Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(79)

Side by Side Diff: src/ia32/lithium-codegen-ia32.cc

Issue 12374046: Tweak register allocation for Math.round and do not use roundsd. (Closed) Base URL: https://v8.googlecode.com/svn/branches/bleeding_edge
Patch Set: Created 7 years, 9 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View unified diff | Download patch | Annotate | Revision Log
« no previous file with comments | « src/ia32/lithium-codegen-ia32.h ('k') | src/ia32/lithium-ia32.h » ('j') | no next file with comments »
Toggle Intra-line Diffs ('i') | Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
OLDNEW
1 // Copyright 2012 the V8 project authors. All rights reserved. 1 // Copyright 2012 the V8 project authors. All rights reserved.
2 // Redistribution and use in source and binary forms, with or without 2 // Redistribution and use in source and binary forms, with or without
3 // modification, are permitted provided that the following conditions are 3 // modification, are permitted provided that the following conditions are
4 // met: 4 // met:
5 // 5 //
6 // * Redistributions of source code must retain the above copyright 6 // * Redistributions of source code must retain the above copyright
7 // notice, this list of conditions and the following disclaimer. 7 // notice, this list of conditions and the following disclaimer.
8 // * Redistributions in binary form must reproduce the above 8 // * Redistributions in binary form must reproduce the above
9 // copyright notice, this list of conditions and the following 9 // copyright notice, this list of conditions and the following
10 // disclaimer in the documentation and/or other materials provided 10 // disclaimer in the documentation and/or other materials provided
(...skipping 3715 matching lines...) Expand 10 before | Expand all | Expand 10 after
3726 __ cvtsi2sd(xmm_scratch, output_reg); 3726 __ cvtsi2sd(xmm_scratch, output_reg);
3727 __ ucomisd(input_reg, xmm_scratch); 3727 __ ucomisd(input_reg, xmm_scratch);
3728 __ j(equal, &done, Label::kNear); 3728 __ j(equal, &done, Label::kNear);
3729 __ sub(output_reg, Immediate(1)); 3729 __ sub(output_reg, Immediate(1));
3730 DeoptimizeIf(overflow, instr->environment()); 3730 DeoptimizeIf(overflow, instr->environment());
3731 3731
3732 __ bind(&done); 3732 __ bind(&done);
3733 } 3733 }
3734 } 3734 }
3735 3735
3736 void LCodeGen::DoMathRound(LUnaryMathOperation* instr) { 3736 void LCodeGen::DoMathRound(LMathRound* instr) {
3737 CpuFeatures::Scope scope(SSE2); 3737 CpuFeatures::Scope scope(SSE2);
3738 Register output_reg = ToRegister(instr->result()); 3738 Register output_reg = ToRegister(instr->result());
3739 XMMRegister input_reg = ToDoubleRegister(instr->value()); 3739 XMMRegister input_reg = ToDoubleRegister(instr->value());
3740 XMMRegister xmm_scratch = xmm0; 3740 XMMRegister xmm_scratch = xmm0;
3741 XMMRegister input_temp = ToDoubleRegister(instr->temp());
3741 ExternalReference one_half = ExternalReference::address_of_one_half(); 3742 ExternalReference one_half = ExternalReference::address_of_one_half();
3742 ExternalReference minus_one_half = 3743 ExternalReference minus_one_half =
3743 ExternalReference::address_of_minus_one_half(); 3744 ExternalReference::address_of_minus_one_half();
3744 bool minus_zero_check =
3745 instr->hydrogen()->CheckFlag(HValue::kBailoutOnMinusZero);
3746 3745
3746 Label done, round_to_zero, below_one_half, do_not_compensate;
3747 __ movdbl(xmm_scratch, Operand::StaticVariable(one_half)); 3747 __ movdbl(xmm_scratch, Operand::StaticVariable(one_half));
3748 __ ucomisd(xmm_scratch, input_reg);
3749 __ j(above, &below_one_half);
3748 3750
3749 if (CpuFeatures::IsSupported(SSE4_1) && !minus_zero_check) { 3751 // CVTTSD2SI rounds towards zero, since 0.5 <= x, we use floor(0.5 + x).
3750 CpuFeatures::Scope scope(SSE4_1); 3752 __ addsd(xmm_scratch, input_reg);
3753 __ cvttsd2si(output_reg, Operand(xmm_scratch));
3754 // Overflow is signalled with minint.
3755 __ cmp(output_reg, 0x80000000u);
3756 __ RecordComment("D2I conversion overflow");
3757 DeoptimizeIf(equal, instr->environment());
3758 __ jmp(&done);
3751 3759
3752 __ addsd(xmm_scratch, input_reg); 3760 __ bind(&below_one_half);
3753 __ roundsd(xmm_scratch, xmm_scratch, Assembler::kRoundDown); 3761 __ movdbl(xmm_scratch, Operand::StaticVariable(minus_one_half));
3754 __ cvttsd2si(output_reg, Operand(xmm_scratch)); 3762 __ ucomisd(xmm_scratch, input_reg);
3755 // Overflow is signalled with minint. 3763 __ j(below_equal, &round_to_zero);
3756 __ cmp(output_reg, 0x80000000u);
3757 __ RecordComment("D2I conversion overflow");
3758 DeoptimizeIf(equal, instr->environment());
3759 } else {
3760 Label done, round_to_zero, below_one_half, do_not_compensate;
3761 __ ucomisd(xmm_scratch, input_reg);
3762 __ j(above, &below_one_half);
3763 3764
3764 // CVTTSD2SI rounds towards zero, since 0.5 <= x, we use floor(0.5 + x). 3765 // CVTTSD2SI rounds towards zero, we use ceil(x - (-0.5)) and then
3765 __ addsd(xmm_scratch, input_reg); 3766 // compare and compensate.
3766 __ cvttsd2si(output_reg, Operand(xmm_scratch)); 3767 __ movsd(input_temp, input_reg); // Do not alter input_reg.
3767 // Overflow is signalled with minint. 3768 __ subsd(input_temp, xmm_scratch);
3768 __ cmp(output_reg, 0x80000000u); 3769 __ cvttsd2si(output_reg, Operand(input_temp));
3769 __ RecordComment("D2I conversion overflow"); 3770 // Catch minint due to overflow, and to prevent overflow when compensating.
3770 DeoptimizeIf(equal, instr->environment()); 3771 __ cmp(output_reg, 0x80000000u);
3771 __ jmp(&done); 3772 __ RecordComment("D2I conversion overflow");
3773 DeoptimizeIf(equal, instr->environment());
3772 3774
3773 __ bind(&below_one_half); 3775 __ cvtsi2sd(xmm_scratch, output_reg);
3774 __ movdbl(xmm_scratch, Operand::StaticVariable(minus_one_half)); 3776 __ ucomisd(xmm_scratch, input_temp);
3775 __ ucomisd(xmm_scratch, input_reg); 3777 __ j(equal, &done);
3776 __ j(below_equal, &round_to_zero); 3778 __ sub(output_reg, Immediate(1));
3779 // No overflow because we already ruled out minint.
3780 __ jmp(&done);
3777 3781
3778 // CVTTSD2SI rounds towards zero, we use ceil(x - (-0.5)) and then 3782 __ bind(&round_to_zero);
3779 // compare and compensate. 3783 // We return 0 for the input range [+0, 0.5[, or [-0.5, 0.5[ if
3780 __ subsd(input_reg, xmm_scratch); 3784 // we can ignore the difference between a result of -0 and +0.
3781 __ cvttsd2si(output_reg, Operand(input_reg)); 3785 if (instr->hydrogen()->CheckFlag(HValue::kBailoutOnMinusZero)) {
3782 // Catch minint due to overflow, and to prevent overflow when compensating. 3786 // If the sign is positive, we return +0.
3783 __ cmp(output_reg, 0x80000000u); 3787 __ movmskpd(output_reg, input_reg);
3784 __ RecordComment("D2I conversion overflow"); 3788 __ test(output_reg, Immediate(1));
3785 DeoptimizeIf(equal, instr->environment()); 3789 __ RecordComment("Minus zero");
3786 3790 DeoptimizeIf(not_zero, instr->environment());
3787 __ cvtsi2sd(xmm_scratch, output_reg);
3788 __ ucomisd(xmm_scratch, input_reg);
3789 __ j(equal, &done);
3790 __ sub(output_reg, Immediate(1));
3791 // No overflow because we already ruled out minint.
3792 __ jmp(&done);
3793
3794 __ bind(&round_to_zero);
3795 // We return 0 for the input range [+0, 0.5[, or [-0.5, 0.5[ if
3796 // we can ignore the difference between a result of -0 and +0.
3797 if (minus_zero_check) {
3798 // If the sign is positive, we return +0.
3799 __ movmskpd(output_reg, input_reg);
3800 __ test(output_reg, Immediate(1));
3801 __ RecordComment("Minus zero");
3802 DeoptimizeIf(not_zero, instr->environment());
3803 }
3804 __ Set(output_reg, Immediate(0));
3805 __ bind(&done);
3806 } 3791 }
3792 __ Set(output_reg, Immediate(0));
3793 __ bind(&done);
3807 } 3794 }
3808 3795
3809 3796
3810 void LCodeGen::DoMathSqrt(LUnaryMathOperation* instr) { 3797 void LCodeGen::DoMathSqrt(LUnaryMathOperation* instr) {
3811 CpuFeatures::Scope scope(SSE2); 3798 CpuFeatures::Scope scope(SSE2);
3812 XMMRegister input_reg = ToDoubleRegister(instr->value()); 3799 XMMRegister input_reg = ToDoubleRegister(instr->value());
3813 ASSERT(ToDoubleRegister(instr->result()).is(input_reg)); 3800 ASSERT(ToDoubleRegister(instr->result()).is(input_reg));
3814 __ sqrtsd(input_reg, input_reg); 3801 __ sqrtsd(input_reg, input_reg);
3815 } 3802 }
3816 3803
(...skipping 211 matching lines...) Expand 10 before | Expand all | Expand 10 after
4028 4015
4029 4016
4030 void LCodeGen::DoUnaryMathOperation(LUnaryMathOperation* instr) { 4017 void LCodeGen::DoUnaryMathOperation(LUnaryMathOperation* instr) {
4031 switch (instr->op()) { 4018 switch (instr->op()) {
4032 case kMathAbs: 4019 case kMathAbs:
4033 DoMathAbs(instr); 4020 DoMathAbs(instr);
4034 break; 4021 break;
4035 case kMathFloor: 4022 case kMathFloor:
4036 DoMathFloor(instr); 4023 DoMathFloor(instr);
4037 break; 4024 break;
4038 case kMathRound:
4039 DoMathRound(instr);
4040 break;
4041 case kMathSqrt: 4025 case kMathSqrt:
4042 DoMathSqrt(instr); 4026 DoMathSqrt(instr);
4043 break; 4027 break;
4044 case kMathCos: 4028 case kMathCos:
4045 DoMathCos(instr); 4029 DoMathCos(instr);
4046 break; 4030 break;
4047 case kMathSin: 4031 case kMathSin:
4048 DoMathSin(instr); 4032 DoMathSin(instr);
4049 break; 4033 break;
4050 case kMathTan: 4034 case kMathTan:
(...skipping 2192 matching lines...) Expand 10 before | Expand all | Expand 10 after
6243 FixedArray::kHeaderSize - kPointerSize)); 6227 FixedArray::kHeaderSize - kPointerSize));
6244 __ bind(&done); 6228 __ bind(&done);
6245 } 6229 }
6246 6230
6247 6231
6248 #undef __ 6232 #undef __
6249 6233
6250 } } // namespace v8::internal 6234 } } // namespace v8::internal
6251 6235
6252 #endif // V8_TARGET_ARCH_IA32 6236 #endif // V8_TARGET_ARCH_IA32
OLDNEW
« no previous file with comments | « src/ia32/lithium-codegen-ia32.h ('k') | src/ia32/lithium-ia32.h » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698