AuroraMiddleware/v8
1
0
Fork 0
Code Issues 2 Pull Requests Releases Wiki Activity

Cleanup: use xorps to zero out an xmm register on ia32 too.

Browse Source 
Encoding is one byte shorter.

R=lrn@chromium.org
BUG=
TEST=

Review URL: http://codereview.chromium.org/6910021

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@7760 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
This commit is contained in:
ager@chromium.org 2011-05-03 12:00:54 +00:00
parent 5197bff4f2
commit eda2a0d4d2
7 changed files with 31 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
src/ia32/assembler-ia32.cc
View File

@ -32,7 +32,7 @@
// The original source code covered by the above license above has been modified
// significantly by Google Inc.
// Copyright 2010 the V8 project authors. All rights reserved.
// Copyright 2011 the V8 project authors. All rights reserved.
#include "v8.h"
@ -2258,6 +2258,15 @@ void Assembler::xorpd(XMMRegister dst, XMMRegister src) {
}
void Assembler::xorps(XMMRegister dst, XMMRegister src) {
EnsureSpace ensure_space(this);
last_pc_ = pc_;
EMIT(0x0F);
EMIT(0x57);
emit_sse_operand(dst, src);
}
void Assembler::sqrtsd(XMMRegister dst, XMMRegister src) {
EnsureSpace ensure_space(this);
last_pc_ = pc_;

1
src/ia32/assembler-ia32.h
View File

@ -954,6 +954,7 @@ class Assembler : public AssemblerBase {
void mulsd(XMMRegister dst, XMMRegister src);
void divsd(XMMRegister dst, XMMRegister src);
void xorpd(XMMRegister dst, XMMRegister src);
void xorps(XMMRegister dst, XMMRegister src);
void sqrtsd(XMMRegister dst, XMMRegister src);
void andpd(XMMRegister dst, XMMRegister src);

4
src/ia32/code-stubs-ia32.cc
View File

@ -2880,7 +2880,7 @@ void MathPowStub::Generate(MacroAssembler* masm) {
// Calculates reciprocal of square root.
// sqrtsd returns -0 when input is -0. ECMA spec requires +0.
__ xorpd(xmm1, xmm1);
__ xorps(xmm1, xmm1);
__ addsd(xmm1, xmm0);
__ sqrtsd(xmm1, xmm1);
__ divsd(xmm3, xmm1);
@ -2897,7 +2897,7 @@ void MathPowStub::Generate(MacroAssembler* masm) {
__ j(not_equal, &call_runtime);
// Calculates square root.
// sqrtsd returns -0 when input is -0. ECMA spec requires +0.
__ xorpd(xmm1, xmm1);
__ xorps(xmm1, xmm1);
__ addsd(xmm1, xmm0);
__ sqrtsd(xmm1, xmm1);

10
src/ia32/disasm-ia32.cc
View File

@ -1,4 +1,4 @@
// Copyright 2007-2008 the V8 project authors. All rights reserved.
// Copyright 2011 the V8 project authors. All rights reserved.
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are
// met:
@ -981,6 +981,14 @@ int DisassemblerIA32::InstructionDecode(v8::internal::Vector<char> out_buffer,
NameOfXMMRegister(regop),
NameOfXMMRegister(rm));
data++;
} else if (f0byte == 0x57) {
data += 2;
int mod, regop, rm;
get_modrm(*data, &mod, &regop, &rm);
AppendToBuffer("xorps %s,%s",
NameOfXMMRegister(regop),
NameOfXMMRegister(rm));
data++;
} else if ((f0byte & 0xF0) == 0x80) {
data += JumpConditional(data, branch_hint);
} else if (f0byte == 0xBE || f0byte == 0xBF || f0byte == 0xB6 ||

2
src/ia32/full-codegen-ia32.cc
View File

@ -2794,7 +2794,7 @@ void FullCodeGenerator::EmitRandomHeapNumber(ZoneList<Expression*>* args) {
__ movd(xmm1, Operand(ebx));
__ movd(xmm0, Operand(eax));
__ cvtss2sd(xmm1, xmm1);
__ pxor(xmm0, xmm1);
__ xorps(xmm0, xmm1);
__ subsd(xmm0, xmm1);
__ movdbl(FieldOperand(edi, HeapNumber::kValueOffset), xmm0);
} else {

14
src/ia32/lithium-codegen-ia32.cc
View File

@ -1139,7 +1139,7 @@ void LCodeGen::DoConstantD(LConstantD* instr) {
// Use xor to produce +0.0 in a fast and compact way, but avoid to
// do so if the constant is -0.0.
if (BitCast<uint64_t, double>(v) == 0) {
__ xorpd(res, res);
__ xorps(res, res);
} else {
Register temp = ToRegister(instr->TempAt(0));
uint64_t int_val = BitCast<uint64_t, double>(v);
@ -1153,7 +1153,7 @@ void LCodeGen::DoConstantD(LConstantD* instr) {
__ Set(temp, Immediate(upper));
__ pinsrd(res, Operand(temp), 1);
} else {
__ xorpd(res, res);
__ xorps(res, res);
__ Set(temp, Immediate(upper));
__ pinsrd(res, Operand(temp), 1);
}
@ -1343,7 +1343,7 @@ void LCodeGen::DoBranch(LBranch* instr) {
EmitBranch(true_block, false_block, not_zero);
} else if (r.IsDouble()) {
XMMRegister reg = ToDoubleRegister(instr->InputAt(0));
__ xorpd(xmm0, xmm0);
__ xorps(xmm0, xmm0);
__ ucomisd(reg, xmm0);
EmitBranch(true_block, false_block, not_equal);
} else {
@ -2720,7 +2720,7 @@ void LCodeGen::DoMathAbs(LUnaryMathOperation* instr) {
if (r.IsDouble()) {
XMMRegister scratch = xmm0;
XMMRegister input_reg = ToDoubleRegister(instr->InputAt(0));
__ pxor(scratch, scratch);
__ xorps(scratch, scratch);
__ subsd(scratch, input_reg);
__ pand(input_reg, scratch);
} else if (r.IsInteger32()) {
@ -2742,7 +2742,7 @@ void LCodeGen::DoMathFloor(LUnaryMathOperation* instr) {
XMMRegister xmm_scratch = xmm0;
Register output_reg = ToRegister(instr->result());
XMMRegister input_reg = ToDoubleRegister(instr->InputAt(0));
__ xorpd(xmm_scratch, xmm_scratch); // Zero the register.
__ xorps(xmm_scratch, xmm_scratch); // Zero the register.
__ ucomisd(input_reg, xmm_scratch);
if (instr->hydrogen()->CheckFlag(HValue::kBailoutOnMinusZero)) {
@ -2818,7 +2818,7 @@ void LCodeGen::DoMathPowHalf(LUnaryMathOperation* instr) {
XMMRegister xmm_scratch = xmm0;
XMMRegister input_reg = ToDoubleRegister(instr->InputAt(0));
ASSERT(ToDoubleRegister(instr->result()).is(input_reg));
__ xorpd(xmm_scratch, xmm_scratch);
__ xorps(xmm_scratch, xmm_scratch);
__ addsd(input_reg, xmm_scratch); // Convert -0 to +0.
__ sqrtsd(input_reg, input_reg);
}
@ -2888,7 +2888,7 @@ void LCodeGen::DoMathLog(LUnaryMathOperation* instr) {
ASSERT(instr->InputAt(0)->Equals(instr->result()));
XMMRegister input_reg = ToDoubleRegister(instr->InputAt(0));
NearLabel positive, done, zero, negative;
__ xorpd(xmm0, xmm0);
__ xorps(xmm0, xmm0);
__ ucomisd(input_reg, xmm0);
__ j(above, &positive);
__ j(equal, &zero);

2
src/x64/lithium-codegen-x64.cc
View File

@ -3597,7 +3597,7 @@ void LCodeGen::DoDoubleToI(LDoubleToI* instr) {
__ cvttsd2siq(result_reg, input_reg);
__ movq(kScratchRegister, V8_INT64_C(0x8000000000000000), RelocInfo::NONE);
__ cmpq(result_reg, kScratchRegister);
DeoptimizeIf(equal, instr->environment());
DeoptimizeIf(equal, instr->environment());
} else {
__ cvttsd2si(result_reg, input_reg);
__ cvtlsi2sd(xmm0, result_reg);