7350 lines
189 KiB
C++
7350 lines
189 KiB
C++
// Copyright (c) Facebook, Inc. and its affiliates.
|
|
// All rights reserved.
|
|
//
|
|
// Copyright 2019 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-igemm-minmax.yaml
|
|
// Generator: tools/generate-gemm-test.py
|
|
|
|
|
|
#include <gtest/gtest.h>
|
|
|
|
#include <xnnpack/allocator.h>
|
|
#include <xnnpack/common.h>
|
|
#include <xnnpack/isa-checks.h>
|
|
|
|
#include <xnnpack/gemm.h>
|
|
#include <xnnpack/igemm.h>
|
|
#include <xnnpack/ppmm.h>
|
|
#include "gemm-microkernel-tester.h"
|
|
|
|
|
|
#if XNN_ARCH_ARM64
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, k_eq_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(4)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(4)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, k_eq_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, k_eq_4_subtile_m) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(8)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, k_eq_4_subtile_n) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, k_lt_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k < 4; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, k_lt_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k < 4; k++) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, k_gt_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 5; k < 8; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, k_gt_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 5; k < 8; k++) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, k_div_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 8; k <= 40; k += 4) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, k_div_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 8; k <= 40; k += 4) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, n_gt_8) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, n_gt_8_strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, n_gt_8_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, n_div_8) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, n_div_8_strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, n_div_8_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, small_kernel_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, n_gt_8_small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, n_div_8_small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, strided_cm_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.cm_stride(11)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, a_offset) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(23)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, zero) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t mz = 0; mz < 1; mz++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(23)
|
|
.zero_index(mz)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, qmin) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(4)
|
|
.qmin(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, qmax) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(4)
|
|
.qmax(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__NEONFP16ARITH_LD64, strided_cm) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(4)
|
|
.cm_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
#endif // XNN_ARCH_ARM64
|
|
|
|
|
|
#if XNN_ARCH_ARM64
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, k_eq_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(4)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(4)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, k_eq_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, k_eq_4_subtile_m) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(8)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, k_eq_4_subtile_n) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, k_lt_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k < 4; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, k_lt_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k < 4; k++) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, k_gt_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 5; k < 8; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, k_gt_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 5; k < 8; k++) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, k_div_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 8; k <= 40; k += 4) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, k_div_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 8; k <= 40; k += 4) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, n_gt_8) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, n_gt_8_strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, n_gt_8_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, n_div_8) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, n_div_8_strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, n_div_8_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, small_kernel_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, n_gt_8_small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, n_div_8_small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, strided_cm_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.cm_stride(11)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, a_offset) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(83)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, zero) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t mz = 0; mz < 4; mz++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(83)
|
|
.zero_index(mz)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, qmin) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(4)
|
|
.qmin(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, qmax) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(4)
|
|
.qmax(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__NEONFP16ARITH_LD64, strided_cm) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(4)
|
|
.cm_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
#endif // XNN_ARCH_ARM64
|
|
|
|
|
|
#if XNN_ARCH_ARM64
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, k_eq_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(4)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(4)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, k_eq_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, k_eq_4_subtile_m) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(8)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, k_eq_4_subtile_n) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, k_lt_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k < 4; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, k_lt_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k < 4; k++) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, k_gt_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 5; k < 8; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, k_gt_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 5; k < 8; k++) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, k_div_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 8; k <= 40; k += 4) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, k_div_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 8; k <= 40; k += 4) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, n_gt_8) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, n_gt_8_strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, n_gt_8_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, n_div_8) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, n_div_8_strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, n_div_8_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, small_kernel_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, n_gt_8_small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, n_div_8_small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, strided_cm_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.cm_stride(11)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, a_offset) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(127)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, zero) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t mz = 0; mz < 6; mz++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(127)
|
|
.zero_index(mz)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, qmin) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(4)
|
|
.qmin(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, qmax) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(4)
|
|
.qmax(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__NEONFP16ARITH_LD64, strided_cm) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(4)
|
|
.cm_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
#endif // XNN_ARCH_ARM64
|
|
|
|
|
|
#if XNN_ARCH_ARM64
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, k_eq_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(8)
|
|
.k(4)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(8)
|
|
.k(4)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, k_eq_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, k_eq_4_subtile_m) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(8)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, k_eq_4_subtile_n) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(n)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, k_lt_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k < 4; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(8)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, k_lt_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k < 4; k++) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, k_gt_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 5; k < 8; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(8)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, k_gt_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 5; k < 8; k++) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, k_div_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 8; k <= 40; k += 4) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(8)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, k_div_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 8; k <= 40; k += 4) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, n_gt_8) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, n_gt_8_strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, n_gt_8_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, n_div_8) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, n_div_8_strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, n_div_8_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, small_kernel_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, n_gt_8_small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, n_div_8_small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, strided_cm_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.cm_stride(11)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, a_offset) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(163)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, zero) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t mz = 0; mz < 8; mz++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(163)
|
|
.zero_index(mz)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, qmin) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(8)
|
|
.k(4)
|
|
.qmin(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, qmax) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(8)
|
|
.k(4)
|
|
.qmax(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X8__NEONFP16ARITH_LD64, strided_cm) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(8)
|
|
.k(4)
|
|
.cm_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x8__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
#endif // XNN_ARCH_ARM64
|
|
|
|
|
|
#if XNN_ARCH_ARM64
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, k_eq_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(4)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(4)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, k_eq_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, k_eq_4_subtile_m) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(16)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, k_eq_4_subtile_n) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, k_lt_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k < 4; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, k_lt_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k < 4; k++) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, k_gt_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 5; k < 8; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, k_gt_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 5; k < 8; k++) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, k_div_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 8; k <= 40; k += 4) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, k_div_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 8; k <= 40; k += 4) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, n_gt_16) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, n_gt_16_strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, n_gt_16_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, n_div_16) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, n_div_16_strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, n_div_16_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, small_kernel_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, n_gt_16_small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, n_div_16_small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, strided_cm_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.cm_stride(19)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, a_offset) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(23)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, zero) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t mz = 0; mz < 1; mz++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(23)
|
|
.zero_index(mz)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, qmin) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(4)
|
|
.qmin(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, qmax) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(4)
|
|
.qmax(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__NEONFP16ARITH_LD64, strided_cm) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(4)
|
|
.cm_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
#endif // XNN_ARCH_ARM64
|
|
|
|
|
|
#if XNN_ARCH_ARM64
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, k_eq_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(4)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(4)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, k_eq_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, k_eq_4_subtile_m) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(16)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, k_eq_4_subtile_n) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, k_lt_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k < 4; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, k_lt_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k < 4; k++) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, k_gt_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 5; k < 8; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, k_gt_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 5; k < 8; k++) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, k_div_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 8; k <= 40; k += 4) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, k_div_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 8; k <= 40; k += 4) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, n_gt_16) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, n_gt_16_strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, n_gt_16_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, n_div_16) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, n_div_16_strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, n_div_16_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, small_kernel_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, n_gt_16_small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, n_div_16_small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, strided_cm_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.cm_stride(19)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, a_offset) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(83)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, zero) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t mz = 0; mz < 4; mz++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(83)
|
|
.zero_index(mz)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, qmin) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(4)
|
|
.qmin(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, qmax) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(4)
|
|
.qmax(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__NEONFP16ARITH_LD64, strided_cm) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(4)
|
|
.cm_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
#endif // XNN_ARCH_ARM64
|
|
|
|
|
|
#if XNN_ARCH_ARM64
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, k_eq_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(16)
|
|
.k(4)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(16)
|
|
.k(4)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, k_eq_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, k_eq_4_subtile_m) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(16)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, k_eq_4_subtile_n) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, k_lt_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k < 4; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(16)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, k_lt_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k < 4; k++) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, k_gt_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 5; k < 8; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(16)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, k_gt_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 5; k < 8; k++) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, k_div_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 8; k <= 40; k += 4) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(16)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, k_div_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 8; k <= 40; k += 4) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, n_gt_16) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, n_gt_16_strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, n_gt_16_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, n_div_16) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, n_div_16_strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, n_div_16_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, small_kernel_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, n_gt_16_small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, n_div_16_small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, strided_cm_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.cm_stride(19)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, a_offset) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(127)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, zero) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t mz = 0; mz < 6; mz++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(127)
|
|
.zero_index(mz)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, qmin) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(16)
|
|
.k(4)
|
|
.qmin(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, qmax) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(16)
|
|
.k(4)
|
|
.qmax(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X16__NEONFP16ARITH_LD64, strided_cm) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(16)
|
|
.k(4)
|
|
.cm_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
#endif // XNN_ARCH_ARM64
|
|
|
|
|
|
#if XNN_ARCH_ARM64
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, k_eq_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(16)
|
|
.k(4)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(16)
|
|
.k(4)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, k_eq_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, k_eq_4_subtile_m) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(16)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, k_eq_4_subtile_n) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(n)
|
|
.k(4)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, k_lt_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k < 4; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(16)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, k_lt_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k < 4; k++) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, k_gt_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 5; k < 8; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(16)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, k_gt_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 5; k < 8; k++) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, k_div_4) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 8; k <= 40; k += 4) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(16)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, k_div_4_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 8; k <= 40; k += 4) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, n_gt_16) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, n_gt_16_strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, n_gt_16_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, n_div_16) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, n_div_16_strided_cn) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, n_div_16_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, small_kernel_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, n_gt_16_small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, n_div_16_small_kernel) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, strided_cm_subtile) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 8; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.cm_stride(19)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, a_offset) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(163)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, zero) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
for (size_t k = 1; k <= 20; k += 5) {
|
|
for (uint32_t mz = 0; mz < 8; mz++) {
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(163)
|
|
.zero_index(mz)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, qmin) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(16)
|
|
.k(4)
|
|
.qmin(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, qmax) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(16)
|
|
.k(4)
|
|
.qmax(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_8X16__NEONFP16ARITH_LD64, strided_cm) {
|
|
TEST_REQUIRES_ARM_NEON_FP16_ARITH;
|
|
GemmMicrokernelTester()
|
|
.mr(8)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(8)
|
|
.n(16)
|
|
.k(4)
|
|
.cm_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_8x16__neonfp16arith_ld64, xnn_init_f16_scaleminmax_neon_params);
|
|
}
|
|
#endif // XNN_ARCH_ARM64
|
|
|
|
|
|
#if XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, k_eq_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(1)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, k_eq_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, k_eq_1_subtile_m) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(8)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, k_eq_1_subtile_n) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, k_gt_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, k_gt_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, n_gt_8) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, n_gt_8_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, n_gt_8_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, n_div_8) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, n_div_8_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, n_div_8_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, small_kernel_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, n_gt_8_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, n_div_8_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, strided_cm_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.cm_stride(11)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, a_offset) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(7)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, zero) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t mz = 0; mz < 1; mz++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(7)
|
|
.zero_index(mz)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, qmin) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(1)
|
|
.qmin(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, qmax) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(1)
|
|
.qmax(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X8__AVX2_BROADCAST, strided_cm) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(8)
|
|
.k(1)
|
|
.cm_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
#endif // XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
|
|
|
|
#if XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, k_eq_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(1)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, k_eq_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, k_eq_1_subtile_m) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(8)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, k_eq_1_subtile_n) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, k_gt_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, k_gt_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, n_gt_8) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, n_gt_8_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, n_gt_8_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, n_div_8) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, n_div_8_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, n_div_8_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, small_kernel_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, n_gt_8_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, n_div_8_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, strided_cm_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.cm_stride(11)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, a_offset) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(23)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, zero) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t mz = 0; mz < 4; mz++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(23)
|
|
.zero_index(mz)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, qmin) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(1)
|
|
.qmin(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, qmax) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(1)
|
|
.qmax(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X8__AVX2_BROADCAST, strided_cm) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(8)
|
|
.k(1)
|
|
.cm_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
#endif // XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
|
|
|
|
#if XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, k_eq_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(8)
|
|
.k(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(8)
|
|
.k(1)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, k_eq_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 5; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, k_eq_1_subtile_m) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t m = 1; m <= 5; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(8)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, k_eq_1_subtile_n) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, k_gt_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(8)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, k_gt_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 5; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, n_gt_8) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, n_gt_8_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, n_gt_8_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 5; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, n_div_8) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, n_div_8_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, n_div_8_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 5; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, small_kernel_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 5; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, n_gt_8_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, n_div_8_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, strided_cm_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 5; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.cm_stride(11)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, a_offset) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(29)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, zero) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t mz = 0; mz < 5; mz++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(29)
|
|
.zero_index(mz)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, qmin) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(8)
|
|
.k(1)
|
|
.qmin(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, qmax) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(8)
|
|
.k(1)
|
|
.qmax(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X8__AVX2_BROADCAST, strided_cm) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(8)
|
|
.k(1)
|
|
.cm_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
#endif // XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
|
|
|
|
#if XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, k_eq_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(1)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, k_eq_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, k_eq_1_subtile_m) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(8)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, k_eq_1_subtile_n) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, k_gt_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, k_gt_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, n_gt_8) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, n_gt_8_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, n_gt_8_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, n_div_8) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, n_div_8_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, n_div_8_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, small_kernel_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, n_gt_8_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, n_div_8_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, strided_cm_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 6; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.cm_stride(11)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, a_offset) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(37)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, zero) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t mz = 0; mz < 6; mz++) {
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(37)
|
|
.zero_index(mz)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, qmin) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(1)
|
|
.qmin(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, qmax) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(1)
|
|
.qmax(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_6X8__AVX2_BROADCAST, strided_cm) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(6)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(6)
|
|
.n(8)
|
|
.k(1)
|
|
.cm_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_6x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
#endif // XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
|
|
|
|
#if XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, k_eq_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(7)
|
|
.n(8)
|
|
.k(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(7)
|
|
.n(8)
|
|
.k(1)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, k_eq_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 7; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, k_eq_1_subtile_m) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t m = 1; m <= 7; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(8)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, k_eq_1_subtile_n) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(7)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, k_gt_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(7)
|
|
.n(8)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, k_gt_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 7; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, n_gt_8) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(7)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, n_gt_8_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(7)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, n_gt_8_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 7; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, n_div_8) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(7)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, n_div_8_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(7)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, n_div_8_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 7; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(7)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, small_kernel_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 7; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, n_gt_8_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 9; n < 16; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(7)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, n_div_8_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 16; n <= 24; n += 8) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(7)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, strided_cm_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 8; n++) {
|
|
for (uint32_t m = 1; m <= 7; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.cm_stride(11)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, a_offset) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(7)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(37)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, zero) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t mz = 0; mz < 7; mz++) {
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(7)
|
|
.n(8)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(37)
|
|
.zero_index(mz)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, qmin) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(7)
|
|
.n(8)
|
|
.k(1)
|
|
.qmin(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, qmax) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(7)
|
|
.n(8)
|
|
.k(1)
|
|
.qmax(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_7X8__AVX2_BROADCAST, strided_cm) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(7)
|
|
.nr(8)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(7)
|
|
.n(8)
|
|
.k(1)
|
|
.cm_stride(11)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_7x8__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
#endif // XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
|
|
|
|
#if XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, k_eq_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(1)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, k_eq_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, k_eq_1_subtile_m) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(16)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, k_eq_1_subtile_n) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, k_gt_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, k_gt_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, n_gt_16) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, n_gt_16_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, n_gt_16_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, n_div_16) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, n_div_16_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, n_div_16_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, small_kernel_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, n_gt_16_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, n_div_16_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, strided_cm_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 1; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.cm_stride(19)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, a_offset) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(7)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, zero) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t mz = 0; mz < 1; mz++) {
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(7)
|
|
.zero_index(mz)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, qmin) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(1)
|
|
.qmin(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, qmax) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(1)
|
|
.qmax(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_1X16__AVX2_BROADCAST, strided_cm) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(1)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(1)
|
|
.n(16)
|
|
.k(1)
|
|
.cm_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_1x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
#endif // XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
|
|
|
|
#if XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, k_eq_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(3)
|
|
.n(16)
|
|
.k(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(3)
|
|
.n(16)
|
|
.k(1)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, k_eq_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 3; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, k_eq_1_subtile_m) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t m = 1; m <= 3; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(16)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, k_eq_1_subtile_n) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(3)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, k_gt_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(3)
|
|
.n(16)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, k_gt_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 3; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, n_gt_16) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(3)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, n_gt_16_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(3)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, n_gt_16_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 3; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, n_div_16) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(3)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, n_div_16_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(3)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, n_div_16_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 3; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(3)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, small_kernel_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 3; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, n_gt_16_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(3)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, n_div_16_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(3)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, strided_cm_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 3; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.cm_stride(19)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, a_offset) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(3)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(17)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, zero) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t mz = 0; mz < 3; mz++) {
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(3)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(17)
|
|
.zero_index(mz)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, qmin) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(3)
|
|
.n(16)
|
|
.k(1)
|
|
.qmin(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, qmax) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(3)
|
|
.n(16)
|
|
.k(1)
|
|
.qmax(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_3X16__AVX2_BROADCAST, strided_cm) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(3)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(3)
|
|
.n(16)
|
|
.k(1)
|
|
.cm_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_3x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
#endif // XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
|
|
|
|
#if XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, k_eq_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(1)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, k_eq_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, k_eq_1_subtile_m) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(16)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, k_eq_1_subtile_n) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, k_gt_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, k_gt_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, n_gt_16) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, n_gt_16_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, n_gt_16_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, n_div_16) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, n_div_16_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, n_div_16_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, small_kernel_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, n_gt_16_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, n_div_16_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, strided_cm_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 4; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.cm_stride(19)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, a_offset) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(23)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, zero) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t mz = 0; mz < 4; mz++) {
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(23)
|
|
.zero_index(mz)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, qmin) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(1)
|
|
.qmin(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, qmax) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(1)
|
|
.qmax(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_4X16__AVX2_BROADCAST, strided_cm) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(4)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(4)
|
|
.n(16)
|
|
.k(1)
|
|
.cm_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
#endif // XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
|
|
|
|
#if XNN_ARCH_X86 || XNN_ARCH_X86_64
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, k_eq_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(16)
|
|
.k(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(16)
|
|
.k(1)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, k_eq_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 5; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, k_eq_1_subtile_m) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t m = 1; m <= 5; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(16)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, k_eq_1_subtile_n) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(n)
|
|
.k(1)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, k_gt_1) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(16)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, k_gt_1_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 2; k < 10; k++) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 5; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, n_gt_16) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, n_gt_16_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, n_gt_16_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 5; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, n_div_16) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(n)
|
|
.k(k)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, n_div_16_strided_cn) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(n)
|
|
.k(k)
|
|
.cn_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, n_div_16_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t m = 1; m <= 5; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, small_kernel_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 5; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, n_gt_16_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 17; n < 32; n++) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, n_div_16_small_kernel) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (uint32_t n = 32; n <= 48; n += 16) {
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(n)
|
|
.k(k)
|
|
.ks(3)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, strided_cm_subtile) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t n = 1; n <= 16; n++) {
|
|
for (uint32_t m = 1; m <= 5; m++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(m)
|
|
.n(n)
|
|
.k(k)
|
|
.cm_stride(19)
|
|
.iterations(1)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, a_offset) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(29)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, zero) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
for (size_t k = 1; k <= 5; k += 2) {
|
|
for (uint32_t mz = 0; mz < 5; mz++) {
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(16)
|
|
.k(k)
|
|
.ks(3)
|
|
.a_offset(29)
|
|
.zero_index(mz)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, qmin) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(16)
|
|
.k(1)
|
|
.qmin(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, qmax) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(16)
|
|
.k(1)
|
|
.qmax(128)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
|
|
TEST(F16_IGEMM_MINMAX_5X16__AVX2_BROADCAST, strided_cm) {
|
|
TEST_REQUIRES_X86_AVX2;
|
|
GemmMicrokernelTester()
|
|
.mr(5)
|
|
.nr(16)
|
|
.kr(1)
|
|
.sr(1)
|
|
.m(5)
|
|
.n(16)
|
|
.k(1)
|
|
.cm_stride(19)
|
|
.Test(xnn_f16_igemm_minmax_ukernel_5x16__avx2_broadcast, xnn_init_f16_scaleminmax_avx_params);
|
|
}
|
|
#endif // XNN_ARCH_X86 || XNN_ARCH_X86_64
|