69 lines · cpp
1//===-- GPU benchmark for log ---------------------------------------------===//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/__support/sign.h"14#include "src/math/log.h"15 16#if defined(NVPTX_MATH_FOUND) || defined(AMDGPU_MATH_FOUND)17#include "platform.h"18#endif19 20#define RANDOM_INPUT_UniformExponent(T, Func, Min, Max, N) \21 [](uint32_t call_index) { \22 using namespace LIBC_NAMESPACE::benchmarks; \23 \24 const UniformExponent<T> dist(Min, Max, LIBC_NAMESPACE::Sign::POS); \25 return MathPerf<T>::template run_throughput<N>(Func, dist, call_index); \26 }27 28#define RANDOM_INPUT_UniformLinear(T, Func, Min, Max, N) \29 [](uint32_t call_index) { \30 using namespace LIBC_NAMESPACE::benchmarks; \31 \32 const UniformLinear<T> dist(Min, Max); \33 return MathPerf<T>::template run_throughput<N>(Func, dist, call_index); \34 }35 36#define BENCH(T, Name, Func, Dist, Min, Max) \37 SINGLE_WAVE_BENCHMARK(LlvmLibcLogGpuBenchmark, Name##_1, \38 RANDOM_INPUT_##Dist(T, Func, Min, Max, 1)); \39 SINGLE_WAVE_BENCHMARK(LlvmLibcLogGpuBenchmark, Name##_128, \40 RANDOM_INPUT_##Dist(T, Func, Min, Max, 128)); \41 SINGLE_WAVE_BENCHMARK(LlvmLibcLogGpuBenchmark, Name##_1024, \42 RANDOM_INPUT_##Dist(T, Func, Min, Max, 1024)); \43 SINGLE_WAVE_BENCHMARK(LlvmLibcLogGpuBenchmark, Name##_4096, \44 RANDOM_INPUT_##Dist(T, Func, Min, Max, 4096))45 46using LIBC_NAMESPACE::log;47 48static constexpr double INV_E = 0x1.78b56362cef38p-2; // exp(-1.0)49static constexpr double E = 0x1.5bf0a8b145769p+1; // exp(+1.0)50 51BENCH(double, LogSubnormal, log, UniformExponent, -1022, -1022);52BENCH(double, LogAroundOne, log, UniformLinear, INV_E, E);53BENCH(double, LogMedMag, log, UniformExponent, -10, 10);54BENCH(double, LogNormal, log, UniformExponent, -1021, 1023);55 56#ifdef NVPTX_MATH_FOUND57BENCH(double, NvLogSubnormal, __nv_log, UniformExponent, -1022, -1022);58BENCH(double, NvLogAroundOne, __nv_log, UniformLinear, INV_E, E);59BENCH(double, NvLogMedMag, __nv_log, UniformExponent, -10, 10);60BENCH(double, NvLogNormal, __nv_log, UniformExponent, -1021, 1023);61#endif62 63#ifdef AMDGPU_MATH_FOUND64BENCH(double, AmdLogSubnormal, __ocml_log_f64, UniformExponent, -1022, -1022);65BENCH(double, AmdLogAroundOne, __ocml_log_f64, UniformLinear, INV_E, E);66BENCH(double, AmdLogMedMag, __ocml_log_f64, UniformExponent, -10, 10);67BENCH(double, AmdLogNormal, __ocml_log_f64, UniformExponent, -1021, 1023);68#endif69