nvc0: upload sample locations on GM20x
[mesa.git] / src / gallium / drivers / nouveau / nvc0 / nvc0_context.c
1 /*
2 * Copyright 2010 Christoph Bumiller
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 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.
21 */
22
23 #include "pipe/p_defines.h"
24 #include "util/u_framebuffer.h"
25
26 #include "nvc0/nvc0_context.h"
27 #include "nvc0/nvc0_screen.h"
28 #include "nvc0/nvc0_resource.h"
29
30 static void
31 nvc0_flush(struct pipe_context *pipe,
32 struct pipe_fence_handle **fence,
33 unsigned flags)
34 {
35 struct nvc0_context *nvc0 = nvc0_context(pipe);
36 struct nouveau_screen *screen = &nvc0->screen->base;
37
38 if (fence)
39 nouveau_fence_ref(screen->fence.current, (struct nouveau_fence **)fence);
40
41 PUSH_KICK(nvc0->base.pushbuf); /* fencing handled in kick_notify */
42
43 nouveau_context_update_frame_stats(&nvc0->base);
44 }
45
46 static void
47 nvc0_texture_barrier(struct pipe_context *pipe)
48 {
49 struct nouveau_pushbuf *push = nvc0_context(pipe)->base.pushbuf;
50
51 IMMED_NVC0(push, NVC0_3D(SERIALIZE), 0);
52 IMMED_NVC0(push, NVC0_3D(TEX_CACHE_CTL), 0);
53 }
54
55 static void
56 nvc0_memory_barrier(struct pipe_context *pipe, unsigned flags)
57 {
58 struct nvc0_context *nvc0 = nvc0_context(pipe);
59 struct nouveau_pushbuf *push = nvc0->base.pushbuf;
60 int i, s;
61
62 if (flags & PIPE_BARRIER_MAPPED_BUFFER) {
63 for (i = 0; i < nvc0->num_vtxbufs; ++i) {
64 if (!nvc0->vtxbuf[i].buffer)
65 continue;
66 if (nvc0->vtxbuf[i].buffer->flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT)
67 nvc0->base.vbo_dirty = true;
68 }
69
70 if (nvc0->idxbuf.buffer &&
71 nvc0->idxbuf.buffer->flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT)
72 nvc0->base.vbo_dirty = true;
73
74 for (s = 0; s < 5 && !nvc0->cb_dirty; ++s) {
75 uint32_t valid = nvc0->constbuf_valid[s];
76
77 while (valid && !nvc0->cb_dirty) {
78 const unsigned i = ffs(valid) - 1;
79 struct pipe_resource *res;
80
81 valid &= ~(1 << i);
82 if (nvc0->constbuf[s][i].user)
83 continue;
84
85 res = nvc0->constbuf[s][i].u.buf;
86 if (!res)
87 continue;
88
89 if (res->flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT)
90 nvc0->cb_dirty = true;
91 }
92 }
93 } else {
94 /* Pretty much any writing by shaders needs a serialize after
95 * it. Especially when moving between 3d and compute pipelines, but even
96 * without that.
97 */
98 IMMED_NVC0(push, NVC0_3D(SERIALIZE), 0);
99 }
100
101 /* If we're going to texture from a buffer/image written by a shader, we
102 * must flush the texture cache.
103 */
104 if (flags & PIPE_BARRIER_TEXTURE)
105 IMMED_NVC0(push, NVC0_3D(TEX_CACHE_CTL), 0);
106
107 if (flags & PIPE_BARRIER_CONSTANT_BUFFER)
108 nvc0->cb_dirty = true;
109 if (flags & (PIPE_BARRIER_VERTEX_BUFFER | PIPE_BARRIER_INDEX_BUFFER))
110 nvc0->base.vbo_dirty = true;
111 }
112
113 static void
114 nvc0_emit_string_marker(struct pipe_context *pipe, const char *str, int len)
115 {
116 struct nouveau_pushbuf *push = nvc0_context(pipe)->base.pushbuf;
117 int string_words = len / 4;
118 int data_words;
119
120 if (len <= 0)
121 return;
122 string_words = MIN2(string_words, NV04_PFIFO_MAX_PACKET_LEN);
123 if (string_words == NV04_PFIFO_MAX_PACKET_LEN)
124 data_words = string_words;
125 else
126 data_words = string_words + !!(len & 3);
127 BEGIN_NIC0(push, SUBC_3D(NV04_GRAPH_NOP), data_words);
128 if (string_words)
129 PUSH_DATAp(push, str, string_words);
130 if (string_words != data_words) {
131 int data = 0;
132 memcpy(&data, &str[string_words * 4], len & 3);
133 PUSH_DATA (push, data);
134 }
135 }
136
137 static void
138 nvc0_context_unreference_resources(struct nvc0_context *nvc0)
139 {
140 unsigned s, i;
141
142 nouveau_bufctx_del(&nvc0->bufctx_3d);
143 nouveau_bufctx_del(&nvc0->bufctx);
144 nouveau_bufctx_del(&nvc0->bufctx_cp);
145
146 util_unreference_framebuffer_state(&nvc0->framebuffer);
147
148 for (i = 0; i < nvc0->num_vtxbufs; ++i)
149 pipe_resource_reference(&nvc0->vtxbuf[i].buffer, NULL);
150
151 pipe_resource_reference(&nvc0->idxbuf.buffer, NULL);
152
153 for (s = 0; s < 6; ++s) {
154 for (i = 0; i < nvc0->num_textures[s]; ++i)
155 pipe_sampler_view_reference(&nvc0->textures[s][i], NULL);
156
157 for (i = 0; i < NVC0_MAX_PIPE_CONSTBUFS; ++i)
158 if (!nvc0->constbuf[s][i].user)
159 pipe_resource_reference(&nvc0->constbuf[s][i].u.buf, NULL);
160
161 for (i = 0; i < NVC0_MAX_BUFFERS; ++i)
162 pipe_resource_reference(&nvc0->buffers[s][i].buffer, NULL);
163
164 for (i = 0; i < NVC0_MAX_IMAGES; ++i) {
165 pipe_resource_reference(&nvc0->images[s][i].resource, NULL);
166 if (nvc0->screen->base.class_3d >= GM107_3D_CLASS)
167 pipe_sampler_view_reference(&nvc0->images_tic[s][i], NULL);
168 }
169 }
170
171 for (s = 0; s < 2; ++s) {
172 for (i = 0; i < NVC0_MAX_SURFACE_SLOTS; ++i)
173 pipe_surface_reference(&nvc0->surfaces[s][i], NULL);
174 }
175
176 for (i = 0; i < nvc0->num_tfbbufs; ++i)
177 pipe_so_target_reference(&nvc0->tfbbuf[i], NULL);
178
179 for (i = 0; i < nvc0->global_residents.size / sizeof(struct pipe_resource *);
180 ++i) {
181 struct pipe_resource **res = util_dynarray_element(
182 &nvc0->global_residents, struct pipe_resource *, i);
183 pipe_resource_reference(res, NULL);
184 }
185 util_dynarray_fini(&nvc0->global_residents);
186
187 if (nvc0->tcp_empty)
188 nvc0->base.pipe.delete_tcs_state(&nvc0->base.pipe, nvc0->tcp_empty);
189 }
190
191 static void
192 nvc0_destroy(struct pipe_context *pipe)
193 {
194 struct nvc0_context *nvc0 = nvc0_context(pipe);
195
196 if (nvc0->screen->cur_ctx == nvc0) {
197 nvc0->screen->cur_ctx = NULL;
198 nvc0->screen->save_state = nvc0->state;
199 nvc0->screen->save_state.tfb = NULL;
200 }
201
202 /* Unset bufctx, we don't want to revalidate any resources after the flush.
203 * Other contexts will always set their bufctx again on action calls.
204 */
205 nouveau_pushbuf_bufctx(nvc0->base.pushbuf, NULL);
206 nouveau_pushbuf_kick(nvc0->base.pushbuf, nvc0->base.pushbuf->channel);
207
208 nvc0_context_unreference_resources(nvc0);
209 nvc0_blitctx_destroy(nvc0);
210
211 nouveau_context_destroy(&nvc0->base);
212 }
213
214 void
215 nvc0_default_kick_notify(struct nouveau_pushbuf *push)
216 {
217 struct nvc0_screen *screen = push->user_priv;
218
219 if (screen) {
220 nouveau_fence_next(&screen->base);
221 nouveau_fence_update(&screen->base, true);
222 if (screen->cur_ctx)
223 screen->cur_ctx->state.flushed = true;
224 NOUVEAU_DRV_STAT(&screen->base, pushbuf_count, 1);
225 }
226 }
227
228 static int
229 nvc0_invalidate_resource_storage(struct nouveau_context *ctx,
230 struct pipe_resource *res,
231 int ref)
232 {
233 struct nvc0_context *nvc0 = nvc0_context(&ctx->pipe);
234 unsigned s, i;
235
236 if (res->bind & PIPE_BIND_RENDER_TARGET) {
237 for (i = 0; i < nvc0->framebuffer.nr_cbufs; ++i) {
238 if (nvc0->framebuffer.cbufs[i] &&
239 nvc0->framebuffer.cbufs[i]->texture == res) {
240 nvc0->dirty_3d |= NVC0_NEW_3D_FRAMEBUFFER;
241 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_FB);
242 if (!--ref)
243 return ref;
244 }
245 }
246 }
247 if (res->bind & PIPE_BIND_DEPTH_STENCIL) {
248 if (nvc0->framebuffer.zsbuf &&
249 nvc0->framebuffer.zsbuf->texture == res) {
250 nvc0->dirty_3d |= NVC0_NEW_3D_FRAMEBUFFER;
251 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_FB);
252 if (!--ref)
253 return ref;
254 }
255 }
256
257 if (res->target == PIPE_BUFFER) {
258 for (i = 0; i < nvc0->num_vtxbufs; ++i) {
259 if (nvc0->vtxbuf[i].buffer == res) {
260 nvc0->dirty_3d |= NVC0_NEW_3D_ARRAYS;
261 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_VTX);
262 if (!--ref)
263 return ref;
264 }
265 }
266
267 if (nvc0->idxbuf.buffer == res) {
268 nvc0->dirty_3d |= NVC0_NEW_3D_IDXBUF;
269 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_IDX);
270 if (!--ref)
271 return ref;
272 }
273
274 for (s = 0; s < 6; ++s) {
275 for (i = 0; i < nvc0->num_textures[s]; ++i) {
276 if (nvc0->textures[s][i] &&
277 nvc0->textures[s][i]->texture == res) {
278 nvc0->textures_dirty[s] |= 1 << i;
279 if (unlikely(s == 5)) {
280 nvc0->dirty_cp |= NVC0_NEW_CP_TEXTURES;
281 nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_TEX(i));
282 } else {
283 nvc0->dirty_3d |= NVC0_NEW_3D_TEXTURES;
284 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_TEX(s, i));
285 }
286 if (!--ref)
287 return ref;
288 }
289 }
290 }
291
292 for (s = 0; s < 6; ++s) {
293 for (i = 0; i < NVC0_MAX_PIPE_CONSTBUFS; ++i) {
294 if (!(nvc0->constbuf_valid[s] & (1 << i)))
295 continue;
296 if (!nvc0->constbuf[s][i].user &&
297 nvc0->constbuf[s][i].u.buf == res) {
298 nvc0->constbuf_dirty[s] |= 1 << i;
299 if (unlikely(s == 5)) {
300 nvc0->dirty_cp |= NVC0_NEW_CP_CONSTBUF;
301 nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_CB(i));
302 } else {
303 nvc0->dirty_3d |= NVC0_NEW_3D_CONSTBUF;
304 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_CB(s, i));
305 }
306 if (!--ref)
307 return ref;
308 }
309 }
310 }
311
312 for (s = 0; s < 6; ++s) {
313 for (i = 0; i < NVC0_MAX_BUFFERS; ++i) {
314 if (nvc0->buffers[s][i].buffer == res) {
315 nvc0->buffers_dirty[s] |= 1 << i;
316 if (unlikely(s == 5)) {
317 nvc0->dirty_cp |= NVC0_NEW_CP_BUFFERS;
318 nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_BUF);
319 } else {
320 nvc0->dirty_3d |= NVC0_NEW_3D_BUFFERS;
321 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_BUF);
322 }
323 if (!--ref)
324 return ref;
325 }
326 }
327 }
328
329 for (s = 0; s < 6; ++s) {
330 for (i = 0; i < NVC0_MAX_IMAGES; ++i) {
331 if (nvc0->images[s][i].resource == res) {
332 nvc0->images_dirty[s] |= 1 << i;
333 if (unlikely(s == 5)) {
334 nvc0->dirty_cp |= NVC0_NEW_CP_SURFACES;
335 nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_SUF);
336 } else {
337 nvc0->dirty_3d |= NVC0_NEW_3D_SURFACES;
338 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_SUF);
339 }
340 }
341 if (!--ref)
342 return ref;
343 }
344 }
345 }
346
347 return ref;
348 }
349
350 static void
351 nvc0_context_get_sample_position(struct pipe_context *, unsigned, unsigned,
352 float *);
353
354 struct pipe_context *
355 nvc0_create(struct pipe_screen *pscreen, void *priv, unsigned ctxflags)
356 {
357 struct nvc0_screen *screen = nvc0_screen(pscreen);
358 struct nvc0_context *nvc0;
359 struct pipe_context *pipe;
360 int ret;
361 uint32_t flags;
362
363 nvc0 = CALLOC_STRUCT(nvc0_context);
364 if (!nvc0)
365 return NULL;
366 pipe = &nvc0->base.pipe;
367
368 if (!nvc0_blitctx_create(nvc0))
369 goto out_err;
370
371 nvc0->base.pushbuf = screen->base.pushbuf;
372 nvc0->base.client = screen->base.client;
373
374 ret = nouveau_bufctx_new(screen->base.client, 2, &nvc0->bufctx);
375 if (!ret)
376 ret = nouveau_bufctx_new(screen->base.client, NVC0_BIND_3D_COUNT,
377 &nvc0->bufctx_3d);
378 if (!ret)
379 ret = nouveau_bufctx_new(screen->base.client, NVC0_BIND_CP_COUNT,
380 &nvc0->bufctx_cp);
381 if (ret)
382 goto out_err;
383
384 nvc0->screen = screen;
385 nvc0->base.screen = &screen->base;
386
387 pipe->screen = pscreen;
388 pipe->priv = priv;
389
390 pipe->destroy = nvc0_destroy;
391
392 pipe->draw_vbo = nvc0_draw_vbo;
393 pipe->clear = nvc0_clear;
394 pipe->launch_grid = (nvc0->screen->base.class_3d >= NVE4_3D_CLASS) ?
395 nve4_launch_grid : nvc0_launch_grid;
396
397 pipe->flush = nvc0_flush;
398 pipe->texture_barrier = nvc0_texture_barrier;
399 pipe->memory_barrier = nvc0_memory_barrier;
400 pipe->get_sample_position = nvc0_context_get_sample_position;
401 pipe->emit_string_marker = nvc0_emit_string_marker;
402
403 nouveau_context_init(&nvc0->base);
404 nvc0_init_query_functions(nvc0);
405 nvc0_init_surface_functions(nvc0);
406 nvc0_init_state_functions(nvc0);
407 nvc0_init_transfer_functions(nvc0);
408 nvc0_init_resource_functions(pipe);
409
410 nvc0->base.invalidate_resource_storage = nvc0_invalidate_resource_storage;
411
412 pipe->create_video_codec = nvc0_create_decoder;
413 pipe->create_video_buffer = nvc0_video_buffer_create;
414
415 /* shader builtin library is per-screen, but we need a context for m2mf */
416 nvc0_program_library_upload(nvc0);
417 nvc0_program_init_tcp_empty(nvc0);
418 if (!nvc0->tcp_empty)
419 goto out_err;
420 /* set the empty tctl prog on next draw in case one is never set */
421 nvc0->dirty_3d |= NVC0_NEW_3D_TCTLPROG;
422
423 /* Do not bind the COMPUTE driver constbuf at screen initialization because
424 * CBs are aliased between 3D and COMPUTE, but make sure it will be bound if
425 * a grid is launched later. */
426 nvc0->dirty_cp |= NVC0_NEW_CP_DRIVERCONST;
427
428 /* now that there are no more opportunities for errors, set the current
429 * context if there isn't already one.
430 */
431 if (!screen->cur_ctx) {
432 nvc0->state = screen->save_state;
433 screen->cur_ctx = nvc0;
434 nouveau_pushbuf_bufctx(screen->base.pushbuf, nvc0->bufctx);
435 }
436 screen->base.pushbuf->kick_notify = nvc0_default_kick_notify;
437
438 /* add permanently resident buffers to bufctxts */
439
440 flags = NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RD;
441
442 BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->text);
443 BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->uniform_bo);
444 BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->txc);
445 if (screen->compute) {
446 BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->text);
447 BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->uniform_bo);
448 BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->txc);
449 }
450
451 flags = NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RDWR;
452
453 if (screen->poly_cache)
454 BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->poly_cache);
455 if (screen->compute)
456 BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->tls);
457
458 flags = NOUVEAU_BO_GART | NOUVEAU_BO_WR;
459
460 BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->fence.bo);
461 BCTX_REFN_bo(nvc0->bufctx, FENCE, flags, screen->fence.bo);
462 if (screen->compute)
463 BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->fence.bo);
464
465 nvc0->base.scratch.bo_size = 2 << 20;
466
467 memset(nvc0->tex_handles, ~0, sizeof(nvc0->tex_handles));
468
469 util_dynarray_init(&nvc0->global_residents);
470
471 return pipe;
472
473 out_err:
474 if (nvc0) {
475 if (nvc0->bufctx_3d)
476 nouveau_bufctx_del(&nvc0->bufctx_3d);
477 if (nvc0->bufctx_cp)
478 nouveau_bufctx_del(&nvc0->bufctx_cp);
479 if (nvc0->bufctx)
480 nouveau_bufctx_del(&nvc0->bufctx);
481 FREE(nvc0->blit);
482 FREE(nvc0);
483 }
484 return NULL;
485 }
486
487 void
488 nvc0_bufctx_fence(struct nvc0_context *nvc0, struct nouveau_bufctx *bufctx,
489 bool on_flush)
490 {
491 struct nouveau_list *list = on_flush ? &bufctx->current : &bufctx->pending;
492 struct nouveau_list *it;
493 NOUVEAU_DRV_STAT_IFD(unsigned count = 0);
494
495 for (it = list->next; it != list; it = it->next) {
496 struct nouveau_bufref *ref = (struct nouveau_bufref *)it;
497 struct nv04_resource *res = ref->priv;
498 if (res)
499 nvc0_resource_validate(res, (unsigned)ref->priv_data);
500 NOUVEAU_DRV_STAT_IFD(count++);
501 }
502 NOUVEAU_DRV_STAT(&nvc0->screen->base, resource_validate_count, count);
503 }
504
505 const void *
506 nvc0_get_sample_locations(unsigned sample_count)
507 {
508 static const uint8_t ms1[1][2] = { { 0x8, 0x8 } };
509 static const uint8_t ms2[2][2] = {
510 { 0x4, 0x4 }, { 0xc, 0xc } }; /* surface coords (0,0), (1,0) */
511 static const uint8_t ms4[4][2] = {
512 { 0x6, 0x2 }, { 0xe, 0x6 }, /* (0,0), (1,0) */
513 { 0x2, 0xa }, { 0xa, 0xe } }; /* (0,1), (1,1) */
514 static const uint8_t ms8[8][2] = {
515 { 0x1, 0x7 }, { 0x5, 0x3 }, /* (0,0), (1,0) */
516 { 0x3, 0xd }, { 0x7, 0xb }, /* (0,1), (1,1) */
517 { 0x9, 0x5 }, { 0xf, 0x1 }, /* (2,0), (3,0) */
518 { 0xb, 0xf }, { 0xd, 0x9 } }; /* (2,1), (3,1) */
519 #if 0
520 /* NOTE: there are alternative modes for MS2 and MS8, currently not used */
521 static const uint8_t ms8_alt[8][2] = {
522 { 0x9, 0x5 }, { 0x7, 0xb }, /* (2,0), (1,1) */
523 { 0xd, 0x9 }, { 0x5, 0x3 }, /* (3,1), (1,0) */
524 { 0x3, 0xd }, { 0x1, 0x7 }, /* (0,1), (0,0) */
525 { 0xb, 0xf }, { 0xf, 0x1 } }; /* (2,1), (3,0) */
526 #endif
527
528 const uint8_t (*ptr)[2];
529
530 switch (sample_count) {
531 case 0:
532 case 1: ptr = ms1; break;
533 case 2: ptr = ms2; break;
534 case 4: ptr = ms4; break;
535 case 8: ptr = ms8; break;
536 default:
537 assert(0);
538 return NULL; /* bad sample count -> undefined locations */
539 }
540 return ptr;
541 }
542
543 static void
544 nvc0_context_get_sample_position(struct pipe_context *pipe,
545 unsigned sample_count, unsigned sample_index,
546 float *xy)
547 {
548 const uint8_t (*ptr)[2];
549
550 ptr = nvc0_get_sample_locations(sample_count);
551 if (!ptr)
552 return;
553
554 xy[0] = ptr[sample_index][0] * 0.0625f;
555 xy[1] = ptr[sample_index][1] * 0.0625f;
556 }