nvc0: fix the driver cb size when draw parameters are used
[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 }
167
168 for (s = 0; s < 2; ++s) {
169 for (i = 0; i < NVC0_MAX_SURFACE_SLOTS; ++i)
170 pipe_surface_reference(&nvc0->surfaces[s][i], NULL);
171 }
172
173 for (i = 0; i < nvc0->num_tfbbufs; ++i)
174 pipe_so_target_reference(&nvc0->tfbbuf[i], NULL);
175
176 for (i = 0; i < nvc0->global_residents.size / sizeof(struct pipe_resource *);
177 ++i) {
178 struct pipe_resource **res = util_dynarray_element(
179 &nvc0->global_residents, struct pipe_resource *, i);
180 pipe_resource_reference(res, NULL);
181 }
182 util_dynarray_fini(&nvc0->global_residents);
183
184 if (nvc0->tcp_empty)
185 nvc0->base.pipe.delete_tcs_state(&nvc0->base.pipe, nvc0->tcp_empty);
186 }
187
188 static void
189 nvc0_destroy(struct pipe_context *pipe)
190 {
191 struct nvc0_context *nvc0 = nvc0_context(pipe);
192
193 if (nvc0->screen->cur_ctx == nvc0) {
194 nvc0->screen->cur_ctx = NULL;
195 nvc0->screen->save_state = nvc0->state;
196 nvc0->screen->save_state.tfb = NULL;
197 }
198
199 /* Unset bufctx, we don't want to revalidate any resources after the flush.
200 * Other contexts will always set their bufctx again on action calls.
201 */
202 nouveau_pushbuf_bufctx(nvc0->base.pushbuf, NULL);
203 nouveau_pushbuf_kick(nvc0->base.pushbuf, nvc0->base.pushbuf->channel);
204
205 nvc0_context_unreference_resources(nvc0);
206 nvc0_blitctx_destroy(nvc0);
207
208 nouveau_context_destroy(&nvc0->base);
209 }
210
211 void
212 nvc0_default_kick_notify(struct nouveau_pushbuf *push)
213 {
214 struct nvc0_screen *screen = push->user_priv;
215
216 if (screen) {
217 nouveau_fence_next(&screen->base);
218 nouveau_fence_update(&screen->base, true);
219 if (screen->cur_ctx)
220 screen->cur_ctx->state.flushed = true;
221 NOUVEAU_DRV_STAT(&screen->base, pushbuf_count, 1);
222 }
223 }
224
225 static int
226 nvc0_invalidate_resource_storage(struct nouveau_context *ctx,
227 struct pipe_resource *res,
228 int ref)
229 {
230 struct nvc0_context *nvc0 = nvc0_context(&ctx->pipe);
231 unsigned s, i;
232
233 if (res->bind & PIPE_BIND_RENDER_TARGET) {
234 for (i = 0; i < nvc0->framebuffer.nr_cbufs; ++i) {
235 if (nvc0->framebuffer.cbufs[i] &&
236 nvc0->framebuffer.cbufs[i]->texture == res) {
237 nvc0->dirty_3d |= NVC0_NEW_3D_FRAMEBUFFER;
238 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_FB);
239 if (!--ref)
240 return ref;
241 }
242 }
243 }
244 if (res->bind & PIPE_BIND_DEPTH_STENCIL) {
245 if (nvc0->framebuffer.zsbuf &&
246 nvc0->framebuffer.zsbuf->texture == res) {
247 nvc0->dirty_3d |= NVC0_NEW_3D_FRAMEBUFFER;
248 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_FB);
249 if (!--ref)
250 return ref;
251 }
252 }
253
254 if (res->target == PIPE_BUFFER) {
255 for (i = 0; i < nvc0->num_vtxbufs; ++i) {
256 if (nvc0->vtxbuf[i].buffer == res) {
257 nvc0->dirty_3d |= NVC0_NEW_3D_ARRAYS;
258 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_VTX);
259 if (!--ref)
260 return ref;
261 }
262 }
263
264 if (nvc0->idxbuf.buffer == res) {
265 nvc0->dirty_3d |= NVC0_NEW_3D_IDXBUF;
266 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_IDX);
267 if (!--ref)
268 return ref;
269 }
270
271 for (s = 0; s < 6; ++s) {
272 for (i = 0; i < nvc0->num_textures[s]; ++i) {
273 if (nvc0->textures[s][i] &&
274 nvc0->textures[s][i]->texture == res) {
275 nvc0->textures_dirty[s] |= 1 << i;
276 if (unlikely(s == 5)) {
277 nvc0->dirty_cp |= NVC0_NEW_CP_TEXTURES;
278 nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_TEX(i));
279 } else {
280 nvc0->dirty_3d |= NVC0_NEW_3D_TEXTURES;
281 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_TEX(s, i));
282 }
283 if (!--ref)
284 return ref;
285 }
286 }
287 }
288
289 for (s = 0; s < 6; ++s) {
290 for (i = 0; i < NVC0_MAX_PIPE_CONSTBUFS; ++i) {
291 if (!(nvc0->constbuf_valid[s] & (1 << i)))
292 continue;
293 if (!nvc0->constbuf[s][i].user &&
294 nvc0->constbuf[s][i].u.buf == res) {
295 nvc0->constbuf_dirty[s] |= 1 << i;
296 if (unlikely(s == 5)) {
297 nvc0->dirty_cp |= NVC0_NEW_CP_CONSTBUF;
298 nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_CB(i));
299 } else {
300 nvc0->dirty_3d |= NVC0_NEW_3D_CONSTBUF;
301 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_CB(s, i));
302 }
303 if (!--ref)
304 return ref;
305 }
306 }
307 }
308
309 for (s = 0; s < 6; ++s) {
310 for (i = 0; i < NVC0_MAX_BUFFERS; ++i) {
311 if (nvc0->buffers[s][i].buffer == res) {
312 nvc0->buffers_dirty[s] |= 1 << i;
313 if (unlikely(s == 5)) {
314 nvc0->dirty_cp |= NVC0_NEW_CP_BUFFERS;
315 nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_BUF);
316 } else {
317 nvc0->dirty_3d |= NVC0_NEW_3D_BUFFERS;
318 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_BUF);
319 }
320 if (!--ref)
321 return ref;
322 }
323 }
324 }
325
326 for (s = 0; s < 6; ++s) {
327 for (i = 0; i < NVC0_MAX_IMAGES; ++i) {
328 if (nvc0->images[s][i].resource == res) {
329 nvc0->images_dirty[s] |= 1 << i;
330 if (unlikely(s == 5)) {
331 nvc0->dirty_cp |= NVC0_NEW_CP_SURFACES;
332 nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_SUF);
333 } else {
334 nvc0->dirty_3d |= NVC0_NEW_3D_SURFACES;
335 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_SUF);
336 }
337 }
338 if (!--ref)
339 return ref;
340 }
341 }
342 }
343
344 return ref;
345 }
346
347 static void
348 nvc0_context_get_sample_position(struct pipe_context *, unsigned, unsigned,
349 float *);
350
351 struct pipe_context *
352 nvc0_create(struct pipe_screen *pscreen, void *priv, unsigned ctxflags)
353 {
354 struct nvc0_screen *screen = nvc0_screen(pscreen);
355 struct nvc0_context *nvc0;
356 struct pipe_context *pipe;
357 int ret;
358 uint32_t flags;
359
360 nvc0 = CALLOC_STRUCT(nvc0_context);
361 if (!nvc0)
362 return NULL;
363 pipe = &nvc0->base.pipe;
364
365 if (!nvc0_blitctx_create(nvc0))
366 goto out_err;
367
368 nvc0->base.pushbuf = screen->base.pushbuf;
369 nvc0->base.client = screen->base.client;
370
371 ret = nouveau_bufctx_new(screen->base.client, 2, &nvc0->bufctx);
372 if (!ret)
373 ret = nouveau_bufctx_new(screen->base.client, NVC0_BIND_3D_COUNT,
374 &nvc0->bufctx_3d);
375 if (!ret)
376 ret = nouveau_bufctx_new(screen->base.client, NVC0_BIND_CP_COUNT,
377 &nvc0->bufctx_cp);
378 if (ret)
379 goto out_err;
380
381 nvc0->screen = screen;
382 nvc0->base.screen = &screen->base;
383
384 pipe->screen = pscreen;
385 pipe->priv = priv;
386
387 pipe->destroy = nvc0_destroy;
388
389 pipe->draw_vbo = nvc0_draw_vbo;
390 pipe->clear = nvc0_clear;
391 pipe->launch_grid = (nvc0->screen->base.class_3d >= NVE4_3D_CLASS) ?
392 nve4_launch_grid : nvc0_launch_grid;
393
394 pipe->flush = nvc0_flush;
395 pipe->texture_barrier = nvc0_texture_barrier;
396 pipe->memory_barrier = nvc0_memory_barrier;
397 pipe->get_sample_position = nvc0_context_get_sample_position;
398 pipe->emit_string_marker = nvc0_emit_string_marker;
399
400 nouveau_context_init(&nvc0->base);
401 nvc0_init_query_functions(nvc0);
402 nvc0_init_surface_functions(nvc0);
403 nvc0_init_state_functions(nvc0);
404 nvc0_init_transfer_functions(nvc0);
405 nvc0_init_resource_functions(pipe);
406
407 nvc0->base.invalidate_resource_storage = nvc0_invalidate_resource_storage;
408
409 pipe->create_video_codec = nvc0_create_decoder;
410 pipe->create_video_buffer = nvc0_video_buffer_create;
411
412 /* shader builtin library is per-screen, but we need a context for m2mf */
413 nvc0_program_library_upload(nvc0);
414 nvc0_program_init_tcp_empty(nvc0);
415 if (!nvc0->tcp_empty)
416 goto out_err;
417 /* set the empty tctl prog on next draw in case one is never set */
418 nvc0->dirty_3d |= NVC0_NEW_3D_TCTLPROG;
419
420 /* Do not bind the COMPUTE driver constbuf at screen initialization because
421 * CBs are aliased between 3D and COMPUTE, but make sure it will be bound if
422 * a grid is launched later. */
423 nvc0->dirty_cp |= NVC0_NEW_CP_DRIVERCONST;
424
425 /* now that there are no more opportunities for errors, set the current
426 * context if there isn't already one.
427 */
428 if (!screen->cur_ctx) {
429 nvc0->state = screen->save_state;
430 screen->cur_ctx = nvc0;
431 nouveau_pushbuf_bufctx(screen->base.pushbuf, nvc0->bufctx);
432 }
433 screen->base.pushbuf->kick_notify = nvc0_default_kick_notify;
434
435 /* add permanently resident buffers to bufctxts */
436
437 flags = NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RD;
438
439 BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->text);
440 BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->uniform_bo);
441 BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->txc);
442 if (screen->compute) {
443 BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->text);
444 BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->uniform_bo);
445 BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->txc);
446 }
447
448 flags = NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RDWR;
449
450 if (screen->poly_cache)
451 BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->poly_cache);
452 if (screen->compute)
453 BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->tls);
454
455 flags = NOUVEAU_BO_GART | NOUVEAU_BO_WR;
456
457 BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->fence.bo);
458 BCTX_REFN_bo(nvc0->bufctx, FENCE, flags, screen->fence.bo);
459 if (screen->compute)
460 BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->fence.bo);
461
462 nvc0->base.scratch.bo_size = 2 << 20;
463
464 memset(nvc0->tex_handles, ~0, sizeof(nvc0->tex_handles));
465
466 util_dynarray_init(&nvc0->global_residents);
467
468 return pipe;
469
470 out_err:
471 if (nvc0) {
472 if (nvc0->bufctx_3d)
473 nouveau_bufctx_del(&nvc0->bufctx_3d);
474 if (nvc0->bufctx_cp)
475 nouveau_bufctx_del(&nvc0->bufctx_cp);
476 if (nvc0->bufctx)
477 nouveau_bufctx_del(&nvc0->bufctx);
478 FREE(nvc0->blit);
479 FREE(nvc0);
480 }
481 return NULL;
482 }
483
484 void
485 nvc0_bufctx_fence(struct nvc0_context *nvc0, struct nouveau_bufctx *bufctx,
486 bool on_flush)
487 {
488 struct nouveau_list *list = on_flush ? &bufctx->current : &bufctx->pending;
489 struct nouveau_list *it;
490 NOUVEAU_DRV_STAT_IFD(unsigned count = 0);
491
492 for (it = list->next; it != list; it = it->next) {
493 struct nouveau_bufref *ref = (struct nouveau_bufref *)it;
494 struct nv04_resource *res = ref->priv;
495 if (res)
496 nvc0_resource_validate(res, (unsigned)ref->priv_data);
497 NOUVEAU_DRV_STAT_IFD(count++);
498 }
499 NOUVEAU_DRV_STAT(&nvc0->screen->base, resource_validate_count, count);
500 }
501
502 static void
503 nvc0_context_get_sample_position(struct pipe_context *pipe,
504 unsigned sample_count, unsigned sample_index,
505 float *xy)
506 {
507 static const uint8_t ms1[1][2] = { { 0x8, 0x8 } };
508 static const uint8_t ms2[2][2] = {
509 { 0x4, 0x4 }, { 0xc, 0xc } }; /* surface coords (0,0), (1,0) */
510 static const uint8_t ms4[4][2] = {
511 { 0x6, 0x2 }, { 0xe, 0x6 }, /* (0,0), (1,0) */
512 { 0x2, 0xa }, { 0xa, 0xe } }; /* (0,1), (1,1) */
513 static const uint8_t ms8[8][2] = {
514 { 0x1, 0x7 }, { 0x5, 0x3 }, /* (0,0), (1,0) */
515 { 0x3, 0xd }, { 0x7, 0xb }, /* (0,1), (1,1) */
516 { 0x9, 0x5 }, { 0xf, 0x1 }, /* (2,0), (3,0) */
517 { 0xb, 0xf }, { 0xd, 0x9 } }; /* (2,1), (3,1) */
518 #if 0
519 /* NOTE: there are alternative modes for MS2 and MS8, currently not used */
520 static const uint8_t ms8_alt[8][2] = {
521 { 0x9, 0x5 }, { 0x7, 0xb }, /* (2,0), (1,1) */
522 { 0xd, 0x9 }, { 0x5, 0x3 }, /* (3,1), (1,0) */
523 { 0x3, 0xd }, { 0x1, 0x7 }, /* (0,1), (0,0) */
524 { 0xb, 0xf }, { 0xf, 0x1 } }; /* (2,1), (3,0) */
525 #endif
526
527 const uint8_t (*ptr)[2];
528
529 switch (sample_count) {
530 case 0:
531 case 1: ptr = ms1; break;
532 case 2: ptr = ms2; break;
533 case 4: ptr = ms4; break;
534 case 8: ptr = ms8; break;
535 default:
536 assert(0);
537 return; /* bad sample count -> undefined locations */
538 }
539 xy[0] = ptr[sample_index][0] * 0.0625f;
540 xy[1] = ptr[sample_index][1] * 0.0625f;
541 }