2 * Copyright © 2016 Red Hat
3 * based on intel anv code:
4 * Copyright © 2015 Intel Corporation
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 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * 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 NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
22 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
26 #include "radv_meta.h"
34 radv_meta_save(struct radv_meta_saved_state
*state
,
35 struct radv_cmd_buffer
*cmd_buffer
, uint32_t flags
)
37 VkPipelineBindPoint bind_point
=
38 flags
& RADV_META_SAVE_GRAPHICS_PIPELINE
?
39 VK_PIPELINE_BIND_POINT_GRAPHICS
:
40 VK_PIPELINE_BIND_POINT_COMPUTE
;
41 struct radv_descriptor_state
*descriptors_state
=
42 radv_get_descriptors_state(cmd_buffer
, bind_point
);
44 assert(flags
& (RADV_META_SAVE_GRAPHICS_PIPELINE
|
45 RADV_META_SAVE_COMPUTE_PIPELINE
));
49 if (state
->flags
& RADV_META_SAVE_GRAPHICS_PIPELINE
) {
50 assert(!(state
->flags
& RADV_META_SAVE_COMPUTE_PIPELINE
));
52 state
->old_pipeline
= cmd_buffer
->state
.pipeline
;
54 /* Save all viewports. */
55 state
->viewport
.count
= cmd_buffer
->state
.dynamic
.viewport
.count
;
56 typed_memcpy(state
->viewport
.viewports
,
57 cmd_buffer
->state
.dynamic
.viewport
.viewports
,
60 /* Save all scissors. */
61 state
->scissor
.count
= cmd_buffer
->state
.dynamic
.scissor
.count
;
62 typed_memcpy(state
->scissor
.scissors
,
63 cmd_buffer
->state
.dynamic
.scissor
.scissors
,
66 /* The most common meta operations all want to have the
67 * viewport reset and any scissors disabled. The rest of the
68 * dynamic state should have no effect.
70 cmd_buffer
->state
.dynamic
.viewport
.count
= 0;
71 cmd_buffer
->state
.dynamic
.scissor
.count
= 0;
72 cmd_buffer
->state
.dirty
|= 1 << VK_DYNAMIC_STATE_VIEWPORT
|
73 1 << VK_DYNAMIC_STATE_SCISSOR
;
76 if (state
->flags
& RADV_META_SAVE_COMPUTE_PIPELINE
) {
77 assert(!(state
->flags
& RADV_META_SAVE_GRAPHICS_PIPELINE
));
79 state
->old_pipeline
= cmd_buffer
->state
.compute_pipeline
;
82 if (state
->flags
& RADV_META_SAVE_DESCRIPTORS
) {
83 state
->old_descriptor_set0
= descriptors_state
->sets
[0];
84 if (!state
->old_descriptor_set0
)
85 state
->flags
&= ~RADV_META_SAVE_DESCRIPTORS
;
88 if (state
->flags
& RADV_META_SAVE_CONSTANTS
) {
89 memcpy(state
->push_constants
, cmd_buffer
->push_constants
,
90 MAX_PUSH_CONSTANTS_SIZE
);
93 if (state
->flags
& RADV_META_SAVE_PASS
) {
94 state
->pass
= cmd_buffer
->state
.pass
;
95 state
->subpass
= cmd_buffer
->state
.subpass
;
96 state
->framebuffer
= cmd_buffer
->state
.framebuffer
;
97 state
->attachments
= cmd_buffer
->state
.attachments
;
98 state
->render_area
= cmd_buffer
->state
.render_area
;
103 radv_meta_restore(const struct radv_meta_saved_state
*state
,
104 struct radv_cmd_buffer
*cmd_buffer
)
106 VkPipelineBindPoint bind_point
=
107 state
->flags
& RADV_META_SAVE_GRAPHICS_PIPELINE
?
108 VK_PIPELINE_BIND_POINT_GRAPHICS
:
109 VK_PIPELINE_BIND_POINT_COMPUTE
;
111 if (state
->flags
& RADV_META_SAVE_GRAPHICS_PIPELINE
) {
112 radv_CmdBindPipeline(radv_cmd_buffer_to_handle(cmd_buffer
),
113 VK_PIPELINE_BIND_POINT_GRAPHICS
,
114 radv_pipeline_to_handle(state
->old_pipeline
));
116 cmd_buffer
->state
.dirty
|= RADV_CMD_DIRTY_PIPELINE
;
118 /* Restore all viewports. */
119 cmd_buffer
->state
.dynamic
.viewport
.count
= state
->viewport
.count
;
120 typed_memcpy(cmd_buffer
->state
.dynamic
.viewport
.viewports
,
121 state
->viewport
.viewports
,
124 /* Restore all scissors. */
125 cmd_buffer
->state
.dynamic
.scissor
.count
= state
->scissor
.count
;
126 typed_memcpy(cmd_buffer
->state
.dynamic
.scissor
.scissors
,
127 state
->scissor
.scissors
,
130 cmd_buffer
->state
.dirty
|= 1 << VK_DYNAMIC_STATE_VIEWPORT
|
131 1 << VK_DYNAMIC_STATE_SCISSOR
;
134 if (state
->flags
& RADV_META_SAVE_COMPUTE_PIPELINE
) {
135 radv_CmdBindPipeline(radv_cmd_buffer_to_handle(cmd_buffer
),
136 VK_PIPELINE_BIND_POINT_COMPUTE
,
137 radv_pipeline_to_handle(state
->old_pipeline
));
140 if (state
->flags
& RADV_META_SAVE_DESCRIPTORS
) {
141 radv_set_descriptor_set(cmd_buffer
, bind_point
,
142 state
->old_descriptor_set0
, 0);
145 if (state
->flags
& RADV_META_SAVE_CONSTANTS
) {
146 memcpy(cmd_buffer
->push_constants
, state
->push_constants
,
147 MAX_PUSH_CONSTANTS_SIZE
);
148 cmd_buffer
->push_constant_stages
|= VK_SHADER_STAGE_COMPUTE_BIT
;
150 if (state
->flags
& RADV_META_SAVE_GRAPHICS_PIPELINE
) {
151 cmd_buffer
->push_constant_stages
|= VK_SHADER_STAGE_ALL_GRAPHICS
;
155 if (state
->flags
& RADV_META_SAVE_PASS
) {
156 cmd_buffer
->state
.pass
= state
->pass
;
157 cmd_buffer
->state
.subpass
= state
->subpass
;
158 cmd_buffer
->state
.framebuffer
= state
->framebuffer
;
159 cmd_buffer
->state
.attachments
= state
->attachments
;
160 cmd_buffer
->state
.render_area
= state
->render_area
;
162 cmd_buffer
->state
.dirty
|= RADV_CMD_DIRTY_FRAMEBUFFER
;
167 radv_meta_get_view_type(const struct radv_image
*image
)
169 switch (image
->type
) {
170 case VK_IMAGE_TYPE_1D
: return VK_IMAGE_VIEW_TYPE_1D
;
171 case VK_IMAGE_TYPE_2D
: return VK_IMAGE_VIEW_TYPE_2D
;
172 case VK_IMAGE_TYPE_3D
: return VK_IMAGE_VIEW_TYPE_3D
;
174 unreachable("bad VkImageViewType");
179 * When creating a destination VkImageView, this function provides the needed
180 * VkImageViewCreateInfo::subresourceRange::baseArrayLayer.
183 radv_meta_get_iview_layer(const struct radv_image
*dest_image
,
184 const VkImageSubresourceLayers
*dest_subresource
,
185 const VkOffset3D
*dest_offset
)
187 switch (dest_image
->type
) {
188 case VK_IMAGE_TYPE_1D
:
189 case VK_IMAGE_TYPE_2D
:
190 return dest_subresource
->baseArrayLayer
;
191 case VK_IMAGE_TYPE_3D
:
192 /* HACK: Vulkan does not allow attaching a 3D image to a framebuffer,
193 * but meta does it anyway. When doing so, we translate the
194 * destination's z offset into an array offset.
196 return dest_offset
->z
;
198 assert(!"bad VkImageType");
204 meta_alloc(void* _device
, size_t size
, size_t alignment
,
205 VkSystemAllocationScope allocationScope
)
207 struct radv_device
*device
= _device
;
208 return device
->alloc
.pfnAllocation(device
->alloc
.pUserData
, size
, alignment
,
209 VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
213 meta_realloc(void* _device
, void *original
, size_t size
, size_t alignment
,
214 VkSystemAllocationScope allocationScope
)
216 struct radv_device
*device
= _device
;
217 return device
->alloc
.pfnReallocation(device
->alloc
.pUserData
, original
,
219 VK_SYSTEM_ALLOCATION_SCOPE_DEVICE
);
223 meta_free(void* _device
, void *data
)
225 struct radv_device
*device
= _device
;
226 return device
->alloc
.pfnFree(device
->alloc
.pUserData
, data
);
230 radv_builtin_cache_path(char *path
)
232 char *xdg_cache_home
= getenv("XDG_CACHE_HOME");
233 const char *suffix
= "/radv_builtin_shaders";
234 const char *suffix2
= "/.cache/radv_builtin_shaders";
235 struct passwd pwd
, *result
;
236 char path2
[PATH_MAX
+ 1]; /* PATH_MAX is not a real max,but suffices here. */
238 if (xdg_cache_home
) {
240 if (strlen(xdg_cache_home
) + strlen(suffix
) > PATH_MAX
)
243 strcpy(path
, xdg_cache_home
);
244 strcat(path
, suffix
);
248 getpwuid_r(getuid(), &pwd
, path2
, PATH_MAX
- strlen(suffix2
), &result
);
252 strcpy(path
, pwd
.pw_dir
);
253 strcat(path
, "/.cache");
256 strcat(path
, suffix
);
261 radv_load_meta_pipeline(struct radv_device
*device
)
263 char path
[PATH_MAX
+ 1];
267 if (!radv_builtin_cache_path(path
))
270 int fd
= open(path
, O_RDONLY
);
275 data
= malloc(st
.st_size
);
278 if(read(fd
, data
, st
.st_size
) == -1)
281 return radv_pipeline_cache_load(&device
->meta_state
.cache
, data
, st
.st_size
);
289 radv_store_meta_pipeline(struct radv_device
*device
)
291 char path
[PATH_MAX
+ 1], path2
[PATH_MAX
+ 7];
295 if (!device
->meta_state
.cache
.modified
)
298 if (radv_GetPipelineCacheData(radv_device_to_handle(device
),
299 radv_pipeline_cache_to_handle(&device
->meta_state
.cache
),
303 if (!radv_builtin_cache_path(path
))
307 strcat(path2
, "XXXXXX");
308 int fd
= mkstemp(path2
);//open(path, O_WRONLY | O_CREAT, 0600);
315 if (radv_GetPipelineCacheData(radv_device_to_handle(device
),
316 radv_pipeline_cache_to_handle(&device
->meta_state
.cache
),
319 if(write(fd
, data
, size
) == -1)
330 radv_device_init_meta(struct radv_device
*device
)
334 memset(&device
->meta_state
, 0, sizeof(device
->meta_state
));
336 device
->meta_state
.alloc
= (VkAllocationCallbacks
) {
338 .pfnAllocation
= meta_alloc
,
339 .pfnReallocation
= meta_realloc
,
340 .pfnFree
= meta_free
,
343 device
->meta_state
.cache
.alloc
= device
->meta_state
.alloc
;
344 radv_pipeline_cache_init(&device
->meta_state
.cache
, device
);
345 bool loaded_cache
= radv_load_meta_pipeline(device
);
346 bool on_demand
= !loaded_cache
;
348 mtx_init(&device
->meta_state
.mtx
, mtx_plain
);
350 result
= radv_device_init_meta_clear_state(device
, on_demand
);
351 if (result
!= VK_SUCCESS
)
354 result
= radv_device_init_meta_resolve_state(device
, on_demand
);
355 if (result
!= VK_SUCCESS
)
358 result
= radv_device_init_meta_blit_state(device
, on_demand
);
359 if (result
!= VK_SUCCESS
)
362 result
= radv_device_init_meta_blit2d_state(device
, on_demand
);
363 if (result
!= VK_SUCCESS
)
366 result
= radv_device_init_meta_bufimage_state(device
);
367 if (result
!= VK_SUCCESS
)
370 result
= radv_device_init_meta_depth_decomp_state(device
, on_demand
);
371 if (result
!= VK_SUCCESS
)
372 goto fail_depth_decomp
;
374 result
= radv_device_init_meta_buffer_state(device
);
375 if (result
!= VK_SUCCESS
)
378 result
= radv_device_init_meta_query_state(device
, on_demand
);
379 if (result
!= VK_SUCCESS
)
382 result
= radv_device_init_meta_fast_clear_flush_state(device
, on_demand
);
383 if (result
!= VK_SUCCESS
)
384 goto fail_fast_clear
;
386 result
= radv_device_init_meta_resolve_compute_state(device
, on_demand
);
387 if (result
!= VK_SUCCESS
)
388 goto fail_resolve_compute
;
390 result
= radv_device_init_meta_resolve_fragment_state(device
, on_demand
);
391 if (result
!= VK_SUCCESS
)
392 goto fail_resolve_fragment
;
395 fail_resolve_fragment
:
396 radv_device_finish_meta_resolve_compute_state(device
);
397 fail_resolve_compute
:
398 radv_device_finish_meta_fast_clear_flush_state(device
);
400 radv_device_finish_meta_query_state(device
);
402 radv_device_finish_meta_buffer_state(device
);
404 radv_device_finish_meta_depth_decomp_state(device
);
406 radv_device_finish_meta_bufimage_state(device
);
408 radv_device_finish_meta_blit2d_state(device
);
410 radv_device_finish_meta_blit_state(device
);
412 radv_device_finish_meta_resolve_state(device
);
414 radv_device_finish_meta_clear_state(device
);
416 mtx_destroy(&device
->meta_state
.mtx
);
417 radv_pipeline_cache_finish(&device
->meta_state
.cache
);
422 radv_device_finish_meta(struct radv_device
*device
)
424 radv_device_finish_meta_clear_state(device
);
425 radv_device_finish_meta_resolve_state(device
);
426 radv_device_finish_meta_blit_state(device
);
427 radv_device_finish_meta_blit2d_state(device
);
428 radv_device_finish_meta_bufimage_state(device
);
429 radv_device_finish_meta_depth_decomp_state(device
);
430 radv_device_finish_meta_query_state(device
);
431 radv_device_finish_meta_buffer_state(device
);
432 radv_device_finish_meta_fast_clear_flush_state(device
);
433 radv_device_finish_meta_resolve_compute_state(device
);
434 radv_device_finish_meta_resolve_fragment_state(device
);
436 radv_store_meta_pipeline(device
);
437 radv_pipeline_cache_finish(&device
->meta_state
.cache
);
438 mtx_destroy(&device
->meta_state
.mtx
);
441 nir_ssa_def
*radv_meta_gen_rect_vertices_comp2(nir_builder
*vs_b
, nir_ssa_def
*comp2
)
444 nir_intrinsic_instr
*vertex_id
= nir_intrinsic_instr_create(vs_b
->shader
, nir_intrinsic_load_vertex_id_zero_base
);
445 nir_ssa_dest_init(&vertex_id
->instr
, &vertex_id
->dest
, 1, 32, "vertexid");
446 nir_builder_instr_insert(vs_b
, &vertex_id
->instr
);
448 /* vertex 0 - -1.0, -1.0 */
449 /* vertex 1 - -1.0, 1.0 */
450 /* vertex 2 - 1.0, -1.0 */
451 /* so channel 0 is vertex_id != 2 ? -1.0 : 1.0
452 channel 1 is vertex id != 1 ? -1.0 : 1.0 */
454 nir_ssa_def
*c0cmp
= nir_ine(vs_b
, &vertex_id
->dest
.ssa
,
455 nir_imm_int(vs_b
, 2));
456 nir_ssa_def
*c1cmp
= nir_ine(vs_b
, &vertex_id
->dest
.ssa
,
457 nir_imm_int(vs_b
, 1));
459 nir_ssa_def
*comp
[4];
460 comp
[0] = nir_bcsel(vs_b
, c0cmp
,
461 nir_imm_float(vs_b
, -1.0),
462 nir_imm_float(vs_b
, 1.0));
464 comp
[1] = nir_bcsel(vs_b
, c1cmp
,
465 nir_imm_float(vs_b
, -1.0),
466 nir_imm_float(vs_b
, 1.0));
468 comp
[3] = nir_imm_float(vs_b
, 1.0);
469 nir_ssa_def
*outvec
= nir_vec(vs_b
, comp
, 4);
474 nir_ssa_def
*radv_meta_gen_rect_vertices(nir_builder
*vs_b
)
476 return radv_meta_gen_rect_vertices_comp2(vs_b
, nir_imm_float(vs_b
, 0.0));
479 /* vertex shader that generates vertices */
481 radv_meta_build_nir_vs_generate_vertices(void)
483 const struct glsl_type
*vec4
= glsl_vec4_type();
486 nir_variable
*v_position
;
488 nir_builder_init_simple_shader(&b
, NULL
, MESA_SHADER_VERTEX
, NULL
);
489 b
.shader
->info
.name
= ralloc_strdup(b
.shader
, "meta_vs_gen_verts");
491 nir_ssa_def
*outvec
= radv_meta_gen_rect_vertices(&b
);
493 v_position
= nir_variable_create(b
.shader
, nir_var_shader_out
, vec4
,
495 v_position
->data
.location
= VARYING_SLOT_POS
;
497 nir_store_var(&b
, v_position
, outvec
, 0xf);
503 radv_meta_build_nir_fs_noop(void)
507 nir_builder_init_simple_shader(&b
, NULL
, MESA_SHADER_FRAGMENT
, NULL
);
508 b
.shader
->info
.name
= ralloc_asprintf(b
.shader
,
514 void radv_meta_build_resolve_shader_core(nir_builder
*b
,
517 nir_variable
*input_img
,
519 nir_ssa_def
*img_coord
)
521 /* do a txf_ms on each sample */
523 nir_if
*outer_if
= NULL
;
525 nir_ssa_def
*input_img_deref
= &nir_build_deref_var(b
, input_img
)->dest
.ssa
;
527 nir_tex_instr
*tex
= nir_tex_instr_create(b
->shader
, 3);
528 tex
->sampler_dim
= GLSL_SAMPLER_DIM_MS
;
529 tex
->op
= nir_texop_txf_ms
;
530 tex
->src
[0].src_type
= nir_tex_src_coord
;
531 tex
->src
[0].src
= nir_src_for_ssa(img_coord
);
532 tex
->src
[1].src_type
= nir_tex_src_ms_index
;
533 tex
->src
[1].src
= nir_src_for_ssa(nir_imm_int(b
, 0));
534 tex
->src
[2].src_type
= nir_tex_src_texture_deref
;
535 tex
->src
[2].src
= nir_src_for_ssa(input_img_deref
);
536 tex
->dest_type
= nir_type_float
;
537 tex
->is_array
= false;
538 tex
->coord_components
= 2;
540 nir_ssa_dest_init(&tex
->instr
, &tex
->dest
, 4, 32, "tex");
541 nir_builder_instr_insert(b
, &tex
->instr
);
543 tmp
= &tex
->dest
.ssa
;
545 if (!is_integer
&& samples
> 1) {
546 nir_tex_instr
*tex_all_same
= nir_tex_instr_create(b
->shader
, 2);
547 tex_all_same
->sampler_dim
= GLSL_SAMPLER_DIM_MS
;
548 tex_all_same
->op
= nir_texop_samples_identical
;
549 tex_all_same
->src
[0].src_type
= nir_tex_src_coord
;
550 tex_all_same
->src
[0].src
= nir_src_for_ssa(img_coord
);
551 tex_all_same
->src
[1].src_type
= nir_tex_src_texture_deref
;
552 tex_all_same
->src
[1].src
= nir_src_for_ssa(input_img_deref
);
553 tex_all_same
->dest_type
= nir_type_float
;
554 tex_all_same
->is_array
= false;
555 tex_all_same
->coord_components
= 2;
557 nir_ssa_dest_init(&tex_all_same
->instr
, &tex_all_same
->dest
, 1, 32, "tex");
558 nir_builder_instr_insert(b
, &tex_all_same
->instr
);
560 nir_ssa_def
*all_same
= nir_ieq(b
, &tex_all_same
->dest
.ssa
, nir_imm_int(b
, 0));
561 nir_if
*if_stmt
= nir_if_create(b
->shader
);
562 if_stmt
->condition
= nir_src_for_ssa(all_same
);
563 nir_cf_node_insert(b
->cursor
, &if_stmt
->cf_node
);
565 b
->cursor
= nir_after_cf_list(&if_stmt
->then_list
);
566 for (int i
= 1; i
< samples
; i
++) {
567 nir_tex_instr
*tex_add
= nir_tex_instr_create(b
->shader
, 3);
568 tex_add
->sampler_dim
= GLSL_SAMPLER_DIM_MS
;
569 tex_add
->op
= nir_texop_txf_ms
;
570 tex_add
->src
[0].src_type
= nir_tex_src_coord
;
571 tex_add
->src
[0].src
= nir_src_for_ssa(img_coord
);
572 tex_add
->src
[1].src_type
= nir_tex_src_ms_index
;
573 tex_add
->src
[1].src
= nir_src_for_ssa(nir_imm_int(b
, i
));
574 tex_add
->src
[2].src_type
= nir_tex_src_texture_deref
;
575 tex_add
->src
[2].src
= nir_src_for_ssa(input_img_deref
);
576 tex_add
->dest_type
= nir_type_float
;
577 tex_add
->is_array
= false;
578 tex_add
->coord_components
= 2;
580 nir_ssa_dest_init(&tex_add
->instr
, &tex_add
->dest
, 4, 32, "tex");
581 nir_builder_instr_insert(b
, &tex_add
->instr
);
583 tmp
= nir_fadd(b
, tmp
, &tex_add
->dest
.ssa
);
586 tmp
= nir_fdiv(b
, tmp
, nir_imm_float(b
, samples
));
587 nir_store_var(b
, color
, tmp
, 0xf);
588 b
->cursor
= nir_after_cf_list(&if_stmt
->else_list
);
591 nir_store_var(b
, color
, &tex
->dest
.ssa
, 0xf);
594 b
->cursor
= nir_after_cf_node(&outer_if
->cf_node
);