aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/gpu/drm/radeon/radeon_benchmark.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/gpu/drm/radeon/radeon_benchmark.c')
-rw-r--r--drivers/gpu/drm/radeon/radeon_benchmark.c247
1 files changed, 171 insertions, 76 deletions
diff --git a/drivers/gpu/drm/radeon/radeon_benchmark.c b/drivers/gpu/drm/radeon/radeon_benchmark.c
index 10191d9372d8..5cafc90de7f8 100644
--- a/drivers/gpu/drm/radeon/radeon_benchmark.c
+++ b/drivers/gpu/drm/radeon/radeon_benchmark.c
@@ -26,21 +26,81 @@
26#include "radeon_reg.h" 26#include "radeon_reg.h"
27#include "radeon.h" 27#include "radeon.h"
28 28
29void radeon_benchmark_move(struct radeon_device *rdev, unsigned bsize, 29#define RADEON_BENCHMARK_COPY_BLIT 1
30 unsigned sdomain, unsigned ddomain) 30#define RADEON_BENCHMARK_COPY_DMA 0
31
32#define RADEON_BENCHMARK_ITERATIONS 1024
33#define RADEON_BENCHMARK_COMMON_MODES_N 17
34
35static 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 r = radeon_fence_create(rdev, &fence);
47 if (r)
48 return r;
49
50 switch (flag) {
51 case RADEON_BENCHMARK_COPY_DMA:
52 r = radeon_copy_dma(rdev, saddr, daddr,
53 size / RADEON_GPU_PAGE_SIZE,
54 fence);
55 break;
56 case RADEON_BENCHMARK_COPY_BLIT:
57 r = radeon_copy_blit(rdev, saddr, daddr,
58 size / RADEON_GPU_PAGE_SIZE,
59 fence);
60 break;
61 default:
62 DRM_ERROR("Unknown copy method\n");
63 r = -EINVAL;
64 }
65 if (r)
66 goto exit_do_move;
67 r = radeon_fence_wait(fence, false);
68 if (r)
69 goto exit_do_move;
70 radeon_fence_unref(&fence);
71 }
72 end_jiffies = jiffies;
73 r = jiffies_to_msecs(end_jiffies - start_jiffies);
74
75exit_do_move:
76 if (fence)
77 radeon_fence_unref(&fence);
78 return r;
79}
80
81
82static void radeon_benchmark_log_results(int n, unsigned size,
83 unsigned int time,
84 unsigned sdomain, unsigned ddomain,
85 char *kind)
86{
87 unsigned int throughput = (n * (size >> 10)) / time;
88 DRM_INFO("radeon: %s %u bo moves of %u kB from"
89 " %d to %d in %u ms, throughput: %u Mb/s or %u MB/s\n",
90 kind, n, size >> 10, sdomain, ddomain, time,
91 throughput * 8, throughput);
92}
93
94static void radeon_benchmark_move(struct radeon_device *rdev, unsigned size,
95 unsigned sdomain, unsigned ddomain)
31{ 96{
32 struct radeon_bo *dobj = NULL; 97 struct radeon_bo *dobj = NULL;
33 struct radeon_bo *sobj = NULL; 98 struct radeon_bo *sobj = NULL;
34 struct radeon_fence *fence = NULL;
35 uint64_t saddr, daddr; 99 uint64_t saddr, daddr;
36 unsigned long start_jiffies; 100 int r, n;
37 unsigned long end_jiffies; 101 unsigned int time;
38 unsigned long time;
39 unsigned i, n, size;
40 int r;
41 102
42 size = bsize; 103 n = RADEON_BENCHMARK_ITERATIONS;
43 n = 1024;
44 r = radeon_bo_create(rdev, size, PAGE_SIZE, true, sdomain, &sobj); 104 r = radeon_bo_create(rdev, size, PAGE_SIZE, true, sdomain, &sobj);
45 if (r) { 105 if (r) {
46 goto out_cleanup; 106 goto out_cleanup;
@@ -67,65 +127,26 @@ void radeon_benchmark_move(struct radeon_device *rdev, unsigned bsize,
67 } 127 }
68 128
69 /* r100 doesn't have dma engine so skip the test */ 129 /* r100 doesn't have dma engine so skip the test */
70 if (rdev->asic->copy_dma) { 130 /* also, VRAM-to-VRAM test doesn't make much sense for DMA */
71 131 /* skip it as well if domains are the same */
72 start_jiffies = jiffies; 132 if ((rdev->asic->copy_dma) && (sdomain != ddomain)) {
73 for (i = 0; i < n; i++) { 133 time = radeon_benchmark_do_move(rdev, size, saddr, daddr,
74 r = radeon_fence_create(rdev, &fence); 134 RADEON_BENCHMARK_COPY_DMA, n);
75 if (r) { 135 if (time < 0)
76 goto out_cleanup;
77 }
78
79 r = radeon_copy_dma(rdev, saddr, daddr,
80 size / RADEON_GPU_PAGE_SIZE, fence);
81
82 if (r) {
83 goto out_cleanup;
84 }
85 r = radeon_fence_wait(fence, false);
86 if (r) {
87 goto out_cleanup;
88 }
89 radeon_fence_unref(&fence);
90 }
91 end_jiffies = jiffies;
92 time = end_jiffies - start_jiffies;
93 time = jiffies_to_msecs(time);
94 if (time > 0) {
95 i = ((n * size) >> 10) / time;
96 printk(KERN_INFO "radeon: dma %u bo moves of %ukb from"
97 " %d to %d in %lums (%ukb/ms %ukb/s %uM/s)\n",
98 n, size >> 10,
99 sdomain, ddomain, time,
100 i, i * 1000, (i * 1000) / 1024);
101 }
102 }
103
104 start_jiffies = jiffies;
105 for (i = 0; i < n; i++) {
106 r = radeon_fence_create(rdev, &fence);
107 if (r) {
108 goto out_cleanup;
109 }
110 r = radeon_copy_blit(rdev, saddr, daddr, size / RADEON_GPU_PAGE_SIZE, fence);
111 if (r) {
112 goto out_cleanup;
113 }
114 r = radeon_fence_wait(fence, false);
115 if (r) {
116 goto out_cleanup; 136 goto out_cleanup;
117 } 137 if (time > 0)
118 radeon_fence_unref(&fence); 138 radeon_benchmark_log_results(n, size, time,
119 } 139 sdomain, ddomain, "dma");
120 end_jiffies = jiffies;
121 time = end_jiffies - start_jiffies;
122 time = jiffies_to_msecs(time);
123 if (time > 0) {
124 i = ((n * size) >> 10) / time;
125 printk(KERN_INFO "radeon: blit %u bo moves of %ukb from %d to %d"
126 " in %lums (%ukb/ms %ukb/s %uM/s)\n", n, size >> 10,
127 sdomain, ddomain, time, i, i * 1000, (i * 1000) / 1024);
128 } 140 }
141
142 time = radeon_benchmark_do_move(rdev, size, saddr, daddr,
143 RADEON_BENCHMARK_COPY_BLIT, n);
144 if (time < 0)
145 goto out_cleanup;
146 if (time > 0)
147 radeon_benchmark_log_results(n, size, time,
148 sdomain, ddomain, "blit");
149
129out_cleanup: 150out_cleanup:
130 if (sobj) { 151 if (sobj) {
131 r = radeon_bo_reserve(sobj, false); 152 r = radeon_bo_reserve(sobj, false);
@@ -143,18 +164,92 @@ out_cleanup:
143 } 164 }
144 radeon_bo_unref(&dobj); 165 radeon_bo_unref(&dobj);
145 } 166 }
146 if (fence) { 167
147 radeon_fence_unref(&fence);
148 }
149 if (r) { 168 if (r) {
150 printk(KERN_WARNING "Error while benchmarking BO move.\n"); 169 DRM_ERROR("Error while benchmarking BO move.\n");
151 } 170 }
152} 171}
153 172
154void radeon_benchmark(struct radeon_device *rdev) 173void radeon_benchmark(struct radeon_device *rdev, int test_number)
155{ 174{
156 radeon_benchmark_move(rdev, 1024*1024, RADEON_GEM_DOMAIN_GTT, 175 int i;
157 RADEON_GEM_DOMAIN_VRAM); 176 int common_modes[RADEON_BENCHMARK_COMMON_MODES_N] = {
158 radeon_benchmark_move(rdev, 1024*1024, RADEON_GEM_DOMAIN_VRAM, 177 640 * 480 * 4,
159 RADEON_GEM_DOMAIN_GTT); 178 720 * 480 * 4,
179 800 * 600 * 4,
180 848 * 480 * 4,
181 1024 * 768 * 4,
182 1152 * 768 * 4,
183 1280 * 720 * 4,
184 1280 * 800 * 4,
185 1280 * 854 * 4,
186 1280 * 960 * 4,
187 1280 * 1024 * 4,
188 1440 * 900 * 4,
189 1400 * 1050 * 4,
190 1680 * 1050 * 4,
191 1600 * 1200 * 4,
192 1920 * 1080 * 4,
193 1920 * 1200 * 4
194 };
195
196 switch (test_number) {
197 case 1:
198 /* simple test, VRAM to GTT and GTT to VRAM */
199 radeon_benchmark_move(rdev, 1024*1024, RADEON_GEM_DOMAIN_GTT,
200 RADEON_GEM_DOMAIN_VRAM);
201 radeon_benchmark_move(rdev, 1024*1024, RADEON_GEM_DOMAIN_VRAM,
202 RADEON_GEM_DOMAIN_GTT);
203 break;
204 case 2:
205 /* simple test, VRAM to VRAM */
206 radeon_benchmark_move(rdev, 1024*1024, RADEON_GEM_DOMAIN_VRAM,
207 RADEON_GEM_DOMAIN_VRAM);
208 break;
209 case 3:
210 /* GTT to VRAM, buffer size sweep, powers of 2 */
211 for (i = 1; i <= 65536; i <<= 1)
212 radeon_benchmark_move(rdev, i*1024,
213 RADEON_GEM_DOMAIN_GTT,
214 RADEON_GEM_DOMAIN_VRAM);
215 break;
216 case 4:
217 /* VRAM to GTT, buffer size sweep, powers of 2 */
218 for (i = 1; i <= 65536; i <<= 1)
219 radeon_benchmark_move(rdev, i*1024,
220 RADEON_GEM_DOMAIN_VRAM,
221 RADEON_GEM_DOMAIN_GTT);
222 break;
223 case 5:
224 /* VRAM to VRAM, buffer size sweep, powers of 2 */
225 for (i = 1; i <= 65536; i <<= 1)
226 radeon_benchmark_move(rdev, i*1024,
227 RADEON_GEM_DOMAIN_VRAM,
228 RADEON_GEM_DOMAIN_VRAM);
229 break;
230 case 6:
231 /* GTT to VRAM, buffer size sweep, common modes */
232 for (i = 1; i < RADEON_BENCHMARK_COMMON_MODES_N; i++)
233 radeon_benchmark_move(rdev, common_modes[i],
234 RADEON_GEM_DOMAIN_GTT,
235 RADEON_GEM_DOMAIN_VRAM);
236 break;
237 case 7:
238 /* VRAM to GTT, buffer size sweep, common modes */
239 for (i = 1; i < RADEON_BENCHMARK_COMMON_MODES_N; i++)
240 radeon_benchmark_move(rdev, common_modes[i],
241 RADEON_GEM_DOMAIN_VRAM,
242 RADEON_GEM_DOMAIN_GTT);
243 break;
244 case 8:
245 /* VRAM to VRAM, buffer size sweep, common modes */
246 for (i = 1; i < RADEON_BENCHMARK_COMMON_MODES_N; i++)
247 radeon_benchmark_move(rdev, common_modes[i],
248 RADEON_GEM_DOMAIN_VRAM,
249 RADEON_GEM_DOMAIN_VRAM);
250 break;
251
252 default:
253 DRM_ERROR("Unknown benchmark\n");
254 }
160} 255}
generated by cgit v1.2.2 (git 2.25.0) at 2025-02-09 01:56:27 -0500