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 if (rscreen
->chip_class
>= EVERGREEN
) {
286 fmask
.bankh
= nr_samples
<= 4 ? 4 : 1;
289 if (rscreen
->ws
->surface_init(rscreen
->ws
, &fmask
)) {
290 R600_ERR("Got error in surface_init while allocating FMASK.\n");
293 assert(fmask
.level
[0].mode
== RADEON_SURF_MODE_2D
);
295 /* Reserve space for FMASK while converting bits back to bytes. */
296 rtex
->fmask_bank_height
= fmask
.bankh
;
297 rtex
->fmask_offset
= align(rtex
->size
, MAX2(256, fmask
.bo_alignment
));
298 rtex
->fmask_size
= (fmask
.bo_size
+ 7) / 8;
299 rtex
->size
= rtex
->fmask_offset
+ rtex
->fmask_size
;
301 printf("FMASK width=%u, height=%i, bits=%u, size=%u\n",
302 fmask
.npix_x
, fmask
.npix_y
, fmask
.bpe
* fmask
.nsamples
, rtex
->fmask_size
);
306 static void r600_texture_allocate_cmask(struct r600_screen
*rscreen
,
307 struct r600_texture
*rtex
)
309 unsigned cmask_tile_width
= 8;
310 unsigned cmask_tile_height
= 8;
311 unsigned cmask_tile_elements
= cmask_tile_width
* cmask_tile_height
;
312 unsigned element_bits
= 4;
313 unsigned cmask_cache_bits
= 1024;
314 unsigned num_pipes
= rscreen
->tiling_info
.num_channels
;
315 unsigned pipe_interleave_bytes
= rscreen
->tiling_info
.group_bytes
;
317 unsigned elements_per_macro_tile
= (cmask_cache_bits
/ element_bits
) * num_pipes
;
318 unsigned pixels_per_macro_tile
= elements_per_macro_tile
* cmask_tile_elements
;
319 unsigned sqrt_pixels_per_macro_tile
= sqrt(pixels_per_macro_tile
);
320 unsigned macro_tile_width
= util_next_power_of_two(sqrt_pixels_per_macro_tile
);
321 unsigned macro_tile_height
= pixels_per_macro_tile
/ macro_tile_width
;
323 unsigned pitch_elements
= align(rtex
->surface
.npix_x
, macro_tile_width
);
324 unsigned height
= align(rtex
->surface
.npix_y
, macro_tile_height
);
326 unsigned base_align
= num_pipes
* pipe_interleave_bytes
;
327 unsigned slice_bytes
=
328 ((pitch_elements
* height
* element_bits
+ 7) / 8) / cmask_tile_elements
;
329 unsigned size
= rtex
->surface
.array_size
* align(slice_bytes
, base_align
);
331 assert(macro_tile_width
% 128 == 0);
332 assert(macro_tile_height
% 128 == 0);
334 rtex
->cmask_slice_tile_max
= ((pitch_elements
* height
) / (128*128)) - 1;
335 rtex
->cmask_offset
= align(rtex
->size
, MAX2(256, base_align
));
336 rtex
->cmask_size
= size
;
337 rtex
->size
= rtex
->cmask_offset
+ rtex
->cmask_size
;
339 printf("CMASK: macro tile width = %u, macro tile height = %u, "
340 "pitch elements = %u, height = %u, slice tile max = %u\n",
341 macro_tile_width
, macro_tile_height
, pitch_elements
, height
,
342 rtex
->cmask_slice_tile_max
);
346 static struct r600_texture
*
347 r600_texture_create_object(struct pipe_screen
*screen
,
348 const struct pipe_resource
*base
,
349 unsigned pitch_in_bytes_override
,
350 struct pb_buffer
*buf
,
352 struct radeon_surface
*surface
)
354 struct r600_texture
*rtex
;
355 struct r600_resource
*resource
;
356 struct r600_screen
*rscreen
= (struct r600_screen
*)screen
;
359 rtex
= CALLOC_STRUCT(r600_texture
);
363 resource
= &rtex
->resource
;
364 resource
->b
.b
= *base
;
365 resource
->b
.vtbl
= &r600_texture_vtbl
;
366 pipe_reference_init(&resource
->b
.b
.reference
, 1);
367 resource
->b
.b
.screen
= screen
;
368 rtex
->pitch_override
= pitch_in_bytes_override
;
370 /* don't include stencil-only formats which we don't support for rendering */
371 rtex
->is_depth
= util_format_has_depth(util_format_description(rtex
->resource
.b
.b
.format
));
373 rtex
->surface
= *surface
;
374 r
= r600_setup_surface(screen
, rtex
,
375 pitch_in_bytes_override
);
381 if (base
->nr_samples
> 1 && !rtex
->is_depth
&& alloc_bo
) {
382 r600_texture_allocate_fmask(rscreen
, rtex
);
383 r600_texture_allocate_cmask(rscreen
, rtex
);
386 if (!rtex
->is_depth
&& base
->nr_samples
> 1 &&
387 (!rtex
->fmask_size
|| !rtex
->cmask_size
)) {
392 /* Now create the backing buffer. */
393 if (!buf
&& alloc_bo
) {
394 unsigned base_align
= rtex
->surface
.bo_alignment
;
395 unsigned usage
= R600_TEX_IS_TILED(rtex
, 0) ? PIPE_USAGE_STATIC
: base
->usage
;
397 if (!r600_init_resource(rscreen
, resource
, rtex
->size
, base_align
, base
->bind
, usage
)) {
403 resource
->cs_buf
= rscreen
->ws
->buffer_get_cs_handle(buf
);
404 resource
->domains
= RADEON_DOMAIN_GTT
| RADEON_DOMAIN_VRAM
;
407 if (rtex
->cmask_size
) {
408 /* Initialize the cmask to 0xCC (= compressed state). */
409 char *ptr
= rscreen
->ws
->buffer_map(resource
->cs_buf
, NULL
, PIPE_TRANSFER_WRITE
);
410 memset(ptr
+ rtex
->cmask_offset
, 0xCC, rtex
->cmask_size
);
411 rscreen
->ws
->buffer_unmap(resource
->cs_buf
);
416 struct pipe_resource
*r600_texture_create(struct pipe_screen
*screen
,
417 const struct pipe_resource
*templ
)
419 struct r600_screen
*rscreen
= (struct r600_screen
*)screen
;
420 struct radeon_surface surface
;
421 unsigned array_mode
= 0;
424 if (!(templ
->flags
& R600_RESOURCE_FLAG_TRANSFER
)) {
425 if (!(templ
->bind
& PIPE_BIND_SCANOUT
) &&
426 templ
->usage
!= PIPE_USAGE_STAGING
&&
427 templ
->usage
!= PIPE_USAGE_STREAM
) {
428 array_mode
= V_038000_ARRAY_2D_TILED_THIN1
;
429 } else if (util_format_is_compressed(templ
->format
)) {
430 array_mode
= V_038000_ARRAY_1D_TILED_THIN1
;
434 /* XXX tiling is broken for the 422 formats */
435 if (util_format_description(templ
->format
)->layout
== UTIL_FORMAT_LAYOUT_SUBSAMPLED
)
436 array_mode
= V_038000_ARRAY_LINEAR_ALIGNED
;
438 r
= r600_init_surface(rscreen
, &surface
, templ
, array_mode
,
439 templ
->flags
& R600_RESOURCE_FLAG_TRANSFER
,
440 templ
->flags
& R600_RESOURCE_FLAG_FLUSHED_DEPTH
);
444 r
= rscreen
->ws
->surface_best(rscreen
->ws
, &surface
);
448 return (struct pipe_resource
*)r600_texture_create_object(screen
, templ
,
449 0, NULL
, TRUE
, &surface
);
452 static struct pipe_surface
*r600_create_surface(struct pipe_context
*pipe
,
453 struct pipe_resource
*texture
,
454 const struct pipe_surface
*templ
)
456 struct r600_texture
*rtex
= (struct r600_texture
*)texture
;
457 struct r600_surface
*surface
= CALLOC_STRUCT(r600_surface
);
458 unsigned level
= templ
->u
.tex
.level
;
460 assert(templ
->u
.tex
.first_layer
== templ
->u
.tex
.last_layer
);
463 pipe_reference_init(&surface
->base
.reference
, 1);
464 pipe_resource_reference(&surface
->base
.texture
, texture
);
465 surface
->base
.context
= pipe
;
466 surface
->base
.format
= templ
->format
;
467 surface
->base
.width
= rtex
->surface
.level
[level
].npix_x
;
468 surface
->base
.height
= rtex
->surface
.level
[level
].npix_y
;
469 surface
->base
.usage
= templ
->usage
;
470 surface
->base
.u
= templ
->u
;
471 return &surface
->base
;
474 static void r600_surface_destroy(struct pipe_context
*pipe
,
475 struct pipe_surface
*surface
)
477 pipe_resource_reference(&surface
->texture
, NULL
);
481 struct pipe_resource
*r600_texture_from_handle(struct pipe_screen
*screen
,
482 const struct pipe_resource
*templ
,
483 struct winsys_handle
*whandle
)
485 struct r600_screen
*rscreen
= (struct r600_screen
*)screen
;
486 struct pb_buffer
*buf
= NULL
;
488 unsigned array_mode
= 0;
489 enum radeon_bo_layout micro
, macro
;
490 struct radeon_surface surface
;
493 /* Support only 2D textures without mipmaps */
494 if ((templ
->target
!= PIPE_TEXTURE_2D
&& templ
->target
!= PIPE_TEXTURE_RECT
) ||
495 templ
->depth0
!= 1 || templ
->last_level
!= 0)
498 buf
= rscreen
->ws
->buffer_from_handle(rscreen
->ws
, whandle
, &stride
);
502 rscreen
->ws
->buffer_get_tiling(buf
, µ
, ¯o
,
503 &surface
.bankw
, &surface
.bankh
,
505 &surface
.stencil_tile_split
,
508 if (macro
== RADEON_LAYOUT_TILED
)
509 array_mode
= V_0280A0_ARRAY_2D_TILED_THIN1
;
510 else if (micro
== RADEON_LAYOUT_TILED
)
511 array_mode
= V_0280A0_ARRAY_1D_TILED_THIN1
;
515 r
= r600_init_surface(rscreen
, &surface
, templ
, array_mode
, false, false);
519 return (struct pipe_resource
*)r600_texture_create_object(screen
, templ
,
520 stride
, buf
, FALSE
, &surface
);
523 bool r600_init_flushed_depth_texture(struct pipe_context
*ctx
,
524 struct pipe_resource
*texture
,
525 struct r600_texture
**staging
)
527 struct r600_texture
*rtex
= (struct r600_texture
*)texture
;
528 struct pipe_resource resource
;
529 struct r600_texture
**flushed_depth_texture
= staging
?
530 staging
: &rtex
->flushed_depth_texture
;
532 if (!staging
&& rtex
->flushed_depth_texture
)
533 return true; /* it's ready */
535 resource
.target
= texture
->target
;
536 resource
.format
= texture
->format
;
537 resource
.width0
= texture
->width0
;
538 resource
.height0
= texture
->height0
;
539 resource
.depth0
= texture
->depth0
;
540 resource
.array_size
= texture
->array_size
;
541 resource
.last_level
= texture
->last_level
;
542 resource
.nr_samples
= texture
->nr_samples
;
543 resource
.usage
= staging
? PIPE_USAGE_STAGING
: PIPE_USAGE_STATIC
;
544 resource
.bind
= texture
->bind
& ~PIPE_BIND_DEPTH_STENCIL
;
545 resource
.flags
= texture
->flags
| R600_RESOURCE_FLAG_FLUSHED_DEPTH
;
548 resource
.flags
|= R600_RESOURCE_FLAG_TRANSFER
;
550 *flushed_depth_texture
= (struct r600_texture
*)ctx
->screen
->resource_create(ctx
->screen
, &resource
);
551 if (*flushed_depth_texture
== NULL
) {
552 R600_ERR("failed to create temporary texture to hold flushed depth\n");
556 (*flushed_depth_texture
)->is_flushing_texture
= TRUE
;
560 /* Needs adjustment for pixelformat:
562 static INLINE
unsigned u_box_volume( const struct pipe_box
*box
)
564 return box
->width
* box
->depth
* box
->height
;
567 struct pipe_transfer
* r600_texture_get_transfer(struct pipe_context
*ctx
,
568 struct pipe_resource
*texture
,
571 const struct pipe_box
*box
)
573 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
574 struct r600_texture
*rtex
= (struct r600_texture
*)texture
;
575 struct pipe_resource resource
;
576 struct r600_transfer
*trans
;
577 boolean use_staging_texture
= FALSE
;
579 /* We cannot map a tiled texture directly because the data is
580 * in a different order, therefore we do detiling using a blit.
582 * Also, use a temporary in GTT memory for read transfers, as
583 * the CPU is much happier reading out of cached system memory
584 * than uncached VRAM.
586 if (R600_TEX_IS_TILED(rtex
, level
)) {
587 use_staging_texture
= TRUE
;
590 if ((usage
& PIPE_TRANSFER_READ
) && u_box_volume(box
) > 1024)
591 use_staging_texture
= TRUE
;
593 /* Use a staging texture for uploads if the underlying BO is busy. */
594 if (!(usage
& PIPE_TRANSFER_READ
) &&
595 (rctx
->ws
->cs_is_buffer_referenced(rctx
->cs
, rtex
->resource
.cs_buf
, RADEON_USAGE_READWRITE
) ||
596 rctx
->ws
->buffer_is_busy(rtex
->resource
.buf
, RADEON_USAGE_READWRITE
))) {
597 use_staging_texture
= TRUE
;
600 if (texture
->flags
& R600_RESOURCE_FLAG_TRANSFER
) {
601 use_staging_texture
= FALSE
;
604 if (use_staging_texture
&& (usage
& PIPE_TRANSFER_MAP_DIRECTLY
)) {
608 trans
= CALLOC_STRUCT(r600_transfer
);
611 pipe_resource_reference(&trans
->transfer
.resource
, texture
);
612 trans
->transfer
.level
= level
;
613 trans
->transfer
.usage
= usage
;
614 trans
->transfer
.box
= *box
;
615 if (rtex
->is_depth
) {
616 /* XXX: only readback the rectangle which is being mapped?
618 /* XXX: when discard is true, no need to read back from depth texture
620 struct r600_texture
*staging_depth
;
622 if (!r600_init_flushed_depth_texture(ctx
, texture
, &staging_depth
)) {
623 R600_ERR("failed to create temporary texture to hold untiled copy\n");
624 pipe_resource_reference(&trans
->transfer
.resource
, NULL
);
629 r600_blit_decompress_depth(ctx
, rtex
, staging_depth
,
631 box
->z
, box
->z
+ box
->depth
- 1,
634 trans
->transfer
.stride
= staging_depth
->surface
.level
[level
].pitch_bytes
;
635 trans
->offset
= r600_texture_get_offset(staging_depth
, level
, box
->z
);
636 trans
->staging
= (struct r600_resource
*)staging_depth
;
637 return &trans
->transfer
;
638 } else if (use_staging_texture
) {
639 resource
.target
= PIPE_TEXTURE_2D
;
640 resource
.format
= texture
->format
;
641 resource
.width0
= box
->width
;
642 resource
.height0
= box
->height
;
644 resource
.array_size
= 1;
645 resource
.last_level
= 0;
646 resource
.nr_samples
= 0;
647 resource
.usage
= PIPE_USAGE_STAGING
;
649 resource
.flags
= R600_RESOURCE_FLAG_TRANSFER
;
650 /* For texture reading, the temporary (detiled) texture is used as
651 * a render target when blitting from a tiled texture. */
652 if (usage
& PIPE_TRANSFER_READ
) {
653 resource
.bind
|= PIPE_BIND_RENDER_TARGET
;
655 /* For texture writing, the temporary texture is used as a sampler
656 * when blitting into a tiled texture. */
657 if (usage
& PIPE_TRANSFER_WRITE
) {
658 resource
.bind
|= PIPE_BIND_SAMPLER_VIEW
;
660 /* Create the temporary texture. */
661 trans
->staging
= (struct r600_resource
*)ctx
->screen
->resource_create(ctx
->screen
, &resource
);
662 if (trans
->staging
== NULL
) {
663 R600_ERR("failed to create temporary texture to hold untiled copy\n");
664 pipe_resource_reference(&trans
->transfer
.resource
, NULL
);
669 trans
->transfer
.stride
=
670 ((struct r600_texture
*)trans
->staging
)->surface
.level
[0].pitch_bytes
;
671 if (usage
& PIPE_TRANSFER_READ
) {
672 r600_copy_to_staging_texture(ctx
, trans
);
673 /* Always referenced in the blit. */
674 r600_flush(ctx
, NULL
, 0);
676 return &trans
->transfer
;
678 trans
->transfer
.stride
= rtex
->surface
.level
[level
].pitch_bytes
;
679 trans
->transfer
.layer_stride
= rtex
->surface
.level
[level
].slice_size
;
680 trans
->offset
= r600_texture_get_offset(rtex
, level
, box
->z
);
681 return &trans
->transfer
;
684 void r600_texture_transfer_destroy(struct pipe_context
*ctx
,
685 struct pipe_transfer
*transfer
)
687 struct r600_transfer
*rtransfer
= (struct r600_transfer
*)transfer
;
688 struct pipe_resource
*texture
= transfer
->resource
;
689 struct r600_texture
*rtex
= (struct r600_texture
*)texture
;
691 if ((transfer
->usage
& PIPE_TRANSFER_WRITE
) && rtransfer
->staging
) {
692 if (rtex
->is_depth
) {
693 ctx
->resource_copy_region(ctx
, texture
, transfer
->level
,
694 transfer
->box
.x
, transfer
->box
.y
, transfer
->box
.z
,
695 &rtransfer
->staging
->b
.b
, transfer
->level
,
698 r600_copy_from_staging_texture(ctx
, rtransfer
);
702 if (rtransfer
->staging
)
703 pipe_resource_reference((struct pipe_resource
**)&rtransfer
->staging
, NULL
);
705 pipe_resource_reference(&transfer
->resource
, NULL
);
709 void* r600_texture_transfer_map(struct pipe_context
*ctx
,
710 struct pipe_transfer
* transfer
)
712 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
713 struct r600_transfer
*rtransfer
= (struct r600_transfer
*)transfer
;
714 struct radeon_winsys_cs_handle
*buf
;
715 struct r600_texture
*rtex
=
716 (struct r600_texture
*)transfer
->resource
;
717 enum pipe_format format
= transfer
->resource
->format
;
721 if ((transfer
->resource
->bind
& PIPE_BIND_GLOBAL
) && transfer
->resource
->target
== PIPE_BUFFER
) {
722 return r600_compute_global_transfer_map(ctx
, transfer
);
725 if (rtransfer
->staging
) {
726 buf
= ((struct r600_resource
*)rtransfer
->staging
)->cs_buf
;
728 buf
= ((struct r600_resource
*)transfer
->resource
)->cs_buf
;
731 if (rtex
->is_depth
|| !rtransfer
->staging
)
732 offset
= rtransfer
->offset
+
733 transfer
->box
.y
/ util_format_get_blockheight(format
) * transfer
->stride
+
734 transfer
->box
.x
/ util_format_get_blockwidth(format
) * util_format_get_blocksize(format
);
736 if (!(map
= rctx
->ws
->buffer_map(buf
, rctx
->cs
, transfer
->usage
))) {
743 void r600_texture_transfer_unmap(struct pipe_context
*ctx
,
744 struct pipe_transfer
* transfer
)
746 struct r600_transfer
*rtransfer
= (struct r600_transfer
*)transfer
;
747 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
748 struct radeon_winsys_cs_handle
*buf
;
750 if ((transfer
->resource
->bind
& PIPE_BIND_GLOBAL
) && transfer
->resource
->target
== PIPE_BUFFER
) {
751 return r600_compute_global_transfer_unmap(ctx
, transfer
);
754 if (rtransfer
->staging
) {
755 buf
= ((struct r600_resource
*)rtransfer
->staging
)->cs_buf
;
757 buf
= ((struct r600_resource
*)transfer
->resource
)->cs_buf
;
759 rctx
->ws
->buffer_unmap(buf
);
762 void r600_init_surface_functions(struct r600_context
*r600
)
764 r600
->context
.create_surface
= r600_create_surface
;
765 r600
->context
.surface_destroy
= r600_surface_destroy
;
768 static unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format
,
769 const unsigned char *swizzle_view
)
772 unsigned char swizzle
[4];
774 const uint32_t swizzle_shift
[4] = {
777 const uint32_t swizzle_bit
[4] = {
782 util_format_compose_swizzles(swizzle_format
, swizzle_view
, swizzle
);
784 memcpy(swizzle
, swizzle_format
, 4);
788 for (i
= 0; i
< 4; i
++) {
789 switch (swizzle
[i
]) {
790 case UTIL_FORMAT_SWIZZLE_Y
:
791 result
|= swizzle_bit
[1] << swizzle_shift
[i
];
793 case UTIL_FORMAT_SWIZZLE_Z
:
794 result
|= swizzle_bit
[2] << swizzle_shift
[i
];
796 case UTIL_FORMAT_SWIZZLE_W
:
797 result
|= swizzle_bit
[3] << swizzle_shift
[i
];
799 case UTIL_FORMAT_SWIZZLE_0
:
800 result
|= V_038010_SQ_SEL_0
<< swizzle_shift
[i
];
802 case UTIL_FORMAT_SWIZZLE_1
:
803 result
|= V_038010_SQ_SEL_1
<< swizzle_shift
[i
];
805 default: /* UTIL_FORMAT_SWIZZLE_X */
806 result
|= swizzle_bit
[0] << swizzle_shift
[i
];
812 /* texture format translate */
813 uint32_t r600_translate_texformat(struct pipe_screen
*screen
,
814 enum pipe_format format
,
815 const unsigned char *swizzle_view
,
816 uint32_t *word4_p
, uint32_t *yuv_format_p
)
818 uint32_t result
= 0, word4
= 0, yuv_format
= 0;
819 const struct util_format_description
*desc
;
820 boolean uniform
= TRUE
;
821 static int r600_enable_s3tc
= -1;
822 bool is_srgb_valid
= FALSE
;
825 const uint32_t sign_bit
[4] = {
826 S_038010_FORMAT_COMP_X(V_038010_SQ_FORMAT_COMP_SIGNED
),
827 S_038010_FORMAT_COMP_Y(V_038010_SQ_FORMAT_COMP_SIGNED
),
828 S_038010_FORMAT_COMP_Z(V_038010_SQ_FORMAT_COMP_SIGNED
),
829 S_038010_FORMAT_COMP_W(V_038010_SQ_FORMAT_COMP_SIGNED
)
831 desc
= util_format_description(format
);
833 word4
|= r600_get_swizzle_combined(desc
->swizzle
, swizzle_view
);
835 /* Colorspace (return non-RGB formats directly). */
836 switch (desc
->colorspace
) {
837 /* Depth stencil formats */
838 case UTIL_FORMAT_COLORSPACE_ZS
:
840 case PIPE_FORMAT_Z16_UNORM
:
843 case PIPE_FORMAT_X24S8_UINT
:
844 word4
|= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT
);
845 case PIPE_FORMAT_Z24X8_UNORM
:
846 case PIPE_FORMAT_Z24_UNORM_S8_UINT
:
849 case PIPE_FORMAT_S8X24_UINT
:
850 word4
|= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT
);
851 case PIPE_FORMAT_X8Z24_UNORM
:
852 case PIPE_FORMAT_S8_UINT_Z24_UNORM
:
855 case PIPE_FORMAT_S8_UINT
:
857 word4
|= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT
);
859 case PIPE_FORMAT_Z32_FLOAT
:
860 result
= FMT_32_FLOAT
;
862 case PIPE_FORMAT_X32_S8X24_UINT
:
863 word4
|= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT
);
864 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
:
865 result
= FMT_X24_8_32_FLOAT
;
871 case UTIL_FORMAT_COLORSPACE_YUV
:
872 yuv_format
|= (1 << 30);
874 case PIPE_FORMAT_UYVY
:
875 case PIPE_FORMAT_YUYV
:
879 goto out_unknown
; /* XXX */
881 case UTIL_FORMAT_COLORSPACE_SRGB
:
882 word4
|= S_038010_FORCE_DEGAMMA(1);
889 if (r600_enable_s3tc
== -1) {
890 struct r600_screen
*rscreen
= (struct r600_screen
*)screen
;
891 if (rscreen
->info
.drm_minor
>= 9)
892 r600_enable_s3tc
= 1;
894 r600_enable_s3tc
= debug_get_bool_option("R600_ENABLE_S3TC", FALSE
);
897 if (desc
->layout
== UTIL_FORMAT_LAYOUT_RGTC
) {
898 if (!r600_enable_s3tc
)
902 case PIPE_FORMAT_RGTC1_SNORM
:
903 case PIPE_FORMAT_LATC1_SNORM
:
904 word4
|= sign_bit
[0];
905 case PIPE_FORMAT_RGTC1_UNORM
:
906 case PIPE_FORMAT_LATC1_UNORM
:
909 case PIPE_FORMAT_RGTC2_SNORM
:
910 case PIPE_FORMAT_LATC2_SNORM
:
911 word4
|= sign_bit
[0] | sign_bit
[1];
912 case PIPE_FORMAT_RGTC2_UNORM
:
913 case PIPE_FORMAT_LATC2_UNORM
:
921 if (desc
->layout
== UTIL_FORMAT_LAYOUT_S3TC
) {
923 if (!r600_enable_s3tc
)
926 if (!util_format_s3tc_enabled
) {
931 case PIPE_FORMAT_DXT1_RGB
:
932 case PIPE_FORMAT_DXT1_RGBA
:
933 case PIPE_FORMAT_DXT1_SRGB
:
934 case PIPE_FORMAT_DXT1_SRGBA
:
936 is_srgb_valid
= TRUE
;
938 case PIPE_FORMAT_DXT3_RGBA
:
939 case PIPE_FORMAT_DXT3_SRGBA
:
941 is_srgb_valid
= TRUE
;
943 case PIPE_FORMAT_DXT5_RGBA
:
944 case PIPE_FORMAT_DXT5_SRGBA
:
946 is_srgb_valid
= TRUE
;
953 if (desc
->layout
== UTIL_FORMAT_LAYOUT_SUBSAMPLED
) {
955 case PIPE_FORMAT_R8G8_B8G8_UNORM
:
956 case PIPE_FORMAT_G8R8_B8R8_UNORM
:
959 case PIPE_FORMAT_G8R8_G8B8_UNORM
:
960 case PIPE_FORMAT_R8G8_R8B8_UNORM
:
968 if (format
== PIPE_FORMAT_R9G9B9E5_FLOAT
) {
969 result
= FMT_5_9_9_9_SHAREDEXP
;
971 } else if (format
== PIPE_FORMAT_R11G11B10_FLOAT
) {
972 result
= FMT_10_11_11_FLOAT
;
977 for (i
= 0; i
< desc
->nr_channels
; i
++) {
978 if (desc
->channel
[i
].type
== UTIL_FORMAT_TYPE_SIGNED
) {
979 word4
|= sign_bit
[i
];
983 /* R8G8Bx_SNORM - XXX CxV8U8 */
985 /* See whether the components are of the same size. */
986 for (i
= 1; i
< desc
->nr_channels
; i
++) {
987 uniform
= uniform
&& desc
->channel
[0].size
== desc
->channel
[i
].size
;
990 /* Non-uniform formats. */
992 if (desc
->colorspace
!= UTIL_FORMAT_COLORSPACE_SRGB
&&
993 desc
->channel
[0].pure_integer
)
994 word4
|= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT
);
995 switch(desc
->nr_channels
) {
997 if (desc
->channel
[0].size
== 5 &&
998 desc
->channel
[1].size
== 6 &&
999 desc
->channel
[2].size
== 5) {
1005 if (desc
->channel
[0].size
== 5 &&
1006 desc
->channel
[1].size
== 5 &&
1007 desc
->channel
[2].size
== 5 &&
1008 desc
->channel
[3].size
== 1) {
1009 result
= FMT_1_5_5_5
;
1012 if (desc
->channel
[0].size
== 10 &&
1013 desc
->channel
[1].size
== 10 &&
1014 desc
->channel
[2].size
== 10 &&
1015 desc
->channel
[3].size
== 2) {
1016 result
= FMT_2_10_10_10
;
1024 /* Find the first non-VOID channel. */
1025 for (i
= 0; i
< 4; i
++) {
1026 if (desc
->channel
[i
].type
!= UTIL_FORMAT_TYPE_VOID
) {
1034 /* uniform formats */
1035 switch (desc
->channel
[i
].type
) {
1036 case UTIL_FORMAT_TYPE_UNSIGNED
:
1037 case UTIL_FORMAT_TYPE_SIGNED
:
1039 if (!desc
->channel
[i
].normalized
&&
1040 desc
->colorspace
!= UTIL_FORMAT_COLORSPACE_SRGB
) {
1044 if (desc
->colorspace
!= UTIL_FORMAT_COLORSPACE_SRGB
&&
1045 desc
->channel
[i
].pure_integer
)
1046 word4
|= S_038010_NUM_FORMAT_ALL(V_038010_SQ_NUM_FORMAT_INT
);
1048 switch (desc
->channel
[i
].size
) {
1050 switch (desc
->nr_channels
) {
1055 result
= FMT_4_4_4_4
;
1060 switch (desc
->nr_channels
) {
1068 result
= FMT_8_8_8_8
;
1069 is_srgb_valid
= TRUE
;
1074 switch (desc
->nr_channels
) {
1082 result
= FMT_16_16_16_16
;
1087 switch (desc
->nr_channels
) {
1095 result
= FMT_32_32_32_32
;
1101 case UTIL_FORMAT_TYPE_FLOAT
:
1102 switch (desc
->channel
[i
].size
) {
1104 switch (desc
->nr_channels
) {
1106 result
= FMT_16_FLOAT
;
1109 result
= FMT_16_16_FLOAT
;
1112 result
= FMT_16_16_16_16_FLOAT
;
1117 switch (desc
->nr_channels
) {
1119 result
= FMT_32_FLOAT
;
1122 result
= FMT_32_32_FLOAT
;
1125 result
= FMT_32_32_32_32_FLOAT
;
1134 if (desc
->colorspace
== UTIL_FORMAT_COLORSPACE_SRGB
&& !is_srgb_valid
)
1139 *yuv_format_p
= yuv_format
;
1142 /* R600_ERR("Unable to handle texformat %d %s\n", format, util_format_name(format)); */