Merge remote-tracking branch 'mesa-public/master' into vulkan
[mesa.git] / src / gallium / drivers / vc4 / vc4_resource.c
1 /*
2 * Copyright © 2014 Broadcom
3 * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
22 * IN THE SOFTWARE.
23 */
24
25 #include "util/u_blit.h"
26 #include "util/u_memory.h"
27 #include "util/u_format.h"
28 #include "util/u_inlines.h"
29 #include "util/u_surface.h"
30 #include "util/u_upload_mgr.h"
31
32 #include "vc4_screen.h"
33 #include "vc4_context.h"
34 #include "vc4_resource.h"
35 #include "vc4_tiling.h"
36
37 static bool miptree_debug = false;
38
39 static bool
40 vc4_resource_bo_alloc(struct vc4_resource *rsc)
41 {
42 struct pipe_resource *prsc = &rsc->base.b;
43 struct pipe_screen *pscreen = prsc->screen;
44 struct vc4_bo *bo;
45
46 if (miptree_debug) {
47 fprintf(stderr, "alloc %p: size %d + offset %d -> %d\n",
48 rsc,
49 rsc->slices[0].size,
50 rsc->slices[0].offset,
51 rsc->slices[0].offset +
52 rsc->slices[0].size +
53 rsc->cube_map_stride * (prsc->array_size - 1));
54 }
55
56 bo = vc4_bo_alloc(vc4_screen(pscreen),
57 rsc->slices[0].offset +
58 rsc->slices[0].size +
59 rsc->cube_map_stride * (prsc->array_size - 1),
60 "resource");
61 if (bo) {
62 vc4_bo_unreference(&rsc->bo);
63 rsc->bo = bo;
64 return true;
65 } else {
66 return false;
67 }
68 }
69
70 static void
71 vc4_resource_transfer_unmap(struct pipe_context *pctx,
72 struct pipe_transfer *ptrans)
73 {
74 struct vc4_context *vc4 = vc4_context(pctx);
75 struct vc4_transfer *trans = vc4_transfer(ptrans);
76
77 if (trans->map) {
78 struct vc4_resource *rsc;
79 struct vc4_resource_slice *slice;
80 if (trans->ss_resource) {
81 rsc = vc4_resource(trans->ss_resource);
82 slice = &rsc->slices[0];
83 } else {
84 rsc = vc4_resource(ptrans->resource);
85 slice = &rsc->slices[ptrans->level];
86 }
87
88 if (ptrans->usage & PIPE_TRANSFER_WRITE) {
89 vc4_store_tiled_image(rsc->bo->map + slice->offset +
90 ptrans->box.z * rsc->cube_map_stride,
91 slice->stride,
92 trans->map, ptrans->stride,
93 slice->tiling, rsc->cpp,
94 &ptrans->box);
95 }
96 free(trans->map);
97 }
98
99 if (trans->ss_resource && (ptrans->usage & PIPE_TRANSFER_WRITE)) {
100 struct pipe_blit_info blit;
101 memset(&blit, 0, sizeof(blit));
102
103 blit.src.resource = trans->ss_resource;
104 blit.src.format = trans->ss_resource->format;
105 blit.src.box.width = trans->ss_box.width;
106 blit.src.box.height = trans->ss_box.height;
107 blit.src.box.depth = 1;
108
109 blit.dst.resource = ptrans->resource;
110 blit.dst.format = ptrans->resource->format;
111 blit.dst.level = ptrans->level;
112 blit.dst.box = trans->ss_box;
113
114 blit.mask = util_format_get_mask(ptrans->resource->format);
115 blit.filter = PIPE_TEX_FILTER_NEAREST;
116
117 pctx->blit(pctx, &blit);
118 vc4_flush(pctx);
119
120 pipe_resource_reference(&trans->ss_resource, NULL);
121 }
122
123 pipe_resource_reference(&ptrans->resource, NULL);
124 util_slab_free(&vc4->transfer_pool, ptrans);
125 }
126
127 static struct pipe_resource *
128 vc4_get_temp_resource(struct pipe_context *pctx,
129 struct pipe_resource *prsc,
130 const struct pipe_box *box)
131 {
132 struct pipe_resource temp_setup;
133
134 memset(&temp_setup, 0, sizeof(temp_setup));
135 temp_setup.target = prsc->target;
136 temp_setup.format = prsc->format;
137 temp_setup.width0 = box->width;
138 temp_setup.height0 = box->height;
139 temp_setup.depth0 = 1;
140 temp_setup.array_size = 1;
141
142 return pctx->screen->resource_create(pctx->screen, &temp_setup);
143 }
144
145 static void *
146 vc4_resource_transfer_map(struct pipe_context *pctx,
147 struct pipe_resource *prsc,
148 unsigned level, unsigned usage,
149 const struct pipe_box *box,
150 struct pipe_transfer **pptrans)
151 {
152 struct vc4_context *vc4 = vc4_context(pctx);
153 struct vc4_resource *rsc = vc4_resource(prsc);
154 struct vc4_transfer *trans;
155 struct pipe_transfer *ptrans;
156 enum pipe_format format = prsc->format;
157 char *buf;
158
159 if (usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE) {
160 if (vc4_resource_bo_alloc(rsc)) {
161
162 /* If it might be bound as one of our vertex buffers,
163 * make sure we re-emit vertex buffer state.
164 */
165 if (prsc->bind & PIPE_BIND_VERTEX_BUFFER)
166 vc4->dirty |= VC4_DIRTY_VTXBUF;
167 } else {
168 /* If we failed to reallocate, flush everything so
169 * that we don't violate any syncing requirements.
170 */
171 vc4_flush(pctx);
172 }
173 } else if (!(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
174 if (vc4_cl_references_bo(pctx, rsc->bo)) {
175 if ((usage & PIPE_TRANSFER_DISCARD_RANGE) &&
176 prsc->last_level == 0 &&
177 prsc->width0 == box->width &&
178 prsc->height0 == box->height &&
179 prsc->depth0 == box->depth &&
180 vc4_resource_bo_alloc(rsc)) {
181 if (prsc->bind & PIPE_BIND_VERTEX_BUFFER)
182 vc4->dirty |= VC4_DIRTY_VTXBUF;
183 } else {
184 vc4_flush(pctx);
185 }
186 }
187 }
188
189 if (usage & PIPE_TRANSFER_WRITE)
190 rsc->writes++;
191
192 trans = util_slab_alloc(&vc4->transfer_pool);
193 if (!trans)
194 return NULL;
195
196 /* XXX: Handle DONTBLOCK, DISCARD_RANGE, PERSISTENT, COHERENT. */
197
198 /* util_slab_alloc() doesn't zero: */
199 memset(trans, 0, sizeof(*trans));
200 ptrans = &trans->base;
201
202 pipe_resource_reference(&ptrans->resource, prsc);
203 ptrans->level = level;
204 ptrans->usage = usage;
205 ptrans->box = *box;
206
207 /* If the resource is multisampled, we need to resolve to single
208 * sample. This seems like it should be handled at a higher layer.
209 */
210 if (prsc->nr_samples > 1) {
211 trans->ss_resource = vc4_get_temp_resource(pctx, prsc, box);
212 if (!trans->ss_resource)
213 goto fail;
214 assert(!trans->ss_resource->nr_samples);
215
216 /* The ptrans->box gets modified for tile alignment, so save
217 * the original box for unmap time.
218 */
219 trans->ss_box = *box;
220
221 if (usage & PIPE_TRANSFER_READ) {
222 struct pipe_blit_info blit;
223 memset(&blit, 0, sizeof(blit));
224
225 blit.src.resource = ptrans->resource;
226 blit.src.format = ptrans->resource->format;
227 blit.src.level = ptrans->level;
228 blit.src.box = trans->ss_box;
229
230 blit.dst.resource = trans->ss_resource;
231 blit.dst.format = trans->ss_resource->format;
232 blit.dst.box.width = trans->ss_box.width;
233 blit.dst.box.height = trans->ss_box.height;
234 blit.dst.box.depth = 1;
235
236 blit.mask = util_format_get_mask(prsc->format);
237 blit.filter = PIPE_TEX_FILTER_NEAREST;
238
239 pctx->blit(pctx, &blit);
240 vc4_flush(pctx);
241 }
242
243 /* The rest of the mapping process should use our temporary. */
244 prsc = trans->ss_resource;
245 rsc = vc4_resource(prsc);
246 ptrans->box.x = 0;
247 ptrans->box.y = 0;
248 ptrans->box.z = 0;
249 }
250
251 /* Note that the current kernel implementation is synchronous, so no
252 * need to do syncing stuff here yet.
253 */
254
255 if (usage & PIPE_TRANSFER_UNSYNCHRONIZED)
256 buf = vc4_bo_map_unsynchronized(rsc->bo);
257 else
258 buf = vc4_bo_map(rsc->bo);
259 if (!buf) {
260 fprintf(stderr, "Failed to map bo\n");
261 goto fail;
262 }
263
264 *pptrans = ptrans;
265
266 struct vc4_resource_slice *slice = &rsc->slices[level];
267 if (rsc->tiled) {
268 uint32_t utile_w = vc4_utile_width(rsc->cpp);
269 uint32_t utile_h = vc4_utile_height(rsc->cpp);
270
271 /* No direct mappings of tiled, since we need to manually
272 * tile/untile.
273 */
274 if (usage & PIPE_TRANSFER_MAP_DIRECTLY)
275 return NULL;
276
277 /* We need to align the box to utile boundaries, since that's
278 * what load/store operate on.
279 */
280 uint32_t orig_width = ptrans->box.width;
281 uint32_t orig_height = ptrans->box.height;
282 uint32_t box_start_x = ptrans->box.x & (utile_w - 1);
283 uint32_t box_start_y = ptrans->box.y & (utile_h - 1);
284 ptrans->box.width += box_start_x;
285 ptrans->box.x -= box_start_x;
286 ptrans->box.height += box_start_y;
287 ptrans->box.y -= box_start_y;
288 ptrans->box.width = align(ptrans->box.width, utile_w);
289 ptrans->box.height = align(ptrans->box.height, utile_h);
290
291 ptrans->stride = ptrans->box.width * rsc->cpp;
292 ptrans->layer_stride = ptrans->stride;
293
294 trans->map = malloc(ptrans->stride * ptrans->box.height);
295 if (usage & PIPE_TRANSFER_READ ||
296 ptrans->box.width != orig_width ||
297 ptrans->box.height != orig_height) {
298 vc4_load_tiled_image(trans->map, ptrans->stride,
299 buf + slice->offset +
300 ptrans->box.z * rsc->cube_map_stride,
301 slice->stride,
302 slice->tiling, rsc->cpp,
303 &ptrans->box);
304 }
305 return (trans->map +
306 box_start_x * rsc->cpp +
307 box_start_y * ptrans->stride);
308 } else {
309 ptrans->stride = slice->stride;
310 ptrans->layer_stride = ptrans->stride;
311
312 return buf + slice->offset +
313 ptrans->box.y / util_format_get_blockheight(format) * ptrans->stride +
314 ptrans->box.x / util_format_get_blockwidth(format) * rsc->cpp +
315 ptrans->box.z * rsc->cube_map_stride;
316 }
317
318
319 fail:
320 vc4_resource_transfer_unmap(pctx, ptrans);
321 return NULL;
322 }
323
324 static void
325 vc4_resource_destroy(struct pipe_screen *pscreen,
326 struct pipe_resource *prsc)
327 {
328 struct vc4_resource *rsc = vc4_resource(prsc);
329 pipe_resource_reference(&rsc->shadow_parent, NULL);
330 vc4_bo_unreference(&rsc->bo);
331 free(rsc);
332 }
333
334 static boolean
335 vc4_resource_get_handle(struct pipe_screen *pscreen,
336 struct pipe_resource *prsc,
337 struct winsys_handle *handle)
338 {
339 struct vc4_resource *rsc = vc4_resource(prsc);
340
341 return vc4_screen_bo_get_handle(pscreen, rsc->bo, rsc->slices[0].stride,
342 handle);
343 }
344
345 static const struct u_resource_vtbl vc4_resource_vtbl = {
346 .resource_get_handle = vc4_resource_get_handle,
347 .resource_destroy = vc4_resource_destroy,
348 .transfer_map = vc4_resource_transfer_map,
349 .transfer_flush_region = u_default_transfer_flush_region,
350 .transfer_unmap = vc4_resource_transfer_unmap,
351 .transfer_inline_write = u_default_transfer_inline_write,
352 };
353
354 static void
355 vc4_setup_slices(struct vc4_resource *rsc)
356 {
357 struct pipe_resource *prsc = &rsc->base.b;
358 uint32_t width = prsc->width0;
359 uint32_t height = prsc->height0;
360 uint32_t pot_width = util_next_power_of_two(width);
361 uint32_t pot_height = util_next_power_of_two(height);
362 uint32_t offset = 0;
363 uint32_t utile_w = vc4_utile_width(rsc->cpp);
364 uint32_t utile_h = vc4_utile_height(rsc->cpp);
365
366 for (int i = prsc->last_level; i >= 0; i--) {
367 struct vc4_resource_slice *slice = &rsc->slices[i];
368
369 uint32_t level_width, level_height;
370 if (i == 0) {
371 level_width = width;
372 level_height = height;
373 } else {
374 level_width = u_minify(pot_width, i);
375 level_height = u_minify(pot_height, i);
376 }
377
378 if (!rsc->tiled) {
379 slice->tiling = VC4_TILING_FORMAT_LINEAR;
380 if (prsc->nr_samples > 1) {
381 /* MSAA (4x) surfaces are stored as raw tile buffer contents. */
382 level_width = align(level_width, 32);
383 level_height = align(level_height, 32);
384 } else {
385 level_width = align(level_width, utile_w);
386 }
387 } else {
388 if (vc4_size_is_lt(level_width, level_height,
389 rsc->cpp)) {
390 slice->tiling = VC4_TILING_FORMAT_LT;
391 level_width = align(level_width, utile_w);
392 level_height = align(level_height, utile_h);
393 } else {
394 slice->tiling = VC4_TILING_FORMAT_T;
395 level_width = align(level_width,
396 4 * 2 * utile_w);
397 level_height = align(level_height,
398 4 * 2 * utile_h);
399 }
400 }
401
402 slice->offset = offset;
403 slice->stride = (level_width * rsc->cpp *
404 MAX2(prsc->nr_samples, 1));
405 slice->size = level_height * slice->stride;
406
407 offset += slice->size;
408
409 if (miptree_debug) {
410 static const char tiling_chars[] = {
411 [VC4_TILING_FORMAT_LINEAR] = 'R',
412 [VC4_TILING_FORMAT_LT] = 'L',
413 [VC4_TILING_FORMAT_T] = 'T'
414 };
415 fprintf(stderr,
416 "rsc setup %p (format %d), %dx%d: "
417 "level %d (%c) -> %dx%d, stride %d@0x%08x\n",
418 rsc, rsc->vc4_format,
419 prsc->width0, prsc->height0,
420 i, tiling_chars[slice->tiling],
421 level_width, level_height,
422 slice->stride, slice->offset);
423 }
424 }
425
426 /* The texture base pointer that has to point to level 0 doesn't have
427 * intra-page bits, so we have to align it, and thus shift up all the
428 * smaller slices.
429 */
430 uint32_t page_align_offset = (align(rsc->slices[0].offset, 4096) -
431 rsc->slices[0].offset);
432 if (page_align_offset) {
433 for (int i = 0; i <= prsc->last_level; i++)
434 rsc->slices[i].offset += page_align_offset;
435 }
436
437 /* Cube map faces appear as whole miptrees at a page-aligned offset
438 * from the first face's miptree.
439 */
440 if (prsc->target == PIPE_TEXTURE_CUBE) {
441 rsc->cube_map_stride = align(rsc->slices[0].offset +
442 rsc->slices[0].size, 4096);
443 }
444 }
445
446 static struct vc4_resource *
447 vc4_resource_setup(struct pipe_screen *pscreen,
448 const struct pipe_resource *tmpl)
449 {
450 struct vc4_resource *rsc = CALLOC_STRUCT(vc4_resource);
451 if (!rsc)
452 return NULL;
453 struct pipe_resource *prsc = &rsc->base.b;
454
455 *prsc = *tmpl;
456
457 pipe_reference_init(&prsc->reference, 1);
458 prsc->screen = pscreen;
459
460 rsc->base.vtbl = &vc4_resource_vtbl;
461 if (prsc->nr_samples <= 1)
462 rsc->cpp = util_format_get_blocksize(tmpl->format);
463 else
464 rsc->cpp = sizeof(uint32_t);
465
466 assert(rsc->cpp);
467
468 return rsc;
469 }
470
471 static enum vc4_texture_data_type
472 get_resource_texture_format(struct pipe_resource *prsc)
473 {
474 struct vc4_resource *rsc = vc4_resource(prsc);
475 uint8_t format = vc4_get_tex_format(prsc->format);
476
477 if (!rsc->tiled) {
478 if (prsc->nr_samples > 1) {
479 return ~0;
480 } else {
481 assert(format == VC4_TEXTURE_TYPE_RGBA8888);
482 return VC4_TEXTURE_TYPE_RGBA32R;
483 }
484 }
485
486 return format;
487 }
488
489 struct pipe_resource *
490 vc4_resource_create(struct pipe_screen *pscreen,
491 const struct pipe_resource *tmpl)
492 {
493 struct vc4_resource *rsc = vc4_resource_setup(pscreen, tmpl);
494 struct pipe_resource *prsc = &rsc->base.b;
495
496 /* We have to make shared be untiled, since we don't have any way to
497 * communicate metadata about tiling currently.
498 */
499 if (tmpl->target == PIPE_BUFFER ||
500 tmpl->nr_samples > 1 ||
501 (tmpl->bind & (PIPE_BIND_SCANOUT |
502 PIPE_BIND_LINEAR |
503 PIPE_BIND_SHARED |
504 PIPE_BIND_CURSOR))) {
505 rsc->tiled = false;
506 } else {
507 rsc->tiled = true;
508 }
509
510 if (tmpl->target != PIPE_BUFFER)
511 rsc->vc4_format = get_resource_texture_format(prsc);
512
513 vc4_setup_slices(rsc);
514 if (!vc4_resource_bo_alloc(rsc))
515 goto fail;
516
517 return prsc;
518 fail:
519 vc4_resource_destroy(pscreen, prsc);
520 return NULL;
521 }
522
523 static struct pipe_resource *
524 vc4_resource_from_handle(struct pipe_screen *pscreen,
525 const struct pipe_resource *tmpl,
526 struct winsys_handle *handle)
527 {
528 struct vc4_resource *rsc = vc4_resource_setup(pscreen, tmpl);
529 struct pipe_resource *prsc = &rsc->base.b;
530 struct vc4_resource_slice *slice = &rsc->slices[0];
531
532 if (!rsc)
533 return NULL;
534
535 rsc->tiled = false;
536 rsc->bo = vc4_screen_bo_from_handle(pscreen, handle);
537 if (!rsc->bo)
538 goto fail;
539
540 if (!using_vc4_simulator)
541 slice->stride = handle->stride;
542 else
543 slice->stride = align(prsc->width0 * rsc->cpp, 16);
544
545 slice->tiling = VC4_TILING_FORMAT_LINEAR;
546
547 rsc->vc4_format = get_resource_texture_format(prsc);
548
549 if (miptree_debug) {
550 fprintf(stderr,
551 "rsc import %p (format %d), %dx%d: "
552 "level 0 (R) -> stride %d@0x%08x\n",
553 rsc, rsc->vc4_format,
554 prsc->width0, prsc->height0,
555 slice->stride, slice->offset);
556 }
557
558 return prsc;
559
560 fail:
561 vc4_resource_destroy(pscreen, prsc);
562 return NULL;
563 }
564
565 static struct pipe_surface *
566 vc4_create_surface(struct pipe_context *pctx,
567 struct pipe_resource *ptex,
568 const struct pipe_surface *surf_tmpl)
569 {
570 struct vc4_surface *surface = CALLOC_STRUCT(vc4_surface);
571 struct vc4_resource *rsc = vc4_resource(ptex);
572
573 if (!surface)
574 return NULL;
575
576 assert(surf_tmpl->u.tex.first_layer == surf_tmpl->u.tex.last_layer);
577
578 struct pipe_surface *psurf = &surface->base;
579 unsigned level = surf_tmpl->u.tex.level;
580
581 pipe_reference_init(&psurf->reference, 1);
582 pipe_resource_reference(&psurf->texture, ptex);
583
584 psurf->context = pctx;
585 psurf->format = surf_tmpl->format;
586 psurf->width = u_minify(ptex->width0, level);
587 psurf->height = u_minify(ptex->height0, level);
588 psurf->u.tex.level = level;
589 psurf->u.tex.first_layer = surf_tmpl->u.tex.first_layer;
590 psurf->u.tex.last_layer = surf_tmpl->u.tex.last_layer;
591 surface->offset = rsc->slices[level].offset;
592 surface->tiling = rsc->slices[level].tiling;
593
594 return &surface->base;
595 }
596
597 static void
598 vc4_surface_destroy(struct pipe_context *pctx, struct pipe_surface *psurf)
599 {
600 pipe_resource_reference(&psurf->texture, NULL);
601 FREE(psurf);
602 }
603
604 static void
605 vc4_dump_surface_non_msaa(struct pipe_surface *psurf)
606 {
607 struct pipe_resource *prsc = psurf->texture;
608 struct vc4_resource *rsc = vc4_resource(prsc);
609 uint32_t *map = vc4_bo_map(rsc->bo);
610 uint32_t stride = rsc->slices[0].stride / 4;
611 uint32_t width = psurf->width;
612 uint32_t height = psurf->height;
613 uint32_t chunk_w = width / 79;
614 uint32_t chunk_h = height / 40;
615 uint32_t found_colors[10];
616 uint32_t num_found_colors = 0;
617
618 if (rsc->vc4_format != VC4_TEXTURE_TYPE_RGBA32R) {
619 fprintf(stderr, "%s: Unsupported format %s\n",
620 __func__, util_format_short_name(psurf->format));
621 return;
622 }
623
624 for (int by = 0; by < height; by += chunk_h) {
625 for (int bx = 0; bx < width; bx += chunk_w) {
626 int all_found_color = -1; /* nothing found */
627
628 for (int y = by; y < MIN2(height, by + chunk_h); y++) {
629 for (int x = bx; x < MIN2(width, bx + chunk_w); x++) {
630 uint32_t pix = map[y * stride + x];
631
632 int i;
633 for (i = 0; i < num_found_colors; i++) {
634 if (pix == found_colors[i])
635 break;
636 }
637 if (i == num_found_colors &&
638 num_found_colors <
639 ARRAY_SIZE(found_colors)) {
640 found_colors[num_found_colors++] = pix;
641 }
642
643 if (i < num_found_colors) {
644 if (all_found_color == -1)
645 all_found_color = i;
646 else if (i != all_found_color)
647 all_found_color = ARRAY_SIZE(found_colors);
648 }
649 }
650 }
651 /* If all pixels for this chunk have a consistent
652 * value, then print a character for it. Either a
653 * fixed name (particularly common for piglit tests),
654 * or a runtime-generated number.
655 */
656 if (all_found_color >= 0 &&
657 all_found_color < ARRAY_SIZE(found_colors)) {
658 static const struct {
659 uint32_t val;
660 const char *c;
661 } named_colors[] = {
662 { 0xff000000, "█" },
663 { 0x00000000, "█" },
664 { 0xffff0000, "r" },
665 { 0xff00ff00, "g" },
666 { 0xff0000ff, "b" },
667 { 0xffffffff, "w" },
668 };
669 int i;
670 for (i = 0; i < ARRAY_SIZE(named_colors); i++) {
671 if (named_colors[i].val ==
672 found_colors[all_found_color]) {
673 fprintf(stderr, "%s",
674 named_colors[i].c);
675 break;
676 }
677 }
678 /* For unnamed colors, print a number and the
679 * numbers will have values printed at the
680 * end.
681 */
682 if (i == ARRAY_SIZE(named_colors)) {
683 fprintf(stderr, "%c",
684 '0' + all_found_color);
685 }
686 } else {
687 /* If there's no consistent color, print this.
688 */
689 fprintf(stderr, ".");
690 }
691 }
692 fprintf(stderr, "\n");
693 }
694
695 for (int i = 0; i < num_found_colors; i++) {
696 fprintf(stderr, "color %d: 0x%08x\n", i, found_colors[i]);
697 }
698 }
699
700 static uint32_t
701 vc4_surface_msaa_get_sample(struct pipe_surface *psurf,
702 uint32_t x, uint32_t y, uint32_t sample)
703 {
704 struct pipe_resource *prsc = psurf->texture;
705 struct vc4_resource *rsc = vc4_resource(prsc);
706 uint32_t tile_w = 32, tile_h = 32;
707 uint32_t tiles_w = DIV_ROUND_UP(psurf->width, 32);
708
709 uint32_t tile_x = x / tile_w;
710 uint32_t tile_y = y / tile_h;
711 uint32_t *tile = (vc4_bo_map(rsc->bo) +
712 VC4_TILE_BUFFER_SIZE * (tile_y * tiles_w + tile_x));
713 uint32_t subtile_x = x % tile_w;
714 uint32_t subtile_y = y % tile_h;
715
716 uint32_t quad_samples = VC4_MAX_SAMPLES * 4;
717 uint32_t tile_stride = quad_samples * tile_w / 2;
718
719 return *((uint32_t *)tile +
720 (subtile_y >> 1) * tile_stride +
721 (subtile_x >> 1) * quad_samples +
722 ((subtile_y & 1) << 1) +
723 (subtile_x & 1) +
724 sample);
725 }
726
727 static void
728 vc4_dump_surface_msaa_char(struct pipe_surface *psurf,
729 uint32_t start_x, uint32_t start_y,
730 uint32_t w, uint32_t h)
731 {
732 bool all_same_color = true;
733 uint32_t all_pix = 0;
734
735 for (int y = start_y; y < start_y + h; y++) {
736 for (int x = start_x; x < start_x + w; x++) {
737 for (int s = 0; s < VC4_MAX_SAMPLES; s++) {
738 uint32_t pix = vc4_surface_msaa_get_sample(psurf,
739 x, y,
740 s);
741 if (x == start_x && y == start_y)
742 all_pix = pix;
743 else if (all_pix != pix)
744 all_same_color = false;
745 }
746 }
747 }
748 if (all_same_color) {
749 static const struct {
750 uint32_t val;
751 const char *c;
752 } named_colors[] = {
753 { 0xff000000, "█" },
754 { 0x00000000, "█" },
755 { 0xffff0000, "r" },
756 { 0xff00ff00, "g" },
757 { 0xff0000ff, "b" },
758 { 0xffffffff, "w" },
759 };
760 int i;
761 for (i = 0; i < ARRAY_SIZE(named_colors); i++) {
762 if (named_colors[i].val == all_pix) {
763 fprintf(stderr, "%s",
764 named_colors[i].c);
765 return;
766 }
767 }
768 fprintf(stderr, "x");
769 } else {
770 fprintf(stderr, ".");
771 }
772 }
773
774 static void
775 vc4_dump_surface_msaa(struct pipe_surface *psurf)
776 {
777 uint32_t tile_w = 32, tile_h = 32;
778 uint32_t tiles_w = DIV_ROUND_UP(psurf->width, tile_w);
779 uint32_t tiles_h = DIV_ROUND_UP(psurf->height, tile_h);
780 uint32_t char_w = 140, char_h = 60;
781 uint32_t char_w_per_tile = char_w / tiles_w - 1;
782 uint32_t char_h_per_tile = char_h / tiles_h - 1;
783 uint32_t found_colors[10];
784 uint32_t num_found_colors = 0;
785
786 fprintf(stderr, "Surface: %dx%d (%dx MSAA)\n",
787 psurf->width, psurf->height, psurf->texture->nr_samples);
788
789 for (int x = 0; x < (char_w_per_tile + 1) * tiles_w; x++)
790 fprintf(stderr, "-");
791 fprintf(stderr, "\n");
792
793 for (int ty = 0; ty < psurf->height; ty += tile_h) {
794 for (int y = 0; y < char_h_per_tile; y++) {
795
796 for (int tx = 0; tx < psurf->width; tx += tile_w) {
797 for (int x = 0; x < char_w_per_tile; x++) {
798 uint32_t bx1 = (x * tile_w /
799 char_w_per_tile);
800 uint32_t bx2 = ((x + 1) * tile_w /
801 char_w_per_tile);
802 uint32_t by1 = (y * tile_h /
803 char_h_per_tile);
804 uint32_t by2 = ((y + 1) * tile_h /
805 char_h_per_tile);
806
807 vc4_dump_surface_msaa_char(psurf,
808 tx + bx1,
809 ty + by1,
810 bx2 - bx1,
811 by2 - by1);
812 }
813 fprintf(stderr, "|");
814 }
815 fprintf(stderr, "\n");
816 }
817
818 for (int x = 0; x < (char_w_per_tile + 1) * tiles_w; x++)
819 fprintf(stderr, "-");
820 fprintf(stderr, "\n");
821 }
822
823 for (int i = 0; i < num_found_colors; i++) {
824 fprintf(stderr, "color %d: 0x%08x\n", i, found_colors[i]);
825 }
826 }
827
828 /** Debug routine to dump the contents of an 8888 surface to the console */
829 void
830 vc4_dump_surface(struct pipe_surface *psurf)
831 {
832 if (!psurf)
833 return;
834
835 if (psurf->texture->nr_samples > 1)
836 vc4_dump_surface_msaa(psurf);
837 else
838 vc4_dump_surface_non_msaa(psurf);
839 }
840
841 static void
842 vc4_flush_resource(struct pipe_context *pctx, struct pipe_resource *resource)
843 {
844 /* All calls to flush_resource are followed by a flush of the context,
845 * so there's nothing to do.
846 */
847 }
848
849 void
850 vc4_update_shadow_baselevel_texture(struct pipe_context *pctx,
851 struct pipe_sampler_view *view)
852 {
853 struct vc4_resource *shadow = vc4_resource(view->texture);
854 struct vc4_resource *orig = vc4_resource(shadow->shadow_parent);
855 assert(orig);
856
857 if (shadow->writes == orig->writes && orig->bo->private)
858 return;
859
860 perf_debug("Updating shadow texture due to %s\n",
861 view->u.tex.first_level ? "base level" : "raster layout");
862
863 for (int i = 0; i <= shadow->base.b.last_level; i++) {
864 unsigned width = u_minify(shadow->base.b.width0, i);
865 unsigned height = u_minify(shadow->base.b.height0, i);
866 struct pipe_blit_info info = {
867 .dst = {
868 .resource = &shadow->base.b,
869 .level = i,
870 .box = {
871 .x = 0,
872 .y = 0,
873 .z = 0,
874 .width = width,
875 .height = height,
876 .depth = 1,
877 },
878 .format = shadow->base.b.format,
879 },
880 .src = {
881 .resource = &orig->base.b,
882 .level = view->u.tex.first_level + i,
883 .box = {
884 .x = 0,
885 .y = 0,
886 .z = 0,
887 .width = width,
888 .height = height,
889 .depth = 1,
890 },
891 .format = orig->base.b.format,
892 },
893 .mask = ~0,
894 };
895 pctx->blit(pctx, &info);
896 }
897
898 shadow->writes = orig->writes;
899 }
900
901 /**
902 * Converts a 4-byte index buffer to 2 bytes.
903 *
904 * Since GLES2 only has support for 1 and 2-byte indices, the hardware doesn't
905 * include 4-byte index support, and we have to shrink it down.
906 *
907 * There's no fallback support for when indices end up being larger than 2^16,
908 * though it will at least assertion fail. Also, if the original index data
909 * was in user memory, it would be nice to not have uploaded it to a VBO
910 * before translating.
911 */
912 struct pipe_resource *
913 vc4_get_shadow_index_buffer(struct pipe_context *pctx,
914 const struct pipe_index_buffer *ib,
915 uint32_t count,
916 uint32_t *shadow_offset)
917 {
918 struct vc4_context *vc4 = vc4_context(pctx);
919 struct vc4_resource *orig = vc4_resource(ib->buffer);
920 perf_debug("Fallback conversion for %d uint indices\n", count);
921
922 void *data;
923 struct pipe_resource *shadow_rsc = NULL;
924 u_upload_alloc(vc4->uploader, 0, count * 2,
925 shadow_offset, &shadow_rsc, &data);
926 uint16_t *dst = data;
927
928 struct pipe_transfer *src_transfer = NULL;
929 const uint32_t *src;
930 if (ib->user_buffer) {
931 src = ib->user_buffer;
932 } else {
933 src = pipe_buffer_map_range(pctx, &orig->base.b,
934 ib->offset,
935 count * 4,
936 PIPE_TRANSFER_READ, &src_transfer);
937 }
938
939 for (int i = 0; i < count; i++) {
940 uint32_t src_index = src[i];
941 assert(src_index <= 0xffff);
942 dst[i] = src_index;
943 }
944
945 if (src_transfer)
946 pctx->transfer_unmap(pctx, src_transfer);
947
948 return shadow_rsc;
949 }
950
951 void
952 vc4_resource_screen_init(struct pipe_screen *pscreen)
953 {
954 pscreen->resource_create = vc4_resource_create;
955 pscreen->resource_from_handle = vc4_resource_from_handle;
956 pscreen->resource_get_handle = u_resource_get_handle_vtbl;
957 pscreen->resource_destroy = u_resource_destroy_vtbl;
958 }
959
960 void
961 vc4_resource_context_init(struct pipe_context *pctx)
962 {
963 pctx->transfer_map = u_transfer_map_vtbl;
964 pctx->transfer_flush_region = u_transfer_flush_region_vtbl;
965 pctx->transfer_unmap = u_transfer_unmap_vtbl;
966 pctx->transfer_inline_write = u_transfer_inline_write_vtbl;
967 pctx->create_surface = vc4_create_surface;
968 pctx->surface_destroy = vc4_surface_destroy;
969 pctx->resource_copy_region = util_resource_copy_region;
970 pctx->blit = vc4_blit;
971 pctx->flush_resource = vc4_flush_resource;
972 }