util: Add cpuid for Solaris Studio.
[mesa.git] / src / gallium / drivers / nvc0 / nvc0_query.c
1 /*
2 * Copyright 2011 Nouveau Project
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 AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
18 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
19 * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
20 * SOFTWARE.
21 *
22 * Authors: Christoph Bumiller
23 */
24
25 #define NVC0_PUSH_EXPLICIT_SPACE_CHECKING
26
27 #include "nvc0_context.h"
28 #include "nouveau/nv_object.xml.h"
29
30 #define NVC0_QUERY_STATE_READY 0
31 #define NVC0_QUERY_STATE_ACTIVE 1
32 #define NVC0_QUERY_STATE_ENDED 2
33 #define NVC0_QUERY_STATE_FLUSHED 3
34
35 struct nvc0_query {
36 uint32_t *data;
37 uint16_t type;
38 uint16_t index;
39 uint32_t sequence;
40 struct nouveau_bo *bo;
41 uint32_t base;
42 uint32_t offset; /* base + i * rotate */
43 uint8_t state;
44 boolean is64bit;
45 uint8_t rotate;
46 int nesting; /* only used for occlusion queries */
47 struct nouveau_mm_allocation *mm;
48 };
49
50 #define NVC0_QUERY_ALLOC_SPACE 256
51
52 static INLINE struct nvc0_query *
53 nvc0_query(struct pipe_query *pipe)
54 {
55 return (struct nvc0_query *)pipe;
56 }
57
58 static boolean
59 nvc0_query_allocate(struct nvc0_context *nvc0, struct nvc0_query *q, int size)
60 {
61 struct nvc0_screen *screen = nvc0->screen;
62 int ret;
63
64 if (q->bo) {
65 nouveau_bo_ref(NULL, &q->bo);
66 if (q->mm) {
67 if (q->state == NVC0_QUERY_STATE_READY)
68 nouveau_mm_free(q->mm);
69 else
70 nouveau_fence_work(screen->base.fence.current,
71 nouveau_mm_free_work, q->mm);
72 }
73 }
74 if (size) {
75 q->mm = nouveau_mm_allocate(screen->base.mm_GART, size, &q->bo, &q->base);
76 if (!q->bo)
77 return FALSE;
78 q->offset = q->base;
79
80 ret = nouveau_bo_map(q->bo, 0, screen->base.client);
81 if (ret) {
82 nvc0_query_allocate(nvc0, q, 0);
83 return FALSE;
84 }
85 q->data = (uint32_t *)((uint8_t *)q->bo->map + q->base);
86 }
87 return TRUE;
88 }
89
90 static void
91 nvc0_query_destroy(struct pipe_context *pipe, struct pipe_query *pq)
92 {
93 nvc0_query_allocate(nvc0_context(pipe), nvc0_query(pq), 0);
94 FREE(nvc0_query(pq));
95 }
96
97 static struct pipe_query *
98 nvc0_query_create(struct pipe_context *pipe, unsigned type)
99 {
100 struct nvc0_context *nvc0 = nvc0_context(pipe);
101 struct nvc0_query *q;
102 unsigned space = NVC0_QUERY_ALLOC_SPACE;
103
104 q = CALLOC_STRUCT(nvc0_query);
105 if (!q)
106 return NULL;
107
108 switch (type) {
109 case PIPE_QUERY_OCCLUSION_COUNTER:
110 case PIPE_QUERY_OCCLUSION_PREDICATE:
111 q->rotate = 32;
112 space = NVC0_QUERY_ALLOC_SPACE;
113 break;
114 case PIPE_QUERY_PIPELINE_STATISTICS:
115 q->is64bit = TRUE;
116 space = 512;
117 break;
118 case PIPE_QUERY_SO_STATISTICS:
119 case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
120 q->is64bit = TRUE;
121 space = 64;
122 break;
123 case PIPE_QUERY_TIME_ELAPSED:
124 case PIPE_QUERY_TIMESTAMP:
125 case PIPE_QUERY_TIMESTAMP_DISJOINT:
126 case PIPE_QUERY_GPU_FINISHED:
127 case PIPE_QUERY_PRIMITIVES_GENERATED:
128 case PIPE_QUERY_PRIMITIVES_EMITTED:
129 space = 32;
130 break;
131 case NVC0_QUERY_TFB_BUFFER_OFFSET:
132 space = 16;
133 break;
134 default:
135 FREE(q);
136 return NULL;
137 }
138 if (!nvc0_query_allocate(nvc0, q, space)) {
139 FREE(q);
140 return NULL;
141 }
142
143 q->type = type;
144
145 if (q->rotate) {
146 /* we advance before query_begin ! */
147 q->offset -= q->rotate;
148 q->data -= q->rotate / sizeof(*q->data);
149 } else
150 if (!q->is64bit)
151 q->data[0] = 0; /* initialize sequence */
152
153 return (struct pipe_query *)q;
154 }
155
156 static void
157 nvc0_query_get(struct nouveau_pushbuf *push, struct nvc0_query *q,
158 unsigned offset, uint32_t get)
159 {
160 offset += q->offset;
161
162 PUSH_SPACE(push, 5);
163 PUSH_REFN (push, q->bo, NOUVEAU_BO_GART | NOUVEAU_BO_WR);
164 BEGIN_NVC0(push, NVC0_3D(QUERY_ADDRESS_HIGH), 4);
165 PUSH_DATAh(push, q->bo->offset + offset);
166 PUSH_DATA (push, q->bo->offset + offset);
167 PUSH_DATA (push, q->sequence);
168 PUSH_DATA (push, get);
169 }
170
171 static void
172 nvc0_query_rotate(struct nvc0_context *nvc0, struct nvc0_query *q)
173 {
174 q->offset += q->rotate;
175 q->data += q->rotate / sizeof(*q->data);
176 if (q->offset - q->base == NVC0_QUERY_ALLOC_SPACE)
177 nvc0_query_allocate(nvc0, q, NVC0_QUERY_ALLOC_SPACE);
178 }
179
180 static void
181 nvc0_query_begin(struct pipe_context *pipe, struct pipe_query *pq)
182 {
183 struct nvc0_context *nvc0 = nvc0_context(pipe);
184 struct nouveau_pushbuf *push = nvc0->base.pushbuf;
185 struct nvc0_query *q = nvc0_query(pq);
186
187 /* For occlusion queries we have to change the storage, because a previous
188 * query might set the initial render conition to FALSE even *after* we re-
189 * initialized it to TRUE.
190 */
191 if (q->rotate) {
192 nvc0_query_rotate(nvc0, q);
193
194 /* XXX: can we do this with the GPU, and sync with respect to a previous
195 * query ?
196 */
197 q->data[0] = q->sequence; /* initialize sequence */
198 q->data[1] = 1; /* initial render condition = TRUE */
199 q->data[4] = q->sequence + 1; /* for comparison COND_MODE */
200 q->data[5] = 0;
201 }
202 q->sequence++;
203
204 switch (q->type) {
205 case PIPE_QUERY_OCCLUSION_COUNTER:
206 case PIPE_QUERY_OCCLUSION_PREDICATE:
207 q->nesting = nvc0->screen->num_occlusion_queries_active++;
208 if (q->nesting) {
209 nvc0_query_get(push, q, 0x10, 0x0100f002);
210 } else {
211 PUSH_SPACE(push, 3);
212 BEGIN_NVC0(push, NVC0_3D(COUNTER_RESET), 1);
213 PUSH_DATA (push, NVC0_3D_COUNTER_RESET_SAMPLECNT);
214 IMMED_NVC0(push, NVC0_3D(SAMPLECNT_ENABLE), 1);
215 }
216 break;
217 case PIPE_QUERY_PRIMITIVES_GENERATED:
218 nvc0_query_get(push, q, 0x10, 0x06805002 | (q->index << 5));
219 break;
220 case PIPE_QUERY_PRIMITIVES_EMITTED:
221 nvc0_query_get(push, q, 0x10, 0x05805002 | (q->index << 5));
222 break;
223 case PIPE_QUERY_SO_STATISTICS:
224 nvc0_query_get(push, q, 0x20, 0x05805002 | (q->index << 5));
225 nvc0_query_get(push, q, 0x30, 0x06805002 | (q->index << 5));
226 break;
227 case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
228 nvc0_query_get(push, q, 0x10, 0x03005002 | (q->index << 5));
229 break;
230 case PIPE_QUERY_TIMESTAMP_DISJOINT:
231 case PIPE_QUERY_TIME_ELAPSED:
232 nvc0_query_get(push, q, 0x10, 0x00005002);
233 break;
234 case PIPE_QUERY_PIPELINE_STATISTICS:
235 nvc0_query_get(push, q, 0xc0 + 0x00, 0x00801002); /* VFETCH, VERTICES */
236 nvc0_query_get(push, q, 0xc0 + 0x10, 0x01801002); /* VFETCH, PRIMS */
237 nvc0_query_get(push, q, 0xc0 + 0x20, 0x02802002); /* VP, LAUNCHES */
238 nvc0_query_get(push, q, 0xc0 + 0x30, 0x03806002); /* GP, LAUNCHES */
239 nvc0_query_get(push, q, 0xc0 + 0x40, 0x04806002); /* GP, PRIMS_OUT */
240 nvc0_query_get(push, q, 0xc0 + 0x50, 0x07804002); /* RAST, PRIMS_IN */
241 nvc0_query_get(push, q, 0xc0 + 0x60, 0x08804002); /* RAST, PRIMS_OUT */
242 nvc0_query_get(push, q, 0xc0 + 0x70, 0x0980a002); /* ROP, PIXELS */
243 nvc0_query_get(push, q, 0xc0 + 0x80, 0x0d808002); /* TCP, LAUNCHES */
244 nvc0_query_get(push, q, 0xc0 + 0x90, 0x0e809002); /* TEP, LAUNCHES */
245 break;
246 default:
247 break;
248 }
249 q->state = NVC0_QUERY_STATE_ACTIVE;
250 }
251
252 static void
253 nvc0_query_end(struct pipe_context *pipe, struct pipe_query *pq)
254 {
255 struct nvc0_context *nvc0 = nvc0_context(pipe);
256 struct nouveau_pushbuf *push = nvc0->base.pushbuf;
257 struct nvc0_query *q = nvc0_query(pq);
258
259 if (q->state != NVC0_QUERY_STATE_ACTIVE) {
260 /* some queries don't require 'begin' to be called (e.g. GPU_FINISHED) */
261 if (q->rotate)
262 nvc0_query_rotate(nvc0, q);
263 q->sequence++;
264 }
265 q->state = NVC0_QUERY_STATE_ENDED;
266
267 switch (q->type) {
268 case PIPE_QUERY_OCCLUSION_COUNTER:
269 case PIPE_QUERY_OCCLUSION_PREDICATE:
270 nvc0_query_get(push, q, 0, 0x0100f002);
271 if (--nvc0->screen->num_occlusion_queries_active == 0) {
272 PUSH_SPACE(push, 1);
273 IMMED_NVC0(push, NVC0_3D(SAMPLECNT_ENABLE), 0);
274 }
275 break;
276 case PIPE_QUERY_PRIMITIVES_GENERATED:
277 nvc0_query_get(push, q, 0, 0x06805002 | (q->index << 5));
278 break;
279 case PIPE_QUERY_PRIMITIVES_EMITTED:
280 nvc0_query_get(push, q, 0, 0x05805002 | (q->index << 5));
281 break;
282 case PIPE_QUERY_SO_STATISTICS:
283 nvc0_query_get(push, q, 0x00, 0x05805002 | (q->index << 5));
284 nvc0_query_get(push, q, 0x10, 0x06805002 | (q->index << 5));
285 break;
286 case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
287 /* TODO: How do we sum over all streams for render condition ? */
288 /* PRIMS_DROPPED doesn't write sequence, use a ZERO query to sync on */
289 nvc0_query_get(push, q, 0x00, 0x03005002 | (q->index << 5));
290 nvc0_query_get(push, q, 0x20, 0x00005002);
291 break;
292 case PIPE_QUERY_TIMESTAMP:
293 case PIPE_QUERY_TIMESTAMP_DISJOINT:
294 case PIPE_QUERY_TIME_ELAPSED:
295 nvc0_query_get(push, q, 0, 0x00005002);
296 break;
297 case PIPE_QUERY_GPU_FINISHED:
298 nvc0_query_get(push, q, 0, 0x1000f010);
299 break;
300 case PIPE_QUERY_PIPELINE_STATISTICS:
301 nvc0_query_get(push, q, 0x00, 0x00801002); /* VFETCH, VERTICES */
302 nvc0_query_get(push, q, 0x10, 0x01801002); /* VFETCH, PRIMS */
303 nvc0_query_get(push, q, 0x20, 0x02802002); /* VP, LAUNCHES */
304 nvc0_query_get(push, q, 0x30, 0x03806002); /* GP, LAUNCHES */
305 nvc0_query_get(push, q, 0x40, 0x04806002); /* GP, PRIMS_OUT */
306 nvc0_query_get(push, q, 0x50, 0x07804002); /* RAST, PRIMS_IN */
307 nvc0_query_get(push, q, 0x60, 0x08804002); /* RAST, PRIMS_OUT */
308 nvc0_query_get(push, q, 0x70, 0x0980a002); /* ROP, PIXELS */
309 nvc0_query_get(push, q, 0x80, 0x0d808002); /* TCP, LAUNCHES */
310 nvc0_query_get(push, q, 0x90, 0x0e809002); /* TEP, LAUNCHES */
311 break;
312 case NVC0_QUERY_TFB_BUFFER_OFFSET:
313 /* indexed by TFB buffer instead of by vertex stream */
314 nvc0_query_get(push, q, 0x00, 0x0d005002 | (q->index << 5));
315 break;
316 default:
317 assert(0);
318 break;
319 }
320 }
321
322 static INLINE void
323 nvc0_query_update(struct nouveau_client *cli, struct nvc0_query *q)
324 {
325 if (q->is64bit) {
326 if (!nouveau_bo_map(q->bo, NOUVEAU_BO_RD | NOUVEAU_BO_NOBLOCK, cli))
327 q->state = NVC0_QUERY_STATE_READY;
328 } else {
329 if (q->data[0] == q->sequence)
330 q->state = NVC0_QUERY_STATE_READY;
331 }
332 }
333
334 static boolean
335 nvc0_query_result(struct pipe_context *pipe, struct pipe_query *pq,
336 boolean wait, union pipe_query_result *result)
337 {
338 struct nvc0_context *nvc0 = nvc0_context(pipe);
339 struct nvc0_query *q = nvc0_query(pq);
340 uint64_t *res64 = (uint64_t*)result;
341 uint32_t *res32 = (uint32_t*)result;
342 boolean *res8 = (boolean*)result;
343 uint64_t *data64 = (uint64_t *)q->data;
344 unsigned i;
345
346 if (q->state != NVC0_QUERY_STATE_READY)
347 nvc0_query_update(nvc0->screen->base.client, q);
348
349 if (q->state != NVC0_QUERY_STATE_READY) {
350 if (!wait) {
351 if (q->state != NVC0_QUERY_STATE_FLUSHED) {
352 q->state = NVC0_QUERY_STATE_FLUSHED;
353 /* flush for silly apps that spin on GL_QUERY_RESULT_AVAILABLE */
354 PUSH_KICK(nvc0->base.pushbuf);
355 }
356 return FALSE;
357 }
358 if (nouveau_bo_wait(q->bo, NOUVEAU_BO_RD, nvc0->screen->base.client))
359 return FALSE;
360 }
361 q->state = NVC0_QUERY_STATE_READY;
362
363 switch (q->type) {
364 case PIPE_QUERY_GPU_FINISHED:
365 res8[0] = TRUE;
366 break;
367 case PIPE_QUERY_OCCLUSION_COUNTER: /* u32 sequence, u32 count, u64 time */
368 res64[0] = q->data[1] - q->data[5];
369 break;
370 case PIPE_QUERY_OCCLUSION_PREDICATE:
371 res8[0] = q->data[1] != q->data[5];
372 break;
373 case PIPE_QUERY_PRIMITIVES_GENERATED: /* u64 count, u64 time */
374 case PIPE_QUERY_PRIMITIVES_EMITTED: /* u64 count, u64 time */
375 res64[0] = data64[0] - data64[2];
376 break;
377 case PIPE_QUERY_SO_STATISTICS:
378 res64[0] = data64[0] - data64[4];
379 res64[1] = data64[2] - data64[6];
380 break;
381 case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
382 res8[0] = data64[0] != data64[2];
383 break;
384 case PIPE_QUERY_TIMESTAMP:
385 res64[0] = data64[1];
386 break;
387 case PIPE_QUERY_TIMESTAMP_DISJOINT: /* u32 sequence, u32 0, u64 time */
388 res64[0] = 1000000000;
389 res8[8] = (data64[1] == data64[3]) ? FALSE : TRUE;
390 break;
391 case PIPE_QUERY_TIME_ELAPSED:
392 res64[0] = data64[1] - data64[3];
393 break;
394 case PIPE_QUERY_PIPELINE_STATISTICS:
395 for (i = 0; i < 10; ++i)
396 res64[i] = data64[i * 2] - data64[24 + i * 2];
397 break;
398 case NVC0_QUERY_TFB_BUFFER_OFFSET:
399 res32[0] = q->data[1];
400 break;
401 default:
402 return FALSE;
403 }
404
405 return TRUE;
406 }
407
408 void
409 nvc0_query_fifo_wait(struct nouveau_pushbuf *push, struct pipe_query *pq)
410 {
411 struct nvc0_query *q = nvc0_query(pq);
412 unsigned offset = q->offset;
413
414 if (q->type == PIPE_QUERY_SO_OVERFLOW_PREDICATE) offset += 0x20;
415
416 PUSH_SPACE(push, 5);
417 PUSH_REFN (push, q->bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
418 BEGIN_NVC0(push, SUBC_3D(NV84_SUBCHAN_SEMAPHORE_ADDRESS_HIGH), 4);
419 PUSH_DATAh(push, q->bo->offset + offset);
420 PUSH_DATA (push, q->bo->offset + offset);
421 PUSH_DATA (push, q->sequence);
422 PUSH_DATA (push, (1 << 12) |
423 NV84_SUBCHAN_SEMAPHORE_TRIGGER_ACQUIRE_EQUAL);
424 }
425
426 static void
427 nvc0_render_condition(struct pipe_context *pipe,
428 struct pipe_query *pq, uint mode)
429 {
430 struct nvc0_context *nvc0 = nvc0_context(pipe);
431 struct nouveau_pushbuf *push = nvc0->base.pushbuf;
432 struct nvc0_query *q;
433 uint32_t cond;
434 boolean negated = FALSE;
435 boolean wait =
436 mode != PIPE_RENDER_COND_NO_WAIT &&
437 mode != PIPE_RENDER_COND_BY_REGION_NO_WAIT;
438
439 if (!pq) {
440 PUSH_SPACE(push, 1);
441 IMMED_NVC0(push, NVC0_3D(COND_MODE), NVC0_3D_COND_MODE_ALWAYS);
442 return;
443 }
444 q = nvc0_query(pq);
445
446 /* NOTE: comparison of 2 queries only works if both have completed */
447 switch (q->type) {
448 case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
449 cond = negated ? NVC0_3D_COND_MODE_EQUAL :
450 NVC0_3D_COND_MODE_NOT_EQUAL;
451 wait = TRUE;
452 break;
453 case PIPE_QUERY_OCCLUSION_COUNTER:
454 case PIPE_QUERY_OCCLUSION_PREDICATE:
455 if (likely(!negated)) {
456 if (unlikely(q->nesting))
457 cond = wait ? NVC0_3D_COND_MODE_NOT_EQUAL :
458 NVC0_3D_COND_MODE_ALWAYS;
459 else
460 cond = NVC0_3D_COND_MODE_RES_NON_ZERO;
461 } else {
462 cond = wait ? NVC0_3D_COND_MODE_EQUAL : NVC0_3D_COND_MODE_ALWAYS;
463 }
464 break;
465 default:
466 assert(!"render condition query not a predicate");
467 mode = NVC0_3D_COND_MODE_ALWAYS;
468 break;
469 }
470
471 if (wait)
472 nvc0_query_fifo_wait(push, pq);
473
474 PUSH_SPACE(push, 4);
475 PUSH_REFN (push, q->bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
476 BEGIN_NVC0(push, NVC0_3D(COND_ADDRESS_HIGH), 3);
477 PUSH_DATAh(push, q->bo->offset + q->offset);
478 PUSH_DATA (push, q->bo->offset + q->offset);
479 PUSH_DATA (push, cond);
480 }
481
482 void
483 nvc0_query_pushbuf_submit(struct nouveau_pushbuf *push,
484 struct pipe_query *pq, unsigned result_offset)
485 {
486 struct nvc0_query *q = nvc0_query(pq);
487
488 #define NVC0_IB_ENTRY_1_NO_PREFETCH (1 << (31 - 8))
489
490 nouveau_pushbuf_space(push, 0, 0, 1);
491 nouveau_pushbuf_data(push, q->bo, q->offset + result_offset, 4 |
492 NVC0_IB_ENTRY_1_NO_PREFETCH);
493 }
494
495 void
496 nvc0_so_target_save_offset(struct pipe_context *pipe,
497 struct pipe_stream_output_target *ptarg,
498 unsigned index, boolean *serialize)
499 {
500 struct nvc0_so_target *targ = nvc0_so_target(ptarg);
501
502 if (*serialize) {
503 *serialize = FALSE;
504 PUSH_SPACE(nvc0_context(pipe)->base.pushbuf, 1);
505 IMMED_NVC0(nvc0_context(pipe)->base.pushbuf, NVC0_3D(SERIALIZE), 0);
506 }
507
508 nvc0_query(targ->pq)->index = index;
509
510 nvc0_query_end(pipe, targ->pq);
511 }
512
513 void
514 nvc0_init_query_functions(struct nvc0_context *nvc0)
515 {
516 struct pipe_context *pipe = &nvc0->base.pipe;
517
518 pipe->create_query = nvc0_query_create;
519 pipe->destroy_query = nvc0_query_destroy;
520 pipe->begin_query = nvc0_query_begin;
521 pipe->end_query = nvc0_query_end;
522 pipe->get_query_result = nvc0_query_result;
523 pipe->render_condition = nvc0_render_condition;
524 }