bzr branch
http://suren.me/webbzr/ani/mrses
1
by Suren A. Chilingaryan
Initial import |
1 |
#define _GNU_SOURCE
|
2 |
||
3 |
#include <stdio.h> |
|
4 |
#include <stdlib.h> |
|
5 |
#include <string.h> |
|
6 |
#include <sys/types.h> |
|
7 |
#include <unistd.h> |
|
8 |
#include <sched.h> |
|
9 |
#include <pthread.h> |
|
10 |
#include <errno.h> |
|
11 |
||
12 |
#include "msg.h" |
|
13 |
#include "hw_sched.h" |
|
14 |
||
15 |
#include "mrses_ppu.h" |
|
16 |
||
17 |
#ifdef HW_HAVE_SPU
|
|
18 |
# include <libspe2.h>
|
|
19 |
# include "mrses_spu.h"
|
|
20 |
||
21 |
extern spe_program_handle_t mrses_spe; |
|
22 |
#endif /* HW_HAVE_SPU */ |
|
23 |
||
24 |
||
25 |
#define MUTEX_INIT(ctx, name) \
|
|
26 |
if (!err) err = pthread_mutex_init(&ctx->name##_mutex, NULL);
|
|
27 |
||
28 |
#define MUTEX_FREE(ctx, name) \
|
|
29 |
pthread_mutex_destroy(&ctx->name##_mutex);
|
|
30 |
||
31 |
#define COND_INIT(ctx, name) \
|
|
32 |
MUTEX_INIT(ctx, name##_cond) \
|
|
33 |
if (!err) { \
|
|
34 |
err = pthread_cond_init(&ctx->name##_cond, NULL); \
|
|
35 |
if (err) { MUTEX_FREE(ctx, name##_cond) } \
|
|
36 |
}
|
|
37 |
||
38 |
#define COND_FREE(ctx, name) \
|
|
39 |
pthread_cond_destroy(&ctx->name##_cond); \
|
|
40 |
MUTEX_FREE(ctx, name##_cond)
|
|
41 |
||
42 |
HWRunFunction ppu_run[] = { |
|
43 |
(HWRunFunction)mrses_ppu_run, |
|
44 |
(HWRunFunction)mrses_ppu_iterate, |
|
45 |
NULL
|
|
46 |
};
|
|
47 |
||
48 |
#ifdef HW_HAVE_SPU
|
|
49 |
HWRunFunction spu_run[] = { |
|
50 |
(HWRunFunction)mrses_spu_run, |
|
51 |
(HWRunFunction)mrses_spu_iterate, |
|
52 |
NULL
|
|
53 |
};
|
|
54 |
#endif /* HW_HAVE_SPU */ |
|
55 |
||
56 |
HWSched hw_sched_create() { |
|
57 |
int i; |
|
58 |
int err = 0; |
|
59 |
cpu_set_t mask; |
|
60 |
||
61 |
int ppu_count; |
|
62 |
HWSched ctx; |
|
63 |
||
64 |
#ifdef HW_HAVE_SPU
|
|
65 |
int spu_count; |
|
66 |
spe_context_ptr_t spe; |
|
67 |
#endif /* HW_HAVE_SPU */ |
|
68 |
||
69 |
ctx = (HWSched)malloc(sizeof(HWSchedS)); |
|
70 |
if (!ctx) return NULL; |
|
71 |
||
72 |
memset(ctx, 0, sizeof(HWSchedS)); |
|
73 |
||
74 |
ctx->status = 1; |
|
75 |
||
76 |
MUTEX_INIT(ctx, data); |
|
77 |
||
78 |
COND_INIT(ctx, compl); |
|
79 |
if (err) { |
|
80 |
MUTEX_FREE(ctx, data); |
|
81 |
}
|
|
82 |
||
83 |
COND_INIT(ctx, job); |
|
84 |
if (err) { |
|
85 |
COND_FREE(ctx, compl); |
|
86 |
MUTEX_FREE(ctx, data); |
|
87 |
} else { |
|
88 |
ctx->sync_init = 1; |
|
89 |
}
|
|
90 |
||
91 |
if (err) { |
|
92 |
fprintf(stderr, "Error initializing conditions and mutexes, errnon: %i\n", errno); |
|
93 |
hw_sched_destroy(ctx); |
|
94 |
return NULL; |
|
95 |
}
|
|
96 |
||
97 |
||
98 |
err = sched_getaffinity(getpid(), sizeof(mask), &mask); |
|
99 |
||
100 |
#ifdef CPU_COUNT
|
|
101 |
ppu_count = CPU_COUNT(&mask); |
|
102 |
#else
|
|
103 |
for (ppu_count = 0; ppu_count < CPU_SETSIZE; ppu_count++) { |
|
104 |
if (!CPU_ISSET(ppu_count, &mask)) break; |
|
105 |
}
|
|
106 |
#endif
|
|
107 |
||
108 |
#ifdef HW_MAX_PPU
|
|
109 |
if (ppu_count > HW_MAX_PPU) ppu_count = HW_MAX_PPU; |
|
110 |
#endif /* HW_MAX_PPU */ |
|
111 |
||
112 |
ctx->n_threads = 0; |
|
113 |
for (i = 0; i < ppu_count; i++) { |
|
114 |
ctx->thread[ctx->n_threads] = hw_thread_create(ctx, ctx->n_threads, NULL, ppu_run, NULL); |
|
115 |
if (ctx->thread[ctx->n_threads]) ++ctx->n_threads; |
|
116 |
}
|
|
117 |
||
118 |
#ifdef HW_HAVE_SPU
|
|
119 |
spu_count = spe_cpu_info_get(SPE_COUNT_USABLE_SPES, -1); |
|
120 |
if ((spu_count + ctx->n_threads) > HW_MAX_THREADS) spu_count = HW_MAX_THREADS - ctx->n_threads; |
|
121 |
||
122 |
#ifdef HW_MAX_SPU
|
|
123 |
if (spu_count > HW_MAX_SPU) spu_count = HW_MAX_SPU; |
|
124 |
#endif /* HW_MAX_SPU */ |
|
125 |
||
126 |
for (i = 0; i < spu_count; i++) { |
|
127 |
spe = spe_context_create (0, NULL); |
|
128 |
if (spe == NULL) { |
|
129 |
reportError("Failed to create SPE context"); |
|
130 |
hw_sched_destroy(ctx); |
|
131 |
return NULL; |
|
132 |
}
|
|
133 |
||
134 |
err = spe_program_load (spe, &mrses_spe); |
|
135 |
if (err) { |
|
136 |
reportError("Failed to load program into the SPE, error: %i", err); |
|
137 |
return NULL; |
|
138 |
}
|
|
139 |
||
140 |
ctx->thread[ctx->n_threads] = hw_thread_create(ctx, ctx->n_threads, spe, spu_run, (HWFreeFunction)spe_context_destroy); |
|
141 |
if (ctx->thread[ctx->n_threads]) ++ctx->n_threads; |
|
142 |
}
|
|
143 |
||
144 |
reportMessage("ppu: %i, spu: %i", ppu_count, spu_count); |
|
145 |
#else /* HW_HAVE_SPU */ |
|
146 |
reportMessage("threads: %i", ppu_count); |
|
147 |
#endif /* HW_HAVE_SPU */ |
|
148 |
||
149 |
||
150 |
return ctx; |
|
151 |
}
|
|
152 |
||
153 |
static int hw_sched_wait_threads(HWSched ctx) { |
|
154 |
int i = 0; |
|
155 |
||
156 |
hw_sched_lock(ctx, compl_cond); |
|
157 |
while (i < ctx->n_threads) { |
|
158 |
for (; i < ctx->n_threads; i++) { |
|
159 |
if (ctx->thread[i]->status == HW_THREAD_STATUS_INIT) { |
|
160 |
hw_sched_wait(ctx, compl); |
|
161 |
break; |
|
162 |
}
|
|
163 |
}
|
|
164 |
||
165 |
}
|
|
166 |
hw_sched_unlock(ctx, compl_cond); |
|
167 |
||
168 |
ctx->started = 1; |
|
169 |
||
170 |
return 0; |
|
171 |
}
|
|
172 |
||
173 |
void hw_sched_destroy(HWSched ctx) { |
|
174 |
int i; |
|
175 |
||
176 |
if (ctx->n_threads > 0) { |
|
177 |
if (!ctx->started) { |
|
178 |
hw_sched_wait_threads(ctx); |
|
179 |
}
|
|
180 |
||
181 |
ctx->status = 0; |
|
182 |
hw_sched_lock(ctx, job_cond); |
|
183 |
hw_sched_broadcast(ctx, job); |
|
184 |
hw_sched_unlock(ctx, job_cond); |
|
185 |
||
186 |
for (i = 0; i < ctx->n_threads; i++) { |
|
187 |
hw_thread_destroy(ctx->thread[i]); |
|
188 |
}
|
|
189 |
}
|
|
190 |
||
191 |
if (ctx->sync_init) { |
|
192 |
COND_FREE(ctx, job); |
|
193 |
COND_FREE(ctx, compl); |
|
194 |
MUTEX_FREE(ctx, data); |
|
195 |
}
|
|
196 |
||
197 |
free(ctx); |
|
198 |
}
|
|
199 |
||
200 |
int hw_sched_set_sequential_mode(HWSched ctx, int *n_blocks, int *cur_block) { |
|
201 |
ctx->mode = HW_SCHED_MODE_SEQUENTIAL; |
|
202 |
ctx->n_blocks = n_blocks; |
|
203 |
ctx->cur_block = cur_block; |
|
204 |
||
205 |
return 0; |
|
206 |
}
|
|
207 |
||
208 |
int hw_sched_get_chunk(HWSched ctx, int thread_id) { |
|
209 |
int block; |
|
210 |
||
211 |
switch (ctx->mode) { |
|
212 |
case HW_SCHED_MODE_PREALLOCATED: |
|
213 |
if (ctx->thread[thread_id]->status == HW_THREAD_STATUS_IDLE) { |
|
214 |
return thread_id; |
|
215 |
} else { |
|
216 |
return -1; |
|
217 |
}
|
|
218 |
case HW_SCHED_MODE_SEQUENTIAL: |
|
219 |
hw_sched_lock(ctx, data); |
|
220 |
block = *ctx->cur_block; |
|
221 |
if (block < *ctx->n_blocks) { |
|
222 |
*ctx->cur_block = *ctx->cur_block + 1; |
|
223 |
} else { |
|
224 |
block = -1; |
|
225 |
}
|
|
226 |
hw_sched_unlock(ctx, data); |
|
227 |
return block; |
|
228 |
default: |
|
229 |
return -1; |
|
230 |
}
|
|
231 |
||
232 |
return -1; |
|
233 |
}
|
|
234 |
||
235 |
||
236 |
int hw_sched_schedule_task(HWSched ctx, void *appctx, int entry) { |
|
237 |
if (!ctx->started) { |
|
238 |
hw_sched_wait_threads(ctx); |
|
239 |
}
|
|
240 |
||
241 |
ctx->ctx = appctx; |
|
242 |
ctx->entry = entry; |
|
243 |
||
244 |
hw_sched_lock(ctx, compl_cond); |
|
245 |
||
246 |
hw_sched_lock(ctx, job_cond); |
|
247 |
hw_sched_broadcast(ctx, job); |
|
248 |
hw_sched_unlock(ctx, job_cond); |
|
249 |
||
250 |
return 0; |
|
251 |
}
|
|
252 |
||
253 |
int hw_sched_wait_task(HWSched ctx) { |
|
254 |
int i = 0; |
|
255 |
||
256 |
while (i < ctx->n_threads) { |
|
257 |
for (; i < ctx->n_threads; i++) { |
|
258 |
if (ctx->thread[i]->status == HW_THREAD_STATUS_DONE) { |
|
259 |
ctx->thread[i]->status = HW_THREAD_STATUS_IDLE; |
|
260 |
} else { |
|
261 |
hw_sched_wait(ctx, compl); |
|
262 |
break; |
|
263 |
}
|
|
264 |
}
|
|
265 |
||
266 |
}
|
|
267 |
||
268 |
/*
|
|
269 |
int i, running = 1;
|
|
270 |
||
271 |
//wait all threads set running mode
|
|
272 |
||
273 |
while (running) {
|
|
274 |
hw_sched_wait(ctx, compl);
|
|
275 |
|
|
276 |
running = 0;
|
|
277 |
for (i = 0; i < ctx->n_threads; i++) {
|
|
278 |
if (ctx->thread[i]->status) {
|
|
279 |
if (ctx->thread[i]->status == HW_THREAD_STATUS_DONE) {
|
|
280 |
ctx->thread[i]->status = HW_THREAD_STATUS_IDLE;
|
|
281 |
} else {
|
|
282 |
running = 1;
|
|
283 |
break;
|
|
284 |
}
|
|
285 |
}
|
|
286 |
}
|
|
287 |
}
|
|
288 |
*/
|
|
289 |
||
290 |
hw_sched_unlock(ctx, compl_cond); |
|
291 |
||
292 |
||
293 |
return 0; |
|
294 |
}
|