1 /*
2 * Copyright © 2012 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 /* Exercises the basic execbuffer using the handle LUT interface */
29
30 #include "igt.h"
31 #include <stdlib.h>
32 #include <stdio.h>
33 #include <string.h>
34 #include <fcntl.h>
35 #include <inttypes.h>
36 #include <errno.h>
37 #include <sys/stat.h>
38 #include <sys/time.h>
39 #include "drm.h"
40
41 IGT_TEST_DESCRIPTION("Exercises the basic execbuffer using the handle LUT"
42 " interface.");
43
44 #define BATCH_SIZE (1024*1024)
45
46 #define LOCAL_I915_EXEC_NO_RELOC (1<<11)
47 #define LOCAL_I915_EXEC_HANDLE_LUT (1<<12)
48
49 #define MAX_NUM_EXEC 2048
50 #define MAX_NUM_RELOC 4096
51
52 #define SKIP_RELOC 0x1
53 #define NO_RELOC 0x2
54 #define CYCLE_BATCH 0x4
55 #define FAULT 0x8
56
57 int target[MAX_NUM_RELOC];
58 struct drm_i915_gem_exec_object2 gem_exec[MAX_NUM_EXEC+1];
59 struct drm_i915_gem_relocation_entry mem_reloc[MAX_NUM_RELOC];
60
61 static uint32_t state = 0x12345678;
62
63 static uint32_t
hars_petruska_f54_1_random(void)64 hars_petruska_f54_1_random (void)
65 {
66 #define rol(x,k) ((x << k) | (x >> (32-k)))
67 return state = (state ^ rol (state, 5) ^ rol (state, 24)) + 0x37798849;
68 #undef rol
69 }
70
has_exec_lut(int fd)71 static int has_exec_lut(int fd)
72 {
73 struct drm_i915_gem_execbuffer2 execbuf;
74
75 memset(&execbuf, 0, sizeof(execbuf));
76 execbuf.buffers_ptr = to_user_pointer((gem_exec + MAX_NUM_EXEC));
77 execbuf.buffer_count = 1;
78 execbuf.flags = LOCAL_I915_EXEC_HANDLE_LUT;
79
80 return __gem_execbuf(fd, &execbuf) == 0;
81 }
82
83 #define ELAPSED(a,b) (1e6*((b)->tv_sec - (a)->tv_sec) + ((b)->tv_usec - (a)->tv_usec))
84 igt_simple_main
85 {
86 uint32_t batch[2] = {MI_BATCH_BUFFER_END};
87 uint32_t cycle[16];
88 int fd, n, m, count, c;
89 const struct {
90 const char *name;
91 unsigned int flags;
92 } pass[] = {
93 { .name = "relocation", .flags = 0 },
94 { .name = "cycle-relocation", .flags = CYCLE_BATCH },
95 { .name = "fault-relocation", .flags = FAULT },
96 { .name = "skip-relocs", .flags = SKIP_RELOC },
97 { .name = "no-relocs", .flags = SKIP_RELOC | NO_RELOC },
98 { .name = NULL },
99 }, *p;
100 struct drm_i915_gem_relocation_entry *reloc;
101 uint32_t reloc_handle;
102 int size;
103
104 igt_skip_on_simulation();
105
106 fd = drm_open_driver(DRIVER_INTEL);
107 igt_require_gem(fd);
108
109 memset(gem_exec, 0, sizeof(gem_exec));
110 for (n = 0; n < MAX_NUM_EXEC; n++)
111 gem_exec[n].handle = gem_create(fd, 4096);
112
113 for (n = 0; n < 16; n++) {
114 cycle[n] = gem_create(fd, 4096);
115 gem_write(fd, cycle[n], 0, batch, sizeof(batch));
116 }
117 gem_exec[MAX_NUM_EXEC].handle = cycle[0];
118
119 memset(mem_reloc, 0, sizeof(mem_reloc));
120 for (n = 0; n < MAX_NUM_RELOC; n++) {
121 mem_reloc[n].offset = 1024;
122 mem_reloc[n].read_domains = I915_GEM_DOMAIN_RENDER;
123 }
124
125 size = ALIGN(sizeof(mem_reloc), 4096);
126 reloc_handle = gem_create(fd, size);
127 reloc = gem_mmap__cpu(fd, reloc_handle, 0, size, PROT_READ | PROT_WRITE);
128 for (n = 0; n < MAX_NUM_RELOC; n++) {
129 reloc[n].offset = 1024;
130 reloc[n].read_domains = I915_GEM_DOMAIN_RENDER;
131 }
132 munmap(reloc, size);
133
134 igt_require(has_exec_lut(fd));
135
136 for (p = pass; p->name != NULL; p++) {
137 if (p->flags & FAULT)
138 igt_disable_prefault();
139 for (n = 1; n <= MAX_NUM_EXEC; n *= 2) {
140 double elapsed[16][2];
141 double s_x, s_y, s_xx, s_xy;
142 double A, B;
143 int i, j;
144
145 for (i = 0, m = 1; m <= MAX_NUM_RELOC; m *= 2, i++) {
146 struct drm_i915_gem_execbuffer2 execbuf;
147 struct drm_i915_gem_exec_object2 *objects;
148 struct timeval start, end;
149
150 if (p->flags & FAULT)
151 reloc = __gem_mmap__cpu(fd, reloc_handle, 0, size, PROT_READ | PROT_WRITE);
152 else
153 reloc = mem_reloc;
154
155 gem_exec[MAX_NUM_EXEC].relocation_count = m;
156 gem_exec[MAX_NUM_EXEC].relocs_ptr = to_user_pointer(reloc);
157 objects = gem_exec + MAX_NUM_EXEC - n;
158
159 memset(&execbuf, 0, sizeof(execbuf));
160 execbuf.buffers_ptr = to_user_pointer(objects);
161 execbuf.buffer_count = n + 1;
162 execbuf.flags = LOCAL_I915_EXEC_HANDLE_LUT;
163 if (p->flags & NO_RELOC)
164 execbuf.flags |= LOCAL_I915_EXEC_NO_RELOC;
165
166 for (j = 0; j < m; j++) {
167 target[j] = hars_petruska_f54_1_random() % n;
168 reloc[j].target_handle = target[j];
169 reloc[j].presumed_offset = -1;
170 }
171
172 gem_execbuf(fd,&execbuf);
173 gettimeofday(&start, NULL);
174 for (count = 0; count < 1000; count++) {
175 if ((p->flags & SKIP_RELOC) == 0) {
176 for (j = 0; j < m; j++)
177 reloc[j].presumed_offset = -1;
178 if (p->flags & CYCLE_BATCH) {
179 c = (c + 1) % 16;
180 gem_exec[MAX_NUM_EXEC].handle = cycle[c];
181 }
182 }
183 if (p->flags & FAULT) {
184 munmap(reloc, size);
185 reloc = __gem_mmap__cpu(fd, reloc_handle, 0, size, PROT_READ | PROT_WRITE);
186 gem_exec[MAX_NUM_EXEC].relocs_ptr = to_user_pointer(reloc);
187 }
188 gem_execbuf(fd, &execbuf);
189 }
190 gettimeofday(&end, NULL);
191 c = 16;
192 do
193 gem_sync(fd, cycle[--c]);
194 while (c != 0);
195 gem_exec[MAX_NUM_EXEC].handle = cycle[c];
196 elapsed[i][1] = ELAPSED(&start, &end);
197
198 execbuf.flags &= ~LOCAL_I915_EXEC_HANDLE_LUT;
199 for (j = 0; j < m; j++)
200 reloc[j].target_handle = objects[target[j]].handle;
201
202 gem_execbuf(fd,&execbuf);
203 gettimeofday(&start, NULL);
204 for (count = 0; count < 1000; count++) {
205 if ((p->flags & SKIP_RELOC) == 0) {
206 for (j = 0; j < m; j++)
207 reloc[j].presumed_offset = -1;
208 if (p->flags & CYCLE_BATCH) {
209 c = (c + 1) % 16;
210 gem_exec[MAX_NUM_EXEC].handle = cycle[c];
211 }
212 }
213 if (p->flags & FAULT) {
214 munmap(reloc, size);
215 reloc = __gem_mmap__cpu(fd, reloc_handle, 0, size, PROT_READ | PROT_WRITE);
216 gem_exec[MAX_NUM_EXEC].relocs_ptr = to_user_pointer(reloc);
217 }
218 gem_execbuf(fd, &execbuf);
219 }
220 gettimeofday(&end, NULL);
221 c = 16;
222 do
223 gem_sync(fd, cycle[--c]);
224 while (c != 0);
225 gem_exec[MAX_NUM_EXEC].handle = cycle[c];
226 elapsed[i][0] = ELAPSED(&start, &end);
227
228 if (p->flags & FAULT)
229 munmap(reloc, size);
230 }
231
232 igt_info("%s: buffers=%4d:", p->name, n);
233
234 s_x = s_y = s_xx = s_xy = 0;
235 for (j = 0; j < i; j++) {
236 int k = 1 << j;
237 s_x += k;
238 s_y += elapsed[j][0];
239 s_xx += k * k;
240 s_xy += k * elapsed[j][0];
241 }
242 B = (s_xy - s_x * s_y / j) / (s_xx - s_x * s_x / j);
243 A = s_y / j - B * s_x / j;
244 igt_info(" old=%7.0f + %.1f*reloc,", A, B);
245
246 s_x = s_y = s_xx = s_xy = 0;
247 for (j = 0; j < i; j++) {
248 int k = 1 << j;
249 s_x += k;
250 s_y += elapsed[j][1];
251 s_xx += k * k;
252 s_xy += k * elapsed[j][1];
253 }
254 B = (s_xy - s_x * s_y / j) / (s_xx - s_x * s_x / j);
255 A = s_y / j - B * s_x / j;
256 igt_info(" lut=%7.0f + %.1f*reloc (ns)", A, B);
257
258 igt_info("\n");
259 }
260 if (p->flags & FAULT)
261 igt_enable_prefault();
262 }
263 }
264