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 if (rscreen
->debug_flags
& DBG_NO_DCC
)
492 rtex
->dcc_buffer
= (struct r600_resource
*)
493 r600_aligned_buffer_create(&rscreen
->b
, PIPE_BIND_CUSTOM
,
494 PIPE_USAGE_DEFAULT
, rtex
->surface
.dcc_size
, rtex
->surface
.dcc_alignment
);
495 if (rtex
->dcc_buffer
== NULL
) {
499 r600_screen_clear_buffer(rscreen
, &rtex
->dcc_buffer
->b
.b
, 0, rtex
->surface
.dcc_size
,
502 rtex
->cb_color_info
|= VI_S_028C70_DCC_ENABLE(1);
505 static unsigned r600_texture_get_htile_size(struct r600_common_screen
*rscreen
,
506 struct r600_texture
*rtex
)
508 unsigned cl_width
, cl_height
, width
, height
;
509 unsigned slice_elements
, slice_bytes
, pipe_interleave_bytes
, base_align
;
510 unsigned num_pipes
= rscreen
->tiling_info
.num_channels
;
512 if (rscreen
->chip_class
<= EVERGREEN
&&
513 rscreen
->info
.drm_major
== 2 && rscreen
->info
.drm_minor
< 26)
516 /* HW bug on R6xx. */
517 if (rscreen
->chip_class
== R600
&&
518 (rtex
->surface
.level
[0].npix_x
> 7680 ||
519 rtex
->surface
.level
[0].npix_y
> 7680))
522 /* HTILE is broken with 1D tiling on old kernels and CIK. */
523 if (rscreen
->chip_class
>= CIK
&&
524 rtex
->surface
.level
[0].mode
== RADEON_SURF_MODE_1D
&&
525 rscreen
->info
.drm_major
== 2 && rscreen
->info
.drm_minor
< 38)
554 width
= align(rtex
->surface
.npix_x
, cl_width
* 8);
555 height
= align(rtex
->surface
.npix_y
, cl_height
* 8);
557 slice_elements
= (width
* height
) / (8 * 8);
558 slice_bytes
= slice_elements
* 4;
560 pipe_interleave_bytes
= rscreen
->tiling_info
.group_bytes
;
561 base_align
= num_pipes
* pipe_interleave_bytes
;
563 return (util_max_layer(&rtex
->resource
.b
.b
, 0) + 1) *
564 align(slice_bytes
, base_align
);
567 static void r600_texture_allocate_htile(struct r600_common_screen
*rscreen
,
568 struct r600_texture
*rtex
)
570 unsigned htile_size
= r600_texture_get_htile_size(rscreen
, rtex
);
575 rtex
->htile_buffer
= (struct r600_resource
*)
576 pipe_buffer_create(&rscreen
->b
, PIPE_BIND_CUSTOM
,
577 PIPE_USAGE_DEFAULT
, htile_size
);
578 if (rtex
->htile_buffer
== NULL
) {
579 /* this is not a fatal error as we can still keep rendering
580 * without htile buffer */
581 R600_ERR("Failed to create buffer object for htile buffer.\n");
583 r600_screen_clear_buffer(rscreen
, &rtex
->htile_buffer
->b
.b
, 0,
584 htile_size
, 0, true);
588 /* Common processing for r600_texture_create and r600_texture_from_handle */
589 static struct r600_texture
*
590 r600_texture_create_object(struct pipe_screen
*screen
,
591 const struct pipe_resource
*base
,
592 unsigned pitch_in_bytes_override
,
593 struct pb_buffer
*buf
,
594 struct radeon_surf
*surface
)
596 struct r600_texture
*rtex
;
597 struct r600_resource
*resource
;
598 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)screen
;
600 rtex
= CALLOC_STRUCT(r600_texture
);
604 resource
= &rtex
->resource
;
605 resource
->b
.b
= *base
;
606 resource
->b
.vtbl
= &r600_texture_vtbl
;
607 pipe_reference_init(&resource
->b
.b
.reference
, 1);
608 resource
->b
.b
.screen
= screen
;
609 rtex
->pitch_override
= pitch_in_bytes_override
;
611 /* don't include stencil-only formats which we don't support for rendering */
612 rtex
->is_depth
= util_format_has_depth(util_format_description(rtex
->resource
.b
.b
.format
));
614 rtex
->surface
= *surface
;
615 if (r600_setup_surface(screen
, rtex
, pitch_in_bytes_override
)) {
620 /* Tiled depth textures utilize the non-displayable tile order.
621 * This must be done after r600_setup_surface.
622 * Applies to R600-Cayman. */
623 rtex
->non_disp_tiling
= rtex
->is_depth
&& rtex
->surface
.level
[0].mode
>= RADEON_SURF_MODE_1D
;
625 if (rtex
->is_depth
) {
626 if (!(base
->flags
& (R600_RESOURCE_FLAG_TRANSFER
|
627 R600_RESOURCE_FLAG_FLUSHED_DEPTH
)) &&
628 !(rscreen
->debug_flags
& DBG_NO_HYPERZ
)) {
630 r600_texture_allocate_htile(rscreen
, rtex
);
633 if (base
->nr_samples
> 1) {
635 r600_texture_allocate_fmask(rscreen
, rtex
);
636 r600_texture_allocate_cmask(rscreen
, rtex
);
637 rtex
->cmask_buffer
= &rtex
->resource
;
639 if (!rtex
->fmask
.size
|| !rtex
->cmask
.size
) {
644 if (rtex
->surface
.dcc_size
)
645 vi_texture_alloc_dcc_separate(rscreen
, rtex
);
648 /* Now create the backing buffer. */
650 if (!r600_init_resource(rscreen
, resource
, rtex
->size
,
651 rtex
->surface
.bo_alignment
, TRUE
)) {
657 resource
->cs_buf
= rscreen
->ws
->buffer_get_cs_handle(buf
);
658 resource
->gpu_address
= rscreen
->ws
->buffer_get_virtual_address(resource
->cs_buf
);
659 resource
->domains
= rscreen
->ws
->buffer_get_initial_domain(resource
->cs_buf
);
662 if (rtex
->cmask
.size
) {
663 /* Initialize the cmask to 0xCC (= compressed state). */
664 r600_screen_clear_buffer(rscreen
, &rtex
->cmask_buffer
->b
.b
,
665 rtex
->cmask
.offset
, rtex
->cmask
.size
,
669 /* Initialize the CMASK base register value. */
670 rtex
->cmask
.base_address_reg
=
671 (rtex
->resource
.gpu_address
+ rtex
->cmask
.offset
) >> 8;
673 if (rscreen
->debug_flags
& DBG_VM
) {
674 fprintf(stderr
, "VM start=0x%"PRIX64
" end=0x%"PRIX64
" | Texture %ix%ix%i, %i levels, %i samples, %s\n",
675 rtex
->resource
.gpu_address
,
676 rtex
->resource
.gpu_address
+ rtex
->resource
.buf
->size
,
677 base
->width0
, base
->height0
, util_max_layer(base
, 0)+1, base
->last_level
+1,
678 base
->nr_samples
? base
->nr_samples
: 1, util_format_short_name(base
->format
));
681 if (rscreen
->debug_flags
& DBG_TEX
) {
682 printf("Texture: npix_x=%u, npix_y=%u, npix_z=%u, blk_w=%u, "
683 "blk_h=%u, blk_d=%u, array_size=%u, last_level=%u, "
684 "bpe=%u, nsamples=%u, flags=0x%x, %s\n",
685 rtex
->surface
.npix_x
, rtex
->surface
.npix_y
,
686 rtex
->surface
.npix_z
, rtex
->surface
.blk_w
,
687 rtex
->surface
.blk_h
, rtex
->surface
.blk_d
,
688 rtex
->surface
.array_size
, rtex
->surface
.last_level
,
689 rtex
->surface
.bpe
, rtex
->surface
.nsamples
,
690 rtex
->surface
.flags
, util_format_short_name(base
->format
));
691 for (int i
= 0; i
<= rtex
->surface
.last_level
; i
++) {
692 printf(" L %i: offset=%"PRIu64
", slice_size=%"PRIu64
", npix_x=%u, "
693 "npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, "
694 "nblk_z=%u, pitch_bytes=%u, mode=%u\n",
695 i
, rtex
->surface
.level
[i
].offset
,
696 rtex
->surface
.level
[i
].slice_size
,
697 u_minify(rtex
->resource
.b
.b
.width0
, i
),
698 u_minify(rtex
->resource
.b
.b
.height0
, i
),
699 u_minify(rtex
->resource
.b
.b
.depth0
, i
),
700 rtex
->surface
.level
[i
].nblk_x
,
701 rtex
->surface
.level
[i
].nblk_y
,
702 rtex
->surface
.level
[i
].nblk_z
,
703 rtex
->surface
.level
[i
].pitch_bytes
,
704 rtex
->surface
.level
[i
].mode
);
706 if (rtex
->surface
.flags
& RADEON_SURF_SBUFFER
) {
707 for (int i
= 0; i
<= rtex
->surface
.last_level
; i
++) {
708 printf(" S %i: offset=%"PRIu64
", slice_size=%"PRIu64
", npix_x=%u, "
709 "npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, "
710 "nblk_z=%u, pitch_bytes=%u, mode=%u\n",
711 i
, rtex
->surface
.stencil_level
[i
].offset
,
712 rtex
->surface
.stencil_level
[i
].slice_size
,
713 u_minify(rtex
->resource
.b
.b
.width0
, i
),
714 u_minify(rtex
->resource
.b
.b
.height0
, i
),
715 u_minify(rtex
->resource
.b
.b
.depth0
, i
),
716 rtex
->surface
.stencil_level
[i
].nblk_x
,
717 rtex
->surface
.stencil_level
[i
].nblk_y
,
718 rtex
->surface
.stencil_level
[i
].nblk_z
,
719 rtex
->surface
.stencil_level
[i
].pitch_bytes
,
720 rtex
->surface
.stencil_level
[i
].mode
);
727 static unsigned r600_choose_tiling(struct r600_common_screen
*rscreen
,
728 const struct pipe_resource
*templ
)
730 const struct util_format_description
*desc
= util_format_description(templ
->format
);
731 bool force_tiling
= templ
->flags
& R600_RESOURCE_FLAG_FORCE_TILING
;
733 /* MSAA resources must be 2D tiled. */
734 if (templ
->nr_samples
> 1)
735 return RADEON_SURF_MODE_2D
;
737 /* Transfer resources should be linear. */
738 if (templ
->flags
& R600_RESOURCE_FLAG_TRANSFER
)
739 return RADEON_SURF_MODE_LINEAR_ALIGNED
;
741 /* r600g: force tiling on TEXTURE_2D and TEXTURE_3D compute resources. */
742 if (rscreen
->chip_class
>= R600
&& rscreen
->chip_class
<= CAYMAN
&&
743 (templ
->bind
& PIPE_BIND_COMPUTE_RESOURCE
) &&
744 (templ
->target
== PIPE_TEXTURE_2D
||
745 templ
->target
== PIPE_TEXTURE_3D
))
748 /* Handle common candidates for the linear mode.
749 * Compressed textures must always be tiled. */
750 if (!force_tiling
&& !util_format_is_compressed(templ
->format
)) {
751 /* Not everything can be linear, so we cannot enforce it
752 * for all textures. */
753 if ((rscreen
->debug_flags
& DBG_NO_TILING
) &&
754 (!util_format_is_depth_or_stencil(templ
->format
) ||
755 !(templ
->flags
& R600_RESOURCE_FLAG_FLUSHED_DEPTH
)))
756 return RADEON_SURF_MODE_LINEAR_ALIGNED
;
758 /* Tiling doesn't work with the 422 (SUBSAMPLED) formats on R600+. */
759 if (desc
->layout
== UTIL_FORMAT_LAYOUT_SUBSAMPLED
)
760 return RADEON_SURF_MODE_LINEAR_ALIGNED
;
762 /* Cursors are linear on SI.
763 * (XXX double-check, maybe also use RADEON_SURF_SCANOUT) */
764 if (rscreen
->chip_class
>= SI
&&
765 (templ
->bind
& PIPE_BIND_CURSOR
))
766 return RADEON_SURF_MODE_LINEAR_ALIGNED
;
768 if (templ
->bind
& PIPE_BIND_LINEAR
)
769 return RADEON_SURF_MODE_LINEAR_ALIGNED
;
771 /* Textures with a very small height are recommended to be linear. */
772 if (templ
->target
== PIPE_TEXTURE_1D
||
773 templ
->target
== PIPE_TEXTURE_1D_ARRAY
||
775 return RADEON_SURF_MODE_LINEAR_ALIGNED
;
777 /* Textures likely to be mapped often. */
778 if (templ
->usage
== PIPE_USAGE_STAGING
||
779 templ
->usage
== PIPE_USAGE_STREAM
)
780 return RADEON_SURF_MODE_LINEAR_ALIGNED
;
783 /* Make small textures 1D tiled. */
784 if (templ
->width0
<= 16 || templ
->height0
<= 16 ||
785 (rscreen
->debug_flags
& DBG_NO_2D_TILING
))
786 return RADEON_SURF_MODE_1D
;
788 /* The allocator will switch to 1D if needed. */
789 return RADEON_SURF_MODE_2D
;
792 struct pipe_resource
*r600_texture_create(struct pipe_screen
*screen
,
793 const struct pipe_resource
*templ
)
795 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)screen
;
796 struct radeon_surf surface
= {0};
799 r
= r600_init_surface(rscreen
, &surface
, templ
,
800 r600_choose_tiling(rscreen
, templ
),
801 templ
->flags
& R600_RESOURCE_FLAG_FLUSHED_DEPTH
);
805 r
= rscreen
->ws
->surface_best(rscreen
->ws
, &surface
);
809 return (struct pipe_resource
*)r600_texture_create_object(screen
, templ
,
813 static struct pipe_resource
*r600_texture_from_handle(struct pipe_screen
*screen
,
814 const struct pipe_resource
*templ
,
815 struct winsys_handle
*whandle
)
817 struct r600_common_screen
*rscreen
= (struct r600_common_screen
*)screen
;
818 struct pb_buffer
*buf
= NULL
;
821 enum radeon_bo_layout micro
, macro
;
822 struct radeon_surf surface
;
826 /* Support only 2D textures without mipmaps */
827 if ((templ
->target
!= PIPE_TEXTURE_2D
&& templ
->target
!= PIPE_TEXTURE_RECT
) ||
828 templ
->depth0
!= 1 || templ
->last_level
!= 0)
831 buf
= rscreen
->ws
->buffer_from_handle(rscreen
->ws
, whandle
, &stride
);
835 rscreen
->ws
->buffer_get_tiling(buf
, µ
, ¯o
,
836 &surface
.bankw
, &surface
.bankh
,
838 &surface
.stencil_tile_split
,
839 &surface
.mtilea
, &scanout
);
841 if (macro
== RADEON_LAYOUT_TILED
)
842 array_mode
= RADEON_SURF_MODE_2D
;
843 else if (micro
== RADEON_LAYOUT_TILED
)
844 array_mode
= RADEON_SURF_MODE_1D
;
846 array_mode
= RADEON_SURF_MODE_LINEAR_ALIGNED
;
848 r
= r600_init_surface(rscreen
, &surface
, templ
, array_mode
, false);
854 surface
.flags
|= RADEON_SURF_SCANOUT
;
856 return (struct pipe_resource
*)r600_texture_create_object(screen
, templ
,
857 stride
, buf
, &surface
);
860 bool r600_init_flushed_depth_texture(struct pipe_context
*ctx
,
861 struct pipe_resource
*texture
,
862 struct r600_texture
**staging
)
864 struct r600_texture
*rtex
= (struct r600_texture
*)texture
;
865 struct pipe_resource resource
;
866 struct r600_texture
**flushed_depth_texture
= staging
?
867 staging
: &rtex
->flushed_depth_texture
;
869 if (!staging
&& rtex
->flushed_depth_texture
)
870 return true; /* it's ready */
872 resource
.target
= texture
->target
;
873 resource
.format
= texture
->format
;
874 resource
.width0
= texture
->width0
;
875 resource
.height0
= texture
->height0
;
876 resource
.depth0
= texture
->depth0
;
877 resource
.array_size
= texture
->array_size
;
878 resource
.last_level
= texture
->last_level
;
879 resource
.nr_samples
= texture
->nr_samples
;
880 resource
.usage
= staging
? PIPE_USAGE_STAGING
: PIPE_USAGE_DEFAULT
;
881 resource
.bind
= texture
->bind
& ~PIPE_BIND_DEPTH_STENCIL
;
882 resource
.flags
= texture
->flags
| R600_RESOURCE_FLAG_FLUSHED_DEPTH
;
885 resource
.flags
|= R600_RESOURCE_FLAG_TRANSFER
;
887 *flushed_depth_texture
= (struct r600_texture
*)ctx
->screen
->resource_create(ctx
->screen
, &resource
);
888 if (*flushed_depth_texture
== NULL
) {
889 R600_ERR("failed to create temporary texture to hold flushed depth\n");
893 (*flushed_depth_texture
)->is_flushing_texture
= TRUE
;
894 (*flushed_depth_texture
)->non_disp_tiling
= false;
899 * Initialize the pipe_resource descriptor to be of the same size as the box,
900 * which is supposed to hold a subregion of the texture "orig" at the given
903 static void r600_init_temp_resource_from_box(struct pipe_resource
*res
,
904 struct pipe_resource
*orig
,
905 const struct pipe_box
*box
,
906 unsigned level
, unsigned flags
)
908 memset(res
, 0, sizeof(*res
));
909 res
->format
= orig
->format
;
910 res
->width0
= box
->width
;
911 res
->height0
= box
->height
;
914 res
->usage
= flags
& R600_RESOURCE_FLAG_TRANSFER
? PIPE_USAGE_STAGING
: PIPE_USAGE_DEFAULT
;
917 /* We must set the correct texture target and dimensions for a 3D box. */
918 if (box
->depth
> 1 && util_max_layer(orig
, level
) > 0)
919 res
->target
= orig
->target
;
921 res
->target
= PIPE_TEXTURE_2D
;
923 switch (res
->target
) {
924 case PIPE_TEXTURE_1D_ARRAY
:
925 case PIPE_TEXTURE_2D_ARRAY
:
926 case PIPE_TEXTURE_CUBE_ARRAY
:
927 res
->array_size
= box
->depth
;
929 case PIPE_TEXTURE_3D
:
930 res
->depth0
= box
->depth
;
936 static void *r600_texture_transfer_map(struct pipe_context
*ctx
,
937 struct pipe_resource
*texture
,
940 const struct pipe_box
*box
,
941 struct pipe_transfer
**ptransfer
)
943 struct r600_common_context
*rctx
= (struct r600_common_context
*)ctx
;
944 struct r600_texture
*rtex
= (struct r600_texture
*)texture
;
945 struct r600_transfer
*trans
;
946 boolean use_staging_texture
= FALSE
;
947 struct r600_resource
*buf
;
951 /* We cannot map a tiled texture directly because the data is
952 * in a different order, therefore we do detiling using a blit.
954 * Also, use a temporary in GTT memory for read transfers, as
955 * the CPU is much happier reading out of cached system memory
956 * than uncached VRAM.
958 if (rtex
->surface
.level
[0].mode
>= RADEON_SURF_MODE_1D
) {
959 use_staging_texture
= TRUE
;
960 } else if ((usage
& PIPE_TRANSFER_READ
) && !(usage
& PIPE_TRANSFER_MAP_DIRECTLY
) &&
961 (rtex
->resource
.domains
== RADEON_DOMAIN_VRAM
)) {
962 /* Untiled buffers in VRAM, which is slow for CPU reads */
963 use_staging_texture
= TRUE
;
964 } else if (!(usage
& PIPE_TRANSFER_READ
) &&
965 (r600_rings_is_buffer_referenced(rctx
, rtex
->resource
.cs_buf
, RADEON_USAGE_READWRITE
) ||
966 !rctx
->ws
->buffer_wait(rtex
->resource
.buf
, 0, RADEON_USAGE_READWRITE
))) {
967 /* Use a staging texture for uploads if the underlying BO is busy. */
968 use_staging_texture
= TRUE
;
971 if (texture
->flags
& R600_RESOURCE_FLAG_TRANSFER
) {
972 use_staging_texture
= FALSE
;
975 if (use_staging_texture
&& (usage
& PIPE_TRANSFER_MAP_DIRECTLY
)) {
979 trans
= CALLOC_STRUCT(r600_transfer
);
982 trans
->transfer
.resource
= texture
;
983 trans
->transfer
.level
= level
;
984 trans
->transfer
.usage
= usage
;
985 trans
->transfer
.box
= *box
;
987 if (rtex
->is_depth
) {
988 struct r600_texture
*staging_depth
;
990 if (rtex
->resource
.b
.b
.nr_samples
> 1) {
991 /* MSAA depth buffers need to be converted to single sample buffers.
993 * Mapping MSAA depth buffers can occur if ReadPixels is called
994 * with a multisample GLX visual.
996 * First downsample the depth buffer to a temporary texture,
997 * then decompress the temporary one to staging.
999 * Only the region being mapped is transfered.
1001 struct pipe_resource resource
;
1003 r600_init_temp_resource_from_box(&resource
, texture
, box
, level
, 0);
1005 if (!r600_init_flushed_depth_texture(ctx
, &resource
, &staging_depth
)) {
1006 R600_ERR("failed to create temporary texture to hold untiled copy\n");
1011 if (usage
& PIPE_TRANSFER_READ
) {
1012 struct pipe_resource
*temp
= ctx
->screen
->resource_create(ctx
->screen
, &resource
);
1014 R600_ERR("failed to create a temporary depth texture\n");
1019 r600_copy_region_with_blit(ctx
, temp
, 0, 0, 0, 0, texture
, level
, box
);
1020 rctx
->blit_decompress_depth(ctx
, (struct r600_texture
*)temp
, staging_depth
,
1021 0, 0, 0, box
->depth
, 0, 0);
1022 pipe_resource_reference((struct pipe_resource
**)&temp
, NULL
);
1026 /* XXX: only readback the rectangle which is being mapped? */
1027 /* XXX: when discard is true, no need to read back from depth texture */
1028 if (!r600_init_flushed_depth_texture(ctx
, texture
, &staging_depth
)) {
1029 R600_ERR("failed to create temporary texture to hold untiled copy\n");
1034 rctx
->blit_decompress_depth(ctx
, rtex
, staging_depth
,
1036 box
->z
, box
->z
+ box
->depth
- 1,
1039 offset
= r600_texture_get_offset(staging_depth
, level
, box
);
1042 trans
->transfer
.stride
= staging_depth
->surface
.level
[level
].pitch_bytes
;
1043 trans
->transfer
.layer_stride
= staging_depth
->surface
.level
[level
].slice_size
;
1044 trans
->staging
= (struct r600_resource
*)staging_depth
;
1045 } else if (use_staging_texture
) {
1046 struct pipe_resource resource
;
1047 struct r600_texture
*staging
;
1049 r600_init_temp_resource_from_box(&resource
, texture
, box
, level
,
1050 R600_RESOURCE_FLAG_TRANSFER
);
1051 resource
.usage
= (usage
& PIPE_TRANSFER_READ
) ?
1052 PIPE_USAGE_STAGING
: PIPE_USAGE_STREAM
;
1054 /* Create the temporary texture. */
1055 staging
= (struct r600_texture
*)ctx
->screen
->resource_create(ctx
->screen
, &resource
);
1056 if (staging
== NULL
) {
1057 R600_ERR("failed to create temporary texture to hold untiled copy\n");
1061 trans
->staging
= &staging
->resource
;
1062 trans
->transfer
.stride
= staging
->surface
.level
[0].pitch_bytes
;
1063 trans
->transfer
.layer_stride
= staging
->surface
.level
[0].slice_size
;
1064 if (usage
& PIPE_TRANSFER_READ
) {
1065 r600_copy_to_staging_texture(ctx
, trans
);
1068 /* the resource is mapped directly */
1069 trans
->transfer
.stride
= rtex
->surface
.level
[level
].pitch_bytes
;
1070 trans
->transfer
.layer_stride
= rtex
->surface
.level
[level
].slice_size
;
1071 offset
= r600_texture_get_offset(rtex
, level
, box
);
1074 if (trans
->staging
) {
1075 buf
= trans
->staging
;
1076 if (!rtex
->is_depth
&& !(usage
& PIPE_TRANSFER_READ
))
1077 usage
|= PIPE_TRANSFER_UNSYNCHRONIZED
;
1079 buf
= &rtex
->resource
;
1082 if (!(map
= r600_buffer_map_sync_with_rings(rctx
, buf
, usage
))) {
1083 pipe_resource_reference((struct pipe_resource
**)&trans
->staging
, NULL
);
1088 *ptransfer
= &trans
->transfer
;
1089 return map
+ offset
;
1092 static void r600_texture_transfer_unmap(struct pipe_context
*ctx
,
1093 struct pipe_transfer
* transfer
)
1095 struct r600_transfer
*rtransfer
= (struct r600_transfer
*)transfer
;
1096 struct pipe_resource
*texture
= transfer
->resource
;
1097 struct r600_texture
*rtex
= (struct r600_texture
*)texture
;
1099 if ((transfer
->usage
& PIPE_TRANSFER_WRITE
) && rtransfer
->staging
) {
1100 if (rtex
->is_depth
&& rtex
->resource
.b
.b
.nr_samples
<= 1) {
1101 ctx
->resource_copy_region(ctx
, texture
, transfer
->level
,
1102 transfer
->box
.x
, transfer
->box
.y
, transfer
->box
.z
,
1103 &rtransfer
->staging
->b
.b
, transfer
->level
,
1106 r600_copy_from_staging_texture(ctx
, rtransfer
);
1110 if (rtransfer
->staging
)
1111 pipe_resource_reference((struct pipe_resource
**)&rtransfer
->staging
, NULL
);
1116 static const struct u_resource_vtbl r600_texture_vtbl
=
1118 NULL
, /* get_handle */
1119 r600_texture_destroy
, /* resource_destroy */
1120 r600_texture_transfer_map
, /* transfer_map */
1121 u_default_transfer_flush_region
, /* transfer_flush_region */
1122 r600_texture_transfer_unmap
, /* transfer_unmap */
1123 NULL
/* transfer_inline_write */
1126 struct pipe_surface
*r600_create_surface_custom(struct pipe_context
*pipe
,
1127 struct pipe_resource
*texture
,
1128 const struct pipe_surface
*templ
,
1129 unsigned width
, unsigned height
)
1131 struct r600_surface
*surface
= CALLOC_STRUCT(r600_surface
);
1133 if (surface
== NULL
)
1136 assert(templ
->u
.tex
.first_layer
<= util_max_layer(texture
, templ
->u
.tex
.level
));
1137 assert(templ
->u
.tex
.last_layer
<= util_max_layer(texture
, templ
->u
.tex
.level
));
1139 pipe_reference_init(&surface
->base
.reference
, 1);
1140 pipe_resource_reference(&surface
->base
.texture
, texture
);
1141 surface
->base
.context
= pipe
;
1142 surface
->base
.format
= templ
->format
;
1143 surface
->base
.width
= width
;
1144 surface
->base
.height
= height
;
1145 surface
->base
.u
= templ
->u
;
1146 return &surface
->base
;
1149 static struct pipe_surface
*r600_create_surface(struct pipe_context
*pipe
,
1150 struct pipe_resource
*tex
,
1151 const struct pipe_surface
*templ
)
1153 unsigned level
= templ
->u
.tex
.level
;
1155 return r600_create_surface_custom(pipe
, tex
, templ
,
1156 u_minify(tex
->width0
, level
),
1157 u_minify(tex
->height0
, level
));
1160 static void r600_surface_destroy(struct pipe_context
*pipe
,
1161 struct pipe_surface
*surface
)
1163 struct r600_surface
*surf
= (struct r600_surface
*)surface
;
1164 pipe_resource_reference((struct pipe_resource
**)&surf
->cb_buffer_fmask
, NULL
);
1165 pipe_resource_reference((struct pipe_resource
**)&surf
->cb_buffer_cmask
, NULL
);
1166 pipe_resource_reference(&surface
->texture
, NULL
);
1170 unsigned r600_translate_colorswap(enum pipe_format format
)
1172 const struct util_format_description
*desc
= util_format_description(format
);
1174 #define HAS_SWIZZLE(chan,swz) (desc->swizzle[chan] == UTIL_FORMAT_SWIZZLE_##swz)
1176 if (format
== PIPE_FORMAT_R11G11B10_FLOAT
) /* isn't plain */
1177 return V_0280A0_SWAP_STD
;
1179 if (desc
->layout
!= UTIL_FORMAT_LAYOUT_PLAIN
)
1182 switch (desc
->nr_channels
) {
1184 if (HAS_SWIZZLE(0,X
))
1185 return V_0280A0_SWAP_STD
; /* X___ */
1186 else if (HAS_SWIZZLE(3,X
))
1187 return V_0280A0_SWAP_ALT_REV
; /* ___X */
1190 if ((HAS_SWIZZLE(0,X
) && HAS_SWIZZLE(1,Y
)) ||
1191 (HAS_SWIZZLE(0,X
) && HAS_SWIZZLE(1,NONE
)) ||
1192 (HAS_SWIZZLE(0,NONE
) && HAS_SWIZZLE(1,Y
)))
1193 return V_0280A0_SWAP_STD
; /* XY__ */
1194 else if ((HAS_SWIZZLE(0,Y
) && HAS_SWIZZLE(1,X
)) ||
1195 (HAS_SWIZZLE(0,Y
) && HAS_SWIZZLE(1,NONE
)) ||
1196 (HAS_SWIZZLE(0,NONE
) && HAS_SWIZZLE(1,X
)))
1197 return V_0280A0_SWAP_STD_REV
; /* YX__ */
1198 else if (HAS_SWIZZLE(0,X
) && HAS_SWIZZLE(3,Y
))
1199 return V_0280A0_SWAP_ALT
; /* X__Y */
1200 else if (HAS_SWIZZLE(0,Y
) && HAS_SWIZZLE(3,X
))
1201 return V_0280A0_SWAP_ALT_REV
; /* Y__X */
1204 if (HAS_SWIZZLE(0,X
))
1205 return V_0280A0_SWAP_STD
; /* XYZ */
1206 else if (HAS_SWIZZLE(0,Z
))
1207 return V_0280A0_SWAP_STD_REV
; /* ZYX */
1210 /* check the middle channels, the 1st and 4th channel can be NONE */
1211 if (HAS_SWIZZLE(1,Y
) && HAS_SWIZZLE(2,Z
))
1212 return V_0280A0_SWAP_STD
; /* XYZW */
1213 else if (HAS_SWIZZLE(1,Z
) && HAS_SWIZZLE(2,Y
))
1214 return V_0280A0_SWAP_STD_REV
; /* WZYX */
1215 else if (HAS_SWIZZLE(1,Y
) && HAS_SWIZZLE(2,X
))
1216 return V_0280A0_SWAP_ALT
; /* ZYXW */
1217 else if (HAS_SWIZZLE(1,X
) && HAS_SWIZZLE(2,Y
))
1218 return V_0280A0_SWAP_ALT_REV
; /* WXYZ */
1224 static void evergreen_set_clear_color(struct r600_texture
*rtex
,
1225 enum pipe_format surface_format
,
1226 const union pipe_color_union
*color
)
1228 union util_color uc
;
1230 memset(&uc
, 0, sizeof(uc
));
1232 if (util_format_is_pure_uint(surface_format
)) {
1233 util_format_write_4ui(surface_format
, color
->ui
, 0, &uc
, 0, 0, 0, 1, 1);
1234 } else if (util_format_is_pure_sint(surface_format
)) {
1235 util_format_write_4i(surface_format
, color
->i
, 0, &uc
, 0, 0, 0, 1, 1);
1237 util_pack_color(color
->f
, surface_format
, &uc
);
1240 memcpy(rtex
->color_clear_value
, &uc
, 2 * sizeof(uint32_t));
1243 static void vi_get_fast_clear_parameters(enum pipe_format surface_format
,
1244 const union pipe_color_union
*color
,
1245 uint32_t* reset_value
,
1246 bool* clear_words_needed
)
1248 bool values
[4] = {};
1250 bool main_value
= false;
1251 bool extra_value
= false;
1253 const struct util_format_description
*desc
= util_format_description(surface_format
);
1255 *clear_words_needed
= true;
1256 *reset_value
= 0x20202020U
;
1258 /* If we want to clear without needing a fast clear eliminate step, we
1259 * can set each channel to 0 or 1 (or 0/max for integer formats). We
1260 * have two sets of flags, one for the last or first channel(extra) and
1261 * one for the other channels(main).
1264 if (surface_format
== PIPE_FORMAT_R11G11B10_FLOAT
||
1265 surface_format
== PIPE_FORMAT_B5G6R5_UNORM
||
1266 surface_format
== PIPE_FORMAT_B5G6R5_SRGB
) {
1268 } else if (desc
->layout
== UTIL_FORMAT_LAYOUT_PLAIN
) {
1269 if(r600_translate_colorswap(surface_format
) <= 1)
1270 extra_channel
= desc
->nr_channels
- 1;
1276 for (i
= 0; i
< 4; ++i
) {
1277 int index
= desc
->swizzle
[i
] - UTIL_FORMAT_SWIZZLE_X
;
1279 if (desc
->swizzle
[i
] < UTIL_FORMAT_SWIZZLE_X
||
1280 desc
->swizzle
[i
] > UTIL_FORMAT_SWIZZLE_W
)
1283 if (util_format_is_pure_sint(surface_format
)) {
1284 values
[i
] = color
->i
[i
] != 0;
1285 if (color
->i
[i
] != 0 && color
->i
[i
] != INT32_MAX
)
1287 } else if (util_format_is_pure_uint(surface_format
)) {
1288 values
[i
] = color
->ui
[i
] != 0U;
1289 if (color
->ui
[i
] != 0U && color
->ui
[i
] != UINT32_MAX
)
1292 values
[i
] = color
->f
[i
] != 0.0F
;
1293 if (color
->f
[i
] != 0.0F
&& color
->f
[i
] != 1.0F
)
1297 if (index
== extra_channel
)
1298 extra_value
= values
[i
];
1300 main_value
= values
[i
];
1303 for (int i
= 0; i
< 4; ++i
)
1304 if (values
[i
] != main_value
&&
1305 desc
->swizzle
[i
] - UTIL_FORMAT_SWIZZLE_X
!= extra_channel
&&
1306 desc
->swizzle
[i
] >= UTIL_FORMAT_SWIZZLE_X
&&
1307 desc
->swizzle
[i
] <= UTIL_FORMAT_SWIZZLE_W
)
1310 *clear_words_needed
= false;
1312 *reset_value
|= 0x80808080U
;
1315 *reset_value
|= 0x40404040U
;
1318 void evergreen_do_fast_color_clear(struct r600_common_context
*rctx
,
1319 struct pipe_framebuffer_state
*fb
,
1320 struct r600_atom
*fb_state
,
1321 unsigned *buffers
, unsigned *dirty_cbufs
,
1322 const union pipe_color_union
*color
)
1326 if (rctx
->render_cond
)
1329 for (i
= 0; i
< fb
->nr_cbufs
; i
++) {
1330 struct r600_texture
*tex
;
1331 unsigned clear_bit
= PIPE_CLEAR_COLOR0
<< i
;
1336 /* if this colorbuffer is not being cleared */
1337 if (!(*buffers
& clear_bit
))
1340 tex
= (struct r600_texture
*)fb
->cbufs
[i
]->texture
;
1342 /* 128-bit formats are unusupported */
1343 if (util_format_get_blocksizebits(fb
->cbufs
[i
]->format
) > 64) {
1347 /* the clear is allowed if all layers are bound */
1348 if (fb
->cbufs
[i
]->u
.tex
.first_layer
!= 0 ||
1349 fb
->cbufs
[i
]->u
.tex
.last_layer
!= util_max_layer(&tex
->resource
.b
.b
, 0)) {
1353 /* cannot clear mipmapped textures */
1354 if (fb
->cbufs
[i
]->texture
->last_level
!= 0) {
1358 /* only supported on tiled surfaces */
1359 if (tex
->surface
.level
[0].mode
< RADEON_SURF_MODE_1D
) {
1363 /* fast color clear with 1D tiling doesn't work on old kernels and CIK */
1364 if (tex
->surface
.level
[0].mode
== RADEON_SURF_MODE_1D
&&
1365 rctx
->chip_class
>= CIK
&&
1366 rctx
->screen
->info
.drm_major
== 2 &&
1367 rctx
->screen
->info
.drm_minor
< 38) {
1371 if (tex
->dcc_buffer
) {
1372 uint32_t reset_value
;
1373 bool clear_words_needed
;
1375 if (rctx
->screen
->debug_flags
& DBG_NO_DCC_CLEAR
)
1378 vi_get_fast_clear_parameters(fb
->cbufs
[i
]->format
, color
, &reset_value
, &clear_words_needed
);
1380 rctx
->clear_buffer(&rctx
->b
, &tex
->dcc_buffer
->b
.b
,
1381 0, tex
->surface
.dcc_size
, reset_value
, true);
1383 if (clear_words_needed
)
1384 tex
->dirty_level_mask
|= 1 << fb
->cbufs
[i
]->u
.tex
.level
;
1386 /* ensure CMASK is enabled */
1387 r600_texture_alloc_cmask_separate(rctx
->screen
, tex
);
1388 if (tex
->cmask
.size
== 0) {
1392 /* Do the fast clear. */
1393 rctx
->clear_buffer(&rctx
->b
, &tex
->cmask_buffer
->b
.b
,
1394 tex
->cmask
.offset
, tex
->cmask
.size
, 0, true);
1396 tex
->dirty_level_mask
|= 1 << fb
->cbufs
[i
]->u
.tex
.level
;
1399 evergreen_set_clear_color(tex
, fb
->cbufs
[i
]->format
, color
);
1402 *dirty_cbufs
|= 1 << i
;
1403 rctx
->set_atom_dirty(rctx
, fb_state
, true);
1404 *buffers
&= ~clear_bit
;
1408 void r600_init_screen_texture_functions(struct r600_common_screen
*rscreen
)
1410 rscreen
->b
.resource_from_handle
= r600_texture_from_handle
;
1411 rscreen
->b
.resource_get_handle
= r600_texture_get_handle
;
1414 void r600_init_context_texture_functions(struct r600_common_context
*rctx
)
1416 rctx
->b
.create_surface
= r600_create_surface
;
1417 rctx
->b
.surface_destroy
= r600_surface_destroy
;