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
;
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
);
119 pipe_resource_reference(&trans
->ss_resource
, NULL
);
122 pipe_resource_reference(&ptrans
->resource
, NULL
);
123 slab_free(&vc4
->transfer_pool
, ptrans
);
126 static struct pipe_resource
*
127 vc4_get_temp_resource(struct pipe_context
*pctx
,
128 struct pipe_resource
*prsc
,
129 const struct pipe_box
*box
)
131 struct pipe_resource temp_setup
;
133 memset(&temp_setup
, 0, sizeof(temp_setup
));
134 temp_setup
.target
= prsc
->target
;
135 temp_setup
.format
= prsc
->format
;
136 temp_setup
.width0
= box
->width
;
137 temp_setup
.height0
= box
->height
;
138 temp_setup
.depth0
= 1;
139 temp_setup
.array_size
= 1;
141 return pctx
->screen
->resource_create(pctx
->screen
, &temp_setup
);
145 vc4_resource_transfer_map(struct pipe_context
*pctx
,
146 struct pipe_resource
*prsc
,
147 unsigned level
, unsigned usage
,
148 const struct pipe_box
*box
,
149 struct pipe_transfer
**pptrans
)
151 struct vc4_context
*vc4
= vc4_context(pctx
);
152 struct vc4_resource
*rsc
= vc4_resource(prsc
);
153 struct vc4_transfer
*trans
;
154 struct pipe_transfer
*ptrans
;
155 enum pipe_format format
= prsc
->format
;
158 /* Upgrade DISCARD_RANGE to WHOLE_RESOURCE if the whole resource is
161 if ((usage
& PIPE_TRANSFER_DISCARD_RANGE
) &&
162 !(usage
& PIPE_TRANSFER_UNSYNCHRONIZED
) &&
163 !(prsc
->flags
& PIPE_RESOURCE_FLAG_MAP_COHERENT
) &&
164 prsc
->last_level
== 0 &&
165 prsc
->width0
== box
->width
&&
166 prsc
->height0
== box
->height
&&
167 prsc
->depth0
== box
->depth
&&
168 prsc
->array_size
== 1) {
169 usage
|= PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
;
172 if (usage
& PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
) {
173 if (vc4_resource_bo_alloc(rsc
)) {
174 /* If it might be bound as one of our vertex buffers,
175 * make sure we re-emit vertex buffer state.
177 if (prsc
->bind
& PIPE_BIND_VERTEX_BUFFER
)
178 vc4
->dirty
|= VC4_DIRTY_VTXBUF
;
180 /* If we failed to reallocate, flush users so that we
181 * don't violate any syncing requirements.
183 vc4_flush_jobs_reading_resource(vc4
, prsc
);
185 } else if (!(usage
& PIPE_TRANSFER_UNSYNCHRONIZED
)) {
186 /* If we're writing and the buffer is being used by the CL, we
187 * have to flush the CL first. If we're only reading, we need
188 * to flush if the CL has written our buffer.
190 if (usage
& PIPE_TRANSFER_WRITE
)
191 vc4_flush_jobs_reading_resource(vc4
, prsc
);
193 vc4_flush_jobs_writing_resource(vc4
, prsc
);
196 if (usage
& PIPE_TRANSFER_WRITE
) {
198 rsc
->initialized_buffers
= ~0;
201 trans
= slab_alloc(&vc4
->transfer_pool
);
205 /* XXX: Handle DONTBLOCK, DISCARD_RANGE, PERSISTENT, COHERENT. */
207 /* slab_alloc_st() doesn't zero: */
208 memset(trans
, 0, sizeof(*trans
));
209 ptrans
= &trans
->base
;
211 pipe_resource_reference(&ptrans
->resource
, prsc
);
212 ptrans
->level
= level
;
213 ptrans
->usage
= usage
;
216 /* If the resource is multisampled, we need to resolve to single
217 * sample. This seems like it should be handled at a higher layer.
219 if (prsc
->nr_samples
> 1) {
220 trans
->ss_resource
= vc4_get_temp_resource(pctx
, prsc
, box
);
221 if (!trans
->ss_resource
)
223 assert(!trans
->ss_resource
->nr_samples
);
225 /* The ptrans->box gets modified for tile alignment, so save
226 * the original box for unmap time.
228 trans
->ss_box
= *box
;
230 if (usage
& PIPE_TRANSFER_READ
) {
231 struct pipe_blit_info blit
;
232 memset(&blit
, 0, sizeof(blit
));
234 blit
.src
.resource
= ptrans
->resource
;
235 blit
.src
.format
= ptrans
->resource
->format
;
236 blit
.src
.level
= ptrans
->level
;
237 blit
.src
.box
= trans
->ss_box
;
239 blit
.dst
.resource
= trans
->ss_resource
;
240 blit
.dst
.format
= trans
->ss_resource
->format
;
241 blit
.dst
.box
.width
= trans
->ss_box
.width
;
242 blit
.dst
.box
.height
= trans
->ss_box
.height
;
243 blit
.dst
.box
.depth
= 1;
245 blit
.mask
= util_format_get_mask(prsc
->format
);
246 blit
.filter
= PIPE_TEX_FILTER_NEAREST
;
248 pctx
->blit(pctx
, &blit
);
249 vc4_flush_jobs_writing_resource(vc4
, blit
.dst
.resource
);
252 /* The rest of the mapping process should use our temporary. */
253 prsc
= trans
->ss_resource
;
254 rsc
= vc4_resource(prsc
);
260 /* Note that the current kernel implementation is synchronous, so no
261 * need to do syncing stuff here yet.
264 if (usage
& PIPE_TRANSFER_UNSYNCHRONIZED
)
265 buf
= vc4_bo_map_unsynchronized(rsc
->bo
);
267 buf
= vc4_bo_map(rsc
->bo
);
269 fprintf(stderr
, "Failed to map bo\n");
275 struct vc4_resource_slice
*slice
= &rsc
->slices
[level
];
277 uint32_t utile_w
= vc4_utile_width(rsc
->cpp
);
278 uint32_t utile_h
= vc4_utile_height(rsc
->cpp
);
280 /* No direct mappings of tiled, since we need to manually
283 if (usage
& PIPE_TRANSFER_MAP_DIRECTLY
)
286 if (format
== PIPE_FORMAT_ETC1_RGB8
) {
287 /* ETC1 is arranged as 64-bit blocks, where each block
288 * is 4x4 pixels. Texture tiling operates on the
289 * 64-bit block the way it would an uncompressed
292 assert(!(ptrans
->box
.x
& 3));
293 assert(!(ptrans
->box
.y
& 3));
296 ptrans
->box
.width
= (ptrans
->box
.width
+ 3) >> 2;
297 ptrans
->box
.height
= (ptrans
->box
.height
+ 3) >> 2;
300 /* We need to align the box to utile boundaries, since that's
301 * what load/store operates on. This may cause us to need to
302 * read out the original contents in that border area. Right
303 * now we just read out the entire contents, including the
304 * middle area that will just get overwritten.
306 uint32_t box_start_x
= ptrans
->box
.x
& (utile_w
- 1);
307 uint32_t box_start_y
= ptrans
->box
.y
& (utile_h
- 1);
308 bool needs_load
= (usage
& PIPE_TRANSFER_READ
) != 0;
311 ptrans
->box
.width
+= box_start_x
;
312 ptrans
->box
.x
-= box_start_x
;
316 ptrans
->box
.height
+= box_start_y
;
317 ptrans
->box
.y
-= box_start_y
;
320 if (ptrans
->box
.width
& (utile_w
- 1)) {
321 /* We only need to force a load if our border region
322 * we're extending into is actually part of the
325 uint32_t slice_width
= u_minify(prsc
->width0
, level
);
326 if (ptrans
->box
.x
+ ptrans
->box
.width
!= slice_width
)
328 ptrans
->box
.width
= align(ptrans
->box
.width
, utile_w
);
330 if (ptrans
->box
.height
& (utile_h
- 1)) {
331 uint32_t slice_height
= u_minify(prsc
->height0
, level
);
332 if (ptrans
->box
.y
+ ptrans
->box
.height
!= slice_height
)
334 ptrans
->box
.height
= align(ptrans
->box
.height
, utile_h
);
337 ptrans
->stride
= ptrans
->box
.width
* rsc
->cpp
;
338 ptrans
->layer_stride
= ptrans
->stride
* ptrans
->box
.height
;
340 trans
->map
= malloc(ptrans
->layer_stride
* ptrans
->box
.depth
);
343 vc4_load_tiled_image(trans
->map
, ptrans
->stride
,
344 buf
+ slice
->offset
+
345 ptrans
->box
.z
* rsc
->cube_map_stride
,
347 slice
->tiling
, rsc
->cpp
,
351 box_start_x
* rsc
->cpp
+
352 box_start_y
* ptrans
->stride
);
354 ptrans
->stride
= slice
->stride
;
355 ptrans
->layer_stride
= ptrans
->stride
;
357 return buf
+ slice
->offset
+
358 ptrans
->box
.y
/ util_format_get_blockheight(format
) * ptrans
->stride
+
359 ptrans
->box
.x
/ util_format_get_blockwidth(format
) * rsc
->cpp
+
360 ptrans
->box
.z
* rsc
->cube_map_stride
;
365 vc4_resource_transfer_unmap(pctx
, ptrans
);
370 vc4_resource_destroy(struct pipe_screen
*pscreen
,
371 struct pipe_resource
*prsc
)
373 struct vc4_resource
*rsc
= vc4_resource(prsc
);
374 pipe_resource_reference(&rsc
->shadow_parent
, NULL
);
375 vc4_bo_unreference(&rsc
->bo
);
380 vc4_resource_get_handle(struct pipe_screen
*pscreen
,
381 struct pipe_context
*pctx
,
382 struct pipe_resource
*prsc
,
383 struct winsys_handle
*whandle
,
386 struct vc4_resource
*rsc
= vc4_resource(prsc
);
388 whandle
->stride
= rsc
->slices
[0].stride
;
390 /* If we're passing some reference to our BO out to some other part of
391 * the system, then we can't do any optimizations about only us being
392 * the ones seeing it (like BO caching or shadow update avoidance).
394 rsc
->bo
->private = false;
396 switch (whandle
->type
) {
397 case DRM_API_HANDLE_TYPE_SHARED
:
398 return vc4_bo_flink(rsc
->bo
, &whandle
->handle
);
399 case DRM_API_HANDLE_TYPE_KMS
:
400 whandle
->handle
= rsc
->bo
->handle
;
402 case DRM_API_HANDLE_TYPE_FD
:
403 whandle
->handle
= vc4_bo_get_dmabuf(rsc
->bo
);
404 return whandle
->handle
!= -1;
411 vc4_setup_slices(struct vc4_resource
*rsc
)
413 struct pipe_resource
*prsc
= &rsc
->base
;
414 uint32_t width
= prsc
->width0
;
415 uint32_t height
= prsc
->height0
;
416 if (prsc
->format
== PIPE_FORMAT_ETC1_RGB8
) {
417 width
= (width
+ 3) >> 2;
418 height
= (height
+ 3) >> 2;
421 uint32_t pot_width
= util_next_power_of_two(width
);
422 uint32_t pot_height
= util_next_power_of_two(height
);
424 uint32_t utile_w
= vc4_utile_width(rsc
->cpp
);
425 uint32_t utile_h
= vc4_utile_height(rsc
->cpp
);
427 for (int i
= prsc
->last_level
; i
>= 0; i
--) {
428 struct vc4_resource_slice
*slice
= &rsc
->slices
[i
];
430 uint32_t level_width
, level_height
;
433 level_height
= height
;
435 level_width
= u_minify(pot_width
, i
);
436 level_height
= u_minify(pot_height
, i
);
440 slice
->tiling
= VC4_TILING_FORMAT_LINEAR
;
441 if (prsc
->nr_samples
> 1) {
442 /* MSAA (4x) surfaces are stored as raw tile buffer contents. */
443 level_width
= align(level_width
, 32);
444 level_height
= align(level_height
, 32);
446 level_width
= align(level_width
, utile_w
);
449 if (vc4_size_is_lt(level_width
, level_height
,
451 slice
->tiling
= VC4_TILING_FORMAT_LT
;
452 level_width
= align(level_width
, utile_w
);
453 level_height
= align(level_height
, utile_h
);
455 slice
->tiling
= VC4_TILING_FORMAT_T
;
456 level_width
= align(level_width
,
458 level_height
= align(level_height
,
463 slice
->offset
= offset
;
464 slice
->stride
= (level_width
* rsc
->cpp
*
465 MAX2(prsc
->nr_samples
, 1));
466 slice
->size
= level_height
* slice
->stride
;
468 offset
+= slice
->size
;
471 static const char tiling_chars
[] = {
472 [VC4_TILING_FORMAT_LINEAR
] = 'R',
473 [VC4_TILING_FORMAT_LT
] = 'L',
474 [VC4_TILING_FORMAT_T
] = 'T'
477 "rsc setup %p (format %s: vc4 %d), %dx%d: "
478 "level %d (%c) -> %dx%d, stride %d@0x%08x\n",
480 util_format_short_name(prsc
->format
),
482 prsc
->width0
, prsc
->height0
,
483 i
, tiling_chars
[slice
->tiling
],
484 level_width
, level_height
,
485 slice
->stride
, slice
->offset
);
489 /* The texture base pointer that has to point to level 0 doesn't have
490 * intra-page bits, so we have to align it, and thus shift up all the
493 uint32_t page_align_offset
= (align(rsc
->slices
[0].offset
, 4096) -
494 rsc
->slices
[0].offset
);
495 if (page_align_offset
) {
496 for (int i
= 0; i
<= prsc
->last_level
; i
++)
497 rsc
->slices
[i
].offset
+= page_align_offset
;
500 /* Cube map faces appear as whole miptrees at a page-aligned offset
501 * from the first face's miptree.
503 if (prsc
->target
== PIPE_TEXTURE_CUBE
) {
504 rsc
->cube_map_stride
= align(rsc
->slices
[0].offset
+
505 rsc
->slices
[0].size
, 4096);
509 static struct vc4_resource
*
510 vc4_resource_setup(struct pipe_screen
*pscreen
,
511 const struct pipe_resource
*tmpl
)
513 struct vc4_resource
*rsc
= CALLOC_STRUCT(vc4_resource
);
516 struct pipe_resource
*prsc
= &rsc
->base
;
520 pipe_reference_init(&prsc
->reference
, 1);
521 prsc
->screen
= pscreen
;
523 if (prsc
->nr_samples
<= 1)
524 rsc
->cpp
= util_format_get_blocksize(tmpl
->format
);
526 rsc
->cpp
= sizeof(uint32_t);
533 static enum vc4_texture_data_type
534 get_resource_texture_format(struct pipe_resource
*prsc
)
536 struct vc4_resource
*rsc
= vc4_resource(prsc
);
537 uint8_t format
= vc4_get_tex_format(prsc
->format
);
540 if (prsc
->nr_samples
> 1) {
543 assert(format
== VC4_TEXTURE_TYPE_RGBA8888
);
544 return VC4_TEXTURE_TYPE_RGBA32R
;
551 struct pipe_resource
*
552 vc4_resource_create(struct pipe_screen
*pscreen
,
553 const struct pipe_resource
*tmpl
)
555 struct vc4_resource
*rsc
= vc4_resource_setup(pscreen
, tmpl
);
556 struct pipe_resource
*prsc
= &rsc
->base
;
558 /* We have to make shared be untiled, since we don't have any way to
559 * communicate metadata about tiling currently.
561 if (tmpl
->target
== PIPE_BUFFER
||
562 tmpl
->nr_samples
> 1 ||
563 (tmpl
->bind
& (PIPE_BIND_SCANOUT
|
566 PIPE_BIND_CURSOR
))) {
572 if (tmpl
->target
!= PIPE_BUFFER
)
573 rsc
->vc4_format
= get_resource_texture_format(prsc
);
575 vc4_setup_slices(rsc
);
576 if (!vc4_resource_bo_alloc(rsc
))
581 vc4_resource_destroy(pscreen
, prsc
);
585 static struct pipe_resource
*
586 vc4_resource_from_handle(struct pipe_screen
*pscreen
,
587 const struct pipe_resource
*tmpl
,
588 struct winsys_handle
*whandle
,
591 struct vc4_screen
*screen
= vc4_screen(pscreen
);
592 struct vc4_resource
*rsc
= vc4_resource_setup(pscreen
, tmpl
);
593 struct pipe_resource
*prsc
= &rsc
->base
;
594 struct vc4_resource_slice
*slice
= &rsc
->slices
[0];
595 uint32_t expected_stride
=
596 align(prsc
->width0
, vc4_utile_width(rsc
->cpp
)) * rsc
->cpp
;
601 if (whandle
->stride
!= expected_stride
) {
602 static bool warned
= false;
606 "Attempting to import %dx%d %s with "
607 "unsupported stride %d instead of %d\n",
608 prsc
->width0
, prsc
->height0
,
609 util_format_short_name(prsc
->format
),
618 if (whandle
->offset
!= 0) {
620 "Attempt to import unsupported winsys offset %u\n",
625 switch (whandle
->type
) {
626 case DRM_API_HANDLE_TYPE_SHARED
:
627 rsc
->bo
= vc4_bo_open_name(screen
,
628 whandle
->handle
, whandle
->stride
);
630 case DRM_API_HANDLE_TYPE_FD
:
631 rsc
->bo
= vc4_bo_open_dmabuf(screen
,
632 whandle
->handle
, whandle
->stride
);
636 "Attempt to import unsupported handle type %d\n",
643 slice
->stride
= whandle
->stride
;
644 slice
->tiling
= VC4_TILING_FORMAT_LINEAR
;
646 rsc
->vc4_format
= get_resource_texture_format(prsc
);
650 "rsc import %p (format %d), %dx%d: "
651 "level 0 (R) -> stride %d@0x%08x\n",
652 rsc
, rsc
->vc4_format
,
653 prsc
->width0
, prsc
->height0
,
654 slice
->stride
, slice
->offset
);
660 vc4_resource_destroy(pscreen
, prsc
);
664 static struct pipe_surface
*
665 vc4_create_surface(struct pipe_context
*pctx
,
666 struct pipe_resource
*ptex
,
667 const struct pipe_surface
*surf_tmpl
)
669 struct vc4_surface
*surface
= CALLOC_STRUCT(vc4_surface
);
670 struct vc4_resource
*rsc
= vc4_resource(ptex
);
675 assert(surf_tmpl
->u
.tex
.first_layer
== surf_tmpl
->u
.tex
.last_layer
);
677 struct pipe_surface
*psurf
= &surface
->base
;
678 unsigned level
= surf_tmpl
->u
.tex
.level
;
680 pipe_reference_init(&psurf
->reference
, 1);
681 pipe_resource_reference(&psurf
->texture
, ptex
);
683 psurf
->context
= pctx
;
684 psurf
->format
= surf_tmpl
->format
;
685 psurf
->width
= u_minify(ptex
->width0
, level
);
686 psurf
->height
= u_minify(ptex
->height0
, level
);
687 psurf
->u
.tex
.level
= level
;
688 psurf
->u
.tex
.first_layer
= surf_tmpl
->u
.tex
.first_layer
;
689 psurf
->u
.tex
.last_layer
= surf_tmpl
->u
.tex
.last_layer
;
690 surface
->offset
= (rsc
->slices
[level
].offset
+
691 psurf
->u
.tex
.first_layer
* rsc
->cube_map_stride
);
692 surface
->tiling
= rsc
->slices
[level
].tiling
;
694 return &surface
->base
;
698 vc4_surface_destroy(struct pipe_context
*pctx
, struct pipe_surface
*psurf
)
700 pipe_resource_reference(&psurf
->texture
, NULL
);
705 vc4_dump_surface_non_msaa(struct pipe_surface
*psurf
)
707 struct pipe_resource
*prsc
= psurf
->texture
;
708 struct vc4_resource
*rsc
= vc4_resource(prsc
);
709 uint32_t *map
= vc4_bo_map(rsc
->bo
);
710 uint32_t stride
= rsc
->slices
[0].stride
/ 4;
711 uint32_t width
= psurf
->width
;
712 uint32_t height
= psurf
->height
;
713 uint32_t chunk_w
= width
/ 79;
714 uint32_t chunk_h
= height
/ 40;
715 uint32_t found_colors
[10];
716 uint32_t num_found_colors
= 0;
718 if (rsc
->vc4_format
!= VC4_TEXTURE_TYPE_RGBA32R
) {
719 fprintf(stderr
, "%s: Unsupported format %s\n",
720 __func__
, util_format_short_name(psurf
->format
));
724 for (int by
= 0; by
< height
; by
+= chunk_h
) {
725 for (int bx
= 0; bx
< width
; bx
+= chunk_w
) {
726 int all_found_color
= -1; /* nothing found */
728 for (int y
= by
; y
< MIN2(height
, by
+ chunk_h
); y
++) {
729 for (int x
= bx
; x
< MIN2(width
, bx
+ chunk_w
); x
++) {
730 uint32_t pix
= map
[y
* stride
+ x
];
733 for (i
= 0; i
< num_found_colors
; i
++) {
734 if (pix
== found_colors
[i
])
737 if (i
== num_found_colors
&&
739 ARRAY_SIZE(found_colors
)) {
740 found_colors
[num_found_colors
++] = pix
;
743 if (i
< num_found_colors
) {
744 if (all_found_color
== -1)
746 else if (i
!= all_found_color
)
747 all_found_color
= ARRAY_SIZE(found_colors
);
751 /* If all pixels for this chunk have a consistent
752 * value, then print a character for it. Either a
753 * fixed name (particularly common for piglit tests),
754 * or a runtime-generated number.
756 if (all_found_color
>= 0 &&
757 all_found_color
< ARRAY_SIZE(found_colors
)) {
758 static const struct {
770 for (i
= 0; i
< ARRAY_SIZE(named_colors
); i
++) {
771 if (named_colors
[i
].val
==
772 found_colors
[all_found_color
]) {
773 fprintf(stderr
, "%s",
778 /* For unnamed colors, print a number and the
779 * numbers will have values printed at the
782 if (i
== ARRAY_SIZE(named_colors
)) {
783 fprintf(stderr
, "%c",
784 '0' + all_found_color
);
787 /* If there's no consistent color, print this.
789 fprintf(stderr
, ".");
792 fprintf(stderr
, "\n");
795 for (int i
= 0; i
< num_found_colors
; i
++) {
796 fprintf(stderr
, "color %d: 0x%08x\n", i
, found_colors
[i
]);
801 vc4_surface_msaa_get_sample(struct pipe_surface
*psurf
,
802 uint32_t x
, uint32_t y
, uint32_t sample
)
804 struct pipe_resource
*prsc
= psurf
->texture
;
805 struct vc4_resource
*rsc
= vc4_resource(prsc
);
806 uint32_t tile_w
= 32, tile_h
= 32;
807 uint32_t tiles_w
= DIV_ROUND_UP(psurf
->width
, 32);
809 uint32_t tile_x
= x
/ tile_w
;
810 uint32_t tile_y
= y
/ tile_h
;
811 uint32_t *tile
= (vc4_bo_map(rsc
->bo
) +
812 VC4_TILE_BUFFER_SIZE
* (tile_y
* tiles_w
+ tile_x
));
813 uint32_t subtile_x
= x
% tile_w
;
814 uint32_t subtile_y
= y
% tile_h
;
816 uint32_t quad_samples
= VC4_MAX_SAMPLES
* 4;
817 uint32_t tile_stride
= quad_samples
* tile_w
/ 2;
819 return *((uint32_t *)tile
+
820 (subtile_y
>> 1) * tile_stride
+
821 (subtile_x
>> 1) * quad_samples
+
822 ((subtile_y
& 1) << 1) +
828 vc4_dump_surface_msaa_char(struct pipe_surface
*psurf
,
829 uint32_t start_x
, uint32_t start_y
,
830 uint32_t w
, uint32_t h
)
832 bool all_same_color
= true;
833 uint32_t all_pix
= 0;
835 for (int y
= start_y
; y
< start_y
+ h
; y
++) {
836 for (int x
= start_x
; x
< start_x
+ w
; x
++) {
837 for (int s
= 0; s
< VC4_MAX_SAMPLES
; s
++) {
838 uint32_t pix
= vc4_surface_msaa_get_sample(psurf
,
841 if (x
== start_x
&& y
== start_y
)
843 else if (all_pix
!= pix
)
844 all_same_color
= false;
848 if (all_same_color
) {
849 static const struct {
861 for (i
= 0; i
< ARRAY_SIZE(named_colors
); i
++) {
862 if (named_colors
[i
].val
== all_pix
) {
863 fprintf(stderr
, "%s",
868 fprintf(stderr
, "x");
870 fprintf(stderr
, ".");
875 vc4_dump_surface_msaa(struct pipe_surface
*psurf
)
877 uint32_t tile_w
= 32, tile_h
= 32;
878 uint32_t tiles_w
= DIV_ROUND_UP(psurf
->width
, tile_w
);
879 uint32_t tiles_h
= DIV_ROUND_UP(psurf
->height
, tile_h
);
880 uint32_t char_w
= 140, char_h
= 60;
881 uint32_t char_w_per_tile
= char_w
/ tiles_w
- 1;
882 uint32_t char_h_per_tile
= char_h
/ tiles_h
- 1;
883 uint32_t found_colors
[10];
884 uint32_t num_found_colors
= 0;
886 fprintf(stderr
, "Surface: %dx%d (%dx MSAA)\n",
887 psurf
->width
, psurf
->height
, psurf
->texture
->nr_samples
);
889 for (int x
= 0; x
< (char_w_per_tile
+ 1) * tiles_w
; x
++)
890 fprintf(stderr
, "-");
891 fprintf(stderr
, "\n");
893 for (int ty
= 0; ty
< psurf
->height
; ty
+= tile_h
) {
894 for (int y
= 0; y
< char_h_per_tile
; y
++) {
896 for (int tx
= 0; tx
< psurf
->width
; tx
+= tile_w
) {
897 for (int x
= 0; x
< char_w_per_tile
; x
++) {
898 uint32_t bx1
= (x
* tile_w
/
900 uint32_t bx2
= ((x
+ 1) * tile_w
/
902 uint32_t by1
= (y
* tile_h
/
904 uint32_t by2
= ((y
+ 1) * tile_h
/
907 vc4_dump_surface_msaa_char(psurf
,
913 fprintf(stderr
, "|");
915 fprintf(stderr
, "\n");
918 for (int x
= 0; x
< (char_w_per_tile
+ 1) * tiles_w
; x
++)
919 fprintf(stderr
, "-");
920 fprintf(stderr
, "\n");
923 for (int i
= 0; i
< num_found_colors
; i
++) {
924 fprintf(stderr
, "color %d: 0x%08x\n", i
, found_colors
[i
]);
928 /** Debug routine to dump the contents of an 8888 surface to the console */
930 vc4_dump_surface(struct pipe_surface
*psurf
)
935 if (psurf
->texture
->nr_samples
> 1)
936 vc4_dump_surface_msaa(psurf
);
938 vc4_dump_surface_non_msaa(psurf
);
942 vc4_flush_resource(struct pipe_context
*pctx
, struct pipe_resource
*resource
)
944 /* All calls to flush_resource are followed by a flush of the context,
945 * so there's nothing to do.
950 vc4_update_shadow_baselevel_texture(struct pipe_context
*pctx
,
951 struct pipe_sampler_view
*view
)
953 struct vc4_resource
*shadow
= vc4_resource(view
->texture
);
954 struct vc4_resource
*orig
= vc4_resource(shadow
->shadow_parent
);
957 if (shadow
->writes
== orig
->writes
&& orig
->bo
->private)
960 perf_debug("Updating %dx%d@%d shadow texture due to %s\n",
961 orig
->base
.width0
, orig
->base
.height0
,
962 view
->u
.tex
.first_level
,
963 view
->u
.tex
.first_level
? "base level" : "raster layout");
965 for (int i
= 0; i
<= shadow
->base
.last_level
; i
++) {
966 unsigned width
= u_minify(shadow
->base
.width0
, i
);
967 unsigned height
= u_minify(shadow
->base
.height0
, i
);
968 struct pipe_blit_info info
= {
970 .resource
= &shadow
->base
,
980 .format
= shadow
->base
.format
,
983 .resource
= &orig
->base
,
984 .level
= view
->u
.tex
.first_level
+ i
,
993 .format
= orig
->base
.format
,
997 pctx
->blit(pctx
, &info
);
1000 shadow
->writes
= orig
->writes
;
1004 * Converts a 4-byte index buffer to 2 bytes.
1006 * Since GLES2 only has support for 1 and 2-byte indices, the hardware doesn't
1007 * include 4-byte index support, and we have to shrink it down.
1009 * There's no fallback support for when indices end up being larger than 2^16,
1010 * though it will at least assertion fail. Also, if the original index data
1011 * was in user memory, it would be nice to not have uploaded it to a VBO
1012 * before translating.
1014 struct pipe_resource
*
1015 vc4_get_shadow_index_buffer(struct pipe_context
*pctx
,
1016 const struct pipe_draw_info
*info
,
1019 uint32_t *shadow_offset
)
1021 struct vc4_context
*vc4
= vc4_context(pctx
);
1022 struct vc4_resource
*orig
= vc4_resource(info
->index
.resource
);
1023 perf_debug("Fallback conversion for %d uint indices\n", count
);
1026 struct pipe_resource
*shadow_rsc
= NULL
;
1027 u_upload_alloc(vc4
->uploader
, 0, count
* 2, 4,
1028 shadow_offset
, &shadow_rsc
, &data
);
1029 uint16_t *dst
= data
;
1031 struct pipe_transfer
*src_transfer
= NULL
;
1032 const uint32_t *src
;
1033 if (info
->has_user_indices
) {
1034 src
= info
->index
.user
;
1036 src
= pipe_buffer_map_range(pctx
, &orig
->base
,
1039 PIPE_TRANSFER_READ
, &src_transfer
);
1042 for (int i
= 0; i
< count
; i
++) {
1043 uint32_t src_index
= src
[i
];
1044 assert(src_index
<= 0xffff);
1049 pctx
->transfer_unmap(pctx
, src_transfer
);
1055 vc4_resource_screen_init(struct pipe_screen
*pscreen
)
1057 pscreen
->resource_create
= vc4_resource_create
;
1058 pscreen
->resource_from_handle
= vc4_resource_from_handle
;
1059 pscreen
->resource_destroy
= u_resource_destroy_vtbl
;
1060 pscreen
->resource_get_handle
= vc4_resource_get_handle
;
1061 pscreen
->resource_destroy
= vc4_resource_destroy
;
1065 vc4_resource_context_init(struct pipe_context
*pctx
)
1067 pctx
->transfer_map
= vc4_resource_transfer_map
;
1068 pctx
->transfer_flush_region
= u_default_transfer_flush_region
;
1069 pctx
->transfer_unmap
= vc4_resource_transfer_unmap
;
1070 pctx
->buffer_subdata
= u_default_buffer_subdata
;
1071 pctx
->texture_subdata
= u_default_texture_subdata
;
1072 pctx
->create_surface
= vc4_create_surface
;
1073 pctx
->surface_destroy
= vc4_surface_destroy
;
1074 pctx
->resource_copy_region
= util_resource_copy_region
;
1075 pctx
->blit
= vc4_blit
;
1076 pctx
->flush_resource
= vc4_flush_resource
;