74794fb937
These instructions were changed from "s8x16" to "i8x16" prefixes in https://github.com/WebAssembly/simd/pull/321. This CL updates all V8 code, including arch-specific code, to match. Bug: v8:10946, v8:10933 Change-Id: I26ef9ad77571f94501d42c1d65f57380fd507f3d Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2432068 Reviewed-by: Zhi An Ng <zhin@chromium.org> Commit-Queue: Adam Klein <adamk@chromium.org> Cr-Commit-Position: refs/heads/master@{#70143}
230 lines
9.2 KiB
C++
230 lines
9.2 KiB
C++
// Copyright 2020 the V8 project authors. All rights reserved.
|
|
// Use of this source code is governed by a BSD-style license that can be
|
|
// found in the LICENSE file.
|
|
//
|
|
// This file contains tests that run only on Liftoff, and each test verifies
|
|
// that the code was compiled by Liftoff. The default behavior is that each
|
|
// function is first attempted to be compiled by Liftoff, and if it fails, fall
|
|
// back to TurboFan. However we want to enforce that Liftoff is the tier that
|
|
// compiles these functions, in order to verify correctness of SIMD
|
|
// implementation in Liftoff.
|
|
|
|
#include "src/codegen/assembler-inl.h"
|
|
#include "src/wasm/wasm-opcodes.h"
|
|
#include "test/cctest/cctest.h"
|
|
#include "test/cctest/wasm/wasm-run-utils.h"
|
|
#include "test/common/wasm/test-signatures.h"
|
|
#include "test/common/wasm/wasm-macro-gen.h"
|
|
|
|
namespace v8 {
|
|
namespace internal {
|
|
namespace wasm {
|
|
namespace test_run_wasm_simd_liftoff {
|
|
|
|
#define WASM_SIMD_LIFTOFF_TEST(name) \
|
|
void RunWasm_##name##_Impl(); \
|
|
TEST(RunWasm_##name##_liftoff) { \
|
|
EXPERIMENTAL_FLAG_SCOPE(simd); \
|
|
RunWasm_##name##_Impl(); \
|
|
} \
|
|
void RunWasm_##name##_Impl()
|
|
|
|
WASM_SIMD_LIFTOFF_TEST(S128Local) {
|
|
WasmRunner<int32_t> r(TestExecutionTier::kLiftoff, kNoLowerSimd);
|
|
byte temp1 = r.AllocateLocal(kWasmS128);
|
|
BUILD(r, WASM_SET_LOCAL(temp1, WASM_GET_LOCAL(temp1)), WASM_ONE);
|
|
CHECK_EQ(1, r.Call());
|
|
}
|
|
|
|
WASM_SIMD_LIFTOFF_TEST(S128Global) {
|
|
WasmRunner<int32_t> r(TestExecutionTier::kLiftoff, kNoLowerSimd);
|
|
|
|
int32_t* g0 = r.builder().AddGlobal<int32_t>(kWasmS128);
|
|
int32_t* g1 = r.builder().AddGlobal<int32_t>(kWasmS128);
|
|
BUILD(r, WASM_SET_GLOBAL(1, WASM_GET_GLOBAL(0)), WASM_ONE);
|
|
|
|
int32_t expected = 0x1234;
|
|
for (int i = 0; i < 4; i++) {
|
|
WriteLittleEndianValue<int32_t>(&g0[i], expected);
|
|
}
|
|
r.Call();
|
|
for (int i = 0; i < 4; i++) {
|
|
int32_t actual = ReadLittleEndianValue<int32_t>(&g1[i]);
|
|
CHECK_EQ(actual, expected);
|
|
}
|
|
}
|
|
|
|
WASM_SIMD_LIFTOFF_TEST(S128Param) {
|
|
// Test how SIMD parameters in functions are processed. There is no easy way
|
|
// to specify a SIMD value when initializing a WasmRunner, so we manually
|
|
// add a new function with the right signature, and call it from main.
|
|
WasmRunner<int32_t> r(TestExecutionTier::kLiftoff, kNoLowerSimd);
|
|
TestSignatures sigs;
|
|
// We use a temp local to materialize a SIMD value, since at this point
|
|
// Liftoff does not support any SIMD operations.
|
|
byte temp1 = r.AllocateLocal(kWasmS128);
|
|
WasmFunctionCompiler& simd_func = r.NewFunction(sigs.i_s());
|
|
BUILD(simd_func, WASM_ONE);
|
|
|
|
BUILD(r,
|
|
WASM_CALL_FUNCTION(simd_func.function_index(), WASM_GET_LOCAL(temp1)));
|
|
|
|
CHECK_EQ(1, r.Call());
|
|
}
|
|
|
|
WASM_SIMD_LIFTOFF_TEST(S128Return) {
|
|
// Test how functions returning SIMD values are processed.
|
|
WasmRunner<int32_t> r(TestExecutionTier::kLiftoff, kNoLowerSimd);
|
|
TestSignatures sigs;
|
|
WasmFunctionCompiler& simd_func = r.NewFunction(sigs.s_i());
|
|
byte temp1 = simd_func.AllocateLocal(kWasmS128);
|
|
BUILD(simd_func, WASM_GET_LOCAL(temp1));
|
|
|
|
BUILD(r, WASM_CALL_FUNCTION(simd_func.function_index(), WASM_ONE), kExprDrop,
|
|
WASM_ONE);
|
|
|
|
CHECK_EQ(1, r.Call());
|
|
}
|
|
|
|
WASM_SIMD_LIFTOFF_TEST(REGRESS_1088273) {
|
|
// TODO(v8:9418): This is a regression test for Liftoff, translated from a
|
|
// mjsunit test. We do not have I64x2Mul lowering yet, so this will cause a
|
|
// crash on arch that don't support SIMD 128 and require lowering, thus
|
|
// explicitly skip them.
|
|
if (!CpuFeatures::SupportsWasmSimd128()) return;
|
|
|
|
WasmRunner<int32_t> r(TestExecutionTier::kLiftoff, kNoLowerSimd);
|
|
TestSignatures sigs;
|
|
WasmFunctionCompiler& simd_func = r.NewFunction(sigs.s_i());
|
|
byte temp1 = simd_func.AllocateLocal(kWasmS128);
|
|
BUILD(simd_func, WASM_GET_LOCAL(temp1));
|
|
|
|
BUILD(r, WASM_SIMD_SPLAT(I8x16, WASM_I32V(0x80)),
|
|
WASM_SIMD_SPLAT(I8x16, WASM_I32V(0x92)),
|
|
WASM_SIMD_I16x8_EXTRACT_LANE_U(0, WASM_SIMD_OP(kExprI64x2Mul)));
|
|
CHECK_EQ(18688, r.Call());
|
|
}
|
|
|
|
// A test to exercise logic in Liftoff's implementation of shuffle. The
|
|
// implementation in Liftoff is a bit more tricky due to shuffle requiring
|
|
// adjacent registers in ARM/ARM64.
|
|
WASM_SIMD_LIFTOFF_TEST(I8x16Shuffle) {
|
|
WasmRunner<int32_t> r(TestExecutionTier::kLiftoff, kNoLowerSimd);
|
|
// Temps to use up registers and force non-adjacent registers for shuffle.
|
|
byte local0 = r.AllocateLocal(kWasmS128);
|
|
byte local1 = r.AllocateLocal(kWasmS128);
|
|
|
|
// g0 and g1 are globals that hold input values for the shuffle,
|
|
// g0 contains byte array [0, 1, ... 15], g1 contains byte array [16, 17,
|
|
// ... 31]. They should never be overwritten - write only to output.
|
|
byte* g0 = r.builder().AddGlobal<byte>(kWasmS128);
|
|
byte* g1 = r.builder().AddGlobal<byte>(kWasmS128);
|
|
for (int i = 0; i < 16; i++) {
|
|
WriteLittleEndianValue<byte>(&g0[i], i);
|
|
WriteLittleEndianValue<byte>(&g1[i], i + 16);
|
|
}
|
|
|
|
// Output global holding a kWasmS128.
|
|
byte* output = r.builder().AddGlobal<byte>(kWasmS128);
|
|
|
|
// i8x16_shuffle(lhs, rhs, pattern) will take the last element of rhs and
|
|
// place it into the last lane of lhs.
|
|
std::array<byte, 16> pattern = {
|
|
{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 31}};
|
|
|
|
// Set up locals so shuffle is called with non-adjacent registers v2 and v0.
|
|
BUILD(r, WASM_SET_LOCAL(local0, WASM_GET_GLOBAL(1)), // local0 is in v0
|
|
WASM_SET_LOCAL(local1, WASM_GET_GLOBAL(0)), // local1 is in v1
|
|
WASM_GET_GLOBAL(0), // global0 is in v2
|
|
WASM_GET_LOCAL(local0), // local0 is in v0
|
|
WASM_SET_GLOBAL(2, WASM_SIMD_I8x16_SHUFFLE_OP(
|
|
kExprI8x16Shuffle, pattern, WASM_NOP, WASM_NOP)),
|
|
WASM_ONE);
|
|
|
|
r.Call();
|
|
|
|
// The shuffle pattern only changes the last element.
|
|
for (int i = 0; i < 15; i++) {
|
|
byte actual = ReadLittleEndianValue<byte>(&output[i]);
|
|
CHECK_EQ(i, actual);
|
|
}
|
|
CHECK_EQ(31, ReadLittleEndianValue<byte>(&output[15]));
|
|
}
|
|
|
|
// Exercise logic in Liftoff's implementation of shuffle when inputs to the
|
|
// shuffle are the same register.
|
|
WASM_SIMD_LIFTOFF_TEST(I8x16Shuffle_SingleOperand) {
|
|
WasmRunner<int32_t> r(TestExecutionTier::kLiftoff, kNoLowerSimd);
|
|
byte local0 = r.AllocateLocal(kWasmS128);
|
|
|
|
byte* g0 = r.builder().AddGlobal<byte>(kWasmS128);
|
|
for (int i = 0; i < 16; i++) {
|
|
WriteLittleEndianValue<byte>(&g0[i], i);
|
|
}
|
|
|
|
byte* output = r.builder().AddGlobal<byte>(kWasmS128);
|
|
|
|
// This pattern reverses first operand. 31 should select the last lane of
|
|
// the second operand, but since the operands are the same, the effect is that
|
|
// the first operand is reversed.
|
|
std::array<byte, 16> pattern = {
|
|
{31, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0}};
|
|
|
|
// Set up locals so shuffle is called with non-adjacent registers v2 and v0.
|
|
BUILD(r, WASM_SET_LOCAL(local0, WASM_GET_GLOBAL(0)), WASM_GET_LOCAL(local0),
|
|
WASM_GET_LOCAL(local0),
|
|
WASM_SET_GLOBAL(1, WASM_SIMD_I8x16_SHUFFLE_OP(
|
|
kExprI8x16Shuffle, pattern, WASM_NOP, WASM_NOP)),
|
|
WASM_ONE);
|
|
|
|
r.Call();
|
|
|
|
for (int i = 0; i < 16; i++) {
|
|
// Check that the output is the reverse of input.
|
|
byte actual = ReadLittleEndianValue<byte>(&output[i]);
|
|
CHECK_EQ(15 - i, actual);
|
|
}
|
|
}
|
|
|
|
// Exercise Liftoff's logic for zero-initializing stack slots. We were using an
|
|
// incorrect instruction for storing zeroes into the slot when the slot offset
|
|
// was too large to fit in the instruction as an immediate.
|
|
WASM_SIMD_LIFTOFF_TEST(FillStackSlotsWithZero_CheckStartOffset) {
|
|
WasmRunner<int64_t> r(TestExecutionTier::kLiftoff, kNoLowerSimd);
|
|
// Function that takes in 32 i64 arguments, returns i64. This gets us a large
|
|
// enough starting offset from which we spill locals.
|
|
// start = 32 * 8 + 16 (instance) = 272 (cannot fit in signed int9).
|
|
FunctionSig* sig =
|
|
r.CreateSig<int64_t, int64_t, int64_t, int64_t, int64_t, int64_t, int64_t,
|
|
int64_t, int64_t, int64_t, int64_t, int64_t, int64_t, int64_t,
|
|
int64_t, int64_t, int64_t, int64_t, int64_t, int64_t, int64_t,
|
|
int64_t, int64_t, int64_t, int64_t, int64_t, int64_t, int64_t,
|
|
int64_t, int64_t, int64_t, int64_t, int64_t>();
|
|
WasmFunctionCompiler& simd_func = r.NewFunction(sig);
|
|
|
|
// We zero 16 bytes at a time using stp, so allocate locals such that we get a
|
|
// remainder, 8 in this case, so we hit the case where we use str.
|
|
simd_func.AllocateLocal(kWasmS128);
|
|
simd_func.AllocateLocal(kWasmI64);
|
|
BUILD(simd_func, WASM_I64V_1(1));
|
|
|
|
BUILD(r, WASM_I64V_1(1), WASM_I64V_1(1), WASM_I64V_1(1), WASM_I64V_1(1),
|
|
WASM_I64V_1(1), WASM_I64V_1(1), WASM_I64V_1(1), WASM_I64V_1(1),
|
|
WASM_I64V_1(1), WASM_I64V_1(1), WASM_I64V_1(1), WASM_I64V_1(1),
|
|
WASM_I64V_1(1), WASM_I64V_1(1), WASM_I64V_1(1), WASM_I64V_1(1),
|
|
WASM_I64V_1(1), WASM_I64V_1(1), WASM_I64V_1(1), WASM_I64V_1(1),
|
|
WASM_I64V_1(1), WASM_I64V_1(1), WASM_I64V_1(1), WASM_I64V_1(1),
|
|
WASM_I64V_1(1), WASM_I64V_1(1), WASM_I64V_1(1), WASM_I64V_1(1),
|
|
WASM_I64V_1(1), WASM_I64V_1(1), WASM_I64V_1(1), WASM_I64V_1(1),
|
|
WASM_CALL_FUNCTION0(simd_func.function_index()));
|
|
|
|
CHECK_EQ(1, r.Call());
|
|
}
|
|
|
|
#undef WASM_SIMD_LIFTOFF_TEST
|
|
|
|
} // namespace test_run_wasm_simd_liftoff
|
|
} // namespace wasm
|
|
} // namespace internal
|
|
} // namespace v8
|