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 "nv_object.xml.h"
30 /* XXX: Nested queries, and simultaneous queries on multiple gallium contexts
31 * (since we use only a single GPU channel per screen) will not work properly.
33 * The first is not that big of an issue because OpenGL does not allow nested
42 struct nouveau_bo *bo;
44 uint32_t offset; /* base + i * 32 */
48 struct nouveau_mm_allocation *mm;
51 #define NV50_QUERY_ALLOC_SPACE 256
53 static INLINE struct nv50_query *
54 nv50_query(struct pipe_query *pipe)
56 return (struct nv50_query *)pipe;
60 nv50_query_allocate(struct nv50_context *nv50, struct nv50_query *q, int size)
62 struct nv50_screen *screen = nv50->screen;
66 nouveau_bo_ref(NULL, &q->bo);
69 nouveau_mm_free(q->mm);
71 nouveau_fence_work(screen->base.fence.current, nouveau_mm_free_work,
76 q->mm = nouveau_mm_allocate(screen->base.mm_GART, size, &q->bo, &q->base);
81 ret = nouveau_bo_map(q->bo, 0, screen->base.client);
83 nv50_query_allocate(nv50, q, 0);
86 q->data = (uint32_t *)((uint8_t *)q->bo->map + q->base);
92 nv50_query_destroy(struct pipe_context *pipe, struct pipe_query *pq)
94 nv50_query_allocate(nv50_context(pipe), nv50_query(pq), 0);
98 static struct pipe_query *
99 nv50_query_create(struct pipe_context *pipe, unsigned type, unsigned index)
101 struct nv50_context *nv50 = nv50_context(pipe);
102 struct nv50_query *q;
104 q = CALLOC_STRUCT(nv50_query);
108 if (!nv50_query_allocate(nv50, q, NV50_QUERY_ALLOC_SPACE)) {
113 q->is64bit = (type == PIPE_QUERY_PRIMITIVES_GENERATED ||
114 type == PIPE_QUERY_PRIMITIVES_EMITTED ||
115 type == PIPE_QUERY_SO_STATISTICS);
118 if (q->type == PIPE_QUERY_OCCLUSION_COUNTER) {
120 q->data -= 32 / sizeof(*q->data); /* we advance before query_begin ! */
123 return (struct pipe_query *)q;
127 nv50_query_get(struct nouveau_pushbuf *push, struct nv50_query *q,
128 unsigned offset, uint32_t get)
133 PUSH_REFN (push, q->bo, NOUVEAU_BO_GART | NOUVEAU_BO_WR);
134 BEGIN_NV04(push, NV50_3D(QUERY_ADDRESS_HIGH), 4);
135 PUSH_DATAh(push, q->bo->offset + offset);
136 PUSH_DATA (push, q->bo->offset + offset);
137 PUSH_DATA (push, q->sequence);
138 PUSH_DATA (push, get);
142 nv50_query_begin(struct pipe_context *pipe, struct pipe_query *pq)
144 struct nv50_context *nv50 = nv50_context(pipe);
145 struct nouveau_pushbuf *push = nv50->base.pushbuf;
146 struct nv50_query *q = nv50_query(pq);
148 /* For occlusion queries we have to change the storage, because a previous
149 * query might set the initial render conition to FALSE even *after* we re-
150 * initialized it to TRUE.
152 if (q->type == PIPE_QUERY_OCCLUSION_COUNTER) {
154 q->data += 32 / sizeof(*q->data);
155 if (q->offset - q->base == NV50_QUERY_ALLOC_SPACE)
156 nv50_query_allocate(nv50, q, NV50_QUERY_ALLOC_SPACE);
158 /* XXX: can we do this with the GPU, and sync with respect to a previous
161 q->data[0] = q->sequence; /* initialize sequence */
162 q->data[1] = 1; /* initial render condition = TRUE */
163 q->data[4] = q->sequence + 1; /* for comparison COND_MODE */
167 q->data[0] = q->sequence++; /* the previously used one */
170 case PIPE_QUERY_OCCLUSION_COUNTER:
172 BEGIN_NV04(push, NV50_3D(COUNTER_RESET), 1);
173 PUSH_DATA (push, NV50_3D_COUNTER_RESET_SAMPLECNT);
174 BEGIN_NV04(push, NV50_3D(SAMPLECNT_ENABLE), 1);
177 case PIPE_QUERY_PRIMITIVES_GENERATED:
178 nv50_query_get(push, q, 0x10, 0x06805002);
180 case PIPE_QUERY_PRIMITIVES_EMITTED:
181 nv50_query_get(push, q, 0x10, 0x05805002);
183 case PIPE_QUERY_SO_STATISTICS:
184 nv50_query_get(push, q, 0x20, 0x05805002);
185 nv50_query_get(push, q, 0x30, 0x06805002);
187 case PIPE_QUERY_PIPELINE_STATISTICS:
188 nv50_query_get(push, q, 0x80, 0x00801002); /* VFETCH, VERTICES */
189 nv50_query_get(push, q, 0x90, 0x01801002); /* VFETCH, PRIMS */
190 nv50_query_get(push, q, 0xa0, 0x02802002); /* VP, LAUNCHES */
191 nv50_query_get(push, q, 0xb0, 0x03806002); /* GP, LAUNCHES */
192 nv50_query_get(push, q, 0xc0, 0x04806002); /* GP, PRIMS_OUT */
193 nv50_query_get(push, q, 0xd0, 0x07804002); /* RAST, PRIMS_IN */
194 nv50_query_get(push, q, 0xe0, 0x08804002); /* RAST, PRIMS_OUT */
195 nv50_query_get(push, q, 0xf0, 0x0980a002); /* ROP, PIXELS */
197 case PIPE_QUERY_TIME_ELAPSED:
198 nv50_query_get(push, q, 0x10, 0x00005002);
208 nv50_query_end(struct pipe_context *pipe, struct pipe_query *pq)
210 struct nv50_context *nv50 = nv50_context(pipe);
211 struct nouveau_pushbuf *push = nv50->base.pushbuf;
212 struct nv50_query *q = nv50_query(pq);
215 case PIPE_QUERY_OCCLUSION_COUNTER:
216 nv50_query_get(push, q, 0, 0x0100f002);
218 BEGIN_NV04(push, NV50_3D(SAMPLECNT_ENABLE), 1);
221 case PIPE_QUERY_PRIMITIVES_GENERATED:
222 nv50_query_get(push, q, 0, 0x06805002);
224 case PIPE_QUERY_PRIMITIVES_EMITTED:
225 nv50_query_get(push, q, 0, 0x05805002);
227 case PIPE_QUERY_SO_STATISTICS:
228 nv50_query_get(push, q, 0x00, 0x05805002);
229 nv50_query_get(push, q, 0x10, 0x06805002);
231 case PIPE_QUERY_PIPELINE_STATISTICS:
232 nv50_query_get(push, q, 0x00, 0x00801002); /* VFETCH, VERTICES */
233 nv50_query_get(push, q, 0x10, 0x01801002); /* VFETCH, PRIMS */
234 nv50_query_get(push, q, 0x20, 0x02802002); /* VP, LAUNCHES */
235 nv50_query_get(push, q, 0x30, 0x03806002); /* GP, LAUNCHES */
236 nv50_query_get(push, q, 0x40, 0x04806002); /* GP, PRIMS_OUT */
237 nv50_query_get(push, q, 0x50, 0x07804002); /* RAST, PRIMS_IN */
238 nv50_query_get(push, q, 0x60, 0x08804002); /* RAST, PRIMS_OUT */
239 nv50_query_get(push, q, 0x70, 0x0980a002); /* ROP, PIXELS */
241 case PIPE_QUERY_TIMESTAMP:
244 case PIPE_QUERY_TIME_ELAPSED:
245 nv50_query_get(push, q, 0, 0x00005002);
247 case PIPE_QUERY_GPU_FINISHED:
249 nv50_query_get(push, q, 0, 0x1000f010);
251 case NVA0_QUERY_STREAM_OUTPUT_BUFFER_OFFSET:
252 nv50_query_get(push, q, 0, 0x0d005002 | (q->index << 5));
254 case PIPE_QUERY_TIMESTAMP_DISJOINT:
255 /* This query is not issued on GPU because disjoint is forced to FALSE */
262 q->ready = q->flushed = FALSE;
265 static INLINE boolean
266 nv50_query_ready(struct nv50_query *q)
268 return q->ready || (!q->is64bit && (q->data[0] == q->sequence));
272 nv50_query_result(struct pipe_context *pipe, struct pipe_query *pq,
273 boolean wait, union pipe_query_result *result)
275 struct nv50_context *nv50 = nv50_context(pipe);
276 struct nv50_query *q = nv50_query(pq);
277 uint64_t *res64 = (uint64_t *)result;
278 uint32_t *res32 = (uint32_t *)result;
279 boolean *res8 = (boolean *)result;
280 uint64_t *data64 = (uint64_t *)q->data;
283 if (!q->ready) /* update ? */
284 q->ready = nv50_query_ready(q);
287 /* for broken apps that spin on GL_QUERY_RESULT_AVAILABLE */
290 PUSH_KICK(nv50->base.pushbuf);
294 if (nouveau_bo_wait(q->bo, NOUVEAU_BO_RD, nv50->screen->base.client))
300 case PIPE_QUERY_GPU_FINISHED:
303 case PIPE_QUERY_OCCLUSION_COUNTER: /* u32 sequence, u32 count, u64 time */
304 res64[0] = q->data[1];
306 case PIPE_QUERY_PRIMITIVES_GENERATED: /* u64 count, u64 time */
307 case PIPE_QUERY_PRIMITIVES_EMITTED: /* u64 count, u64 time */
308 res64[0] = data64[0] - data64[2];
310 case PIPE_QUERY_SO_STATISTICS:
311 res64[0] = data64[0] - data64[4];
312 res64[1] = data64[2] - data64[6];
314 case PIPE_QUERY_PIPELINE_STATISTICS:
315 for (i = 0; i < 8; ++i)
316 res64[i] = data64[i * 2] - data64[16 + i * 2];
318 case PIPE_QUERY_TIMESTAMP:
319 res64[0] = data64[1];
321 case PIPE_QUERY_TIMESTAMP_DISJOINT:
322 res64[0] = 1000000000;
325 case PIPE_QUERY_TIME_ELAPSED:
326 res64[0] = data64[1] - data64[3];
328 case NVA0_QUERY_STREAM_OUTPUT_BUFFER_OFFSET:
329 res32[0] = q->data[1];
339 nv84_query_fifo_wait(struct nouveau_pushbuf *push, struct pipe_query *pq)
341 struct nv50_query *q = nv50_query(pq);
342 unsigned offset = q->offset;
345 PUSH_REFN (push, q->bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
346 BEGIN_NV04(push, SUBC_3D(NV84_SUBCHAN_SEMAPHORE_ADDRESS_HIGH), 4);
347 PUSH_DATAh(push, q->bo->offset + offset);
348 PUSH_DATA (push, q->bo->offset + offset);
349 PUSH_DATA (push, q->sequence);
350 PUSH_DATA (push, NV84_SUBCHAN_SEMAPHORE_TRIGGER_ACQUIRE_EQUAL);
354 nv50_render_condition(struct pipe_context *pipe,
355 struct pipe_query *pq,
356 boolean condition, uint mode)
358 struct nv50_context *nv50 = nv50_context(pipe);
359 struct nouveau_pushbuf *push = nv50->base.pushbuf;
360 struct nv50_query *q;
363 mode != PIPE_RENDER_COND_NO_WAIT &&
364 mode != PIPE_RENDER_COND_BY_REGION_NO_WAIT;
367 cond = NV50_3D_COND_MODE_ALWAYS;
371 /* NOTE: comparison of 2 queries only works if both have completed */
373 case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
374 cond = condition ? NV50_3D_COND_MODE_EQUAL :
375 NV50_3D_COND_MODE_NOT_EQUAL;
378 case PIPE_QUERY_OCCLUSION_COUNTER:
379 case PIPE_QUERY_OCCLUSION_PREDICATE:
380 if (likely(!condition)) {
381 /* XXX: Placeholder, handle nesting here if available */
383 cond = wait ? NV50_3D_COND_MODE_NOT_EQUAL :
384 NV50_3D_COND_MODE_ALWAYS;
386 cond = NV50_3D_COND_MODE_RES_NON_ZERO;
388 cond = wait ? NV50_3D_COND_MODE_EQUAL : NV50_3D_COND_MODE_ALWAYS;
392 assert(!"render condition query not a predicate");
393 cond = NV50_3D_COND_MODE_ALWAYS;
398 nv50->cond_query = pq;
399 nv50->cond_cond = condition;
400 nv50->cond_condmode = cond;
401 nv50->cond_mode = mode;
405 BEGIN_NV04(push, NV50_3D(COND_MODE), 1);
406 PUSH_DATA (push, cond);
413 BEGIN_NV04(push, SUBC_3D(NV50_GRAPH_SERIALIZE), 1);
417 PUSH_REFN (push, q->bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
418 BEGIN_NV04(push, NV50_3D(COND_ADDRESS_HIGH), 3);
419 PUSH_DATAh(push, q->bo->offset + q->offset);
420 PUSH_DATA (push, q->bo->offset + q->offset);
421 PUSH_DATA (push, cond);
423 BEGIN_NV04(push, NV50_2D(COND_ADDRESS_HIGH), 2);
424 PUSH_DATAh(push, q->bo->offset + q->offset);
425 PUSH_DATA (push, q->bo->offset + q->offset);
429 nv50_query_pushbuf_submit(struct nouveau_pushbuf *push,
430 struct pipe_query *pq, unsigned result_offset)
432 struct nv50_query *q = nv50_query(pq);
434 /* XXX: does this exist ? */
435 #define NV50_IB_ENTRY_1_NO_PREFETCH (0 << (31 - 8))
437 nouveau_pushbuf_space(push, 0, 0, 1);
438 nouveau_pushbuf_data(push, q->bo, q->offset + result_offset, 4 |
439 NV50_IB_ENTRY_1_NO_PREFETCH);
443 nva0_so_target_save_offset(struct pipe_context *pipe,
444 struct pipe_stream_output_target *ptarg,
445 unsigned index, boolean serialize)
447 struct nv50_so_target *targ = nv50_so_target(ptarg);
450 struct nouveau_pushbuf *push = nv50_context(pipe)->base.pushbuf;
452 BEGIN_NV04(push, SUBC_3D(NV50_GRAPH_SERIALIZE), 1);
456 nv50_query(targ->pq)->index = index;
457 nv50_query_end(pipe, targ->pq);
461 nv50_init_query_functions(struct nv50_context *nv50)
463 struct pipe_context *pipe = &nv50->base.pipe;
465 pipe->create_query = nv50_query_create;
466 pipe->destroy_query = nv50_query_destroy;
467 pipe->begin_query = nv50_query_begin;
468 pipe->end_query = nv50_query_end;
469 pipe->get_query_result = nv50_query_result;
470 pipe->render_condition = nv50_render_condition;