[libc-commits] [libc] [libc] Add N Threads Benchmark Helper (PR #99834)
via libc-commits
libc-commits at lists.llvm.org
Sun Jul 21 19:34:51 PDT 2024
https://github.com/jameshu15869 created https://github.com/llvm/llvm-project/pull/99834
This PR adds a `BENCHMARK_N_THREADS()` helper to register benchmarks with a specific number of threads. This PR replaces the flags used originally to allow any amount of threads.
>From 19b1100b902d7c93e0c447c5e2832a54ea7f1f05 Mon Sep 17 00:00:00 2001
From: jameshu15869 <jhudson15869 at gmail.com>
Date: Sun, 21 Jul 2024 22:33:17 -0400
Subject: [PATCH] add n threads benchmark helper
---
libc/benchmarks/gpu/LibcGpuBenchmark.cpp | 5 +----
libc/benchmarks/gpu/LibcGpuBenchmark.h | 26 +++++++++++++-----------
2 files changed, 15 insertions(+), 16 deletions(-)
diff --git a/libc/benchmarks/gpu/LibcGpuBenchmark.cpp b/libc/benchmarks/gpu/LibcGpuBenchmark.cpp
index 59de18c20417d..ac139cc21b0b6 100644
--- a/libc/benchmarks/gpu/LibcGpuBenchmark.cpp
+++ b/libc/benchmarks/gpu/LibcGpuBenchmark.cpp
@@ -151,10 +151,7 @@ void Benchmark::run_benchmarks() {
all_results.reset();
gpu::sync_threads();
- if (!b->flags ||
- ((b->flags & BenchmarkFlags::SINGLE_THREADED) && id == 0) ||
- ((b->flags & BenchmarkFlags::SINGLE_WAVE) &&
- id < gpu::get_lane_size())) {
+ if (b->num_threads == static_cast<uint32_t>(-1) || id < b->num_threads) {
auto current_result = b->run();
all_results.update(current_result);
}
diff --git a/libc/benchmarks/gpu/LibcGpuBenchmark.h b/libc/benchmarks/gpu/LibcGpuBenchmark.h
index c07fab9ccfbe3..f5cf4822f6fd3 100644
--- a/libc/benchmarks/gpu/LibcGpuBenchmark.h
+++ b/libc/benchmarks/gpu/LibcGpuBenchmark.h
@@ -74,8 +74,6 @@ struct BenchmarkResult {
clock_t total_time = 0;
};
-enum BenchmarkFlags { SINGLE_THREADED = 0x1, SINGLE_WAVE = 0x2 };
-
BenchmarkResult benchmark(const BenchmarkOptions &options,
cpp::function<uint64_t(void)> wrapper_func);
@@ -83,12 +81,13 @@ class Benchmark {
const cpp::function<uint64_t(void)> func;
const cpp::string_view suite_name;
const cpp::string_view test_name;
- const uint8_t flags;
+ const uint32_t num_threads;
public:
Benchmark(cpp::function<uint64_t(void)> func, char const *suite_name,
- char const *test_name, uint8_t flags)
- : func(func), suite_name(suite_name), test_name(test_name), flags(flags) {
+ char const *test_name, uint32_t num_threads)
+ : func(func), suite_name(suite_name), test_name(test_name),
+ num_threads(num_threads) {
add_benchmark(this);
}
@@ -108,18 +107,21 @@ class Benchmark {
} // namespace benchmarks
} // namespace LIBC_NAMESPACE_DECL
+// Passing -1 indicates the benchmark should be run with as many threads as
+// allocated by the user in the benchmark's CMake.
#define BENCHMARK(SuiteName, TestName, Func) \
LIBC_NAMESPACE::benchmarks::Benchmark SuiteName##_##TestName##_Instance( \
- Func, #SuiteName, #TestName, 0)
+ Func, #SuiteName, #TestName, -1)
-#define SINGLE_THREADED_BENCHMARK(SuiteName, TestName, Func) \
+#define BENCHMARK_N_THREADS(SuiteName, TestName, Func, NumThreads) \
LIBC_NAMESPACE::benchmarks::Benchmark SuiteName##_##TestName##_Instance( \
- Func, #SuiteName, #TestName, \
- LIBC_NAMESPACE::benchmarks::BenchmarkFlags::SINGLE_THREADED)
+ Func, #SuiteName, #TestName, NumThreads)
+
+#define SINGLE_THREADED_BENCHMARK(SuiteName, TestName, Func) \
+ BENCHMARK_N_THREADS(SuiteName, TestName, Func, 1)
#define SINGLE_WAVE_BENCHMARK(SuiteName, TestName, Func) \
- LIBC_NAMESPACE::benchmarks::Benchmark SuiteName##_##TestName##_Instance( \
- Func, #SuiteName, #TestName, \
- LIBC_NAMESPACE::benchmarks::BenchmarkFlags::SINGLE_WAVE)
+ BENCHMARK_N_THREADS(SuiteName, TestName, Func, \
+ LIBC_NAMESPACE::gpu::get_lane_size())
#endif
More information about the libc-commits
mailing list