1 /*
2 * Copyright © 2011 Intel Corporation
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 (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 *
23 * Authors:
24 * Chris Wilson <chris@chris-wilson.co.uk>
25 *
26 */
27
28 #include "igt.h"
29 #include <unistd.h>
30 #include <stdlib.h>
31 #include <stdint.h>
32 #include <stdio.h>
33 #include <string.h>
34 #include <fcntl.h>
35 #include <inttypes.h>
36 #include <errno.h>
37 #include <math.h>
38 #include <sys/stat.h>
39 #include <sys/ioctl.h>
40 #include <sys/time.h>
41 #include <time.h>
42
43 #include "drm.h"
44
45 #define LOCAL_I915_EXEC_NO_RELOC (1<<11)
46 #define LOCAL_I915_EXEC_HANDLE_LUT (1<<12)
47
48 #define COPY_BLT_CMD (2<<29|0x53<<22|0x6)
49 #define BLT_WRITE_ALPHA (1<<21)
50 #define BLT_WRITE_RGB (1<<20)
51 #define BLT_SRC_TILED (1<<15)
52 #define BLT_DST_TILED (1<<11)
53
54 static int has_64bit_reloc;
55
56 static double
elapsed(const struct timespec * start,const struct timespec * end)57 elapsed(const struct timespec *start, const struct timespec *end)
58 {
59 return (end->tv_sec - start->tv_sec) + 1e-9*(end->tv_nsec - start->tv_nsec);
60 }
61
baseline(uint64_t bytes,int milliseconds)62 static int baseline(uint64_t bytes, int milliseconds)
63 {
64 struct timespec start, end;
65 const int size = 64*1024*1024;
66 int count = 0;
67 void *mem;
68
69 mem = malloc(size);
70 if (mem == NULL)
71 return 1;
72
73 clock_gettime(CLOCK_MONOTONIC, &start);
74 do {
75 memset(mem, count, size);
76 count++;
77 clock_gettime(CLOCK_MONOTONIC, &end);
78 if (elapsed(&start, &end) > 0.1)
79 break;
80 } while (1);
81
82 free(mem);
83
84 return ceil(1e-3*milliseconds/elapsed(&start, &end) * count * size / bytes);
85 }
86
gem_linear_blt(int fd,uint32_t * batch,int offset,uint32_t src,uint32_t dst,uint32_t length,struct drm_i915_gem_relocation_entry * reloc)87 static int gem_linear_blt(int fd,
88 uint32_t *batch,
89 int offset,
90 uint32_t src,
91 uint32_t dst,
92 uint32_t length,
93 struct drm_i915_gem_relocation_entry *reloc)
94 {
95 uint32_t *b = batch + offset/4;
96 int height = length / (16 * 1024);
97
98 igt_assert_lte(height, 1 << 16);
99
100 if (height) {
101 int i = 0;
102 b[i++] = COPY_BLT_CMD | BLT_WRITE_ALPHA | BLT_WRITE_RGB;
103 if (has_64bit_reloc)
104 b[i-1]+=2;
105 b[i++] = 0xcc << 16 | 1 << 25 | 1 << 24 | (16*1024);
106 b[i++] = 0;
107 b[i++] = height << 16 | (4*1024);
108 b[i++] = 0;
109 reloc->offset = (b-batch+4) * sizeof(uint32_t);
110 reloc->delta = 0;
111 reloc->target_handle = dst;
112 reloc->read_domains = I915_GEM_DOMAIN_RENDER;
113 reloc->write_domain = I915_GEM_DOMAIN_RENDER;
114 reloc->presumed_offset = 0;
115 reloc++;
116 if (has_64bit_reloc)
117 b[i++] = 0; /* FIXME */
118
119 b[i++] = 0;
120 b[i++] = 16*1024;
121 b[i++] = 0;
122 reloc->offset = (b-batch+7) * sizeof(uint32_t);
123 if (has_64bit_reloc)
124 reloc->offset += sizeof(uint32_t);
125 reloc->delta = 0;
126 reloc->target_handle = src;
127 reloc->read_domains = I915_GEM_DOMAIN_RENDER;
128 reloc->write_domain = 0;
129 reloc->presumed_offset = 0;
130 reloc++;
131 if (has_64bit_reloc)
132 b[i++] = 0; /* FIXME */
133
134 b += i;
135 length -= height * 16*1024;
136 }
137
138 if (length) {
139 int i = 0;
140 b[i++] = COPY_BLT_CMD | BLT_WRITE_ALPHA | BLT_WRITE_RGB;
141 if (has_64bit_reloc)
142 b[i-1]+=2;
143 b[i++] = 0xcc << 16 | 1 << 25 | 1 << 24 | (16*1024);
144 b[i++] = height << 16;
145 b[i++] = (1+height) << 16 | (length / 4);
146 b[i++] = 0;
147 reloc->offset = (b-batch+4) * sizeof(uint32_t);
148 reloc->delta = 0;
149 reloc->target_handle = dst;
150 reloc->read_domains = I915_GEM_DOMAIN_RENDER;
151 reloc->write_domain = I915_GEM_DOMAIN_RENDER;
152 reloc->presumed_offset = 0;
153 reloc++;
154 if (has_64bit_reloc)
155 b[i++] = 0; /* FIXME */
156
157 b[i++] = height << 16;
158 b[i++] = 16*1024;
159 b[i++] = 0;
160 reloc->offset = (b-batch+7) * sizeof(uint32_t);
161 if (has_64bit_reloc)
162 reloc->offset += sizeof(uint32_t);
163 reloc->delta = 0;
164 reloc->target_handle = src;
165 reloc->read_domains = I915_GEM_DOMAIN_RENDER;
166 reloc->write_domain = 0;
167 reloc->presumed_offset = 0;
168 reloc++;
169 if (has_64bit_reloc)
170 b[i++] = 0; /* FIXME */
171
172 b += i;
173 }
174
175 b[0] = MI_BATCH_BUFFER_END;
176 b[1] = 0;
177
178 return (b+2 - batch) * sizeof(uint32_t);
179 }
180
181 #define SYNC 0x1
182 #define NOCMD 0x2
183
run(int object,int batch,int time,int reps,int ncpus,unsigned flags)184 static int run(int object, int batch, int time, int reps, int ncpus, unsigned flags)
185 {
186 struct drm_i915_gem_execbuffer2 execbuf;
187 struct drm_i915_gem_exec_object2 exec[3];
188 struct drm_i915_gem_relocation_entry *reloc;
189 uint32_t *buf, handle, src, dst;
190 int fd, len, gen, size, nreloc;
191 int ring, count;
192 double *shared;
193
194 shared = mmap(0, 4096, PROT_WRITE, MAP_SHARED | MAP_ANON, -1, 0);
195
196 size = ALIGN(batch * 64, 4096);
197 reloc = malloc(sizeof(*reloc)*size/32*2);
198
199 fd = drm_open_driver(DRIVER_INTEL);
200 handle = gem_create(fd, size);
201 buf = gem_mmap__cpu(fd, handle, 0, size, PROT_WRITE);
202
203 gen = intel_gen(intel_get_drm_devid(fd));
204 has_64bit_reloc = gen >= 8;
205
206 src = gem_create(fd, ALIGN(object, 4096));
207 dst = gem_create(fd, ALIGN(object, 4096));
208
209 len = gem_linear_blt(fd, buf, 0, 0, 1, object, reloc);
210 if (has_64bit_reloc)
211 nreloc = len > 56 ? 4 : 2;
212 else
213 nreloc = len > 40 ? 4 : 2;
214
215 memset(exec, 0, sizeof(exec));
216 exec[0].handle = src;
217 exec[1].handle = dst;
218
219 exec[2].handle = handle;
220 exec[2].relocs_ptr = (uintptr_t)reloc;
221 exec[2].relocation_count = nreloc;
222
223 ring = 0;
224 if (gen >= 6)
225 ring = I915_EXEC_BLT;
226
227 memset(&execbuf, 0, sizeof(execbuf));
228 execbuf.buffers_ptr = (uintptr_t)exec;
229 execbuf.buffer_count = 3;
230 execbuf.batch_len = len;
231 execbuf.flags = ring;
232 execbuf.flags |= LOCAL_I915_EXEC_HANDLE_LUT;
233
234 if (__gem_execbuf(fd, &execbuf)) {
235 gem_set_domain(fd, handle, I915_GEM_DOMAIN_CPU, I915_GEM_DOMAIN_CPU);
236 len = gem_linear_blt(fd, buf, 0, src, dst, object, reloc);
237 igt_assert(len == execbuf.batch_len);
238 execbuf.flags = ring;
239 gem_execbuf(fd, &execbuf);
240 }
241 gem_sync(fd, handle);
242
243 if (batch > 1) {
244 if (execbuf.flags & LOCAL_I915_EXEC_HANDLE_LUT) {
245 src = 0;
246 dst = 1;
247 }
248
249 gem_set_domain(fd, handle, I915_GEM_DOMAIN_CPU, I915_GEM_DOMAIN_CPU);
250 for (int i = 1; i < batch; i++) {
251 len = gem_linear_blt(fd, buf, len - 8,
252 src, dst, object,
253 reloc + nreloc * i);
254 }
255 exec[2].relocation_count = nreloc * batch;
256 execbuf.batch_len = len;
257
258 gem_execbuf(fd, &execbuf);
259 gem_sync(fd, handle);
260 }
261 if (execbuf.flags & LOCAL_I915_EXEC_HANDLE_LUT)
262 execbuf.flags |= LOCAL_I915_EXEC_NO_RELOC;
263
264 /* Guess how many loops we need for 0.1s */
265 count = baseline((uint64_t)object * batch, 100) / ncpus;
266 if (flags & SYNC) {
267 time *= count / 2;
268 count = 1;
269 }
270 if (flags & NOCMD) {
271 drm_i915_getparam_t gp;
272 int v;
273
274 gp.param = I915_PARAM_CMD_PARSER_VERSION;
275 gp.value = &v;
276 drmIoctl(fd, DRM_IOCTL_I915_GETPARAM, &gp);
277 if (v < 1)
278 return 77;
279
280 execbuf.batch_len = 0;
281 }
282
283 while (reps--) {
284 memset(shared, 0, 4096);
285
286 igt_fork(child, ncpus) {
287 double min = HUGE_VAL;
288
289 for (int s = 0; s <= time / 100; s++) {
290 struct timespec start, end;
291 double t;
292
293 clock_gettime(CLOCK_MONOTONIC, &start);
294 for (int loop = 0; loop < count; loop++)
295 gem_execbuf(fd, &execbuf);
296 gem_sync(fd, handle);
297 clock_gettime(CLOCK_MONOTONIC, &end);
298
299 t = elapsed(&start, &end);
300 if (t < min)
301 min = t;
302 }
303
304 shared[child] = object/(1024*1024.)*batch*count/min;
305 }
306 igt_waitchildren();
307
308 for (int child = 0; child < ncpus; child++)
309 shared[ncpus] += shared[child];
310 printf("%7.3f\n", shared[ncpus] / ncpus);
311 }
312
313 close(fd);
314 return 0;
315 }
316
main(int argc,char ** argv)317 int main(int argc, char **argv)
318 {
319 int size = 1024*1024;
320 int reps = 13;
321 int time = 2000;
322 int ncpus = 1;
323 int batch = 1;
324 unsigned flags = 0;
325 int c;
326
327 while ((c = getopt (argc, argv, "CSs:b:r:t:f")) != -1) {
328 switch (c) {
329 case 's':
330 size = atoi(optarg);
331 size = ALIGN(size, 4);
332 if (size < 4)
333 size = 4;
334 break;
335
336 case 'S':
337 flags |= SYNC;
338 break;
339
340 case 'C':
341 flags |= NOCMD;
342 break;
343
344 case 't':
345 time = atoi(optarg);
346 if (time < 1)
347 time = 1;
348 break;
349
350 case 'r':
351 reps = atoi(optarg);
352 if (reps < 1)
353 reps = 1;
354 break;
355
356 case 'b':
357 batch = atoi(optarg);
358 if (batch < 1)
359 batch = 1;
360 break;
361
362 case 'f':
363 ncpus = sysconf(_SC_NPROCESSORS_ONLN);
364 break;
365
366 default:
367 break;
368 }
369 }
370
371 return run(size, batch, time, reps, ncpus, flags);
372 }
373