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 * $FreeBSD: head/sys/dev/drm2/radeon/radeon_benchmark.c 254885 2013-08-25 19:37:15Z dumbbell $ 25 */ 26 27 #include <drm/drmP.h> 28 #include <uapi_drm/radeon_drm.h> 29 #include "radeon_reg.h" 30 #include "radeon.h" 31 32 #define RADEON_BENCHMARK_COPY_BLIT 1 33 #define RADEON_BENCHMARK_COPY_DMA 0 34 35 #define RADEON_BENCHMARK_ITERATIONS 1024 36 #define RADEON_BENCHMARK_COMMON_MODES_N 17 37 38 static int radeon_benchmark_do_move(struct radeon_device *rdev, unsigned size, 39 uint64_t saddr, uint64_t daddr, 40 int flag, int n) 41 { 42 unsigned long start_jiffies; 43 unsigned long end_jiffies; 44 struct radeon_fence *fence = NULL; 45 int i, r; 46 47 start_jiffies = jiffies; 48 for (i = 0; i < n; i++) { 49 switch (flag) { 50 case RADEON_BENCHMARK_COPY_DMA: 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_copy_blit(rdev, saddr, daddr, 57 size / RADEON_GPU_PAGE_SIZE, 58 &fence); 59 break; 60 default: 61 DRM_ERROR("Unknown copy method\n"); 62 r = -EINVAL; 63 } 64 if (r) 65 goto exit_do_move; 66 r = radeon_fence_wait(fence, false); 67 if (r) 68 goto exit_do_move; 69 radeon_fence_unref(&fence); 70 } 71 end_jiffies = jiffies; 72 r = jiffies_to_msecs(end_jiffies - start_jiffies); 73 74 exit_do_move: 75 if (fence) 76 radeon_fence_unref(&fence); 77 return r; 78 } 79 80 81 static void radeon_benchmark_log_results(int n, unsigned size, 82 unsigned int time, 83 unsigned sdomain, unsigned ddomain, 84 char *kind) 85 { 86 unsigned int throughput = (n * (size >> 10)) / time; 87 DRM_INFO("radeon: %s %u bo moves of %u kB from" 88 " %d to %d in %u ms, throughput: %u Mb/s or %u MB/s\n", 89 kind, n, size >> 10, sdomain, ddomain, time, 90 throughput * 8, throughput); 91 } 92 93 static void radeon_benchmark_move(struct radeon_device *rdev, unsigned size, 94 unsigned sdomain, unsigned ddomain) 95 { 96 struct radeon_bo *dobj = NULL; 97 struct radeon_bo *sobj = NULL; 98 uint64_t saddr, daddr; 99 int r, n; 100 int time; 101 102 n = RADEON_BENCHMARK_ITERATIONS; 103 r = radeon_bo_create(rdev, size, PAGE_SIZE, true, sdomain, NULL, &sobj); 104 if (r) { 105 goto out_cleanup; 106 } 107 r = radeon_bo_reserve(sobj, false); 108 if (unlikely(r != 0)) 109 goto out_cleanup; 110 r = radeon_bo_pin(sobj, sdomain, &saddr); 111 radeon_bo_unreserve(sobj); 112 if (r) { 113 goto out_cleanup; 114 } 115 r = radeon_bo_create(rdev, size, PAGE_SIZE, true, ddomain, NULL, &dobj); 116 if (r) { 117 goto out_cleanup; 118 } 119 r = radeon_bo_reserve(dobj, false); 120 if (unlikely(r != 0)) 121 goto out_cleanup; 122 r = radeon_bo_pin(dobj, ddomain, &daddr); 123 radeon_bo_unreserve(dobj); 124 if (r) { 125 goto out_cleanup; 126 } 127 128 /* r100 doesn't have dma engine so skip the test */ 129 /* also, VRAM-to-VRAM test doesn't make much sense for DMA */ 130 /* skip it as well if domains are the same */ 131 if ((rdev->asic->copy.dma) && (sdomain != ddomain)) { 132 time = radeon_benchmark_do_move(rdev, size, saddr, daddr, 133 RADEON_BENCHMARK_COPY_DMA, n); 134 if (time < 0) 135 goto out_cleanup; 136 if (time > 0) 137 radeon_benchmark_log_results(n, size, time, 138 sdomain, ddomain, "dma"); 139 } 140 141 time = radeon_benchmark_do_move(rdev, size, saddr, daddr, 142 RADEON_BENCHMARK_COPY_BLIT, n); 143 if (time < 0) 144 goto out_cleanup; 145 if (time > 0) 146 radeon_benchmark_log_results(n, size, time, 147 sdomain, ddomain, "blit"); 148 149 out_cleanup: 150 if (sobj) { 151 r = radeon_bo_reserve(sobj, false); 152 if (likely(r == 0)) { 153 radeon_bo_unpin(sobj); 154 radeon_bo_unreserve(sobj); 155 } 156 radeon_bo_unref(&sobj); 157 } 158 if (dobj) { 159 r = radeon_bo_reserve(dobj, false); 160 if (likely(r == 0)) { 161 radeon_bo_unpin(dobj); 162 radeon_bo_unreserve(dobj); 163 } 164 radeon_bo_unref(&dobj); 165 } 166 167 if (r) { 168 DRM_ERROR("Error while benchmarking BO move.\n"); 169 } 170 } 171 172 void radeon_benchmark(struct radeon_device *rdev, int test_number) 173 { 174 int i; 175 int common_modes[RADEON_BENCHMARK_COMMON_MODES_N] = { 176 640 * 480 * 4, 177 720 * 480 * 4, 178 800 * 600 * 4, 179 848 * 480 * 4, 180 1024 * 768 * 4, 181 1152 * 768 * 4, 182 1280 * 720 * 4, 183 1280 * 800 * 4, 184 1280 * 854 * 4, 185 1280 * 960 * 4, 186 1280 * 1024 * 4, 187 1440 * 900 * 4, 188 1400 * 1050 * 4, 189 1680 * 1050 * 4, 190 1600 * 1200 * 4, 191 1920 * 1080 * 4, 192 1920 * 1200 * 4 193 }; 194 195 switch (test_number) { 196 case 1: 197 /* simple test, VRAM to GTT and GTT to VRAM */ 198 radeon_benchmark_move(rdev, 1024*1024, RADEON_GEM_DOMAIN_GTT, 199 RADEON_GEM_DOMAIN_VRAM); 200 radeon_benchmark_move(rdev, 1024*1024, RADEON_GEM_DOMAIN_VRAM, 201 RADEON_GEM_DOMAIN_GTT); 202 break; 203 case 2: 204 /* simple test, VRAM to VRAM */ 205 radeon_benchmark_move(rdev, 1024*1024, RADEON_GEM_DOMAIN_VRAM, 206 RADEON_GEM_DOMAIN_VRAM); 207 break; 208 case 3: 209 /* GTT to VRAM, buffer size sweep, powers of 2 */ 210 for (i = 1; i <= 16384; i <<= 1) 211 radeon_benchmark_move(rdev, i * RADEON_GPU_PAGE_SIZE, 212 RADEON_GEM_DOMAIN_GTT, 213 RADEON_GEM_DOMAIN_VRAM); 214 break; 215 case 4: 216 /* VRAM to GTT, buffer size sweep, powers of 2 */ 217 for (i = 1; i <= 16384; i <<= 1) 218 radeon_benchmark_move(rdev, i * RADEON_GPU_PAGE_SIZE, 219 RADEON_GEM_DOMAIN_VRAM, 220 RADEON_GEM_DOMAIN_GTT); 221 break; 222 case 5: 223 /* VRAM to VRAM, buffer size sweep, powers of 2 */ 224 for (i = 1; i <= 16384; i <<= 1) 225 radeon_benchmark_move(rdev, i * RADEON_GPU_PAGE_SIZE, 226 RADEON_GEM_DOMAIN_VRAM, 227 RADEON_GEM_DOMAIN_VRAM); 228 break; 229 case 6: 230 /* GTT to VRAM, buffer size sweep, common modes */ 231 for (i = 0; i < RADEON_BENCHMARK_COMMON_MODES_N; i++) 232 radeon_benchmark_move(rdev, common_modes[i], 233 RADEON_GEM_DOMAIN_GTT, 234 RADEON_GEM_DOMAIN_VRAM); 235 break; 236 case 7: 237 /* VRAM to GTT, buffer size sweep, common modes */ 238 for (i = 0; i < RADEON_BENCHMARK_COMMON_MODES_N; i++) 239 radeon_benchmark_move(rdev, common_modes[i], 240 RADEON_GEM_DOMAIN_VRAM, 241 RADEON_GEM_DOMAIN_GTT); 242 break; 243 case 8: 244 /* VRAM to VRAM, buffer size sweep, common modes */ 245 for (i = 0; i < RADEON_BENCHMARK_COMMON_MODES_N; i++) 246 radeon_benchmark_move(rdev, common_modes[i], 247 RADEON_GEM_DOMAIN_VRAM, 248 RADEON_GEM_DOMAIN_VRAM); 249 break; 250 251 default: 252 DRM_ERROR("Unknown benchmark\n"); 253 } 254 } 255