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