2 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
3 * Copyright 2015 Advanced Micro Devices, Inc.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * on the rights to use, copy, modify, merge, publish, distribute, sub
10 * license, and/or sell copies of the Software, and to permit persons to whom
11 * the Software is furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
21 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
22 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
23 * USE OR OTHER DEALINGS IN THE SOFTWARE.
27 #include "si_compute.h"
28 #include "util/u_format.h"
29 #include "util/u_log.h"
30 #include "util/u_surface.h"
33 SI_COPY
= SI_SAVE_FRAMEBUFFER
| SI_SAVE_TEXTURES
|
34 SI_SAVE_FRAGMENT_STATE
| SI_DISABLE_RENDER_COND
,
36 SI_BLIT
= SI_SAVE_FRAMEBUFFER
| SI_SAVE_TEXTURES
|
37 SI_SAVE_FRAGMENT_STATE
,
39 SI_DECOMPRESS
= SI_SAVE_FRAMEBUFFER
| SI_SAVE_FRAGMENT_STATE
|
40 SI_DISABLE_RENDER_COND
,
42 SI_COLOR_RESOLVE
= SI_SAVE_FRAMEBUFFER
| SI_SAVE_FRAGMENT_STATE
45 void si_blitter_begin(struct si_context
*sctx
, enum si_blitter_op op
)
47 util_blitter_save_vertex_shader(sctx
->blitter
, sctx
->vs_shader
.cso
);
48 util_blitter_save_tessctrl_shader(sctx
->blitter
, sctx
->tcs_shader
.cso
);
49 util_blitter_save_tesseval_shader(sctx
->blitter
, sctx
->tes_shader
.cso
);
50 util_blitter_save_geometry_shader(sctx
->blitter
, sctx
->gs_shader
.cso
);
51 util_blitter_save_so_targets(sctx
->blitter
, sctx
->streamout
.num_targets
,
52 (struct pipe_stream_output_target
**)sctx
->streamout
.targets
);
53 util_blitter_save_rasterizer(sctx
->blitter
, sctx
->queued
.named
.rasterizer
);
55 if (op
& SI_SAVE_FRAGMENT_STATE
) {
56 util_blitter_save_blend(sctx
->blitter
, sctx
->queued
.named
.blend
);
57 util_blitter_save_depth_stencil_alpha(sctx
->blitter
, sctx
->queued
.named
.dsa
);
58 util_blitter_save_stencil_ref(sctx
->blitter
, &sctx
->stencil_ref
.state
);
59 util_blitter_save_fragment_shader(sctx
->blitter
, sctx
->ps_shader
.cso
);
60 util_blitter_save_sample_mask(sctx
->blitter
, sctx
->sample_mask
);
61 util_blitter_save_scissor(sctx
->blitter
, &sctx
->scissors
[0]);
62 util_blitter_save_window_rectangles(sctx
->blitter
,
63 sctx
->window_rectangles_include
,
64 sctx
->num_window_rectangles
,
65 sctx
->window_rectangles
);
68 if (op
& SI_SAVE_FRAMEBUFFER
)
69 util_blitter_save_framebuffer(sctx
->blitter
, &sctx
->framebuffer
.state
);
71 if (op
& SI_SAVE_TEXTURES
) {
72 util_blitter_save_fragment_sampler_states(
74 (void**)sctx
->samplers
[PIPE_SHADER_FRAGMENT
].sampler_states
);
76 util_blitter_save_fragment_sampler_views(sctx
->blitter
, 2,
77 sctx
->samplers
[PIPE_SHADER_FRAGMENT
].views
);
80 if (op
& SI_DISABLE_RENDER_COND
)
81 sctx
->render_cond_force_off
= true;
83 if (sctx
->screen
->dpbb_allowed
) {
84 sctx
->dpbb_force_off
= true;
85 si_mark_atom_dirty(sctx
, &sctx
->atoms
.s
.dpbb_state
);
89 void si_blitter_end(struct si_context
*sctx
)
91 if (sctx
->screen
->dpbb_allowed
) {
92 sctx
->dpbb_force_off
= false;
93 si_mark_atom_dirty(sctx
, &sctx
->atoms
.s
.dpbb_state
);
96 sctx
->render_cond_force_off
= false;
98 /* Restore shader pointers because the VS blit shader changed all
99 * non-global VS user SGPRs. */
100 sctx
->shader_pointers_dirty
|= SI_DESCS_SHADER_MASK(VERTEX
);
101 sctx
->vertex_buffer_pointer_dirty
= sctx
->vb_descriptors_buffer
!= NULL
;
102 si_mark_atom_dirty(sctx
, &sctx
->atoms
.s
.shader_pointers
);
105 static unsigned u_max_sample(struct pipe_resource
*r
)
107 return r
->nr_samples
? r
->nr_samples
- 1 : 0;
111 si_blit_dbcb_copy(struct si_context
*sctx
,
112 struct si_texture
*src
,
113 struct si_texture
*dst
,
114 unsigned planes
, unsigned level_mask
,
115 unsigned first_layer
, unsigned last_layer
,
116 unsigned first_sample
, unsigned last_sample
)
118 struct pipe_surface surf_tmpl
= {{0}};
119 unsigned layer
, sample
, checked_last_layer
, max_layer
;
120 unsigned fully_copied_levels
= 0;
122 if (planes
& PIPE_MASK_Z
)
123 sctx
->dbcb_depth_copy_enabled
= true;
124 if (planes
& PIPE_MASK_S
)
125 sctx
->dbcb_stencil_copy_enabled
= true;
126 si_mark_atom_dirty(sctx
, &sctx
->atoms
.s
.db_render_state
);
128 assert(sctx
->dbcb_depth_copy_enabled
|| sctx
->dbcb_stencil_copy_enabled
);
130 sctx
->decompression_enabled
= true;
133 unsigned level
= u_bit_scan(&level_mask
);
135 /* The smaller the mipmap level, the less layers there are
136 * as far as 3D textures are concerned. */
137 max_layer
= util_max_layer(&src
->buffer
.b
.b
, level
);
138 checked_last_layer
= MIN2(last_layer
, max_layer
);
140 surf_tmpl
.u
.tex
.level
= level
;
142 for (layer
= first_layer
; layer
<= checked_last_layer
; layer
++) {
143 struct pipe_surface
*zsurf
, *cbsurf
;
145 surf_tmpl
.format
= src
->buffer
.b
.b
.format
;
146 surf_tmpl
.u
.tex
.first_layer
= layer
;
147 surf_tmpl
.u
.tex
.last_layer
= layer
;
149 zsurf
= sctx
->b
.create_surface(&sctx
->b
, &src
->buffer
.b
.b
, &surf_tmpl
);
151 surf_tmpl
.format
= dst
->buffer
.b
.b
.format
;
152 cbsurf
= sctx
->b
.create_surface(&sctx
->b
, &dst
->buffer
.b
.b
, &surf_tmpl
);
154 for (sample
= first_sample
; sample
<= last_sample
; sample
++) {
155 if (sample
!= sctx
->dbcb_copy_sample
) {
156 sctx
->dbcb_copy_sample
= sample
;
157 si_mark_atom_dirty(sctx
, &sctx
->atoms
.s
.db_render_state
);
160 si_blitter_begin(sctx
, SI_DECOMPRESS
);
161 util_blitter_custom_depth_stencil(sctx
->blitter
, zsurf
, cbsurf
, 1 << sample
,
162 sctx
->custom_dsa_flush
, 1.0f
);
163 si_blitter_end(sctx
);
166 pipe_surface_reference(&zsurf
, NULL
);
167 pipe_surface_reference(&cbsurf
, NULL
);
170 if (first_layer
== 0 && last_layer
>= max_layer
&&
171 first_sample
== 0 && last_sample
>= u_max_sample(&src
->buffer
.b
.b
))
172 fully_copied_levels
|= 1u << level
;
175 sctx
->decompression_enabled
= false;
176 sctx
->dbcb_depth_copy_enabled
= false;
177 sctx
->dbcb_stencil_copy_enabled
= false;
178 si_mark_atom_dirty(sctx
, &sctx
->atoms
.s
.db_render_state
);
180 return fully_copied_levels
;
183 /* Helper function for si_blit_decompress_zs_in_place.
186 si_blit_decompress_zs_planes_in_place(struct si_context
*sctx
,
187 struct si_texture
*texture
,
188 unsigned planes
, unsigned level_mask
,
189 unsigned first_layer
, unsigned last_layer
)
191 struct pipe_surface
*zsurf
, surf_tmpl
= {{0}};
192 unsigned layer
, max_layer
, checked_last_layer
;
193 unsigned fully_decompressed_mask
= 0;
198 if (planes
& PIPE_MASK_S
)
199 sctx
->db_flush_stencil_inplace
= true;
200 if (planes
& PIPE_MASK_Z
)
201 sctx
->db_flush_depth_inplace
= true;
202 si_mark_atom_dirty(sctx
, &sctx
->atoms
.s
.db_render_state
);
204 surf_tmpl
.format
= texture
->buffer
.b
.b
.format
;
206 sctx
->decompression_enabled
= true;
209 unsigned level
= u_bit_scan(&level_mask
);
211 surf_tmpl
.u
.tex
.level
= level
;
213 /* The smaller the mipmap level, the less layers there are
214 * as far as 3D textures are concerned. */
215 max_layer
= util_max_layer(&texture
->buffer
.b
.b
, level
);
216 checked_last_layer
= MIN2(last_layer
, max_layer
);
218 for (layer
= first_layer
; layer
<= checked_last_layer
; layer
++) {
219 surf_tmpl
.u
.tex
.first_layer
= layer
;
220 surf_tmpl
.u
.tex
.last_layer
= layer
;
222 zsurf
= sctx
->b
.create_surface(&sctx
->b
, &texture
->buffer
.b
.b
, &surf_tmpl
);
224 si_blitter_begin(sctx
, SI_DECOMPRESS
);
225 util_blitter_custom_depth_stencil(sctx
->blitter
, zsurf
, NULL
, ~0,
226 sctx
->custom_dsa_flush
,
228 si_blitter_end(sctx
);
230 pipe_surface_reference(&zsurf
, NULL
);
233 /* The texture will always be dirty if some layers aren't flushed.
234 * I don't think this case occurs often though. */
235 if (first_layer
== 0 && last_layer
>= max_layer
) {
236 fully_decompressed_mask
|= 1u << level
;
240 if (planes
& PIPE_MASK_Z
)
241 texture
->dirty_level_mask
&= ~fully_decompressed_mask
;
242 if (planes
& PIPE_MASK_S
)
243 texture
->stencil_dirty_level_mask
&= ~fully_decompressed_mask
;
245 sctx
->decompression_enabled
= false;
246 sctx
->db_flush_depth_inplace
= false;
247 sctx
->db_flush_stencil_inplace
= false;
248 si_mark_atom_dirty(sctx
, &sctx
->atoms
.s
.db_render_state
);
251 /* Helper function of si_flush_depth_texture: decompress the given levels
252 * of Z and/or S planes in place.
255 si_blit_decompress_zs_in_place(struct si_context
*sctx
,
256 struct si_texture
*texture
,
257 unsigned levels_z
, unsigned levels_s
,
258 unsigned first_layer
, unsigned last_layer
)
260 unsigned both
= levels_z
& levels_s
;
262 /* First, do combined Z & S decompresses for levels that need it. */
264 si_blit_decompress_zs_planes_in_place(
265 sctx
, texture
, PIPE_MASK_Z
| PIPE_MASK_S
,
267 first_layer
, last_layer
);
272 /* Now do separate Z and S decompresses. */
274 si_blit_decompress_zs_planes_in_place(
275 sctx
, texture
, PIPE_MASK_Z
,
277 first_layer
, last_layer
);
281 si_blit_decompress_zs_planes_in_place(
282 sctx
, texture
, PIPE_MASK_S
,
284 first_layer
, last_layer
);
289 si_decompress_depth(struct si_context
*sctx
,
290 struct si_texture
*tex
,
291 unsigned required_planes
,
292 unsigned first_level
, unsigned last_level
,
293 unsigned first_layer
, unsigned last_layer
)
295 unsigned inplace_planes
= 0;
296 unsigned copy_planes
= 0;
297 unsigned level_mask
= u_bit_consecutive(first_level
, last_level
- first_level
+ 1);
298 unsigned levels_z
= 0;
299 unsigned levels_s
= 0;
301 if (required_planes
& PIPE_MASK_Z
) {
302 levels_z
= level_mask
& tex
->dirty_level_mask
;
305 if (si_can_sample_zs(tex
, false))
306 inplace_planes
|= PIPE_MASK_Z
;
308 copy_planes
|= PIPE_MASK_Z
;
311 if (required_planes
& PIPE_MASK_S
) {
312 levels_s
= level_mask
& tex
->stencil_dirty_level_mask
;
315 if (si_can_sample_zs(tex
, true))
316 inplace_planes
|= PIPE_MASK_S
;
318 copy_planes
|= PIPE_MASK_S
;
322 if (unlikely(sctx
->log
))
323 u_log_printf(sctx
->log
,
324 "\n------------------------------------------------\n"
325 "Decompress Depth (levels %u - %u, levels Z: 0x%x S: 0x%x)\n\n",
326 first_level
, last_level
, levels_z
, levels_s
);
328 /* We may have to allocate the flushed texture here when called from
329 * si_decompress_subresource.
332 (tex
->flushed_depth_texture
||
333 si_init_flushed_depth_texture(&sctx
->b
, &tex
->buffer
.b
.b
))) {
334 struct si_texture
*dst
= tex
->flushed_depth_texture
;
335 unsigned fully_copied_levels
;
338 assert(tex
->flushed_depth_texture
);
340 if (util_format_is_depth_and_stencil(dst
->buffer
.b
.b
.format
))
341 copy_planes
= PIPE_MASK_Z
| PIPE_MASK_S
;
343 if (copy_planes
& PIPE_MASK_Z
) {
347 if (copy_planes
& PIPE_MASK_S
) {
352 fully_copied_levels
= si_blit_dbcb_copy(
353 sctx
, tex
, dst
, copy_planes
, levels
,
354 first_layer
, last_layer
,
355 0, u_max_sample(&tex
->buffer
.b
.b
));
357 if (copy_planes
& PIPE_MASK_Z
)
358 tex
->dirty_level_mask
&= ~fully_copied_levels
;
359 if (copy_planes
& PIPE_MASK_S
)
360 tex
->stencil_dirty_level_mask
&= ~fully_copied_levels
;
363 if (inplace_planes
) {
364 bool has_htile
= si_htile_enabled(tex
, first_level
);
365 bool tc_compat_htile
= vi_tc_compat_htile_enabled(tex
, first_level
);
367 /* Don't decompress if there is no HTILE or when HTILE is
369 if (has_htile
&& !tc_compat_htile
) {
370 si_blit_decompress_zs_in_place(
373 first_layer
, last_layer
);
375 /* This is only a cache flush.
377 * Only clear the mask that we are flushing, because
378 * si_make_DB_shader_coherent() treats different levels
379 * and depth and stencil differently.
381 if (inplace_planes
& PIPE_MASK_Z
)
382 tex
->dirty_level_mask
&= ~levels_z
;
383 if (inplace_planes
& PIPE_MASK_S
)
384 tex
->stencil_dirty_level_mask
&= ~levels_s
;
387 /* Only in-place decompression needs to flush DB caches, or
388 * when we don't decompress but TC-compatible planes are dirty.
390 si_make_DB_shader_coherent(sctx
, tex
->buffer
.b
.b
.nr_samples
,
391 inplace_planes
& PIPE_MASK_S
,
394 /* set_framebuffer_state takes care of coherency for single-sample.
395 * The DB->CB copy uses CB for the final writes.
397 if (copy_planes
&& tex
->buffer
.b
.b
.nr_samples
> 1)
398 si_make_CB_shader_coherent(sctx
, tex
->buffer
.b
.b
.nr_samples
,
399 false, true /* no DCC */);
403 si_decompress_sampler_depth_textures(struct si_context
*sctx
,
404 struct si_samplers
*textures
)
407 unsigned mask
= textures
->needs_depth_decompress_mask
;
410 struct pipe_sampler_view
*view
;
411 struct si_sampler_view
*sview
;
412 struct si_texture
*tex
;
414 i
= u_bit_scan(&mask
);
416 view
= textures
->views
[i
];
418 sview
= (struct si_sampler_view
*)view
;
420 tex
= (struct si_texture
*)view
->texture
;
421 assert(tex
->db_compatible
);
423 si_decompress_depth(sctx
, tex
,
424 sview
->is_stencil_sampler
? PIPE_MASK_S
: PIPE_MASK_Z
,
425 view
->u
.tex
.first_level
, view
->u
.tex
.last_level
,
426 0, util_max_layer(&tex
->buffer
.b
.b
, view
->u
.tex
.first_level
));
430 static void si_blit_decompress_color(struct si_context
*sctx
,
431 struct si_texture
*tex
,
432 unsigned first_level
, unsigned last_level
,
433 unsigned first_layer
, unsigned last_layer
,
434 bool need_dcc_decompress
)
437 unsigned layer
, checked_last_layer
, max_layer
;
438 unsigned level_mask
=
439 u_bit_consecutive(first_level
, last_level
- first_level
+ 1);
441 if (!need_dcc_decompress
)
442 level_mask
&= tex
->dirty_level_mask
;
446 if (unlikely(sctx
->log
))
447 u_log_printf(sctx
->log
,
448 "\n------------------------------------------------\n"
449 "Decompress Color (levels %u - %u, mask 0x%x)\n\n",
450 first_level
, last_level
, level_mask
);
452 if (need_dcc_decompress
) {
453 custom_blend
= sctx
->custom_blend_dcc_decompress
;
455 assert(tex
->dcc_offset
);
457 /* disable levels without DCC */
458 for (int i
= first_level
; i
<= last_level
; i
++) {
459 if (!vi_dcc_enabled(tex
, i
))
460 level_mask
&= ~(1 << i
);
462 } else if (tex
->surface
.fmask_size
) {
463 custom_blend
= sctx
->custom_blend_fmask_decompress
;
465 custom_blend
= sctx
->custom_blend_eliminate_fastclear
;
468 sctx
->decompression_enabled
= true;
471 unsigned level
= u_bit_scan(&level_mask
);
473 /* The smaller the mipmap level, the less layers there are
474 * as far as 3D textures are concerned. */
475 max_layer
= util_max_layer(&tex
->buffer
.b
.b
, level
);
476 checked_last_layer
= MIN2(last_layer
, max_layer
);
478 for (layer
= first_layer
; layer
<= checked_last_layer
; layer
++) {
479 struct pipe_surface
*cbsurf
, surf_tmpl
;
481 surf_tmpl
.format
= tex
->buffer
.b
.b
.format
;
482 surf_tmpl
.u
.tex
.level
= level
;
483 surf_tmpl
.u
.tex
.first_layer
= layer
;
484 surf_tmpl
.u
.tex
.last_layer
= layer
;
485 cbsurf
= sctx
->b
.create_surface(&sctx
->b
, &tex
->buffer
.b
.b
, &surf_tmpl
);
487 /* Required before and after FMASK and DCC_DECOMPRESS. */
488 if (custom_blend
== sctx
->custom_blend_fmask_decompress
||
489 custom_blend
== sctx
->custom_blend_dcc_decompress
)
490 sctx
->flags
|= SI_CONTEXT_FLUSH_AND_INV_CB
;
492 si_blitter_begin(sctx
, SI_DECOMPRESS
);
493 util_blitter_custom_color(sctx
->blitter
, cbsurf
, custom_blend
);
494 si_blitter_end(sctx
);
496 if (custom_blend
== sctx
->custom_blend_fmask_decompress
||
497 custom_blend
== sctx
->custom_blend_dcc_decompress
)
498 sctx
->flags
|= SI_CONTEXT_FLUSH_AND_INV_CB
;
500 pipe_surface_reference(&cbsurf
, NULL
);
503 /* The texture will always be dirty if some layers aren't flushed.
504 * I don't think this case occurs often though. */
505 if (first_layer
== 0 && last_layer
>= max_layer
) {
506 tex
->dirty_level_mask
&= ~(1 << level
);
510 sctx
->decompression_enabled
= false;
511 si_make_CB_shader_coherent(sctx
, tex
->buffer
.b
.b
.nr_samples
,
512 vi_dcc_enabled(tex
, first_level
),
513 tex
->surface
.u
.gfx9
.dcc
.pipe_aligned
);
517 si_decompress_color_texture(struct si_context
*sctx
, struct si_texture
*tex
,
518 unsigned first_level
, unsigned last_level
)
520 /* CMASK or DCC can be discarded and we can still end up here. */
521 if (!tex
->cmask_buffer
&& !tex
->surface
.fmask_size
&& !tex
->dcc_offset
)
524 si_blit_decompress_color(sctx
, tex
, first_level
, last_level
, 0,
525 util_max_layer(&tex
->buffer
.b
.b
, first_level
),
530 si_decompress_sampler_color_textures(struct si_context
*sctx
,
531 struct si_samplers
*textures
)
534 unsigned mask
= textures
->needs_color_decompress_mask
;
537 struct pipe_sampler_view
*view
;
538 struct si_texture
*tex
;
540 i
= u_bit_scan(&mask
);
542 view
= textures
->views
[i
];
545 tex
= (struct si_texture
*)view
->texture
;
547 si_decompress_color_texture(sctx
, tex
, view
->u
.tex
.first_level
,
548 view
->u
.tex
.last_level
);
553 si_decompress_image_color_textures(struct si_context
*sctx
,
554 struct si_images
*images
)
557 unsigned mask
= images
->needs_color_decompress_mask
;
560 const struct pipe_image_view
*view
;
561 struct si_texture
*tex
;
563 i
= u_bit_scan(&mask
);
565 view
= &images
->views
[i
];
566 assert(view
->resource
->target
!= PIPE_BUFFER
);
568 tex
= (struct si_texture
*)view
->resource
;
570 si_decompress_color_texture(sctx
, tex
, view
->u
.tex
.level
,
575 static void si_check_render_feedback_texture(struct si_context
*sctx
,
576 struct si_texture
*tex
,
577 unsigned first_level
,
579 unsigned first_layer
,
582 bool render_feedback
= false;
584 if (!tex
->dcc_offset
)
587 for (unsigned j
= 0; j
< sctx
->framebuffer
.state
.nr_cbufs
; ++j
) {
588 struct si_surface
* surf
;
590 if (!sctx
->framebuffer
.state
.cbufs
[j
])
593 surf
= (struct si_surface
*)sctx
->framebuffer
.state
.cbufs
[j
];
595 if (tex
== (struct si_texture
*)surf
->base
.texture
&&
596 surf
->base
.u
.tex
.level
>= first_level
&&
597 surf
->base
.u
.tex
.level
<= last_level
&&
598 surf
->base
.u
.tex
.first_layer
<= last_layer
&&
599 surf
->base
.u
.tex
.last_layer
>= first_layer
) {
600 render_feedback
= true;
606 si_texture_disable_dcc(sctx
, tex
);
609 static void si_check_render_feedback_textures(struct si_context
*sctx
,
610 struct si_samplers
*textures
)
612 uint32_t mask
= textures
->enabled_mask
;
615 const struct pipe_sampler_view
*view
;
616 struct si_texture
*tex
;
618 unsigned i
= u_bit_scan(&mask
);
620 view
= textures
->views
[i
];
621 if(view
->texture
->target
== PIPE_BUFFER
)
624 tex
= (struct si_texture
*)view
->texture
;
626 si_check_render_feedback_texture(sctx
, tex
,
627 view
->u
.tex
.first_level
,
628 view
->u
.tex
.last_level
,
629 view
->u
.tex
.first_layer
,
630 view
->u
.tex
.last_layer
);
634 static void si_check_render_feedback_images(struct si_context
*sctx
,
635 struct si_images
*images
)
637 uint32_t mask
= images
->enabled_mask
;
640 const struct pipe_image_view
*view
;
641 struct si_texture
*tex
;
643 unsigned i
= u_bit_scan(&mask
);
645 view
= &images
->views
[i
];
646 if (view
->resource
->target
== PIPE_BUFFER
)
649 tex
= (struct si_texture
*)view
->resource
;
651 si_check_render_feedback_texture(sctx
, tex
,
654 view
->u
.tex
.first_layer
,
655 view
->u
.tex
.last_layer
);
659 static void si_check_render_feedback_resident_textures(struct si_context
*sctx
)
661 util_dynarray_foreach(&sctx
->resident_tex_handles
,
662 struct si_texture_handle
*, tex_handle
) {
663 struct pipe_sampler_view
*view
;
664 struct si_texture
*tex
;
666 view
= (*tex_handle
)->view
;
667 if (view
->texture
->target
== PIPE_BUFFER
)
670 tex
= (struct si_texture
*)view
->texture
;
672 si_check_render_feedback_texture(sctx
, tex
,
673 view
->u
.tex
.first_level
,
674 view
->u
.tex
.last_level
,
675 view
->u
.tex
.first_layer
,
676 view
->u
.tex
.last_layer
);
680 static void si_check_render_feedback_resident_images(struct si_context
*sctx
)
682 util_dynarray_foreach(&sctx
->resident_img_handles
,
683 struct si_image_handle
*, img_handle
) {
684 struct pipe_image_view
*view
;
685 struct si_texture
*tex
;
687 view
= &(*img_handle
)->view
;
688 if (view
->resource
->target
== PIPE_BUFFER
)
691 tex
= (struct si_texture
*)view
->resource
;
693 si_check_render_feedback_texture(sctx
, tex
,
696 view
->u
.tex
.first_layer
,
697 view
->u
.tex
.last_layer
);
701 static void si_check_render_feedback(struct si_context
*sctx
)
703 if (!sctx
->need_check_render_feedback
)
706 /* There is no render feedback if color writes are disabled.
707 * (e.g. a pixel shader with image stores)
709 if (!si_get_total_colormask(sctx
))
712 for (int i
= 0; i
< SI_NUM_SHADERS
; ++i
) {
713 si_check_render_feedback_images(sctx
, &sctx
->images
[i
]);
714 si_check_render_feedback_textures(sctx
, &sctx
->samplers
[i
]);
717 si_check_render_feedback_resident_images(sctx
);
718 si_check_render_feedback_resident_textures(sctx
);
720 sctx
->need_check_render_feedback
= false;
723 static void si_decompress_resident_textures(struct si_context
*sctx
)
725 util_dynarray_foreach(&sctx
->resident_tex_needs_color_decompress
,
726 struct si_texture_handle
*, tex_handle
) {
727 struct pipe_sampler_view
*view
= (*tex_handle
)->view
;
728 struct si_texture
*tex
= (struct si_texture
*)view
->texture
;
730 si_decompress_color_texture(sctx
, tex
, view
->u
.tex
.first_level
,
731 view
->u
.tex
.last_level
);
734 util_dynarray_foreach(&sctx
->resident_tex_needs_depth_decompress
,
735 struct si_texture_handle
*, tex_handle
) {
736 struct pipe_sampler_view
*view
= (*tex_handle
)->view
;
737 struct si_sampler_view
*sview
= (struct si_sampler_view
*)view
;
738 struct si_texture
*tex
= (struct si_texture
*)view
->texture
;
740 si_decompress_depth(sctx
, tex
,
741 sview
->is_stencil_sampler
? PIPE_MASK_S
: PIPE_MASK_Z
,
742 view
->u
.tex
.first_level
, view
->u
.tex
.last_level
,
743 0, util_max_layer(&tex
->buffer
.b
.b
, view
->u
.tex
.first_level
));
747 static void si_decompress_resident_images(struct si_context
*sctx
)
749 util_dynarray_foreach(&sctx
->resident_img_needs_color_decompress
,
750 struct si_image_handle
*, img_handle
) {
751 struct pipe_image_view
*view
= &(*img_handle
)->view
;
752 struct si_texture
*tex
= (struct si_texture
*)view
->resource
;
754 si_decompress_color_texture(sctx
, tex
, view
->u
.tex
.level
,
759 void si_decompress_textures(struct si_context
*sctx
, unsigned shader_mask
)
761 unsigned compressed_colortex_counter
, mask
;
763 if (sctx
->blitter
->running
)
766 /* Update the compressed_colortex_mask if necessary. */
767 compressed_colortex_counter
= p_atomic_read(&sctx
->screen
->compressed_colortex_counter
);
768 if (compressed_colortex_counter
!= sctx
->last_compressed_colortex_counter
) {
769 sctx
->last_compressed_colortex_counter
= compressed_colortex_counter
;
770 si_update_needs_color_decompress_masks(sctx
);
773 /* Decompress color & depth textures if needed. */
774 mask
= sctx
->shader_needs_decompress_mask
& shader_mask
;
776 unsigned i
= u_bit_scan(&mask
);
778 if (sctx
->samplers
[i
].needs_depth_decompress_mask
) {
779 si_decompress_sampler_depth_textures(sctx
, &sctx
->samplers
[i
]);
781 if (sctx
->samplers
[i
].needs_color_decompress_mask
) {
782 si_decompress_sampler_color_textures(sctx
, &sctx
->samplers
[i
]);
784 if (sctx
->images
[i
].needs_color_decompress_mask
) {
785 si_decompress_image_color_textures(sctx
, &sctx
->images
[i
]);
789 if (shader_mask
& u_bit_consecutive(0, SI_NUM_GRAPHICS_SHADERS
)) {
790 if (sctx
->uses_bindless_samplers
)
791 si_decompress_resident_textures(sctx
);
792 if (sctx
->uses_bindless_images
)
793 si_decompress_resident_images(sctx
);
795 if (sctx
->ps_uses_fbfetch
) {
796 struct pipe_surface
*cb0
= sctx
->framebuffer
.state
.cbufs
[0];
797 si_decompress_color_texture(sctx
,
798 (struct si_texture
*)cb0
->texture
,
799 cb0
->u
.tex
.first_layer
,
800 cb0
->u
.tex
.last_layer
);
803 si_check_render_feedback(sctx
);
804 } else if (shader_mask
& (1 << PIPE_SHADER_COMPUTE
)) {
805 if (sctx
->cs_shader_state
.program
->uses_bindless_samplers
)
806 si_decompress_resident_textures(sctx
);
807 if (sctx
->cs_shader_state
.program
->uses_bindless_images
)
808 si_decompress_resident_images(sctx
);
812 /* Helper for decompressing a portion of a color or depth resource before
813 * blitting if any decompression is needed.
814 * The driver doesn't decompress resources automatically while u_blitter is
816 static void si_decompress_subresource(struct pipe_context
*ctx
,
817 struct pipe_resource
*tex
,
818 unsigned planes
, unsigned level
,
819 unsigned first_layer
, unsigned last_layer
)
821 struct si_context
*sctx
= (struct si_context
*)ctx
;
822 struct si_texture
*stex
= (struct si_texture
*)tex
;
824 if (stex
->db_compatible
) {
825 planes
&= PIPE_MASK_Z
| PIPE_MASK_S
;
827 if (!stex
->surface
.has_stencil
)
828 planes
&= ~PIPE_MASK_S
;
830 /* If we've rendered into the framebuffer and it's a blitting
831 * source, make sure the decompression pass is invoked
832 * by dirtying the framebuffer.
834 if (sctx
->framebuffer
.state
.zsbuf
&&
835 sctx
->framebuffer
.state
.zsbuf
->u
.tex
.level
== level
&&
836 sctx
->framebuffer
.state
.zsbuf
->texture
== tex
)
837 si_update_fb_dirtiness_after_rendering(sctx
);
839 si_decompress_depth(sctx
, stex
, planes
,
841 first_layer
, last_layer
);
842 } else if (stex
->surface
.fmask_size
|| stex
->cmask_buffer
|| stex
->dcc_offset
) {
843 /* If we've rendered into the framebuffer and it's a blitting
844 * source, make sure the decompression pass is invoked
845 * by dirtying the framebuffer.
847 for (unsigned i
= 0; i
< sctx
->framebuffer
.state
.nr_cbufs
; i
++) {
848 if (sctx
->framebuffer
.state
.cbufs
[i
] &&
849 sctx
->framebuffer
.state
.cbufs
[i
]->u
.tex
.level
== level
&&
850 sctx
->framebuffer
.state
.cbufs
[i
]->texture
== tex
) {
851 si_update_fb_dirtiness_after_rendering(sctx
);
856 si_blit_decompress_color(sctx
, stex
, level
, level
,
857 first_layer
, last_layer
, false);
861 struct texture_orig_info
{
871 void si_resource_copy_region(struct pipe_context
*ctx
,
872 struct pipe_resource
*dst
,
874 unsigned dstx
, unsigned dsty
, unsigned dstz
,
875 struct pipe_resource
*src
,
877 const struct pipe_box
*src_box
)
879 struct si_context
*sctx
= (struct si_context
*)ctx
;
880 struct si_texture
*ssrc
= (struct si_texture
*)src
;
881 struct si_texture
*sdst
= (struct si_texture
*)dst
;
882 struct pipe_surface
*dst_view
, dst_templ
;
883 struct pipe_sampler_view src_templ
, *src_view
;
884 unsigned dst_width
, dst_height
, src_width0
, src_height0
;
885 unsigned dst_width0
, dst_height0
, src_force_level
= 0;
886 struct pipe_box sbox
, dstbox
;
888 /* Handle buffers first. */
889 if (dst
->target
== PIPE_BUFFER
&& src
->target
== PIPE_BUFFER
) {
890 si_copy_buffer(sctx
, dst
, src
, dstx
, src_box
->x
, src_box
->width
);
894 if (!util_format_is_compressed(src
->format
) &&
895 !util_format_is_compressed(dst
->format
) &&
896 !util_format_is_depth_or_stencil(src
->format
) &&
897 src
->nr_samples
<= 1 &&
899 !(dst
->target
!= src
->target
&&
900 (src
->target
== PIPE_TEXTURE_1D_ARRAY
|| dst
->target
== PIPE_TEXTURE_1D_ARRAY
))) {
901 si_compute_copy_image(sctx
, dst
, dst_level
, src
, src_level
, dstx
, dsty
, dstz
, src_box
);
905 assert(u_max_sample(dst
) == u_max_sample(src
));
907 /* The driver doesn't decompress resources automatically while
908 * u_blitter is rendering. */
909 si_decompress_subresource(ctx
, src
, PIPE_MASK_RGBAZS
, src_level
,
910 src_box
->z
, src_box
->z
+ src_box
->depth
- 1);
912 dst_width
= u_minify(dst
->width0
, dst_level
);
913 dst_height
= u_minify(dst
->height0
, dst_level
);
914 dst_width0
= dst
->width0
;
915 dst_height0
= dst
->height0
;
916 src_width0
= src
->width0
;
917 src_height0
= src
->height0
;
919 util_blitter_default_dst_texture(&dst_templ
, dst
, dst_level
, dstz
);
920 util_blitter_default_src_texture(sctx
->blitter
, &src_templ
, src
, src_level
);
922 if (util_format_is_compressed(src
->format
) ||
923 util_format_is_compressed(dst
->format
)) {
924 unsigned blocksize
= ssrc
->surface
.bpe
;
927 src_templ
.format
= PIPE_FORMAT_R16G16B16A16_UINT
; /* 64-bit block */
929 src_templ
.format
= PIPE_FORMAT_R32G32B32A32_UINT
; /* 128-bit block */
930 dst_templ
.format
= src_templ
.format
;
932 dst_width
= util_format_get_nblocksx(dst
->format
, dst_width
);
933 dst_height
= util_format_get_nblocksy(dst
->format
, dst_height
);
934 dst_width0
= util_format_get_nblocksx(dst
->format
, dst_width0
);
935 dst_height0
= util_format_get_nblocksy(dst
->format
, dst_height0
);
936 src_width0
= util_format_get_nblocksx(src
->format
, src_width0
);
937 src_height0
= util_format_get_nblocksy(src
->format
, src_height0
);
939 dstx
= util_format_get_nblocksx(dst
->format
, dstx
);
940 dsty
= util_format_get_nblocksy(dst
->format
, dsty
);
942 sbox
.x
= util_format_get_nblocksx(src
->format
, src_box
->x
);
943 sbox
.y
= util_format_get_nblocksy(src
->format
, src_box
->y
);
945 sbox
.width
= util_format_get_nblocksx(src
->format
, src_box
->width
);
946 sbox
.height
= util_format_get_nblocksy(src
->format
, src_box
->height
);
947 sbox
.depth
= src_box
->depth
;
950 src_force_level
= src_level
;
951 } else if (!util_blitter_is_copy_supported(sctx
->blitter
, dst
, src
)) {
952 if (util_format_is_subsampled_422(src
->format
)) {
953 src_templ
.format
= PIPE_FORMAT_R8G8B8A8_UINT
;
954 dst_templ
.format
= PIPE_FORMAT_R8G8B8A8_UINT
;
956 dst_width
= util_format_get_nblocksx(dst
->format
, dst_width
);
957 dst_width0
= util_format_get_nblocksx(dst
->format
, dst_width0
);
958 src_width0
= util_format_get_nblocksx(src
->format
, src_width0
);
960 dstx
= util_format_get_nblocksx(dst
->format
, dstx
);
963 sbox
.x
= util_format_get_nblocksx(src
->format
, src_box
->x
);
964 sbox
.width
= util_format_get_nblocksx(src
->format
, src_box
->width
);
967 unsigned blocksize
= ssrc
->surface
.bpe
;
971 dst_templ
.format
= PIPE_FORMAT_R8_UNORM
;
972 src_templ
.format
= PIPE_FORMAT_R8_UNORM
;
975 dst_templ
.format
= PIPE_FORMAT_R8G8_UNORM
;
976 src_templ
.format
= PIPE_FORMAT_R8G8_UNORM
;
979 dst_templ
.format
= PIPE_FORMAT_R8G8B8A8_UNORM
;
980 src_templ
.format
= PIPE_FORMAT_R8G8B8A8_UNORM
;
983 dst_templ
.format
= PIPE_FORMAT_R16G16B16A16_UINT
;
984 src_templ
.format
= PIPE_FORMAT_R16G16B16A16_UINT
;
987 dst_templ
.format
= PIPE_FORMAT_R32G32B32A32_UINT
;
988 src_templ
.format
= PIPE_FORMAT_R32G32B32A32_UINT
;
991 fprintf(stderr
, "Unhandled format %s with blocksize %u\n",
992 util_format_short_name(src
->format
), blocksize
);
998 /* SNORM8 blitting has precision issues on some chips. Use the SINT
999 * equivalent instead, which doesn't force DCC decompression.
1000 * Note that some chips avoid this issue by using SDMA.
1002 if (util_format_is_snorm8(dst_templ
.format
)) {
1003 dst_templ
.format
= src_templ
.format
=
1004 util_format_snorm8_to_sint8(dst_templ
.format
);
1007 vi_disable_dcc_if_incompatible_format(sctx
, dst
, dst_level
,
1009 vi_disable_dcc_if_incompatible_format(sctx
, src
, src_level
,
1012 /* Initialize the surface. */
1013 dst_view
= si_create_surface_custom(ctx
, dst
, &dst_templ
,
1014 dst_width0
, dst_height0
,
1015 dst_width
, dst_height
);
1017 /* Initialize the sampler view. */
1018 src_view
= si_create_sampler_view_custom(ctx
, src
, &src_templ
,
1019 src_width0
, src_height0
,
1022 u_box_3d(dstx
, dsty
, dstz
, abs(src_box
->width
), abs(src_box
->height
),
1023 abs(src_box
->depth
), &dstbox
);
1026 si_blitter_begin(sctx
, SI_COPY
);
1027 util_blitter_blit_generic(sctx
->blitter
, dst_view
, &dstbox
,
1028 src_view
, src_box
, src_width0
, src_height0
,
1029 PIPE_MASK_RGBAZS
, PIPE_TEX_FILTER_NEAREST
, NULL
,
1031 si_blitter_end(sctx
);
1033 pipe_surface_reference(&dst_view
, NULL
);
1034 pipe_sampler_view_reference(&src_view
, NULL
);
1037 static void si_do_CB_resolve(struct si_context
*sctx
,
1038 const struct pipe_blit_info
*info
,
1039 struct pipe_resource
*dst
,
1040 unsigned dst_level
, unsigned dst_z
,
1041 enum pipe_format format
)
1043 /* Required before and after CB_RESOLVE. */
1044 sctx
->flags
|= SI_CONTEXT_FLUSH_AND_INV_CB
;
1046 si_blitter_begin(sctx
, SI_COLOR_RESOLVE
|
1047 (info
->render_condition_enable
? 0 : SI_DISABLE_RENDER_COND
));
1048 util_blitter_custom_resolve_color(sctx
->blitter
, dst
, dst_level
, dst_z
,
1049 info
->src
.resource
, info
->src
.box
.z
,
1050 ~0, sctx
->custom_blend_resolve
,
1052 si_blitter_end(sctx
);
1054 /* Flush caches for possible texturing. */
1055 si_make_CB_shader_coherent(sctx
, 1, false, true /* no DCC */);
1058 static bool do_hardware_msaa_resolve(struct pipe_context
*ctx
,
1059 const struct pipe_blit_info
*info
)
1061 struct si_context
*sctx
= (struct si_context
*)ctx
;
1062 struct si_texture
*src
= (struct si_texture
*)info
->src
.resource
;
1063 struct si_texture
*dst
= (struct si_texture
*)info
->dst
.resource
;
1064 MAYBE_UNUSED
struct si_texture
*stmp
;
1065 unsigned dst_width
= u_minify(info
->dst
.resource
->width0
, info
->dst
.level
);
1066 unsigned dst_height
= u_minify(info
->dst
.resource
->height0
, info
->dst
.level
);
1067 enum pipe_format format
= info
->src
.format
;
1068 struct pipe_resource
*tmp
, templ
;
1069 struct pipe_blit_info blit
;
1071 /* Check basic requirements for hw resolve. */
1072 if (!(info
->src
.resource
->nr_samples
> 1 &&
1073 info
->dst
.resource
->nr_samples
<= 1 &&
1074 !util_format_is_pure_integer(format
) &&
1075 !util_format_is_depth_or_stencil(format
) &&
1076 util_max_layer(info
->src
.resource
, 0) == 0))
1079 /* Hardware MSAA resolve doesn't work if SPI format = NORM16_ABGR and
1080 * the format is R16G16. Use R16A16, which does work.
1082 if (format
== PIPE_FORMAT_R16G16_UNORM
)
1083 format
= PIPE_FORMAT_R16A16_UNORM
;
1084 if (format
== PIPE_FORMAT_R16G16_SNORM
)
1085 format
= PIPE_FORMAT_R16A16_SNORM
;
1087 /* Check the remaining requirements for hw resolve. */
1088 if (util_max_layer(info
->dst
.resource
, info
->dst
.level
) == 0 &&
1089 !info
->scissor_enable
&&
1090 (info
->mask
& PIPE_MASK_RGBA
) == PIPE_MASK_RGBA
&&
1091 util_is_format_compatible(util_format_description(info
->src
.format
),
1092 util_format_description(info
->dst
.format
)) &&
1093 dst_width
== info
->src
.resource
->width0
&&
1094 dst_height
== info
->src
.resource
->height0
&&
1095 info
->dst
.box
.x
== 0 &&
1096 info
->dst
.box
.y
== 0 &&
1097 info
->dst
.box
.width
== dst_width
&&
1098 info
->dst
.box
.height
== dst_height
&&
1099 info
->dst
.box
.depth
== 1 &&
1100 info
->src
.box
.x
== 0 &&
1101 info
->src
.box
.y
== 0 &&
1102 info
->src
.box
.width
== dst_width
&&
1103 info
->src
.box
.height
== dst_height
&&
1104 info
->src
.box
.depth
== 1 &&
1105 !dst
->surface
.is_linear
&&
1106 (!dst
->cmask_buffer
|| !dst
->dirty_level_mask
)) { /* dst cannot be fast-cleared */
1107 /* Check the last constraint. */
1108 if (src
->surface
.micro_tile_mode
!= dst
->surface
.micro_tile_mode
) {
1109 /* The next fast clear will switch to this mode to
1110 * get direct hw resolve next time if the mode is
1113 * TODO-GFX10: This does not work in GFX10 because MSAA
1114 * is restricted to 64KB_R_X and 64KB_Z_X swizzle modes.
1115 * In some cases we could change the swizzle of the
1116 * destination texture instead, but the more general
1117 * solution is to implement compute shader resolve.
1119 src
->last_msaa_resolve_target_micro_mode
=
1120 dst
->surface
.micro_tile_mode
;
1121 goto resolve_to_temp
;
1124 /* Resolving into a surface with DCC is unsupported. Since
1125 * it's being overwritten anyway, clear it to uncompressed.
1126 * This is still the fastest codepath even with this clear.
1128 if (vi_dcc_enabled(dst
, info
->dst
.level
)) {
1129 /* TODO: Implement per-level DCC clears for GFX9. */
1130 if (sctx
->chip_class
>= GFX9
&&
1131 info
->dst
.resource
->last_level
!= 0)
1132 goto resolve_to_temp
;
1134 /* This can happen with mipmapping. */
1135 if (sctx
->chip_class
== GFX8
&&
1136 !dst
->surface
.u
.legacy
.level
[info
->dst
.level
].dcc_fast_clear_size
)
1137 goto resolve_to_temp
;
1139 vi_dcc_clear_level(sctx
, dst
, info
->dst
.level
,
1141 dst
->dirty_level_mask
&= ~(1 << info
->dst
.level
);
1144 /* Resolve directly from src to dst. */
1145 si_do_CB_resolve(sctx
, info
, info
->dst
.resource
,
1146 info
->dst
.level
, info
->dst
.box
.z
, format
);
1151 /* Shader-based resolve is VERY SLOW. Instead, resolve into
1152 * a temporary texture and blit.
1154 memset(&templ
, 0, sizeof(templ
));
1155 templ
.target
= PIPE_TEXTURE_2D
;
1156 templ
.format
= info
->src
.resource
->format
;
1157 templ
.width0
= info
->src
.resource
->width0
;
1158 templ
.height0
= info
->src
.resource
->height0
;
1160 templ
.array_size
= 1;
1161 templ
.usage
= PIPE_USAGE_DEFAULT
;
1162 templ
.flags
= SI_RESOURCE_FLAG_FORCE_MSAA_TILING
|
1163 SI_RESOURCE_FLAG_DISABLE_DCC
;
1165 /* The src and dst microtile modes must be the same. */
1166 if (src
->surface
.micro_tile_mode
== RADEON_MICRO_MODE_DISPLAY
)
1167 templ
.bind
= PIPE_BIND_SCANOUT
;
1171 tmp
= ctx
->screen
->resource_create(ctx
->screen
, &templ
);
1174 stmp
= (struct si_texture
*)tmp
;
1176 assert(!stmp
->surface
.is_linear
);
1177 assert(src
->surface
.micro_tile_mode
== stmp
->surface
.micro_tile_mode
);
1180 si_do_CB_resolve(sctx
, info
, tmp
, 0, 0, format
);
1184 blit
.src
.resource
= tmp
;
1187 si_blitter_begin(sctx
, SI_BLIT
|
1188 (info
->render_condition_enable
? 0 : SI_DISABLE_RENDER_COND
));
1189 util_blitter_blit(sctx
->blitter
, &blit
);
1190 si_blitter_end(sctx
);
1192 pipe_resource_reference(&tmp
, NULL
);
1196 static void si_blit(struct pipe_context
*ctx
,
1197 const struct pipe_blit_info
*info
)
1199 struct si_context
*sctx
= (struct si_context
*)ctx
;
1200 struct si_texture
*dst
= (struct si_texture
*)info
->dst
.resource
;
1202 if (do_hardware_msaa_resolve(ctx
, info
)) {
1206 /* Using SDMA for copying to a linear texture in GTT is much faster.
1207 * This improves DRI PRIME performance.
1209 * resource_copy_region can't do this yet, because dma_copy calls it
1210 * on failure (recursion).
1212 if (dst
->surface
.is_linear
&&
1214 util_can_blit_via_copy_region(info
, false)) {
1215 sctx
->dma_copy(ctx
, info
->dst
.resource
, info
->dst
.level
,
1216 info
->dst
.box
.x
, info
->dst
.box
.y
,
1218 info
->src
.resource
, info
->src
.level
,
1223 assert(util_blitter_is_blit_supported(sctx
->blitter
, info
));
1225 /* The driver doesn't decompress resources automatically while
1226 * u_blitter is rendering. */
1227 vi_disable_dcc_if_incompatible_format(sctx
, info
->src
.resource
,
1230 vi_disable_dcc_if_incompatible_format(sctx
, info
->dst
.resource
,
1233 si_decompress_subresource(ctx
, info
->src
.resource
, PIPE_MASK_RGBAZS
,
1236 info
->src
.box
.z
+ info
->src
.box
.depth
- 1);
1238 if (sctx
->screen
->debug_flags
& DBG(FORCE_DMA
) &&
1239 util_try_blit_via_copy_region(ctx
, info
))
1242 si_blitter_begin(sctx
, SI_BLIT
|
1243 (info
->render_condition_enable
? 0 : SI_DISABLE_RENDER_COND
));
1244 util_blitter_blit(sctx
->blitter
, info
);
1245 si_blitter_end(sctx
);
1248 static boolean
si_generate_mipmap(struct pipe_context
*ctx
,
1249 struct pipe_resource
*tex
,
1250 enum pipe_format format
,
1251 unsigned base_level
, unsigned last_level
,
1252 unsigned first_layer
, unsigned last_layer
)
1254 struct si_context
*sctx
= (struct si_context
*)ctx
;
1255 struct si_texture
*stex
= (struct si_texture
*)tex
;
1257 if (!util_blitter_is_copy_supported(sctx
->blitter
, tex
, tex
))
1260 /* The driver doesn't decompress resources automatically while
1261 * u_blitter is rendering. */
1262 vi_disable_dcc_if_incompatible_format(sctx
, tex
, base_level
,
1264 si_decompress_subresource(ctx
, tex
, PIPE_MASK_RGBAZS
,
1265 base_level
, first_layer
, last_layer
);
1267 /* Clear dirty_level_mask for the levels that will be overwritten. */
1268 assert(base_level
< last_level
);
1269 stex
->dirty_level_mask
&= ~u_bit_consecutive(base_level
+ 1,
1270 last_level
- base_level
);
1272 sctx
->generate_mipmap_for_depth
= stex
->is_depth
;
1274 si_blitter_begin(sctx
, SI_BLIT
| SI_DISABLE_RENDER_COND
);
1275 util_blitter_generate_mipmap(sctx
->blitter
, tex
, format
,
1276 base_level
, last_level
,
1277 first_layer
, last_layer
);
1278 si_blitter_end(sctx
);
1280 sctx
->generate_mipmap_for_depth
= false;
1284 static void si_flush_resource(struct pipe_context
*ctx
,
1285 struct pipe_resource
*res
)
1287 struct si_context
*sctx
= (struct si_context
*)ctx
;
1288 struct si_texture
*tex
= (struct si_texture
*)res
;
1290 assert(res
->target
!= PIPE_BUFFER
);
1291 assert(!tex
->dcc_separate_buffer
|| tex
->dcc_gather_statistics
);
1293 /* st/dri calls flush twice per frame (not a bug), this prevents double
1295 if (tex
->dcc_separate_buffer
&& !tex
->separate_dcc_dirty
)
1298 if (!tex
->is_depth
&& (tex
->cmask_buffer
|| tex
->dcc_offset
)) {
1299 si_blit_decompress_color(sctx
, tex
, 0, res
->last_level
,
1300 0, util_max_layer(res
, 0),
1301 tex
->dcc_separate_buffer
!= NULL
);
1303 if (tex
->display_dcc_offset
)
1304 si_retile_dcc(sctx
, tex
);
1307 /* Always do the analysis even if DCC is disabled at the moment. */
1308 if (tex
->dcc_gather_statistics
) {
1309 bool separate_dcc_dirty
= tex
->separate_dcc_dirty
;
1311 /* If the color buffer hasn't been unbound and fast clear hasn't
1312 * been used, separate_dcc_dirty is false, but there may have been
1313 * new rendering. Check if the color buffer is bound and assume
1316 * Note that DRI2 never unbinds window colorbuffers, which means
1317 * the DCC pipeline statistics query would never be re-set and would
1318 * keep adding new results until all free memory is exhausted if we
1321 if (!separate_dcc_dirty
) {
1322 for (unsigned i
= 0; i
< sctx
->framebuffer
.state
.nr_cbufs
; i
++) {
1323 if (sctx
->framebuffer
.state
.cbufs
[i
] &&
1324 sctx
->framebuffer
.state
.cbufs
[i
]->texture
== res
) {
1325 separate_dcc_dirty
= true;
1331 if (separate_dcc_dirty
) {
1332 tex
->separate_dcc_dirty
= false;
1333 vi_separate_dcc_process_and_reset_stats(ctx
, tex
);
1338 void si_decompress_dcc(struct si_context
*sctx
, struct si_texture
*tex
)
1340 /* If graphics is disabled, we can't decompress DCC, but it shouldn't
1341 * be compressed either. The caller should simply discard it.
1343 if (!tex
->dcc_offset
|| !sctx
->has_graphics
)
1346 si_blit_decompress_color(sctx
, tex
, 0, tex
->buffer
.b
.b
.last_level
,
1347 0, util_max_layer(&tex
->buffer
.b
.b
, 0),
1351 void si_init_blit_functions(struct si_context
*sctx
)
1353 sctx
->b
.resource_copy_region
= si_resource_copy_region
;
1355 if (sctx
->has_graphics
) {
1356 sctx
->b
.blit
= si_blit
;
1357 sctx
->b
.flush_resource
= si_flush_resource
;
1358 sctx
->b
.generate_mipmap
= si_generate_mipmap
;