+ return _processed.load_relaxed();
}
void work(uint worker_id) override {
@@ -648,7 +664,7 @@ public:
}
}
- AtomicAccess::add(&_processed, processed, memory_order_relaxed);
+ _processed.add_then_fetch(processed, memory_order_relaxed);
}
};
@@ -669,8 +685,8 @@ class ShenandoahVerifierMarkedRegionTask : public WorkerTask {
ShenandoahHeap *_heap;
MarkBitMap* _bitmap;
ShenandoahLivenessData* _ld;
- volatile size_t _claimed;
- volatile size_t _processed;
+ Atomic _claimed;
+ Atomic _processed;
ShenandoahGeneration* _generation;
public:
@@ -690,7 +706,7 @@ public:
_generation(generation) {}
size_t processed() {
- return AtomicAccess::load(&_processed);
+ return _processed.load_relaxed();
}
void work(uint worker_id) override {
@@ -705,7 +721,7 @@ public:
_options);
while (true) {
- size_t v = AtomicAccess::fetch_then_add(&_claimed, 1u, memory_order_relaxed);
+ size_t v = _claimed.fetch_then_add(1u, memory_order_relaxed);
if (v < _heap->num_regions()) {
ShenandoahHeapRegion* r = _heap->get_region(v);
if (!in_generation(r)) {
@@ -733,7 +749,7 @@ public:
if (_generation->complete_marking_context()->is_marked(cast_to_oop(obj))) {
verify_and_follow(obj, stack, cl, &processed);
}
- AtomicAccess::add(&_processed, processed, memory_order_relaxed);
+ _processed.add_then_fetch(processed, memory_order_relaxed);
}
virtual void work_regular(ShenandoahHeapRegion *r, ShenandoahVerifierStack &stack, ShenandoahVerifyOopClosure &cl) {
@@ -766,7 +782,7 @@ public:
}
}
- AtomicAccess::add(&_processed, processed, memory_order_relaxed);
+ _processed.add_then_fetch(processed, memory_order_relaxed);
}
void verify_and_follow(HeapWord *addr, ShenandoahVerifierStack &stack, ShenandoahVerifyOopClosure &cl, size_t *processed) {
@@ -1035,12 +1051,12 @@ void ShenandoahVerifier::verify_at_safepoint(ShenandoahGeneration* generation,
if (r->is_humongous()) {
// For humongous objects, test if start region is marked live, and if so,
// all humongous regions in that chain have live data equal to their "used".
- juint start_live = AtomicAccess::load(&ld[r->humongous_start_region()->index()]);
+ juint start_live = ld[r->humongous_start_region()->index()].load_relaxed();
if (start_live > 0) {
verf_live = (juint)(r->used() / HeapWordSize);
}
} else {
- verf_live = AtomicAccess::load(&ld[r->index()]);
+ verf_live = ld[r->index()].load_relaxed();
}
size_t reg_live = r->get_live_data_words();
@@ -1232,7 +1248,9 @@ private:
void do_oop_work(T* p) {
T o = RawAccess<>::oop_load(p);
if (!CompressedOops::is_null(o)) {
- oop obj = CompressedOops::decode_not_null(o);
+ oop obj = CompressedOops::decode_raw_not_null(o);
+ ShenandoahAsserts::assert_correct(p, obj, __FILE__, __LINE__);
+
oop fwd = ShenandoahForwarding::get_forwardee_raw_unchecked(obj);
if (obj != fwd) {
ShenandoahAsserts::print_failure(ShenandoahAsserts::_safe_all, obj, p, nullptr,
@@ -1252,7 +1270,9 @@ private:
void do_oop_work(T* p) {
T o = RawAccess<>::oop_load(p);
if (!CompressedOops::is_null(o)) {
- oop obj = CompressedOops::decode_not_null(o);
+ oop obj = CompressedOops::decode_raw_not_null(o);
+ ShenandoahAsserts::assert_correct(p, obj, __FILE__, __LINE__);
+
ShenandoahHeap* heap = ShenandoahHeap::heap();
if (!heap->marking_context()->is_marked_or_old(obj)) {
@@ -1306,7 +1326,9 @@ public:
inline void work(T* p) {
T o = RawAccess<>::oop_load(p);
if (!CompressedOops::is_null(o)) {
- oop obj = CompressedOops::decode_not_null(o);
+ oop obj = CompressedOops::decode_raw_not_null(o);
+ ShenandoahAsserts::assert_correct(p, obj, __FILE__, __LINE__);
+
if (_heap->is_in_young(obj) && !_scanner->is_card_dirty((HeapWord*) p)) {
ShenandoahAsserts::print_failure(ShenandoahAsserts::_safe_all, obj, p, nullptr,
_message, "clean card, it should be dirty.", __FILE__, __LINE__);
diff --git a/src/hotspot/share/gc/shenandoah/shenandoahVerifier.hpp b/src/hotspot/share/gc/shenandoah/shenandoahVerifier.hpp
index f66d7bbec77..7e683cf7af8 100644
--- a/src/hotspot/share/gc/shenandoah/shenandoahVerifier.hpp
+++ b/src/hotspot/share/gc/shenandoah/shenandoahVerifier.hpp
@@ -52,7 +52,7 @@ private:
};
typedef Stack ShenandoahVerifierStack;
-typedef volatile juint ShenandoahLivenessData;
+typedef Atomic ShenandoahLivenessData;
class ShenandoahVerifier : public CHeapObj {
private:
diff --git a/src/hotspot/share/gc/shenandoah/vmStructs_shenandoah.hpp b/src/hotspot/share/gc/shenandoah/vmStructs_shenandoah.hpp
index 3968575d089..e5e2b14a3a1 100644
--- a/src/hotspot/share/gc/shenandoah/vmStructs_shenandoah.hpp
+++ b/src/hotspot/share/gc/shenandoah/vmStructs_shenandoah.hpp
@@ -29,21 +29,22 @@
#include "gc/shenandoah/shenandoahHeap.hpp"
#include "gc/shenandoah/shenandoahHeapRegion.hpp"
#include "gc/shenandoah/shenandoahMonitoringSupport.hpp"
+#include "runtime/atomic.hpp"
-#define VM_STRUCTS_SHENANDOAH(nonstatic_field, volatile_nonstatic_field, static_field) \
- nonstatic_field(ShenandoahHeap, _num_regions, size_t) \
- nonstatic_field(ShenandoahHeap, _regions, ShenandoahHeapRegion**) \
- nonstatic_field(ShenandoahHeap, _log_min_obj_alignment_in_bytes, int) \
- nonstatic_field(ShenandoahHeap, _free_set, ShenandoahFreeSet*) \
- volatile_nonstatic_field(ShenandoahHeap, _committed, size_t) \
- static_field(ShenandoahHeapRegion, RegionSizeBytes, size_t) \
- static_field(ShenandoahHeapRegion, RegionSizeBytesShift, size_t) \
- volatile_nonstatic_field(ShenandoahHeapRegion, _state, ShenandoahHeapRegion::RegionState) \
- nonstatic_field(ShenandoahHeapRegion, _index, size_t const) \
- nonstatic_field(ShenandoahHeapRegion, _bottom, HeapWord* const) \
- nonstatic_field(ShenandoahHeapRegion, _top, HeapWord*) \
- nonstatic_field(ShenandoahHeapRegion, _end, HeapWord* const) \
- nonstatic_field(ShenandoahFreeSet, _total_global_used, size_t) \
+#define VM_STRUCTS_SHENANDOAH(nonstatic_field, volatile_nonstatic_field, static_field) \
+ nonstatic_field(ShenandoahHeap, _num_regions, size_t) \
+ nonstatic_field(ShenandoahHeap, _regions, ShenandoahHeapRegion**) \
+ nonstatic_field(ShenandoahHeap, _log_min_obj_alignment_in_bytes, int) \
+ nonstatic_field(ShenandoahHeap, _free_set, ShenandoahFreeSet*) \
+ volatile_nonstatic_field(ShenandoahHeap, _committed, size_t) \
+ static_field(ShenandoahHeapRegion, RegionSizeBytes, size_t) \
+ static_field(ShenandoahHeapRegion, RegionSizeBytesShift, size_t) \
+ nonstatic_field(ShenandoahHeapRegion, _state, Atomic) \
+ nonstatic_field(ShenandoahHeapRegion, _index, size_t const) \
+ nonstatic_field(ShenandoahHeapRegion, _bottom, HeapWord* const) \
+ nonstatic_field(ShenandoahHeapRegion, _top, HeapWord*) \
+ nonstatic_field(ShenandoahHeapRegion, _end, HeapWord* const) \
+ nonstatic_field(ShenandoahFreeSet, _total_global_used, size_t) \
#define VM_INT_CONSTANTS_SHENANDOAH(declare_constant, declare_constant_with_value) \
declare_constant(ShenandoahHeapRegion::_empty_uncommitted) \
@@ -65,7 +66,7 @@
declare_toplevel_type(ShenandoahHeapRegion) \
declare_toplevel_type(ShenandoahHeap*) \
declare_toplevel_type(ShenandoahHeapRegion*) \
- declare_toplevel_type(ShenandoahHeapRegion::RegionState) \
+ declare_toplevel_type(Atomic) \
declare_toplevel_type(ShenandoahFreeSet) \
declare_toplevel_type(ShenandoahFreeSet*) \
diff --git a/src/hotspot/share/gc/z/zAddress.inline.hpp b/src/hotspot/share/gc/z/zAddress.inline.hpp
index c8c8ec7ae3a..0b99802729b 100644
--- a/src/hotspot/share/gc/z/zAddress.inline.hpp
+++ b/src/hotspot/share/gc/z/zAddress.inline.hpp
@@ -199,18 +199,18 @@ CREATE_ZOFFSET_OPERATORS(zoffset)
inline uintptr_t untype(zbacking_offset offset) {
const uintptr_t value = static_cast(offset);
- assert(value < ZBackingOffsetMax, "Offset out of bounds (" PTR_FORMAT " < " PTR_FORMAT ")", value, ZAddressOffsetMax);
+ assert(value < ZBackingOffsetMax, "Offset out of bounds (" PTR_FORMAT " < " PTR_FORMAT ")", value, ZBackingOffsetMax);
return value;
}
inline uintptr_t untype(zbacking_offset_end offset) {
const uintptr_t value = static_cast(offset);
- assert(value <= ZBackingOffsetMax, "Offset out of bounds (" PTR_FORMAT " <= " PTR_FORMAT ")", value, ZAddressOffsetMax);
+ assert(value <= ZBackingOffsetMax, "Offset out of bounds (" PTR_FORMAT " <= " PTR_FORMAT ")", value, ZBackingOffsetMax);
return value;
}
inline zbacking_offset to_zbacking_offset(uintptr_t value) {
- assert(value < ZBackingOffsetMax, "Value out of bounds (" PTR_FORMAT " < " PTR_FORMAT ")", value, ZAddressOffsetMax);
+ assert(value < ZBackingOffsetMax, "Value out of bounds (" PTR_FORMAT " < " PTR_FORMAT ")", value, ZBackingOffsetMax);
return zbacking_offset(value);
}
@@ -227,7 +227,7 @@ inline zbacking_offset_end to_zbacking_offset_end(zbacking_offset start, size_t
}
inline zbacking_offset_end to_zbacking_offset_end(uintptr_t value) {
- assert(value <= ZBackingOffsetMax, "Value out of bounds (" PTR_FORMAT " <= " PTR_FORMAT ")", value, ZAddressOffsetMax);
+ assert(value <= ZBackingOffsetMax, "Value out of bounds (" PTR_FORMAT " <= " PTR_FORMAT ")", value, ZBackingOffsetMax);
return zbacking_offset_end(value);
}
diff --git a/src/hotspot/share/gc/z/zMarkTerminate.hpp b/src/hotspot/share/gc/z/zMarkTerminate.hpp
index cff1f8e73fa..9def3e72120 100644
--- a/src/hotspot/share/gc/z/zMarkTerminate.hpp
+++ b/src/hotspot/share/gc/z/zMarkTerminate.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017, 2023, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2017, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -25,6 +25,7 @@
#define SHARE_GC_Z_ZMARKTERMINATE_HPP
#include "gc/z/zLock.hpp"
+#include "runtime/atomic.hpp"
#include "utilities/globalDefinitions.hpp"
class ZMarkStripeSet;
@@ -32,9 +33,9 @@ class ZMarkStripeSet;
class ZMarkTerminate {
private:
uint _nworkers;
- volatile uint _nworking;
- volatile uint _nawakening;
- volatile bool _resurrected;
+ Atomic _nworking;
+ Atomic _nawakening;
+ Atomic _resurrected;
ZConditionLock _lock;
void maybe_reduce_stripes(ZMarkStripeSet* stripes, size_t used_nstripes);
diff --git a/src/hotspot/share/gc/z/zMarkTerminate.inline.hpp b/src/hotspot/share/gc/z/zMarkTerminate.inline.hpp
index 575044e3a39..84a545623bc 100644
--- a/src/hotspot/share/gc/z/zMarkTerminate.inline.hpp
+++ b/src/hotspot/share/gc/z/zMarkTerminate.inline.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2017, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -30,7 +30,6 @@
#include "gc/z/zLock.inline.hpp"
#include "gc/z/zMarkStack.hpp"
#include "logging/log.hpp"
-#include "runtime/atomicAccess.hpp"
#include "runtime/osThread.hpp"
#include "runtime/thread.inline.hpp"
@@ -42,24 +41,23 @@ inline ZMarkTerminate::ZMarkTerminate()
_lock() {}
inline void ZMarkTerminate::reset(uint nworkers) {
- AtomicAccess::store(&_nworkers, nworkers);
- AtomicAccess::store(&_nworking, nworkers);
- _nawakening = 0;
+ _nworkers = nworkers;
+ _nworking.store_relaxed(nworkers);
+ _nawakening.store_relaxed(0u);
}
inline void ZMarkTerminate::leave() {
SuspendibleThreadSetLeaver sts_leaver;
ZLocker locker(&_lock);
- AtomicAccess::store(&_nworking, _nworking - 1);
- if (_nworking == 0) {
+ if (_nworking.sub_then_fetch(1u, memory_order_relaxed) == 0) {
// Last thread leaving; notify waiters
_lock.notify_all();
}
}
inline void ZMarkTerminate::maybe_reduce_stripes(ZMarkStripeSet* stripes, size_t used_nstripes) {
- size_t nstripes = stripes->nstripes();
+ const size_t nstripes = stripes->nstripes();
if (used_nstripes == nstripes && nstripes > 1u) {
stripes->try_set_nstripes(nstripes, nstripes >> 1);
}
@@ -69,8 +67,7 @@ inline bool ZMarkTerminate::try_terminate(ZMarkStripeSet* stripes, size_t used_n
SuspendibleThreadSetLeaver sts_leaver;
ZLocker locker(&_lock);
- AtomicAccess::store(&_nworking, _nworking - 1);
- if (_nworking == 0) {
+ if (_nworking.sub_then_fetch(1u, memory_order_relaxed) == 0) {
// Last thread entering termination: success
_lock.notify_all();
return true;
@@ -83,24 +80,24 @@ inline bool ZMarkTerminate::try_terminate(ZMarkStripeSet* stripes, size_t used_n
// We either got notification about more work
// or got a spurious wakeup; don't terminate
- if (_nawakening > 0) {
- AtomicAccess::store(&_nawakening, _nawakening - 1);
+ if (_nawakening.load_relaxed() > 0) {
+ _nawakening.sub_then_fetch(1u, memory_order_relaxed);
}
- if (_nworking == 0) {
+ if (_nworking.load_relaxed() == 0) {
// We got notified all work is done; terminate
return true;
}
- AtomicAccess::store(&_nworking, _nworking + 1);
+ _nworking.add_then_fetch(1u, memory_order_relaxed);
return false;
}
inline void ZMarkTerminate::wake_up() {
- uint nworking = AtomicAccess::load(&_nworking);
- uint nawakening = AtomicAccess::load(&_nawakening);
- if (nworking + nawakening == AtomicAccess::load(&_nworkers)) {
+ const uint nworking = _nworking.load_relaxed();
+ const uint nawakening = _nawakening.load_relaxed();
+ if (nworking + nawakening == _nworkers) {
// Everyone is working or about to
return;
}
@@ -111,24 +108,24 @@ inline void ZMarkTerminate::wake_up() {
}
ZLocker locker(&_lock);
- if (_nworking + _nawakening != _nworkers) {
+ if (_nworking.load_relaxed() + _nawakening.load_relaxed() != _nworkers) {
// Everyone is not working
- AtomicAccess::store(&_nawakening, _nawakening + 1);
+ _nawakening.add_then_fetch(1u, memory_order_relaxed);
_lock.notify();
}
}
inline bool ZMarkTerminate::saturated() const {
- uint nworking = AtomicAccess::load(&_nworking);
- uint nawakening = AtomicAccess::load(&_nawakening);
+ const uint nworking = _nworking.load_relaxed();
+ const uint nawakening = _nawakening.load_relaxed();
- return nworking + nawakening == AtomicAccess::load(&_nworkers);
+ return nworking + nawakening == _nworkers;
}
inline void ZMarkTerminate::set_resurrected(bool value) {
// Update resurrected if it changed
if (resurrected() != value) {
- AtomicAccess::store(&_resurrected, value);
+ _resurrected.store_relaxed(value);
if (value) {
log_debug(gc, marking)("Resurrection broke termination");
} else {
@@ -138,7 +135,7 @@ inline void ZMarkTerminate::set_resurrected(bool value) {
}
inline bool ZMarkTerminate::resurrected() const {
- return AtomicAccess::load(&_resurrected);
+ return _resurrected.load_relaxed();
}
#endif // SHARE_GC_Z_ZMARKTERMINATE_INLINE_HPP
diff --git a/src/hotspot/share/gc/z/zNMethodTableIteration.cpp b/src/hotspot/share/gc/z/zNMethodTableIteration.cpp
index bdd4270ddac..423e98dd28b 100644
--- a/src/hotspot/share/gc/z/zNMethodTableIteration.cpp
+++ b/src/hotspot/share/gc/z/zNMethodTableIteration.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2017, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -24,7 +24,6 @@
#include "gc/z/zNMethodTableEntry.hpp"
#include "gc/z/zNMethodTableIteration.hpp"
#include "memory/iterator.hpp"
-#include "runtime/atomicAccess.hpp"
#include "utilities/debug.hpp"
#include "utilities/globalDefinitions.hpp"
@@ -42,11 +41,11 @@ void ZNMethodTableIteration::nmethods_do_begin(ZNMethodTableEntry* table, size_t
_table = table;
_size = size;
- _claimed = 0;
+ _claimed.store_relaxed(0u);
}
void ZNMethodTableIteration::nmethods_do_end() {
- assert(_claimed >= _size, "Failed to claim all table entries");
+ assert(_claimed.load_relaxed() >= _size, "Failed to claim all table entries");
// Finish iteration
_table = nullptr;
@@ -57,7 +56,7 @@ void ZNMethodTableIteration::nmethods_do(NMethodClosure* cl) {
// Claim table partition. Each partition is currently sized to span
// two cache lines. This number is just a guess, but seems to work well.
const size_t partition_size = (ZCacheLineSize * 2) / sizeof(ZNMethodTableEntry);
- const size_t partition_start = MIN2(AtomicAccess::fetch_then_add(&_claimed, partition_size), _size);
+ const size_t partition_start = MIN2(_claimed.fetch_then_add(partition_size), _size);
const size_t partition_end = MIN2(partition_start + partition_size, _size);
if (partition_start == partition_end) {
// End of table
diff --git a/src/hotspot/share/gc/z/zNMethodTableIteration.hpp b/src/hotspot/share/gc/z/zNMethodTableIteration.hpp
index 064c3eafca0..12dc096491d 100644
--- a/src/hotspot/share/gc/z/zNMethodTableIteration.hpp
+++ b/src/hotspot/share/gc/z/zNMethodTableIteration.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2019, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2019, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -25,15 +25,16 @@
#define SHARE_GC_Z_ZNMETHODTABLEITERATION_HPP
#include "gc/z/zGlobals.hpp"
+#include "runtime/atomic.hpp"
class NMethodClosure;
class ZNMethodTableEntry;
class ZNMethodTableIteration {
private:
- ZNMethodTableEntry* _table;
- size_t _size;
- ZCACHE_ALIGNED volatile size_t _claimed;
+ ZNMethodTableEntry* _table;
+ size_t _size;
+ ZCACHE_ALIGNED Atomic _claimed;
bool in_progress() const;
diff --git a/src/hotspot/share/gc/z/zRelocate.cpp b/src/hotspot/share/gc/z/zRelocate.cpp
index da07f67d859..d51cf5abbae 100644
--- a/src/hotspot/share/gc/z/zRelocate.cpp
+++ b/src/hotspot/share/gc/z/zRelocate.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2015, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2015, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -64,31 +64,31 @@ ZRelocateQueue::ZRelocateQueue()
_needs_attention(0) {}
bool ZRelocateQueue::needs_attention() const {
- return AtomicAccess::load(&_needs_attention) != 0;
+ return _needs_attention.load_relaxed() != 0;
}
void ZRelocateQueue::inc_needs_attention() {
- const int needs_attention = AtomicAccess::add(&_needs_attention, 1);
+ const int needs_attention = _needs_attention.add_then_fetch(1);
assert(needs_attention == 1 || needs_attention == 2, "Invalid state");
}
void ZRelocateQueue::dec_needs_attention() {
- const int needs_attention = AtomicAccess::sub(&_needs_attention, 1);
+ const int needs_attention = _needs_attention.sub_then_fetch(1);
assert(needs_attention == 0 || needs_attention == 1, "Invalid state");
}
void ZRelocateQueue::activate(uint nworkers) {
- _is_active = true;
+ _is_active.store_relaxed(true);
join(nworkers);
}
void ZRelocateQueue::deactivate() {
- AtomicAccess::store(&_is_active, false);
+ _is_active.store_relaxed(false);
clear();
}
bool ZRelocateQueue::is_active() const {
- return AtomicAccess::load(&_is_active);
+ return _is_active.load_relaxed();
}
void ZRelocateQueue::join(uint nworkers) {
@@ -453,7 +453,7 @@ static void retire_target_page(ZGeneration* generation, ZPage* page) {
class ZRelocateSmallAllocator {
private:
ZGeneration* const _generation;
- volatile size_t _in_place_count;
+ Atomic _in_place_count;
public:
ZRelocateSmallAllocator(ZGeneration* generation)
@@ -463,7 +463,7 @@ public:
ZPage* alloc_and_retire_target_page(ZForwarding* forwarding, ZPage* target) {
ZPage* const page = alloc_page(forwarding);
if (page == nullptr) {
- AtomicAccess::inc(&_in_place_count);
+ _in_place_count.add_then_fetch(1u);
}
if (target != nullptr) {
@@ -493,7 +493,7 @@ public:
}
size_t in_place_count() const {
- return _in_place_count;
+ return _in_place_count.load_relaxed();
}
};
@@ -503,7 +503,7 @@ private:
ZConditionLock _lock;
ZRelocationTargets* _shared_targets;
bool _in_place;
- volatile size_t _in_place_count;
+ Atomic _in_place_count;
public:
ZRelocateMediumAllocator(ZGeneration* generation, ZRelocationTargets* shared_targets)
@@ -539,7 +539,7 @@ public:
ZPage* const to_page = alloc_page(forwarding);
_shared_targets->set(partition_id, to_age, to_page);
if (to_page == nullptr) {
- AtomicAccess::inc(&_in_place_count);
+ _in_place_count.add_then_fetch(1u);
_in_place = true;
}
@@ -579,7 +579,7 @@ public:
}
size_t in_place_count() const {
- return _in_place_count;
+ return _in_place_count.load_relaxed();
}
};
diff --git a/src/hotspot/share/gc/z/zRelocate.hpp b/src/hotspot/share/gc/z/zRelocate.hpp
index 038efba83eb..d25536e1bce 100644
--- a/src/hotspot/share/gc/z/zRelocate.hpp
+++ b/src/hotspot/share/gc/z/zRelocate.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2015, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2015, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -28,6 +28,7 @@
#include "gc/z/zPageAge.hpp"
#include "gc/z/zRelocationSet.hpp"
#include "gc/z/zValue.hpp"
+#include "runtime/atomic.hpp"
class ZForwarding;
class ZGeneration;
@@ -42,8 +43,8 @@ private:
uint _nworkers;
uint _nsynchronized;
bool _synchronize;
- volatile bool _is_active;
- volatile int _needs_attention;
+ Atomic _is_active;
+ Atomic _needs_attention;
bool needs_attention() const;
void inc_needs_attention();
diff --git a/src/hotspot/share/gc/z/zTLABUsage.cpp b/src/hotspot/share/gc/z/zTLABUsage.cpp
index c2e3db6bedf..d3613cf8632 100644
--- a/src/hotspot/share/gc/z/zTLABUsage.cpp
+++ b/src/hotspot/share/gc/z/zTLABUsage.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2025, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -23,24 +23,23 @@
#include "gc/z/zTLABUsage.hpp"
#include "logging/log.hpp"
-#include "runtime/atomicAccess.hpp"
ZTLABUsage::ZTLABUsage()
: _used(0),
_used_history() {}
void ZTLABUsage::increase_used(size_t size) {
- AtomicAccess::add(&_used, size, memory_order_relaxed);
+ _used.add_then_fetch(size, memory_order_relaxed);
}
void ZTLABUsage::decrease_used(size_t size) {
- precond(size <= _used);
+ precond(size <= _used.load_relaxed());
- AtomicAccess::sub(&_used, size, memory_order_relaxed);
+ _used.sub_then_fetch(size, memory_order_relaxed);
}
void ZTLABUsage::reset() {
- const size_t used = AtomicAccess::xchg(&_used, (size_t) 0);
+ const size_t used = _used.exchange(0u);
// Avoid updates when nothing has been allocated since the last YC
if (used == 0) {
diff --git a/src/hotspot/share/gc/z/zTLABUsage.hpp b/src/hotspot/share/gc/z/zTLABUsage.hpp
index 3d1b084fe16..68c7620c26e 100644
--- a/src/hotspot/share/gc/z/zTLABUsage.hpp
+++ b/src/hotspot/share/gc/z/zTLABUsage.hpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2025, 2026, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
@@ -24,6 +24,7 @@
#ifndef SHARE_GC_Z_ZTLABUSAGE_HPP
#define SHARE_GC_Z_ZTLABUSAGE_HPP
+#include "runtime/atomic.hpp"
#include "utilities/globalDefinitions.hpp"
#include "utilities/numberSeq.hpp"
@@ -42,9 +43,9 @@
class ZTLABUsage {
private:
// Accounting TLAB used until the next GC cycle
- volatile size_t _used;
+ Atomic