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 "pipe/p_defines.h"
26 #include "util/u_blit.h"
27 #include "util/u_memory.h"
28 #include "util/u_format.h"
29 #include "util/u_inlines.h"
30 #include "util/u_surface.h"
31 #include "util/u_transfer_helper.h"
32 #include "util/u_upload_mgr.h"
34 #include "drm-uapi/drm_fourcc.h"
35 #include "drm-uapi/vc4_drm.h"
36 #include "vc4_screen.h"
37 #include "vc4_context.h"
38 #include "vc4_resource.h"
39 #include "vc4_tiling.h"
42 vc4_resource_bo_alloc(struct vc4_resource
*rsc
)
44 struct pipe_resource
*prsc
= &rsc
->base
;
45 struct pipe_screen
*pscreen
= prsc
->screen
;
48 if (vc4_debug
& VC4_DEBUG_SURFACE
) {
49 fprintf(stderr
, "alloc %p: size %d + offset %d -> %d\n",
52 rsc
->slices
[0].offset
,
53 rsc
->slices
[0].offset
+
55 rsc
->cube_map_stride
* (prsc
->array_size
- 1));
58 bo
= vc4_bo_alloc(vc4_screen(pscreen
),
59 rsc
->slices
[0].offset
+
61 rsc
->cube_map_stride
* (prsc
->array_size
- 1),
64 vc4_bo_unreference(&rsc
->bo
);
73 vc4_resource_transfer_unmap(struct pipe_context
*pctx
,
74 struct pipe_transfer
*ptrans
)
76 struct vc4_context
*vc4
= vc4_context(pctx
);
77 struct vc4_transfer
*trans
= vc4_transfer(ptrans
);
80 struct vc4_resource
*rsc
= vc4_resource(ptrans
->resource
);
81 struct vc4_resource_slice
*slice
= &rsc
->slices
[ptrans
->level
];
83 if (ptrans
->usage
& PIPE_TRANSFER_WRITE
) {
84 vc4_store_tiled_image(rsc
->bo
->map
+ slice
->offset
+
85 ptrans
->box
.z
* rsc
->cube_map_stride
,
87 trans
->map
, ptrans
->stride
,
88 slice
->tiling
, rsc
->cpp
,
94 pipe_resource_reference(&ptrans
->resource
, NULL
);
95 slab_free(&vc4
->transfer_pool
, ptrans
);
99 vc4_resource_transfer_map(struct pipe_context
*pctx
,
100 struct pipe_resource
*prsc
,
101 unsigned level
, unsigned usage
,
102 const struct pipe_box
*box
,
103 struct pipe_transfer
**pptrans
)
105 struct vc4_context
*vc4
= vc4_context(pctx
);
106 struct vc4_resource
*rsc
= vc4_resource(prsc
);
107 struct vc4_transfer
*trans
;
108 struct pipe_transfer
*ptrans
;
109 enum pipe_format format
= prsc
->format
;
112 /* Upgrade DISCARD_RANGE to WHOLE_RESOURCE if the whole resource is
115 if ((usage
& PIPE_TRANSFER_DISCARD_RANGE
) &&
116 !(usage
& PIPE_TRANSFER_UNSYNCHRONIZED
) &&
117 !(prsc
->flags
& PIPE_RESOURCE_FLAG_MAP_PERSISTENT
) &&
118 prsc
->last_level
== 0 &&
119 prsc
->width0
== box
->width
&&
120 prsc
->height0
== box
->height
&&
121 prsc
->depth0
== box
->depth
&&
122 prsc
->array_size
== 1 &&
124 usage
|= PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
;
127 if (usage
& PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
) {
128 if (vc4_resource_bo_alloc(rsc
)) {
129 /* If it might be bound as one of our vertex buffers,
130 * make sure we re-emit vertex buffer state.
132 if (prsc
->bind
& PIPE_BIND_VERTEX_BUFFER
)
133 vc4
->dirty
|= VC4_DIRTY_VTXBUF
;
135 /* If we failed to reallocate, flush users so that we
136 * don't violate any syncing requirements.
138 vc4_flush_jobs_reading_resource(vc4
, prsc
);
140 } else if (!(usage
& PIPE_TRANSFER_UNSYNCHRONIZED
)) {
141 /* If we're writing and the buffer is being used by the CL, we
142 * have to flush the CL first. If we're only reading, we need
143 * to flush if the CL has written our buffer.
145 if (usage
& PIPE_TRANSFER_WRITE
)
146 vc4_flush_jobs_reading_resource(vc4
, prsc
);
148 vc4_flush_jobs_writing_resource(vc4
, prsc
);
151 if (usage
& PIPE_TRANSFER_WRITE
) {
153 rsc
->initialized_buffers
= ~0;
156 trans
= slab_alloc(&vc4
->transfer_pool
);
160 /* XXX: Handle DONTBLOCK, DISCARD_RANGE, PERSISTENT, COHERENT. */
162 /* slab_alloc_st() doesn't zero: */
163 memset(trans
, 0, sizeof(*trans
));
164 ptrans
= &trans
->base
;
166 pipe_resource_reference(&ptrans
->resource
, prsc
);
167 ptrans
->level
= level
;
168 ptrans
->usage
= usage
;
171 if (usage
& PIPE_TRANSFER_UNSYNCHRONIZED
)
172 buf
= vc4_bo_map_unsynchronized(rsc
->bo
);
174 buf
= vc4_bo_map(rsc
->bo
);
176 fprintf(stderr
, "Failed to map bo\n");
182 struct vc4_resource_slice
*slice
= &rsc
->slices
[level
];
184 /* No direct mappings of tiled, since we need to manually
187 if (usage
& PIPE_TRANSFER_MAP_DIRECTLY
)
190 if (format
== PIPE_FORMAT_ETC1_RGB8
) {
191 /* ETC1 is arranged as 64-bit blocks, where each block
192 * is 4x4 pixels. Texture tiling operates on the
193 * 64-bit block the way it would an uncompressed
196 assert(!(ptrans
->box
.x
& 3));
197 assert(!(ptrans
->box
.y
& 3));
200 ptrans
->box
.width
= (ptrans
->box
.width
+ 3) >> 2;
201 ptrans
->box
.height
= (ptrans
->box
.height
+ 3) >> 2;
204 ptrans
->stride
= ptrans
->box
.width
* rsc
->cpp
;
205 ptrans
->layer_stride
= ptrans
->stride
* ptrans
->box
.height
;
207 trans
->map
= malloc(ptrans
->layer_stride
* ptrans
->box
.depth
);
209 if (usage
& PIPE_TRANSFER_READ
) {
210 vc4_load_tiled_image(trans
->map
, ptrans
->stride
,
211 buf
+ slice
->offset
+
212 ptrans
->box
.z
* rsc
->cube_map_stride
,
214 slice
->tiling
, rsc
->cpp
,
219 ptrans
->stride
= slice
->stride
;
220 ptrans
->layer_stride
= ptrans
->stride
;
222 return buf
+ slice
->offset
+
223 ptrans
->box
.y
/ util_format_get_blockheight(format
) * ptrans
->stride
+
224 ptrans
->box
.x
/ util_format_get_blockwidth(format
) * rsc
->cpp
+
225 ptrans
->box
.z
* rsc
->cube_map_stride
;
230 vc4_resource_transfer_unmap(pctx
, ptrans
);
235 vc4_texture_subdata(struct pipe_context
*pctx
,
236 struct pipe_resource
*prsc
,
239 const struct pipe_box
*box
,
242 unsigned layer_stride
)
244 struct vc4_resource
*rsc
= vc4_resource(prsc
);
245 struct vc4_resource_slice
*slice
= &rsc
->slices
[level
];
247 /* For a direct mapping, we can just take the u_transfer path. */
250 (usage
& PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE
)) {
251 return u_default_texture_subdata(pctx
, prsc
, level
, usage
, box
,
252 data
, stride
, layer_stride
);
255 /* Otherwise, map and store the texture data directly into the tiled
259 if (usage
& PIPE_TRANSFER_UNSYNCHRONIZED
)
260 buf
= vc4_bo_map_unsynchronized(rsc
->bo
);
262 buf
= vc4_bo_map(rsc
->bo
);
264 vc4_store_tiled_image(buf
+ slice
->offset
+
265 box
->z
* rsc
->cube_map_stride
,
267 (void *)data
, stride
,
268 slice
->tiling
, rsc
->cpp
,
273 vc4_resource_destroy(struct pipe_screen
*pscreen
,
274 struct pipe_resource
*prsc
)
276 struct vc4_screen
*screen
= vc4_screen(pscreen
);
277 struct vc4_resource
*rsc
= vc4_resource(prsc
);
278 vc4_bo_unreference(&rsc
->bo
);
281 renderonly_scanout_destroy(rsc
->scanout
, screen
->ro
);
287 vc4_resource_get_handle(struct pipe_screen
*pscreen
,
288 struct pipe_context
*pctx
,
289 struct pipe_resource
*prsc
,
290 struct winsys_handle
*whandle
,
293 struct vc4_screen
*screen
= vc4_screen(pscreen
);
294 struct vc4_resource
*rsc
= vc4_resource(prsc
);
296 whandle
->stride
= rsc
->slices
[0].stride
;
299 /* If we're passing some reference to our BO out to some other part of
300 * the system, then we can't do any optimizations about only us being
301 * the ones seeing it (like BO caching or shadow update avoidance).
303 rsc
->bo
->private = false;
306 whandle
->modifier
= DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED
;
308 whandle
->modifier
= DRM_FORMAT_MOD_LINEAR
;
310 switch (whandle
->type
) {
311 case WINSYS_HANDLE_TYPE_SHARED
:
313 /* This could probably be supported, assuming that a
314 * control node was used for pl111.
316 fprintf(stderr
, "flink unsupported with pl111\n");
320 return vc4_bo_flink(rsc
->bo
, &whandle
->handle
);
321 case WINSYS_HANDLE_TYPE_KMS
:
323 assert(rsc
->scanout
);
324 return renderonly_get_handle(rsc
->scanout
, whandle
);
326 whandle
->handle
= rsc
->bo
->handle
;
328 case WINSYS_HANDLE_TYPE_FD
:
329 /* FDs are cross-device, so we can export directly from vc4.
331 whandle
->handle
= vc4_bo_get_dmabuf(rsc
->bo
);
332 return whandle
->handle
!= -1;
339 vc4_setup_slices(struct vc4_resource
*rsc
, const char *caller
)
341 struct pipe_resource
*prsc
= &rsc
->base
;
342 uint32_t width
= prsc
->width0
;
343 uint32_t height
= prsc
->height0
;
344 if (prsc
->format
== PIPE_FORMAT_ETC1_RGB8
) {
345 width
= (width
+ 3) >> 2;
346 height
= (height
+ 3) >> 2;
349 uint32_t pot_width
= util_next_power_of_two(width
);
350 uint32_t pot_height
= util_next_power_of_two(height
);
352 uint32_t utile_w
= vc4_utile_width(rsc
->cpp
);
353 uint32_t utile_h
= vc4_utile_height(rsc
->cpp
);
355 for (int i
= prsc
->last_level
; i
>= 0; i
--) {
356 struct vc4_resource_slice
*slice
= &rsc
->slices
[i
];
358 uint32_t level_width
, level_height
;
361 level_height
= height
;
363 level_width
= u_minify(pot_width
, i
);
364 level_height
= u_minify(pot_height
, i
);
368 slice
->tiling
= VC4_TILING_FORMAT_LINEAR
;
369 if (prsc
->nr_samples
> 1) {
370 /* MSAA (4x) surfaces are stored as raw tile buffer contents. */
371 level_width
= align(level_width
, 32);
372 level_height
= align(level_height
, 32);
374 level_width
= align(level_width
, utile_w
);
377 if (vc4_size_is_lt(level_width
, level_height
,
379 slice
->tiling
= VC4_TILING_FORMAT_LT
;
380 level_width
= align(level_width
, utile_w
);
381 level_height
= align(level_height
, utile_h
);
383 slice
->tiling
= VC4_TILING_FORMAT_T
;
384 level_width
= align(level_width
,
386 level_height
= align(level_height
,
391 slice
->offset
= offset
;
392 slice
->stride
= (level_width
* rsc
->cpp
*
393 MAX2(prsc
->nr_samples
, 1));
394 slice
->size
= level_height
* slice
->stride
;
396 offset
+= slice
->size
;
398 if (vc4_debug
& VC4_DEBUG_SURFACE
) {
399 static const char tiling_chars
[] = {
400 [VC4_TILING_FORMAT_LINEAR
] = 'R',
401 [VC4_TILING_FORMAT_LT
] = 'L',
402 [VC4_TILING_FORMAT_T
] = 'T'
405 "rsc %s %p (format %s: vc4 %d), %dx%d: "
406 "level %d (%c) -> %dx%d, stride %d@0x%08x\n",
408 util_format_short_name(prsc
->format
),
410 prsc
->width0
, prsc
->height0
,
411 i
, tiling_chars
[slice
->tiling
],
412 level_width
, level_height
,
413 slice
->stride
, slice
->offset
);
417 /* The texture base pointer that has to point to level 0 doesn't have
418 * intra-page bits, so we have to align it, and thus shift up all the
421 uint32_t page_align_offset
= (align(rsc
->slices
[0].offset
, 4096) -
422 rsc
->slices
[0].offset
);
423 if (page_align_offset
) {
424 for (int i
= 0; i
<= prsc
->last_level
; i
++)
425 rsc
->slices
[i
].offset
+= page_align_offset
;
428 /* Cube map faces appear as whole miptrees at a page-aligned offset
429 * from the first face's miptree.
431 if (prsc
->target
== PIPE_TEXTURE_CUBE
) {
432 rsc
->cube_map_stride
= align(rsc
->slices
[0].offset
+
433 rsc
->slices
[0].size
, 4096);
437 static struct vc4_resource
*
438 vc4_resource_setup(struct pipe_screen
*pscreen
,
439 const struct pipe_resource
*tmpl
)
441 struct vc4_resource
*rsc
= CALLOC_STRUCT(vc4_resource
);
444 struct pipe_resource
*prsc
= &rsc
->base
;
448 pipe_reference_init(&prsc
->reference
, 1);
449 prsc
->screen
= pscreen
;
451 if (prsc
->nr_samples
<= 1)
452 rsc
->cpp
= util_format_get_blocksize(tmpl
->format
);
454 rsc
->cpp
= sizeof(uint32_t);
461 static enum vc4_texture_data_type
462 get_resource_texture_format(struct pipe_resource
*prsc
)
464 struct vc4_resource
*rsc
= vc4_resource(prsc
);
465 uint8_t format
= vc4_get_tex_format(prsc
->format
);
468 if (prsc
->nr_samples
> 1) {
471 if (format
== VC4_TEXTURE_TYPE_RGBA8888
)
472 return VC4_TEXTURE_TYPE_RGBA32R
;
482 find_modifier(uint64_t needle
, const uint64_t *haystack
, int count
)
486 for (i
= 0; i
< count
; i
++) {
487 if (haystack
[i
] == needle
)
494 static struct pipe_resource
*
495 vc4_resource_create_with_modifiers(struct pipe_screen
*pscreen
,
496 const struct pipe_resource
*tmpl
,
497 const uint64_t *modifiers
,
500 struct vc4_screen
*screen
= vc4_screen(pscreen
);
501 struct vc4_resource
*rsc
= vc4_resource_setup(pscreen
, tmpl
);
502 struct pipe_resource
*prsc
= &rsc
->base
;
503 bool linear_ok
= find_modifier(DRM_FORMAT_MOD_LINEAR
, modifiers
, count
);
504 /* Use a tiled layout if we can, for better 3D performance. */
505 bool should_tile
= true;
507 /* VBOs/PBOs are untiled (and 1 height). */
508 if (tmpl
->target
== PIPE_BUFFER
)
511 /* MSAA buffers are linear. */
512 if (tmpl
->nr_samples
> 1)
515 /* No tiling when we're sharing with another device (pl111). */
516 if (screen
->ro
&& (tmpl
->bind
& PIPE_BIND_SCANOUT
))
519 /* Cursors are always linear, and the user can request linear as well.
521 if (tmpl
->bind
& (PIPE_BIND_LINEAR
| PIPE_BIND_CURSOR
))
524 /* No shared objects with LT format -- the kernel only has T-format
525 * metadata. LT objects are small enough it's not worth the trouble to
526 * give them metadata to tile.
528 if ((tmpl
->bind
& (PIPE_BIND_SHARED
| PIPE_BIND_SCANOUT
)) &&
529 vc4_size_is_lt(prsc
->width0
, prsc
->height0
, rsc
->cpp
))
532 /* If we're sharing or scanning out, we need the ioctl present to
533 * inform the kernel or the other side.
535 if ((tmpl
->bind
& (PIPE_BIND_SHARED
|
536 PIPE_BIND_SCANOUT
)) && !screen
->has_tiling_ioctl
)
539 /* No user-specified modifier; determine our own. */
540 if (count
== 1 && modifiers
[0] == DRM_FORMAT_MOD_INVALID
) {
542 rsc
->tiled
= should_tile
;
543 } else if (should_tile
&&
544 find_modifier(DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED
,
547 } else if (linear_ok
) {
550 fprintf(stderr
, "Unsupported modifier requested\n");
554 if (tmpl
->target
!= PIPE_BUFFER
)
555 rsc
->vc4_format
= get_resource_texture_format(prsc
);
557 vc4_setup_slices(rsc
, "create");
558 if (!vc4_resource_bo_alloc(rsc
))
561 if (screen
->has_tiling_ioctl
) {
564 modifier
= DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED
;
566 modifier
= DRM_FORMAT_MOD_LINEAR
;
567 struct drm_vc4_set_tiling set_tiling
= {
568 .handle
= rsc
->bo
->handle
,
569 .modifier
= modifier
,
571 int ret
= vc4_ioctl(screen
->fd
, DRM_IOCTL_VC4_SET_TILING
,
577 /* Set up the "scanout resource" (the dmabuf export of our buffer to
578 * the KMS handle) if the buffer might ever have
579 * resource_get_handle(WINSYS_HANDLE_TYPE_KMS) called on it.
580 * create_with_modifiers() doesn't give us usage flags, so we have to
581 * assume that all calls with modifiers are scanout-possible.
584 ((tmpl
->bind
& PIPE_BIND_SCANOUT
) ||
585 !(count
== 1 && modifiers
[0] == DRM_FORMAT_MOD_INVALID
))) {
587 renderonly_scanout_for_resource(prsc
, screen
->ro
, NULL
);
592 vc4_bo_label(screen
, rsc
->bo
, "%sresource %dx%d@%d/%d",
593 (tmpl
->bind
& PIPE_BIND_SCANOUT
) ? "scanout " : "",
594 tmpl
->width0
, tmpl
->height0
,
595 rsc
->cpp
* 8, prsc
->last_level
);
599 vc4_resource_destroy(pscreen
, prsc
);
603 struct pipe_resource
*
604 vc4_resource_create(struct pipe_screen
*pscreen
,
605 const struct pipe_resource
*tmpl
)
607 const uint64_t mod
= DRM_FORMAT_MOD_INVALID
;
608 return vc4_resource_create_with_modifiers(pscreen
, tmpl
, &mod
, 1);
611 static struct pipe_resource
*
612 vc4_resource_from_handle(struct pipe_screen
*pscreen
,
613 const struct pipe_resource
*tmpl
,
614 struct winsys_handle
*whandle
,
617 struct vc4_screen
*screen
= vc4_screen(pscreen
);
618 struct vc4_resource
*rsc
= vc4_resource_setup(pscreen
, tmpl
);
619 struct pipe_resource
*prsc
= &rsc
->base
;
620 struct vc4_resource_slice
*slice
= &rsc
->slices
[0];
625 switch (whandle
->type
) {
626 case WINSYS_HANDLE_TYPE_SHARED
:
627 rsc
->bo
= vc4_bo_open_name(screen
, whandle
->handle
);
629 case WINSYS_HANDLE_TYPE_FD
:
630 rsc
->bo
= vc4_bo_open_dmabuf(screen
, whandle
->handle
);
634 "Attempt to import unsupported handle type %d\n",
641 struct drm_vc4_get_tiling get_tiling
= {
642 .handle
= rsc
->bo
->handle
,
644 int ret
= vc4_ioctl(screen
->fd
, DRM_IOCTL_VC4_GET_TILING
, &get_tiling
);
647 whandle
->modifier
= DRM_FORMAT_MOD_LINEAR
;
648 } else if (whandle
->modifier
== DRM_FORMAT_MOD_INVALID
) {
649 whandle
->modifier
= get_tiling
.modifier
;
650 } else if (whandle
->modifier
!= get_tiling
.modifier
) {
652 "Modifier 0x%llx vs. tiling (0x%llx) mismatch\n",
653 (long long)whandle
->modifier
, get_tiling
.modifier
);
657 switch (whandle
->modifier
) {
658 case DRM_FORMAT_MOD_LINEAR
:
661 case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED
:
666 "Attempt to import unsupported modifier 0x%llx\n",
667 (long long)whandle
->modifier
);
671 rsc
->vc4_format
= get_resource_texture_format(prsc
);
672 vc4_setup_slices(rsc
, "import");
674 if (whandle
->offset
!= 0) {
677 "Attempt to import unsupported "
678 "winsys offset %u\n",
683 rsc
->slices
[0].offset
+= whandle
->offset
;
685 if (rsc
->slices
[0].offset
+ rsc
->slices
[0].size
>
687 fprintf(stderr
, "Attempt to import "
688 "with overflowing offset (%d + %d > %d)\n",
697 /* Make sure that renderonly has a handle to our buffer in the
698 * display's fd, so that a later renderonly_get_handle()
699 * returns correct handles or GEM names.
702 renderonly_create_gpu_import_for_resource(prsc
,
709 if (rsc
->tiled
&& whandle
->stride
!= slice
->stride
) {
710 static bool warned
= false;
714 "Attempting to import %dx%d %s with "
715 "unsupported stride %d instead of %d\n",
716 prsc
->width0
, prsc
->height0
,
717 util_format_short_name(prsc
->format
),
722 } else if (!rsc
->tiled
) {
723 slice
->stride
= whandle
->stride
;
729 vc4_resource_destroy(pscreen
, prsc
);
733 static struct pipe_surface
*
734 vc4_create_surface(struct pipe_context
*pctx
,
735 struct pipe_resource
*ptex
,
736 const struct pipe_surface
*surf_tmpl
)
738 struct vc4_surface
*surface
= CALLOC_STRUCT(vc4_surface
);
739 struct vc4_resource
*rsc
= vc4_resource(ptex
);
744 assert(surf_tmpl
->u
.tex
.first_layer
== surf_tmpl
->u
.tex
.last_layer
);
746 struct pipe_surface
*psurf
= &surface
->base
;
747 unsigned level
= surf_tmpl
->u
.tex
.level
;
749 pipe_reference_init(&psurf
->reference
, 1);
750 pipe_resource_reference(&psurf
->texture
, ptex
);
752 psurf
->context
= pctx
;
753 psurf
->format
= surf_tmpl
->format
;
754 psurf
->width
= u_minify(ptex
->width0
, level
);
755 psurf
->height
= u_minify(ptex
->height0
, level
);
756 psurf
->u
.tex
.level
= level
;
757 psurf
->u
.tex
.first_layer
= surf_tmpl
->u
.tex
.first_layer
;
758 psurf
->u
.tex
.last_layer
= surf_tmpl
->u
.tex
.last_layer
;
759 surface
->offset
= (rsc
->slices
[level
].offset
+
760 psurf
->u
.tex
.first_layer
* rsc
->cube_map_stride
);
761 surface
->tiling
= rsc
->slices
[level
].tiling
;
763 return &surface
->base
;
767 vc4_surface_destroy(struct pipe_context
*pctx
, struct pipe_surface
*psurf
)
769 pipe_resource_reference(&psurf
->texture
, NULL
);
774 vc4_dump_surface_non_msaa(struct pipe_surface
*psurf
)
776 struct pipe_resource
*prsc
= psurf
->texture
;
777 struct vc4_resource
*rsc
= vc4_resource(prsc
);
778 uint32_t *map
= vc4_bo_map(rsc
->bo
);
779 uint32_t stride
= rsc
->slices
[0].stride
/ 4;
780 uint32_t width
= psurf
->width
;
781 uint32_t height
= psurf
->height
;
782 uint32_t chunk_w
= width
/ 79;
783 uint32_t chunk_h
= height
/ 40;
784 uint32_t found_colors
[10];
785 uint32_t num_found_colors
= 0;
787 if (rsc
->vc4_format
!= VC4_TEXTURE_TYPE_RGBA32R
) {
788 fprintf(stderr
, "%s: Unsupported format %s\n",
789 __func__
, util_format_short_name(psurf
->format
));
793 for (int by
= 0; by
< height
; by
+= chunk_h
) {
794 for (int bx
= 0; bx
< width
; bx
+= chunk_w
) {
795 int all_found_color
= -1; /* nothing found */
797 for (int y
= by
; y
< MIN2(height
, by
+ chunk_h
); y
++) {
798 for (int x
= bx
; x
< MIN2(width
, bx
+ chunk_w
); x
++) {
799 uint32_t pix
= map
[y
* stride
+ x
];
802 for (i
= 0; i
< num_found_colors
; i
++) {
803 if (pix
== found_colors
[i
])
806 if (i
== num_found_colors
&&
808 ARRAY_SIZE(found_colors
)) {
809 found_colors
[num_found_colors
++] = pix
;
812 if (i
< num_found_colors
) {
813 if (all_found_color
== -1)
815 else if (i
!= all_found_color
)
816 all_found_color
= ARRAY_SIZE(found_colors
);
820 /* If all pixels for this chunk have a consistent
821 * value, then print a character for it. Either a
822 * fixed name (particularly common for piglit tests),
823 * or a runtime-generated number.
825 if (all_found_color
>= 0 &&
826 all_found_color
< ARRAY_SIZE(found_colors
)) {
827 static const struct {
839 for (i
= 0; i
< ARRAY_SIZE(named_colors
); i
++) {
840 if (named_colors
[i
].val
==
841 found_colors
[all_found_color
]) {
842 fprintf(stderr
, "%s",
847 /* For unnamed colors, print a number and the
848 * numbers will have values printed at the
851 if (i
== ARRAY_SIZE(named_colors
)) {
852 fprintf(stderr
, "%c",
853 '0' + all_found_color
);
856 /* If there's no consistent color, print this.
858 fprintf(stderr
, ".");
861 fprintf(stderr
, "\n");
864 for (int i
= 0; i
< num_found_colors
; i
++) {
865 fprintf(stderr
, "color %d: 0x%08x\n", i
, found_colors
[i
]);
870 vc4_surface_msaa_get_sample(struct pipe_surface
*psurf
,
871 uint32_t x
, uint32_t y
, uint32_t sample
)
873 struct pipe_resource
*prsc
= psurf
->texture
;
874 struct vc4_resource
*rsc
= vc4_resource(prsc
);
875 uint32_t tile_w
= 32, tile_h
= 32;
876 uint32_t tiles_w
= DIV_ROUND_UP(psurf
->width
, 32);
878 uint32_t tile_x
= x
/ tile_w
;
879 uint32_t tile_y
= y
/ tile_h
;
880 uint32_t *tile
= (vc4_bo_map(rsc
->bo
) +
881 VC4_TILE_BUFFER_SIZE
* (tile_y
* tiles_w
+ tile_x
));
882 uint32_t subtile_x
= x
% tile_w
;
883 uint32_t subtile_y
= y
% tile_h
;
885 uint32_t quad_samples
= VC4_MAX_SAMPLES
* 4;
886 uint32_t tile_stride
= quad_samples
* tile_w
/ 2;
888 return *((uint32_t *)tile
+
889 (subtile_y
>> 1) * tile_stride
+
890 (subtile_x
>> 1) * quad_samples
+
891 ((subtile_y
& 1) << 1) +
897 vc4_dump_surface_msaa_char(struct pipe_surface
*psurf
,
898 uint32_t start_x
, uint32_t start_y
,
899 uint32_t w
, uint32_t h
)
901 bool all_same_color
= true;
902 uint32_t all_pix
= 0;
904 for (int y
= start_y
; y
< start_y
+ h
; y
++) {
905 for (int x
= start_x
; x
< start_x
+ w
; x
++) {
906 for (int s
= 0; s
< VC4_MAX_SAMPLES
; s
++) {
907 uint32_t pix
= vc4_surface_msaa_get_sample(psurf
,
910 if (x
== start_x
&& y
== start_y
)
912 else if (all_pix
!= pix
)
913 all_same_color
= false;
917 if (all_same_color
) {
918 static const struct {
930 for (i
= 0; i
< ARRAY_SIZE(named_colors
); i
++) {
931 if (named_colors
[i
].val
== all_pix
) {
932 fprintf(stderr
, "%s",
937 fprintf(stderr
, "x");
939 fprintf(stderr
, ".");
944 vc4_dump_surface_msaa(struct pipe_surface
*psurf
)
946 uint32_t tile_w
= 32, tile_h
= 32;
947 uint32_t tiles_w
= DIV_ROUND_UP(psurf
->width
, tile_w
);
948 uint32_t tiles_h
= DIV_ROUND_UP(psurf
->height
, tile_h
);
949 uint32_t char_w
= 140, char_h
= 60;
950 uint32_t char_w_per_tile
= char_w
/ tiles_w
- 1;
951 uint32_t char_h_per_tile
= char_h
/ tiles_h
- 1;
953 fprintf(stderr
, "Surface: %dx%d (%dx MSAA)\n",
954 psurf
->width
, psurf
->height
, psurf
->texture
->nr_samples
);
956 for (int x
= 0; x
< (char_w_per_tile
+ 1) * tiles_w
; x
++)
957 fprintf(stderr
, "-");
958 fprintf(stderr
, "\n");
960 for (int ty
= 0; ty
< psurf
->height
; ty
+= tile_h
) {
961 for (int y
= 0; y
< char_h_per_tile
; y
++) {
963 for (int tx
= 0; tx
< psurf
->width
; tx
+= tile_w
) {
964 for (int x
= 0; x
< char_w_per_tile
; x
++) {
965 uint32_t bx1
= (x
* tile_w
/
967 uint32_t bx2
= ((x
+ 1) * tile_w
/
969 uint32_t by1
= (y
* tile_h
/
971 uint32_t by2
= ((y
+ 1) * tile_h
/
974 vc4_dump_surface_msaa_char(psurf
,
980 fprintf(stderr
, "|");
982 fprintf(stderr
, "\n");
985 for (int x
= 0; x
< (char_w_per_tile
+ 1) * tiles_w
; x
++)
986 fprintf(stderr
, "-");
987 fprintf(stderr
, "\n");
991 /** Debug routine to dump the contents of an 8888 surface to the console */
993 vc4_dump_surface(struct pipe_surface
*psurf
)
998 if (psurf
->texture
->nr_samples
> 1)
999 vc4_dump_surface_msaa(psurf
);
1001 vc4_dump_surface_non_msaa(psurf
);
1005 vc4_flush_resource(struct pipe_context
*pctx
, struct pipe_resource
*resource
)
1007 /* All calls to flush_resource are followed by a flush of the context,
1008 * so there's nothing to do.
1013 vc4_update_shadow_baselevel_texture(struct pipe_context
*pctx
,
1014 struct pipe_sampler_view
*pview
)
1016 struct vc4_context
*vc4
= vc4_context(pctx
);
1017 struct vc4_sampler_view
*view
= vc4_sampler_view(pview
);
1018 struct vc4_resource
*shadow
= vc4_resource(view
->texture
);
1019 struct vc4_resource
*orig
= vc4_resource(pview
->texture
);
1021 assert(view
->texture
!= pview
->texture
);
1023 if (shadow
->writes
== orig
->writes
&& orig
->bo
->private)
1026 perf_debug("Updating %dx%d@%d shadow texture due to %s\n",
1027 orig
->base
.width0
, orig
->base
.height0
,
1028 pview
->u
.tex
.first_level
,
1029 pview
->u
.tex
.first_level
? "base level" : "raster layout");
1031 for (int i
= 0; i
<= shadow
->base
.last_level
; i
++) {
1032 unsigned width
= u_minify(shadow
->base
.width0
, i
);
1033 unsigned height
= u_minify(shadow
->base
.height0
, i
);
1034 struct pipe_blit_info info
= {
1036 .resource
= &shadow
->base
,
1046 .format
= shadow
->base
.format
,
1049 .resource
= &orig
->base
,
1050 .level
= pview
->u
.tex
.first_level
+ i
,
1059 .format
= orig
->base
.format
,
1063 pctx
->blit(pctx
, &info
);
1066 shadow
->writes
= orig
->writes
;
1070 * Converts a 4-byte index buffer to 2 bytes.
1072 * Since GLES2 only has support for 1 and 2-byte indices, the hardware doesn't
1073 * include 4-byte index support, and we have to shrink it down.
1075 * There's no fallback support for when indices end up being larger than 2^16,
1076 * though it will at least assertion fail. Also, if the original index data
1077 * was in user memory, it would be nice to not have uploaded it to a VBO
1078 * before translating.
1080 struct pipe_resource
*
1081 vc4_get_shadow_index_buffer(struct pipe_context
*pctx
,
1082 const struct pipe_draw_info
*info
,
1085 uint32_t *shadow_offset
)
1087 struct vc4_context
*vc4
= vc4_context(pctx
);
1088 struct vc4_resource
*orig
= vc4_resource(info
->index
.resource
);
1089 perf_debug("Fallback conversion for %d uint indices\n", count
);
1092 struct pipe_resource
*shadow_rsc
= NULL
;
1093 u_upload_alloc(vc4
->uploader
, 0, count
* 2, 4,
1094 shadow_offset
, &shadow_rsc
, &data
);
1095 uint16_t *dst
= data
;
1097 struct pipe_transfer
*src_transfer
= NULL
;
1098 const uint32_t *src
;
1099 if (info
->has_user_indices
) {
1100 src
= info
->index
.user
;
1102 src
= pipe_buffer_map_range(pctx
, &orig
->base
,
1105 PIPE_TRANSFER_READ
, &src_transfer
);
1108 for (int i
= 0; i
< count
; i
++) {
1109 uint32_t src_index
= src
[i
];
1110 assert(src_index
<= 0xffff);
1115 pctx
->transfer_unmap(pctx
, src_transfer
);
1120 static const struct u_transfer_vtbl transfer_vtbl
= {
1121 .resource_create
= vc4_resource_create
,
1122 .resource_destroy
= vc4_resource_destroy
,
1123 .transfer_map
= vc4_resource_transfer_map
,
1124 .transfer_unmap
= vc4_resource_transfer_unmap
,
1125 .transfer_flush_region
= u_default_transfer_flush_region
,
1129 vc4_resource_screen_init(struct pipe_screen
*pscreen
)
1131 struct vc4_screen
*screen
= vc4_screen(pscreen
);
1133 pscreen
->resource_create
= vc4_resource_create
;
1134 pscreen
->resource_create_with_modifiers
=
1135 vc4_resource_create_with_modifiers
;
1136 pscreen
->resource_from_handle
= vc4_resource_from_handle
;
1137 pscreen
->resource_destroy
= u_resource_destroy_vtbl
;
1138 pscreen
->resource_get_handle
= vc4_resource_get_handle
;
1139 pscreen
->resource_destroy
= vc4_resource_destroy
;
1140 pscreen
->transfer_helper
= u_transfer_helper_create(&transfer_vtbl
,
1144 /* Test if the kernel has GET_TILING; it will return -EINVAL if the
1145 * ioctl does not exist, but -ENOENT if we pass an impossible handle.
1146 * 0 cannot be a valid GEM object, so use that.
1148 struct drm_vc4_get_tiling get_tiling
= {
1151 int ret
= vc4_ioctl(screen
->fd
, DRM_IOCTL_VC4_GET_TILING
, &get_tiling
);
1152 if (ret
== -1 && errno
== ENOENT
)
1153 screen
->has_tiling_ioctl
= true;
1157 vc4_resource_context_init(struct pipe_context
*pctx
)
1159 pctx
->transfer_map
= u_transfer_helper_transfer_map
;
1160 pctx
->transfer_flush_region
= u_transfer_helper_transfer_flush_region
;
1161 pctx
->transfer_unmap
= u_transfer_helper_transfer_unmap
;
1162 pctx
->buffer_subdata
= u_default_buffer_subdata
;
1163 pctx
->texture_subdata
= vc4_texture_subdata
;
1164 pctx
->create_surface
= vc4_create_surface
;
1165 pctx
->surface_destroy
= vc4_surface_destroy
;
1166 pctx
->resource_copy_region
= util_resource_copy_region
;
1167 pctx
->blit
= vc4_blit
;
1168 pctx
->flush_resource
= vc4_flush_resource
;