1 /* $NetBSD: amdgpu_benchmark.c,v 1.3 2018/08/27 14:04:50 riastradh Exp $ */ 2 3 /* 4 * Copyright 2009 Jerome Glisse. 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a 7 * copy of this software and associated documentation files (the "Software"), 8 * to deal in the Software without restriction, including without limitation 9 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 10 * and/or sell copies of the Software, and to permit persons to whom the 11 * Software is furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 22 * OTHER DEALINGS IN THE SOFTWARE. 23 * 24 * Authors: Jerome Glisse 25 */ 26 #include <sys/cdefs.h> 27 __KERNEL_RCSID(0, "$NetBSD: amdgpu_benchmark.c,v 1.3 2018/08/27 14:04:50 riastradh Exp $"); 28 29 #include <drm/drmP.h> 30 #include <drm/amdgpu_drm.h> 31 #include "amdgpu.h" 32 33 #define AMDGPU_BENCHMARK_ITERATIONS 1024 34 #define AMDGPU_BENCHMARK_COMMON_MODES_N 17 35 36 static int amdgpu_benchmark_do_move(struct amdgpu_device *adev, unsigned size, 37 uint64_t saddr, uint64_t daddr, int n) 38 { 39 unsigned long start_jiffies; 40 unsigned long end_jiffies; 41 struct fence *fence = NULL; 42 int i, r; 43 44 start_jiffies = jiffies; 45 for (i = 0; i < n; i++) { 46 struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring; 47 r = amdgpu_copy_buffer(ring, saddr, daddr, size, NULL, &fence); 48 if (r) 49 goto exit_do_move; 50 r = fence_wait(fence, false); 51 if (r) 52 goto exit_do_move; 53 fence_put(fence); 54 } 55 end_jiffies = jiffies; 56 r = jiffies_to_msecs(end_jiffies - start_jiffies); 57 58 exit_do_move: 59 if (fence) 60 fence_put(fence); 61 return r; 62 } 63 64 65 static void amdgpu_benchmark_log_results(int n, unsigned size, 66 unsigned int time, 67 unsigned sdomain, unsigned ddomain, 68 const char *kind) 69 { 70 unsigned int throughput = (n * (size >> 10)) / time; 71 DRM_INFO("amdgpu: %s %u bo moves of %u kB from" 72 " %d to %d in %u ms, throughput: %u Mb/s or %u MB/s\n", 73 kind, n, size >> 10, sdomain, ddomain, time, 74 throughput * 8, throughput); 75 } 76 77 static void amdgpu_benchmark_move(struct amdgpu_device *adev, unsigned size, 78 unsigned sdomain, unsigned ddomain) 79 { 80 struct amdgpu_bo *dobj = NULL; 81 struct amdgpu_bo *sobj = NULL; 82 uint64_t saddr, daddr; 83 int r, n; 84 int time; 85 86 n = AMDGPU_BENCHMARK_ITERATIONS; 87 r = amdgpu_bo_create(adev, size, PAGE_SIZE, true, sdomain, 0, NULL, 88 NULL, &sobj); 89 if (r) { 90 goto out_cleanup; 91 } 92 r = amdgpu_bo_reserve(sobj, false); 93 if (unlikely(r != 0)) 94 goto out_cleanup; 95 r = amdgpu_bo_pin(sobj, sdomain, &saddr); 96 amdgpu_bo_unreserve(sobj); 97 if (r) { 98 goto out_cleanup; 99 } 100 r = amdgpu_bo_create(adev, size, PAGE_SIZE, true, ddomain, 0, NULL, 101 NULL, &dobj); 102 if (r) { 103 goto out_cleanup; 104 } 105 r = amdgpu_bo_reserve(dobj, false); 106 if (unlikely(r != 0)) 107 goto out_cleanup; 108 r = amdgpu_bo_pin(dobj, ddomain, &daddr); 109 amdgpu_bo_unreserve(dobj); 110 if (r) { 111 goto out_cleanup; 112 } 113 114 if (adev->mman.buffer_funcs) { 115 time = amdgpu_benchmark_do_move(adev, size, saddr, daddr, n); 116 if (time < 0) 117 goto out_cleanup; 118 if (time > 0) 119 amdgpu_benchmark_log_results(n, size, time, 120 sdomain, ddomain, "dma"); 121 } 122 123 out_cleanup: 124 if (sobj) { 125 r = amdgpu_bo_reserve(sobj, false); 126 if (likely(r == 0)) { 127 amdgpu_bo_unpin(sobj); 128 amdgpu_bo_unreserve(sobj); 129 } 130 amdgpu_bo_unref(&sobj); 131 } 132 if (dobj) { 133 r = amdgpu_bo_reserve(dobj, false); 134 if (likely(r == 0)) { 135 amdgpu_bo_unpin(dobj); 136 amdgpu_bo_unreserve(dobj); 137 } 138 amdgpu_bo_unref(&dobj); 139 } 140 141 if (r) { 142 DRM_ERROR("Error while benchmarking BO move.\n"); 143 } 144 } 145 146 void amdgpu_benchmark(struct amdgpu_device *adev, int test_number) 147 { 148 int i; 149 int common_modes[AMDGPU_BENCHMARK_COMMON_MODES_N] = { 150 640 * 480 * 4, 151 720 * 480 * 4, 152 800 * 600 * 4, 153 848 * 480 * 4, 154 1024 * 768 * 4, 155 1152 * 768 * 4, 156 1280 * 720 * 4, 157 1280 * 800 * 4, 158 1280 * 854 * 4, 159 1280 * 960 * 4, 160 1280 * 1024 * 4, 161 1440 * 900 * 4, 162 1400 * 1050 * 4, 163 1680 * 1050 * 4, 164 1600 * 1200 * 4, 165 1920 * 1080 * 4, 166 1920 * 1200 * 4 167 }; 168 169 switch (test_number) { 170 case 1: 171 /* simple test, VRAM to GTT and GTT to VRAM */ 172 amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_GTT, 173 AMDGPU_GEM_DOMAIN_VRAM); 174 amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_VRAM, 175 AMDGPU_GEM_DOMAIN_GTT); 176 break; 177 case 2: 178 /* simple test, VRAM to VRAM */ 179 amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_VRAM, 180 AMDGPU_GEM_DOMAIN_VRAM); 181 break; 182 case 3: 183 /* GTT to VRAM, buffer size sweep, powers of 2 */ 184 for (i = 1; i <= 16384; i <<= 1) 185 amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE, 186 AMDGPU_GEM_DOMAIN_GTT, 187 AMDGPU_GEM_DOMAIN_VRAM); 188 break; 189 case 4: 190 /* VRAM to GTT, buffer size sweep, powers of 2 */ 191 for (i = 1; i <= 16384; i <<= 1) 192 amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE, 193 AMDGPU_GEM_DOMAIN_VRAM, 194 AMDGPU_GEM_DOMAIN_GTT); 195 break; 196 case 5: 197 /* VRAM to VRAM, buffer size sweep, powers of 2 */ 198 for (i = 1; i <= 16384; i <<= 1) 199 amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE, 200 AMDGPU_GEM_DOMAIN_VRAM, 201 AMDGPU_GEM_DOMAIN_VRAM); 202 break; 203 case 6: 204 /* GTT to VRAM, buffer size sweep, common modes */ 205 for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++) 206 amdgpu_benchmark_move(adev, common_modes[i], 207 AMDGPU_GEM_DOMAIN_GTT, 208 AMDGPU_GEM_DOMAIN_VRAM); 209 break; 210 case 7: 211 /* VRAM to GTT, buffer size sweep, common modes */ 212 for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++) 213 amdgpu_benchmark_move(adev, common_modes[i], 214 AMDGPU_GEM_DOMAIN_VRAM, 215 AMDGPU_GEM_DOMAIN_GTT); 216 break; 217 case 8: 218 /* VRAM to VRAM, buffer size sweep, common modes */ 219 for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++) 220 amdgpu_benchmark_move(adev, common_modes[i], 221 AMDGPU_GEM_DOMAIN_VRAM, 222 AMDGPU_GEM_DOMAIN_VRAM); 223 break; 224 225 default: 226 DRM_ERROR("Unknown benchmark\n"); 227 } 228 } 229