Skip to content

V
V8
Commit 4f38417d authored by Zhi An Ng's avatar Zhi An Ng Committed by Commit Bot
Browse files
Options

Revert "[wasm-simd][ia32] Optimize some signed integer widening sequences" 

This reverts commit 0ef2eea7.

Reason for revert: broke noavx https://ci.chromium.org/ui/p/v8/builders/ci/V8%20Linux%20-%20debug/34035/overview

Original change's description:
> [wasm-simd][ia32] Optimize some signed integer widening sequences
>
> Optimize ia32 code sequences. This is the same sequences as x64, which
> have been optimized based on supported extensions.
>
> Bug: v8:11464
> Change-Id: I10396a928a431cdd2de9b22bb8a395bc0adb4694
> Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2704897
> Reviewed-by: Deepti Gandluri <gdeepti@chromium.org>
> Commit-Queue: Zhi An Ng <zhin@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#72926}

Bug: v8:11464
Change-Id: Ibeaf35b5f6aa75d10e24f1fb57843dbc0791d37a
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2713208
Auto-Submit: Zhi An Ng <zhin@chromium.org>
Reviewed-by: 's avatarBill Budge <bbudge@chromium.org>
Commit-Queue: Bill Budge <bbudge@chromium.org>
Cr-Commit-Position: refs/heads/master@{#72929}
parent 4cd74d7d
Hide whitespace changes
Inline Side-by-side
Showing with 20 additions and 104 deletions
src/codegen/ia32/macro-assembler-ia32.cc
View file @ 4f38417d
......@@ -777,92 +777,6 @@ void TurboAssembler::I64x2UConvertI32x4High(XMMRegister dst, XMMRegister src,
}
}
void TurboAssembler::I32x4SConvertI16x8High(XMMRegister dst, XMMRegister src) {
if (CpuFeatures::IsSupported(AVX)) {
CpuFeatureScope avx_scope(this, AVX);
// src = |a|b|c|d|e|f|g|h| (high)
// dst = |e|e|f|f|g|g|h|h|
vpunpckhwd(dst, src, src);
vpsrad(dst, dst, 16);
} else {
if (dst == src) {
// 2 bytes shorter than pshufd, but has depdency on dst.
movhlps(dst, src);
pmovsxwd(dst, dst);
} else {
// No dependency on dst.
pshufd(dst, src, 0xEE);
pmovsxwd(dst, dst);
}
}
}
void TurboAssembler::I32x4UConvertI16x8High(XMMRegister dst, XMMRegister src,
XMMRegister scratch) {
if (CpuFeatures::IsSupported(AVX)) {
CpuFeatureScope avx_scope(this, AVX);
// scratch = |0|0|0|0|0|0|0|0|
// src = |a|b|c|d|e|f|g|h|
// dst = |0|a|0|b|0|c|0|d|
XMMRegister tmp = dst == src ? scratch : dst;
vpxor(tmp, tmp, tmp);
vpunpckhwd(dst, src, tmp);
} else {
if (dst == src) {
// xorps can be executed on more ports than pshufd.
xorps(scratch, scratch);
punpckhwd(dst, scratch);
} else {
// No dependency on dst.
pshufd(dst, src, 0xEE);
pmovzxwd(dst, dst);
}
}
}
void TurboAssembler::I16x8SConvertI8x16High(XMMRegister dst, XMMRegister src) {
if (CpuFeatures::IsSupported(AVX)) {
CpuFeatureScope avx_scope(this, AVX);
// src = |a|b|c|d|e|f|g|h|i|j|k|l|m|n|o|p| (high)
// dst = |i|i|j|j|k|k|l|l|m|m|n|n|o|o|p|p|
vpunpckhbw(dst, src, src);
vpsraw(dst, dst, 8);
} else {
if (dst == src) {
// 2 bytes shorter than pshufd, but has depdency on dst.
movhlps(dst, src);
pmovsxbw(dst, dst);
} else {
// No dependency on dst.
pshufd(dst, src, 0xEE);
pmovsxbw(dst, dst);
}
}
}
void TurboAssembler::I16x8UConvertI8x16High(XMMRegister dst, XMMRegister src,
XMMRegister scratch) {
if (CpuFeatures::IsSupported(AVX)) {
CpuFeatureScope avx_scope(this, AVX);
// tmp = |0|0|0|0|0|0|0|0 | 0|0|0|0|0|0|0|0|
// src = |a|b|c|d|e|f|g|h | i|j|k|l|m|n|o|p|
// dst = |0|a|0|b|0|c|0|d | 0|e|0|f|0|g|0|h|
XMMRegister tmp = dst == src ? scratch : dst;
vpxor(tmp, tmp, tmp);
vpunpckhbw(dst, src, tmp);
} else {
if (dst == src) {
// xorps can be executed on more ports than pshufd.
xorps(scratch, scratch);
punpckhbw(dst, scratch);
} else {
// No dependency on dst.
pshufd(dst, src, 0xEE);
pmovzxbw(dst, dst);
}
}
}
void TurboAssembler::I16x8Q15MulRSatS(XMMRegister dst, XMMRegister src1,
XMMRegister src2, XMMRegister scratch) {
// k = i16x8.splat(0x8000)
......
src/codegen/ia32/macro-assembler-ia32.h
View file @ 4f38417d
......@@ -661,12 +661,6 @@ class V8_EXPORT_PRIVATE TurboAssembler : public TurboAssemblerBase {
void I64x2SConvertI32x4High(XMMRegister dst, XMMRegister src);
void I64x2UConvertI32x4High(XMMRegister dst, XMMRegister src,
XMMRegister scratch);
void I32x4SConvertI16x8High(XMMRegister dst, XMMRegister src);
void I32x4UConvertI16x8High(XMMRegister dst, XMMRegister src,
XMMRegister scratch);
void I16x8SConvertI8x16High(XMMRegister dst, XMMRegister src);
void I16x8UConvertI8x16High(XMMRegister dst, XMMRegister src,
XMMRegister scratch);
void I16x8Q15MulRSatS(XMMRegister dst, XMMRegister src1, XMMRegister src2,
XMMRegister scratch);
void S128Store32Lane(Operand dst, XMMRegister src, uint8_t laneidx);
......
src/compiler/backend/ia32/code-generator-ia32.cc
View file @ 4f38417d
......@@ -2666,8 +2666,9 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
break;
}
case kIA32I32x4SConvertI16x8High: {
__ I32x4SConvertI16x8High(i.OutputSimd128Register(),
i.InputSimd128Register(0));
XMMRegister dst = i.OutputSimd128Register();
__ Palignr(dst, i.InputOperand(0), 8);
__ Pmovsxwd(dst, dst);
break;
}
case kIA32I32x4Neg: {
......@@ -2875,8 +2876,9 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
break;
}
case kIA32I32x4UConvertI16x8High: {
__ I32x4UConvertI16x8High(i.OutputSimd128Register(),
i.InputSimd128Register(0), kScratchDoubleReg);
XMMRegister dst = i.OutputSimd128Register();
__ Palignr(dst, i.InputOperand(0), 8);
__ Pmovzxwd(dst, dst);
break;
}
case kIA32I32x4ShrU: {
......@@ -2977,8 +2979,9 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
break;
}
case kIA32I16x8SConvertI8x16High: {
__ I16x8SConvertI8x16High(i.OutputSimd128Register(),
i.InputSimd128Register(0));
XMMRegister dst = i.OutputSimd128Register();
__ Palignr(dst, i.InputOperand(0), 8);
__ Pmovsxbw(dst, dst);
break;
}
case kIA32I16x8Neg: {
......@@ -3160,8 +3163,9 @@ CodeGenerator::CodeGenResult CodeGenerator::AssembleArchInstruction(
break;
}
case kIA32I16x8UConvertI8x16High: {
__ I16x8UConvertI8x16High(i.OutputSimd128Register(),
i.InputSimd128Register(0), kScratchDoubleReg);
XMMRegister dst = i.OutputSimd128Register();
__ Palignr(dst, i.InputOperand(0), 8);
__ Pmovzxbw(dst, dst);
break;
}
case kIA32I16x8ShrU: {
......
src/wasm/baseline/ia32/liftoff-assembler-ia32.h
View file @ 4f38417d
......@@ -4551,7 +4551,8 @@ void LiftoffAssembler::emit_i16x8_sconvert_i8x16_low(LiftoffRegister dst,
void LiftoffAssembler::emit_i16x8_sconvert_i8x16_high(LiftoffRegister dst,
LiftoffRegister src) {
I16x8SConvertI8x16High(dst.fp(), src.fp());
Palignr(dst.fp(), src.fp(), static_cast<uint8_t>(8));
Pmovsxbw(dst.fp(), dst.fp());
}
void LiftoffAssembler::emit_i16x8_uconvert_i8x16_low(LiftoffRegister dst,
......@@ -4561,7 +4562,8 @@ void LiftoffAssembler::emit_i16x8_uconvert_i8x16_low(LiftoffRegister dst,
void LiftoffAssembler::emit_i16x8_uconvert_i8x16_high(LiftoffRegister dst,
LiftoffRegister src) {
I16x8UConvertI8x16High(dst.fp(), src.fp(), liftoff::kScratchDoubleReg);
Palignr(dst.fp(), src.fp(), static_cast<uint8_t>(8));
Pmovzxbw(dst.fp(), dst.fp());
}
void LiftoffAssembler::emit_i32x4_sconvert_i16x8_low(LiftoffRegister dst,
......@@ -4571,7 +4573,8 @@ void LiftoffAssembler::emit_i32x4_sconvert_i16x8_low(LiftoffRegister dst,
void LiftoffAssembler::emit_i32x4_sconvert_i16x8_high(LiftoffRegister dst,
LiftoffRegister src) {
I32x4SConvertI16x8High(dst.fp(), src.fp());
Palignr(dst.fp(), src.fp(), static_cast<uint8_t>(8));
Pmovsxwd(dst.fp(), dst.fp());
}
void LiftoffAssembler::emit_i32x4_uconvert_i16x8_low(LiftoffRegister dst,
......@@ -4581,7 +4584,8 @@ void LiftoffAssembler::emit_i32x4_uconvert_i16x8_low(LiftoffRegister dst,
void LiftoffAssembler::emit_i32x4_uconvert_i16x8_high(LiftoffRegister dst,
LiftoffRegister src) {
I32x4UConvertI16x8High(dst.fp(), src.fp(), liftoff::kScratchDoubleReg);
Palignr(dst.fp(), src.fp(), static_cast<uint8_t>(8));
Pmovzxwd(dst.fp(), dst.fp());
}
void LiftoffAssembler::emit_i32x4_trunc_sat_f64x2_s_zero(LiftoffRegister dst,
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment