AuroraMiddleware/v8
1
0
Fork 0
Code Issues 2 Pull Requests Releases Wiki Activity

[wasm] Fix tier-up budget tracking for recursive calls

Browse Source 
In the previous implementation, functions overwrote any budget
decrements caused by recursive invocations of themselves, which
could cause tier-up decisions for certain unlucky functions to
get delayed unreasonably long.
This patch avoids this by working with the on-instance value
directly instead of caching it in a stack slot. That generates
the same amount of Liftoff code as the status quo, but handles
recursive functions properly.
The "barista3" benchmark's peak performance improves by almost 20%.

Bug: v8:12281
Change-Id: I8b487a88da99c2d22e132f2cc72bdf36aa5f6e63
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3693710
Commit-Queue: Jakob Kummerow <jkummerow@chromium.org>
Reviewed-by: Clemens Backes <clemensb@chromium.org>
Cr-Commit-Position: refs/heads/main@{#81249}
This commit is contained in:
Jakob Kummerow 2022-06-20 13:43:46 +02:00 committed by V8 LUCI CQ
parent 3c34137ad2
commit 15f372afaf
13 changed files with 26 additions and 51 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
src/runtime/runtime-wasm.cc
View File

@ -332,6 +332,9 @@ RUNTIME_FUNCTION(Runtime_WasmTriggerTierUp) {
DCHECK_EQ(instance, frame_finder.frame()->wasm_instance());
wasm::TriggerTierUp(instance, func_index);
int array_index =
wasm::declared_function_index(instance.module(), func_index);
instance.tiering_budget_array()[array_index] = FLAG_wasm_tiering_budget;
return ReadOnlyRoots(isolate).undefined_value();
}

3
src/wasm/baseline/arm/liftoff-assembler-arm.h
View File

@ -71,7 +71,6 @@ static_assert(2 * kSystemPointerSize == LiftoffAssembler::kStackSlotSize,
"Slot size should be twice the size of the 32 bit pointer.");
constexpr int kInstanceOffset = 2 * kSystemPointerSize;
constexpr int kFeedbackVectorOffset = 3 * kSystemPointerSize;
constexpr int kTierupBudgetOffset = 4 * kSystemPointerSize;
// kPatchInstructionsRequired sets a maximum limit of how many instructions that
// PatchPrepareStackFrame will use in order to increase the stack appropriately.
// Three instructions are required to sub a large constant, movw + movt + sub.
@ -562,7 +561,7 @@ void LiftoffAssembler::AbortCompilation() { AbortedCodeGeneration(); }
// static
constexpr int LiftoffAssembler::StaticStackFrameSize() {
return liftoff::kTierupBudgetOffset;
return liftoff::kFeedbackVectorOffset;
}
int LiftoffAssembler::SlotSizeForType(ValueKind kind) {

3
src/wasm/baseline/arm64/liftoff-assembler-arm64.h
View File

@ -71,7 +71,6 @@ inline constexpr Condition ToCondition(LiftoffCondition liftoff_cond) {
constexpr int kInstanceOffset = 2 * kSystemPointerSize;
constexpr int kFeedbackVectorOffset = 3 * kSystemPointerSize;
constexpr int kTierupBudgetOffset = 4 * kSystemPointerSize;
inline MemOperand GetStackSlot(int offset) { return MemOperand(fp, -offset); }
@ -387,7 +386,7 @@ void LiftoffAssembler::AbortCompilation() { AbortedCodeGeneration(); }
// static
constexpr int LiftoffAssembler::StaticStackFrameSize() {
return liftoff::kTierupBudgetOffset;
return liftoff::kFeedbackVectorOffset;
}
int LiftoffAssembler::SlotSizeForType(ValueKind kind) {

3
src/wasm/baseline/ia32/liftoff-assembler-ia32.h
View File

@ -52,7 +52,6 @@ inline constexpr Condition ToCondition(LiftoffCondition liftoff_cond) {
// ebp-4 holds the stack marker, ebp-8 is the instance parameter.
constexpr int kInstanceOffset = 8;
constexpr int kFeedbackVectorOffset = 12; // ebp-12 is the feedback vector.
constexpr int kTierupBudgetOffset = 16; // ebp-16 is the tiering budget.
inline Operand GetStackSlot(int offset) { return Operand(ebp, -offset); }
@ -314,7 +313,7 @@ void LiftoffAssembler::AbortCompilation() {}
// static
constexpr int LiftoffAssembler::StaticStackFrameSize() {
return liftoff::kTierupBudgetOffset;
return liftoff::kFeedbackVectorOffset;
}
int LiftoffAssembler::SlotSizeForType(ValueKind kind) {

42
src/wasm/baseline/liftoff-compiler.cc
View File

@ -742,8 +742,17 @@ class LiftoffCompiler {
const int kMax = FLAG_wasm_tiering_budget / 4;
if (budget_used > kMax) budget_used = kMax;
LiftoffRegister budget_reg = __ GetUnusedRegister(kGpReg, {});
__ Fill(budget_reg, liftoff::kTierupBudgetOffset, ValueKind::kI32);
LiftoffRegList pinned;
LiftoffRegister budget_reg =
pinned.set(__ GetUnusedRegister(kGpReg, pinned));
LiftoffRegister array_reg =
pinned.set(__ GetUnusedRegister(kGpReg, pinned));
LOAD_INSTANCE_FIELD(array_reg.gp(), TieringBudgetArray, kSystemPointerSize,
pinned);
uint32_t offset =
kInt32Size * declared_function_index(env_->module, func_index_);
__ Load(budget_reg, array_reg.gp(), no_reg, offset, LoadType::kI32Load,
pinned);
LiftoffRegList regs_to_save = __ cache_state()->used_registers;
// The cached instance will be reloaded separately.
if (__ cache_state()->cached_instance != no_reg) {
@ -763,7 +772,8 @@ class LiftoffCompiler {
OutOfLineCode& ool = out_of_line_code_.back();
__ emit_i32_subi_jump_negative(budget_reg.gp(), budget_used,
ool.label.get());
__ Spill(liftoff::kTierupBudgetOffset, budget_reg, ValueKind::kI32);
__ Store(array_reg.gp(), no_reg, offset, budget_reg, StoreType::kI32Store,
pinned);
__ bind(ool.continuation.get());
}
@ -848,17 +858,6 @@ class LiftoffCompiler {
pinned);
__ Spill(liftoff::kFeedbackVectorOffset, tmp, kPointerKind);
}
if (dynamic_tiering()) {
CODE_COMMENT("load tier up budget");
LiftoffRegList pinned = kGpParamRegisters;
LiftoffRegister tmp = pinned.set(__ GetUnusedRegister(kGpReg, pinned));
LOAD_INSTANCE_FIELD(tmp.gp(), TieringBudgetArray, kSystemPointerSize,
pinned);
uint32_t offset =
kInt32Size * declared_function_index(env_->module, func_index_);
__ Load(tmp, tmp.gp(), no_reg, offset, LoadType::kI32Load, pinned);
__ Spill(liftoff::kTierupBudgetOffset, tmp, ValueKind::kI32);
}
if (for_debugging_) __ ResetOSRTarget();
// Process parameters.
@ -986,11 +985,6 @@ class LiftoffCompiler {
__ RecordSpillsInSafepoint(safepoint, gp_regs,
ool->safepoint_info->spills, index);
}
if (is_tierup) {
// Reset the budget.
__ Spill(liftoff::kTierupBudgetOffset,
WasmValue(FLAG_wasm_tiering_budget));
}
DCHECK_EQ(!debug_sidetable_builder_, !ool->debug_sidetable_entry_builder);
if (V8_UNLIKELY(ool->debug_sidetable_entry_builder)) {
@ -2246,16 +2240,6 @@ class LiftoffCompiler {
void TierupCheckOnExit(FullDecoder* decoder) {
if (!dynamic_tiering()) return;
TierupCheck(decoder, decoder->position(), __ pc_offset());
CODE_COMMENT("update tiering budget");
LiftoffRegList pinned;
LiftoffRegister budget = pinned.set(__ GetUnusedRegister(kGpReg, pinned));
LiftoffRegister array = pinned.set(__ GetUnusedRegister(kGpReg, pinned));
LOAD_INSTANCE_FIELD(array.gp(), TieringBudgetArray, kSystemPointerSize,
pinned);
uint32_t offset =
kInt32Size * declared_function_index(env_->module, func_index_);
__ Fill(budget, liftoff::kTierupBudgetOffset, ValueKind::kI32);
__ Store(array.gp(), no_reg, offset, budget, StoreType::kI32Store, pinned);
}
void DoReturn(FullDecoder* decoder, uint32_t /* drop_values */) {

3
src/wasm/baseline/loong64/liftoff-assembler-loong64.h
View File

@ -72,7 +72,6 @@ inline constexpr Condition ToCondition(LiftoffCondition liftoff_cond) {
constexpr int kInstanceOffset = 2 * kSystemPointerSize;
constexpr int kFeedbackVectorOffset = 3 * kSystemPointerSize;
constexpr int kTierupBudgetOffset = 4 * kSystemPointerSize;
inline MemOperand GetStackSlot(int offset) { return MemOperand(fp, -offset); }
@ -293,7 +292,7 @@ void LiftoffAssembler::AbortCompilation() {}
// static
constexpr int LiftoffAssembler::StaticStackFrameSize() {
return liftoff::kTierupBudgetOffset;
return liftoff::kFeedbackVectorOffset;
}
int LiftoffAssembler::SlotSizeForType(ValueKind kind) {

3
src/wasm/baseline/mips/liftoff-assembler-mips.h
View File

@ -75,7 +75,6 @@ constexpr int32_t kHighWordOffset = 4;
constexpr int kInstanceOffset = 2 * kSystemPointerSize;
constexpr int kFeedbackVectorOffset = 3 * kSystemPointerSize;
constexpr int kTierupBudgetOffset = 4 * kSystemPointerSize;
inline MemOperand GetStackSlot(int offset) { return MemOperand(fp, -offset); }
@ -422,7 +421,7 @@ void LiftoffAssembler::AbortCompilation() {}
// static
constexpr int LiftoffAssembler::StaticStackFrameSize() {
return liftoff::kTierupBudgetOffset;
return liftoff::kFeedbackVectorOffset;
}
int LiftoffAssembler::SlotSizeForType(ValueKind kind) {

3
src/wasm/baseline/mips64/liftoff-assembler-mips64.h
View File

@ -72,7 +72,6 @@ inline constexpr Condition ToCondition(LiftoffCondition liftoff_cond) {
constexpr int kInstanceOffset = 2 * kSystemPointerSize;
constexpr int kFeedbackVectorOffset = 3 * kSystemPointerSize;
constexpr int kTierupBudgetOffset = 4 * kSystemPointerSize;
inline MemOperand GetStackSlot(int offset) { return MemOperand(fp, -offset); }
@ -409,7 +408,7 @@ void LiftoffAssembler::AbortCompilation() {}
// static
constexpr int LiftoffAssembler::StaticStackFrameSize() {
return liftoff::kTierupBudgetOffset;
return liftoff::kFeedbackVectorOffset;
}
int LiftoffAssembler::SlotSizeForType(ValueKind kind) {

4
src/wasm/baseline/ppc/liftoff-assembler-ppc.h
View File

@ -49,8 +49,6 @@ constexpr int32_t kInstanceOffset =
(FLAG_enable_embedded_constant_pool.value() ? 3 : 2) * kSystemPointerSize;
constexpr int kFeedbackVectorOffset =
(FLAG_enable_embedded_constant_pool.value() ? 4 : 3) * kSystemPointerSize;
constexpr int kTierupBudgetOffset =
(FLAG_enable_embedded_constant_pool.value() ? 5 : 4) * kSystemPointerSize;
inline MemOperand GetHalfStackSlot(int offset, RegPairHalf half) {
int32_t half_offset =
@ -222,7 +220,7 @@ void LiftoffAssembler::AbortCompilation() { FinishCode(); }
// static
constexpr int LiftoffAssembler::StaticStackFrameSize() {
return liftoff::kTierupBudgetOffset;
return liftoff::kFeedbackVectorOffset;
}
int LiftoffAssembler::SlotSizeForType(ValueKind kind) {

3
src/wasm/baseline/riscv64/liftoff-assembler-riscv64.h
View File

@ -72,7 +72,6 @@ inline constexpr Condition ToCondition(LiftoffCondition liftoff_cond) {
// fp-8 holds the stack marker, fp-16 is the instance parameter.
constexpr int kInstanceOffset = 2 * kSystemPointerSize;
constexpr int kFeedbackVectorOffset = 3 * kSystemPointerSize;
constexpr int kTierupBudgetOffset = 4 * kSystemPointerSize;
inline MemOperand GetStackSlot(int offset) { return MemOperand(fp, -offset); }
@ -393,7 +392,7 @@ void LiftoffAssembler::AbortCompilation() { AbortedCodeGeneration(); }
// static
constexpr int LiftoffAssembler::StaticStackFrameSize() {
return liftoff::kTierupBudgetOffset;
return liftoff::kFeedbackVectorOffset;
}
int LiftoffAssembler::SlotSizeForType(ValueKind kind) {

3
src/wasm/baseline/s390/liftoff-assembler-s390.h
View File

@ -85,7 +85,6 @@ inline constexpr bool UseSignedOp(LiftoffCondition liftoff_cond) {
//
constexpr int32_t kInstanceOffset = 2 * kSystemPointerSize;
constexpr int kFeedbackVectorOffset = 3 * kSystemPointerSize;
constexpr int kTierupBudgetOffset = 4 * kSystemPointerSize;
inline MemOperand GetStackSlot(uint32_t offset) {
return MemOperand(fp, -offset);
}
@ -200,7 +199,7 @@ void LiftoffAssembler::AbortCompilation() { AbortedCodeGeneration(); }
// static
constexpr int LiftoffAssembler::StaticStackFrameSize() {
return liftoff::kTierupBudgetOffset;
return liftoff::kFeedbackVectorOffset;
}
int LiftoffAssembler::SlotSizeForType(ValueKind kind) {

2
src/wasm/baseline/x64/liftoff-assembler-x64.h
View File

@ -69,8 +69,6 @@ static_assert((kLiftoffAssemblerFpCacheRegs &
constexpr int kInstanceOffset = 16;
// rbp-24 is the feedback vector.
constexpr int kFeedbackVectorOffset = 24;
// rbp-32 is the remaining tier-up budget.
constexpr int kTierupBudgetOffset = 32;
inline constexpr Operand GetStackSlot(int offset) {
return Operand(rbp, -offset);

2
src/wasm/wasm-constants.h
View File

@ -190,7 +190,7 @@ constexpr uint32_t kMinimumSupertypeArraySize = 3;
constexpr int kMaxPolymorphism = 4;
#if V8_TARGET_ARCH_X64
constexpr int32_t kOSRTargetOffset = 5 * kSystemPointerSize;
constexpr int32_t kOSRTargetOffset = 4 * kSystemPointerSize;
#endif
} // namespace wasm