2 * Copyright 2010 Christoph Bumiller
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.
23 #include "pipe/p_defines.h"
24 #include "util/u_framebuffer.h"
26 #include "nvc0/nvc0_context.h"
27 #include "nvc0/nvc0_screen.h"
28 #include "nvc0/nvc0_resource.h"
31 nvc0_flush(struct pipe_context *pipe,
32 struct pipe_fence_handle **fence,
35 struct nvc0_context *nvc0 = nvc0_context(pipe);
36 struct nouveau_screen *screen = &nvc0->screen->base;
39 nouveau_fence_ref(screen->fence.current, (struct nouveau_fence **)fence);
41 pipe_mutex_lock(screen->push_mutex);
42 PUSH_KICK(nvc0->base.pushbuf); /* fencing handled in kick_notify */
43 pipe_mutex_unlock(screen->push_mutex);
45 nouveau_context_update_frame_stats(&nvc0->base);
49 nvc0_texture_barrier(struct pipe_context *pipe)
51 struct nouveau_pushbuf *push = nvc0_context(pipe)->base.pushbuf;
53 pipe_mutex_lock(nvc0_context(pipe)->screen->base.push_mutex);
54 IMMED_NVC0(push, NVC0_3D(SERIALIZE), 0);
55 IMMED_NVC0(push, NVC0_3D(TEX_CACHE_CTL), 0);
56 pipe_mutex_unlock(nvc0_context(pipe)->screen->base.push_mutex);
60 nvc0_memory_barrier(struct pipe_context *pipe, unsigned flags)
62 struct nvc0_context *nvc0 = nvc0_context(pipe);
63 struct nouveau_pushbuf *push = nvc0->base.pushbuf;
66 pipe_mutex_lock(nvc0_context(pipe)->screen->base.push_mutex);
68 if (flags & PIPE_BARRIER_MAPPED_BUFFER) {
69 for (i = 0; i < nvc0->num_vtxbufs; ++i) {
70 if (!nvc0->vtxbuf[i].buffer)
72 if (nvc0->vtxbuf[i].buffer->flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT)
73 nvc0->base.vbo_dirty = true;
76 if (nvc0->idxbuf.buffer &&
77 nvc0->idxbuf.buffer->flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT)
78 nvc0->base.vbo_dirty = true;
80 for (s = 0; s < 5 && !nvc0->cb_dirty; ++s) {
81 uint32_t valid = nvc0->constbuf_valid[s];
83 while (valid && !nvc0->cb_dirty) {
84 const unsigned i = ffs(valid) - 1;
85 struct pipe_resource *res;
88 if (nvc0->constbuf[s][i].user)
91 res = nvc0->constbuf[s][i].u.buf;
95 if (res->flags & PIPE_RESOURCE_FLAG_MAP_PERSISTENT)
96 nvc0->cb_dirty = true;
100 /* Pretty much any writing by shaders needs a serialize after
101 * it. Especially when moving between 3d and compute pipelines, but even
104 IMMED_NVC0(push, NVC0_3D(SERIALIZE), 0);
107 /* If we're going to texture from a buffer/image written by a shader, we
108 * must flush the texture cache.
110 if (flags & PIPE_BARRIER_TEXTURE)
111 IMMED_NVC0(push, NVC0_3D(TEX_CACHE_CTL), 0);
113 if (flags & PIPE_BARRIER_CONSTANT_BUFFER)
114 nvc0->cb_dirty = true;
115 if (flags & (PIPE_BARRIER_VERTEX_BUFFER | PIPE_BARRIER_INDEX_BUFFER))
116 nvc0->base.vbo_dirty = true;
118 pipe_mutex_unlock(nvc0_context(pipe)->screen->base.push_mutex);
122 nvc0_emit_string_marker(struct pipe_context *pipe, const char *str, int len)
124 struct nouveau_pushbuf *push = nvc0_context(pipe)->base.pushbuf;
125 int string_words = len / 4;
130 string_words = MIN2(string_words, NV04_PFIFO_MAX_PACKET_LEN);
131 if (string_words == NV04_PFIFO_MAX_PACKET_LEN)
132 data_words = string_words;
134 data_words = string_words + !!(len & 3);
135 pipe_mutex_lock(nvc0_context(pipe)->screen->base.push_mutex);
136 BEGIN_NIC0(push, SUBC_3D(NV04_GRAPH_NOP), data_words);
138 PUSH_DATAp(push, str, string_words);
139 if (string_words != data_words) {
141 memcpy(&data, &str[string_words * 4], len & 3);
142 PUSH_DATA (push, data);
144 pipe_mutex_unlock(nvc0_context(pipe)->screen->base.push_mutex);
148 nvc0_context_unreference_resources(struct nvc0_context *nvc0)
152 nouveau_bufctx_del(&nvc0->bufctx_3d);
153 nouveau_bufctx_del(&nvc0->bufctx);
154 nouveau_bufctx_del(&nvc0->bufctx_cp);
156 util_unreference_framebuffer_state(&nvc0->framebuffer);
158 for (i = 0; i < nvc0->num_vtxbufs; ++i)
159 pipe_resource_reference(&nvc0->vtxbuf[i].buffer, NULL);
161 pipe_resource_reference(&nvc0->idxbuf.buffer, NULL);
163 for (s = 0; s < 6; ++s) {
164 for (i = 0; i < nvc0->num_textures[s]; ++i)
165 pipe_sampler_view_reference(&nvc0->textures[s][i], NULL);
167 for (i = 0; i < NVC0_MAX_PIPE_CONSTBUFS; ++i)
168 if (!nvc0->constbuf[s][i].user)
169 pipe_resource_reference(&nvc0->constbuf[s][i].u.buf, NULL);
171 for (i = 0; i < NVC0_MAX_BUFFERS; ++i)
172 pipe_resource_reference(&nvc0->buffers[s][i].buffer, NULL);
174 for (i = 0; i < NVC0_MAX_IMAGES; ++i)
175 pipe_resource_reference(&nvc0->images[s][i].resource, NULL);
178 for (s = 0; s < 2; ++s) {
179 for (i = 0; i < NVC0_MAX_SURFACE_SLOTS; ++i)
180 pipe_surface_reference(&nvc0->surfaces[s][i], NULL);
183 for (i = 0; i < nvc0->num_tfbbufs; ++i)
184 pipe_so_target_reference(&nvc0->tfbbuf[i], NULL);
186 for (i = 0; i < nvc0->global_residents.size / sizeof(struct pipe_resource *);
188 struct pipe_resource **res = util_dynarray_element(
189 &nvc0->global_residents, struct pipe_resource *, i);
190 pipe_resource_reference(res, NULL);
192 util_dynarray_fini(&nvc0->global_residents);
195 nvc0->base.pipe.delete_tcs_state(&nvc0->base.pipe, nvc0->tcp_empty);
199 nvc0_destroy(struct pipe_context *pipe)
201 struct nvc0_context *nvc0 = nvc0_context(pipe);
203 if (nvc0->screen->cur_ctx == nvc0) {
204 nvc0->screen->cur_ctx = NULL;
205 nvc0->screen->save_state = nvc0->state;
206 nvc0->screen->save_state.tfb = NULL;
209 /* Unset bufctx, we don't want to revalidate any resources after the flush.
210 * Other contexts will always set their bufctx again on action calls.
212 nouveau_pushbuf_bufctx(nvc0->base.pushbuf, NULL);
213 nouveau_pushbuf_kick(nvc0->base.pushbuf, nvc0->base.pushbuf->channel);
215 nvc0_context_unreference_resources(nvc0);
216 nvc0_blitctx_destroy(nvc0);
218 nouveau_context_destroy(&nvc0->base);
222 nvc0_default_kick_notify(struct nouveau_pushbuf *push)
224 struct nvc0_screen *screen = push->user_priv;
227 nouveau_fence_next(&screen->base);
228 nouveau_fence_update(&screen->base, true);
230 screen->cur_ctx->state.flushed = true;
231 NOUVEAU_DRV_STAT(&screen->base, pushbuf_count, 1);
236 nvc0_invalidate_resource_storage(struct nouveau_context *ctx,
237 struct pipe_resource *res,
240 struct nvc0_context *nvc0 = nvc0_context(&ctx->pipe);
243 if (res->bind & PIPE_BIND_RENDER_TARGET) {
244 for (i = 0; i < nvc0->framebuffer.nr_cbufs; ++i) {
245 if (nvc0->framebuffer.cbufs[i] &&
246 nvc0->framebuffer.cbufs[i]->texture == res) {
247 nvc0->dirty_3d |= NVC0_NEW_3D_FRAMEBUFFER;
248 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_FB);
254 if (res->bind & PIPE_BIND_DEPTH_STENCIL) {
255 if (nvc0->framebuffer.zsbuf &&
256 nvc0->framebuffer.zsbuf->texture == res) {
257 nvc0->dirty_3d |= NVC0_NEW_3D_FRAMEBUFFER;
258 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_FB);
264 if (res->target == PIPE_BUFFER) {
265 for (i = 0; i < nvc0->num_vtxbufs; ++i) {
266 if (nvc0->vtxbuf[i].buffer == res) {
267 nvc0->dirty_3d |= NVC0_NEW_3D_ARRAYS;
268 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_VTX);
274 if (nvc0->idxbuf.buffer == res) {
275 nvc0->dirty_3d |= NVC0_NEW_3D_IDXBUF;
276 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_IDX);
281 for (s = 0; s < 6; ++s) {
282 for (i = 0; i < nvc0->num_textures[s]; ++i) {
283 if (nvc0->textures[s][i] &&
284 nvc0->textures[s][i]->texture == res) {
285 nvc0->textures_dirty[s] |= 1 << i;
286 if (unlikely(s == 5)) {
287 nvc0->dirty_cp |= NVC0_NEW_CP_TEXTURES;
288 nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_TEX(i));
290 nvc0->dirty_3d |= NVC0_NEW_3D_TEXTURES;
291 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_TEX(s, i));
299 for (s = 0; s < 6; ++s) {
300 for (i = 0; i < NVC0_MAX_PIPE_CONSTBUFS; ++i) {
301 if (!(nvc0->constbuf_valid[s] & (1 << i)))
303 if (!nvc0->constbuf[s][i].user &&
304 nvc0->constbuf[s][i].u.buf == res) {
305 nvc0->constbuf_dirty[s] |= 1 << i;
306 if (unlikely(s == 5)) {
307 nvc0->dirty_cp |= NVC0_NEW_CP_CONSTBUF;
308 nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_CB(i));
310 nvc0->dirty_3d |= NVC0_NEW_3D_CONSTBUF;
311 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_CB(s, i));
319 for (s = 0; s < 6; ++s) {
320 for (i = 0; i < NVC0_MAX_BUFFERS; ++i) {
321 if (nvc0->buffers[s][i].buffer == res) {
322 nvc0->buffers_dirty[s] |= 1 << i;
323 if (unlikely(s == 5)) {
324 nvc0->dirty_cp |= NVC0_NEW_CP_BUFFERS;
325 nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_BUF);
327 nvc0->dirty_3d |= NVC0_NEW_3D_BUFFERS;
328 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_BUF);
336 for (s = 0; s < 6; ++s) {
337 for (i = 0; i < NVC0_MAX_IMAGES; ++i) {
338 if (nvc0->images[s][i].resource == res) {
339 nvc0->images_dirty[s] |= 1 << i;
340 if (unlikely(s == 5)) {
341 nvc0->dirty_cp |= NVC0_NEW_CP_SURFACES;
342 nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_SUF);
344 nvc0->dirty_3d |= NVC0_NEW_3D_SURFACES;
345 nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_SUF);
358 nvc0_context_get_sample_position(struct pipe_context *, unsigned, unsigned,
361 struct pipe_context *
362 nvc0_create(struct pipe_screen *pscreen, void *priv, unsigned ctxflags)
364 struct nvc0_screen *screen = nvc0_screen(pscreen);
365 struct nvc0_context *nvc0;
366 struct pipe_context *pipe;
370 nvc0 = CALLOC_STRUCT(nvc0_context);
373 pipe = &nvc0->base.pipe;
375 pipe_mutex_lock(screen->base.push_mutex);
377 if (!nvc0_blitctx_create(nvc0))
380 nvc0->base.pushbuf = screen->base.pushbuf;
381 nvc0->base.client = screen->base.client;
383 ret = nouveau_bufctx_new(screen->base.client, 2, &nvc0->bufctx);
385 ret = nouveau_bufctx_new(screen->base.client, NVC0_BIND_3D_COUNT,
388 ret = nouveau_bufctx_new(screen->base.client, NVC0_BIND_CP_COUNT,
393 nvc0->screen = screen;
394 nvc0->base.screen = &screen->base;
396 pipe->screen = pscreen;
399 pipe->destroy = nvc0_destroy;
401 pipe->draw_vbo = nvc0_draw_vbo;
402 pipe->clear = nvc0_clear;
403 pipe->launch_grid = (nvc0->screen->base.class_3d >= NVE4_3D_CLASS) ?
404 nve4_launch_grid : nvc0_launch_grid;
406 pipe->flush = nvc0_flush;
407 pipe->texture_barrier = nvc0_texture_barrier;
408 pipe->memory_barrier = nvc0_memory_barrier;
409 pipe->get_sample_position = nvc0_context_get_sample_position;
410 pipe->emit_string_marker = nvc0_emit_string_marker;
412 nouveau_context_init(&nvc0->base);
413 nvc0_init_query_functions(nvc0);
414 nvc0_init_surface_functions(nvc0);
415 nvc0_init_state_functions(nvc0);
416 nvc0_init_transfer_functions(nvc0);
417 nvc0_init_resource_functions(pipe);
419 nvc0->base.invalidate_resource_storage = nvc0_invalidate_resource_storage;
421 pipe->create_video_codec = nvc0_create_decoder;
422 pipe->create_video_buffer = nvc0_video_buffer_create;
424 /* shader builtin library is per-screen, but we need a context for m2mf */
425 nvc0_program_library_upload(nvc0);
426 nvc0_program_init_tcp_empty(nvc0);
427 if (!nvc0->tcp_empty)
429 /* set the empty tctl prog on next draw in case one is never set */
430 nvc0->dirty_3d |= NVC0_NEW_3D_TCTLPROG;
432 /* Do not bind the COMPUTE driver constbuf at screen initialization because
433 * CBs are aliased between 3D and COMPUTE, but make sure it will be bound if
434 * a grid is launched later. */
435 nvc0->dirty_cp |= NVC0_NEW_CP_DRIVERCONST;
437 /* now that there are no more opportunities for errors, set the current
438 * context if there isn't already one.
440 if (!screen->cur_ctx) {
441 nvc0->state = screen->save_state;
442 screen->cur_ctx = nvc0;
443 nouveau_pushbuf_bufctx(screen->base.pushbuf, nvc0->bufctx);
445 screen->base.pushbuf->kick_notify = nvc0_default_kick_notify;
447 /* add permanently resident buffers to bufctxts */
449 flags = NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RD;
451 BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->text);
452 BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->uniform_bo);
453 BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->txc);
454 if (screen->compute) {
455 BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->text);
456 BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->uniform_bo);
457 BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->txc);
460 flags = NV_VRAM_DOMAIN(&screen->base) | NOUVEAU_BO_RDWR;
462 if (screen->poly_cache)
463 BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->poly_cache);
465 BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->tls);
467 flags = NOUVEAU_BO_GART | NOUVEAU_BO_WR;
469 BCTX_REFN_bo(nvc0->bufctx_3d, 3D_SCREEN, flags, screen->fence.bo);
470 BCTX_REFN_bo(nvc0->bufctx, FENCE, flags, screen->fence.bo);
472 BCTX_REFN_bo(nvc0->bufctx_cp, CP_SCREEN, flags, screen->fence.bo);
474 nvc0->base.scratch.bo_size = 2 << 20;
476 memset(nvc0->tex_handles, ~0, sizeof(nvc0->tex_handles));
478 util_dynarray_init(&nvc0->global_residents);
480 pipe_mutex_unlock(screen->base.push_mutex);
485 pipe_mutex_unlock(screen->base.push_mutex);
488 nouveau_bufctx_del(&nvc0->bufctx_3d);
490 nouveau_bufctx_del(&nvc0->bufctx_cp);
492 nouveau_bufctx_del(&nvc0->bufctx);
500 nvc0_bufctx_fence(struct nvc0_context *nvc0, struct nouveau_bufctx *bufctx,
503 struct nouveau_list *list = on_flush ? &bufctx->current : &bufctx->pending;
504 struct nouveau_list *it;
505 NOUVEAU_DRV_STAT_IFD(unsigned count = 0);
507 for (it = list->next; it != list; it = it->next) {
508 struct nouveau_bufref *ref = (struct nouveau_bufref *)it;
509 struct nv04_resource *res = ref->priv;
511 nvc0_resource_validate(res, (unsigned)ref->priv_data);
512 NOUVEAU_DRV_STAT_IFD(count++);
514 NOUVEAU_DRV_STAT(&nvc0->screen->base, resource_validate_count, count);
518 nvc0_get_sample_locations(unsigned sample_count)
520 static const uint8_t ms1[1][2] = { { 0x8, 0x8 } };
521 static const uint8_t ms2[2][2] = {
522 { 0x4, 0x4 }, { 0xc, 0xc } }; /* surface coords (0,0), (1,0) */
523 static const uint8_t ms4[4][2] = {
524 { 0x6, 0x2 }, { 0xe, 0x6 }, /* (0,0), (1,0) */
525 { 0x2, 0xa }, { 0xa, 0xe } }; /* (0,1), (1,1) */
526 static const uint8_t ms8[8][2] = {
527 { 0x1, 0x7 }, { 0x5, 0x3 }, /* (0,0), (1,0) */
528 { 0x3, 0xd }, { 0x7, 0xb }, /* (0,1), (1,1) */
529 { 0x9, 0x5 }, { 0xf, 0x1 }, /* (2,0), (3,0) */
530 { 0xb, 0xf }, { 0xd, 0x9 } }; /* (2,1), (3,1) */
532 /* NOTE: there are alternative modes for MS2 and MS8, currently not used */
533 static const uint8_t ms8_alt[8][2] = {
534 { 0x9, 0x5 }, { 0x7, 0xb }, /* (2,0), (1,1) */
535 { 0xd, 0x9 }, { 0x5, 0x3 }, /* (3,1), (1,0) */
536 { 0x3, 0xd }, { 0x1, 0x7 }, /* (0,1), (0,0) */
537 { 0xb, 0xf }, { 0xf, 0x1 } }; /* (2,1), (3,0) */
540 const uint8_t (*ptr)[2];
542 switch (sample_count) {
544 case 1: ptr = ms1; break;
545 case 2: ptr = ms2; break;
546 case 4: ptr = ms4; break;
547 case 8: ptr = ms8; break;
550 return NULL; /* bad sample count -> undefined locations */
556 nvc0_context_get_sample_position(struct pipe_context *pipe,
557 unsigned sample_count, unsigned sample_index,
560 const uint8_t (*ptr)[2];
562 ptr = nvc0_get_sample_locations(sample_count);
566 xy[0] = ptr[sample_index][0] * 0.0625f;
567 xy[1] = ptr[sample_index][1] * 0.0625f;