63 lines · cpp
1//===-- GPU benchmark for logf16 ------------------------------------------===//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/macros/properties/types.h"14#include "src/__support/sign.h"15#include "src/math/logf16.h"16 17#if defined(NVPTX_MATH_FOUND) || defined(AMDGPU_MATH_FOUND)18#include "platform.h"19#endif20 21#define RANDOM_INPUT_UniformExponent(T, Func, Min, Max, N) \22 [](uint32_t call_index) { \23 using namespace LIBC_NAMESPACE::benchmarks; \24 \25 const UniformExponent<T> dist(Min, Max, LIBC_NAMESPACE::Sign::POS); \26 return MathPerf<T>::template run_throughput<N>(Func, dist, call_index); \27 }28 29#define RANDOM_INPUT_UniformLinear(T, Func, Min, Max, N) \30 [](uint32_t call_index) { \31 using namespace LIBC_NAMESPACE::benchmarks; \32 \33 const UniformLinear<T> dist(Min, Max); \34 return MathPerf<T>::template run_throughput<N>(Func, dist, call_index); \35 }36 37#define BENCH(T, Name, Func, Dist, Min, Max) \38 SINGLE_WAVE_BENCHMARK(LlvmLibcLogf16GpuBenchmark, Name##_1, \39 RANDOM_INPUT_##Dist(T, Func, Min, Max, 1)); \40 SINGLE_WAVE_BENCHMARK(LlvmLibcLogf16GpuBenchmark, Name##_128, \41 RANDOM_INPUT_##Dist(T, Func, Min, Max, 128)); \42 SINGLE_WAVE_BENCHMARK(LlvmLibcLogf16GpuBenchmark, Name##_1024, \43 RANDOM_INPUT_##Dist(T, Func, Min, Max, 1024)); \44 SINGLE_WAVE_BENCHMARK(LlvmLibcLogf16GpuBenchmark, Name##_4096, \45 RANDOM_INPUT_##Dist(T, Func, Min, Max, 4096))46 47using LIBC_NAMESPACE::logf16;48 49static constexpr float16 INV_E = 0x1.78b56362cef38p-2f16; // exp(-1.0)50static constexpr float16 E = 0x1.5bf0a8b145769p+1f16; // exp(+1.0)51 52BENCH(float16, Logf16Subnormal, logf16, UniformExponent, -14, -14);53BENCH(float16, Logf16AroundOne, logf16, UniformLinear, INV_E, E);54BENCH(float16, Logf16MedMag, logf16, UniformExponent, -10, 10);55BENCH(float16, Logf16Normal, logf16, UniformExponent, -13, 15);56 57#ifdef AMDGPU_MATH_FOUND58BENCH(float16, AmdLogf16Subnormal, __ocml_log_f16, UniformExponent, -14, -14);59BENCH(float16, AmdLogf16AroundOne, __ocml_log_f16, UniformLinear, INV_E, E);60BENCH(float16, AmdLogf16MedMag, __ocml_log_f16, UniformExponent, -10, 10);61BENCH(float16, AmdLogf16Normal, __ocml_log_f16, UniformExponent, -13, 15);62#endif63