Reapply "Precompilation/x64: Load float vector constants via Thread."

runtime/vm/assembler_x64_test.cc

 // Copyright (c) 2012, the Dart project authors.  Please see the AUTHORS file
 // for details. All rights reserved. Use of this source code is governed by a
 // BSD-style license that can be found in the LICENSE file.
 
 #include "vm/globals.h"
 #if defined(TARGET_ARCH_X64)
 
 #include "vm/assembler.h"
 #include "vm/os.h"
 #include "vm/unit_test.h"
@@ skipping 2003 matching lines @@
 
 
 ASSEMBLER_TEST_RUN(PackedDoubleSub, test) {
   typedef double (*PackedDoubleSub)();
   double res = reinterpret_cast<PackedDoubleSub>(test->entry())();
   EXPECT_FLOAT_EQ(-2.0, res, 0.000001f);
 }
 
 
 static void EnterTestFrame(Assembler* assembler) {
+  COMPILE_ASSERT(THR != CallingConventions::kArg1Reg);
+  COMPILE_ASSERT(CODE_REG != CallingConventions::kArg2Reg);
   __ EnterFrame(0);
   __ pushq(CODE_REG);
   __ pushq(PP);
+  __ pushq(THR);
   __ movq(CODE_REG, Address(CallingConventions::kArg1Reg,
                             VMHandles::kOffsetOfRawPtrInHandle));
+  __ movq(THR, CallingConventions::kArg2Reg);
   __ LoadPoolPointer(PP);
 }
 
 
 static void LeaveTestFrame(Assembler* assembler) {
+  __ popq(THR);
   __ popq(PP);
   __ popq(CODE_REG);
   __ LeaveFrame();
 }
 
 
 ASSEMBLER_TEST_GENERATE(PackedDoubleNegate, assembler) {
   static const struct ALIGN16 {
     double a;
     double b;
   } constant0 = { 1.0, 2.0 };
   EnterTestFrame(assembler);
   __ movq(RAX, Immediate(reinterpret_cast<uword>(&constant0)));
   __ movups(XMM10, Address(RAX, 0));
   __ negatepd(XMM10);
   __ movaps(XMM0, XMM10);
   LeaveTestFrame(assembler);
   __ ret();
 }
 
 
 ASSEMBLER_TEST_RUN(PackedDoubleNegate, test) {
-  double res = test->InvokeWithCode<double>();
+  double res = test->InvokeWithCodeAndThread<double>();
   EXPECT_FLOAT_EQ(-1.0, res, 0.000001f);
 }
 
 
 ASSEMBLER_TEST_GENERATE(PackedDoubleAbsolute, assembler) {
   static const struct ALIGN16 {
     double a;
     double b;
   } constant0 = { -1.0, 2.0 };
   EnterTestFrame(assembler);
   __ movq(RAX, Immediate(reinterpret_cast<uword>(&constant0)));
   __ movups(XMM10, Address(RAX, 0));
   __ abspd(XMM10);
   __ movaps(XMM0, XMM10);
   LeaveTestFrame(assembler);
   __ ret();
 }
 
 
 ASSEMBLER_TEST_RUN(PackedDoubleAbsolute, test) {
-  double res = test->InvokeWithCode<double>();
+  double res = test->InvokeWithCodeAndThread<double>();
   EXPECT_FLOAT_EQ(1.0, res, 0.000001f);
 }
 
 
 ASSEMBLER_TEST_GENERATE(PackedDoubleMul, assembler) {
   static const struct ALIGN16 {
     double a;
     double b;
   } constant0 = { 3.0, 2.0 };
   static const struct ALIGN16 {
@@ skipping 425 matching lines @@
   __ movd(XMM0, RAX);
   __ shufps(XMM0, XMM0, Immediate(0x0));
   __ negateps(XMM0);
   __ shufps(XMM0, XMM0, Immediate(0xAA));  // Copy third lane into all 4 lanes.
   LeaveTestFrame(assembler);
   __ ret();
 }
 
 
 ASSEMBLER_TEST_RUN(PackedNegate, test) {
-  float res = test->InvokeWithCode<float>();
+  float res = test->InvokeWithCodeAndThread<float>();
   EXPECT_FLOAT_EQ(-12.3f, res, 0.001f);
 }
 
 
 ASSEMBLER_TEST_GENERATE(PackedAbsolute, assembler) {
   EnterTestFrame(assembler);
   __ movl(RAX, Immediate(bit_cast<int32_t, float>(-15.3f)));
   __ movd(XMM0, RAX);
   __ shufps(XMM0, XMM0, Immediate(0x0));
   __ absps(XMM0);
   __ shufps(XMM0, XMM0, Immediate(0xAA));  // Copy third lane into all 4 lanes.
   LeaveTestFrame(assembler);
   __ ret();
 }
 
 
 ASSEMBLER_TEST_RUN(PackedAbsolute, test) {
-  float res = test->InvokeWithCode<float>();
+  float res = test->InvokeWithCodeAndThread<float>();
   EXPECT_FLOAT_EQ(15.3f, res, 0.001f);
 }
 
 
 ASSEMBLER_TEST_GENERATE(PackedSetWZero, assembler) {
   EnterTestFrame(assembler);
   __ set1ps(XMM0, RAX, Immediate(bit_cast<int32_t, float>(12.3f)));
   __ zerowps(XMM0);
   __ shufps(XMM0, XMM0, Immediate(0xFF));  // Copy the W lane which is now 0.0.
   LeaveTestFrame(assembler);
   __ ret();
 }
 
 
 ASSEMBLER_TEST_RUN(PackedSetWZero, test) {
-  float res = test->InvokeWithCode<float>();
+  float res = test->InvokeWithCodeAndThread<float>();
   EXPECT_FLOAT_EQ(0.0f, res, 0.001f);
 }
 
 
 ASSEMBLER_TEST_GENERATE(PackedMin, assembler) {
   __ set1ps(XMM0, RAX, Immediate(bit_cast<int32_t, float>(2.0f)));
   __ set1ps(XMM1, RAX, Immediate(bit_cast<int32_t, float>(4.0f)));
   __ minps(XMM0, XMM1);
   __ ret();
 }
@@ skipping 103 matching lines @@
   __ movaps(XMM0, XMM9);
   __ pushq(RAX);
   __ movss(Address(RSP, 0), XMM0);
   __ popq(RAX);
   LeaveTestFrame(assembler);
   __ ret();
 }
 
 
 ASSEMBLER_TEST_RUN(PackedLogicalNot, test) {
-  uint32_t res = test->InvokeWithCode<uint32_t>();
+  uint32_t res = test->InvokeWithCodeAndThread<uint32_t>();
   EXPECT_EQ(static_cast<uword>(0x0), res);
 }
 
 
 ASSEMBLER_TEST_GENERATE(PackedMoveHighLow, assembler) {
   static const struct ALIGN16 {
     float a;
     float b;
     float c;
     float d;
@@ skipping 393 matching lines @@
   __ StoreObject(Address(RSP, 0), obj);
   __ popq(RCX);
   __ CompareObject(RCX, obj);
   __ j(NOT_EQUAL, &fail);
   __ pushq(RAX);
   __ StoreObject(Address(RSP, 0), smi);
   __ popq(RCX);
   __ CompareObject(RCX, smi);
   __ j(NOT_EQUAL, &fail);
   __ movl(RAX, Immediate(1));  // OK
-  __ popq(PP);  // Restore caller's pool pointer.
-  __ LeaveFrame();
+  LeaveTestFrame(assembler);
   __ ret();
   __ Bind(&fail);
   __ movl(RAX, Immediate(0));  // Fail.
   LeaveTestFrame(assembler);
   __ ret();
 }
 
 
 ASSEMBLER_TEST_RUN(TestObjectCompare, test) {
-  bool res = test->InvokeWithCode<bool>();
+  bool res = test->InvokeWithCodeAndThread<bool>();
   EXPECT_EQ(true, res);
 }
 
 
 ASSEMBLER_TEST_GENERATE(TestNop, assembler) {
   __ nop(1);
   __ nop(2);
   __ nop(3);
   __ nop(4);
   __ nop(5);
@@ skipping 291 matching lines @@
   res = reinterpret_cast<DoubleToDoubleTruncCode>(test->entry())(-12.3);
   EXPECT_EQ(-12.0, res);
   res = reinterpret_cast<DoubleToDoubleTruncCode>(test->entry())(-12.8);
   EXPECT_EQ(-12.0, res);
 }
 
 
 ASSEMBLER_TEST_GENERATE(DoubleAbs, assembler) {
   EnterTestFrame(assembler);
 #if defined(TARGET_OS_WINDOWS)
-  // First argument is code object, MSVC passes second argument in XMM1.
-  __ DoubleAbs(XMM1);
-  __ movaps(XMM0, XMM1);
+  // First argument is code object, second argument is thread. MSVC passes
+  // third argument in XMM2.
+  __ DoubleAbs(XMM2);
+  __ movaps(XMM0, XMM2);
 #else
+  // SysV ABI allocates integral and double registers for arguments
+  // independently.
   __ DoubleAbs(XMM0);
 #endif
   LeaveTestFrame(assembler);
   __ ret();
 }
 
 
 ASSEMBLER_TEST_RUN(DoubleAbs, test) {
   double val = -12.45;
-  double res =  test->InvokeWithCode<double, double>(val);
+  double res =  test->InvokeWithCodeAndThread<double, double>(val);
   EXPECT_FLOAT_EQ(-val, res, 0.001);
   val = 12.45;
-  res = test->InvokeWithCode<double, double>(val);
+  res = test->InvokeWithCodeAndThread<double, double>(val);
   EXPECT_FLOAT_EQ(val, res, 0.001);
 }
 
 
 ASSEMBLER_TEST_GENERATE(ExtractSignBits, assembler) {
   __ movmskpd(RAX, XMM0);
   __ andq(RAX, Immediate(0x1));
   __ ret();
 }
 
@@ skipping 182 matching lines @@
   EXPECT_EQ(ICData::kInt64RangeBit, range_of(Integer::New(kMaxInt64)));
   EXPECT_EQ(ICData::kInt64RangeBit, range_of(Integer::New(kMinInt64)));
 
   EXPECT_EQ(0, range_of(Bool::True().raw()));
 }
 
 
 }  // namespace dart
 
 #endif  // defined TARGET_ARCH_X64