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
25 #include <sys/cdefs.h>
26 __FBSDID("$FreeBSD$");
27
28 #include <dev/drm2/drmP.h>
29 #include <dev/drm2/radeon/radeon_drm.h>
30 #include "radeon_reg.h"
31 #include "radeon.h"
32
33 #define RADEON_BENCHMARK_COPY_BLIT 1
34 #define RADEON_BENCHMARK_COPY_DMA 0
35
36 #define RADEON_BENCHMARK_ITERATIONS 1024
37 #define RADEON_BENCHMARK_COMMON_MODES_N 17
38
radeon_benchmark_do_move(struct radeon_device * rdev,unsigned size,uint64_t saddr,uint64_t daddr,int flag,int n)39 static int radeon_benchmark_do_move(struct radeon_device *rdev, unsigned size,
40 uint64_t saddr, uint64_t daddr,
41 int flag, int n)
42 {
43 unsigned long start_jiffies;
44 unsigned long end_jiffies;
45 struct radeon_fence *fence = NULL;
46 int i, r;
47
48 start_jiffies = jiffies;
49 for (i = 0; i < n; i++) {
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
75 exit_do_move:
76 if (fence)
77 radeon_fence_unref(&fence);
78 return r;
79 }
80
81
radeon_benchmark_log_results(int n,unsigned size,unsigned int time,unsigned sdomain,unsigned ddomain,char * kind)82 static 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
radeon_benchmark_move(struct radeon_device * rdev,unsigned size,unsigned sdomain,unsigned ddomain)94 static void radeon_benchmark_move(struct radeon_device *rdev, unsigned size,
95 unsigned sdomain, unsigned ddomain)
96 {
97 struct radeon_bo *dobj = NULL;
98 struct radeon_bo *sobj = NULL;
99 uint64_t saddr, daddr;
100 int r, n;
101 int time;
102
103 n = RADEON_BENCHMARK_ITERATIONS;
104 r = radeon_bo_create(rdev, size, PAGE_SIZE, true, sdomain, NULL, &sobj);
105 if (r) {
106 goto out_cleanup;
107 }
108 r = radeon_bo_reserve(sobj, false);
109 if (unlikely(r != 0))
110 goto out_cleanup;
111 r = radeon_bo_pin(sobj, sdomain, &saddr);
112 radeon_bo_unreserve(sobj);
113 if (r) {
114 goto out_cleanup;
115 }
116 r = radeon_bo_create(rdev, size, PAGE_SIZE, true, ddomain, NULL, &dobj);
117 if (r) {
118 goto out_cleanup;
119 }
120 r = radeon_bo_reserve(dobj, false);
121 if (unlikely(r != 0))
122 goto out_cleanup;
123 r = radeon_bo_pin(dobj, ddomain, &daddr);
124 radeon_bo_unreserve(dobj);
125 if (r) {
126 goto out_cleanup;
127 }
128
129 /* r100 doesn't have dma engine so skip the test */
130 /* also, VRAM-to-VRAM test doesn't make much sense for DMA */
131 /* skip it as well if domains are the same */
132 if ((rdev->asic->copy.dma) && (sdomain != ddomain)) {
133 time = radeon_benchmark_do_move(rdev, size, saddr, daddr,
134 RADEON_BENCHMARK_COPY_DMA, n);
135 if (time < 0)
136 goto out_cleanup;
137 if (time > 0)
138 radeon_benchmark_log_results(n, size, time,
139 sdomain, ddomain, "dma");
140 }
141
142 if (rdev->asic->copy.blit) {
143 time = radeon_benchmark_do_move(rdev, size, saddr, daddr,
144 RADEON_BENCHMARK_COPY_BLIT, n);
145 if (time < 0)
146 goto out_cleanup;
147 if (time > 0)
148 radeon_benchmark_log_results(n, size, time,
149 sdomain, ddomain, "blit");
150 }
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
radeon_benchmark(struct radeon_device * rdev,int test_number)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