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_copy_dma(rdev, saddr, daddr, 49 size / RADEON_GPU_PAGE_SIZE, 50 &fence); 51 break; 52 case RADEON_BENCHMARK_COPY_BLIT: 53 r = radeon_copy_blit(rdev, saddr, daddr, 54 size / RADEON_GPU_PAGE_SIZE, 55 &fence); 56 break; 57 default: 58 DRM_ERROR("Unknown copy method\n"); 59 r = -EINVAL; 60 } 61 if (r) 62 goto exit_do_move; 63 r = radeon_fence_wait(fence, false); 64 if (r) 65 goto exit_do_move; 66 radeon_fence_unref(&fence); 67 } 68 end_jiffies = jiffies; 69 r = jiffies_to_msecs(end_jiffies - start_jiffies); 70 71 exit_do_move: 72 if (fence) 73 radeon_fence_unref(&fence); 74 return r; 75 } 76 77 78 static void radeon_benchmark_log_results(int n, unsigned size, 79 unsigned int time, 80 unsigned sdomain, unsigned ddomain, 81 char *kind) 82 { 83 unsigned int throughput = (n * (size >> 10)) / time; 84 DRM_INFO("radeon: %s %u bo moves of %u kB from" 85 " %d to %d in %u ms, throughput: %u Mb/s or %u MB/s\n", 86 kind, n, size >> 10, sdomain, ddomain, time, 87 throughput * 8, throughput); 88 } 89 90 static void radeon_benchmark_move(struct radeon_device *rdev, unsigned size, 91 unsigned sdomain, unsigned ddomain) 92 { 93 struct radeon_bo *dobj = NULL; 94 struct radeon_bo *sobj = NULL; 95 uint64_t saddr, daddr; 96 int r, n; 97 int time; 98 99 n = RADEON_BENCHMARK_ITERATIONS; 100 r = radeon_bo_create(rdev, size, PAGE_SIZE, true, sdomain, 0, NULL, &sobj); 101 if (r) { 102 goto out_cleanup; 103 } 104 r = radeon_bo_reserve(sobj, false); 105 if (unlikely(r != 0)) 106 goto out_cleanup; 107 r = radeon_bo_pin(sobj, sdomain, (u64 *)&saddr); 108 radeon_bo_unreserve(sobj); 109 if (r) { 110 goto out_cleanup; 111 } 112 r = radeon_bo_create(rdev, size, PAGE_SIZE, true, ddomain, 0, NULL, &dobj); 113 if (r) { 114 goto out_cleanup; 115 } 116 r = radeon_bo_reserve(dobj, false); 117 if (unlikely(r != 0)) 118 goto out_cleanup; 119 r = radeon_bo_pin(dobj, ddomain, (u64 *)&daddr); 120 radeon_bo_unreserve(dobj); 121 if (r) { 122 goto out_cleanup; 123 } 124 125 if (rdev->asic->copy.dma) { 126 time = radeon_benchmark_do_move(rdev, size, saddr, daddr, 127 RADEON_BENCHMARK_COPY_DMA, n); 128 if (time < 0) 129 goto out_cleanup; 130 if (time > 0) 131 radeon_benchmark_log_results(n, size, time, 132 sdomain, ddomain, "dma"); 133 } 134 135 if (rdev->asic->copy.blit) { 136 time = radeon_benchmark_do_move(rdev, size, saddr, daddr, 137 RADEON_BENCHMARK_COPY_BLIT, n); 138 if (time < 0) 139 goto out_cleanup; 140 if (time > 0) 141 radeon_benchmark_log_results(n, size, time, 142 sdomain, ddomain, "blit"); 143 } 144 145 out_cleanup: 146 if (sobj) { 147 r = radeon_bo_reserve(sobj, false); 148 if (likely(r == 0)) { 149 radeon_bo_unpin(sobj); 150 radeon_bo_unreserve(sobj); 151 } 152 radeon_bo_unref(&sobj); 153 } 154 if (dobj) { 155 r = radeon_bo_reserve(dobj, false); 156 if (likely(r == 0)) { 157 radeon_bo_unpin(dobj); 158 radeon_bo_unreserve(dobj); 159 } 160 radeon_bo_unref(&dobj); 161 } 162 163 if (r) { 164 DRM_ERROR("Error while benchmarking BO move.\n"); 165 } 166 } 167 168 void radeon_benchmark(struct radeon_device *rdev, int test_number) 169 { 170 int i; 171 int common_modes[RADEON_BENCHMARK_COMMON_MODES_N] = { 172 640 * 480 * 4, 173 720 * 480 * 4, 174 800 * 600 * 4, 175 848 * 480 * 4, 176 1024 * 768 * 4, 177 1152 * 768 * 4, 178 1280 * 720 * 4, 179 1280 * 800 * 4, 180 1280 * 854 * 4, 181 1280 * 960 * 4, 182 1280 * 1024 * 4, 183 1440 * 900 * 4, 184 1400 * 1050 * 4, 185 1680 * 1050 * 4, 186 1600 * 1200 * 4, 187 1920 * 1080 * 4, 188 1920 * 1200 * 4 189 }; 190 191 switch (test_number) { 192 case 1: 193 /* simple test, VRAM to GTT and GTT to VRAM */ 194 radeon_benchmark_move(rdev, 1024*1024, RADEON_GEM_DOMAIN_GTT, 195 RADEON_GEM_DOMAIN_VRAM); 196 radeon_benchmark_move(rdev, 1024*1024, RADEON_GEM_DOMAIN_VRAM, 197 RADEON_GEM_DOMAIN_GTT); 198 break; 199 case 2: 200 /* simple test, VRAM to VRAM */ 201 radeon_benchmark_move(rdev, 1024*1024, RADEON_GEM_DOMAIN_VRAM, 202 RADEON_GEM_DOMAIN_VRAM); 203 break; 204 case 3: 205 /* GTT to VRAM, buffer size sweep, powers of 2 */ 206 for (i = 1; i <= 16384; i <<= 1) 207 radeon_benchmark_move(rdev, i * RADEON_GPU_PAGE_SIZE, 208 RADEON_GEM_DOMAIN_GTT, 209 RADEON_GEM_DOMAIN_VRAM); 210 break; 211 case 4: 212 /* VRAM to GTT, 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_VRAM, 216 RADEON_GEM_DOMAIN_GTT); 217 break; 218 case 5: 219 /* VRAM to VRAM, 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_VRAM); 224 break; 225 case 6: 226 /* GTT to VRAM, buffer size sweep, common modes */ 227 for (i = 0; i < RADEON_BENCHMARK_COMMON_MODES_N; i++) 228 radeon_benchmark_move(rdev, common_modes[i], 229 RADEON_GEM_DOMAIN_GTT, 230 RADEON_GEM_DOMAIN_VRAM); 231 break; 232 case 7: 233 /* VRAM to GTT, 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_VRAM, 237 RADEON_GEM_DOMAIN_GTT); 238 break; 239 case 8: 240 /* VRAM to VRAM, 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_VRAM); 245 break; 246 247 default: 248 DRM_ERROR("Unknown benchmark\n"); 249 } 250 } 251