2 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
27 #include "r600_formats.h"
31 #include "util/u_format_s3tc.h"
32 #include "util/u_memory.h"
34 /* Copy from a full GPU texture to a transfer's staging one. */
35 static void r600_copy_to_staging_texture(struct pipe_context
*ctx
, struct r600_transfer
*rtransfer
)
37 struct pipe_transfer
*transfer
= (struct pipe_transfer
*)rtransfer
;
38 struct pipe_resource
*texture
= transfer
->resource
;
40 ctx
->resource_copy_region(ctx
, &rtransfer
->staging
->b
.b
,
41 0, 0, 0, 0, texture
, transfer
->level
,
46 /* Copy from a transfer's staging texture to a full GPU one. */
47 static void r600_copy_from_staging_texture(struct pipe_context
*ctx
, struct r600_transfer
*rtransfer
)
49 struct pipe_transfer
*transfer
= (struct pipe_transfer
*)rtransfer
;
50 struct pipe_resource
*texture
= transfer
->resource
;
53 u_box_origin_2d(transfer
->box
.width
, transfer
->box
.height
, &sbox
);
55 ctx
->resource_copy_region(ctx
, texture
, transfer
->level
,
56 transfer
->box
.x
, transfer
->box
.y
, transfer
->box
.z
,
57 &rtransfer
->staging
->b
.b
,
61 unsigned r600_texture_get_offset(struct r600_texture
*rtex
,
62 unsigned level
, unsigned layer
)
64 return rtex
->surface
.level
[level
].offset
+
65 layer
* rtex
->surface
.level
[level
].slice_size
;
68 static int r600_init_surface(struct r600_screen
*rscreen
,
69 struct radeon_surface
*surface
,
70 const struct pipe_resource
*ptex
,
72 bool is_transfer
, bool is_flushed_depth
)
74 const struct util_format_description
*desc
=
75 util_format_description(ptex
->format
);
76 bool is_depth
, is_stencil
;
78 is_depth
= util_format_has_depth(desc
);
79 is_stencil
= util_format_has_stencil(desc
);
81 surface
->npix_x
= ptex
->width0
;
82 surface
->npix_y
= ptex
->height0
;
83 surface
->npix_z
= ptex
->depth0
;
84 surface
->blk_w
= util_format_get_blockwidth(ptex
->format
);
85 surface
->blk_h
= util_format_get_blockheight(ptex
->format
);
87 surface
->array_size
= 1;
88 surface
->last_level
= ptex
->last_level
;
90 if (rscreen
->chip_class
>= EVERGREEN
&&
91 !is_transfer
&& !is_flushed_depth
&&
92 ptex
->format
== PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
) {
93 surface
->bpe
= 4; /* stencil is allocated separately on evergreen */
95 surface
->bpe
= util_format_get_blocksize(ptex
->format
);
96 /* align byte per element on dword */
97 if (surface
->bpe
== 3) {
102 surface
->nsamples
= ptex
->nr_samples
? ptex
->nr_samples
: 1;
105 switch (array_mode
) {
106 case V_038000_ARRAY_1D_TILED_THIN1
:
107 surface
->flags
|= RADEON_SURF_SET(RADEON_SURF_MODE_1D
, MODE
);
109 case V_038000_ARRAY_2D_TILED_THIN1
:
110 surface
->flags
|= RADEON_SURF_SET(RADEON_SURF_MODE_2D
, MODE
);
112 case V_038000_ARRAY_LINEAR_ALIGNED
:
113 surface
->flags
|= RADEON_SURF_SET(RADEON_SURF_MODE_LINEAR_ALIGNED
, MODE
);
115 case V_038000_ARRAY_LINEAR_GENERAL
:
117 surface
->flags
|= RADEON_SURF_SET(RADEON_SURF_MODE_LINEAR
, MODE
);
120 switch (ptex
->target
) {
121 case PIPE_TEXTURE_1D
:
122 surface
->flags
|= RADEON_SURF_SET(RADEON_SURF_TYPE_1D
, TYPE
);
124 case PIPE_TEXTURE_RECT
:
125 case PIPE_TEXTURE_2D
:
126 surface
->flags
|= RADEON_SURF_SET(RADEON_SURF_TYPE_2D
, TYPE
);
128 case PIPE_TEXTURE_3D
:
129 surface
->flags
|= RADEON_SURF_SET(RADEON_SURF_TYPE_3D
, TYPE
);
131 case PIPE_TEXTURE_1D_ARRAY
:
132 surface
->flags
|= RADEON_SURF_SET(RADEON_SURF_TYPE_1D_ARRAY
, TYPE
);
133 surface
->array_size
= ptex
->array_size
;
135 case PIPE_TEXTURE_2D_ARRAY
:
136 surface
->flags
|= RADEON_SURF_SET(RADEON_SURF_TYPE_2D_ARRAY
, TYPE
);
137 surface
->array_size
= ptex
->array_size
;
139 case PIPE_TEXTURE_CUBE
:
140 surface
->flags
|= RADEON_SURF_SET(RADEON_SURF_TYPE_CUBEMAP
, TYPE
);
146 if (ptex
->bind
& PIPE_BIND_SCANOUT
) {
147 surface
->flags
|= RADEON_SURF_SCANOUT
;
150 if (!is_transfer
&& !is_flushed_depth
&& is_depth
) {
151 surface
->flags
|= RADEON_SURF_ZBUFFER
;
154 surface
->flags
|= RADEON_SURF_SBUFFER
;
160 static int r600_setup_surface(struct pipe_screen
*screen
,
161 struct r600_texture
*rtex
,
162 unsigned pitch_in_bytes_override
)
164 struct pipe_resource
*ptex
= &rtex
->resource
.b
.b
;
165 struct r600_screen
*rscreen
= (struct r600_screen
*)screen
;
169 r
= rscreen
->ws
->surface_init(rscreen
->ws
, &rtex
->surface
);
173 rtex
->size
= rtex
->surface
.bo_size
;
174 if (pitch_in_bytes_override
&& pitch_in_bytes_override
!= rtex
->surface
.level
[0].pitch_bytes
) {
175 /* old ddx on evergreen over estimate alignment for 1d, only 1 level
178 rtex
->surface
.level
[0].nblk_x
= pitch_in_bytes_override
/ rtex
->surface
.bpe
;
179 rtex
->surface
.level
[0].pitch_bytes
= pitch_in_bytes_override
;
180 rtex
->surface
.level
[0].slice_size
= pitch_in_bytes_override
* rtex
->surface
.level
[0].nblk_y
;
181 if (rtex
->surface
.flags
& RADEON_SURF_SBUFFER
) {
182 rtex
->surface
.stencil_offset
= rtex
->surface
.level
[0].slice_size
;
185 for (i
= 0; i
<= ptex
->last_level
; i
++) {
186 switch (rtex
->surface
.level
[i
].mode
) {
187 case RADEON_SURF_MODE_LINEAR_ALIGNED
:
188 rtex
->array_mode
[i
] = V_038000_ARRAY_LINEAR_ALIGNED
;
190 case RADEON_SURF_MODE_1D
:
191 rtex
->array_mode
[i
] = V_038000_ARRAY_1D_TILED_THIN1
;
193 case RADEON_SURF_MODE_2D
:
194 rtex
->array_mode
[i
] = V_038000_ARRAY_2D_TILED_THIN1
;
197 case RADEON_SURF_MODE_LINEAR
:
198 rtex
->array_mode
[i
] = 0;
205 static boolean
r600_texture_get_handle(struct pipe_screen
* screen
,
206 struct pipe_resource
*ptex
,
207 struct winsys_handle
*whandle
)
209 struct r600_texture
*rtex
= (struct r600_texture
*)ptex
;
210 struct r600_resource
*resource
= &rtex
->resource
;
211 struct radeon_surface
*surface
= &rtex
->surface
;
212 struct r600_screen
*rscreen
= (struct r600_screen
*)screen
;
214 rscreen
->ws
->buffer_set_tiling(resource
->buf
,
216 surface
->level
[0].mode
>= RADEON_SURF_MODE_1D
?
217 RADEON_LAYOUT_TILED
: RADEON_LAYOUT_LINEAR
,
218 surface
->level
[0].mode
>= RADEON_SURF_MODE_2D
?
219 RADEON_LAYOUT_TILED
: RADEON_LAYOUT_LINEAR
,
220 surface
->bankw
, surface
->bankh
,
222 surface
->stencil_tile_split
,
224 rtex
->surface
.level
[0].pitch_bytes
);
226 return rscreen
->ws
->buffer_get_handle(resource
->buf
,
227 rtex
->surface
.level
[0].pitch_bytes
, whandle
);
230 static void r600_texture_destroy(struct pipe_screen
*screen
,
231 struct pipe_resource
*ptex
)
233 struct r600_texture
*rtex
= (struct r600_texture
*)ptex
;
234 struct r600_resource
*resource
= &rtex
->resource
;
236 if (rtex
->flushed_depth_texture
)
237 pipe_resource_reference((struct pipe_resource
**)&rtex
->flushed_depth_texture
, NULL
);
239 pb_reference(&resource
->buf
, NULL
);
243 static const struct u_resource_vtbl r600_texture_vtbl
=
245 r600_texture_get_handle
, /* get_handle */
246 r600_texture_destroy
, /* resource_destroy */
247 r600_texture_get_transfer
, /* get_transfer */
248 r600_texture_transfer_destroy
, /* transfer_destroy */
249 r600_texture_transfer_map
, /* transfer_map */
250 NULL
, /* transfer_flush_region */
251 r600_texture_transfer_unmap
, /* transfer_unmap */
252 NULL
/* transfer_inline_write */
255 static void r600_texture_allocate_fmask(struct r600_screen
*rscreen
,
256 struct r600_texture
*rtex
)
258 /* FMASK is allocated pretty much like an ordinary texture.
259 * Here we use bpe in the units of bits, not bytes. */
260 struct radeon_surface fmask
= rtex
->surface
;
261 unsigned nr_samples
= rtex
->resource
.b
.b
.nr_samples
;
263 switch (nr_samples
) {
265 /* This should be 8,1, but we should set nsamples > 1
266 * for the allocator to treat it as a multisample surface.
267 * Let's set 4,2 then. */
281 R600_ERR("Invalid sample count for FMASK allocation.\n");
285 /* R600-R700 errata? Anyway, this fixes colorbuffer corruption. */
286 if (rscreen
->chip_class
<= R700
) {
290 if (rscreen
->chip_class
>= EVERGREEN
) {
291 fmask
.bankh
= nr_samples
<= 4 ? 4 : 1;
294 if (rscreen
->ws
->surface_init(rscreen
->ws
, &fmask
)) {
295 R600_ERR("Got error in surface_init while allocating FMASK.\n");
298 assert(fmask
.level
[0].mode
== RADEON_SURF_MODE_2D
);
300 /* Reserve space for FMASK while converting bits back to bytes. */
301 rtex
->fmask_bank_height
= fmask
.bankh
;
302 rtex
->fmask_offset
= align(rtex
->size
, MAX2(256, fmask
.bo_alignment
));
303 rtex
->fmask_size
= (fmask
.bo_size
+ 7) / 8;
304 rtex
->size
= rtex
->fmask_offset
+ rtex
->fmask_size
;
306 printf("FMASK width=%u, height=%i, bits=%u, size=%u\n",
307 fmask
.npix_x
, fmask
.npix_y
, fmask
.bpe
* fmask
.nsamples
, rtex
->fmask_size
);
311 static void r600_texture_allocate_cmask(struct r600_screen
*rscreen
,
312 struct r600_texture
*rtex
)
314 unsigned cmask_tile_width
= 8;
315 unsigned cmask_tile_height
= 8;
316 unsigned cmask_tile_elements
= cmask_tile_width
* cmask_tile_height
;
317 unsigned element_bits
= 4;
318 unsigned cmask_cache_bits
= 1024;
319 unsigned num_pipes
= rscreen
->tiling_info
.num_channels
;
320 unsigned pipe_interleave_bytes
= rscreen
->tiling_info
.group_bytes
;
322 unsigned elements_per_macro_tile
= (cmask_cache_bits
/ element_bits
) * num_pipes
;
323 unsigned pixels_per_macro_tile
= elements_per_macro_tile
* cmask_tile_elements
;
324 unsigned sqrt_pixels_per_macro_tile
= sqrt(pixels_per_macro_tile
);
325 unsigned macro_tile_width
= util_next_power_of_two(sqrt_pixels_per_macro_tile
);
326 unsigned macro_tile_height
= pixels_per_macro_tile
/ macro_tile_width
;
328 unsigned pitch_elements
= align(rtex
->surface
.npix_x
, macro_tile_width
);
329 unsigned height
= align(rtex
->surface
.npix_y
, macro_tile_height
);
331 unsigned base_align
= num_pipes
* pipe_interleave_bytes
;
332 unsigned slice_bytes
=
333 ((pitch_elements
* height
* element_bits
+ 7) / 8) / cmask_tile_elements
;
334 unsigned size
= rtex
->surface
.array_size
* align(slice_bytes
, base_align
);
336 assert(macro_tile_width
% 128 == 0);
337 assert(macro_tile_height
% 128 == 0);
339 rtex
->cmask_slice_tile_max
= ((pitch_elements
* height
) / (128*128)) - 1;
340 rtex
->cmask_offset
= align(rtex
->size
, MAX2(256, base_align
));
341 rtex
->cmask_size
= size
;
342 rtex
->size
= rtex
->cmask_offset
+ rtex
->cmask_size
;
344 printf("CMASK: macro tile width = %u, macro tile height = %u, "
345 "pitch elements = %u, height = %u, slice tile max = %u\n",
346 macro_tile_width
, macro_tile_height
, pitch_elements
, height
,
347 rtex
->cmask_slice_tile_max
);
351 static struct r600_texture
*
352 r600_texture_create_object(struct pipe_screen
*screen
,
353 const struct pipe_resource
*base
,
354 unsigned pitch_in_bytes_override
,
355 struct pb_buffer
*buf
,
357 struct radeon_surface
*surface
)
359 struct r600_texture
*rtex
;
360 struct r600_resource
*resource
;
361 struct r600_screen
*rscreen
= (struct r600_screen
*)screen
;
364 rtex
= CALLOC_STRUCT(r600_texture
);
368 resource
= &rtex
->resource
;
369 resource
->b
.b
= *base
;
370 resource
->b
.vtbl
= &r600_texture_vtbl
;
371 pipe_reference_init(&resource
->b
.b
.reference
, 1);
372 resource
->b
.b
.screen
= screen
;
373 rtex
->pitch_override
= pitch_in_bytes_override
;
375 /* don't include stencil-only formats which we don't support for rendering */
376 rtex
->is_depth
= util_format_has_depth(util_format_description(rtex
->resource
.b
.b
.format
));
378 rtex
->surface
= *surface
;
379 r
= r600_setup_surface(screen
, rtex
,
380 pitch_in_bytes_override
);
386 if (base
->nr_samples
> 1 && !rtex
->is_depth
&& alloc_bo
) {
387 r600_texture_allocate_cmask(rscreen
, rtex
);
388 r600_texture_allocate_fmask(rscreen
, rtex
);
391 if (!rtex
->is_depth
&& base
->nr_samples
> 1 &&
392 (!rtex
->fmask_size
|| !rtex
->cmask_size
)) {
397 /* Now create the backing buffer. */
398 if (!buf
&& alloc_bo
) {
399 unsigned base_align
= rtex
->surface
.bo_alignment
;
400 unsigned usage
= R600_TEX_IS_TILED(rtex
, 0) ? PIPE_USAGE_STATIC
: base
->usage
;
402 if (!r600_init_resource(rscreen
, resource
, rtex
->size
, base_align
, base
->bind
, usage
)) {
408 resource
->cs_buf
= rscreen
->ws
->buffer_get_cs_handle(buf
);
409 resource
->domains
= RADEON_DOMAIN_GTT
| RADEON_DOMAIN_VRAM
;
412 if (rtex
->cmask_size
) {
413 /* Initialize the cmask to 0xCC (= compressed state). */
414 char *ptr
= rscreen
->ws
->buffer_map(resource
->cs_buf
, NULL
, PIPE_TRANSFER_WRITE
);
415 memset(ptr
+ rtex
->cmask_offset
, 0xCC, rtex
->cmask_size
);
416 rscreen
->ws
->buffer_unmap(resource
->cs_buf
);
421 struct pipe_resource
*r600_texture_create(struct pipe_screen
*screen
,
422 const struct pipe_resource
*templ
)
424 struct r600_screen
*rscreen
= (struct r600_screen
*)screen
;
425 struct radeon_surface surface
;
426 unsigned array_mode
= 0;
429 if (!(templ
->flags
& R600_RESOURCE_FLAG_TRANSFER
)) {
430 if (!(templ
->bind
& PIPE_BIND_SCANOUT
) &&
431 templ
->usage
!= PIPE_USAGE_STAGING
&&
432 templ
->usage
!= PIPE_USAGE_STREAM
) {
433 array_mode
= V_038000_ARRAY_2D_TILED_THIN1
;
434 } else if (util_format_is_compressed(templ
->format
)) {
435 array_mode
= V_038000_ARRAY_1D_TILED_THIN1
;
439 /* XXX tiling is broken for the 422 formats */
440 if (util_format_description(templ
->format
)->layout
== UTIL_FORMAT_LAYOUT_SUBSAMPLED
)
441 array_mode
= V_038000_ARRAY_LINEAR_ALIGNED
;
443 r
= r600_init_surface(rscreen
, &surface
, templ
, array_mode
,
444 templ
->flags
& R600_RESOURCE_FLAG_TRANSFER
,
445 templ
->flags
& R600_RESOURCE_FLAG_FLUSHED_DEPTH
);
449 r
= rscreen
->ws
->surface_best(rscreen
->ws
, &surface
);
453 return (struct pipe_resource
*)r600_texture_create_object(screen
, templ
,
454 0, NULL
, TRUE
, &surface
);
457 static struct pipe_surface
*r600_create_surface(struct pipe_context
*pipe
,
458 struct pipe_resource
*texture
,
459 const struct pipe_surface
*templ
)
461 struct r600_texture
*rtex
= (struct r600_texture
*)texture
;
462 struct r600_surface
*surface
= CALLOC_STRUCT(r600_surface
);
463 unsigned level
= templ
->u
.tex
.level
;
465 assert(templ
->u
.tex
.first_layer
== templ
->u
.tex
.last_layer
);
468 pipe_reference_init(&surface
->base
.reference
, 1);
469 pipe_resource_reference(&surface
->base
.texture
, texture
);
470 surface
->base
.context
= pipe
;
471 surface
->base
.format
= templ
->format
;
472 surface
->base
.width
= rtex
->surface
.level
[level
].npix_x
;
473 surface
->base
.height
= rtex
->surface
.level
[level
].npix_y
;
474 surface
->base
.usage
= templ
->usage
;
475 surface
->base
.u
= templ
->u
;
476 return &surface
->base
;
479 static void r600_surface_destroy(struct pipe_context
*pipe
,
480 struct pipe_surface
*surface
)
482 pipe_resource_reference(&surface
->texture
, NULL
);
486 struct pipe_resource
*r600_texture_from_handle(struct pipe_screen
*screen
,
487 const struct pipe_resource
*templ
,
488 struct winsys_handle
*whandle
)
490 struct r600_screen
*rscreen
= (struct r600_screen
*)screen
;
491 struct pb_buffer
*buf
= NULL
;
493 unsigned array_mode
= 0;
494 enum radeon_bo_layout micro
, macro
;
495 struct radeon_surface surface
;
498 /* Support only 2D textures without mipmaps */
499 if ((templ
->target
!= PIPE_TEXTURE_2D
&& templ
->target
!= PIPE_TEXTURE_RECT
) ||
500 templ
->depth0
!= 1 || templ
->last_level
!= 0)
503 buf
= rscreen
->ws
->buffer_from_handle(rscreen
->ws
, whandle
, &stride
);
507 rscreen
->ws
->buffer_get_tiling(buf
, µ
, ¯o
,
508 &surface
.bankw
, &surface
.bankh
,
510 &surface
.stencil_tile_split
,
513 if (macro
== RADEON_LAYOUT_TILED
)
514 array_mode
= V_0280A0_ARRAY_2D_TILED_THIN1
;
515 else if (micro
== RADEON_LAYOUT_TILED
)
516 array_mode
= V_0280A0_ARRAY_1D_TILED_THIN1
;
520 r
= r600_init_surface(rscreen
, &surface
, templ
, array_mode
, false, false);
524 return (struct pipe_resource
*)r600_texture_create_object(screen
, templ
,
525 stride
, buf
, FALSE
, &surface
);
528 bool r600_init_flushed_depth_texture(struct pipe_context
*ctx
,
529 struct pipe_resource
*texture
,
530 struct r600_texture
**staging
)
532 struct r600_texture
*rtex
= (struct r600_texture
*)texture
;
533 struct pipe_resource resource
;
534 struct r600_texture
**flushed_depth_texture
= staging
?
535 staging
: &rtex
->flushed_depth_texture
;
537 if (!staging
&& rtex
->flushed_depth_texture
)
538 return true; /* it's ready */
540 resource
.target
= texture
->target
;
541 resource
.format
= texture
->format
;
542 resource
.width0
= texture
->width0
;
543 resource
.height0
= texture
->height0
;
544 resource
.depth0
= texture
->depth0
;
545 resource
.array_size
= texture
->array_size
;
546 resource
.last_level
= texture
->last_level
;
547 resource
.nr_samples
= texture
->nr_samples
;
548 resource
.usage
= staging
? PIPE_USAGE_STAGING
: PIPE_USAGE_STATIC
;
549 resource
.bind
= texture
->bind
& ~PIPE_BIND_DEPTH_STENCIL
;
550 resource
.flags
= texture
->flags
| R600_RESOURCE_FLAG_FLUSHED_DEPTH
;
553 resource
.flags
|= R600_RESOURCE_FLAG_TRANSFER
;
555 *flushed_depth_texture
= (struct r600_texture
*)ctx
->screen
->resource_create(ctx
->screen
, &resource
);
556 if (*flushed_depth_texture
== NULL
) {
557 R600_ERR("failed to create temporary texture to hold flushed depth\n");
561 (*flushed_depth_texture
)->is_flushing_texture
= TRUE
;
565 /* Needs adjustment for pixelformat:
567 static INLINE
unsigned u_box_volume( const struct pipe_box
*box
)
569 return box
->width
* box
->depth
* box
->height
;
572 struct pipe_transfer
* r600_texture_get_transfer(struct pipe_context
*ctx
,
573 struct pipe_resource
*texture
,
576 const struct pipe_box
*box
)
578 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
579 struct r600_texture
*rtex
= (struct r600_texture
*)texture
;
580 struct pipe_resource resource
;
581 struct r600_transfer
*trans
;
582 boolean use_staging_texture
= FALSE
;
584 /* We cannot map a tiled texture directly because the data is
585 * in a different order, therefore we do detiling using a blit.
587 * Also, use a temporary in GTT memory for read transfers, as
588 * the CPU is much happier reading out of cached system memory
589 * than uncached VRAM.
591 if (R600_TEX_IS_TILED(rtex
, level
)) {
592 use_staging_texture
= TRUE
;
595 if ((usage
& PIPE_TRANSFER_READ
) && u_box_volume(box
) > 1024)
596 use_staging_texture
= TRUE
;
598 /* Use a staging texture for uploads if the underlying BO is busy. */
599 if (!(usage
& PIPE_TRANSFER_READ
) &&
600 (rctx
->ws
->cs_is_buffer_referenced(rctx
->cs
, rtex
->resource
.cs_buf
, RADEON_USAGE_READWRITE
) ||
601 rctx
->ws
->buffer_is_busy(rtex
->resource
.buf
, RADEON_USAGE_READWRITE
))) {
602 use_staging_texture
= TRUE
;
605 if (texture
->flags
& R600_RESOURCE_FLAG_TRANSFER
) {
606 use_staging_texture
= FALSE
;
609 if (use_staging_texture
&& (usage
& PIPE_TRANSFER_MAP_DIRECTLY
)) {
613 trans
= CALLOC_STRUCT(r600_transfer
);
616 pipe_resource_reference(&trans
->transfer
.resource
, texture
);
617 trans
->transfer
.level
= level
;
618 trans
->transfer
.usage
= usage
;
619 trans
->transfer
.box
= *box
;
620 if (rtex
->is_depth
) {
621 /* XXX: only readback the rectangle which is being mapped?
623 /* XXX: when discard is true, no need to read back from depth texture
625 struct r600_texture
*staging_depth
;
627 if (!r600_init_flushed_depth_texture(ctx
, texture
, &staging_depth
)) {
628 R600_ERR("failed to create temporary texture to hold untiled copy\n");
629 pipe_resource_reference(&trans
->transfer
.resource
, NULL
);
634 r600_blit_decompress_depth(ctx
, rtex
, staging_depth
,
636 box
->z
, box
->z
+ box
->depth
- 1,
639 trans
->transfer
.stride
= staging_depth
->surface
.level
[level
].pitch_bytes
;
640 trans
->offset
= r600_texture_get_offset(staging_depth
, level
, box
->z
);
641 trans
->staging
= (struct r600_resource
*)staging_depth
;
642 return &trans
->transfer
;
643 } else if (use_staging_texture
) {
644 resource
.target
= PIPE_TEXTURE_2D
;
645 resource
.format
= texture
->format
;
646 resource
.width0
= box
->width
;
647 resource
.height0
= box
->height
;
649 resource
.array_size
= 1;
650 resource
.last_level
= 0;
651 resource
.nr_samples
= 0;
652 resource
.usage
= PIPE_USAGE_STAGING
;
654 resource
.flags
= R600_RESOURCE_FLAG_TRANSFER
;
655 /* For texture reading, the temporary (detiled) texture is used as
656 * a render target when blitting from a tiled texture. */
657 if (usage
& PIPE_TRANSFER_READ
) {
658 resource
.bind
|= PIPE_BIND_RENDER_TARGET
;
660 /* For texture writing, the temporary texture is used as a sampler
661 * when blitting into a tiled texture. */
662 if (usage
& PIPE_TRANSFER_WRITE
) {
663 resource
.bind
|= PIPE_BIND_SAMPLER_VIEW
;
665 /* Create the temporary texture. */
666 trans
->staging
= (struct r600_resource
*)ctx
->screen
->resource_create(ctx
->screen
, &resource
);
667 if (trans
->staging
== NULL
) {
668 R600_ERR("failed to create temporary texture to hold untiled copy\n");
669 pipe_resource_reference(&trans
->transfer
.resource
, NULL
);
674 trans
->transfer
.stride
=
675 ((struct r600_texture
*)trans
->staging
)->surface
.level
[0].pitch_bytes
;
676 if (usage
& PIPE_TRANSFER_READ
) {
677 r600_copy_to_staging_texture(ctx
, trans
);
678 /* Always referenced in the blit. */
679 r600_flush(ctx
, NULL
, 0);
681 return &trans
->transfer
;
683 trans
->transfer
.stride
= rtex
->surface
.level
[level
].pitch_bytes
;
684 trans
->transfer
.layer_stride
= rtex
->surface
.level
[level
].slice_size
;
685 trans
->offset
= r600_texture_get_offset(rtex
, level
, box
->z
);
686 return &trans
->transfer
;
689 void r600_texture_transfer_destroy(struct pipe_context
*ctx
,
690 struct pipe_transfer
*transfer
)
692 struct r600_transfer
*rtransfer
= (struct r600_transfer
*)transfer
;
693 struct pipe_resource
*texture
= transfer
->resource
;
694 struct r600_texture
*rtex
= (struct r600_texture
*)texture
;
696 if ((transfer
->usage
& PIPE_TRANSFER_WRITE
) && rtransfer
->staging
) {
697 if (rtex
->is_depth
) {
698 ctx
->resource_copy_region(ctx
, texture
, transfer
->level
,
699 transfer
->box
.x
, transfer
->box
.y
, transfer
->box
.z
,
700 &rtransfer
->staging
->b
.b
, transfer
->level
,
703 r600_copy_from_staging_texture(ctx
, rtransfer
);
707 if (rtransfer
->staging
)
708 pipe_resource_reference((struct pipe_resource
**)&rtransfer
->staging
, NULL
);
710 pipe_resource_reference(&transfer
->resource
, NULL
);
714 void* r600_texture_transfer_map(struct pipe_context
*ctx
,
715 struct pipe_transfer
* transfer
)
717 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
718 struct r600_transfer
*rtransfer
= (struct r600_transfer
*)transfer
;
719 struct radeon_winsys_cs_handle
*buf
;
720 struct r600_texture
*rtex
=
721 (struct r600_texture
*)transfer
->resource
;
722 enum pipe_format format
= transfer
->resource
->format
;
726 if ((transfer
->resource
->bind
& PIPE_BIND_GLOBAL
) && transfer
->resource
->target
== PIPE_BUFFER
) {
727 return r600_compute_global_transfer_map(ctx
, transfer
);
730 if (rtransfer
->staging
) {
731 buf
= ((struct r600_resource
*)rtransfer
->staging
)->cs_buf
;
733 buf
= ((struct r600_resource
*)transfer
->resource
)->cs_buf
;
736 if (rtex
->is_depth
|| !rtransfer
->staging
)
737 offset
= rtransfer
->offset
+
738 transfer
->box
.y
/ util_format_get_blockheight(format
) * transfer
->stride
+
739 transfer
->box
.x
/ util_format_get_blockwidth(format
) * util_format_get_blocksize(format
);
741 if (!(map
= rctx
->ws
->buffer_map(buf
, rctx
->cs
, transfer
->usage
))) {
748 void r600_texture_transfer_unmap(struct pipe_context
*ctx
,
749 struct pipe_transfer
* transfer
)
751 struct r600_transfer
*rtransfer
= (struct r600_transfer
*)transfer
;
752 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
753 struct radeon_winsys_cs_handle
*buf
;
755 if ((transfer
->resource
->bind
& PIPE_BIND_GLOBAL
) && transfer
->resource
->target
== PIPE_BUFFER
) {
756 return r600_compute_global_transfer_unmap(ctx
, transfer
);
759 if (rtransfer
->staging
) {
760 buf
= ((struct r600_resource
*)rtransfer
->staging
)->cs_buf
;
762 buf
= ((struct r600_resource
*)transfer
->resource
)->cs_buf
;
764 rctx
->ws
->buffer_unmap(buf
);
767 void r600_init_surface_functions(struct r600_context
*r600
)
769 r600
->context
.create_surface
= r600_create_surface
;
770 r600
->context
.surface_destroy
= r600_surface_destroy
;
773 static unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format
,
774 const unsigned char *swizzle_view
)
777 unsigned char swizzle
[4];
779 const uint32_t swizzle_shift
[4] = {
782 const uint32_t swizzle_bit
[4] = {
787 util_format_compose_swizzles(swizzle_format
, swizzle_view
, swizzle
);
789 memcpy(swizzle
, swizzle_format
, 4);
793 for (i
= 0; i
< 4; i
++) {
794 switch (swizzle
[i
]) {
795 case UTIL_FORMAT_SWIZZLE_Y
:
796 result
|= swizzle_bit
[1] << swizzle_shift
[i
];
798 case UTIL_FORMAT_SWIZZLE_Z
:
799 result
|= swizzle_bit
[2] << swizzle_shift
[i
];
801 case UTIL_FORMAT_SWIZZLE_W
:
802 result
|= swizzle_bit
[3] << swizzle_shift
[i
];
804 case UTIL_FORMAT_SWIZZLE_0
:
805 result
|= V_038010_SQ_SEL_0
<< swizzle_shift
[i
];
807 case UTIL_FORMAT_SWIZZLE_1
:
808 result
|= V_038010_SQ_SEL_1
<< swizzle_shift
[i
];
810 default: /* UTIL_FORMAT_SWIZZLE_X */
811 result
|= swizzle_bit
[0] << swizzle_shift
[i
];
817 /* texture format translate */
818 uint32_t r600_translate_texformat(struct pipe_screen
*screen
,
819 enum pipe_format format
,
820 const unsigned char *swizzle_view
,
821 uint32_t *word4_p
, uint32_t *yuv_format_p
)
823 uint32_t result
= 0, word4
= 0, yuv_format
= 0;
824 const struct util_format_description
*desc
;
825 boolean uniform
= TRUE
;
826 static int r600_enable_s3tc
= -1;
827 bool is_srgb_valid
= FALSE
;
830 const uint32_t sign_bit
[4] = {
831 S_038010_FORMAT_COMP_X(V_038010_SQ_FORMAT_COMP_SIGNED
),
832 S_038010_FORMAT_COMP_Y(V_038010_SQ_FORMAT_COMP_SIGNED
),
833 S_038010_FORMAT_COMP_Z(V_038010_SQ_FORMAT_COMP_SIGNED
),
834 S_038010_FORMAT_COMP_W(V_038010_SQ_FORMAT_COMP_SIGNED
)
836 desc
= util_format_description(format
);
838 word4
|= r600_get_swizzle_combined(desc
->swizzle
, swizzle_view
);
840 /* Colorspace (return non-RGB formats directly). */
841 switch (desc
->colorspace
) {
842 /* Depth stencil formats */
843 case UTIL_FORMAT_COLORSPACE_ZS
:
845 case PIPE_FORMAT_Z16_UNORM
:
848 case PIPE_FORMAT_X24S8_UINT
:
849 word4
|= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT
);
850 case PIPE_FORMAT_Z24X8_UNORM
:
851 case PIPE_FORMAT_Z24_UNORM_S8_UINT
:
854 case PIPE_FORMAT_S8X24_UINT
:
855 word4
|= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT
);
856 case PIPE_FORMAT_X8Z24_UNORM
:
857 case PIPE_FORMAT_S8_UINT_Z24_UNORM
:
860 case PIPE_FORMAT_S8_UINT
:
862 word4
|= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT
);
864 case PIPE_FORMAT_Z32_FLOAT
:
865 result
= FMT_32_FLOAT
;
867 case PIPE_FORMAT_X32_S8X24_UINT
:
868 word4
|= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT
);
869 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
:
870 result
= FMT_X24_8_32_FLOAT
;
876 case UTIL_FORMAT_COLORSPACE_YUV
:
877 yuv_format
|= (1 << 30);
879 case PIPE_FORMAT_UYVY
:
880 case PIPE_FORMAT_YUYV
:
884 goto out_unknown
; /* XXX */
886 case UTIL_FORMAT_COLORSPACE_SRGB
:
887 word4
|= S_038010_FORCE_DEGAMMA(1);
894 if (r600_enable_s3tc
== -1) {
895 struct r600_screen
*rscreen
= (struct r600_screen
*)screen
;
896 if (rscreen
->info
.drm_minor
>= 9)
897 r600_enable_s3tc
= 1;
899 r600_enable_s3tc
= debug_get_bool_option("R600_ENABLE_S3TC", FALSE
);
902 if (desc
->layout
== UTIL_FORMAT_LAYOUT_RGTC
) {
903 if (!r600_enable_s3tc
)
907 case PIPE_FORMAT_RGTC1_SNORM
:
908 case PIPE_FORMAT_LATC1_SNORM
:
909 word4
|= sign_bit
[0];
910 case PIPE_FORMAT_RGTC1_UNORM
:
911 case PIPE_FORMAT_LATC1_UNORM
:
914 case PIPE_FORMAT_RGTC2_SNORM
:
915 case PIPE_FORMAT_LATC2_SNORM
:
916 word4
|= sign_bit
[0] | sign_bit
[1];
917 case PIPE_FORMAT_RGTC2_UNORM
:
918 case PIPE_FORMAT_LATC2_UNORM
:
926 if (desc
->layout
== UTIL_FORMAT_LAYOUT_S3TC
) {
928 if (!r600_enable_s3tc
)
931 if (!util_format_s3tc_enabled
) {
936 case PIPE_FORMAT_DXT1_RGB
:
937 case PIPE_FORMAT_DXT1_RGBA
:
938 case PIPE_FORMAT_DXT1_SRGB
:
939 case PIPE_FORMAT_DXT1_SRGBA
:
941 is_srgb_valid
= TRUE
;
943 case PIPE_FORMAT_DXT3_RGBA
:
944 case PIPE_FORMAT_DXT3_SRGBA
:
946 is_srgb_valid
= TRUE
;
948 case PIPE_FORMAT_DXT5_RGBA
:
949 case PIPE_FORMAT_DXT5_SRGBA
:
951 is_srgb_valid
= TRUE
;
958 if (desc
->layout
== UTIL_FORMAT_LAYOUT_SUBSAMPLED
) {
960 case PIPE_FORMAT_R8G8_B8G8_UNORM
:
961 case PIPE_FORMAT_G8R8_B8R8_UNORM
:
964 case PIPE_FORMAT_G8R8_G8B8_UNORM
:
965 case PIPE_FORMAT_R8G8_R8B8_UNORM
:
973 if (format
== PIPE_FORMAT_R9G9B9E5_FLOAT
) {
974 result
= FMT_5_9_9_9_SHAREDEXP
;
976 } else if (format
== PIPE_FORMAT_R11G11B10_FLOAT
) {
977 result
= FMT_10_11_11_FLOAT
;
982 for (i
= 0; i
< desc
->nr_channels
; i
++) {
983 if (desc
->channel
[i
].type
== UTIL_FORMAT_TYPE_SIGNED
) {
984 word4
|= sign_bit
[i
];
988 /* R8G8Bx_SNORM - XXX CxV8U8 */
990 /* See whether the components are of the same size. */
991 for (i
= 1; i
< desc
->nr_channels
; i
++) {
992 uniform
= uniform
&& desc
->channel
[0].size
== desc
->channel
[i
].size
;
995 /* Non-uniform formats. */
997 if (desc
->colorspace
!= UTIL_FORMAT_COLORSPACE_SRGB
&&
998 desc
->channel
[0].pure_integer
)
999 word4
|= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT
);
1000 switch(desc
->nr_channels
) {
1002 if (desc
->channel
[0].size
== 5 &&
1003 desc
->channel
[1].size
== 6 &&
1004 desc
->channel
[2].size
== 5) {
1010 if (desc
->channel
[0].size
== 5 &&
1011 desc
->channel
[1].size
== 5 &&
1012 desc
->channel
[2].size
== 5 &&
1013 desc
->channel
[3].size
== 1) {
1014 result
= FMT_1_5_5_5
;
1017 if (desc
->channel
[0].size
== 10 &&
1018 desc
->channel
[1].size
== 10 &&
1019 desc
->channel
[2].size
== 10 &&
1020 desc
->channel
[3].size
== 2) {
1021 result
= FMT_2_10_10_10
;
1029 /* Find the first non-VOID channel. */
1030 for (i
= 0; i
< 4; i
++) {
1031 if (desc
->channel
[i
].type
!= UTIL_FORMAT_TYPE_VOID
) {
1039 /* uniform formats */
1040 switch (desc
->channel
[i
].type
) {
1041 case UTIL_FORMAT_TYPE_UNSIGNED
:
1042 case UTIL_FORMAT_TYPE_SIGNED
:
1044 if (!desc
->channel
[i
].normalized
&&
1045 desc
->colorspace
!= UTIL_FORMAT_COLORSPACE_SRGB
) {
1049 if (desc
->colorspace
!= UTIL_FORMAT_COLORSPACE_SRGB
&&
1050 desc
->channel
[i
].pure_integer
)
1051 word4
|= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT
);
1053 switch (desc
->channel
[i
].size
) {
1055 switch (desc
->nr_channels
) {
1060 result
= FMT_4_4_4_4
;
1065 switch (desc
->nr_channels
) {
1073 result
= FMT_8_8_8_8
;
1074 is_srgb_valid
= TRUE
;
1079 switch (desc
->nr_channels
) {
1087 result
= FMT_16_16_16_16
;
1092 switch (desc
->nr_channels
) {
1100 result
= FMT_32_32_32_32
;
1106 case UTIL_FORMAT_TYPE_FLOAT
:
1107 switch (desc
->channel
[i
].size
) {
1109 switch (desc
->nr_channels
) {
1111 result
= FMT_16_FLOAT
;
1114 result
= FMT_16_16_FLOAT
;
1117 result
= FMT_16_16_16_16_FLOAT
;
1122 switch (desc
->nr_channels
) {
1124 result
= FMT_32_FLOAT
;
1127 result
= FMT_32_32_FLOAT
;
1130 result
= FMT_32_32_32_32_FLOAT
;
1139 if (desc
->colorspace
== UTIL_FORMAT_COLORSPACE_SRGB
&& !is_srgb_valid
)
1144 *yuv_format_p
= yuv_format
;
1147 /* R600_ERR("Unable to handle texformat %d %s\n", format, util_format_name(format)); */