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_memory.h"
26 #include "util/u_format.h"
27 #include "util/u_inlines.h"
28 #include "util/u_surface.h"
29 #include "util/u_upload_mgr.h"
31 #include "vc4_screen.h"
32 #include "vc4_context.h"
33 #include "vc4_resource.h"
34 #include "vc4_tiling.h"
36 static bool miptree_debug
= false;
39 vc4_resource_bo_alloc(struct vc4_resource
*rsc
)
41 struct pipe_resource
*prsc
= &rsc
->base
.b
;
42 struct pipe_screen
*pscreen
= prsc
->screen
;
45 fprintf(stderr
, "alloc %p: size %d + offset %d -> %d\n",
48 rsc
->slices
[0].offset
,
49 rsc
->slices
[0].offset
+
51 rsc
->cube_map_stride
* (prsc
->array_size
- 1));
54 vc4_bo_unreference(&rsc
->bo
);
55 rsc
->bo
= vc4_bo_alloc(vc4_screen(pscreen
),
56 rsc
->slices
[0].offset
+
58 rsc
->cube_map_stride
* (prsc
->array_size
- 1),
63 vc4_resource_transfer_unmap(struct pipe_context
*pctx
,
64 struct pipe_transfer
*ptrans
)
66 struct vc4_context
*vc4
= vc4_context(pctx
);
67 struct vc4_transfer
*trans
= vc4_transfer(ptrans
);
68 struct pipe_resource
*prsc
= ptrans
->resource
;
69 struct vc4_resource
*rsc
= vc4_resource(prsc
);
70 struct vc4_resource_slice
*slice
= &rsc
->slices
[ptrans
->level
];
73 if (ptrans
->usage
& PIPE_TRANSFER_WRITE
) {
74 vc4_store_tiled_image(rsc
->bo
->map
+ slice
->offset
+
75 ptrans
->box
.z
* rsc
->cube_map_stride
,
77 trans
->map
, ptrans
->stride
,
78 slice
->tiling
, rsc
->cpp
,
84 pipe_resource_reference(&ptrans
->resource
, NULL
);
85 util_slab_free(&vc4
->transfer_pool
, ptrans
);
89 vc4_resource_transfer_map(struct pipe_context
*pctx
,
90 struct pipe_resource
*prsc
,
91 unsigned level
, unsigned usage
,
92 const struct pipe_box
*box
,
93 struct pipe_transfer
**pptrans
)
95 struct vc4_context
*vc4
= vc4_context(pctx
);
96 struct vc4_resource
*rsc
= vc4_resource(prsc
);
97 struct vc4_resource_slice
*slice
= &rsc
->slices
[level
];
98 struct vc4_transfer
*trans
;
99 struct pipe_transfer
*ptrans
;
100 enum pipe_format format
= prsc
->format
;
103 if (usage
& PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
) {
104 vc4_resource_bo_alloc(rsc
);
106 /* If it might be bound as one of our vertex buffers, make
107 * sure we re-emit vertex buffer state.
109 if (prsc
->bind
& PIPE_BIND_VERTEX_BUFFER
)
110 vc4
->dirty
|= VC4_DIRTY_VTXBUF
;
111 } else if (!(usage
& PIPE_TRANSFER_UNSYNCHRONIZED
)) {
112 if (vc4_cl_references_bo(pctx
, rsc
->bo
)) {
113 if ((usage
& PIPE_TRANSFER_DISCARD_RANGE
) &&
114 prsc
->last_level
== 0 &&
115 prsc
->width0
== box
->width
&&
116 prsc
->height0
== box
->height
&&
117 prsc
->depth0
== box
->depth
) {
118 vc4_resource_bo_alloc(rsc
);
119 if (prsc
->bind
& PIPE_BIND_VERTEX_BUFFER
)
120 vc4
->dirty
|= VC4_DIRTY_VTXBUF
;
127 if (usage
& PIPE_TRANSFER_WRITE
)
130 trans
= util_slab_alloc(&vc4
->transfer_pool
);
134 /* XXX: Handle DONTBLOCK, DISCARD_RANGE, PERSISTENT, COHERENT. */
136 /* util_slab_alloc() doesn't zero: */
137 memset(trans
, 0, sizeof(*trans
));
138 ptrans
= &trans
->base
;
140 pipe_resource_reference(&ptrans
->resource
, prsc
);
141 ptrans
->level
= level
;
142 ptrans
->usage
= usage
;
145 /* Note that the current kernel implementation is synchronous, so no
146 * need to do syncing stuff here yet.
149 if (usage
& PIPE_TRANSFER_UNSYNCHRONIZED
)
150 buf
= vc4_bo_map_unsynchronized(rsc
->bo
);
152 buf
= vc4_bo_map(rsc
->bo
);
154 fprintf(stderr
, "Failed to map bo\n");
161 uint32_t utile_w
= vc4_utile_width(rsc
->cpp
);
162 uint32_t utile_h
= vc4_utile_height(rsc
->cpp
);
164 /* No direct mappings of tiled, since we need to manually
167 if (usage
& PIPE_TRANSFER_MAP_DIRECTLY
)
170 /* We need to align the box to utile boundaries, since that's
171 * what load/store operate on.
173 uint32_t orig_width
= ptrans
->box
.width
;
174 uint32_t orig_height
= ptrans
->box
.height
;
175 uint32_t box_start_x
= ptrans
->box
.x
& (utile_w
- 1);
176 uint32_t box_start_y
= ptrans
->box
.y
& (utile_h
- 1);
177 ptrans
->box
.width
+= box_start_x
;
178 ptrans
->box
.x
-= box_start_x
;
179 ptrans
->box
.height
+= box_start_y
;
180 ptrans
->box
.y
-= box_start_y
;
181 ptrans
->box
.width
= align(ptrans
->box
.width
, utile_w
);
182 ptrans
->box
.height
= align(ptrans
->box
.height
, utile_h
);
184 ptrans
->stride
= ptrans
->box
.width
* rsc
->cpp
;
185 ptrans
->layer_stride
= ptrans
->stride
;
187 trans
->map
= malloc(ptrans
->stride
* ptrans
->box
.height
);
188 if (usage
& PIPE_TRANSFER_READ
||
189 ptrans
->box
.width
!= orig_width
||
190 ptrans
->box
.height
!= orig_height
) {
191 vc4_load_tiled_image(trans
->map
, ptrans
->stride
,
192 buf
+ slice
->offset
+
193 box
->z
* rsc
->cube_map_stride
,
195 slice
->tiling
, rsc
->cpp
,
199 box_start_x
* rsc
->cpp
+
200 box_start_y
* ptrans
->stride
);
202 ptrans
->stride
= slice
->stride
;
203 ptrans
->layer_stride
= ptrans
->stride
;
205 return buf
+ slice
->offset
+
206 box
->y
/ util_format_get_blockheight(format
) * ptrans
->stride
+
207 box
->x
/ util_format_get_blockwidth(format
) * rsc
->cpp
+
208 box
->z
* rsc
->cube_map_stride
;
213 vc4_resource_transfer_unmap(pctx
, ptrans
);
218 vc4_resource_destroy(struct pipe_screen
*pscreen
,
219 struct pipe_resource
*prsc
)
221 struct vc4_resource
*rsc
= vc4_resource(prsc
);
222 pipe_resource_reference(&rsc
->shadow_parent
, NULL
);
223 vc4_bo_unreference(&rsc
->bo
);
228 vc4_resource_get_handle(struct pipe_screen
*pscreen
,
229 struct pipe_resource
*prsc
,
230 struct winsys_handle
*handle
)
232 struct vc4_resource
*rsc
= vc4_resource(prsc
);
234 return vc4_screen_bo_get_handle(pscreen
, rsc
->bo
, rsc
->slices
[0].stride
,
238 static const struct u_resource_vtbl vc4_resource_vtbl
= {
239 .resource_get_handle
= vc4_resource_get_handle
,
240 .resource_destroy
= vc4_resource_destroy
,
241 .transfer_map
= vc4_resource_transfer_map
,
242 .transfer_flush_region
= u_default_transfer_flush_region
,
243 .transfer_unmap
= vc4_resource_transfer_unmap
,
244 .transfer_inline_write
= u_default_transfer_inline_write
,
248 vc4_setup_slices(struct vc4_resource
*rsc
)
250 struct pipe_resource
*prsc
= &rsc
->base
.b
;
251 uint32_t width
= prsc
->width0
;
252 uint32_t height
= prsc
->height0
;
253 uint32_t pot_width
= util_next_power_of_two(width
);
254 uint32_t pot_height
= util_next_power_of_two(height
);
256 uint32_t utile_w
= vc4_utile_width(rsc
->cpp
);
257 uint32_t utile_h
= vc4_utile_height(rsc
->cpp
);
259 for (int i
= prsc
->last_level
; i
>= 0; i
--) {
260 struct vc4_resource_slice
*slice
= &rsc
->slices
[i
];
262 uint32_t level_width
, level_height
;
265 level_height
= height
;
267 level_width
= u_minify(pot_width
, i
);
268 level_height
= u_minify(pot_height
, i
);
272 slice
->tiling
= VC4_TILING_FORMAT_LINEAR
;
273 level_width
= align(level_width
, utile_w
);
275 if (vc4_size_is_lt(level_width
, level_height
,
277 slice
->tiling
= VC4_TILING_FORMAT_LT
;
278 level_width
= align(level_width
, utile_w
);
279 level_height
= align(level_height
, utile_h
);
281 slice
->tiling
= VC4_TILING_FORMAT_T
;
282 level_width
= align(level_width
,
284 level_height
= align(level_height
,
289 slice
->offset
= offset
;
290 slice
->stride
= level_width
* rsc
->cpp
;
291 slice
->size
= level_height
* slice
->stride
;
293 offset
+= slice
->size
;
296 static const char tiling_chars
[] = {
297 [VC4_TILING_FORMAT_LINEAR
] = 'R',
298 [VC4_TILING_FORMAT_LT
] = 'L',
299 [VC4_TILING_FORMAT_T
] = 'T'
302 "rsc setup %p (format %d), %dx%d: "
303 "level %d (%c) -> %dx%d, stride %d@0x%08x\n",
304 rsc
, rsc
->vc4_format
,
305 prsc
->width0
, prsc
->height0
,
306 i
, tiling_chars
[slice
->tiling
],
307 level_width
, level_height
,
308 slice
->stride
, slice
->offset
);
312 /* The texture base pointer that has to point to level 0 doesn't have
313 * intra-page bits, so we have to align it, and thus shift up all the
316 uint32_t page_align_offset
= (align(rsc
->slices
[0].offset
, 4096) -
317 rsc
->slices
[0].offset
);
318 if (page_align_offset
) {
319 for (int i
= 0; i
<= prsc
->last_level
; i
++)
320 rsc
->slices
[i
].offset
+= page_align_offset
;
323 /* Cube map faces appear as whole miptrees at a page-aligned offset
324 * from the first face's miptree.
326 if (prsc
->target
== PIPE_TEXTURE_CUBE
) {
327 rsc
->cube_map_stride
= align(rsc
->slices
[0].offset
+
328 rsc
->slices
[0].size
, 4096);
332 static struct vc4_resource
*
333 vc4_resource_setup(struct pipe_screen
*pscreen
,
334 const struct pipe_resource
*tmpl
)
336 struct vc4_resource
*rsc
= CALLOC_STRUCT(vc4_resource
);
339 struct pipe_resource
*prsc
= &rsc
->base
.b
;
343 pipe_reference_init(&prsc
->reference
, 1);
344 prsc
->screen
= pscreen
;
346 rsc
->base
.vtbl
= &vc4_resource_vtbl
;
347 rsc
->cpp
= util_format_get_blocksize(tmpl
->format
);
354 static enum vc4_texture_data_type
355 get_resource_texture_format(struct pipe_resource
*prsc
)
357 struct vc4_resource
*rsc
= vc4_resource(prsc
);
358 uint8_t format
= vc4_get_tex_format(prsc
->format
);
361 assert(format
== VC4_TEXTURE_TYPE_RGBA8888
);
362 return VC4_TEXTURE_TYPE_RGBA32R
;
368 struct pipe_resource
*
369 vc4_resource_create(struct pipe_screen
*pscreen
,
370 const struct pipe_resource
*tmpl
)
372 struct vc4_resource
*rsc
= vc4_resource_setup(pscreen
, tmpl
);
373 struct pipe_resource
*prsc
= &rsc
->base
.b
;
375 /* We have to make shared be untiled, since we don't have any way to
376 * communicate metadata about tiling currently.
378 if (tmpl
->target
== PIPE_BUFFER
||
379 (tmpl
->bind
& (PIPE_BIND_SCANOUT
|
382 PIPE_BIND_CURSOR
))) {
388 if (tmpl
->target
!= PIPE_BUFFER
)
389 rsc
->vc4_format
= get_resource_texture_format(prsc
);
391 vc4_setup_slices(rsc
);
392 vc4_resource_bo_alloc(rsc
);
398 vc4_resource_destroy(pscreen
, prsc
);
402 static struct pipe_resource
*
403 vc4_resource_from_handle(struct pipe_screen
*pscreen
,
404 const struct pipe_resource
*tmpl
,
405 struct winsys_handle
*handle
)
407 struct vc4_resource
*rsc
= vc4_resource_setup(pscreen
, tmpl
);
408 struct pipe_resource
*prsc
= &rsc
->base
.b
;
409 struct vc4_resource_slice
*slice
= &rsc
->slices
[0];
415 rsc
->bo
= vc4_screen_bo_from_handle(pscreen
, handle
);
419 if (!using_vc4_simulator
)
420 slice
->stride
= handle
->stride
;
422 slice
->stride
= align(prsc
->width0
* rsc
->cpp
, 16);
424 slice
->tiling
= VC4_TILING_FORMAT_LINEAR
;
426 rsc
->vc4_format
= get_resource_texture_format(prsc
);
430 "rsc import %p (format %d), %dx%d: "
431 "level 0 (R) -> stride %d@0x%08x\n",
432 rsc
, rsc
->vc4_format
,
433 prsc
->width0
, prsc
->height0
,
434 slice
->stride
, slice
->offset
);
440 vc4_resource_destroy(pscreen
, prsc
);
444 static struct pipe_surface
*
445 vc4_create_surface(struct pipe_context
*pctx
,
446 struct pipe_resource
*ptex
,
447 const struct pipe_surface
*surf_tmpl
)
449 struct vc4_surface
*surface
= CALLOC_STRUCT(vc4_surface
);
450 struct vc4_resource
*rsc
= vc4_resource(ptex
);
455 assert(surf_tmpl
->u
.tex
.first_layer
== surf_tmpl
->u
.tex
.last_layer
);
457 struct pipe_surface
*psurf
= &surface
->base
;
458 unsigned level
= surf_tmpl
->u
.tex
.level
;
460 pipe_reference_init(&psurf
->reference
, 1);
461 pipe_resource_reference(&psurf
->texture
, ptex
);
463 psurf
->context
= pctx
;
464 psurf
->format
= surf_tmpl
->format
;
465 psurf
->width
= u_minify(ptex
->width0
, level
);
466 psurf
->height
= u_minify(ptex
->height0
, level
);
467 psurf
->u
.tex
.level
= level
;
468 psurf
->u
.tex
.first_layer
= surf_tmpl
->u
.tex
.first_layer
;
469 psurf
->u
.tex
.last_layer
= surf_tmpl
->u
.tex
.last_layer
;
470 surface
->offset
= rsc
->slices
[level
].offset
;
471 surface
->tiling
= rsc
->slices
[level
].tiling
;
473 return &surface
->base
;
477 vc4_surface_destroy(struct pipe_context
*pctx
, struct pipe_surface
*psurf
)
479 pipe_resource_reference(&psurf
->texture
, NULL
);
483 /** Debug routine to dump the contents of an 8888 surface to the console */
485 vc4_dump_surface(struct pipe_surface
*psurf
)
490 struct pipe_resource
*prsc
= psurf
->texture
;
491 struct vc4_resource
*rsc
= vc4_resource(prsc
);
492 uint32_t *map
= vc4_bo_map(rsc
->bo
);
493 uint32_t stride
= rsc
->slices
[0].stride
/ 4;
494 uint32_t width
= psurf
->width
;
495 uint32_t height
= psurf
->height
;
496 uint32_t chunk_w
= width
/ 79;
497 uint32_t chunk_h
= height
/ 40;
498 uint32_t found_colors
[10];
499 uint32_t num_found_colors
= 0;
501 if (rsc
->vc4_format
!= VC4_TEXTURE_TYPE_RGBA32R
) {
502 fprintf(stderr
, "%s: Unsupported format %s\n",
503 __func__
, util_format_short_name(psurf
->format
));
507 for (int by
= 0; by
< height
; by
+= chunk_h
) {
508 for (int bx
= 0; bx
< width
; bx
+= chunk_w
) {
509 int all_found_color
= -1; /* nothing found */
511 for (int y
= by
; y
< MIN2(height
, by
+ chunk_h
); y
++) {
512 for (int x
= bx
; x
< MIN2(width
, bx
+ chunk_w
); x
++) {
513 uint32_t pix
= map
[y
* stride
+ x
];
516 for (i
= 0; i
< num_found_colors
; i
++) {
517 if (pix
== found_colors
[i
])
520 if (i
== num_found_colors
&&
522 ARRAY_SIZE(found_colors
)) {
523 found_colors
[num_found_colors
++] = pix
;
526 if (i
< num_found_colors
) {
527 if (all_found_color
== -1)
529 else if (i
!= all_found_color
)
530 all_found_color
= ARRAY_SIZE(found_colors
);
534 /* If all pixels for this chunk have a consistent
535 * value, then print a character for it. Either a
536 * fixed name (particularly common for piglit tests),
537 * or a runtime-generated number.
539 if (all_found_color
>= 0 &&
540 all_found_color
< ARRAY_SIZE(found_colors
)) {
541 static const struct {
553 for (i
= 0; i
< ARRAY_SIZE(named_colors
); i
++) {
554 if (named_colors
[i
].val
==
555 found_colors
[all_found_color
]) {
556 fprintf(stderr
, "%s",
561 /* For unnamed colors, print a number and the
562 * numbers will have values printed at the
565 if (i
== ARRAY_SIZE(named_colors
)) {
566 fprintf(stderr
, "%c",
567 '0' + all_found_color
);
570 /* If there's no consistent color, print this.
572 fprintf(stderr
, ".");
575 fprintf(stderr
, "\n");
578 for (int i
= 0; i
< num_found_colors
; i
++) {
579 fprintf(stderr
, "color %d: 0x%08x\n", i
, found_colors
[i
]);
584 vc4_flush_resource(struct pipe_context
*pctx
, struct pipe_resource
*resource
)
586 /* All calls to flush_resource are followed by a flush of the context,
587 * so there's nothing to do.
592 vc4_update_shadow_baselevel_texture(struct pipe_context
*pctx
,
593 struct pipe_sampler_view
*view
)
595 struct vc4_resource
*shadow
= vc4_resource(view
->texture
);
596 struct vc4_resource
*orig
= vc4_resource(shadow
->shadow_parent
);
599 if (shadow
->writes
== orig
->writes
&& orig
->bo
->private)
602 perf_debug("Updating shadow texture due to %s\n",
603 view
->u
.tex
.first_level
? "base level" : "raster layout");
605 for (int i
= 0; i
<= shadow
->base
.b
.last_level
; i
++) {
606 unsigned width
= u_minify(shadow
->base
.b
.width0
, i
);
607 unsigned height
= u_minify(shadow
->base
.b
.height0
, i
);
608 struct pipe_blit_info info
= {
610 .resource
= &shadow
->base
.b
,
620 .format
= shadow
->base
.b
.format
,
623 .resource
= &orig
->base
.b
,
624 .level
= view
->u
.tex
.first_level
+ i
,
633 .format
= orig
->base
.b
.format
,
637 pctx
->blit(pctx
, &info
);
640 shadow
->writes
= orig
->writes
;
644 * Converts a 4-byte index buffer to 2 bytes.
646 * Since GLES2 only has support for 1 and 2-byte indices, the hardware doesn't
647 * include 4-byte index support, and we have to shrink it down.
649 * There's no fallback support for when indices end up being larger than 2^16,
650 * though it will at least assertion fail. Also, if the original index data
651 * was in user memory, it would be nice to not have uploaded it to a VBO
652 * before translating.
654 struct pipe_resource
*
655 vc4_get_shadow_index_buffer(struct pipe_context
*pctx
,
656 const struct pipe_index_buffer
*ib
,
658 uint32_t *shadow_offset
)
660 struct vc4_context
*vc4
= vc4_context(pctx
);
661 struct vc4_resource
*orig
= vc4_resource(ib
->buffer
);
662 perf_debug("Fallback conversion for %d uint indices\n", count
);
665 struct pipe_resource
*shadow_rsc
= NULL
;
666 u_upload_alloc(vc4
->uploader
, 0, count
* 2,
667 shadow_offset
, &shadow_rsc
, &data
);
668 uint16_t *dst
= data
;
670 struct pipe_transfer
*src_transfer
= NULL
;
672 if (ib
->user_buffer
) {
673 src
= ib
->user_buffer
;
675 src
= pipe_buffer_map_range(pctx
, &orig
->base
.b
,
678 PIPE_TRANSFER_READ
, &src_transfer
);
681 for (int i
= 0; i
< count
; i
++) {
682 uint32_t src_index
= src
[i
];
683 assert(src_index
<= 0xffff);
688 pctx
->transfer_unmap(pctx
, src_transfer
);
694 vc4_resource_screen_init(struct pipe_screen
*pscreen
)
696 pscreen
->resource_create
= vc4_resource_create
;
697 pscreen
->resource_from_handle
= vc4_resource_from_handle
;
698 pscreen
->resource_get_handle
= u_resource_get_handle_vtbl
;
699 pscreen
->resource_destroy
= u_resource_destroy_vtbl
;
703 vc4_resource_context_init(struct pipe_context
*pctx
)
705 pctx
->transfer_map
= u_transfer_map_vtbl
;
706 pctx
->transfer_flush_region
= u_transfer_flush_region_vtbl
;
707 pctx
->transfer_unmap
= u_transfer_unmap_vtbl
;
708 pctx
->transfer_inline_write
= u_transfer_inline_write_vtbl
;
709 pctx
->create_surface
= vc4_create_surface
;
710 pctx
->surface_destroy
= vc4_surface_destroy
;
711 pctx
->resource_copy_region
= util_resource_copy_region
;
712 pctx
->blit
= vc4_blit
;
713 pctx
->flush_resource
= vc4_flush_resource
;