jdk/test/hotspot/jtreg/compiler/vectorization/TestVectorAlgorithms.java
2025-12-15 13:36:25 +01:00

486 lines
20 KiB
Java

/*
* Copyright (c) 2025, 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.
*/
/*
* @test id=vanilla
* @bug 8373026
* @summary Test auto vectorization and Vector API with some vector
* algorithms. Related benchmark: VectorAlgorithms.java
* @library /test/lib /
* @modules jdk.incubator.vector
* @run driver ${test.main.class} vanilla
*/
/*
* @test id=noSuperWord
* @bug 8373026
* @library /test/lib /
* @modules jdk.incubator.vector
* @run driver ${test.main.class} noSuperWord
*/
/*
* @test id=noOptimizeFill
* @bug 8373026
* @library /test/lib /
* @modules jdk.incubator.vector
* @run driver ${test.main.class} noOptimizeFill
*/
package compiler.vectorization;
import java.util.Map;
import java.util.HashMap;
import jdk.test.lib.Utils;
import java.util.Random;
import java.lang.foreign.*;
import compiler.lib.ir_framework.*;
import compiler.lib.generators.*;
import static compiler.lib.generators.Generators.G;
import compiler.lib.verify.*;
/**
* The goal of this benchmark is to show the power of auto vectorization
* and the Vector API.
*
* Please only modify this benchark in synchronization with the JMH benchmark:
* micro/org/openjdk/bench/vm/compiler/VectorAlgorithms.java
*/
public class TestVectorAlgorithms {
private static final Random RANDOM = Utils.getRandomInstance();
private static final RestrictableGenerator<Integer> INT_GEN = Generators.G.ints();
interface TestFunction {
Object run();
}
Map<String, Map<String, TestFunction>> testGroups = new HashMap<String, Map<String, TestFunction>>();
int[] aI;
int[] rI1;
int[] rI2;
int[] rI3;
int[] rI4;
int eI;
int[] oopsX4;
int[] memX4;
public static void main(String[] args) {
TestFramework framework = new TestFramework();
framework.addFlags("--add-modules=jdk.incubator.vector", "-XX:CompileCommand=inline,*VectorAlgorithmsImpl::*");
switch (args[0]) {
case "vanilla" -> { /* no extra flags */ }
case "noSuperWord" -> { framework.addFlags("-XX:-UseSuperWord"); }
case "noOptimizeFill" -> { framework.addFlags("-XX:-OptimizeFill"); }
default -> { throw new RuntimeException("Test argument not recognized: " + args[0]); }
}
framework.start();
}
public TestVectorAlgorithms () {
// IMPORTANT:
// If you want to use some array but do NOT modify it: just use it.
// If you want to use it and DO want to modify it: clone it. This
// ensures that each test gets a separate copy, and that when we
// capture the modified arrays they are different for every method
// and run.
// An alternative to cloning is to use different return arrays for
// different implementations of the same group, e.g. rI1, rI2, ...
testGroups.put("fillI", new HashMap<String,TestFunction>());
testGroups.get("fillI").put("fillI_loop", () -> { return fillI_loop(rI1); });
testGroups.get("fillI").put("fillI_VectorAPI", () -> { return fillI_VectorAPI(rI1); });
testGroups.get("fillI").put("fillI_Arrays", () -> { return fillI_Arrays(rI1); });
testGroups.put("iotaI", new HashMap<String,TestFunction>());
testGroups.get("iotaI").put("iotaI_loop", () -> { return iotaI_loop(rI1); });
testGroups.get("iotaI").put("iotaI_VectorAPI", () -> { return iotaI_VectorAPI(rI1); });
testGroups.put("copyI", new HashMap<String,TestFunction>());
testGroups.get("copyI").put("copyI_loop", () -> { return copyI_loop(aI, rI1); });
testGroups.get("copyI").put("copyI_VectorAPI", () -> { return copyI_VectorAPI(aI, rI1); });
testGroups.get("copyI").put("copyI_System_arraycopy", () -> { return copyI_System_arraycopy(aI, rI1); });
testGroups.put("mapI", new HashMap<String,TestFunction>());
testGroups.get("mapI").put("mapI_loop", () -> { return mapI_loop(aI, rI1); });
testGroups.get("mapI").put("mapI_VectorAPI", () -> { return mapI_VectorAPI(aI, rI1); });
testGroups.put("reduceAddI", new HashMap<String,TestFunction>());
testGroups.get("reduceAddI").put("reduceAddI_loop", () -> { return reduceAddI_loop(aI); });
testGroups.get("reduceAddI").put("reduceAddI_reassociate", () -> { return reduceAddI_reassociate(aI); });
testGroups.get("reduceAddI").put("reduceAddI_VectorAPI_naive", () -> { return reduceAddI_VectorAPI_naive(aI); });
testGroups.get("reduceAddI").put("reduceAddI_VectorAPI_reduction_after_loop", () -> { return reduceAddI_VectorAPI_reduction_after_loop(aI); });
testGroups.put("scanAddI", new HashMap<String,TestFunction>());
testGroups.get("scanAddI").put("scanAddI_loop", () -> { return scanAddI_loop(aI, rI1); });
testGroups.get("scanAddI").put("scanAddI_loop_reassociate", () -> { return scanAddI_loop_reassociate(aI, rI2); });
testGroups.get("scanAddI").put("scanAddI_VectorAPI_permute_add", () -> { return scanAddI_VectorAPI_permute_add(aI, rI4); });
testGroups.put("findMinIndexI", new HashMap<String,TestFunction>());
testGroups.get("findMinIndexI").put("findMinIndexI_loop", () -> { return findMinIndexI_loop(aI); });
testGroups.get("findMinIndexI").put("findMinIndexI_VectorAPI", () -> { return findMinIndexI_VectorAPI(aI); });
testGroups.put("findI", new HashMap<String,TestFunction>());
testGroups.get("findI").put("findI_loop", () -> { return findI_loop(aI, eI); });
testGroups.get("findI").put("findI_VectorAPI", () -> { return findI_VectorAPI(aI, eI); });
testGroups.put("reverseI", new HashMap<String,TestFunction>());
testGroups.get("reverseI").put("reverseI_loop", () -> { return reverseI_loop(aI, rI1); });
testGroups.get("reverseI").put("reverseI_VectorAPI", () -> { return reverseI_VectorAPI(aI, rI2); });
testGroups.put("filterI", new HashMap<String,TestFunction>());
testGroups.get("filterI").put("filterI_loop", () -> { return filterI_loop(aI, rI1, eI); });
testGroups.get("filterI").put("filterI_VectorAPI", () -> { return filterI_VectorAPI(aI, rI2, eI); });
testGroups.put("reduceAddIFieldsX4", new HashMap<String,TestFunction>());
testGroups.get("reduceAddIFieldsX4").put("reduceAddIFieldsX4_loop", () -> { return reduceAddIFieldsX4_loop(oopsX4, memX4); });
testGroups.get("reduceAddIFieldsX4").put("reduceAddIFieldsX4_VectorAPI", () -> { return reduceAddIFieldsX4_VectorAPI(oopsX4, memX4); });
}
@Warmup(100)
@Run(test = {"fillI_loop",
"fillI_VectorAPI",
"fillI_Arrays",
"iotaI_loop",
"iotaI_VectorAPI",
"copyI_loop",
"copyI_VectorAPI",
"copyI_System_arraycopy",
"mapI_loop",
"mapI_VectorAPI",
"reduceAddI_loop",
"reduceAddI_reassociate",
"reduceAddI_VectorAPI_naive",
"reduceAddI_VectorAPI_reduction_after_loop",
"scanAddI_loop",
"scanAddI_loop_reassociate",
"scanAddI_VectorAPI_permute_add",
"findMinIndexI_loop",
"findMinIndexI_VectorAPI",
"findI_loop",
"findI_VectorAPI",
"reverseI_loop",
"reverseI_VectorAPI",
"filterI_loop",
"filterI_VectorAPI",
"reduceAddIFieldsX4_loop",
"reduceAddIFieldsX4_VectorAPI"})
public void runTests(RunInfo info) {
// Repeat many times, so that we also have multiple iterations for post-warmup to potentially recompile
int iters = info.isWarmUp() ? 1 : 20;
for (int iter = 0; iter < iters; iter++) {
// Set up random inputs, random size is important to stress tails.
int size = 100_000 + RANDOM.nextInt(10_000);
aI = new int[size];
G.fill(INT_GEN, aI);
// Pick some random element. Most of the time it is in aI, sometimes not.
eI = (RANDOM.nextInt(10) == 0) ? RANDOM.nextInt() : aI[RANDOM.nextInt(size)];
//for (int i = 0; i < aI.length; i++) { aI[i] = i; }
rI1 = new int[size];
rI2 = new int[size];
rI3 = new int[size];
rI4 = new int[size];
// X4 oop setup.
oopsX4 = new int[size];
int numX4 = 10_000;
for (int i = 0; i < size; i++) {
// assign either a zero=null, or assign a random oop.
oopsX4[i] = (RANDOM.nextInt(10) == 0) ? 0 : RANDOM.nextInt(numX4) * 4;
}
// Just fill the whole array with random values.
// The relevant field is only at every "4 * i + 3" though.
memX4 = new int[4 * numX4];
for (int i = 0; i < 4 * numX4; i++) {
memX4[i] = RANDOM.nextInt();
}
// Run all tests
for (Map.Entry<String, Map<String,TestFunction>> group_entry : testGroups.entrySet()) {
String group_name = group_entry.getKey();
Map<String, TestFunction> group = group_entry.getValue();
Object gold = null;
String gold_name = "NONE";
for (Map.Entry<String,TestFunction> entry : group.entrySet()) {
String name = entry.getKey();
TestFunction test = entry.getValue();
Object result = test.run();
if (gold == null) {
gold = result;
gold_name = name;
} else {
try {
Verify.checkEQ(gold, result);
} catch (VerifyException e) {
throw new RuntimeException("Verify.checkEQ failed for group " + group_name +
", gold " + gold_name + ", test " + name, e);
}
}
}
}
}
}
@Test
@IR(counts = {IRNode.REPLICATE_I, "= 1",
IRNode.STORE_VECTOR, "> 0"},
applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"},
applyIfAnd = {"UseSuperWord", "true", "OptimizeFill", "false"})
@IR(counts = {".*CallLeafNoFP.*arrayof_jint_fill.*", "= 1"},
phase = CompilePhase.BEFORE_MATCHING,
applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"},
applyIf = {"OptimizeFill", "true"})
// By default, the fill intrinsic "arrayof_jint_fill" is used, but we can disable
// the detection of the fill loop, and then we auto vectorize.
public Object fillI_loop(int[] r) {
return VectorAlgorithmsImpl.fillI_loop(r);
}
@Test
@IR(counts = {IRNode.REPLICATE_I, "= 1",
IRNode.STORE_VECTOR, "> 0"},
applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"})
public Object fillI_VectorAPI(int[] r) {
return VectorAlgorithmsImpl.fillI_VectorAPI(r);
}
@Test
// Arrays.fill is not necessarily inlined, so we can't check
// for vectors in the IR.
public Object fillI_Arrays(int[] r) {
return VectorAlgorithmsImpl.fillI_Arrays(r);
}
@Test
@IR(counts = {IRNode.POPULATE_INDEX, "> 0",
IRNode.STORE_VECTOR, "> 0"},
applyIfCPUFeatureOr = {"avx2", "true", "sve", "true"},
applyIf = {"UseSuperWord", "true"})
// Note: the Vector API example below can also vectorize for AVX,
// because it does not use a PopulateIndex.
public Object iotaI_loop(int[] r) {
return VectorAlgorithmsImpl.iotaI_loop(r);
}
@Test
@IR(counts = {IRNode.ADD_VI, "> 0",
IRNode.STORE_VECTOR, "> 0"},
applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"})
public Object iotaI_VectorAPI(int[] r) {
return VectorAlgorithmsImpl.iotaI_VectorAPI(r);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "> 0",
IRNode.STORE_VECTOR, "> 0"},
applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"},
applyIf = {"UseSuperWord", "true"})
public Object copyI_loop(int[] a, int[] r) {
return VectorAlgorithmsImpl.copyI_loop(a, r);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "> 0",
IRNode.STORE_VECTOR, "> 0"},
applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"})
public Object copyI_VectorAPI(int[] a, int[] r) {
return VectorAlgorithmsImpl.copyI_VectorAPI(a, r);
}
@Test
@IR(counts = {".*CallLeafNoFP.*arrayof_jint_disjoint_arraycopy.*", "= 1"},
phase = CompilePhase.BEFORE_MATCHING,
applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"})
public Object copyI_System_arraycopy(int[] a, int[] r) {
return VectorAlgorithmsImpl.copyI_System_arraycopy(a, r);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "> 0",
IRNode.MUL_VI, "> 0",
IRNode.STORE_VECTOR, "> 0"},
applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"},
applyIf = {"UseSuperWord", "true"})
public Object mapI_loop(int[] a, int[] r) {
return VectorAlgorithmsImpl.mapI_loop(a, r);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "> 0",
IRNode.MUL_VI, "> 0",
IRNode.STORE_VECTOR, "> 0"},
applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"})
public Object mapI_VectorAPI(int[] a, int[] r) {
return VectorAlgorithmsImpl.mapI_VectorAPI(a, r);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "> 0",
IRNode.ADD_REDUCTION_VI, "> 0",
IRNode.ADD_VI, "> 0"},
applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"},
applyIf = {"UseSuperWord", "true"})
public int reduceAddI_loop(int[] a) {
return VectorAlgorithmsImpl.reduceAddI_loop(a);
}
@Test
public int reduceAddI_reassociate(int[] a) {
return VectorAlgorithmsImpl.reduceAddI_reassociate(a);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "> 0",
IRNode.ADD_REDUCTION_VI, "> 0"}, // reduceLanes inside loop
applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"})
public int reduceAddI_VectorAPI_naive(int[] a) {
return VectorAlgorithmsImpl.reduceAddI_VectorAPI_naive(aI);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "> 0",
IRNode.ADD_REDUCTION_VI, "> 0",
IRNode.ADD_VI, "> 0"},
applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"})
public int reduceAddI_VectorAPI_reduction_after_loop(int[] a) {
return VectorAlgorithmsImpl.reduceAddI_VectorAPI_reduction_after_loop(aI);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "= 0",
IRNode.STORE_VECTOR, "= 0"})
// Currently does not vectorize, but might in the future.
public Object scanAddI_loop(int[] a, int[] r) {
return VectorAlgorithmsImpl.scanAddI_loop(a, r);
}
@Test
public Object scanAddI_loop_reassociate(int[] a, int[] r) {
return VectorAlgorithmsImpl.scanAddI_loop_reassociate(a, r);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "> 0",
IRNode.REARRANGE_VI, "> 0",
IRNode.AND_VI, "> 0",
IRNode.ADD_VI, "> 0",
IRNode.STORE_VECTOR, "> 0"},
applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"},
applyIf = {"MaxVectorSize", ">=64"})
public Object scanAddI_VectorAPI_permute_add(int[] a, int[] r) {
return VectorAlgorithmsImpl.scanAddI_VectorAPI_permute_add(a, r);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "= 0"})
// Currently does not vectorize, but might in the future.
public int findMinIndexI_loop(int[] a) {
return VectorAlgorithmsImpl.findMinIndexI_loop(a);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "> 0",
IRNode.VECTOR_MASK_CMP, "> 0",
IRNode.VECTOR_BLEND_I, "> 0",
IRNode.MIN_REDUCTION_V, "> 0",
IRNode.ADD_VI, "> 0"},
applyIfCPUFeatureOr = {"avx", "true", "asimd", "true"})
public int findMinIndexI_VectorAPI(int[] a) {
return VectorAlgorithmsImpl.findMinIndexI_VectorAPI(a);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "= 0"})
// Currently does not vectorize, but might in the future.
public int findI_loop(int[] a, int e) {
return VectorAlgorithmsImpl.findI_loop(a, e);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "> 0",
IRNode.VECTOR_MASK_CMP, "> 0",
IRNode.VECTOR_TEST, "> 0"},
applyIfCPUFeatureOr = {"avx", "true", "asimd", "true"})
public int findI_VectorAPI(int[] a, int e) {
return VectorAlgorithmsImpl.findI_VectorAPI(a, e);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "= 0",
IRNode.STORE_VECTOR, "= 0"})
// Currently does not vectorize, but might in the future.
public Object reverseI_loop(int[] a, int[] r) {
return VectorAlgorithmsImpl.reverseI_loop(a, r);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "> 0",
IRNode.REARRANGE_VI, "> 0",
IRNode.AND_VI, "> 0",
IRNode.STORE_VECTOR, "> 0"},
applyIfCPUFeatureOr = {"sse4.1", "true", "asimd", "true"})
public Object reverseI_VectorAPI(int[] a, int[] r) {
return VectorAlgorithmsImpl.reverseI_VectorAPI(a, r);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "= 0",
IRNode.STORE_VECTOR, "= 0"})
public Object filterI_loop(int[] a, int[] r, int threshold) {
return VectorAlgorithmsImpl.filterI_loop(a, r, threshold);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "> 0",
IRNode.VECTOR_MASK_CMP, "> 0",
IRNode.VECTOR_TEST, "> 0",
IRNode.VECTOR_LONG_TO_MASK, "> 0",
IRNode.STORE_VECTOR_MASKED, "> 0"},
applyIfCPUFeatureOr = {"avx", "true", "asimd", "true"})
public Object filterI_VectorAPI(int[] a, int[] r, int threshold) {
return VectorAlgorithmsImpl.filterI_VectorAPI(a, r, threshold);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "= 0"})
// Currently does not vectorize, but might in the future.
public int reduceAddIFieldsX4_loop(int[] oops, int[] mem) {
return VectorAlgorithmsImpl.reduceAddIFieldsX4_loop(oops, mem);
}
@Test
@IR(counts = {IRNode.LOAD_VECTOR_I, "> 0",
IRNode.VECTOR_MASK_CMP, "> 0",
IRNode.VECTOR_TEST, "> 0",
IRNode.LOAD_VECTOR_GATHER_MASKED, "> 0",
IRNode.OR_V_MASK, "> 0",
IRNode.ADD_VI, "> 0",
IRNode.ADD_REDUCTION_VI, "> 0"},
applyIfCPUFeatureOr = {"avx512", "true", "sve", "true"})
public int reduceAddIFieldsX4_VectorAPI(int[] oops, int[] mem) {
return VectorAlgorithmsImpl.reduceAddIFieldsX4_VectorAPI(oops, mem);
}
}