1 /* 2 * Copyright 2009 Jerome Glisse. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 * 22 * Authors: Jerome Glisse 23 */ 24 #include <drm/drmP.h> 25 #include <drm/radeon_drm.h> 26 #include "radeon_reg.h" 27 #include "radeon.h" 28 29 #define RADEON_BENCHMARK_COPY_BLIT 1 30 #define RADEON_BENCHMARK_COPY_DMA 0 31 32 #define RADEON_BENCHMARK_ITERATIONS 1024 33 #define RADEON_BENCHMARK_COMMON_MODES_N 17 34 35 static int radeon_benchmark_do_move(struct radeon_device *rdev, unsigned size, 36 uint64_t saddr, uint64_t daddr, 37 int flag, int n) 38 { 39 unsigned long start_jiffies; 40 unsigned long end_jiffies; 41 struct radeon_fence *fence = NULL; 42 int i, r; 43 44 start_jiffies = jiffies; 45 for (i = 0; i < n; i++) { 46 switch (flag) { 47 case RADEON_BENCHMARK_COPY_DMA: 48 r = radeon_fence_create(rdev, &fence, radeon_copy_dma_ring_index(rdev)); 49 if (r) 50 return r; 51 r = radeon_copy_dma(rdev, saddr, daddr, 52 size / RADEON_GPU_PAGE_SIZE, 53 fence); 54 break; 55 case RADEON_BENCHMARK_COPY_BLIT: 56 r = radeon_fence_create(rdev, &fence, radeon_copy_blit_ring_index(rdev)); 57 if (r) 58 return r; 59 r = radeon_copy_blit(rdev, saddr, daddr, 60 size / RADEON_GPU_PAGE_SIZE, 61 fence); 62 break; 63 default: 64 DRM_ERROR("Unknown copy method\n"); 65 r = -EINVAL; 66 } 67 if (r) 68 goto exit_do_move; 69 r = radeon_fence_wait(fence, false); 70 if (r) 71 goto exit_do_move; 72 radeon_fence_unref(&fence); 73 } 74 end_jiffies = jiffies; 75 r = jiffies_to_msecs(end_jiffies - start_jiffies); 76 77 exit_do_move: 78 if (fence) 79 radeon_fence_unref(&fence); 80 return r; 81 } 82 83 84 static void radeon_benchmark_log_results(int n, unsigned size, 85 unsigned int time, 86 unsigned sdomain, unsigned ddomain, 87 char *kind) 88 { 89 unsigned int throughput = (n * (size >> 10)) / time; 90 DRM_INFO("radeon: %s %u bo moves of %u kB from" 91 " %d to %d in %u ms, throughput: %u Mb/s or %u MB/s\n", 92 kind, n, size >> 10, sdomain, ddomain, time, 93 throughput * 8, throughput); 94 } 95 96 static void radeon_benchmark_move(struct radeon_device *rdev, unsigned size, 97 unsigned sdomain, unsigned ddomain) 98 { 99 struct radeon_bo *dobj = NULL; 100 struct radeon_bo *sobj = NULL; 101 uint64_t saddr, daddr; 102 int r, n; 103 int time; 104 105 n = RADEON_BENCHMARK_ITERATIONS; 106 r = radeon_bo_create(rdev, size, PAGE_SIZE, true, sdomain, &sobj); 107 if (r) { 108 goto out_cleanup; 109 } 110 r = radeon_bo_reserve(sobj, false); 111 if (unlikely(r != 0)) 112 goto out_cleanup; 113 r = radeon_bo_pin(sobj, sdomain, &saddr); 114 radeon_bo_unreserve(sobj); 115 if (r) { 116 goto out_cleanup; 117 } 118 r = radeon_bo_create(rdev, size, PAGE_SIZE, true, ddomain, &dobj); 119 if (r) { 120 goto out_cleanup; 121 } 122 r = radeon_bo_reserve(dobj, false); 123 if (unlikely(r != 0)) 124 goto out_cleanup; 125 r = radeon_bo_pin(dobj, ddomain, &daddr); 126 radeon_bo_unreserve(dobj); 127 if (r) { 128 goto out_cleanup; 129 } 130 131 /* r100 doesn't have dma engine so skip the test */ 132 /* also, VRAM-to-VRAM test doesn't make much sense for DMA */ 133 /* skip it as well if domains are the same */ 134 if ((rdev->asic->copy.dma) && (sdomain != ddomain)) { 135 time = radeon_benchmark_do_move(rdev, size, saddr, daddr, 136 RADEON_BENCHMARK_COPY_DMA, n); 137 if (time < 0) 138 goto out_cleanup; 139 if (time > 0) 140 radeon_benchmark_log_results(n, size, time, 141 sdomain, ddomain, "dma"); 142 } 143 144 time = radeon_benchmark_do_move(rdev, size, saddr, daddr, 145 RADEON_BENCHMARK_COPY_BLIT, n); 146 if (time < 0) 147 goto out_cleanup; 148 if (time > 0) 149 radeon_benchmark_log_results(n, size, time, 150 sdomain, ddomain, "blit"); 151 152 out_cleanup: 153 if (sobj) { 154 r = radeon_bo_reserve(sobj, false); 155 if (likely(r == 0)) { 156 radeon_bo_unpin(sobj); 157 radeon_bo_unreserve(sobj); 158 } 159 radeon_bo_unref(&sobj); 160 } 161 if (dobj) { 162 r = radeon_bo_reserve(dobj, false); 163 if (likely(r == 0)) { 164 radeon_bo_unpin(dobj); 165 radeon_bo_unreserve(dobj); 166 } 167 radeon_bo_unref(&dobj); 168 } 169 170 if (r) { 171 DRM_ERROR("Error while benchmarking BO move.\n"); 172 } 173 } 174 175 void radeon_benchmark(struct radeon_device *rdev, int test_number) 176 { 177 int i; 178 int common_modes[RADEON_BENCHMARK_COMMON_MODES_N] = { 179 640 * 480 * 4, 180 720 * 480 * 4, 181 800 * 600 * 4, 182 848 * 480 * 4, 183 1024 * 768 * 4, 184 1152 * 768 * 4, 185 1280 * 720 * 4, 186 1280 * 800 * 4, 187 1280 * 854 * 4, 188 1280 * 960 * 4, 189 1280 * 1024 * 4, 190 1440 * 900 * 4, 191 1400 * 1050 * 4, 192 1680 * 1050 * 4, 193 1600 * 1200 * 4, 194 1920 * 1080 * 4, 195 1920 * 1200 * 4 196 }; 197 198 switch (test_number) { 199 case 1: 200 /* simple test, VRAM to GTT and GTT to VRAM */ 201 radeon_benchmark_move(rdev, 1024*1024, RADEON_GEM_DOMAIN_GTT, 202 RADEON_GEM_DOMAIN_VRAM); 203 radeon_benchmark_move(rdev, 1024*1024, RADEON_GEM_DOMAIN_VRAM, 204 RADEON_GEM_DOMAIN_GTT); 205 break; 206 case 2: 207 /* simple test, VRAM to VRAM */ 208 radeon_benchmark_move(rdev, 1024*1024, RADEON_GEM_DOMAIN_VRAM, 209 RADEON_GEM_DOMAIN_VRAM); 210 break; 211 case 3: 212 /* GTT to VRAM, buffer size sweep, powers of 2 */ 213 for (i = 1; i <= 16384; i <<= 1) 214 radeon_benchmark_move(rdev, i * RADEON_GPU_PAGE_SIZE, 215 RADEON_GEM_DOMAIN_GTT, 216 RADEON_GEM_DOMAIN_VRAM); 217 break; 218 case 4: 219 /* VRAM to GTT, buffer size sweep, powers of 2 */ 220 for (i = 1; i <= 16384; i <<= 1) 221 radeon_benchmark_move(rdev, i * RADEON_GPU_PAGE_SIZE, 222 RADEON_GEM_DOMAIN_VRAM, 223 RADEON_GEM_DOMAIN_GTT); 224 break; 225 case 5: 226 /* VRAM to VRAM, buffer size sweep, powers of 2 */ 227 for (i = 1; i <= 16384; i <<= 1) 228 radeon_benchmark_move(rdev, i * RADEON_GPU_PAGE_SIZE, 229 RADEON_GEM_DOMAIN_VRAM, 230 RADEON_GEM_DOMAIN_VRAM); 231 break; 232 case 6: 233 /* GTT to VRAM, buffer size sweep, common modes */ 234 for (i = 0; i < RADEON_BENCHMARK_COMMON_MODES_N; i++) 235 radeon_benchmark_move(rdev, common_modes[i], 236 RADEON_GEM_DOMAIN_GTT, 237 RADEON_GEM_DOMAIN_VRAM); 238 break; 239 case 7: 240 /* VRAM to GTT, buffer size sweep, common modes */ 241 for (i = 0; i < RADEON_BENCHMARK_COMMON_MODES_N; i++) 242 radeon_benchmark_move(rdev, common_modes[i], 243 RADEON_GEM_DOMAIN_VRAM, 244 RADEON_GEM_DOMAIN_GTT); 245 break; 246 case 8: 247 /* VRAM to VRAM, buffer size sweep, common modes */ 248 for (i = 0; i < RADEON_BENCHMARK_COMMON_MODES_N; i++) 249 radeon_benchmark_move(rdev, common_modes[i], 250 RADEON_GEM_DOMAIN_VRAM, 251 RADEON_GEM_DOMAIN_VRAM); 252 break; 253 254 default: 255 DRM_ERROR("Unknown benchmark\n"); 256 } 257 } 258