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
radeon_benchmark_do_move(struct radeon_device * rdev,unsigned size,uint64_t saddr,uint64_t daddr,int flag,int n)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
radeon_benchmark_log_results(int n,unsigned size,unsigned int time,unsigned sdomain,unsigned ddomain,char * kind)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
radeon_benchmark_move(struct radeon_device * rdev,unsigned size,unsigned sdomain,unsigned ddomain)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 if (rdev->asic->copy.blit) {
145 time = radeon_benchmark_do_move(rdev, size, saddr, daddr,
146 RADEON_BENCHMARK_COPY_BLIT, n);
147 if (time < 0)
148 goto out_cleanup;
149 if (time > 0)
150 radeon_benchmark_log_results(n, size, time,
151 sdomain, ddomain, "blit");
152 }
153
154 out_cleanup:
155 if (sobj) {
156 r = radeon_bo_reserve(sobj, false);
157 if (likely(r == 0)) {
158 radeon_bo_unpin(sobj);
159 radeon_bo_unreserve(sobj);
160 }
161 radeon_bo_unref(&sobj);
162 }
163 if (dobj) {
164 r = radeon_bo_reserve(dobj, false);
165 if (likely(r == 0)) {
166 radeon_bo_unpin(dobj);
167 radeon_bo_unreserve(dobj);
168 }
169 radeon_bo_unref(&dobj);
170 }
171
172 if (r) {
173 DRM_ERROR("Error while benchmarking BO move.\n");
174 }
175 }
176
radeon_benchmark(struct radeon_device * rdev,int test_number)177 void radeon_benchmark(struct radeon_device *rdev, int test_number)
178 {
179 int i;
180 int common_modes[RADEON_BENCHMARK_COMMON_MODES_N] = {
181 640 * 480 * 4,
182 720 * 480 * 4,
183 800 * 600 * 4,
184 848 * 480 * 4,
185 1024 * 768 * 4,
186 1152 * 768 * 4,
187 1280 * 720 * 4,
188 1280 * 800 * 4,
189 1280 * 854 * 4,
190 1280 * 960 * 4,
191 1280 * 1024 * 4,
192 1440 * 900 * 4,
193 1400 * 1050 * 4,
194 1680 * 1050 * 4,
195 1600 * 1200 * 4,
196 1920 * 1080 * 4,
197 1920 * 1200 * 4
198 };
199
200 switch (test_number) {
201 case 1:
202 /* simple test, VRAM to GTT and GTT to VRAM */
203 radeon_benchmark_move(rdev, 1024*1024, RADEON_GEM_DOMAIN_GTT,
204 RADEON_GEM_DOMAIN_VRAM);
205 radeon_benchmark_move(rdev, 1024*1024, RADEON_GEM_DOMAIN_VRAM,
206 RADEON_GEM_DOMAIN_GTT);
207 break;
208 case 2:
209 /* simple test, VRAM to VRAM */
210 radeon_benchmark_move(rdev, 1024*1024, RADEON_GEM_DOMAIN_VRAM,
211 RADEON_GEM_DOMAIN_VRAM);
212 break;
213 case 3:
214 /* GTT to VRAM, buffer size sweep, powers of 2 */
215 for (i = 1; i <= 16384; i <<= 1)
216 radeon_benchmark_move(rdev, i * RADEON_GPU_PAGE_SIZE,
217 RADEON_GEM_DOMAIN_GTT,
218 RADEON_GEM_DOMAIN_VRAM);
219 break;
220 case 4:
221 /* VRAM to GTT, buffer size sweep, powers of 2 */
222 for (i = 1; i <= 16384; i <<= 1)
223 radeon_benchmark_move(rdev, i * RADEON_GPU_PAGE_SIZE,
224 RADEON_GEM_DOMAIN_VRAM,
225 RADEON_GEM_DOMAIN_GTT);
226 break;
227 case 5:
228 /* VRAM to VRAM, buffer size sweep, powers of 2 */
229 for (i = 1; i <= 16384; i <<= 1)
230 radeon_benchmark_move(rdev, i * RADEON_GPU_PAGE_SIZE,
231 RADEON_GEM_DOMAIN_VRAM,
232 RADEON_GEM_DOMAIN_VRAM);
233 break;
234 case 6:
235 /* GTT to VRAM, buffer size sweep, common modes */
236 for (i = 0; i < RADEON_BENCHMARK_COMMON_MODES_N; i++)
237 radeon_benchmark_move(rdev, common_modes[i],
238 RADEON_GEM_DOMAIN_GTT,
239 RADEON_GEM_DOMAIN_VRAM);
240 break;
241 case 7:
242 /* VRAM to GTT, buffer size sweep, common modes */
243 for (i = 0; i < RADEON_BENCHMARK_COMMON_MODES_N; i++)
244 radeon_benchmark_move(rdev, common_modes[i],
245 RADEON_GEM_DOMAIN_VRAM,
246 RADEON_GEM_DOMAIN_GTT);
247 break;
248 case 8:
249 /* VRAM to VRAM, buffer size sweep, common modes */
250 for (i = 0; i < RADEON_BENCHMARK_COMMON_MODES_N; i++)
251 radeon_benchmark_move(rdev, common_modes[i],
252 RADEON_GEM_DOMAIN_VRAM,
253 RADEON_GEM_DOMAIN_VRAM);
254 break;
255
256 default:
257 DRM_ERROR("Unknown benchmark\n");
258 }
259 }
260