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_pipe_common.h"
29 #include "util/u_format.h"
30 #include "util/u_memory.h"
31 #include "util/u_pack_color.h"
35 /* Same as resource_copy_region, except that both upsampling and downsampling are allowed. */
36 static void r600_copy_region_with_blit(struct pipe_context
*pipe
,
37 struct pipe_resource
*dst
,
39 unsigned dstx
, unsigned dsty
, unsigned dstz
,
40 struct pipe_resource
*src
,
42 const struct pipe_box
*src_box
)
44 struct pipe_blit_info blit
;
46 memset(&blit
, 0, sizeof(blit
));
47 blit
.src
.resource
= src
;
48 blit
.src
.format
= src
->format
;
49 blit
.src
.level
= src_level
;
50 blit
.src
.box
= *src_box
;
51 blit
.dst
.resource
= dst
;
52 blit
.dst
.format
= dst
->format
;
53 blit
.dst
.level
= dst_level
;
54 blit
.dst
.box
.x
= dstx
;
55 blit
.dst
.box
.y
= dsty
;
56 blit
.dst
.box
.z
= dstz
;
57 blit
.dst
.box
.width
= src_box
->width
;
58 blit
.dst
.box
.height
= src_box
->height
;
59 blit
.dst
.box
.depth
= src_box
->depth
;
60 blit
.mask
= util_format_get_mask(src
->format
) &
61 util_format_get_mask(dst
->format
);
62 blit
.filter
= PIPE_TEX_FILTER_NEAREST
;
65 pipe
->blit(pipe
, &blit
);
69 /* Copy from a full GPU texture to a transfer's staging one. */
70 static void r600_copy_to_staging_texture(struct pipe_context
*ctx
, struct r600_transfer
*rtransfer
)
72 struct r600_common_context
*rctx
= (struct r600_common_context
*)ctx
;
73 struct pipe_transfer
*transfer
= (struct pipe_transfer
*)rtransfer
;
74 struct pipe_resource
*dst
= &rtransfer
->staging
->b
.b
;
75 struct pipe_resource
*src
= transfer
->resource
;
77 if (src
->nr_samples
> 1) {
78 r600_copy_region_with_blit(ctx
, dst
, 0, 0, 0, 0,
79 src
, transfer
->level
, &transfer
->box
);
83 rctx
->dma_copy(ctx
, dst
, 0, 0, 0, 0, src
, transfer
->level
,
87 /* Copy from a transfer's staging texture to a full GPU one. */
88 static void r600_copy_from_staging_texture(struct pipe_context
*ctx
, struct r600_transfer
*rtransfer
)
90 struct r600_common_context
*rctx
= (struct r600_common_context
*)ctx
;
91 struct pipe_transfer
*transfer
= (struct pipe_transfer
*)rtransfer
;
92 struct pipe_resource
*dst
= transfer
->resource
;
93 struct pipe_resource
*src
= &rtransfer
->staging
->b
.b
;
96 u_box_3d(0, 0, 0, transfer
->box
.width
, transfer
->box
.height
, transfer
->box
.depth
, &sbox
);
98 if (dst
->nr_samples
> 1) {
99 r600_copy_region_with_blit(ctx
, dst
, transfer
->level
,
100 transfer
->box
.x
, transfer
->box
.y
, transfer
->box
.z
,
105 rctx
->dma_copy(ctx
, dst
, transfer
->level
,
106 transfer
->box
.x
, transfer
->box
.y
, transfer
->box
.z
,
110 static unsigned r600_texture_get_offset(struct r600_texture
*rtex
, unsigned level
,
111 const struct pipe_box
*box
)
113 enum pipe_format format
= rtex
->resource
.b
.b
.format
;
115 return rtex
->surface
.level
[level
].offset
+
116 box
->z
* rtex
->surface
.level
[level
].slice_size
+
117 box
->y
/ util_format_get_blockheight(format
) * rtex
->surface
.level
[level
].pitch_bytes
+
118 box
->x
/ util_format_get_blockwidth(format
) * util_format_get_blocksize(format
);
121 static int r600_init_surface(struct r600_common_screen
*rscreen
,
122 struct radeon_surf
*surface
,
123 const struct pipe_resource
*ptex
,
125 bool is_flushed_depth
)
127 const struct util_format_description
*desc
=
128 util_format_description(ptex
->format
);
129 bool is_depth
, is_stencil
;
131 is_depth
= util_format_has_depth(desc
);
132 is_stencil
= util_format_has_stencil(desc
);
134 surface
->npix_x
= ptex
->width0
;
135 surface
->npix_y
= ptex
->height0
;
136 surface
->npix_z
= ptex
->depth0
;
137 surface
->blk_w
= util_format_get_blockwidth(ptex
->format
);
138 surface
->blk_h
= util_format_get_blockheight(ptex
->format
);
140 surface
->array_size
= 1;
141 surface
->last_level
= ptex
->last_level
;
143 if (rscreen
->chip_class
>= EVERGREEN
&& !is_flushed_depth
&&
144 ptex
->format
== PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
) {
145 surface
->bpe
= 4; /* stencil is allocated separately on evergreen */
147 surface
->bpe
= util_format_get_blocksize(ptex
->format
);
148 /* align byte per element on dword */
149 if (surface
->bpe
== 3) {
154 surface
->nsamples
= ptex
->nr_samples
? ptex
->nr_samples
: 1;
155 surface
->flags
= RADEON_SURF_SET(array_mode
, MODE
);
157 switch (ptex
->target
) {
158 case PIPE_TEXTURE_1D
:
159 surface
->flags
|= RADEON_SURF_SET(RADEON_SURF_TYPE_1D
, TYPE
);
161 case PIPE_TEXTURE_RECT
:
162 case PIPE_TEXTURE_2D
:
163 surface
->flags
|= RADEON_SURF_SET(RADEON_SURF_TYPE_2D
, TYPE
);
165 case PIPE_TEXTURE_3D
:
166 surface
->flags
|= RADEON_SURF_SET(RADEON_SURF_TYPE_3D
, TYPE
);
168 case PIPE_TEXTURE_1D_ARRAY
:
169 surface
->flags
|= RADEON_SURF_SET(RADEON_SURF_TYPE_1D_ARRAY
, TYPE
);
170 surface
->array_size
= ptex
->array_size
;
172 case PIPE_TEXTURE_2D_ARRAY
:
173 case PIPE_TEXTURE_CUBE_ARRAY
: /* cube array layout like 2d array */
174 surface
->flags
|= RADEON_SURF_SET(RADEON_SURF_TYPE_2D_ARRAY
, TYPE
);
175 surface
->array_size
= ptex
->array_size
;
177 case PIPE_TEXTURE_CUBE
:
178 surface
->flags
|= RADEON_SURF_SET(RADEON_SURF_TYPE_CUBEMAP
, TYPE
);
184 if (ptex
->bind
& PIPE_BIND_SCANOUT
) {
185 surface
->flags
|= RADEON_SURF_SCANOUT
;
188 if (!is_flushed_depth
&& is_depth
) {
189 surface
->flags
|= RADEON_SURF_ZBUFFER
;
192 surface
->flags
|= RADEON_SURF_SBUFFER
|
193 RADEON_SURF_HAS_SBUFFER_MIPTREE
;
196 if (rscreen
->chip_class
>= SI
) {
197 surface
->flags
|= RADEON_SURF_HAS_TILE_MODE_INDEX
;
202 static int r600_setup_surface(struct pipe_screen
*screen
,
203 struct r600_texture
*rtex
,
204 unsigned pitch_in_bytes_override
)
206 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)screen
;
209 r
= rscreen
->ws
->surface_init(rscreen
->ws
, &rtex
->surface
);
214 rtex
->size
= rtex
->surface
.bo_size
;
216 if (pitch_in_bytes_override
&& pitch_in_bytes_override
!= rtex
->surface
.level
[0].pitch_bytes
) {
217 /* old ddx on evergreen over estimate alignment for 1d, only 1 level
220 rtex
->surface
.level
[0].nblk_x
= pitch_in_bytes_override
/ rtex
->surface
.bpe
;
221 rtex
->surface
.level
[0].pitch_bytes
= pitch_in_bytes_override
;
222 rtex
->surface
.level
[0].slice_size
= pitch_in_bytes_override
* rtex
->surface
.level
[0].nblk_y
;
223 if (rtex
->surface
.flags
& RADEON_SURF_SBUFFER
) {
224 rtex
->surface
.stencil_offset
=
225 rtex
->surface
.stencil_level
[0].offset
= rtex
->surface
.level
[0].slice_size
;
231 static boolean
r600_texture_get_handle(struct pipe_screen
* screen
,
232 struct pipe_resource
*ptex
,
233 struct winsys_handle
*whandle
)
235 struct r600_texture
*rtex
= (struct r600_texture
*)ptex
;
236 struct r600_resource
*resource
= &rtex
->resource
;
237 struct radeon_surf
*surface
= &rtex
->surface
;
238 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)screen
;
240 rscreen
->ws
->buffer_set_tiling(resource
->buf
,
242 surface
->level
[0].mode
>= RADEON_SURF_MODE_1D
?
243 RADEON_LAYOUT_TILED
: RADEON_LAYOUT_LINEAR
,
244 surface
->level
[0].mode
>= RADEON_SURF_MODE_2D
?
245 RADEON_LAYOUT_TILED
: RADEON_LAYOUT_LINEAR
,
246 surface
->pipe_config
,
247 surface
->bankw
, surface
->bankh
,
249 surface
->stencil_tile_split
,
250 surface
->mtilea
, surface
->num_banks
,
251 surface
->level
[0].pitch_bytes
,
252 (surface
->flags
& RADEON_SURF_SCANOUT
) != 0);
254 return rscreen
->ws
->buffer_get_handle(resource
->buf
,
255 surface
->level
[0].pitch_bytes
, whandle
);
258 static void r600_texture_destroy(struct pipe_screen
*screen
,
259 struct pipe_resource
*ptex
)
261 struct r600_texture
*rtex
= (struct r600_texture
*)ptex
;
262 struct r600_resource
*resource
= &rtex
->resource
;
264 if (rtex
->flushed_depth_texture
)
265 pipe_resource_reference((struct pipe_resource
**)&rtex
->flushed_depth_texture
, NULL
);
267 pipe_resource_reference((struct pipe_resource
**)&rtex
->htile_buffer
, NULL
);
268 if (rtex
->cmask_buffer
!= &rtex
->resource
) {
269 pipe_resource_reference((struct pipe_resource
**)&rtex
->cmask_buffer
, NULL
);
271 pipe_resource_reference((struct pipe_resource
**)&rtex
->dcc_buffer
, NULL
);
272 pb_reference(&resource
->buf
, NULL
);
276 static const struct u_resource_vtbl r600_texture_vtbl
;
278 /* The number of samples can be specified independently of the texture. */
279 void r600_texture_get_fmask_info(struct r600_common_screen
*rscreen
,
280 struct r600_texture
*rtex
,
282 struct r600_fmask_info
*out
)
284 /* FMASK is allocated like an ordinary texture. */
285 struct radeon_surf fmask
= rtex
->surface
;
287 memset(out
, 0, sizeof(*out
));
289 fmask
.bo_alignment
= 0;
292 fmask
.flags
|= RADEON_SURF_FMASK
;
294 /* Force 2D tiling if it wasn't set. This may occur when creating
295 * FMASK for MSAA resolve on R6xx. On R6xx, the single-sample
296 * destination buffer must have an FMASK too. */
297 fmask
.flags
= RADEON_SURF_CLR(fmask
.flags
, MODE
);
298 fmask
.flags
|= RADEON_SURF_SET(RADEON_SURF_MODE_2D
, MODE
);
300 if (rscreen
->chip_class
>= SI
) {
301 fmask
.flags
|= RADEON_SURF_HAS_TILE_MODE_INDEX
;
304 switch (nr_samples
) {
308 if (rscreen
->chip_class
<= CAYMAN
) {
316 R600_ERR("Invalid sample count for FMASK allocation.\n");
320 /* Overallocate FMASK on R600-R700 to fix colorbuffer corruption.
321 * This can be fixed by writing a separate FMASK allocator specifically
322 * for R600-R700 asics. */
323 if (rscreen
->chip_class
<= R700
) {
327 if (rscreen
->ws
->surface_init(rscreen
->ws
, &fmask
)) {
328 R600_ERR("Got error in surface_init while allocating FMASK.\n");
332 assert(fmask
.level
[0].mode
== RADEON_SURF_MODE_2D
);
334 out
->slice_tile_max
= (fmask
.level
[0].nblk_x
* fmask
.level
[0].nblk_y
) / 64;
335 if (out
->slice_tile_max
)
336 out
->slice_tile_max
-= 1;
338 out
->tile_mode_index
= fmask
.tiling_index
[0];
339 out
->pitch
= fmask
.level
[0].nblk_x
;
340 out
->bank_height
= fmask
.bankh
;
341 out
->alignment
= MAX2(256, fmask
.bo_alignment
);
342 out
->size
= fmask
.bo_size
;
345 static void r600_texture_allocate_fmask(struct r600_common_screen
*rscreen
,
346 struct r600_texture
*rtex
)
348 r600_texture_get_fmask_info(rscreen
, rtex
,
349 rtex
->resource
.b
.b
.nr_samples
, &rtex
->fmask
);
351 rtex
->fmask
.offset
= align(rtex
->size
, rtex
->fmask
.alignment
);
352 rtex
->size
= rtex
->fmask
.offset
+ rtex
->fmask
.size
;
355 void r600_texture_get_cmask_info(struct r600_common_screen
*rscreen
,
356 struct r600_texture
*rtex
,
357 struct r600_cmask_info
*out
)
359 unsigned cmask_tile_width
= 8;
360 unsigned cmask_tile_height
= 8;
361 unsigned cmask_tile_elements
= cmask_tile_width
* cmask_tile_height
;
362 unsigned element_bits
= 4;
363 unsigned cmask_cache_bits
= 1024;
364 unsigned num_pipes
= rscreen
->tiling_info
.num_channels
;
365 unsigned pipe_interleave_bytes
= rscreen
->tiling_info
.group_bytes
;
367 unsigned elements_per_macro_tile
= (cmask_cache_bits
/ element_bits
) * num_pipes
;
368 unsigned pixels_per_macro_tile
= elements_per_macro_tile
* cmask_tile_elements
;
369 unsigned sqrt_pixels_per_macro_tile
= sqrt(pixels_per_macro_tile
);
370 unsigned macro_tile_width
= util_next_power_of_two(sqrt_pixels_per_macro_tile
);
371 unsigned macro_tile_height
= pixels_per_macro_tile
/ macro_tile_width
;
373 unsigned pitch_elements
= align(rtex
->surface
.npix_x
, macro_tile_width
);
374 unsigned height
= align(rtex
->surface
.npix_y
, macro_tile_height
);
376 unsigned base_align
= num_pipes
* pipe_interleave_bytes
;
377 unsigned slice_bytes
=
378 ((pitch_elements
* height
* element_bits
+ 7) / 8) / cmask_tile_elements
;
380 assert(macro_tile_width
% 128 == 0);
381 assert(macro_tile_height
% 128 == 0);
383 out
->slice_tile_max
= ((pitch_elements
* height
) / (128*128)) - 1;
384 out
->alignment
= MAX2(256, base_align
);
385 out
->size
= (util_max_layer(&rtex
->resource
.b
.b
, 0) + 1) *
386 align(slice_bytes
, base_align
);
389 static void si_texture_get_cmask_info(struct r600_common_screen
*rscreen
,
390 struct r600_texture
*rtex
,
391 struct r600_cmask_info
*out
)
393 unsigned pipe_interleave_bytes
= rscreen
->tiling_info
.group_bytes
;
394 unsigned num_pipes
= rscreen
->tiling_info
.num_channels
;
395 unsigned cl_width
, cl_height
;
410 case 16: /* Hawaii */
419 unsigned base_align
= num_pipes
* pipe_interleave_bytes
;
421 unsigned width
= align(rtex
->surface
.npix_x
, cl_width
*8);
422 unsigned height
= align(rtex
->surface
.npix_y
, cl_height
*8);
423 unsigned slice_elements
= (width
* height
) / (8*8);
425 /* Each element of CMASK is a nibble. */
426 unsigned slice_bytes
= slice_elements
/ 2;
428 out
->slice_tile_max
= (width
* height
) / (128*128);
429 if (out
->slice_tile_max
)
430 out
->slice_tile_max
-= 1;
432 out
->alignment
= MAX2(256, base_align
);
433 out
->size
= (util_max_layer(&rtex
->resource
.b
.b
, 0) + 1) *
434 align(slice_bytes
, base_align
);
437 static void r600_texture_allocate_cmask(struct r600_common_screen
*rscreen
,
438 struct r600_texture
*rtex
)
440 if (rscreen
->chip_class
>= SI
) {
441 si_texture_get_cmask_info(rscreen
, rtex
, &rtex
->cmask
);
443 r600_texture_get_cmask_info(rscreen
, rtex
, &rtex
->cmask
);
446 rtex
->cmask
.offset
= align(rtex
->size
, rtex
->cmask
.alignment
);
447 rtex
->size
= rtex
->cmask
.offset
+ rtex
->cmask
.size
;
449 if (rscreen
->chip_class
>= SI
)
450 rtex
->cb_color_info
|= SI_S_028C70_FAST_CLEAR(1);
452 rtex
->cb_color_info
|= EG_S_028C70_FAST_CLEAR(1);
455 static void r600_texture_alloc_cmask_separate(struct r600_common_screen
*rscreen
,
456 struct r600_texture
*rtex
)
458 if (rtex
->cmask_buffer
)
461 assert(rtex
->cmask
.size
== 0);
463 if (rscreen
->chip_class
>= SI
) {
464 si_texture_get_cmask_info(rscreen
, rtex
, &rtex
->cmask
);
466 r600_texture_get_cmask_info(rscreen
, rtex
, &rtex
->cmask
);
469 rtex
->cmask_buffer
= (struct r600_resource
*)
470 pipe_buffer_create(&rscreen
->b
, PIPE_BIND_CUSTOM
,
471 PIPE_USAGE_DEFAULT
, rtex
->cmask
.size
);
472 if (rtex
->cmask_buffer
== NULL
) {
473 rtex
->cmask
.size
= 0;
477 /* update colorbuffer state bits */
478 rtex
->cmask
.base_address_reg
= rtex
->cmask_buffer
->gpu_address
>> 8;
480 if (rscreen
->chip_class
>= SI
)
481 rtex
->cb_color_info
|= SI_S_028C70_FAST_CLEAR(1);
483 rtex
->cb_color_info
|= EG_S_028C70_FAST_CLEAR(1);
486 static void vi_texture_alloc_dcc_separate(struct r600_common_screen
*rscreen
,
487 struct r600_texture
*rtex
)
489 rtex
->dcc_buffer
= (struct r600_resource
*)
490 r600_aligned_buffer_create(&rscreen
->b
, PIPE_BIND_CUSTOM
,
491 PIPE_USAGE_DEFAULT
, rtex
->surface
.dcc_size
, rtex
->surface
.dcc_alignment
);
492 if (rtex
->dcc_buffer
== NULL
) {
496 r600_screen_clear_buffer(rscreen
, &rtex
->dcc_buffer
->b
.b
, 0, rtex
->surface
.dcc_size
,
499 rtex
->cb_color_info
|= VI_S_028C70_DCC_ENABLE(1);
502 static unsigned r600_texture_get_htile_size(struct r600_common_screen
*rscreen
,
503 struct r600_texture
*rtex
)
505 unsigned cl_width
, cl_height
, width
, height
;
506 unsigned slice_elements
, slice_bytes
, pipe_interleave_bytes
, base_align
;
507 unsigned num_pipes
= rscreen
->tiling_info
.num_channels
;
509 if (rscreen
->chip_class
<= EVERGREEN
&&
510 rscreen
->info
.drm_major
== 2 && rscreen
->info
.drm_minor
< 26)
513 /* HW bug on R6xx. */
514 if (rscreen
->chip_class
== R600
&&
515 (rtex
->surface
.level
[0].npix_x
> 7680 ||
516 rtex
->surface
.level
[0].npix_y
> 7680))
519 /* HTILE is broken with 1D tiling on old kernels and CIK. */
520 if (rscreen
->chip_class
>= CIK
&&
521 rtex
->surface
.level
[0].mode
== RADEON_SURF_MODE_1D
&&
522 rscreen
->info
.drm_major
== 2 && rscreen
->info
.drm_minor
< 38)
551 width
= align(rtex
->surface
.npix_x
, cl_width
* 8);
552 height
= align(rtex
->surface
.npix_y
, cl_height
* 8);
554 slice_elements
= (width
* height
) / (8 * 8);
555 slice_bytes
= slice_elements
* 4;
557 pipe_interleave_bytes
= rscreen
->tiling_info
.group_bytes
;
558 base_align
= num_pipes
* pipe_interleave_bytes
;
560 return (util_max_layer(&rtex
->resource
.b
.b
, 0) + 1) *
561 align(slice_bytes
, base_align
);
564 static void r600_texture_allocate_htile(struct r600_common_screen
*rscreen
,
565 struct r600_texture
*rtex
)
567 unsigned htile_size
= r600_texture_get_htile_size(rscreen
, rtex
);
572 rtex
->htile_buffer
= (struct r600_resource
*)
573 pipe_buffer_create(&rscreen
->b
, PIPE_BIND_CUSTOM
,
574 PIPE_USAGE_DEFAULT
, htile_size
);
575 if (rtex
->htile_buffer
== NULL
) {
576 /* this is not a fatal error as we can still keep rendering
577 * without htile buffer */
578 R600_ERR("Failed to create buffer object for htile buffer.\n");
580 r600_screen_clear_buffer(rscreen
, &rtex
->htile_buffer
->b
.b
, 0,
581 htile_size
, 0, true);
585 /* Common processing for r600_texture_create and r600_texture_from_handle */
586 static struct r600_texture
*
587 r600_texture_create_object(struct pipe_screen
*screen
,
588 const struct pipe_resource
*base
,
589 unsigned pitch_in_bytes_override
,
590 struct pb_buffer
*buf
,
591 struct radeon_surf
*surface
)
593 struct r600_texture
*rtex
;
594 struct r600_resource
*resource
;
595 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)screen
;
597 rtex
= CALLOC_STRUCT(r600_texture
);
601 resource
= &rtex
->resource
;
602 resource
->b
.b
= *base
;
603 resource
->b
.vtbl
= &r600_texture_vtbl
;
604 pipe_reference_init(&resource
->b
.b
.reference
, 1);
605 resource
->b
.b
.screen
= screen
;
606 rtex
->pitch_override
= pitch_in_bytes_override
;
608 /* don't include stencil-only formats which we don't support for rendering */
609 rtex
->is_depth
= util_format_has_depth(util_format_description(rtex
->resource
.b
.b
.format
));
611 rtex
->surface
= *surface
;
612 if (r600_setup_surface(screen
, rtex
, pitch_in_bytes_override
)) {
617 /* Tiled depth textures utilize the non-displayable tile order.
618 * This must be done after r600_setup_surface.
619 * Applies to R600-Cayman. */
620 rtex
->non_disp_tiling
= rtex
->is_depth
&& rtex
->surface
.level
[0].mode
>= RADEON_SURF_MODE_1D
;
622 if (rtex
->is_depth
) {
623 if (!(base
->flags
& (R600_RESOURCE_FLAG_TRANSFER
|
624 R600_RESOURCE_FLAG_FLUSHED_DEPTH
)) &&
625 !(rscreen
->debug_flags
& DBG_NO_HYPERZ
)) {
627 r600_texture_allocate_htile(rscreen
, rtex
);
630 if (base
->nr_samples
> 1) {
632 r600_texture_allocate_fmask(rscreen
, rtex
);
633 r600_texture_allocate_cmask(rscreen
, rtex
);
634 rtex
->cmask_buffer
= &rtex
->resource
;
636 if (!rtex
->fmask
.size
|| !rtex
->cmask
.size
) {
641 if (rtex
->surface
.dcc_enabled
) {
642 vi_texture_alloc_dcc_separate(rscreen
, rtex
);
646 /* Now create the backing buffer. */
648 if (!r600_init_resource(rscreen
, resource
, rtex
->size
,
649 rtex
->surface
.bo_alignment
, TRUE
)) {
655 resource
->cs_buf
= rscreen
->ws
->buffer_get_cs_handle(buf
);
656 resource
->gpu_address
= rscreen
->ws
->buffer_get_virtual_address(resource
->cs_buf
);
657 resource
->domains
= rscreen
->ws
->buffer_get_initial_domain(resource
->cs_buf
);
660 if (rtex
->cmask
.size
) {
661 /* Initialize the cmask to 0xCC (= compressed state). */
662 r600_screen_clear_buffer(rscreen
, &rtex
->cmask_buffer
->b
.b
,
663 rtex
->cmask
.offset
, rtex
->cmask
.size
,
667 /* Initialize the CMASK base register value. */
668 rtex
->cmask
.base_address_reg
=
669 (rtex
->resource
.gpu_address
+ rtex
->cmask
.offset
) >> 8;
671 if (rscreen
->debug_flags
& DBG_VM
) {
672 fprintf(stderr
, "VM start=0x%"PRIX64
" end=0x%"PRIX64
" | Texture %ix%ix%i, %i levels, %i samples, %s\n",
673 rtex
->resource
.gpu_address
,
674 rtex
->resource
.gpu_address
+ rtex
->resource
.buf
->size
,
675 base
->width0
, base
->height0
, util_max_layer(base
, 0)+1, base
->last_level
+1,
676 base
->nr_samples
? base
->nr_samples
: 1, util_format_short_name(base
->format
));
679 if (rscreen
->debug_flags
& DBG_TEX
||
680 (rtex
->resource
.b
.b
.last_level
> 0 && rscreen
->debug_flags
& DBG_TEXMIP
)) {
681 printf("Texture: npix_x=%u, npix_y=%u, npix_z=%u, blk_w=%u, "
682 "blk_h=%u, blk_d=%u, array_size=%u, last_level=%u, "
683 "bpe=%u, nsamples=%u, flags=0x%x, %s\n",
684 rtex
->surface
.npix_x
, rtex
->surface
.npix_y
,
685 rtex
->surface
.npix_z
, rtex
->surface
.blk_w
,
686 rtex
->surface
.blk_h
, rtex
->surface
.blk_d
,
687 rtex
->surface
.array_size
, rtex
->surface
.last_level
,
688 rtex
->surface
.bpe
, rtex
->surface
.nsamples
,
689 rtex
->surface
.flags
, util_format_short_name(base
->format
));
690 for (int i
= 0; i
<= rtex
->surface
.last_level
; i
++) {
691 printf(" L %i: offset=%"PRIu64
", slice_size=%"PRIu64
", npix_x=%u, "
692 "npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, "
693 "nblk_z=%u, pitch_bytes=%u, mode=%u\n",
694 i
, rtex
->surface
.level
[i
].offset
,
695 rtex
->surface
.level
[i
].slice_size
,
696 u_minify(rtex
->resource
.b
.b
.width0
, i
),
697 u_minify(rtex
->resource
.b
.b
.height0
, i
),
698 u_minify(rtex
->resource
.b
.b
.depth0
, i
),
699 rtex
->surface
.level
[i
].nblk_x
,
700 rtex
->surface
.level
[i
].nblk_y
,
701 rtex
->surface
.level
[i
].nblk_z
,
702 rtex
->surface
.level
[i
].pitch_bytes
,
703 rtex
->surface
.level
[i
].mode
);
705 if (rtex
->surface
.flags
& RADEON_SURF_SBUFFER
) {
706 for (int i
= 0; i
<= rtex
->surface
.last_level
; i
++) {
707 printf(" S %i: offset=%"PRIu64
", slice_size=%"PRIu64
", npix_x=%u, "
708 "npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, "
709 "nblk_z=%u, pitch_bytes=%u, mode=%u\n",
710 i
, rtex
->surface
.stencil_level
[i
].offset
,
711 rtex
->surface
.stencil_level
[i
].slice_size
,
712 u_minify(rtex
->resource
.b
.b
.width0
, i
),
713 u_minify(rtex
->resource
.b
.b
.height0
, i
),
714 u_minify(rtex
->resource
.b
.b
.depth0
, i
),
715 rtex
->surface
.stencil_level
[i
].nblk_x
,
716 rtex
->surface
.stencil_level
[i
].nblk_y
,
717 rtex
->surface
.stencil_level
[i
].nblk_z
,
718 rtex
->surface
.stencil_level
[i
].pitch_bytes
,
719 rtex
->surface
.stencil_level
[i
].mode
);
726 static unsigned r600_choose_tiling(struct r600_common_screen
*rscreen
,
727 const struct pipe_resource
*templ
)
729 const struct util_format_description
*desc
= util_format_description(templ
->format
);
730 bool force_tiling
= templ
->flags
& R600_RESOURCE_FLAG_FORCE_TILING
;
732 /* MSAA resources must be 2D tiled. */
733 if (templ
->nr_samples
> 1)
734 return RADEON_SURF_MODE_2D
;
736 /* Transfer resources should be linear. */
737 if (templ
->flags
& R600_RESOURCE_FLAG_TRANSFER
)
738 return RADEON_SURF_MODE_LINEAR_ALIGNED
;
740 /* r600g: force tiling on TEXTURE_2D and TEXTURE_3D compute resources. */
741 if (rscreen
->chip_class
>= R600
&& rscreen
->chip_class
<= CAYMAN
&&
742 (templ
->bind
& PIPE_BIND_COMPUTE_RESOURCE
) &&
743 (templ
->target
== PIPE_TEXTURE_2D
||
744 templ
->target
== PIPE_TEXTURE_3D
))
747 /* Handle common candidates for the linear mode.
748 * Compressed textures must always be tiled. */
749 if (!force_tiling
&& !util_format_is_compressed(templ
->format
)) {
750 /* Not everything can be linear, so we cannot enforce it
751 * for all textures. */
752 if ((rscreen
->debug_flags
& DBG_NO_TILING
) &&
753 (!util_format_is_depth_or_stencil(templ
->format
) ||
754 !(templ
->flags
& R600_RESOURCE_FLAG_FLUSHED_DEPTH
)))
755 return RADEON_SURF_MODE_LINEAR_ALIGNED
;
757 /* Tiling doesn't work with the 422 (SUBSAMPLED) formats on R600+. */
758 if (desc
->layout
== UTIL_FORMAT_LAYOUT_SUBSAMPLED
)
759 return RADEON_SURF_MODE_LINEAR_ALIGNED
;
761 /* Cursors are linear on SI.
762 * (XXX double-check, maybe also use RADEON_SURF_SCANOUT) */
763 if (rscreen
->chip_class
>= SI
&&
764 (templ
->bind
& PIPE_BIND_CURSOR
))
765 return RADEON_SURF_MODE_LINEAR_ALIGNED
;
767 if (templ
->bind
& PIPE_BIND_LINEAR
)
768 return RADEON_SURF_MODE_LINEAR_ALIGNED
;
770 /* Textures with a very small height are recommended to be linear. */
771 if (templ
->target
== PIPE_TEXTURE_1D
||
772 templ
->target
== PIPE_TEXTURE_1D_ARRAY
||
774 return RADEON_SURF_MODE_LINEAR_ALIGNED
;
776 /* Textures likely to be mapped often. */
777 if (templ
->usage
== PIPE_USAGE_STAGING
||
778 templ
->usage
== PIPE_USAGE_STREAM
)
779 return RADEON_SURF_MODE_LINEAR_ALIGNED
;
782 /* Make small textures 1D tiled. */
783 if (templ
->width0
<= 16 || templ
->height0
<= 16 ||
784 (rscreen
->debug_flags
& DBG_NO_2D_TILING
))
785 return RADEON_SURF_MODE_1D
;
787 /* The allocator will switch to 1D if needed. */
788 return RADEON_SURF_MODE_2D
;
791 struct pipe_resource
*r600_texture_create(struct pipe_screen
*screen
,
792 const struct pipe_resource
*templ
)
794 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)screen
;
795 struct radeon_surf surface
= {0};
798 r
= r600_init_surface(rscreen
, &surface
, templ
,
799 r600_choose_tiling(rscreen
, templ
),
800 templ
->flags
& R600_RESOURCE_FLAG_FLUSHED_DEPTH
);
804 r
= rscreen
->ws
->surface_best(rscreen
->ws
, &surface
);
808 return (struct pipe_resource
*)r600_texture_create_object(screen
, templ
,
812 static struct pipe_resource
*r600_texture_from_handle(struct pipe_screen
*screen
,
813 const struct pipe_resource
*templ
,
814 struct winsys_handle
*whandle
)
816 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)screen
;
817 struct pb_buffer
*buf
= NULL
;
820 enum radeon_bo_layout micro
, macro
;
821 struct radeon_surf surface
;
825 /* Support only 2D textures without mipmaps */
826 if ((templ
->target
!= PIPE_TEXTURE_2D
&& templ
->target
!= PIPE_TEXTURE_RECT
) ||
827 templ
->depth0
!= 1 || templ
->last_level
!= 0)
830 buf
= rscreen
->ws
->buffer_from_handle(rscreen
->ws
, whandle
, &stride
);
834 rscreen
->ws
->buffer_get_tiling(buf
, µ
, ¯o
,
835 &surface
.bankw
, &surface
.bankh
,
837 &surface
.stencil_tile_split
,
838 &surface
.mtilea
, &scanout
);
840 if (macro
== RADEON_LAYOUT_TILED
)
841 array_mode
= RADEON_SURF_MODE_2D
;
842 else if (micro
== RADEON_LAYOUT_TILED
)
843 array_mode
= RADEON_SURF_MODE_1D
;
845 array_mode
= RADEON_SURF_MODE_LINEAR_ALIGNED
;
847 r
= r600_init_surface(rscreen
, &surface
, templ
, array_mode
, false);
853 surface
.flags
|= RADEON_SURF_SCANOUT
;
855 return (struct pipe_resource
*)r600_texture_create_object(screen
, templ
,
856 stride
, buf
, &surface
);
859 bool r600_init_flushed_depth_texture(struct pipe_context
*ctx
,
860 struct pipe_resource
*texture
,
861 struct r600_texture
**staging
)
863 struct r600_texture
*rtex
= (struct r600_texture
*)texture
;
864 struct pipe_resource resource
;
865 struct r600_texture
**flushed_depth_texture
= staging
?
866 staging
: &rtex
->flushed_depth_texture
;
868 if (!staging
&& rtex
->flushed_depth_texture
)
869 return true; /* it's ready */
871 resource
.target
= texture
->target
;
872 resource
.format
= texture
->format
;
873 resource
.width0
= texture
->width0
;
874 resource
.height0
= texture
->height0
;
875 resource
.depth0
= texture
->depth0
;
876 resource
.array_size
= texture
->array_size
;
877 resource
.last_level
= texture
->last_level
;
878 resource
.nr_samples
= texture
->nr_samples
;
879 resource
.usage
= staging
? PIPE_USAGE_STAGING
: PIPE_USAGE_DEFAULT
;
880 resource
.bind
= texture
->bind
& ~PIPE_BIND_DEPTH_STENCIL
;
881 resource
.flags
= texture
->flags
| R600_RESOURCE_FLAG_FLUSHED_DEPTH
;
884 resource
.flags
|= R600_RESOURCE_FLAG_TRANSFER
;
886 *flushed_depth_texture
= (struct r600_texture
*)ctx
->screen
->resource_create(ctx
->screen
, &resource
);
887 if (*flushed_depth_texture
== NULL
) {
888 R600_ERR("failed to create temporary texture to hold flushed depth\n");
892 (*flushed_depth_texture
)->is_flushing_texture
= TRUE
;
893 (*flushed_depth_texture
)->non_disp_tiling
= false;
898 * Initialize the pipe_resource descriptor to be of the same size as the box,
899 * which is supposed to hold a subregion of the texture "orig" at the given
902 static void r600_init_temp_resource_from_box(struct pipe_resource
*res
,
903 struct pipe_resource
*orig
,
904 const struct pipe_box
*box
,
905 unsigned level
, unsigned flags
)
907 memset(res
, 0, sizeof(*res
));
908 res
->format
= orig
->format
;
909 res
->width0
= box
->width
;
910 res
->height0
= box
->height
;
913 res
->usage
= flags
& R600_RESOURCE_FLAG_TRANSFER
? PIPE_USAGE_STAGING
: PIPE_USAGE_DEFAULT
;
916 /* We must set the correct texture target and dimensions for a 3D box. */
917 if (box
->depth
> 1 && util_max_layer(orig
, level
) > 0)
918 res
->target
= orig
->target
;
920 res
->target
= PIPE_TEXTURE_2D
;
922 switch (res
->target
) {
923 case PIPE_TEXTURE_1D_ARRAY
:
924 case PIPE_TEXTURE_2D_ARRAY
:
925 case PIPE_TEXTURE_CUBE_ARRAY
:
926 res
->array_size
= box
->depth
;
928 case PIPE_TEXTURE_3D
:
929 res
->depth0
= box
->depth
;
935 static void *r600_texture_transfer_map(struct pipe_context
*ctx
,
936 struct pipe_resource
*texture
,
939 const struct pipe_box
*box
,
940 struct pipe_transfer
**ptransfer
)
942 struct r600_common_context
*rctx
= (struct r600_common_context
*)ctx
;
943 struct r600_texture
*rtex
= (struct r600_texture
*)texture
;
944 struct r600_transfer
*trans
;
945 boolean use_staging_texture
= FALSE
;
946 struct r600_resource
*buf
;
950 /* We cannot map a tiled texture directly because the data is
951 * in a different order, therefore we do detiling using a blit.
953 * Also, use a temporary in GTT memory for read transfers, as
954 * the CPU is much happier reading out of cached system memory
955 * than uncached VRAM.
957 if (rtex
->surface
.level
[0].mode
>= RADEON_SURF_MODE_1D
) {
958 use_staging_texture
= TRUE
;
959 } else if ((usage
& PIPE_TRANSFER_READ
) && !(usage
& PIPE_TRANSFER_MAP_DIRECTLY
) &&
960 (rtex
->resource
.domains
== RADEON_DOMAIN_VRAM
)) {
961 /* Untiled buffers in VRAM, which is slow for CPU reads */
962 use_staging_texture
= TRUE
;
963 } else if (!(usage
& PIPE_TRANSFER_READ
) &&
964 (r600_rings_is_buffer_referenced(rctx
, rtex
->resource
.cs_buf
, RADEON_USAGE_READWRITE
) ||
965 !rctx
->ws
->buffer_wait(rtex
->resource
.buf
, 0, RADEON_USAGE_READWRITE
))) {
966 /* Use a staging texture for uploads if the underlying BO is busy. */
967 use_staging_texture
= TRUE
;
970 if (texture
->flags
& R600_RESOURCE_FLAG_TRANSFER
) {
971 use_staging_texture
= FALSE
;
974 if (use_staging_texture
&& (usage
& PIPE_TRANSFER_MAP_DIRECTLY
)) {
978 trans
= CALLOC_STRUCT(r600_transfer
);
981 trans
->transfer
.resource
= texture
;
982 trans
->transfer
.level
= level
;
983 trans
->transfer
.usage
= usage
;
984 trans
->transfer
.box
= *box
;
986 if (rtex
->is_depth
) {
987 struct r600_texture
*staging_depth
;
989 if (rtex
->resource
.b
.b
.nr_samples
> 1) {
990 /* MSAA depth buffers need to be converted to single sample buffers.
992 * Mapping MSAA depth buffers can occur if ReadPixels is called
993 * with a multisample GLX visual.
995 * First downsample the depth buffer to a temporary texture,
996 * then decompress the temporary one to staging.
998 * Only the region being mapped is transfered.
1000 struct pipe_resource resource
;
1002 r600_init_temp_resource_from_box(&resource
, texture
, box
, level
, 0);
1004 if (!r600_init_flushed_depth_texture(ctx
, &resource
, &staging_depth
)) {
1005 R600_ERR("failed to create temporary texture to hold untiled copy\n");
1010 if (usage
& PIPE_TRANSFER_READ
) {
1011 struct pipe_resource
*temp
= ctx
->screen
->resource_create(ctx
->screen
, &resource
);
1013 R600_ERR("failed to create a temporary depth texture\n");
1018 r600_copy_region_with_blit(ctx
, temp
, 0, 0, 0, 0, texture
, level
, box
);
1019 rctx
->blit_decompress_depth(ctx
, (struct r600_texture
*)temp
, staging_depth
,
1020 0, 0, 0, box
->depth
, 0, 0);
1021 pipe_resource_reference((struct pipe_resource
**)&temp
, NULL
);
1025 /* XXX: only readback the rectangle which is being mapped? */
1026 /* XXX: when discard is true, no need to read back from depth texture */
1027 if (!r600_init_flushed_depth_texture(ctx
, texture
, &staging_depth
)) {
1028 R600_ERR("failed to create temporary texture to hold untiled copy\n");
1033 rctx
->blit_decompress_depth(ctx
, rtex
, staging_depth
,
1035 box
->z
, box
->z
+ box
->depth
- 1,
1038 offset
= r600_texture_get_offset(staging_depth
, level
, box
);
1041 trans
->transfer
.stride
= staging_depth
->surface
.level
[level
].pitch_bytes
;
1042 trans
->transfer
.layer_stride
= staging_depth
->surface
.level
[level
].slice_size
;
1043 trans
->staging
= (struct r600_resource
*)staging_depth
;
1044 } else if (use_staging_texture
) {
1045 struct pipe_resource resource
;
1046 struct r600_texture
*staging
;
1048 r600_init_temp_resource_from_box(&resource
, texture
, box
, level
,
1049 R600_RESOURCE_FLAG_TRANSFER
);
1050 resource
.usage
= (usage
& PIPE_TRANSFER_READ
) ?
1051 PIPE_USAGE_STAGING
: PIPE_USAGE_STREAM
;
1053 /* Create the temporary texture. */
1054 staging
= (struct r600_texture
*)ctx
->screen
->resource_create(ctx
->screen
, &resource
);
1055 if (staging
== NULL
) {
1056 R600_ERR("failed to create temporary texture to hold untiled copy\n");
1060 trans
->staging
= &staging
->resource
;
1061 trans
->transfer
.stride
= staging
->surface
.level
[0].pitch_bytes
;
1062 trans
->transfer
.layer_stride
= staging
->surface
.level
[0].slice_size
;
1063 if (usage
& PIPE_TRANSFER_READ
) {
1064 r600_copy_to_staging_texture(ctx
, trans
);
1067 /* the resource is mapped directly */
1068 trans
->transfer
.stride
= rtex
->surface
.level
[level
].pitch_bytes
;
1069 trans
->transfer
.layer_stride
= rtex
->surface
.level
[level
].slice_size
;
1070 offset
= r600_texture_get_offset(rtex
, level
, box
);
1073 if (trans
->staging
) {
1074 buf
= trans
->staging
;
1075 if (!rtex
->is_depth
&& !(usage
& PIPE_TRANSFER_READ
))
1076 usage
|= PIPE_TRANSFER_UNSYNCHRONIZED
;
1078 buf
= &rtex
->resource
;
1081 if (!(map
= r600_buffer_map_sync_with_rings(rctx
, buf
, usage
))) {
1082 pipe_resource_reference((struct pipe_resource
**)&trans
->staging
, NULL
);
1087 *ptransfer
= &trans
->transfer
;
1088 return map
+ offset
;
1091 static void r600_texture_transfer_unmap(struct pipe_context
*ctx
,
1092 struct pipe_transfer
* transfer
)
1094 struct r600_transfer
*rtransfer
= (struct r600_transfer
*)transfer
;
1095 struct pipe_resource
*texture
= transfer
->resource
;
1096 struct r600_texture
*rtex
= (struct r600_texture
*)texture
;
1098 if ((transfer
->usage
& PIPE_TRANSFER_WRITE
) && rtransfer
->staging
) {
1099 if (rtex
->is_depth
&& rtex
->resource
.b
.b
.nr_samples
<= 1) {
1100 ctx
->resource_copy_region(ctx
, texture
, transfer
->level
,
1101 transfer
->box
.x
, transfer
->box
.y
, transfer
->box
.z
,
1102 &rtransfer
->staging
->b
.b
, transfer
->level
,
1105 r600_copy_from_staging_texture(ctx
, rtransfer
);
1109 if (rtransfer
->staging
)
1110 pipe_resource_reference((struct pipe_resource
**)&rtransfer
->staging
, NULL
);
1115 static const struct u_resource_vtbl r600_texture_vtbl
=
1117 NULL
, /* get_handle */
1118 r600_texture_destroy
, /* resource_destroy */
1119 r600_texture_transfer_map
, /* transfer_map */
1120 u_default_transfer_flush_region
, /* transfer_flush_region */
1121 r600_texture_transfer_unmap
, /* transfer_unmap */
1122 NULL
/* transfer_inline_write */
1125 struct pipe_surface
*r600_create_surface_custom(struct pipe_context
*pipe
,
1126 struct pipe_resource
*texture
,
1127 const struct pipe_surface
*templ
,
1128 unsigned width
, unsigned height
)
1130 struct r600_surface
*surface
= CALLOC_STRUCT(r600_surface
);
1132 if (surface
== NULL
)
1135 assert(templ
->u
.tex
.first_layer
<= util_max_layer(texture
, templ
->u
.tex
.level
));
1136 assert(templ
->u
.tex
.last_layer
<= util_max_layer(texture
, templ
->u
.tex
.level
));
1138 pipe_reference_init(&surface
->base
.reference
, 1);
1139 pipe_resource_reference(&surface
->base
.texture
, texture
);
1140 surface
->base
.context
= pipe
;
1141 surface
->base
.format
= templ
->format
;
1142 surface
->base
.width
= width
;
1143 surface
->base
.height
= height
;
1144 surface
->base
.u
= templ
->u
;
1145 return &surface
->base
;
1148 static struct pipe_surface
*r600_create_surface(struct pipe_context
*pipe
,
1149 struct pipe_resource
*tex
,
1150 const struct pipe_surface
*templ
)
1152 unsigned level
= templ
->u
.tex
.level
;
1154 return r600_create_surface_custom(pipe
, tex
, templ
,
1155 u_minify(tex
->width0
, level
),
1156 u_minify(tex
->height0
, level
));
1159 static void r600_surface_destroy(struct pipe_context
*pipe
,
1160 struct pipe_surface
*surface
)
1162 struct r600_surface
*surf
= (struct r600_surface
*)surface
;
1163 pipe_resource_reference((struct pipe_resource
**)&surf
->cb_buffer_fmask
, NULL
);
1164 pipe_resource_reference((struct pipe_resource
**)&surf
->cb_buffer_cmask
, NULL
);
1165 pipe_resource_reference(&surface
->texture
, NULL
);
1169 unsigned r600_translate_colorswap(enum pipe_format format
)
1171 const struct util_format_description
*desc
= util_format_description(format
);
1173 #define HAS_SWIZZLE(chan,swz) (desc->swizzle[chan] == UTIL_FORMAT_SWIZZLE_##swz)
1175 if (format
== PIPE_FORMAT_R11G11B10_FLOAT
) /* isn't plain */
1176 return V_0280A0_SWAP_STD
;
1178 if (desc
->layout
!= UTIL_FORMAT_LAYOUT_PLAIN
)
1181 switch (desc
->nr_channels
) {
1183 if (HAS_SWIZZLE(0,X
))
1184 return V_0280A0_SWAP_STD
; /* X___ */
1185 else if (HAS_SWIZZLE(3,X
))
1186 return V_0280A0_SWAP_ALT_REV
; /* ___X */
1189 if ((HAS_SWIZZLE(0,X
) && HAS_SWIZZLE(1,Y
)) ||
1190 (HAS_SWIZZLE(0,X
) && HAS_SWIZZLE(1,NONE
)) ||
1191 (HAS_SWIZZLE(0,NONE
) && HAS_SWIZZLE(1,Y
)))
1192 return V_0280A0_SWAP_STD
; /* XY__ */
1193 else if ((HAS_SWIZZLE(0,Y
) && HAS_SWIZZLE(1,X
)) ||
1194 (HAS_SWIZZLE(0,Y
) && HAS_SWIZZLE(1,NONE
)) ||
1195 (HAS_SWIZZLE(0,NONE
) && HAS_SWIZZLE(1,X
)))
1196 return V_0280A0_SWAP_STD_REV
; /* YX__ */
1197 else if (HAS_SWIZZLE(0,X
) && HAS_SWIZZLE(3,Y
))
1198 return V_0280A0_SWAP_ALT
; /* X__Y */
1199 else if (HAS_SWIZZLE(0,Y
) && HAS_SWIZZLE(3,X
))
1200 return V_0280A0_SWAP_ALT_REV
; /* Y__X */
1203 if (HAS_SWIZZLE(0,X
))
1204 return V_0280A0_SWAP_STD
; /* XYZ */
1205 else if (HAS_SWIZZLE(0,Z
))
1206 return V_0280A0_SWAP_STD_REV
; /* ZYX */
1209 /* check the middle channels, the 1st and 4th channel can be NONE */
1210 if (HAS_SWIZZLE(1,Y
) && HAS_SWIZZLE(2,Z
))
1211 return V_0280A0_SWAP_STD
; /* XYZW */
1212 else if (HAS_SWIZZLE(1,Z
) && HAS_SWIZZLE(2,Y
))
1213 return V_0280A0_SWAP_STD_REV
; /* WZYX */
1214 else if (HAS_SWIZZLE(1,Y
) && HAS_SWIZZLE(2,X
))
1215 return V_0280A0_SWAP_ALT
; /* ZYXW */
1216 else if (HAS_SWIZZLE(1,X
) && HAS_SWIZZLE(2,Y
))
1217 return V_0280A0_SWAP_ALT_REV
; /* WXYZ */
1223 static void evergreen_set_clear_color(struct r600_texture
*rtex
,
1224 enum pipe_format surface_format
,
1225 const union pipe_color_union
*color
)
1227 union util_color uc
;
1229 memset(&uc
, 0, sizeof(uc
));
1231 if (util_format_is_pure_uint(surface_format
)) {
1232 util_format_write_4ui(surface_format
, color
->ui
, 0, &uc
, 0, 0, 0, 1, 1);
1233 } else if (util_format_is_pure_sint(surface_format
)) {
1234 util_format_write_4i(surface_format
, color
->i
, 0, &uc
, 0, 0, 0, 1, 1);
1236 util_pack_color(color
->f
, surface_format
, &uc
);
1239 memcpy(rtex
->color_clear_value
, &uc
, 2 * sizeof(uint32_t));
1242 static void vi_get_fast_clear_parameters(enum pipe_format surface_format
,
1243 const union pipe_color_union
*color
,
1244 uint32_t* reset_value
,
1245 bool* clear_words_needed
)
1247 bool values
[4] = {};
1249 bool main_value
= false;
1250 bool extra_value
= false;
1252 const struct util_format_description
*desc
= util_format_description(surface_format
);
1254 *clear_words_needed
= true;
1255 *reset_value
= 0x20202020U
;
1257 /* If we want to clear without needing a fast clear eliminate step, we
1258 * can set each channel to 0 or 1 (or 0/max for integer formats). We
1259 * have two sets of flags, one for the last or first channel(extra) and
1260 * one for the other channels(main).
1263 if (surface_format
== PIPE_FORMAT_R11G11B10_FLOAT
||
1264 surface_format
== PIPE_FORMAT_B5G6R5_UNORM
||
1265 surface_format
== PIPE_FORMAT_B5G6R5_SRGB
) {
1267 } else if (desc
->layout
== UTIL_FORMAT_LAYOUT_PLAIN
) {
1268 if(r600_translate_colorswap(surface_format
) <= 1)
1269 extra_channel
= desc
->nr_channels
- 1;
1275 for (i
= 0; i
< 4; ++i
) {
1276 int index
= desc
->swizzle
[i
] - UTIL_FORMAT_SWIZZLE_X
;
1278 if (desc
->swizzle
[i
] < UTIL_FORMAT_SWIZZLE_X
||
1279 desc
->swizzle
[i
] > UTIL_FORMAT_SWIZZLE_W
)
1282 if (util_format_is_pure_sint(surface_format
)) {
1283 values
[i
] = color
->i
[i
] != 0;
1284 if (color
->i
[i
] != 0 && color
->i
[i
] != INT32_MAX
)
1286 } else if (util_format_is_pure_uint(surface_format
)) {
1287 values
[i
] = color
->ui
[i
] != 0U;
1288 if (color
->ui
[i
] != 0U && color
->ui
[i
] != UINT32_MAX
)
1291 values
[i
] = color
->f
[i
] != 0.0F
;
1292 if (color
->f
[i
] != 0.0F
&& color
->f
[i
] != 1.0F
)
1296 if (index
== extra_channel
)
1297 extra_value
= values
[i
];
1299 main_value
= values
[i
];
1302 for (int i
= 0; i
< 4; ++i
)
1303 if (values
[i
] != main_value
&&
1304 desc
->swizzle
[i
] - UTIL_FORMAT_SWIZZLE_X
!= extra_channel
&&
1305 desc
->swizzle
[i
] >= UTIL_FORMAT_SWIZZLE_X
&&
1306 desc
->swizzle
[i
] <= UTIL_FORMAT_SWIZZLE_W
)
1309 *clear_words_needed
= false;
1311 *reset_value
|= 0x80808080U
;
1314 *reset_value
|= 0x40404040U
;
1317 void evergreen_do_fast_color_clear(struct r600_common_context
*rctx
,
1318 struct pipe_framebuffer_state
*fb
,
1319 struct r600_atom
*fb_state
,
1320 unsigned *buffers
, unsigned *dirty_cbufs
,
1321 const union pipe_color_union
*color
)
1325 if (rctx
->current_render_cond
)
1328 for (i
= 0; i
< fb
->nr_cbufs
; i
++) {
1329 struct r600_texture
*tex
;
1330 unsigned clear_bit
= PIPE_CLEAR_COLOR0
<< i
;
1335 /* if this colorbuffer is not being cleared */
1336 if (!(*buffers
& clear_bit
))
1339 tex
= (struct r600_texture
*)fb
->cbufs
[i
]->texture
;
1341 /* 128-bit formats are unusupported */
1342 if (util_format_get_blocksizebits(fb
->cbufs
[i
]->format
) > 64) {
1346 /* the clear is allowed if all layers are bound */
1347 if (fb
->cbufs
[i
]->u
.tex
.first_layer
!= 0 ||
1348 fb
->cbufs
[i
]->u
.tex
.last_layer
!= util_max_layer(&tex
->resource
.b
.b
, 0)) {
1352 /* cannot clear mipmapped textures */
1353 if (fb
->cbufs
[i
]->texture
->last_level
!= 0) {
1357 /* only supported on tiled surfaces */
1358 if (tex
->surface
.level
[0].mode
< RADEON_SURF_MODE_1D
) {
1362 /* fast color clear with 1D tiling doesn't work on old kernels and CIK */
1363 if (tex
->surface
.level
[0].mode
== RADEON_SURF_MODE_1D
&&
1364 rctx
->chip_class
>= CIK
&&
1365 rctx
->screen
->info
.drm_major
== 2 &&
1366 rctx
->screen
->info
.drm_minor
< 38) {
1370 if (tex
->dcc_buffer
) {
1371 uint32_t reset_value
;
1372 bool clear_words_needed
;
1374 vi_get_fast_clear_parameters(fb
->cbufs
[i
]->format
, color
, &reset_value
, &clear_words_needed
);
1376 rctx
->clear_buffer(&rctx
->b
, &tex
->dcc_buffer
->b
.b
,
1377 0, tex
->surface
.dcc_size
, reset_value
, true);
1379 if (clear_words_needed
)
1380 tex
->dirty_level_mask
|= 1 << fb
->cbufs
[i
]->u
.tex
.level
;
1382 /* ensure CMASK is enabled */
1383 r600_texture_alloc_cmask_separate(rctx
->screen
, tex
);
1384 if (tex
->cmask
.size
== 0) {
1388 /* Do the fast clear. */
1389 rctx
->clear_buffer(&rctx
->b
, &tex
->cmask_buffer
->b
.b
,
1390 tex
->cmask
.offset
, tex
->cmask
.size
, 0, true);
1392 tex
->dirty_level_mask
|= 1 << fb
->cbufs
[i
]->u
.tex
.level
;
1395 evergreen_set_clear_color(tex
, fb
->cbufs
[i
]->format
, color
);
1398 *dirty_cbufs
|= 1 << i
;
1399 rctx
->set_atom_dirty(rctx
, fb_state
, true);
1400 *buffers
&= ~clear_bit
;
1404 void r600_init_screen_texture_functions(struct r600_common_screen
*rscreen
)
1406 rscreen
->b
.resource_from_handle
= r600_texture_from_handle
;
1407 rscreen
->b
.resource_get_handle
= r600_texture_get_handle
;
1410 void r600_init_context_texture_functions(struct r600_common_context
*rctx
)
1412 rctx
->b
.create_surface
= r600_create_surface
;
1413 rctx
->b
.surface_destroy
= r600_surface_destroy
;