125 lines
4.6 KiB
C++
125 lines
4.6 KiB
C++
// Copyright 2024 Google LLC
|
|
//
|
|
// This source code is licensed under the BSD-style license found in the
|
|
// LICENSE file in the root directory of this source tree.
|
|
//
|
|
// Auto-generated file. Do not edit!
|
|
// Specification: test/f16-rsum.yaml
|
|
// Generator: tools/generate-rdsum-benchmark.py
|
|
|
|
#include "rsum-benchmark.h"
|
|
#include "utils.h"
|
|
#include <benchmark/benchmark.h>
|
|
|
|
#include "xnnpack.h"
|
|
#include "xnnpack/buffer.h"
|
|
#include "xnnpack/common.h"
|
|
#include "xnnpack/reduce.h"
|
|
#include "xnnpack/microfnptr.h"
|
|
#include "xnnpack/microparams-init.h"
|
|
|
|
|
|
#if XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)
|
|
BENCHMARK_CAPTURE(f16_rsum, neonfp16arith_u8,
|
|
xnn_f16_rsum_ukernel__neonfp16arith_u8,
|
|
xnn_init_f16_scale_scalar_params,
|
|
benchmark::utils::CheckNEONFP16ARITH)
|
|
->Apply(BenchmarkRSUM)
|
|
->UseRealTime();
|
|
#endif // XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)
|
|
|
|
|
|
#if XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)
|
|
BENCHMARK_CAPTURE(f16_rsum, neonfp16arith_u16_acc2,
|
|
xnn_f16_rsum_ukernel__neonfp16arith_u16_acc2,
|
|
xnn_init_f16_scale_scalar_params,
|
|
benchmark::utils::CheckNEONFP16ARITH)
|
|
->Apply(BenchmarkRSUM)
|
|
->UseRealTime();
|
|
#endif // XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)
|
|
|
|
|
|
#if XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)
|
|
BENCHMARK_CAPTURE(f16_rsum, neonfp16arith_u24_acc3,
|
|
xnn_f16_rsum_ukernel__neonfp16arith_u24_acc3,
|
|
xnn_init_f16_scale_scalar_params,
|
|
benchmark::utils::CheckNEONFP16ARITH)
|
|
->Apply(BenchmarkRSUM)
|
|
->UseRealTime();
|
|
#endif // XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)
|
|
|
|
|
|
#if XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)
|
|
BENCHMARK_CAPTURE(f16_rsum, neonfp16arith_u32_acc2,
|
|
xnn_f16_rsum_ukernel__neonfp16arith_u32_acc2,
|
|
xnn_init_f16_scale_scalar_params,
|
|
benchmark::utils::CheckNEONFP16ARITH)
|
|
->Apply(BenchmarkRSUM)
|
|
->UseRealTime();
|
|
#endif // XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)
|
|
|
|
|
|
#if XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)
|
|
BENCHMARK_CAPTURE(f16_rsum, neonfp16arith_u32_acc4,
|
|
xnn_f16_rsum_ukernel__neonfp16arith_u32_acc4,
|
|
xnn_init_f16_scale_scalar_params,
|
|
benchmark::utils::CheckNEONFP16ARITH)
|
|
->Apply(BenchmarkRSUM)
|
|
->UseRealTime();
|
|
#endif // XNN_ENABLE_ARM_FP16_VECTOR && (XNN_ARCH_ARM || XNN_ARCH_ARM64)
|
|
|
|
|
|
#if XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
|
|
BENCHMARK_CAPTURE(f16_rsum, avx512fp16_u32,
|
|
xnn_f16_rsum_ukernel__avx512fp16_u32,
|
|
xnn_init_f16_scale_scalar_params,
|
|
benchmark::utils::CheckAVX512FP16)
|
|
->Apply(BenchmarkRSUM)
|
|
->UseRealTime();
|
|
#endif // XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
|
|
|
|
|
|
#if XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
|
|
BENCHMARK_CAPTURE(f16_rsum, avx512fp16_u64_acc2,
|
|
xnn_f16_rsum_ukernel__avx512fp16_u64_acc2,
|
|
xnn_init_f16_scale_scalar_params,
|
|
benchmark::utils::CheckAVX512FP16)
|
|
->Apply(BenchmarkRSUM)
|
|
->UseRealTime();
|
|
#endif // XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
|
|
|
|
|
|
#if XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
|
|
BENCHMARK_CAPTURE(f16_rsum, avx512fp16_u96_acc3,
|
|
xnn_f16_rsum_ukernel__avx512fp16_u96_acc3,
|
|
xnn_init_f16_scale_scalar_params,
|
|
benchmark::utils::CheckAVX512FP16)
|
|
->Apply(BenchmarkRSUM)
|
|
->UseRealTime();
|
|
#endif // XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
|
|
|
|
|
|
#if XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
|
|
BENCHMARK_CAPTURE(f16_rsum, avx512fp16_u128_acc2,
|
|
xnn_f16_rsum_ukernel__avx512fp16_u128_acc2,
|
|
xnn_init_f16_scale_scalar_params,
|
|
benchmark::utils::CheckAVX512FP16)
|
|
->Apply(BenchmarkRSUM)
|
|
->UseRealTime();
|
|
#endif // XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
|
|
|
|
|
|
#if XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
|
|
BENCHMARK_CAPTURE(f16_rsum, avx512fp16_u128_acc4,
|
|
xnn_f16_rsum_ukernel__avx512fp16_u128_acc4,
|
|
xnn_init_f16_scale_scalar_params,
|
|
benchmark::utils::CheckAVX512FP16)
|
|
->Apply(BenchmarkRSUM)
|
|
->UseRealTime();
|
|
#endif // XNN_ENABLE_AVX512FP16 && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
|
|
|
|
|
|
#ifndef XNNPACK_BENCHMARK_NO_MAIN
|
|
BENCHMARK_MAIN();
|
|
#endif
|