60 lines · cpp
1//===-- GPU benchmark for expf --------------------------------------------===//2//3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.4// See https://llvm.org/LICENSE.txt for license information.5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception6//7//===----------------------------------------------------------------------===//8 9#include "benchmarks/gpu/LibcGpuBenchmark.h"10#include "benchmarks/gpu/Random.h"11 12#include "hdr/stdint_proxy.h"13#include "src/math/expf.h"14 15#if defined(NVPTX_MATH_FOUND) || defined(AMDGPU_MATH_FOUND)16#include "platform.h"17#endif18 19#define RANDOM_INPUT(T, Func, Dist, Min, Max, N) \20 [](uint32_t call_index) { \21 using namespace LIBC_NAMESPACE::benchmarks; \22 \23 const Dist<T> dist(Min, Max); \24 return MathPerf<T>::template run_throughput<N>(Func, dist, call_index); \25 }26 27#define BENCH(T, Name, Func, Dist, Min, Max) \28 SINGLE_WAVE_BENCHMARK(LlvmLibcExpfGpuBenchmark, Name##_1, \29 RANDOM_INPUT(T, Func, Dist, Min, Max, 1)); \30 SINGLE_WAVE_BENCHMARK(LlvmLibcExpfGpuBenchmark, Name##_128, \31 RANDOM_INPUT(T, Func, Dist, Min, Max, 128)); \32 SINGLE_WAVE_BENCHMARK(LlvmLibcExpfGpuBenchmark, Name##_1024, \33 RANDOM_INPUT(T, Func, Dist, Min, Max, 1024)); \34 SINGLE_WAVE_BENCHMARK(LlvmLibcExpfGpuBenchmark, Name##_4096, \35 RANDOM_INPUT(T, Func, Dist, Min, Max, 4096))36 37using LIBC_NAMESPACE::expf;38 39BENCH(float, ExpfSubnormal, expf, UniformExponent, -126, -126);40BENCH(float, ExpfCoreRange, expf, UniformLinear, -10.0f, 10.0f);41BENCH(float, ExpfFinite, expf, UniformLinear, -103.0f, 88.0f);42BENCH(float, ExpfUnderflow, expf, UniformLinear, -104.0f, -103.0f);43BENCH(float, ExpfOverflow, expf, UniformLinear, 88.0f, 89.0f);44 45#ifdef NVPTX_MATH_FOUND46BENCH(float, NvExpfSubnormal, __nv_expf, UniformExponent, -126, -126);47BENCH(float, NvExpfCoreRange, __nv_expf, UniformLinear, -10.0f, 10.0f);48BENCH(float, NvExpfFinite, __nv_expf, UniformLinear, -103.0f, 88.0f);49BENCH(float, NvExpfUnderflow, __nv_expf, UniformLinear, -104.0f, -103.0f);50BENCH(float, NvExpfOverflow, __nv_expf, UniformLinear, 88.0f, 89.0f);51#endif52 53#ifdef AMDGPU_MATH_FOUND54BENCH(float, AmdExpfSubnormal, __ocml_exp_f32, UniformExponent, -126, -126);55BENCH(float, AmdExpfCoreRange, __ocml_exp_f32, UniformLinear, -10.0f, 10.0f);56BENCH(float, AmdExpfFinite, __ocml_exp_f32, UniformLinear, -103.0f, 88.0f);57BENCH(float, AmdExpfUnderflow, __ocml_exp_f32, UniformLinear, -104.0f, -103.0f);58BENCH(float, AmdExpfOverflow, __ocml_exp_f32, UniformLinear, 88.0f, 89.0f);59#endif60