1286 lines
50 KiB
C++
1286 lines
50 KiB
C++
// Copyright 2023 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/qd8-f32-qb4w-gemm-minmax.yaml
|
|
// Generator: tools/generate-gemm-test.py
|
|
|
|
#include <benchmark/benchmark.h>
|
|
#include "gemm-benchmark.h"
|
|
#include "utils.h"
|
|
#include "xnnpack/common.h"
|
|
#include "xnnpack/gemm.h"
|
|
#include "xnnpack/isa-checks.h"
|
|
#include "xnnpack/microfnptr.h"
|
|
#include "xnnpack/microparams-init.h"
|
|
#include "xnnpack/pack.h"
|
|
#include "xnnpack/packw.h"
|
|
|
|
|
|
#if XNN_ENABLE_ARM_I8MM && XNN_ARCH_ARM64
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x8c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x8c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/8, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x8c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x32c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x32c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/32, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x32c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_2x8c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x8c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/2, /*nr=*/8, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_2x8c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_2x16c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x16c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/2, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_2x16c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_2x32c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x32c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/2, /*nr=*/32, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_2x32c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_3x8c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x8c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/3, /*nr=*/8, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_3x8c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_3x16c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x16c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/3, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_3x16c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_3x32c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x32c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/3, /*nr=*/32, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_3x32c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_4x8c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x8c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/4, /*nr=*/8, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_4x8c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_4x16c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x16c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/4, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_4x16c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_4x32c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x32c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/4, /*nr=*/32, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_4x32c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_5x8c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x8c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/5, /*nr=*/8, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_5x8c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/5, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_5x32c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x32c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/5, /*nr=*/32, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_5x32c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_6x8c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_6x8c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/6, /*nr=*/8, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_6x8c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_6x16c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_6x16c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/6, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_6x16c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_6x32c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_6x32c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/6, /*nr=*/32, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_6x32c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_7x8c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_7x8c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/7, /*nr=*/8, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_7x8c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/7, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_7x32c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_7x32c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/7, /*nr=*/32, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_7x32c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_8x8c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_8x8c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/8, /*nr=*/8, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_8x8c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/8, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__neoni8mm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_8x32c8__neoni8mm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_8x32c8__neoni8mm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/8, /*nr=*/32, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckNEONI8MM);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_8x32c8__neoni8mm)
|
|
#endif // XNN_ENABLE_ARM_I8MM && XNN_ARCH_ARM64
|
|
|
|
|
|
#if XNN_ENABLE_ARM_DOTPROD && (XNN_ARCH_ARM || XNN_ARCH_ARM64)
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x8c4__neondot(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x8c4__neondot,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/8, /*kr=*/4, /*sr=*/1,
|
|
benchmark::utils::CheckNEONDOT);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x8c4__neondot)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x16c4__neondot(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x16c4__neondot,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/16, /*kr=*/4, /*sr=*/1,
|
|
benchmark::utils::CheckNEONDOT);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x16c4__neondot)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_2x8c4__neondot(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x8c4__neondot,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/2, /*nr=*/8, /*kr=*/4, /*sr=*/1,
|
|
benchmark::utils::CheckNEONDOT);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_2x8c4__neondot)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_2x16c4__neondot(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x16c4__neondot,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/2, /*nr=*/16, /*kr=*/4, /*sr=*/1,
|
|
benchmark::utils::CheckNEONDOT);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_2x16c4__neondot)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_3x8c4__neondot(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x8c4__neondot,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/3, /*nr=*/8, /*kr=*/4, /*sr=*/1,
|
|
benchmark::utils::CheckNEONDOT);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_3x8c4__neondot)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_3x16c4__neondot(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x16c4__neondot,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/3, /*nr=*/16, /*kr=*/4, /*sr=*/1,
|
|
benchmark::utils::CheckNEONDOT);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_3x16c4__neondot)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_4x8c4__neondot(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x8c4__neondot,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/4, /*nr=*/8, /*kr=*/4, /*sr=*/1,
|
|
benchmark::utils::CheckNEONDOT);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_4x8c4__neondot)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_4x16c4__neondot(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x16c4__neondot,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/4, /*nr=*/16, /*kr=*/4, /*sr=*/1,
|
|
benchmark::utils::CheckNEONDOT);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_4x16c4__neondot)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_5x8c4__neondot(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x8c4__neondot,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/5, /*nr=*/8, /*kr=*/4, /*sr=*/1,
|
|
benchmark::utils::CheckNEONDOT);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_5x8c4__neondot)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_5x16c4__neondot(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x16c4__neondot,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/5, /*nr=*/16, /*kr=*/4, /*sr=*/1,
|
|
benchmark::utils::CheckNEONDOT);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_5x16c4__neondot)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_6x8c4__neondot(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_6x8c4__neondot,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/6, /*nr=*/8, /*kr=*/4, /*sr=*/1,
|
|
benchmark::utils::CheckNEONDOT);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_6x8c4__neondot)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_6x16c4__neondot(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_6x16c4__neondot,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/6, /*nr=*/16, /*kr=*/4, /*sr=*/1,
|
|
benchmark::utils::CheckNEONDOT);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_6x16c4__neondot)
|
|
#endif // XNN_ENABLE_ARM_DOTPROD && (XNN_ARCH_ARM || XNN_ARCH_ARM64)
|
|
|
|
|
|
#if XNN_ARCH_ARM || XNN_ARCH_ARM64
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x16__neon_mlal_lane(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x16__neon_mlal_lane,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/16, /*kr=*/1, /*sr=*/1,
|
|
benchmark::utils::CheckNEON);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x16__neon_mlal_lane)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x16__neon_mlal_lane_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x16__neon_mlal_lane_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/16, /*kr=*/1, /*sr=*/1,
|
|
benchmark::utils::CheckNEON);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x16__neon_mlal_lane_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_2x16__neon_mlal_lane(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x16__neon_mlal_lane,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/2, /*nr=*/16, /*kr=*/1, /*sr=*/1,
|
|
benchmark::utils::CheckNEON);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_2x16__neon_mlal_lane)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_2x16__neon_mlal_lane_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x16__neon_mlal_lane_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/2, /*nr=*/16, /*kr=*/1, /*sr=*/1,
|
|
benchmark::utils::CheckNEON);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_2x16__neon_mlal_lane_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_3x16__neon_mlal_lane(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x16__neon_mlal_lane,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/3, /*nr=*/16, /*kr=*/1, /*sr=*/1,
|
|
benchmark::utils::CheckNEON);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_3x16__neon_mlal_lane)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_3x16__neon_mlal_lane_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x16__neon_mlal_lane_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/3, /*nr=*/16, /*kr=*/1, /*sr=*/1,
|
|
benchmark::utils::CheckNEON);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_3x16__neon_mlal_lane_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_4x16__neon_mlal_lane(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x16__neon_mlal_lane,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/4, /*nr=*/16, /*kr=*/1, /*sr=*/1,
|
|
benchmark::utils::CheckNEON);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_4x16__neon_mlal_lane)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_4x16__neon_mlal_lane_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x16__neon_mlal_lane_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/4, /*nr=*/16, /*kr=*/1, /*sr=*/1,
|
|
benchmark::utils::CheckNEON);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_4x16__neon_mlal_lane_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_6x16__neon_mlal_lane(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_6x16__neon_mlal_lane,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/6, /*nr=*/16, /*kr=*/1, /*sr=*/1,
|
|
benchmark::utils::CheckNEON);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_6x16__neon_mlal_lane)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_6x16__neon_mlal_lane_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_6x16__neon_mlal_lane_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/6, /*nr=*/16, /*kr=*/1, /*sr=*/1,
|
|
benchmark::utils::CheckNEON);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_6x16__neon_mlal_lane_prfm)
|
|
#endif // XNN_ARCH_ARM || XNN_ARCH_ARM64
|
|
|
|
|
|
#if XNN_ENABLE_AVX512VNNIGFNI && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__avx512vnnigfni(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__avx512vnnigfni,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNIGFNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__avx512vnnigfni)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__avx512vnnigfni(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__avx512vnnigfni,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/5, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNIGFNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__avx512vnnigfni)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__avx512vnnigfni(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__avx512vnnigfni,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/7, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNIGFNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__avx512vnnigfni)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__avx512vnnigfni(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__avx512vnnigfni,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/8, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNIGFNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__avx512vnnigfni)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_9x16c8__avx512vnnigfni(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_9x16c8__avx512vnnigfni,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/9, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNIGFNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_9x16c8__avx512vnnigfni)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_10x16c8__avx512vnnigfni(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_10x16c8__avx512vnnigfni,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/10, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNIGFNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_10x16c8__avx512vnnigfni)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_12x16c8__avx512vnnigfni(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_12x16c8__avx512vnnigfni,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/12, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNIGFNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_12x16c8__avx512vnnigfni)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_14x16c8__avx512vnnigfni(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_14x16c8__avx512vnnigfni,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/14, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNIGFNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_14x16c8__avx512vnnigfni)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__avx512vnnigfni_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__avx512vnnigfni_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNIGFNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__avx512vnnigfni_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__avx512vnnigfni_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__avx512vnnigfni_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/5, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNIGFNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__avx512vnnigfni_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__avx512vnnigfni_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__avx512vnnigfni_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/7, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNIGFNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__avx512vnnigfni_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__avx512vnnigfni_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__avx512vnnigfni_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/8, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNIGFNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__avx512vnnigfni_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_9x16c8__avx512vnnigfni_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_9x16c8__avx512vnnigfni_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/9, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNIGFNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_9x16c8__avx512vnnigfni_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_10x16c8__avx512vnnigfni_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_10x16c8__avx512vnnigfni_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/10, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNIGFNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_10x16c8__avx512vnnigfni_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_12x16c8__avx512vnnigfni_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_12x16c8__avx512vnnigfni_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/12, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNIGFNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_12x16c8__avx512vnnigfni_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_14x16c8__avx512vnnigfni_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_14x16c8__avx512vnnigfni_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/14, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNIGFNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_14x16c8__avx512vnnigfni_prfm)
|
|
#endif // XNN_ENABLE_AVX512VNNIGFNI && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
|
|
|
|
|
|
#if XNN_ENABLE_AVX512VNNI && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__avx512vnni(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__avx512vnni,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__avx512vnni)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__avx512vnni(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__avx512vnni,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/5, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__avx512vnni)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__avx512vnni(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__avx512vnni,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/7, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__avx512vnni)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__avx512vnni(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__avx512vnni,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/8, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__avx512vnni)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_9x16c8__avx512vnni(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_9x16c8__avx512vnni,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/9, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_9x16c8__avx512vnni)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_10x16c8__avx512vnni(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_10x16c8__avx512vnni,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/10, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_10x16c8__avx512vnni)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_12x16c8__avx512vnni(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_12x16c8__avx512vnni,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/12, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_12x16c8__avx512vnni)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_14x16c8__avx512vnni(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_14x16c8__avx512vnni,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/14, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_14x16c8__avx512vnni)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__avx512vnni_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__avx512vnni_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x16c8__avx512vnni_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__avx512vnni_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__avx512vnni_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/5, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_5x16c8__avx512vnni_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__avx512vnni_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__avx512vnni_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/7, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_7x16c8__avx512vnni_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__avx512vnni_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__avx512vnni_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/8, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_8x16c8__avx512vnni_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_9x16c8__avx512vnni_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_9x16c8__avx512vnni_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/9, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_9x16c8__avx512vnni_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_10x16c8__avx512vnni_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_10x16c8__avx512vnni_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/10, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_10x16c8__avx512vnni_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_12x16c8__avx512vnni_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_12x16c8__avx512vnni_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/12, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_12x16c8__avx512vnni_prfm)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_14x16c8__avx512vnni_prfm(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_14x16c8__avx512vnni_prfm,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/14, /*nr=*/16, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX512VNNI);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_14x16c8__avx512vnni_prfm)
|
|
#endif // XNN_ENABLE_AVX512VNNI && (XNN_ARCH_X86 || XNN_ARCH_X86_64)
|
|
|
|
|
|
#if XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x8c8__avx2(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x8c8__avx2,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/8, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX2);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x8c8__avx2)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_2x8c8__avx2(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x8c8__avx2,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/2, /*nr=*/8, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX2);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_2x8c8__avx2)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_3x8c8__avx2(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x8c8__avx2,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/3, /*nr=*/8, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX2);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_3x8c8__avx2)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_4x8c8__avx2(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x8c8__avx2,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/4, /*nr=*/8, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX2);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_4x8c8__avx2)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__avx_ld128(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__avx_ld128,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__avx_ld128)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__avx_ld128(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__avx_ld128,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/2, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__avx_ld128)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__avx_ld128(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__avx_ld128,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/3, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__avx_ld128)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__avx_ld128(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__avx_ld128,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/4, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__avx_ld128)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__avx_ld64(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__avx_ld64,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__avx_ld64)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__avx_ld64(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__avx_ld64,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/2, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__avx_ld64)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__avx_ld64(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__avx_ld64,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/3, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__avx_ld64)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__avx_ld64(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__avx_ld64,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/4, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckAVX);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__avx_ld64)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__sse41_ld128(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__sse41_ld128,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckSSE41);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__sse41_ld128)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__sse41_ld128(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__sse41_ld128,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/2, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckSSE41);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__sse41_ld128)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__sse41_ld128(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__sse41_ld128,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/3, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckSSE41);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__sse41_ld128)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__sse41_ld128(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__sse41_ld128,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/4, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckSSE41);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__sse41_ld128)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__sse41_ld64(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__sse41_ld64,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckSSE41);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__sse41_ld64)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__sse41_ld64(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__sse41_ld64,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/2, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckSSE41);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__sse41_ld64)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__sse41_ld64(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__sse41_ld64,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/3, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckSSE41);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__sse41_ld64)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__sse41_ld64(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__sse41_ld64,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/4, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
benchmark::utils::CheckSSE41);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__sse41_ld64)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__sse2_ld128(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__sse2_ld128,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
/*isa_check=*/nullptr);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__sse2_ld128)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__sse2_ld128(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__sse2_ld128,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/2, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
/*isa_check=*/nullptr);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__sse2_ld128)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__sse2_ld128(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__sse2_ld128,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/3, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
/*isa_check=*/nullptr);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__sse2_ld128)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__sse2_ld128(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__sse2_ld128,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/4, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
/*isa_check=*/nullptr);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__sse2_ld128)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__sse2_ld64(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__sse2_ld64,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
/*isa_check=*/nullptr);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x4c8__sse2_ld64)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__sse2_ld64(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__sse2_ld64,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/2, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
/*isa_check=*/nullptr);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_2x4c8__sse2_ld64)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__sse2_ld64(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__sse2_ld64,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/3, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
/*isa_check=*/nullptr);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_3x4c8__sse2_ld64)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__sse2_ld64(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__sse2_ld64,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/4, /*nr=*/4, /*kr=*/8, /*sr=*/1,
|
|
/*isa_check=*/nullptr);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_4x4c8__sse2_ld64)
|
|
#endif // XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x2__scalar(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x2__scalar,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/2, /*kr=*/1, /*sr=*/1,
|
|
/*isa_check=*/nullptr);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x2__scalar)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x4__scalar(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x4__scalar,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/4, /*kr=*/1, /*sr=*/1,
|
|
/*isa_check=*/nullptr);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x4__scalar)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_1x8__scalar(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_1x8__scalar,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/1, /*nr=*/8, /*kr=*/1, /*sr=*/1,
|
|
/*isa_check=*/nullptr);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_1x8__scalar)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_2x2__scalar(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x2__scalar,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/2, /*nr=*/2, /*kr=*/1, /*sr=*/1,
|
|
/*isa_check=*/nullptr);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_2x2__scalar)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_2x4__scalar(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x4__scalar,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/2, /*nr=*/4, /*kr=*/1, /*sr=*/1,
|
|
/*isa_check=*/nullptr);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_2x4__scalar)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_2x8__scalar(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_2x8__scalar,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/2, /*nr=*/8, /*kr=*/1, /*sr=*/1,
|
|
/*isa_check=*/nullptr);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_2x8__scalar)
|
|
|
|
static void qd8_f32_qb4w_gemm_minmax_ukernel_4x4__scalar(benchmark::State& state, const char* net) {
|
|
GEMMBenchmark(state,
|
|
xnn_qd8_f32_qb4w_gemm_minmax_ukernel_4x4__scalar,
|
|
xnn_init_f32_qb4w_minmax_scalar_params,
|
|
xnn_pack_qs8_qb4w_gemm_goi_w,
|
|
/*mr=*/4, /*nr=*/4, /*kr=*/1, /*sr=*/1,
|
|
/*isa_check=*/nullptr);
|
|
}
|
|
|
|
BENCHMARK_GEMM_BL(qd8_f32_qb4w_gemm_minmax_ukernel_4x4__scalar)
|
|
|
|
#ifndef XNNPACK_BENCHMARK_NO_MAIN
|
|
BENCHMARK_MAIN();
|
|
#endif
|