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 <drm/amdgpu_drm.h>
26 #include "amdgpu.h"
27 
28 #define AMDGPU_BENCHMARK_ITERATIONS 1024
29 #define AMDGPU_BENCHMARK_COMMON_MODES_N 17
30 
amdgpu_benchmark_do_move(struct amdgpu_device * adev,unsigned size,uint64_t saddr,uint64_t daddr,int n,s64 * time_ms)31 static int amdgpu_benchmark_do_move(struct amdgpu_device *adev, unsigned size,
32 				    uint64_t saddr, uint64_t daddr, int n, s64 *time_ms)
33 {
34 	ktime_t stime, etime;
35 	struct dma_fence *fence;
36 	int i, r;
37 
38 	stime = ktime_get();
39 	for (i = 0; i < n; i++) {
40 		struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
41 		r = amdgpu_copy_buffer(ring, saddr, daddr, size, NULL, &fence,
42 				       false, false, false);
43 		if (r)
44 			goto exit_do_move;
45 		r = dma_fence_wait(fence, false);
46 		dma_fence_put(fence);
47 		if (r)
48 			goto exit_do_move;
49 	}
50 
51 exit_do_move:
52 	etime = ktime_get();
53 	*time_ms = ktime_ms_delta(etime, stime);
54 
55 	return r;
56 }
57 
58 
amdgpu_benchmark_log_results(struct amdgpu_device * adev,int n,unsigned size,s64 time_ms,unsigned sdomain,unsigned ddomain,char * kind)59 static void amdgpu_benchmark_log_results(struct amdgpu_device *adev,
60 					 int n, unsigned size,
61 					 s64 time_ms,
62 					 unsigned sdomain, unsigned ddomain,
63 					 char *kind)
64 {
65 	s64 throughput = (n * (size >> 10));
66 
67 	throughput = div64_s64(throughput, time_ms);
68 
69 	dev_info(adev->dev, "amdgpu: %s %u bo moves of %u kB from"
70 		 " %d to %d in %lld ms, throughput: %lld Mb/s or %lld MB/s\n",
71 		 kind, n, size >> 10, sdomain, ddomain, time_ms,
72 		 throughput * 8, throughput);
73 }
74 
amdgpu_benchmark_move(struct amdgpu_device * adev,unsigned size,unsigned sdomain,unsigned ddomain)75 static int amdgpu_benchmark_move(struct amdgpu_device *adev, unsigned size,
76 				 unsigned sdomain, unsigned ddomain)
77 {
78 	struct amdgpu_bo *dobj = NULL;
79 	struct amdgpu_bo *sobj = NULL;
80 	uint64_t saddr, daddr;
81 	s64 time_ms;
82 	int r, n;
83 
84 	n = AMDGPU_BENCHMARK_ITERATIONS;
85 
86 	r = amdgpu_bo_create_kernel(adev, size,
87 				    PAGE_SIZE, sdomain,
88 				    &sobj,
89 				    &saddr,
90 				    NULL);
91 	if (r)
92 		goto out_cleanup;
93 	r = amdgpu_bo_create_kernel(adev, size,
94 				    PAGE_SIZE, ddomain,
95 				    &dobj,
96 				    &daddr,
97 				    NULL);
98 	if (r)
99 		goto out_cleanup;
100 
101 	if (adev->mman.buffer_funcs) {
102 		r = amdgpu_benchmark_do_move(adev, size, saddr, daddr, n, &time_ms);
103 		if (r)
104 			goto out_cleanup;
105 		else
106 			amdgpu_benchmark_log_results(adev, n, size, time_ms,
107 						     sdomain, ddomain, "dma");
108 	}
109 
110 out_cleanup:
111 	/* Check error value now. The value can be overwritten when clean up.*/
112 	if (r < 0)
113 		dev_info(adev->dev, "Error while benchmarking BO move.\n");
114 
115 	if (sobj)
116 		amdgpu_bo_free_kernel(&sobj, &saddr, NULL);
117 	if (dobj)
118 		amdgpu_bo_free_kernel(&dobj, &daddr, NULL);
119 	return r;
120 }
121 
amdgpu_benchmark(struct amdgpu_device * adev,int test_number)122 int amdgpu_benchmark(struct amdgpu_device *adev, int test_number)
123 {
124 	int i, r;
125 	static const int common_modes[AMDGPU_BENCHMARK_COMMON_MODES_N] = {
126 		640 * 480 * 4,
127 		720 * 480 * 4,
128 		800 * 600 * 4,
129 		848 * 480 * 4,
130 		1024 * 768 * 4,
131 		1152 * 768 * 4,
132 		1280 * 720 * 4,
133 		1280 * 800 * 4,
134 		1280 * 854 * 4,
135 		1280 * 960 * 4,
136 		1280 * 1024 * 4,
137 		1440 * 900 * 4,
138 		1400 * 1050 * 4,
139 		1680 * 1050 * 4,
140 		1600 * 1200 * 4,
141 		1920 * 1080 * 4,
142 		1920 * 1200 * 4
143 	};
144 
145 	mutex_lock(&adev->benchmark_mutex);
146 	switch (test_number) {
147 	case 1:
148 		dev_info(adev->dev,
149 			 "benchmark test: %d (simple test, VRAM to GTT and GTT to VRAM)\n",
150 			 test_number);
151 		/* simple test, VRAM to GTT and GTT to VRAM */
152 		r = amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_GTT,
153 					  AMDGPU_GEM_DOMAIN_VRAM);
154 		if (r)
155 			goto done;
156 		r = amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_VRAM,
157 					  AMDGPU_GEM_DOMAIN_GTT);
158 		if (r)
159 			goto done;
160 		break;
161 	case 2:
162 		dev_info(adev->dev,
163 			 "benchmark test: %d (simple test, VRAM to VRAM)\n",
164 			 test_number);
165 		/* simple test, VRAM to VRAM */
166 		r = amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_VRAM,
167 					  AMDGPU_GEM_DOMAIN_VRAM);
168 		if (r)
169 			goto done;
170 		break;
171 	case 3:
172 		dev_info(adev->dev,
173 			 "benchmark test: %d (GTT to VRAM, buffer size sweep, powers of 2)\n",
174 			 test_number);
175 		/* GTT to VRAM, buffer size sweep, powers of 2 */
176 		for (i = 1; i <= 16384; i <<= 1) {
177 			r = amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE,
178 						  AMDGPU_GEM_DOMAIN_GTT,
179 						  AMDGPU_GEM_DOMAIN_VRAM);
180 			if (r)
181 				goto done;
182 		}
183 		break;
184 	case 4:
185 		dev_info(adev->dev,
186 			 "benchmark test: %d (VRAM to GTT, buffer size sweep, powers of 2)\n",
187 			 test_number);
188 		/* VRAM to GTT, buffer size sweep, powers of 2 */
189 		for (i = 1; i <= 16384; i <<= 1) {
190 			r = amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE,
191 						  AMDGPU_GEM_DOMAIN_VRAM,
192 						  AMDGPU_GEM_DOMAIN_GTT);
193 			if (r)
194 				goto done;
195 		}
196 		break;
197 	case 5:
198 		dev_info(adev->dev,
199 			 "benchmark test: %d (VRAM to VRAM, buffer size sweep, powers of 2)\n",
200 			 test_number);
201 		/* VRAM to VRAM, buffer size sweep, powers of 2 */
202 		for (i = 1; i <= 16384; i <<= 1) {
203 			r = amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE,
204 						  AMDGPU_GEM_DOMAIN_VRAM,
205 						  AMDGPU_GEM_DOMAIN_VRAM);
206 			if (r)
207 				goto done;
208 		}
209 		break;
210 	case 6:
211 		dev_info(adev->dev,
212 			 "benchmark test: %d (GTT to VRAM, buffer size sweep, common modes)\n",
213 			 test_number);
214 		/* GTT to VRAM, buffer size sweep, common modes */
215 		for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++) {
216 			r = amdgpu_benchmark_move(adev, common_modes[i],
217 						  AMDGPU_GEM_DOMAIN_GTT,
218 						  AMDGPU_GEM_DOMAIN_VRAM);
219 			if (r)
220 				goto done;
221 		}
222 		break;
223 	case 7:
224 		dev_info(adev->dev,
225 			 "benchmark test: %d (VRAM to GTT, buffer size sweep, common modes)\n",
226 			 test_number);
227 		/* VRAM to GTT, buffer size sweep, common modes */
228 		for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++) {
229 			r = amdgpu_benchmark_move(adev, common_modes[i],
230 						  AMDGPU_GEM_DOMAIN_VRAM,
231 						  AMDGPU_GEM_DOMAIN_GTT);
232 			if (r)
233 				goto done;
234 		}
235 		break;
236 	case 8:
237 		dev_info(adev->dev,
238 			 "benchmark test: %d (VRAM to VRAM, buffer size sweep, common modes)\n",
239 			 test_number);
240 		/* VRAM to VRAM, buffer size sweep, common modes */
241 		for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++) {
242 			r = amdgpu_benchmark_move(adev, common_modes[i],
243 					      AMDGPU_GEM_DOMAIN_VRAM,
244 					      AMDGPU_GEM_DOMAIN_VRAM);
245 			if (r)
246 				goto done;
247 		}
248 		break;
249 
250 	default:
251 		dev_info(adev->dev, "Unknown benchmark %d\n", test_number);
252 		r = -EINVAL;
253 		break;
254 	}
255 
256 done:
257 	mutex_unlock(&adev->benchmark_mutex);
258 
259 	return r;
260 }
261