2 * Copyright 2011 Nouveau Project
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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 OR COPYRIGHT HOLDERS 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.
22 * Authors: Christoph Bumiller
25 #define NV50_PUSH_EXPLICIT_SPACE_CHECKING
27 #include "nv50/nv50_context.h"
28 #include "nv50/nv50_query.h"
29 #include "nv_object.xml.h"
31 #define NV50_QUERY_STATE_READY 0
32 #define NV50_QUERY_STATE_ACTIVE 1
33 #define NV50_QUERY_STATE_ENDED 2
34 #define NV50_QUERY_STATE_FLUSHED 3
36 /* XXX: Nested queries, and simultaneous queries on multiple gallium contexts
37 * (since we use only a single GPU channel per screen) will not work properly.
39 * The first is not that big of an issue because OpenGL does not allow nested
43 #define NV50_QUERY_ALLOC_SPACE 256
46 nv50_query_allocate(struct nv50_context *nv50, struct nv50_query *q, int size)
48 struct nv50_screen *screen = nv50->screen;
52 nouveau_bo_ref(NULL, &q->bo);
54 if (q->state == NV50_QUERY_STATE_READY)
55 nouveau_mm_free(q->mm);
57 nouveau_fence_work(screen->base.fence.current, nouveau_mm_free_work,
62 q->mm = nouveau_mm_allocate(screen->base.mm_GART, size, &q->bo, &q->base);
67 ret = nouveau_bo_map(q->bo, 0, screen->base.client);
69 nv50_query_allocate(nv50, q, 0);
72 q->data = (uint32_t *)((uint8_t *)q->bo->map + q->base);
78 nv50_query_destroy(struct pipe_context *pipe, struct pipe_query *pq)
80 nv50_query_allocate(nv50_context(pipe), nv50_query(pq), 0);
81 nouveau_fence_ref(NULL, &nv50_query(pq)->fence);
85 static struct pipe_query *
86 nv50_query_create(struct pipe_context *pipe, unsigned type, unsigned index)
88 struct nv50_context *nv50 = nv50_context(pipe);
91 q = CALLOC_STRUCT(nv50_query);
95 if (!nv50_query_allocate(nv50, q, NV50_QUERY_ALLOC_SPACE)) {
100 q->is64bit = (type == PIPE_QUERY_PRIMITIVES_GENERATED ||
101 type == PIPE_QUERY_PRIMITIVES_EMITTED ||
102 type == PIPE_QUERY_SO_STATISTICS ||
103 type == PIPE_QUERY_PIPELINE_STATISTICS);
106 if (q->type == PIPE_QUERY_OCCLUSION_COUNTER) {
108 q->data -= 32 / sizeof(*q->data); /* we advance before query_begin ! */
111 return (struct pipe_query *)q;
115 nv50_query_get(struct nouveau_pushbuf *push, struct nv50_query *q,
116 unsigned offset, uint32_t get)
121 PUSH_REFN (push, q->bo, NOUVEAU_BO_GART | NOUVEAU_BO_WR);
122 BEGIN_NV04(push, NV50_3D(QUERY_ADDRESS_HIGH), 4);
123 PUSH_DATAh(push, q->bo->offset + offset);
124 PUSH_DATA (push, q->bo->offset + offset);
125 PUSH_DATA (push, q->sequence);
126 PUSH_DATA (push, get);
130 nv50_query_begin(struct pipe_context *pipe, struct pipe_query *pq)
132 struct nv50_context *nv50 = nv50_context(pipe);
133 struct nouveau_pushbuf *push = nv50->base.pushbuf;
134 struct nv50_query *q = nv50_query(pq);
136 /* For occlusion queries we have to change the storage, because a previous
137 * query might set the initial render conition to false even *after* we re-
138 * initialized it to true.
140 if (q->type == PIPE_QUERY_OCCLUSION_COUNTER) {
142 q->data += 32 / sizeof(*q->data);
143 if (q->offset - q->base == NV50_QUERY_ALLOC_SPACE)
144 nv50_query_allocate(nv50, q, NV50_QUERY_ALLOC_SPACE);
146 /* XXX: can we do this with the GPU, and sync with respect to a previous
149 q->data[0] = q->sequence; /* initialize sequence */
150 q->data[1] = 1; /* initial render condition = true */
151 q->data[4] = q->sequence + 1; /* for comparison COND_MODE */
155 q->data[0] = q->sequence++; /* the previously used one */
158 case PIPE_QUERY_OCCLUSION_COUNTER:
159 q->nesting = nv50->screen->num_occlusion_queries_active++;
161 nv50_query_get(push, q, 0x10, 0x0100f002);
164 BEGIN_NV04(push, NV50_3D(COUNTER_RESET), 1);
165 PUSH_DATA (push, NV50_3D_COUNTER_RESET_SAMPLECNT);
166 BEGIN_NV04(push, NV50_3D(SAMPLECNT_ENABLE), 1);
170 case PIPE_QUERY_PRIMITIVES_GENERATED:
171 nv50_query_get(push, q, 0x10, 0x06805002);
173 case PIPE_QUERY_PRIMITIVES_EMITTED:
174 nv50_query_get(push, q, 0x10, 0x05805002);
176 case PIPE_QUERY_SO_STATISTICS:
177 nv50_query_get(push, q, 0x20, 0x05805002);
178 nv50_query_get(push, q, 0x30, 0x06805002);
180 case PIPE_QUERY_PIPELINE_STATISTICS:
181 nv50_query_get(push, q, 0x80, 0x00801002); /* VFETCH, VERTICES */
182 nv50_query_get(push, q, 0x90, 0x01801002); /* VFETCH, PRIMS */
183 nv50_query_get(push, q, 0xa0, 0x02802002); /* VP, LAUNCHES */
184 nv50_query_get(push, q, 0xb0, 0x03806002); /* GP, LAUNCHES */
185 nv50_query_get(push, q, 0xc0, 0x04806002); /* GP, PRIMS_OUT */
186 nv50_query_get(push, q, 0xd0, 0x07804002); /* RAST, PRIMS_IN */
187 nv50_query_get(push, q, 0xe0, 0x08804002); /* RAST, PRIMS_OUT */
188 nv50_query_get(push, q, 0xf0, 0x0980a002); /* ROP, PIXELS */
190 case PIPE_QUERY_TIME_ELAPSED:
191 nv50_query_get(push, q, 0x10, 0x00005002);
196 q->state = NV50_QUERY_STATE_ACTIVE;
201 nv50_query_end(struct pipe_context *pipe, struct pipe_query *pq)
203 struct nv50_context *nv50 = nv50_context(pipe);
204 struct nouveau_pushbuf *push = nv50->base.pushbuf;
205 struct nv50_query *q = nv50_query(pq);
207 q->state = NV50_QUERY_STATE_ENDED;
210 case PIPE_QUERY_OCCLUSION_COUNTER:
211 nv50_query_get(push, q, 0, 0x0100f002);
212 if (--nv50->screen->num_occlusion_queries_active == 0) {
214 BEGIN_NV04(push, NV50_3D(SAMPLECNT_ENABLE), 1);
218 case PIPE_QUERY_PRIMITIVES_GENERATED:
219 nv50_query_get(push, q, 0, 0x06805002);
221 case PIPE_QUERY_PRIMITIVES_EMITTED:
222 nv50_query_get(push, q, 0, 0x05805002);
224 case PIPE_QUERY_SO_STATISTICS:
225 nv50_query_get(push, q, 0x00, 0x05805002);
226 nv50_query_get(push, q, 0x10, 0x06805002);
228 case PIPE_QUERY_PIPELINE_STATISTICS:
229 nv50_query_get(push, q, 0x00, 0x00801002); /* VFETCH, VERTICES */
230 nv50_query_get(push, q, 0x10, 0x01801002); /* VFETCH, PRIMS */
231 nv50_query_get(push, q, 0x20, 0x02802002); /* VP, LAUNCHES */
232 nv50_query_get(push, q, 0x30, 0x03806002); /* GP, LAUNCHES */
233 nv50_query_get(push, q, 0x40, 0x04806002); /* GP, PRIMS_OUT */
234 nv50_query_get(push, q, 0x50, 0x07804002); /* RAST, PRIMS_IN */
235 nv50_query_get(push, q, 0x60, 0x08804002); /* RAST, PRIMS_OUT */
236 nv50_query_get(push, q, 0x70, 0x0980a002); /* ROP, PIXELS */
238 case PIPE_QUERY_TIMESTAMP:
241 case PIPE_QUERY_TIME_ELAPSED:
242 nv50_query_get(push, q, 0, 0x00005002);
244 case PIPE_QUERY_GPU_FINISHED:
246 nv50_query_get(push, q, 0, 0x1000f010);
248 case NVA0_QUERY_STREAM_OUTPUT_BUFFER_OFFSET:
250 nv50_query_get(push, q, 0, 0x0d005002 | (q->index << 5));
252 case PIPE_QUERY_TIMESTAMP_DISJOINT:
253 /* This query is not issued on GPU because disjoint is forced to false */
254 q->state = NV50_QUERY_STATE_READY;
262 nouveau_fence_ref(nv50->screen->base.fence.current, &q->fence);
266 nv50_query_update(struct nv50_query *q)
269 if (nouveau_fence_signalled(q->fence))
270 q->state = NV50_QUERY_STATE_READY;
272 if (q->data[0] == q->sequence)
273 q->state = NV50_QUERY_STATE_READY;
278 nv50_query_result(struct pipe_context *pipe, struct pipe_query *pq,
279 boolean wait, union pipe_query_result *result)
281 struct nv50_context *nv50 = nv50_context(pipe);
282 struct nv50_query *q = nv50_query(pq);
283 uint64_t *res64 = (uint64_t *)result;
284 uint32_t *res32 = (uint32_t *)result;
285 uint8_t *res8 = (uint8_t *)result;
286 uint64_t *data64 = (uint64_t *)q->data;
289 if (q->state != NV50_QUERY_STATE_READY)
290 nv50_query_update(q);
292 if (q->state != NV50_QUERY_STATE_READY) {
294 /* for broken apps that spin on GL_QUERY_RESULT_AVAILABLE */
295 if (q->state != NV50_QUERY_STATE_FLUSHED) {
296 q->state = NV50_QUERY_STATE_FLUSHED;
297 PUSH_KICK(nv50->base.pushbuf);
301 if (nouveau_bo_wait(q->bo, NOUVEAU_BO_RD, nv50->screen->base.client))
304 q->state = NV50_QUERY_STATE_READY;
307 case PIPE_QUERY_GPU_FINISHED:
310 case PIPE_QUERY_OCCLUSION_COUNTER: /* u32 sequence, u32 count, u64 time */
311 res64[0] = q->data[1] - q->data[5];
313 case PIPE_QUERY_PRIMITIVES_GENERATED: /* u64 count, u64 time */
314 case PIPE_QUERY_PRIMITIVES_EMITTED: /* u64 count, u64 time */
315 res64[0] = data64[0] - data64[2];
317 case PIPE_QUERY_SO_STATISTICS:
318 res64[0] = data64[0] - data64[4];
319 res64[1] = data64[2] - data64[6];
321 case PIPE_QUERY_PIPELINE_STATISTICS:
322 for (i = 0; i < 8; ++i)
323 res64[i] = data64[i * 2] - data64[16 + i * 2];
325 case PIPE_QUERY_TIMESTAMP:
326 res64[0] = data64[1];
328 case PIPE_QUERY_TIMESTAMP_DISJOINT:
329 res64[0] = 1000000000;
332 case PIPE_QUERY_TIME_ELAPSED:
333 res64[0] = data64[1] - data64[3];
335 case NVA0_QUERY_STREAM_OUTPUT_BUFFER_OFFSET:
336 res32[0] = q->data[1];
346 nv84_query_fifo_wait(struct nouveau_pushbuf *push, struct nv50_query *q)
348 unsigned offset = q->offset;
351 PUSH_REFN (push, q->bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
352 BEGIN_NV04(push, SUBC_3D(NV84_SUBCHAN_SEMAPHORE_ADDRESS_HIGH), 4);
353 PUSH_DATAh(push, q->bo->offset + offset);
354 PUSH_DATA (push, q->bo->offset + offset);
355 PUSH_DATA (push, q->sequence);
356 PUSH_DATA (push, NV84_SUBCHAN_SEMAPHORE_TRIGGER_ACQUIRE_EQUAL);
360 nv50_render_condition(struct pipe_context *pipe,
361 struct pipe_query *pq,
362 boolean condition, uint mode)
364 struct nv50_context *nv50 = nv50_context(pipe);
365 struct nouveau_pushbuf *push = nv50->base.pushbuf;
366 struct nv50_query *q;
369 mode != PIPE_RENDER_COND_NO_WAIT &&
370 mode != PIPE_RENDER_COND_BY_REGION_NO_WAIT;
373 cond = NV50_3D_COND_MODE_ALWAYS;
377 /* NOTE: comparison of 2 queries only works if both have completed */
379 case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
380 cond = condition ? NV50_3D_COND_MODE_EQUAL :
381 NV50_3D_COND_MODE_NOT_EQUAL;
384 case PIPE_QUERY_OCCLUSION_COUNTER:
385 case PIPE_QUERY_OCCLUSION_PREDICATE:
386 if (likely(!condition)) {
387 if (unlikely(q->nesting))
388 cond = wait ? NV50_3D_COND_MODE_NOT_EQUAL :
389 NV50_3D_COND_MODE_ALWAYS;
391 cond = NV50_3D_COND_MODE_RES_NON_ZERO;
393 cond = wait ? NV50_3D_COND_MODE_EQUAL : NV50_3D_COND_MODE_ALWAYS;
397 assert(!"render condition query not a predicate");
398 cond = NV50_3D_COND_MODE_ALWAYS;
403 nv50->cond_query = pq;
404 nv50->cond_cond = condition;
405 nv50->cond_condmode = cond;
406 nv50->cond_mode = mode;
410 BEGIN_NV04(push, NV50_3D(COND_MODE), 1);
411 PUSH_DATA (push, cond);
418 BEGIN_NV04(push, SUBC_3D(NV50_GRAPH_SERIALIZE), 1);
422 PUSH_REFN (push, q->bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
423 BEGIN_NV04(push, NV50_3D(COND_ADDRESS_HIGH), 3);
424 PUSH_DATAh(push, q->bo->offset + q->offset);
425 PUSH_DATA (push, q->bo->offset + q->offset);
426 PUSH_DATA (push, cond);
428 BEGIN_NV04(push, NV50_2D(COND_ADDRESS_HIGH), 2);
429 PUSH_DATAh(push, q->bo->offset + q->offset);
430 PUSH_DATA (push, q->bo->offset + q->offset);
434 nv50_query_pushbuf_submit(struct nouveau_pushbuf *push, uint16_t method,
435 struct nv50_query *q, unsigned result_offset)
437 nv50_query_update(q);
438 if (q->state != NV50_QUERY_STATE_READY)
439 nouveau_bo_wait(q->bo, NOUVEAU_BO_RD, push->client);
440 q->state = NV50_QUERY_STATE_READY;
442 BEGIN_NV04(push, SUBC_3D(method), 1);
443 PUSH_DATA (push, q->data[result_offset / 4]);
447 nv50_init_query_functions(struct nv50_context *nv50)
449 struct pipe_context *pipe = &nv50->base.pipe;
451 pipe->create_query = nv50_query_create;
452 pipe->destroy_query = nv50_query_destroy;
453 pipe->begin_query = nv50_query_begin;
454 pipe->end_query = nv50_query_end;
455 pipe->get_query_result = nv50_query_result;
456 pipe->render_condition = nv50_render_condition;