mozilla
/
gecko-dev
зеркало из https://github.com/mozilla/gecko-dev.git
1
0
Форкнуть 0
Код Задачи Пакеты Проекты Релизы Вики Активность

Bug 1669938 - Promote fp rounding wasm SIMD instructions to accepted status. r=jseward 

Background: https://github.com/WebAssembly/simd/pull/232

For all the rounding SIMD instructions:

- remove the internal 'Experimental' opcode suffix in the C++ code
- remove the guard on experimental Wasm instructions in all the C++ decoders
- move the test cases from simd/experimental.js to simd/ad-hack.js

I have checked that current V8 and wasm-tools use the same opcode
mappings.  V8 in turn guarantees the correct mapping for LLVM and
binaryen.

Drive-by bug fix: the test predicate for f64 square root was wrong, it
would round its argument to float.  This did not matter for the test
inputs we had but started to matter when I added more difficult inputs
for testing rounding.

Differential Revision: https://phabricator.services.mozilla.com/D92926
This commit is contained in:
Lars T Hansen 2020-10-14 09:50:46 +00:00
Родитель e015128a86
Коммит dec2b41f88
9 изменённых файлов: 68 добавлений и 109 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

21
js/src/jit-test/tests/wasm/simd/ad-hack.js
Просмотреть файл

@ -271,6 +271,7 @@ BigInt64Array.rectify = (x) => BigInt(x);
Float32Array.inputs = [[1, -1, 1e10, -1e10],
[-1, -2, -1e10, 1e10],
[5.1, -1.1, -4.3, -0],
...permute([1, -10, NaN, Infinity])];
Float32Array.rectify = (x) => Math.fround(x);
@ -812,8 +813,16 @@ function iabs(bits) { return (a) => zero_extend(a < 0 ? -a : a, bits) }
function fneg(a) { return -a }
function fabs(a) { return Math.abs(a) }
function fsqrt(a) { return Math.fround(Math.sqrt(Math.fround(a))) }
function sqrt(a) { return Math.sqrt(Math.fround(a)) }
function dsqrt(a) { return Math.sqrt(a) }
function bitnot(a) { return (~a) & 255 }
function ffloor(x) { return Math.fround(Math.floor(x)) }
function fceil(x) { return Math.fround(Math.ceil(x)) }
function ftrunc(x) { return Math.fround(Math.sign(x)*Math.floor(Math.abs(x))) }
function fnearest(x) { return Math.fround(Math.round(x)) }
function dfloor(x) { return Math.floor(x) }
function dceil(x) { return Math.ceil(x) }
function dtrunc(x) { return Math.sign(x)*Math.floor(Math.abs(x)) }
function dnearest(x) { return Math.round(x) }
for ( let [op, memtype, rop, resultmemtype] of
[['i8x16.neg', Int8Array, ineg(8)],
@ -828,7 +837,15 @@ for ( let [op, memtype, rop, resultmemtype] of
['f32x4.abs', Float32Array, fabs],
['f64x2.abs', Float64Array, fabs],
['f32x4.sqrt', Float32Array, fsqrt],
['f64x2.sqrt', Float64Array, sqrt],
['f64x2.sqrt', Float64Array, dsqrt],
['f32x4.ceil', Float32Array, fceil],
['f32x4.floor', Float32Array, ffloor],
['f32x4.trunc', Float32Array, ftrunc],
['f32x4.nearest', Float32Array, fnearest],
['f64x2.ceil', Float64Array, dceil],
['f64x2.floor', Float64Array, dfloor],
['f64x2.trunc', Float64Array, dtrunc],
['f64x2.nearest', Float64Array, dnearest],
['v128.not', Uint8Array, bitnot],
])
{

44
js/src/jit-test/tests/wasm/simd/experimental.js
Просмотреть файл

@ -53,16 +53,6 @@ function iota(len) {
function pmin(x, y) { return y < x ? y : x }
function pmax(x, y) { return x < y ? y : x }
function ffloor(x) { return Math.fround(Math.floor(x)) }
function fceil(x) { return Math.fround(Math.ceil(x)) }
function ftrunc(x) { return Math.fround(Math.sign(x)*Math.floor(Math.abs(x))) }
function fnearest(x) { return Math.fround(Math.round(x)) }
function dfloor(x) { return Math.floor(x) }
function dceil(x) { return Math.ceil(x) }
function dtrunc(x) { return Math.sign(x)*Math.floor(Math.abs(x)) }
function dnearest(x) { return Math.round(x) }
const v2vSig = {args:[], ret:VoidCode};
function V128Load(addr) {
@ -138,40 +128,6 @@ ins.exports.run();
var result = get(mem32, 0, 4);
assertSame(result, ans);
// Rounding, https://github.com/WebAssembly/simd/pull/232
var fxs = [5.1, -1.1, -4.3, 0];
var dxs = [5.1, -1.1];
for ( let [opcode, xs, operator] of [[F32x4CeilCode, fxs, fceil],
[F32x4FloorCode, fxs, ffloor],
[F32x4TruncCode, fxs, ftrunc],
[F32x4NearestCode, fxs, fnearest],
[F64x2CeilCode, dxs, dceil],
[F64x2FloorCode, dxs, dfloor],
[F64x2TruncCode, dxs, dtrunc],
[F64x2NearestCode, dxs, dnearest]] ) {
var k = xs.length;
var ans = xs.map(operator);
var ins = wasmEval(moduleWithSections([
sigSection([v2vSig]),
declSection([0]),
memorySection(1),
exportSection([{funcIndex: 0, name: "run"},
{memIndex: 0, name: "mem"}]),
bodySection([
funcBody({locals:[],
body: [...V128StoreExpr(0, [...V128Load(16),
SimdPrefix, varU32(opcode)])]})])]));
var mem = new (k == 4 ? Float32Array : Float64Array)(ins.exports.mem.buffer);
set(mem, k, xs);
ins.exports.run();
var result = get(mem, 0, k);
assertSame(result, ans);
}
// Zero-extending SIMD load, https://github.com/WebAssembly/simd/pull/237
for ( let [opcode, k, log2align, cons, cast] of [[V128Load32ZeroCode, 4, 2, Int32Array, Number],

3
js/src/jit/MacroAssembler.h
Просмотреть файл

@ -2674,8 +2674,7 @@ class MacroAssembler : public MacroAssemblerSpecific {
inline void widenDotInt16x8(FloatRegister rhs, FloatRegister lhsDest)
DEFINED_ON(x86_shared, arm64);
// Floating point rounding (experimental as of August, 2020)
// https://github.com/WebAssembly/simd/pull/232
// Floating point rounding
inline void ceilFloat32x4(FloatRegister src, FloatRegister dest)
DEFINED_ON(x86_shared, arm64);

16
js/src/jit/x86-shared/CodeGenerator-x86-shared.cpp
Просмотреть файл

@ -3136,28 +3136,28 @@ void CodeGenerator::visitWasmUnarySimd128(LWasmUnarySimd128* ins) {
case wasm::SimdOp::I32x4Abs:
masm.absInt32x4(src, dest);
break;
case wasm::SimdOp::F32x4CeilExperimental:
case wasm::SimdOp::F32x4Ceil:
masm.ceilFloat32x4(src, dest);
break;
case wasm::SimdOp::F32x4FloorExperimental:
case wasm::SimdOp::F32x4Floor:
masm.floorFloat32x4(src, dest);
break;
case wasm::SimdOp::F32x4TruncExperimental:
case wasm::SimdOp::F32x4Trunc:
masm.truncFloat32x4(src, dest);
break;
case wasm::SimdOp::F32x4NearestExperimental:
case wasm::SimdOp::F32x4Nearest:
masm.nearestFloat32x4(src, dest);
break;
case wasm::SimdOp::F64x2CeilExperimental:
case wasm::SimdOp::F64x2Ceil:
masm.ceilFloat64x2(src, dest);
break;
case wasm::SimdOp::F64x2FloorExperimental:
case wasm::SimdOp::F64x2Floor:
masm.floorFloat64x2(src, dest);
break;
case wasm::SimdOp::F64x2TruncExperimental:
case wasm::SimdOp::F64x2Trunc:
masm.truncFloat64x2(src, dest);
break;
case wasm::SimdOp::F64x2NearestExperimental:
case wasm::SimdOp::F64x2Nearest:
masm.nearestFloat64x2(src, dest);
break;
default:

24
js/src/wasm/WasmBaselineCompile.cpp
Просмотреть файл

@ -15037,29 +15037,21 @@ bool BaseCompiler::emitBody() {
CHECK_NEXT(dispatchVectorUnary(AbsI16x8));
case uint32_t(SimdOp::I32x4Abs):
CHECK_NEXT(dispatchVectorUnary(AbsI32x4));
case uint32_t(SimdOp::F32x4CeilExperimental):
CHECK_SIMD_EXPERIMENTAL();
case uint32_t(SimdOp::F32x4Ceil):
CHECK_NEXT(dispatchVectorUnary(CeilF32x4));
case uint32_t(SimdOp::F32x4FloorExperimental):
CHECK_SIMD_EXPERIMENTAL();
case uint32_t(SimdOp::F32x4Floor):
CHECK_NEXT(dispatchVectorUnary(FloorF32x4));
case uint32_t(SimdOp::F32x4TruncExperimental):
CHECK_SIMD_EXPERIMENTAL();
case uint32_t(SimdOp::F32x4Trunc):
CHECK_NEXT(dispatchVectorUnary(TruncF32x4));
case uint32_t(SimdOp::F32x4NearestExperimental):
CHECK_SIMD_EXPERIMENTAL();
case uint32_t(SimdOp::F32x4Nearest):
CHECK_NEXT(dispatchVectorUnary(NearestF32x4));
case uint32_t(SimdOp::F64x2CeilExperimental):
CHECK_SIMD_EXPERIMENTAL();
case uint32_t(SimdOp::F64x2Ceil):
CHECK_NEXT(dispatchVectorUnary(CeilF64x2));
case uint32_t(SimdOp::F64x2FloorExperimental):
CHECK_SIMD_EXPERIMENTAL();
case uint32_t(SimdOp::F64x2Floor):
CHECK_NEXT(dispatchVectorUnary(FloorF64x2));
case uint32_t(SimdOp::F64x2TruncExperimental):
CHECK_SIMD_EXPERIMENTAL();
case uint32_t(SimdOp::F64x2Trunc):
CHECK_NEXT(dispatchVectorUnary(TruncF64x2));
case uint32_t(SimdOp::F64x2NearestExperimental):
CHECK_SIMD_EXPERIMENTAL();
case uint32_t(SimdOp::F64x2Nearest):
CHECK_NEXT(dispatchVectorUnary(NearestF64x2));
case uint32_t(SimdOp::I8x16Shl):
CHECK_NEXT(dispatchVectorVariableShift(ShiftLeftI8x16));

16
js/src/wasm/WasmConstants.h
Просмотреть файл

@ -642,14 +642,14 @@ enum class SimdOp {
I64x2Mul = 0xd5,
// MinS = 0xd6
// MinU = 0xd7
F32x4CeilExperimental = 0xd8,
F32x4FloorExperimental = 0xd9,
F32x4TruncExperimental = 0xda,
F32x4NearestExperimental = 0xdb,
F64x2CeilExperimental = 0xdc,
F64x2FloorExperimental = 0xdd,
F64x2TruncExperimental = 0xde,
F64x2NearestExperimental = 0xdf,
F32x4Ceil = 0xd8,
F32x4Floor = 0xd9,
F32x4Trunc = 0xda,
F32x4Nearest = 0xdb,
F64x2Ceil = 0xdc,
F64x2Floor = 0xdd,
F64x2Trunc = 0xde,
F64x2Nearest = 0xdf,
F32x4Abs = 0xe0,
F32x4Neg = 0xe1,
// Round = 0xe2

18
js/src/wasm/WasmIonCompile.cpp
Просмотреть файл

@ -4950,16 +4950,14 @@ static bool EmitBodyExprs(FunctionCompiler& f) {
case uint32_t(SimdOp::I8x16Abs):
case uint32_t(SimdOp::I16x8Abs):
case uint32_t(SimdOp::I32x4Abs):
CHECK(EmitUnarySimd128(f, SimdOp(op.b1)));
case uint32_t(SimdOp::F32x4CeilExperimental):
case uint32_t(SimdOp::F32x4FloorExperimental):
case uint32_t(SimdOp::F32x4TruncExperimental):
case uint32_t(SimdOp::F32x4NearestExperimental):
case uint32_t(SimdOp::F64x2CeilExperimental):
case uint32_t(SimdOp::F64x2FloorExperimental):
case uint32_t(SimdOp::F64x2TruncExperimental):
case uint32_t(SimdOp::F64x2NearestExperimental):
CHECK_SIMD_EXPERIMENTAL();
case uint32_t(SimdOp::F32x4Ceil):
case uint32_t(SimdOp::F32x4Floor):
case uint32_t(SimdOp::F32x4Trunc):
case uint32_t(SimdOp::F32x4Nearest):
case uint32_t(SimdOp::F64x2Ceil):
case uint32_t(SimdOp::F64x2Floor):
case uint32_t(SimdOp::F64x2Trunc):
case uint32_t(SimdOp::F64x2Nearest):
CHECK(EmitUnarySimd128(f, SimdOp(op.b1)));
case uint32_t(SimdOp::I8x16AnyTrue):
case uint32_t(SimdOp::I16x8AnyTrue):

16
js/src/wasm/WasmOpIter.cpp
Просмотреть файл

@ -462,14 +462,14 @@ OpKind wasm::Classify(OpBytes op) {
case SimdOp::I8x16Abs:
case SimdOp::I16x8Abs:
case SimdOp::I32x4Abs:
case SimdOp::F32x4CeilExperimental:
case SimdOp::F32x4FloorExperimental:
case SimdOp::F32x4TruncExperimental:
case SimdOp::F32x4NearestExperimental:
case SimdOp::F64x2CeilExperimental:
case SimdOp::F64x2FloorExperimental:
case SimdOp::F64x2TruncExperimental:
case SimdOp::F64x2NearestExperimental:
case SimdOp::F32x4Ceil:
case SimdOp::F32x4Floor:
case SimdOp::F32x4Trunc:
case SimdOp::F32x4Nearest:
case SimdOp::F64x2Ceil:
case SimdOp::F64x2Floor:
case SimdOp::F64x2Trunc:
case SimdOp::F64x2Nearest:
WASM_SIMD_OP(OpKind::Unary);
case SimdOp::I8x16Shl:
case SimdOp::I8x16ShrS:

19
js/src/wasm/WasmValidate.cpp
Просмотреть файл

@ -1088,17 +1088,14 @@ static bool DecodeFunctionBodyExprs(const ModuleEnvironment& env,
case uint32_t(SimdOp::I8x16Abs):
case uint32_t(SimdOp::I16x8Abs):
case uint32_t(SimdOp::I32x4Abs):
CHECK(iter.readUnary(ValType::V128, &nothing));
case uint32_t(SimdOp::F32x4CeilExperimental):
case uint32_t(SimdOp::F32x4FloorExperimental):
case uint32_t(SimdOp::F32x4TruncExperimental):
case uint32_t(SimdOp::F32x4NearestExperimental):
case uint32_t(SimdOp::F64x2CeilExperimental):
case uint32_t(SimdOp::F64x2FloorExperimental):
case uint32_t(SimdOp::F64x2TruncExperimental):
case uint32_t(SimdOp::F64x2NearestExperimental):
CHECK_SIMD_EXPERIMENTAL();
case uint32_t(SimdOp::F32x4Ceil):
case uint32_t(SimdOp::F32x4Floor):
case uint32_t(SimdOp::F32x4Trunc):
case uint32_t(SimdOp::F32x4Nearest):
case uint32_t(SimdOp::F64x2Ceil):
case uint32_t(SimdOp::F64x2Floor):
case uint32_t(SimdOp::F64x2Trunc):
case uint32_t(SimdOp::F64x2Nearest):
CHECK(iter.readUnary(ValType::V128, &nothing));
case uint32_t(SimdOp::I8x16Shl):