XNNPACK Team | b455b12 | 2019-09-27 18:10:33 -0700 | [diff] [blame^] | 1 | // Copyright 2019 Google LLC |
| 2 | // |
| 3 | // This source code is licensed under the BSD-style license found in the |
| 4 | // LICENSE file in the root directory of this source tree. |
| 5 | |
| 6 | #include <cpuinfo.h> |
| 7 | #include <gtest/gtest.h> |
| 8 | |
| 9 | #include <xnnpack/isa-checks.h> |
| 10 | #include <xnnpack/rmax.h> |
| 11 | |
| 12 | #include "rmax-microkernel-tester.h" |
| 13 | |
| 14 | |
| 15 | #if CPUINFO_ARCH_ARM || CPUINFO_ARCH_ARM64 |
| 16 | TEST(F32_RMAX__NEON, n_lt_16) { |
| 17 | TEST_REQUIRES_ARM_NEON; |
| 18 | for (size_t n = 1; n < 16; n++) { |
| 19 | RMaxMicrokernelTester() |
| 20 | .n(n) |
| 21 | .Test(xnn_f32_rmax_ukernel__neon); |
| 22 | } |
| 23 | } |
| 24 | |
| 25 | TEST(F32_RMAX__NEON, n_eq_16) { |
| 26 | TEST_REQUIRES_ARM_NEON; |
| 27 | RMaxMicrokernelTester() |
| 28 | .n(16) |
| 29 | .Test(xnn_f32_rmax_ukernel__neon); |
| 30 | } |
| 31 | |
| 32 | TEST(F32_RMAX__NEON, n_div_16) { |
| 33 | TEST_REQUIRES_ARM_NEON; |
| 34 | for (size_t n = 16; n < 128; n += 16) { |
| 35 | RMaxMicrokernelTester() |
| 36 | .n(n) |
| 37 | .Test(xnn_f32_rmax_ukernel__neon); |
| 38 | } |
| 39 | } |
| 40 | |
| 41 | TEST(F32_RMAX__NEON, n_gt_16) { |
| 42 | TEST_REQUIRES_ARM_NEON; |
| 43 | for (size_t n = 16; n < 32; n++) { |
| 44 | RMaxMicrokernelTester() |
| 45 | .n(n) |
| 46 | .Test(xnn_f32_rmax_ukernel__neon); |
| 47 | } |
| 48 | } |
| 49 | #endif // CPUINFO_ARCH_ARM || CPUINFO_ARCH_ARM64 |
| 50 | |
| 51 | #if CPUINFO_ARCH_X86 || CPUINFO_ARCH_X86_64 |
| 52 | TEST(F32_RMAX__SSE, n_lt_16) { |
| 53 | TEST_REQUIRES_X86_SSE2; |
| 54 | for (size_t n = 1; n < 16; n++) { |
| 55 | RMaxMicrokernelTester() |
| 56 | .n(n) |
| 57 | .Test(xnn_f32_rmax_ukernel__sse); |
| 58 | } |
| 59 | } |
| 60 | |
| 61 | TEST(F32_RMAX__SSE, n_eq_16) { |
| 62 | TEST_REQUIRES_X86_SSE2; |
| 63 | RMaxMicrokernelTester() |
| 64 | .n(16) |
| 65 | .Test(xnn_f32_rmax_ukernel__sse); |
| 66 | } |
| 67 | |
| 68 | TEST(F32_RMAX__SSE, n_div_16) { |
| 69 | TEST_REQUIRES_X86_SSE2; |
| 70 | for (size_t n = 16; n < 128; n += 16) { |
| 71 | RMaxMicrokernelTester() |
| 72 | .n(n) |
| 73 | .Test(xnn_f32_rmax_ukernel__sse); |
| 74 | } |
| 75 | } |
| 76 | |
| 77 | TEST(F32_RMAX__SSE, n_gt_16) { |
| 78 | TEST_REQUIRES_X86_SSE2; |
| 79 | for (size_t n = 16; n < 32; n++) { |
| 80 | RMaxMicrokernelTester() |
| 81 | .n(n) |
| 82 | .Test(xnn_f32_rmax_ukernel__sse); |
| 83 | } |
| 84 | } |
| 85 | |
| 86 | TEST(F32_RMAX__AVX, n_lt_32) { |
| 87 | TEST_REQUIRES_X86_AVX; |
| 88 | for (size_t n = 1; n < 32; n++) { |
| 89 | RMaxMicrokernelTester() |
| 90 | .n(n) |
| 91 | .Test(xnn_f32_rmax_ukernel__avx); |
| 92 | } |
| 93 | } |
| 94 | |
| 95 | TEST(F32_RMAX__AVX, n_eq_32) { |
| 96 | TEST_REQUIRES_X86_AVX; |
| 97 | RMaxMicrokernelTester() |
| 98 | .n(32) |
| 99 | .Test(xnn_f32_rmax_ukernel__avx); |
| 100 | } |
| 101 | |
| 102 | TEST(F32_RMAX__AVX, n_div_32) { |
| 103 | TEST_REQUIRES_X86_AVX; |
| 104 | for (size_t n = 32; n < 256; n += 32) { |
| 105 | RMaxMicrokernelTester() |
| 106 | .n(n) |
| 107 | .Test(xnn_f32_rmax_ukernel__avx); |
| 108 | } |
| 109 | } |
| 110 | |
| 111 | TEST(F32_RMAX__AVX, n_gt_32) { |
| 112 | TEST_REQUIRES_X86_AVX; |
| 113 | for (size_t n = 32; n < 64; n++) { |
| 114 | RMaxMicrokernelTester() |
| 115 | .n(n) |
| 116 | .Test(xnn_f32_rmax_ukernel__avx); |
| 117 | } |
| 118 | } |
| 119 | |
| 120 | TEST(F32_RMAX__AVX512F, n_lt_64) { |
| 121 | TEST_REQUIRES_X86_AVX512F; |
| 122 | for (size_t n = 1; n < 64; n++) { |
| 123 | RMaxMicrokernelTester() |
| 124 | .n(n) |
| 125 | .Test(xnn_f32_rmax_ukernel__avx512f); |
| 126 | } |
| 127 | } |
| 128 | |
| 129 | TEST(F32_RMAX__AVX512F, n_eq_64) { |
| 130 | TEST_REQUIRES_X86_AVX512F; |
| 131 | RMaxMicrokernelTester() |
| 132 | .n(64) |
| 133 | .Test(xnn_f32_rmax_ukernel__avx512f); |
| 134 | } |
| 135 | |
| 136 | TEST(F32_RMAX__AVX512F, n_div_64) { |
| 137 | TEST_REQUIRES_X86_AVX512F; |
| 138 | for (size_t n = 64; n < 512; n += 64) { |
| 139 | RMaxMicrokernelTester() |
| 140 | .n(n) |
| 141 | .Test(xnn_f32_rmax_ukernel__avx512f); |
| 142 | } |
| 143 | } |
| 144 | |
| 145 | TEST(F32_RMAX__AVX512F, n_gt_64) { |
| 146 | TEST_REQUIRES_X86_AVX512F; |
| 147 | for (size_t n = 64; n < 128; n++) { |
| 148 | RMaxMicrokernelTester() |
| 149 | .n(n) |
| 150 | .Test(xnn_f32_rmax_ukernel__avx512f); |
| 151 | } |
| 152 | } |
| 153 | #endif // CPUINFO_ARCH_X86 || CPUINFO_ARCH_X86_64 |
| 154 | |
| 155 | TEST(F32_RMAX__SCALAR, n_lt_4) { |
| 156 | for (size_t n = 1; n < 4; n++) { |
| 157 | RMaxMicrokernelTester() |
| 158 | .n(n) |
| 159 | .Test(xnn_f32_rmax_ukernel__scalar); |
| 160 | } |
| 161 | } |
| 162 | |
| 163 | TEST(F32_RMAX__SCALAR, n_eq_4) { |
| 164 | RMaxMicrokernelTester() |
| 165 | .n(4) |
| 166 | .Test(xnn_f32_rmax_ukernel__scalar); |
| 167 | } |
| 168 | |
| 169 | TEST(F32_RMAX__SCALAR, n_div_4) { |
| 170 | for (size_t n = 4; n < 32; n += 4) { |
| 171 | RMaxMicrokernelTester() |
| 172 | .n(n) |
| 173 | .Test(xnn_f32_rmax_ukernel__scalar); |
| 174 | } |
| 175 | } |
| 176 | |
| 177 | TEST(F32_RMAX__SCALAR, n_gt_4) { |
| 178 | for (size_t n = 4; n < 8; n++) { |
| 179 | RMaxMicrokernelTester() |
| 180 | .n(n) |
| 181 | .Test(xnn_f32_rmax_ukernel__scalar); |
| 182 | } |
| 183 | } |