8331934: [s390x] Add support for primitive array C1 clone intrinsic

Reviewed-by: mdoerr, sjayagond

src/hotspot/cpu/s390/c1_LIRAssembler_s390.cpp

@@ -2274,7 +2274,9 @@ void LIR_Assembler::emit_arraycopy(LIR_OpArrayCopy* op) {
   address entry = StubRoutines::select_arraycopy_function(basic_type, aligned, disjoint, name, false);
   __ call_VM_leaf(entry);
 
-  __ bind(*stub->continuation());
+  if (stub != nullptr) {
+    __ bind(*stub->continuation());
+  }
 }
 
 void LIR_Assembler::shift_op(LIR_Code code, LIR_Opr left, LIR_Opr count, LIR_Opr dest, LIR_Opr tmp) {
@@ -2385,7 +2387,8 @@ void LIR_Assembler::emit_alloc_array(LIR_OpAllocArray* op) {
                       arrayOopDesc::base_offset_in_bytes(op->type()),
                       type2aelembytes(op->type()),
                       op->klass()->as_register(),
-                      *op->stub()->entry());
+                      *op->stub()->entry(),
+                      op->zero_array());
   }
   __ bind(*op->stub()->continuation());
 }

src/hotspot/cpu/s390/c1_LIRGenerator_s390.cpp

@@ -1,6 +1,6 @@
 /*
  * Copyright (c) 2016, 2024, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2016, 2017 SAP SE. All rights reserved.
+ * Copyright (c) 2016, 2024 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
@@ -710,7 +710,13 @@ void LIRGenerator::do_ArrayCopy(Intrinsic* x) {
   frame_map()->c_calling_convention(&sig);
 
   // Make all state_for calls early since they can emit code.
-  CodeEmitInfo* info = state_for (x, x->state());
+  CodeEmitInfo* info = nullptr;
+  if (x->state_before() != nullptr && x->state_before()->force_reexecute()) {
+    info = state_for(x, x->state_before());
+    info->set_force_reexecute();
+  } else {
+    info = state_for(x, x->state());
+  }
 
   LIRItem src(x->argument_at(0), this);
   LIRItem src_pos(x->argument_at(1), this);
@@ -735,6 +741,9 @@ void LIRGenerator::do_ArrayCopy(Intrinsic* x) {
   int flags;
   ciArrayKlass* expected_type;
   arraycopy_helper(x, &flags, &expected_type);
+  if (x->check_flag(Instruction::OmitChecksFlag)) {
+    flags = 0;
+  }
 
   __ arraycopy(src.result(), src_pos.result(), dst.result(), dst_pos.result(),
                length.result(), tmp, expected_type, flags, info); // does add_safepoint
@@ -769,7 +778,13 @@ void LIRGenerator::do_NewInstance(NewInstance* x) {
 }
 
 void LIRGenerator::do_NewTypeArray(NewTypeArray* x) {
-  CodeEmitInfo* info = state_for (x, x->state());
+  CodeEmitInfo* info = nullptr;
+  if (x->state_before() != nullptr && x->state_before()->force_reexecute()) {
+    info = state_for(x, x->state_before());
+    info->set_force_reexecute();
+  } else {
+    info = state_for(x, x->state());
+  }
 
   LIRItem length(x->length(), this);
   length.load_item();
@@ -786,7 +801,8 @@ void LIRGenerator::do_NewTypeArray(NewTypeArray* x) {
   __ metadata2reg(ciTypeArrayKlass::make(elem_type)->constant_encoding(), klass_reg);
 
   CodeStub* slow_path = new NewTypeArrayStub(klass_reg, len, reg, info);
-  __ allocate_array(reg, len, tmp1, tmp2, tmp3, tmp4, elem_type, klass_reg, slow_path);
+  __ allocate_array(reg, len, tmp1, tmp2, tmp3, tmp4, elem_type,
+                    klass_reg, slow_path, x->zero_array());
 
   LIR_Opr result = rlock_result(x);
   __ move(reg, result);

src/hotspot/cpu/s390/c1_MacroAssembler_s390.cpp

@@ -1,6 +1,6 @@
 /*
  * Copyright (c) 2016, 2024, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2016, 2023 SAP SE. All rights reserved.
+ * Copyright (c) 2016, 2024 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
@@ -191,7 +191,6 @@ void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register
 }
 
 void C1_MacroAssembler::initialize_body(Register objectFields, Register len_in_bytes, Register Rzero) {
-  Label done;
   assert_different_registers(objectFields, len_in_bytes, Rzero);
 
   // Initialize object fields.
@@ -203,7 +202,6 @@ void C1_MacroAssembler::initialize_body(Register objectFields, Register len_in_b
   // Use Rzero as src length, then mvcle will copy nothing
   // and fill the object with the padding value 0.
   move_long_ext(objectFields, as_Register(Rzero->encoding()-1), 0);
-  bind(done);
 }
 
 void C1_MacroAssembler::allocate_object(
@@ -274,7 +272,8 @@ void C1_MacroAssembler::allocate_array(
   int      base_offset_in_bytes,       // elements offset in bytes
   int      elt_size,                   // element size in bytes
   Register klass,                      // object klass
-  Label&   slow_case                   // Continuation point if fast allocation fails.
+  Label&   slow_case,                  // Continuation point if fast allocation fails.
+  bool     zero_array                  // zero the allocated array or not
 ) {
   assert_different_registers(obj, len, t1, t2, klass);
 
@@ -305,15 +304,17 @@ void C1_MacroAssembler::allocate_array(
   initialize_header(obj, klass, len, noreg, t1);
 
   // Clear rest of allocated space.
-  Label done;
-  Register object_fields = t1;
-  Register Rzero = Z_R1_scratch;
-  z_aghi(arr_size, -base_offset_in_bytes);
-  z_bre(done); // Jump if size of fields is zero.
-  z_la(object_fields, base_offset_in_bytes, obj);
-  z_xgr(Rzero, Rzero);
-  initialize_body(object_fields, arr_size, Rzero);
-  bind(done);
+  if (zero_array) {
+    Label done;
+    Register object_fields = t1;
+    Register Rzero = Z_R1_scratch;
+    z_aghi(arr_size, -base_offset_in_bytes);
+    z_bre(done); // Jump if size of fields is zero.
+    z_la(object_fields, base_offset_in_bytes, obj);
+    z_xgr(Rzero, Rzero);
+    initialize_body(object_fields, arr_size, Rzero);
+    bind(done);
+  }
 
   // Dtrace support is unimplemented.
   // if (CURRENT_ENV->dtrace_alloc_probes()) {

src/hotspot/cpu/s390/c1_MacroAssembler_s390.hpp

@@ -1,6 +1,6 @@
 /*
  * Copyright (c) 2016, 2024, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2016, 2023 SAP SE. All rights reserved.
+ * Copyright (c) 2016, 2024 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
@@ -89,7 +89,8 @@
     int      base_offset_in_bytes,     // elements offset in bytes
     int      elt_size,                 // element size in bytes
     Register klass,                    // object klass
-    Label&   slow_case                 // Continuation point if fast allocation fails.
+    Label&   slow_case,                // Continuation point if fast allocation fails.
+    bool     zero_array                // zero the allocated array or not
   );
 
   // Invalidates registers in this window.

src/hotspot/share/c1/c1_Compiler.cpp

@@ -235,7 +235,7 @@ bool Compiler::is_intrinsic_supported(vmIntrinsics::ID id) {
   case vmIntrinsics::_counterTime:
 #endif
   case vmIntrinsics::_getObjectSize:
-#if defined(X86) || defined(AARCH64)
+#if defined(X86) || defined(AARCH64) || defined(S390)
   case vmIntrinsics::_clone:
 #endif
     break;

src/hotspot/share/c1/c1_LIR.cpp

@@ -353,7 +353,7 @@ LIR_OpArrayCopy::LIR_OpArrayCopy(LIR_Opr src, LIR_Opr src_pos, LIR_Opr dst, LIR_
   , _tmp(tmp)
   , _expected_type(expected_type)
   , _flags(flags) {
-#if defined(X86) || defined(AARCH64)
+#if defined(X86) || defined(AARCH64) || defined(S390)
   if (expected_type != nullptr && flags == 0) {
     _stub = nullptr;
   } else {