2 * Copyright © 2014 Broadcom
3 * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
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:
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
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
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"
32 #include "vc4_screen.h"
33 #include "vc4_context.h"
34 #include "vc4_resource.h"
35 #include "vc4_tiling.h"
37 static bool miptree_debug
= false;
40 vc4_resource_bo_alloc(struct vc4_resource
*rsc
)
42 struct pipe_resource
*prsc
= &rsc
->base
.b
;
43 struct pipe_screen
*pscreen
= prsc
->screen
;
47 fprintf(stderr
, "alloc %p: size %d + offset %d -> %d\n",
50 rsc
->slices
[0].offset
,
51 rsc
->slices
[0].offset
+
53 rsc
->cube_map_stride
* (prsc
->array_size
- 1));
56 bo
= vc4_bo_alloc(vc4_screen(pscreen
),
57 rsc
->slices
[0].offset
+
59 rsc
->cube_map_stride
* (prsc
->array_size
- 1),
62 vc4_bo_unreference(&rsc
->bo
);
71 vc4_resource_transfer_unmap(struct pipe_context
*pctx
,
72 struct pipe_transfer
*ptrans
)
74 struct vc4_context
*vc4
= vc4_context(pctx
);
75 struct vc4_transfer
*trans
= vc4_transfer(ptrans
);
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];
84 rsc
= vc4_resource(ptrans
->resource
);
85 slice
= &rsc
->slices
[ptrans
->level
];
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
,
92 trans
->map
, ptrans
->stride
,
93 slice
->tiling
, rsc
->cpp
,
99 if (trans
->ss_resource
&& (ptrans
->usage
& PIPE_TRANSFER_WRITE
)) {
100 struct pipe_blit_info blit
;
101 memset(&blit
, 0, sizeof(blit
));
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;
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
;
114 blit
.mask
= util_format_get_mask(ptrans
->resource
->format
);
115 blit
.filter
= PIPE_TEX_FILTER_NEAREST
;
117 pctx
->blit(pctx
, &blit
);
120 pipe_resource_reference(&trans
->ss_resource
, NULL
);
123 pipe_resource_reference(&ptrans
->resource
, NULL
);
124 util_slab_free(&vc4
->transfer_pool
, ptrans
);
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
)
132 struct pipe_resource temp_setup
;
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;
142 return pctx
->screen
->resource_create(pctx
->screen
, &temp_setup
);
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
)
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
;
159 if (usage
& PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
) {
160 if (vc4_resource_bo_alloc(rsc
)) {
162 /* If it might be bound as one of our vertex buffers,
163 * make sure we re-emit vertex buffer state.
165 if (prsc
->bind
& PIPE_BIND_VERTEX_BUFFER
)
166 vc4
->dirty
|= VC4_DIRTY_VTXBUF
;
168 /* If we failed to reallocate, flush everything so
169 * that we don't violate any syncing requirements.
173 } else if (!(usage
& PIPE_TRANSFER_UNSYNCHRONIZED
)) {
174 /* If we're writing and the buffer is being used by the CL, we
175 * have to flush the CL first. If we're only reading, we need
176 * to flush if the CL has written our buffer.
178 if (vc4_cl_references_bo(pctx
, rsc
->bo
,
179 usage
& PIPE_TRANSFER_WRITE
)) {
180 if ((usage
& PIPE_TRANSFER_DISCARD_RANGE
) &&
181 prsc
->last_level
== 0 &&
182 prsc
->width0
== box
->width
&&
183 prsc
->height0
== box
->height
&&
184 prsc
->depth0
== box
->depth
&&
185 vc4_resource_bo_alloc(rsc
)) {
186 if (prsc
->bind
& PIPE_BIND_VERTEX_BUFFER
)
187 vc4
->dirty
|= VC4_DIRTY_VTXBUF
;
194 if (usage
& PIPE_TRANSFER_WRITE
)
197 trans
= util_slab_alloc(&vc4
->transfer_pool
);
201 /* XXX: Handle DONTBLOCK, DISCARD_RANGE, PERSISTENT, COHERENT. */
203 /* util_slab_alloc() doesn't zero: */
204 memset(trans
, 0, sizeof(*trans
));
205 ptrans
= &trans
->base
;
207 pipe_resource_reference(&ptrans
->resource
, prsc
);
208 ptrans
->level
= level
;
209 ptrans
->usage
= usage
;
212 /* If the resource is multisampled, we need to resolve to single
213 * sample. This seems like it should be handled at a higher layer.
215 if (prsc
->nr_samples
> 1) {
216 trans
->ss_resource
= vc4_get_temp_resource(pctx
, prsc
, box
);
217 if (!trans
->ss_resource
)
219 assert(!trans
->ss_resource
->nr_samples
);
221 /* The ptrans->box gets modified for tile alignment, so save
222 * the original box for unmap time.
224 trans
->ss_box
= *box
;
226 if (usage
& PIPE_TRANSFER_READ
) {
227 struct pipe_blit_info blit
;
228 memset(&blit
, 0, sizeof(blit
));
230 blit
.src
.resource
= ptrans
->resource
;
231 blit
.src
.format
= ptrans
->resource
->format
;
232 blit
.src
.level
= ptrans
->level
;
233 blit
.src
.box
= trans
->ss_box
;
235 blit
.dst
.resource
= trans
->ss_resource
;
236 blit
.dst
.format
= trans
->ss_resource
->format
;
237 blit
.dst
.box
.width
= trans
->ss_box
.width
;
238 blit
.dst
.box
.height
= trans
->ss_box
.height
;
239 blit
.dst
.box
.depth
= 1;
241 blit
.mask
= util_format_get_mask(prsc
->format
);
242 blit
.filter
= PIPE_TEX_FILTER_NEAREST
;
244 pctx
->blit(pctx
, &blit
);
248 /* The rest of the mapping process should use our temporary. */
249 prsc
= trans
->ss_resource
;
250 rsc
= vc4_resource(prsc
);
256 /* Note that the current kernel implementation is synchronous, so no
257 * need to do syncing stuff here yet.
260 if (usage
& PIPE_TRANSFER_UNSYNCHRONIZED
)
261 buf
= vc4_bo_map_unsynchronized(rsc
->bo
);
263 buf
= vc4_bo_map(rsc
->bo
);
265 fprintf(stderr
, "Failed to map bo\n");
271 struct vc4_resource_slice
*slice
= &rsc
->slices
[level
];
273 uint32_t utile_w
= vc4_utile_width(rsc
->cpp
);
274 uint32_t utile_h
= vc4_utile_height(rsc
->cpp
);
276 /* No direct mappings of tiled, since we need to manually
279 if (usage
& PIPE_TRANSFER_MAP_DIRECTLY
)
282 /* We need to align the box to utile boundaries, since that's
283 * what load/store operate on.
285 uint32_t orig_width
= ptrans
->box
.width
;
286 uint32_t orig_height
= ptrans
->box
.height
;
287 uint32_t box_start_x
= ptrans
->box
.x
& (utile_w
- 1);
288 uint32_t box_start_y
= ptrans
->box
.y
& (utile_h
- 1);
289 ptrans
->box
.width
+= box_start_x
;
290 ptrans
->box
.x
-= box_start_x
;
291 ptrans
->box
.height
+= box_start_y
;
292 ptrans
->box
.y
-= box_start_y
;
293 ptrans
->box
.width
= align(ptrans
->box
.width
, utile_w
);
294 ptrans
->box
.height
= align(ptrans
->box
.height
, utile_h
);
296 ptrans
->stride
= ptrans
->box
.width
* rsc
->cpp
;
297 ptrans
->layer_stride
= ptrans
->stride
;
299 trans
->map
= malloc(ptrans
->stride
* ptrans
->box
.height
);
300 if (usage
& PIPE_TRANSFER_READ
||
301 ptrans
->box
.width
!= orig_width
||
302 ptrans
->box
.height
!= orig_height
) {
303 vc4_load_tiled_image(trans
->map
, ptrans
->stride
,
304 buf
+ slice
->offset
+
305 ptrans
->box
.z
* rsc
->cube_map_stride
,
307 slice
->tiling
, rsc
->cpp
,
311 box_start_x
* rsc
->cpp
+
312 box_start_y
* ptrans
->stride
);
314 ptrans
->stride
= slice
->stride
;
315 ptrans
->layer_stride
= ptrans
->stride
;
317 return buf
+ slice
->offset
+
318 ptrans
->box
.y
/ util_format_get_blockheight(format
) * ptrans
->stride
+
319 ptrans
->box
.x
/ util_format_get_blockwidth(format
) * rsc
->cpp
+
320 ptrans
->box
.z
* rsc
->cube_map_stride
;
325 vc4_resource_transfer_unmap(pctx
, ptrans
);
330 vc4_resource_destroy(struct pipe_screen
*pscreen
,
331 struct pipe_resource
*prsc
)
333 struct vc4_resource
*rsc
= vc4_resource(prsc
);
334 pipe_resource_reference(&rsc
->shadow_parent
, NULL
);
335 vc4_bo_unreference(&rsc
->bo
);
340 vc4_resource_get_handle(struct pipe_screen
*pscreen
,
341 struct pipe_resource
*prsc
,
342 struct winsys_handle
*handle
)
344 struct vc4_resource
*rsc
= vc4_resource(prsc
);
346 return vc4_screen_bo_get_handle(pscreen
, rsc
->bo
, rsc
->slices
[0].stride
,
350 static const struct u_resource_vtbl vc4_resource_vtbl
= {
351 .resource_get_handle
= vc4_resource_get_handle
,
352 .resource_destroy
= vc4_resource_destroy
,
353 .transfer_map
= vc4_resource_transfer_map
,
354 .transfer_flush_region
= u_default_transfer_flush_region
,
355 .transfer_unmap
= vc4_resource_transfer_unmap
,
356 .transfer_inline_write
= u_default_transfer_inline_write
,
360 vc4_setup_slices(struct vc4_resource
*rsc
)
362 struct pipe_resource
*prsc
= &rsc
->base
.b
;
363 uint32_t width
= prsc
->width0
;
364 uint32_t height
= prsc
->height0
;
365 uint32_t pot_width
= util_next_power_of_two(width
);
366 uint32_t pot_height
= util_next_power_of_two(height
);
368 uint32_t utile_w
= vc4_utile_width(rsc
->cpp
);
369 uint32_t utile_h
= vc4_utile_height(rsc
->cpp
);
371 for (int i
= prsc
->last_level
; i
>= 0; i
--) {
372 struct vc4_resource_slice
*slice
= &rsc
->slices
[i
];
374 uint32_t level_width
, level_height
;
377 level_height
= height
;
379 level_width
= u_minify(pot_width
, i
);
380 level_height
= u_minify(pot_height
, i
);
384 slice
->tiling
= VC4_TILING_FORMAT_LINEAR
;
385 if (prsc
->nr_samples
> 1) {
386 /* MSAA (4x) surfaces are stored as raw tile buffer contents. */
387 level_width
= align(level_width
, 32);
388 level_height
= align(level_height
, 32);
390 level_width
= align(level_width
, utile_w
);
393 if (vc4_size_is_lt(level_width
, level_height
,
395 slice
->tiling
= VC4_TILING_FORMAT_LT
;
396 level_width
= align(level_width
, utile_w
);
397 level_height
= align(level_height
, utile_h
);
399 slice
->tiling
= VC4_TILING_FORMAT_T
;
400 level_width
= align(level_width
,
402 level_height
= align(level_height
,
407 slice
->offset
= offset
;
408 slice
->stride
= (level_width
* rsc
->cpp
*
409 MAX2(prsc
->nr_samples
, 1));
410 slice
->size
= level_height
* slice
->stride
;
412 offset
+= slice
->size
;
415 static const char tiling_chars
[] = {
416 [VC4_TILING_FORMAT_LINEAR
] = 'R',
417 [VC4_TILING_FORMAT_LT
] = 'L',
418 [VC4_TILING_FORMAT_T
] = 'T'
421 "rsc setup %p (format %d), %dx%d: "
422 "level %d (%c) -> %dx%d, stride %d@0x%08x\n",
423 rsc
, rsc
->vc4_format
,
424 prsc
->width0
, prsc
->height0
,
425 i
, tiling_chars
[slice
->tiling
],
426 level_width
, level_height
,
427 slice
->stride
, slice
->offset
);
431 /* The texture base pointer that has to point to level 0 doesn't have
432 * intra-page bits, so we have to align it, and thus shift up all the
435 uint32_t page_align_offset
= (align(rsc
->slices
[0].offset
, 4096) -
436 rsc
->slices
[0].offset
);
437 if (page_align_offset
) {
438 for (int i
= 0; i
<= prsc
->last_level
; i
++)
439 rsc
->slices
[i
].offset
+= page_align_offset
;
442 /* Cube map faces appear as whole miptrees at a page-aligned offset
443 * from the first face's miptree.
445 if (prsc
->target
== PIPE_TEXTURE_CUBE
) {
446 rsc
->cube_map_stride
= align(rsc
->slices
[0].offset
+
447 rsc
->slices
[0].size
, 4096);
451 static struct vc4_resource
*
452 vc4_resource_setup(struct pipe_screen
*pscreen
,
453 const struct pipe_resource
*tmpl
)
455 struct vc4_resource
*rsc
= CALLOC_STRUCT(vc4_resource
);
458 struct pipe_resource
*prsc
= &rsc
->base
.b
;
462 pipe_reference_init(&prsc
->reference
, 1);
463 prsc
->screen
= pscreen
;
465 rsc
->base
.vtbl
= &vc4_resource_vtbl
;
466 if (prsc
->nr_samples
<= 1)
467 rsc
->cpp
= util_format_get_blocksize(tmpl
->format
);
469 rsc
->cpp
= sizeof(uint32_t);
476 static enum vc4_texture_data_type
477 get_resource_texture_format(struct pipe_resource
*prsc
)
479 struct vc4_resource
*rsc
= vc4_resource(prsc
);
480 uint8_t format
= vc4_get_tex_format(prsc
->format
);
483 if (prsc
->nr_samples
> 1) {
486 assert(format
== VC4_TEXTURE_TYPE_RGBA8888
);
487 return VC4_TEXTURE_TYPE_RGBA32R
;
494 struct pipe_resource
*
495 vc4_resource_create(struct pipe_screen
*pscreen
,
496 const struct pipe_resource
*tmpl
)
498 struct vc4_resource
*rsc
= vc4_resource_setup(pscreen
, tmpl
);
499 struct pipe_resource
*prsc
= &rsc
->base
.b
;
501 /* We have to make shared be untiled, since we don't have any way to
502 * communicate metadata about tiling currently.
504 if (tmpl
->target
== PIPE_BUFFER
||
505 tmpl
->nr_samples
> 1 ||
506 (tmpl
->bind
& (PIPE_BIND_SCANOUT
|
509 PIPE_BIND_CURSOR
))) {
515 if (tmpl
->target
!= PIPE_BUFFER
)
516 rsc
->vc4_format
= get_resource_texture_format(prsc
);
518 vc4_setup_slices(rsc
);
519 if (!vc4_resource_bo_alloc(rsc
))
524 vc4_resource_destroy(pscreen
, prsc
);
528 static struct pipe_resource
*
529 vc4_resource_from_handle(struct pipe_screen
*pscreen
,
530 const struct pipe_resource
*tmpl
,
531 struct winsys_handle
*handle
,
534 struct vc4_resource
*rsc
= vc4_resource_setup(pscreen
, tmpl
);
535 struct pipe_resource
*prsc
= &rsc
->base
.b
;
536 struct vc4_resource_slice
*slice
= &rsc
->slices
[0];
537 uint32_t expected_stride
= align(prsc
->width0
/ rsc
->cpp
,
538 vc4_utile_width(rsc
->cpp
));
543 if (handle
->stride
!= expected_stride
) {
544 static bool warned
= false;
548 "Attempting to import %dx%d %s with "
549 "unsupported stride %d instead of %d\n",
550 prsc
->width0
, prsc
->height0
,
551 util_format_short_name(prsc
->format
),
559 rsc
->bo
= vc4_screen_bo_from_handle(pscreen
, handle
);
563 slice
->stride
= handle
->stride
;
564 slice
->tiling
= VC4_TILING_FORMAT_LINEAR
;
566 rsc
->vc4_format
= get_resource_texture_format(prsc
);
570 "rsc import %p (format %d), %dx%d: "
571 "level 0 (R) -> stride %d@0x%08x\n",
572 rsc
, rsc
->vc4_format
,
573 prsc
->width0
, prsc
->height0
,
574 slice
->stride
, slice
->offset
);
580 vc4_resource_destroy(pscreen
, prsc
);
584 static struct pipe_surface
*
585 vc4_create_surface(struct pipe_context
*pctx
,
586 struct pipe_resource
*ptex
,
587 const struct pipe_surface
*surf_tmpl
)
589 struct vc4_surface
*surface
= CALLOC_STRUCT(vc4_surface
);
590 struct vc4_resource
*rsc
= vc4_resource(ptex
);
595 assert(surf_tmpl
->u
.tex
.first_layer
== surf_tmpl
->u
.tex
.last_layer
);
597 struct pipe_surface
*psurf
= &surface
->base
;
598 unsigned level
= surf_tmpl
->u
.tex
.level
;
600 pipe_reference_init(&psurf
->reference
, 1);
601 pipe_resource_reference(&psurf
->texture
, ptex
);
603 psurf
->context
= pctx
;
604 psurf
->format
= surf_tmpl
->format
;
605 psurf
->width
= u_minify(ptex
->width0
, level
);
606 psurf
->height
= u_minify(ptex
->height0
, level
);
607 psurf
->u
.tex
.level
= level
;
608 psurf
->u
.tex
.first_layer
= surf_tmpl
->u
.tex
.first_layer
;
609 psurf
->u
.tex
.last_layer
= surf_tmpl
->u
.tex
.last_layer
;
610 surface
->offset
= rsc
->slices
[level
].offset
;
611 surface
->tiling
= rsc
->slices
[level
].tiling
;
613 return &surface
->base
;
617 vc4_surface_destroy(struct pipe_context
*pctx
, struct pipe_surface
*psurf
)
619 pipe_resource_reference(&psurf
->texture
, NULL
);
624 vc4_dump_surface_non_msaa(struct pipe_surface
*psurf
)
626 struct pipe_resource
*prsc
= psurf
->texture
;
627 struct vc4_resource
*rsc
= vc4_resource(prsc
);
628 uint32_t *map
= vc4_bo_map(rsc
->bo
);
629 uint32_t stride
= rsc
->slices
[0].stride
/ 4;
630 uint32_t width
= psurf
->width
;
631 uint32_t height
= psurf
->height
;
632 uint32_t chunk_w
= width
/ 79;
633 uint32_t chunk_h
= height
/ 40;
634 uint32_t found_colors
[10];
635 uint32_t num_found_colors
= 0;
637 if (rsc
->vc4_format
!= VC4_TEXTURE_TYPE_RGBA32R
) {
638 fprintf(stderr
, "%s: Unsupported format %s\n",
639 __func__
, util_format_short_name(psurf
->format
));
643 for (int by
= 0; by
< height
; by
+= chunk_h
) {
644 for (int bx
= 0; bx
< width
; bx
+= chunk_w
) {
645 int all_found_color
= -1; /* nothing found */
647 for (int y
= by
; y
< MIN2(height
, by
+ chunk_h
); y
++) {
648 for (int x
= bx
; x
< MIN2(width
, bx
+ chunk_w
); x
++) {
649 uint32_t pix
= map
[y
* stride
+ x
];
652 for (i
= 0; i
< num_found_colors
; i
++) {
653 if (pix
== found_colors
[i
])
656 if (i
== num_found_colors
&&
658 ARRAY_SIZE(found_colors
)) {
659 found_colors
[num_found_colors
++] = pix
;
662 if (i
< num_found_colors
) {
663 if (all_found_color
== -1)
665 else if (i
!= all_found_color
)
666 all_found_color
= ARRAY_SIZE(found_colors
);
670 /* If all pixels for this chunk have a consistent
671 * value, then print a character for it. Either a
672 * fixed name (particularly common for piglit tests),
673 * or a runtime-generated number.
675 if (all_found_color
>= 0 &&
676 all_found_color
< ARRAY_SIZE(found_colors
)) {
677 static const struct {
689 for (i
= 0; i
< ARRAY_SIZE(named_colors
); i
++) {
690 if (named_colors
[i
].val
==
691 found_colors
[all_found_color
]) {
692 fprintf(stderr
, "%s",
697 /* For unnamed colors, print a number and the
698 * numbers will have values printed at the
701 if (i
== ARRAY_SIZE(named_colors
)) {
702 fprintf(stderr
, "%c",
703 '0' + all_found_color
);
706 /* If there's no consistent color, print this.
708 fprintf(stderr
, ".");
711 fprintf(stderr
, "\n");
714 for (int i
= 0; i
< num_found_colors
; i
++) {
715 fprintf(stderr
, "color %d: 0x%08x\n", i
, found_colors
[i
]);
720 vc4_surface_msaa_get_sample(struct pipe_surface
*psurf
,
721 uint32_t x
, uint32_t y
, uint32_t sample
)
723 struct pipe_resource
*prsc
= psurf
->texture
;
724 struct vc4_resource
*rsc
= vc4_resource(prsc
);
725 uint32_t tile_w
= 32, tile_h
= 32;
726 uint32_t tiles_w
= DIV_ROUND_UP(psurf
->width
, 32);
728 uint32_t tile_x
= x
/ tile_w
;
729 uint32_t tile_y
= y
/ tile_h
;
730 uint32_t *tile
= (vc4_bo_map(rsc
->bo
) +
731 VC4_TILE_BUFFER_SIZE
* (tile_y
* tiles_w
+ tile_x
));
732 uint32_t subtile_x
= x
% tile_w
;
733 uint32_t subtile_y
= y
% tile_h
;
735 uint32_t quad_samples
= VC4_MAX_SAMPLES
* 4;
736 uint32_t tile_stride
= quad_samples
* tile_w
/ 2;
738 return *((uint32_t *)tile
+
739 (subtile_y
>> 1) * tile_stride
+
740 (subtile_x
>> 1) * quad_samples
+
741 ((subtile_y
& 1) << 1) +
747 vc4_dump_surface_msaa_char(struct pipe_surface
*psurf
,
748 uint32_t start_x
, uint32_t start_y
,
749 uint32_t w
, uint32_t h
)
751 bool all_same_color
= true;
752 uint32_t all_pix
= 0;
754 for (int y
= start_y
; y
< start_y
+ h
; y
++) {
755 for (int x
= start_x
; x
< start_x
+ w
; x
++) {
756 for (int s
= 0; s
< VC4_MAX_SAMPLES
; s
++) {
757 uint32_t pix
= vc4_surface_msaa_get_sample(psurf
,
760 if (x
== start_x
&& y
== start_y
)
762 else if (all_pix
!= pix
)
763 all_same_color
= false;
767 if (all_same_color
) {
768 static const struct {
780 for (i
= 0; i
< ARRAY_SIZE(named_colors
); i
++) {
781 if (named_colors
[i
].val
== all_pix
) {
782 fprintf(stderr
, "%s",
787 fprintf(stderr
, "x");
789 fprintf(stderr
, ".");
794 vc4_dump_surface_msaa(struct pipe_surface
*psurf
)
796 uint32_t tile_w
= 32, tile_h
= 32;
797 uint32_t tiles_w
= DIV_ROUND_UP(psurf
->width
, tile_w
);
798 uint32_t tiles_h
= DIV_ROUND_UP(psurf
->height
, tile_h
);
799 uint32_t char_w
= 140, char_h
= 60;
800 uint32_t char_w_per_tile
= char_w
/ tiles_w
- 1;
801 uint32_t char_h_per_tile
= char_h
/ tiles_h
- 1;
802 uint32_t found_colors
[10];
803 uint32_t num_found_colors
= 0;
805 fprintf(stderr
, "Surface: %dx%d (%dx MSAA)\n",
806 psurf
->width
, psurf
->height
, psurf
->texture
->nr_samples
);
808 for (int x
= 0; x
< (char_w_per_tile
+ 1) * tiles_w
; x
++)
809 fprintf(stderr
, "-");
810 fprintf(stderr
, "\n");
812 for (int ty
= 0; ty
< psurf
->height
; ty
+= tile_h
) {
813 for (int y
= 0; y
< char_h_per_tile
; y
++) {
815 for (int tx
= 0; tx
< psurf
->width
; tx
+= tile_w
) {
816 for (int x
= 0; x
< char_w_per_tile
; x
++) {
817 uint32_t bx1
= (x
* tile_w
/
819 uint32_t bx2
= ((x
+ 1) * tile_w
/
821 uint32_t by1
= (y
* tile_h
/
823 uint32_t by2
= ((y
+ 1) * tile_h
/
826 vc4_dump_surface_msaa_char(psurf
,
832 fprintf(stderr
, "|");
834 fprintf(stderr
, "\n");
837 for (int x
= 0; x
< (char_w_per_tile
+ 1) * tiles_w
; x
++)
838 fprintf(stderr
, "-");
839 fprintf(stderr
, "\n");
842 for (int i
= 0; i
< num_found_colors
; i
++) {
843 fprintf(stderr
, "color %d: 0x%08x\n", i
, found_colors
[i
]);
847 /** Debug routine to dump the contents of an 8888 surface to the console */
849 vc4_dump_surface(struct pipe_surface
*psurf
)
854 if (psurf
->texture
->nr_samples
> 1)
855 vc4_dump_surface_msaa(psurf
);
857 vc4_dump_surface_non_msaa(psurf
);
861 vc4_flush_resource(struct pipe_context
*pctx
, struct pipe_resource
*resource
)
863 /* All calls to flush_resource are followed by a flush of the context,
864 * so there's nothing to do.
869 vc4_update_shadow_baselevel_texture(struct pipe_context
*pctx
,
870 struct pipe_sampler_view
*view
)
872 struct vc4_resource
*shadow
= vc4_resource(view
->texture
);
873 struct vc4_resource
*orig
= vc4_resource(shadow
->shadow_parent
);
876 if (shadow
->writes
== orig
->writes
&& orig
->bo
->private)
879 perf_debug("Updating shadow texture due to %s\n",
880 view
->u
.tex
.first_level
? "base level" : "raster layout");
882 for (int i
= 0; i
<= shadow
->base
.b
.last_level
; i
++) {
883 unsigned width
= u_minify(shadow
->base
.b
.width0
, i
);
884 unsigned height
= u_minify(shadow
->base
.b
.height0
, i
);
885 struct pipe_blit_info info
= {
887 .resource
= &shadow
->base
.b
,
897 .format
= shadow
->base
.b
.format
,
900 .resource
= &orig
->base
.b
,
901 .level
= view
->u
.tex
.first_level
+ i
,
910 .format
= orig
->base
.b
.format
,
914 pctx
->blit(pctx
, &info
);
917 shadow
->writes
= orig
->writes
;
921 * Converts a 4-byte index buffer to 2 bytes.
923 * Since GLES2 only has support for 1 and 2-byte indices, the hardware doesn't
924 * include 4-byte index support, and we have to shrink it down.
926 * There's no fallback support for when indices end up being larger than 2^16,
927 * though it will at least assertion fail. Also, if the original index data
928 * was in user memory, it would be nice to not have uploaded it to a VBO
929 * before translating.
931 struct pipe_resource
*
932 vc4_get_shadow_index_buffer(struct pipe_context
*pctx
,
933 const struct pipe_index_buffer
*ib
,
935 uint32_t *shadow_offset
)
937 struct vc4_context
*vc4
= vc4_context(pctx
);
938 struct vc4_resource
*orig
= vc4_resource(ib
->buffer
);
939 perf_debug("Fallback conversion for %d uint indices\n", count
);
942 struct pipe_resource
*shadow_rsc
= NULL
;
943 u_upload_alloc(vc4
->uploader
, 0, count
* 2, 4,
944 shadow_offset
, &shadow_rsc
, &data
);
945 uint16_t *dst
= data
;
947 struct pipe_transfer
*src_transfer
= NULL
;
949 if (ib
->user_buffer
) {
950 src
= ib
->user_buffer
;
952 src
= pipe_buffer_map_range(pctx
, &orig
->base
.b
,
955 PIPE_TRANSFER_READ
, &src_transfer
);
958 for (int i
= 0; i
< count
; i
++) {
959 uint32_t src_index
= src
[i
];
960 assert(src_index
<= 0xffff);
965 pctx
->transfer_unmap(pctx
, src_transfer
);
971 vc4_resource_screen_init(struct pipe_screen
*pscreen
)
973 pscreen
->resource_create
= vc4_resource_create
;
974 pscreen
->resource_from_handle
= vc4_resource_from_handle
;
975 pscreen
->resource_get_handle
= u_resource_get_handle_vtbl
;
976 pscreen
->resource_destroy
= u_resource_destroy_vtbl
;
980 vc4_resource_context_init(struct pipe_context
*pctx
)
982 pctx
->transfer_map
= u_transfer_map_vtbl
;
983 pctx
->transfer_flush_region
= u_transfer_flush_region_vtbl
;
984 pctx
->transfer_unmap
= u_transfer_unmap_vtbl
;
985 pctx
->transfer_inline_write
= u_transfer_inline_write_vtbl
;
986 pctx
->create_surface
= vc4_create_surface
;
987 pctx
->surface_destroy
= vc4_surface_destroy
;
988 pctx
->resource_copy_region
= util_resource_copy_region
;
989 pctx
->blit
= vc4_blit
;
990 pctx
->flush_resource
= vc4_flush_resource
;