1 /* $NetBSD: radeon_benchmark.c,v 1.4 2021/12/18 23:45:43 riastradh Exp $ */
2
3 /*
4 * Copyright 2009 Jerome Glisse.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22 * OTHER DEALINGS IN THE SOFTWARE.
23 *
24 * Authors: Jerome Glisse
25 */
26
27 #include <sys/cdefs.h>
28 __KERNEL_RCSID(0, "$NetBSD: radeon_benchmark.c,v 1.4 2021/12/18 23:45:43 riastradh Exp $");
29
30 #include <drm/radeon_drm.h>
31 #include "radeon_reg.h"
32 #include "radeon.h"
33
34 #define RADEON_BENCHMARK_COPY_BLIT 1
35 #define RADEON_BENCHMARK_COPY_DMA 0
36
37 #define RADEON_BENCHMARK_ITERATIONS 1024
38 #define RADEON_BENCHMARK_COMMON_MODES_N 17
39
radeon_benchmark_do_move(struct radeon_device * rdev,unsigned size,uint64_t saddr,uint64_t daddr,int flag,int n,struct dma_resv * resv)40 static int radeon_benchmark_do_move(struct radeon_device *rdev, unsigned size,
41 uint64_t saddr, uint64_t daddr,
42 int flag, int n,
43 struct dma_resv *resv)
44 {
45 unsigned long start_jiffies;
46 unsigned long end_jiffies;
47 struct radeon_fence *fence = NULL;
48 int i, r;
49
50 start_jiffies = jiffies;
51 for (i = 0; i < n; i++) {
52 switch (flag) {
53 case RADEON_BENCHMARK_COPY_DMA:
54 fence = radeon_copy_dma(rdev, saddr, daddr,
55 size / RADEON_GPU_PAGE_SIZE,
56 resv);
57 break;
58 case RADEON_BENCHMARK_COPY_BLIT:
59 fence = radeon_copy_blit(rdev, saddr, daddr,
60 size / RADEON_GPU_PAGE_SIZE,
61 resv);
62 break;
63 default:
64 DRM_ERROR("Unknown copy method\n");
65 return -EINVAL;
66 }
67 if (IS_ERR(fence))
68 return PTR_ERR(fence);
69
70 r = radeon_fence_wait(fence, false);
71 radeon_fence_unref(&fence);
72 if (r)
73 return r;
74 }
75 end_jiffies = jiffies;
76 return jiffies_to_msecs(end_jiffies - start_jiffies);
77 }
78
79
radeon_benchmark_log_results(int n,unsigned size,unsigned int time,unsigned sdomain,unsigned ddomain,const char * kind)80 static void radeon_benchmark_log_results(int n, unsigned size,
81 unsigned int time,
82 unsigned sdomain, unsigned ddomain,
83 const char *kind)
84 {
85 unsigned int throughput = (n * (size >> 10)) / time;
86 DRM_INFO("radeon: %s %u bo moves of %u kB from"
87 " %d to %d in %u ms, throughput: %u Mb/s or %u MB/s\n",
88 kind, n, size >> 10, sdomain, ddomain, time,
89 throughput * 8, throughput);
90 }
91
radeon_benchmark_move(struct radeon_device * rdev,unsigned size,unsigned sdomain,unsigned ddomain)92 static void radeon_benchmark_move(struct radeon_device *rdev, unsigned size,
93 unsigned sdomain, unsigned ddomain)
94 {
95 struct radeon_bo *dobj = NULL;
96 struct radeon_bo *sobj = NULL;
97 uint64_t saddr, daddr;
98 int r, n;
99 int time;
100
101 n = RADEON_BENCHMARK_ITERATIONS;
102 r = radeon_bo_create(rdev, size, PAGE_SIZE, true, sdomain, 0, NULL, NULL, &sobj);
103 if (r) {
104 goto out_cleanup;
105 }
106 r = radeon_bo_reserve(sobj, false);
107 if (unlikely(r != 0))
108 goto out_cleanup;
109 r = radeon_bo_pin(sobj, sdomain, &saddr);
110 radeon_bo_unreserve(sobj);
111 if (r) {
112 goto out_cleanup;
113 }
114 r = radeon_bo_create(rdev, size, PAGE_SIZE, true, ddomain, 0, NULL, NULL, &dobj);
115 if (r) {
116 goto out_cleanup;
117 }
118 r = radeon_bo_reserve(dobj, false);
119 if (unlikely(r != 0))
120 goto out_cleanup;
121 r = radeon_bo_pin(dobj, ddomain, &daddr);
122 radeon_bo_unreserve(dobj);
123 if (r) {
124 goto out_cleanup;
125 }
126
127 if (rdev->asic->copy.dma) {
128 time = radeon_benchmark_do_move(rdev, size, saddr, daddr,
129 RADEON_BENCHMARK_COPY_DMA, n,
130 dobj->tbo.base.resv);
131 if (time < 0)
132 goto out_cleanup;
133 if (time > 0)
134 radeon_benchmark_log_results(n, size, time,
135 sdomain, ddomain, "dma");
136 }
137
138 if (rdev->asic->copy.blit) {
139 time = radeon_benchmark_do_move(rdev, size, saddr, daddr,
140 RADEON_BENCHMARK_COPY_BLIT, n,
141 dobj->tbo.base.resv);
142 if (time < 0)
143 goto out_cleanup;
144 if (time > 0)
145 radeon_benchmark_log_results(n, size, time,
146 sdomain, ddomain, "blit");
147 }
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
radeon_benchmark(struct radeon_device * rdev,int test_number)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