Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(6)

Side by Side Diff: src/ia32/code-stubs-ia32.cc

Issue 8817015: Remove early NaN/Infinity check in MathPowStub. (Closed) Base URL: https://v8.googlecode.com/svn/branches/bleeding_edge
Patch Set: Change comments Created 9 years 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/assembler.cc ('k') | src/runtime.cc » ('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 2011 the V8 project authors. All rights reserved. 1 // Copyright 2011 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 2928 matching lines...) Expand 10 before | Expand all | Expand 10 after
2939 2939
2940 void MathPowStub::Generate(MacroAssembler* masm) { 2940 void MathPowStub::Generate(MacroAssembler* masm) {
2941 CpuFeatures::Scope use_sse2(SSE2); 2941 CpuFeatures::Scope use_sse2(SSE2);
2942 Factory* factory = masm->isolate()->factory(); 2942 Factory* factory = masm->isolate()->factory();
2943 Label double_int_runtime, generic_runtime, done; 2943 Label double_int_runtime, generic_runtime, done;
2944 Label base_is_smi, unpack_exponent, exponent_not_smi, int_exponent; 2944 Label base_is_smi, unpack_exponent, exponent_not_smi, int_exponent;
2945 // Save 1 in xmm3 - we need this several times later on. 2945 // Save 1 in xmm3 - we need this several times later on.
2946 __ mov(ecx, Immediate(1)); 2946 __ mov(ecx, Immediate(1));
2947 __ cvtsi2sd(xmm3, ecx); 2947 __ cvtsi2sd(xmm3, ecx);
2948 2948
2949 switch (exponent_type_) { 2949 if (exponent_type_ == ON_STACK) {
2950 case ON_STACK: 2950 // The exponent (and base) are supplied as arguments on the stack.
2951 // The exponent (and base) are supplied as arguments on the stack. 2951 // This can only happen if the stub is called from non-optimized code.
2952 // This can only happen if the stub is called from non-optimized code. 2952 // Load input parameters from stack
2953 // Load input parameters from stack 2953 __ mov(edx, Operand(esp, 2 * kPointerSize));
2954 __ mov(edx, Operand(esp, 2 * kPointerSize)); 2954 __ mov(eax, Operand(esp, 1 * kPointerSize));
2955 __ mov(eax, Operand(esp, 1 * kPointerSize)); 2955 // edx: base (smi or heap number)
2956 // edx: base (smi or heap number) 2956 // eax: exponent (smi or heap number)
2957 // eax: exponent (smi or heap number) 2957 __ JumpIfSmi(edx, &base_is_smi, Label::kNear);
2958 __ JumpIfSmi(edx, &base_is_smi, Label::kNear); 2958 __ cmp(FieldOperand(edx, HeapObject::kMapOffset),
2959 __ cmp(FieldOperand(edx, HeapObject::kMapOffset), 2959 factory->heap_number_map());
2960 factory->heap_number_map()); 2960 __ j(not_equal, &generic_runtime);
2961 __ j(not_equal, &generic_runtime);
2962 2961
2963 // Check base for NaN or +/-Infinity 2962 __ movdbl(xmm1, FieldOperand(edx, HeapNumber::kValueOffset));
2964 __ mov(ecx, FieldOperand(edx, HeapNumber::kExponentOffset)); 2963 __ jmp(&unpack_exponent, Label::kNear);
2965 __ and_(ecx, HeapNumber::kExponentMask);
2966 __ cmp(ecx, Immediate(HeapNumber::kExponentMask));
2967 __ j(equal, &generic_runtime);
2968 __ movdbl(xmm1, FieldOperand(edx, HeapNumber::kValueOffset));
2969 __ jmp(&unpack_exponent, Label::kNear);
2970 2964
2971 __ bind(&base_is_smi); 2965 __ bind(&base_is_smi);
2972 __ SmiUntag(edx); 2966 __ SmiUntag(edx);
2973 __ cvtsi2sd(xmm1, edx); 2967 __ cvtsi2sd(xmm1, edx);
2974 __ bind(&unpack_exponent); 2968 __ bind(&unpack_exponent);
2975 2969
2976 __ JumpIfNotSmi(eax, &exponent_not_smi, Label::kNear); 2970 __ JumpIfNotSmi(eax, &exponent_not_smi, Label::kNear);
2977 __ SmiUntag(eax); 2971 __ SmiUntag(eax);
2978 __ jmp(&int_exponent); 2972 __ jmp(&int_exponent);
2979 2973
2980 __ bind(&exponent_not_smi); 2974 __ bind(&exponent_not_smi);
2981 __ cmp(FieldOperand(eax, HeapObject::kMapOffset), 2975 __ cmp(FieldOperand(eax, HeapObject::kMapOffset),
2982 factory->heap_number_map()); 2976 factory->heap_number_map());
2983 __ j(not_equal, &generic_runtime); 2977 __ j(not_equal, &generic_runtime);
2984 __ movdbl(xmm2, FieldOperand(eax, HeapNumber::kValueOffset)); 2978 __ movdbl(xmm2, FieldOperand(eax, HeapNumber::kValueOffset));
2985 break; 2979 } else if (exponent_type_ == TAGGED) {
2980 // xmm1: base as double
2981 // eax: exponent (smi or heap number)
2982 __ JumpIfNotSmi(eax, &exponent_not_smi, Label::kNear);
2983 __ SmiUntag(eax);
2984 __ jmp(&int_exponent);
2986 2985
2987 case TAGGED: 2986 __ bind(&exponent_not_smi);
2988 // xmm1: base as double 2987 __ movdbl(xmm2, FieldOperand(eax, HeapNumber::kValueOffset));
2989 // eax: exponent (smi or heap number)
2990 __ JumpIfNotSmi(eax, &exponent_not_smi, Label::kNear);
2991 __ SmiUntag(eax);
2992 __ jmp(&int_exponent);
2993
2994 __ bind(&exponent_not_smi);
2995 __ movdbl(xmm2, FieldOperand(eax, HeapNumber::kValueOffset));
2996 // Fall through intended
2997 case INTEGER:
2998 // xmm1: base as double
2999 // eax: exponent as untagged integer
3000 case DOUBLE:
3001 // xmm1: base as double
3002 // xmm2: exponent as double
3003 // Check base in xmm1 for NaN or +/-Infinity
3004 const int kExponentShift = kBitsPerByte *
3005 (HeapNumber::kExponentOffset - HeapNumber::kMantissaOffset);
3006 __ movsd(xmm4, xmm1);
3007 __ psrlq(xmm4, kExponentShift);
3008 __ movd(ecx, xmm4);
3009 __ and_(ecx, HeapNumber::kExponentMask);
3010 __ cmp(ecx, Immediate(HeapNumber::kExponentMask));
3011 __ j(equal, &generic_runtime);
3012 break;
3013 } 2988 }
3014 2989
3015 if (exponent_type_ != INTEGER) { 2990 if (exponent_type_ != INTEGER) {
3016 Label not_minus_half, fast_power; 2991 Label not_minus_half, fast_power;
3017 // xmm1: base as double that is not +/- Infinity or NaN 2992 // xmm1: base as double that is not +/- Infinity or NaN
3018 // xmm2: exponent as double 2993 // xmm2: exponent as double
3019 // Detect integer exponents stored as double. 2994 // Detect integer exponents stored as double.
3020 __ cvttsd2si(eax, Operand(xmm2)); 2995 __ cvttsd2si(eax, Operand(xmm2));
3021 // Skip to runtime if possibly NaN (indicated by the indefinite integer). 2996 // Skip to runtime if possibly NaN (indicated by the indefinite integer).
3022 __ cmp(eax, Immediate(0x80000000u)); 2997 __ cmp(eax, Immediate(0x80000000u));
3023 __ j(equal, &generic_runtime); 2998 __ j(equal, &generic_runtime);
3024 __ cvtsi2sd(xmm4, eax); 2999 __ cvtsi2sd(xmm4, eax);
3025 __ ucomisd(xmm2, xmm4); 3000 __ ucomisd(xmm2, xmm4);
3026 __ j(equal, &int_exponent); 3001 __ j(equal, &int_exponent);
3027 3002
3028 if (exponent_type_ == ON_STACK) { 3003 if (exponent_type_ == ON_STACK) {
3029 // Detect square root case. Crankshaft detects constant +/-0.5 at 3004 // Detect square root case. Crankshaft detects constant +/-0.5 at
3030 // compile time and uses DoMathPowHalf instead. We then skip this check 3005 // compile time and uses DoMathPowHalf instead. We then skip this check
3031 // for non-constant cases of +/-0.5 as these hardly occur. 3006 // for non-constant cases of +/-0.5 as these hardly occur.
3032 3007
3008 Label continue_sqrt, continue_rsqrt;
3033 // Test for -0.5. 3009 // Test for -0.5.
3034 // Load xmm4 with -0.5. 3010 // Load xmm4 with -0.5.
3035 __ mov(ecx, Immediate(0xBF000000u)); 3011 __ mov(ecx, Immediate(0xBF000000u));
3036 __ movd(xmm4, ecx); 3012 __ movd(xmm4, ecx);
3037 __ cvtss2sd(xmm4, xmm4); 3013 __ cvtss2sd(xmm4, xmm4);
3038 // xmm3 now has -0.5. 3014 // xmm3 now has -0.5.
ulan 2011/12/06 11:55:19 Typo in the comment xmm3 -> xmm4.
3039 __ ucomisd(xmm4, xmm2); 3015 __ ucomisd(xmm4, xmm2);
3040 __ j(not_equal, &not_minus_half, Label::kNear); 3016 __ j(not_equal, &not_minus_half, Label::kNear);
3041 3017
3042 // Calculates reciprocal of square root.eax 3018 // Calculates reciprocal of square root. Check for the special case of
3019 // Math.pow(-Infinity, -0.5) == 0 (ECMA spec, 15.8.2.13).
3020 // According to IEEE-754, single-precision -Infinity has the highest
3021 // 9 bits set and the lowest 23 bits cleared.
3022 __ mov(ecx, 0xFF800000u);
3023 __ movd(xmm4, ecx);
3024 __ cvtss2sd(xmm4, xmm4);
3025 __ ucomisd(xmm1, xmm4);
3026 __ j(not_equal, &continue_rsqrt, Label::kNear);
3027
3028 // Set result to 0 in the special case.
3029 __ xorps(xmm3, xmm3);
3030 __ jmp(&done);
3031
3032 __ bind(&continue_rsqrt);
3043 // sqrtsd returns -0 when input is -0. ECMA spec requires +0. 3033 // sqrtsd returns -0 when input is -0. ECMA spec requires +0.
3044 __ xorps(xmm2, xmm2); 3034 __ xorps(xmm2, xmm2);
3045 __ addsd(xmm2, xmm1); 3035 __ addsd(xmm2, xmm1); // Convert -0 to +0.
3046 __ sqrtsd(xmm2, xmm2); 3036 __ sqrtsd(xmm2, xmm2);
3047 __ divsd(xmm3, xmm2); 3037 __ divsd(xmm3, xmm2);
3048 __ jmp(&done); 3038 __ jmp(&done);
3049 3039
3050 // Test for 0.5. 3040 // Test for 0.5.
3051 __ bind(&not_minus_half); 3041 __ bind(&not_minus_half);
3052 // Load xmm2 with 0.5. 3042 // Load xmm2 with 0.5.
3053 // Since xmm3 is 1 and xmm4 is -0.5 this is simply xmm4 + xmm3. 3043 // Since xmm3 is 1 and xmm4 is -0.5 this is simply xmm4 + xmm3.
3054 __ addsd(xmm4, xmm3); 3044 __ addsd(xmm4, xmm3);
3055 // xmm2 now has 0.5. 3045 // xmm2 now has 0.5.
ulan 2011/12/06 11:55:19 Typo in the comment xmm2 -> xmm4.
3056 __ ucomisd(xmm4, xmm2); 3046 __ ucomisd(xmm4, xmm2);
3057 __ j(not_equal, &fast_power, Label::kNear); 3047 __ j(not_equal, &fast_power, Label::kNear);
3058 // Calculates square root. 3048
3049 // Calculates square root. Check for the special case of
3050 // Math.pow(-Infinity, 0.5) == Infinity (ECMA spec, 15.8.2.13).
3051 // According to IEEE-754, single-precision -Infinity has the highest
3052 // 9 bits set and the lowest 23 bits cleared.
3053 __ mov(ecx, 0xFF800000u);
3054 __ movd(xmm4, ecx);
3055 __ cvtss2sd(xmm4, xmm4);
3056 __ ucomisd(xmm1, xmm4);
3057 __ j(not_equal, &continue_sqrt, Label::kNear);
3058
3059 // Set result to Infinity in the special case.
3060 __ xorps(xmm3, xmm3);
3061 __ subsd(xmm3, xmm4);
3062 __ jmp(&done);
3063
3064 __ bind(&continue_sqrt);
3059 // sqrtsd returns -0 when input is -0. ECMA spec requires +0. 3065 // sqrtsd returns -0 when input is -0. ECMA spec requires +0.
3060 __ xorps(xmm4, xmm4); 3066 __ xorps(xmm4, xmm4);
3061 __ addsd(xmm4, xmm1); 3067 __ addsd(xmm4, xmm1); // Convert -0 to +0.
3062 __ sqrtsd(xmm3, xmm4); 3068 __ sqrtsd(xmm3, xmm4);
3063 __ jmp(&done); 3069 __ jmp(&done);
3064 } 3070 }
3065 3071
3066 // Using FPU instructions to calculate power. 3072 // Using FPU instructions to calculate power.
3067 Label fast_power_failed; 3073 Label fast_power_failed;
3068 __ bind(&fast_power); 3074 __ bind(&fast_power);
3069 __ fnclex(); // Clear flags to catch exceptions later. 3075 __ fnclex(); // Clear flags to catch exceptions later.
3070 // Transfer (B)ase and (E)xponent onto the FPU register stack. 3076 // Transfer (B)ase and (E)xponent onto the FPU register stack.
3071 __ sub(esp, Immediate(kDoubleSize)); 3077 __ sub(esp, Immediate(kDoubleSize));
(...skipping 4207 matching lines...) Expand 10 before | Expand all | Expand 10 after
7279 false); 7285 false);
7280 __ pop(edx); 7286 __ pop(edx);
7281 __ ret(0); 7287 __ ret(0);
7282 } 7288 }
7283 7289
7284 #undef __ 7290 #undef __
7285 7291
7286 } } // namespace v8::internal 7292 } } // namespace v8::internal
7287 7293
7288 #endif // V8_TARGET_ARCH_IA32 7294 #endif // V8_TARGET_ARCH_IA32
OLDNEW
« no previous file with comments | « src/assembler.cc ('k') | src/runtime.cc » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698