-
Notifications
You must be signed in to change notification settings - Fork 390
[MicroBenchmarks] Add initial early-exit vectorization microbenchmarks. #317
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base: main
Are you sure you want to change the base?
Changes from all commits
File filter
Filter by extension
Conversations
Jump to
Diff view
Diff view
There are no files selected for viewing
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,196 @@ | ||
| //===- EarlyExit.cpp - Early Exit Vectorization Benchmarks ------*- C++ -*-===// | ||
| /// | ||
| /// \file | ||
| /// Benchmarks for early exit loop vectorization with varying trip counts | ||
| /// and element types. | ||
| /// | ||
| //===----------------------------------------------------------------------===// | ||
| #include <algorithm> | ||
| #include <cstdint> | ||
| #include <memory> | ||
|
|
||
| #include "benchmark/benchmark.h" | ||
|
|
||
| // Check for builtin availability using __has_builtin when available. | ||
| #ifndef __has_builtin | ||
| #define __has_builtin(x) 0 | ||
| #endif | ||
|
|
||
| #if __has_builtin(__builtin_assume_aligned) | ||
| #define ASSUME_ALIGNED(ptr, align) __builtin_assume_aligned(ptr, align) | ||
| #else | ||
| #define ASSUME_ALIGNED(ptr, align) (ptr) | ||
| #endif | ||
|
|
||
| #if __has_builtin(__builtin_assume_dereferenceable) | ||
| #define ASSUME_DEREFERENCEABLE(ptr, size) __builtin_assume_dereferenceable(ptr, size) | ||
| #else | ||
| #define ASSUME_DEREFERENCEABLE(ptr, size) ((void)0) | ||
| #endif | ||
|
|
||
| uint64_t Sum = 0; | ||
|
|
||
| /// Initialize array \p A with 1s. | ||
| template <typename Ty> | ||
| static void init_data(const std::unique_ptr<Ty[]> &A, unsigned N) { | ||
| std::fill_n(A.get(), N, static_cast<Ty>(1)); | ||
| } | ||
|
|
||
| // Helper to block optimizing \p F based on its arguments. | ||
| template <typename F, typename... Args> | ||
| __attribute__((optnone)) static uint64_t callThroughOptnone(F &&f, | ||
| Args &&...args) { | ||
| return f(std::forward<Args>(args)...); | ||
| } | ||
|
|
||
| enum class EarlyExitPosition { None, First, Mid }; | ||
|
|
||
| template <typename Ty> | ||
| static void __attribute__((always_inline)) | ||
| runEarlyExitBenchmark(benchmark::State &state, | ||
| uint64_t (*Fn)(Ty *, Ty *, Ty *, int), | ||
| EarlyExitPosition ExitPos) { | ||
| auto Iterations = state.range(0); | ||
| std::unique_ptr<Ty[]> A(new Ty[Iterations]); | ||
| std::unique_ptr<Ty[]> B(new Ty[Iterations]); | ||
| std::unique_ptr<Ty[]> C(new Ty[Iterations]); | ||
| init_data(A, Iterations); | ||
| init_data(B, Iterations); | ||
| init_data(C, Iterations); | ||
|
|
||
| switch (ExitPos) { | ||
| case EarlyExitPosition::None: | ||
| break; | ||
| case EarlyExitPosition::First: | ||
| A[0] = 0; | ||
| break; | ||
| case EarlyExitPosition::Mid: | ||
| A[Iterations / 2] = 0; | ||
| break; | ||
| } | ||
|
|
||
| for (auto _ : state) { | ||
| benchmark::DoNotOptimize(A); | ||
| benchmark::DoNotOptimize(B); | ||
| benchmark::DoNotOptimize(C); | ||
| benchmark::DoNotOptimize(Sum); | ||
| benchmark::ClobberMemory(); | ||
| Sum += callThroughOptnone(Fn, &A[0], &B[0], &C[0], Iterations); | ||
| } | ||
| } | ||
|
|
||
| template <typename Ty> | ||
| static uint64_t __attribute__((noinline)) | ||
| singleEarlyExitWithSingleLoad(Ty *A, Ty *B, Ty *C, int Iterations) { | ||
| auto AlignedA = static_cast<Ty *>(ASSUME_ALIGNED(A, sizeof(Ty))); | ||
| ASSUME_DEREFERENCEABLE(AlignedA, Iterations * sizeof(Ty)); | ||
| for (int J = 0; J < Iterations; J++) { | ||
| if (AlignedA[J] == 0) | ||
| return J; | ||
| } | ||
| return 0; | ||
| } | ||
|
|
||
| template <typename Ty> | ||
| static uint64_t __attribute__((noinline)) | ||
| singleEarlyExitWithTwoLoads(Ty *A, Ty *B, Ty *C, int Iterations) { | ||
| auto AlignedA = static_cast<Ty *>(ASSUME_ALIGNED(A, sizeof(Ty))); | ||
| auto AlignedB = static_cast<Ty *>(ASSUME_ALIGNED(B, sizeof(Ty))); | ||
| ASSUME_DEREFERENCEABLE(AlignedA, Iterations * sizeof(Ty)); | ||
| ASSUME_DEREFERENCEABLE(AlignedB, Iterations * sizeof(Ty)); | ||
| for (int J = 0; J < Iterations; J++) { | ||
| if (AlignedA[J] != AlignedB[J]) | ||
| return J; | ||
| } | ||
| return 0; | ||
| } | ||
|
|
||
| template <typename Ty> | ||
| static uint64_t __attribute__((noinline)) | ||
| singleEarlyExitWithThreeLoadsAndCompute(Ty *A, Ty *B, Ty *C, int Iterations) { | ||
| auto AlignedA = static_cast<Ty *>(ASSUME_ALIGNED(A, sizeof(Ty))); | ||
| auto AlignedB = static_cast<Ty *>(ASSUME_ALIGNED(B, sizeof(Ty))); | ||
| auto AlignedC = static_cast<Ty *>(ASSUME_ALIGNED(C, sizeof(Ty))); | ||
| ASSUME_DEREFERENCEABLE(AlignedA, Iterations * sizeof(Ty)); | ||
| ASSUME_DEREFERENCEABLE(AlignedB, Iterations * sizeof(Ty)); | ||
| ASSUME_DEREFERENCEABLE(AlignedC, Iterations * sizeof(Ty)); | ||
| for (int J = 0; J < Iterations; J++) { | ||
| // Exits when A[J] == 0: (0 + 1) * 1 == 1 | ||
| // Continues when A[J] == 1: (1 + 1) * 1 == 2 | ||
| if ((AlignedA[J] + AlignedB[J]) * AlignedC[J] == 1) | ||
| return J; | ||
| } | ||
| return 0; | ||
| } | ||
|
|
||
| template <typename Ty> | ||
| void autovec_no_early_exit_single_load(benchmark::State &state) { | ||
| runEarlyExitBenchmark<Ty>(state, &singleEarlyExitWithSingleLoad<Ty>, | ||
| EarlyExitPosition::None); | ||
| } | ||
|
|
||
| template <typename Ty> | ||
| void autovec_no_early_exit_two_loads(benchmark::State &state) { | ||
| runEarlyExitBenchmark<Ty>(state, &singleEarlyExitWithTwoLoads<Ty>, | ||
| EarlyExitPosition::None); | ||
| } | ||
|
|
||
| template <typename Ty> | ||
| void autovec_early_exit_taken_first_single_load(benchmark::State &state) { | ||
| runEarlyExitBenchmark<Ty>(state, &singleEarlyExitWithSingleLoad<Ty>, | ||
| EarlyExitPosition::First); | ||
| } | ||
|
|
||
| template <typename Ty> | ||
| void autovec_early_exit_taken_mid_single_load(benchmark::State &state) { | ||
| runEarlyExitBenchmark<Ty>(state, &singleEarlyExitWithSingleLoad<Ty>, | ||
| EarlyExitPosition::Mid); | ||
| } | ||
|
|
||
| template <typename Ty> | ||
| void autovec_early_exit_taken_first_two_loads(benchmark::State &state) { | ||
| runEarlyExitBenchmark<Ty>(state, &singleEarlyExitWithTwoLoads<Ty>, | ||
| EarlyExitPosition::First); | ||
| } | ||
|
|
||
| template <typename Ty> | ||
| void autovec_early_exit_taken_mid_two_loads(benchmark::State &state) { | ||
| runEarlyExitBenchmark<Ty>(state, &singleEarlyExitWithTwoLoads<Ty>, | ||
| EarlyExitPosition::Mid); | ||
| } | ||
|
|
||
| template <typename Ty> | ||
| void autovec_no_early_exit_three_loads(benchmark::State &state) { | ||
| runEarlyExitBenchmark<Ty>(state, &singleEarlyExitWithThreeLoadsAndCompute<Ty>, | ||
| EarlyExitPosition::None); | ||
| } | ||
|
|
||
| template <typename Ty> | ||
| void autovec_early_exit_taken_first_three_loads(benchmark::State &state) { | ||
| runEarlyExitBenchmark<Ty>(state, &singleEarlyExitWithThreeLoadsAndCompute<Ty>, | ||
| EarlyExitPosition::First); | ||
| } | ||
|
|
||
| template <typename Ty> | ||
| void autovec_early_exit_taken_mid_three_loads(benchmark::State &state) { | ||
| runEarlyExitBenchmark<Ty>(state, &singleEarlyExitWithThreeLoadsAndCompute<Ty>, | ||
| EarlyExitPosition::Mid); | ||
| } | ||
|
|
||
| #define BENCHMARK_ARGS ->Arg(10)->Arg(16)->Arg(28)->Arg(51)->Arg(256)->Arg(999) | ||
|
|
||
| #define REGISTER_BENCHMARK_FOR_TYPES(BenchFunc) \ | ||
| BENCHMARK_TEMPLATE(BenchFunc, uint8_t) BENCHMARK_ARGS; \ | ||
| BENCHMARK_TEMPLATE(BenchFunc, uint16_t) BENCHMARK_ARGS; \ | ||
| BENCHMARK_TEMPLATE(BenchFunc, uint32_t) BENCHMARK_ARGS; \ | ||
| BENCHMARK_TEMPLATE(BenchFunc, uint64_t) BENCHMARK_ARGS; | ||
|
|
||
| REGISTER_BENCHMARK_FOR_TYPES(autovec_no_early_exit_single_load) | ||
| REGISTER_BENCHMARK_FOR_TYPES(autovec_early_exit_taken_first_single_load) | ||
|
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Any reason why there's no
Contributor
Author
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Just an oversight, registered |
||
| REGISTER_BENCHMARK_FOR_TYPES(autovec_early_exit_taken_mid_single_load) | ||
| REGISTER_BENCHMARK_FOR_TYPES(autovec_no_early_exit_two_loads) | ||
| REGISTER_BENCHMARK_FOR_TYPES(autovec_early_exit_taken_first_two_loads) | ||
| REGISTER_BENCHMARK_FOR_TYPES(autovec_early_exit_taken_mid_two_loads) | ||
| REGISTER_BENCHMARK_FOR_TYPES(autovec_no_early_exit_three_loads) | ||
| REGISTER_BENCHMARK_FOR_TYPES(autovec_early_exit_taken_first_three_loads) | ||
| REGISTER_BENCHMARK_FOR_TYPES(autovec_early_exit_taken_mid_three_loads) | ||
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Why is
Suma global?There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This may just be a bit paranoid, to ensure the result of the loops are used
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
benchmark::DoNotOptimize(Sum);does this already, no?