2 * Copyright © 2016 Red Hat.
3 * Copyright © 2016 Bas Nieuwenhuizen
5 * based in part on anv driver which is:
6 * Copyright © 2015 Intel Corporation
8 * Permission is hereby granted, free of charge, to any person obtaining a
9 * copy of this software and associated documentation files (the "Software"),
10 * to deal in the Software without restriction, including without limitation
11 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
12 * and/or sell copies of the Software, and to permit persons to whom the
13 * Software is furnished to do so, subject to the following conditions:
15 * The above copyright notice and this permission notice (including the next
16 * paragraph) shall be included in all copies or substantial portions of the
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
20 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
21 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
22 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
23 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
24 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
28 #include "radv_private.h"
29 #include "radv_radeon_winsys.h"
32 #include "vk_format.h"
33 #include "radv_meta.h"
37 static void radv_handle_image_transition(struct radv_cmd_buffer
*cmd_buffer
,
38 struct radv_image
*image
,
39 VkImageLayout src_layout
,
40 VkImageLayout dst_layout
,
43 const VkImageSubresourceRange
*range
,
44 VkImageAspectFlags pending_clears
);
46 const struct radv_dynamic_state default_dynamic_state
= {
59 .blend_constants
= { 0.0f
, 0.0f
, 0.0f
, 0.0f
},
64 .stencil_compare_mask
= {
68 .stencil_write_mask
= {
72 .stencil_reference
= {
79 radv_dynamic_state_copy(struct radv_dynamic_state
*dest
,
80 const struct radv_dynamic_state
*src
,
83 if (copy_mask
& (1 << VK_DYNAMIC_STATE_VIEWPORT
)) {
84 dest
->viewport
.count
= src
->viewport
.count
;
85 typed_memcpy(dest
->viewport
.viewports
, src
->viewport
.viewports
,
89 if (copy_mask
& (1 << VK_DYNAMIC_STATE_SCISSOR
)) {
90 dest
->scissor
.count
= src
->scissor
.count
;
91 typed_memcpy(dest
->scissor
.scissors
, src
->scissor
.scissors
,
95 if (copy_mask
& (1 << VK_DYNAMIC_STATE_LINE_WIDTH
))
96 dest
->line_width
= src
->line_width
;
98 if (copy_mask
& (1 << VK_DYNAMIC_STATE_DEPTH_BIAS
))
99 dest
->depth_bias
= src
->depth_bias
;
101 if (copy_mask
& (1 << VK_DYNAMIC_STATE_BLEND_CONSTANTS
))
102 typed_memcpy(dest
->blend_constants
, src
->blend_constants
, 4);
104 if (copy_mask
& (1 << VK_DYNAMIC_STATE_DEPTH_BOUNDS
))
105 dest
->depth_bounds
= src
->depth_bounds
;
107 if (copy_mask
& (1 << VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK
))
108 dest
->stencil_compare_mask
= src
->stencil_compare_mask
;
110 if (copy_mask
& (1 << VK_DYNAMIC_STATE_STENCIL_WRITE_MASK
))
111 dest
->stencil_write_mask
= src
->stencil_write_mask
;
113 if (copy_mask
& (1 << VK_DYNAMIC_STATE_STENCIL_REFERENCE
))
114 dest
->stencil_reference
= src
->stencil_reference
;
117 bool radv_cmd_buffer_uses_mec(struct radv_cmd_buffer
*cmd_buffer
)
119 return cmd_buffer
->queue_family_index
== RADV_QUEUE_COMPUTE
&&
120 cmd_buffer
->device
->physical_device
->rad_info
.chip_class
>= CIK
;
123 enum ring_type
radv_queue_family_to_ring(int f
) {
125 case RADV_QUEUE_GENERAL
:
127 case RADV_QUEUE_COMPUTE
:
129 case RADV_QUEUE_TRANSFER
:
132 unreachable("Unknown queue family");
136 static VkResult
radv_create_cmd_buffer(
137 struct radv_device
* device
,
138 struct radv_cmd_pool
* pool
,
139 VkCommandBufferLevel level
,
140 VkCommandBuffer
* pCommandBuffer
)
142 struct radv_cmd_buffer
*cmd_buffer
;
145 cmd_buffer
= vk_alloc(&pool
->alloc
, sizeof(*cmd_buffer
), 8,
146 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
147 if (cmd_buffer
== NULL
)
148 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
150 memset(cmd_buffer
, 0, sizeof(*cmd_buffer
));
151 cmd_buffer
->_loader_data
.loaderMagic
= ICD_LOADER_MAGIC
;
152 cmd_buffer
->device
= device
;
153 cmd_buffer
->pool
= pool
;
154 cmd_buffer
->level
= level
;
157 list_addtail(&cmd_buffer
->pool_link
, &pool
->cmd_buffers
);
158 cmd_buffer
->queue_family_index
= pool
->queue_family_index
;
161 /* Init the pool_link so we can safefly call list_del when we destroy
164 list_inithead(&cmd_buffer
->pool_link
);
165 cmd_buffer
->queue_family_index
= RADV_QUEUE_GENERAL
;
168 ring
= radv_queue_family_to_ring(cmd_buffer
->queue_family_index
);
170 cmd_buffer
->cs
= device
->ws
->cs_create(device
->ws
, ring
);
171 if (!cmd_buffer
->cs
) {
172 result
= VK_ERROR_OUT_OF_HOST_MEMORY
;
176 *pCommandBuffer
= radv_cmd_buffer_to_handle(cmd_buffer
);
178 cmd_buffer
->upload
.offset
= 0;
179 cmd_buffer
->upload
.size
= 0;
180 list_inithead(&cmd_buffer
->upload
.list
);
185 vk_free(&cmd_buffer
->pool
->alloc
, cmd_buffer
);
191 radv_cmd_buffer_resize_upload_buf(struct radv_cmd_buffer
*cmd_buffer
,
195 struct radeon_winsys_bo
*bo
;
196 struct radv_cmd_buffer_upload
*upload
;
197 struct radv_device
*device
= cmd_buffer
->device
;
199 new_size
= MAX2(min_needed
, 16 * 1024);
200 new_size
= MAX2(new_size
, 2 * cmd_buffer
->upload
.size
);
202 bo
= device
->ws
->buffer_create(device
->ws
,
205 RADEON_FLAG_CPU_ACCESS
);
208 cmd_buffer
->record_fail
= true;
212 device
->ws
->cs_add_buffer(cmd_buffer
->cs
, bo
, 8);
213 if (cmd_buffer
->upload
.upload_bo
) {
214 upload
= malloc(sizeof(*upload
));
217 cmd_buffer
->record_fail
= true;
218 device
->ws
->buffer_destroy(bo
);
222 memcpy(upload
, &cmd_buffer
->upload
, sizeof(*upload
));
223 list_add(&upload
->list
, &cmd_buffer
->upload
.list
);
226 cmd_buffer
->upload
.upload_bo
= bo
;
227 cmd_buffer
->upload
.size
= new_size
;
228 cmd_buffer
->upload
.offset
= 0;
229 cmd_buffer
->upload
.map
= device
->ws
->buffer_map(cmd_buffer
->upload
.upload_bo
);
231 if (!cmd_buffer
->upload
.map
) {
232 cmd_buffer
->record_fail
= true;
240 radv_cmd_buffer_upload_alloc(struct radv_cmd_buffer
*cmd_buffer
,
243 unsigned *out_offset
,
246 uint64_t offset
= align(cmd_buffer
->upload
.offset
, alignment
);
247 if (offset
+ size
> cmd_buffer
->upload
.size
) {
248 if (!radv_cmd_buffer_resize_upload_buf(cmd_buffer
, size
))
253 *out_offset
= offset
;
254 *ptr
= cmd_buffer
->upload
.map
+ offset
;
256 cmd_buffer
->upload
.offset
= offset
+ size
;
261 radv_cmd_buffer_upload_data(struct radv_cmd_buffer
*cmd_buffer
,
262 unsigned size
, unsigned alignment
,
263 const void *data
, unsigned *out_offset
)
267 if (!radv_cmd_buffer_upload_alloc(cmd_buffer
, size
, alignment
,
268 out_offset
, (void **)&ptr
))
272 memcpy(ptr
, data
, size
);
277 void radv_cmd_buffer_trace_emit(struct radv_cmd_buffer
*cmd_buffer
)
279 struct radv_device
*device
= cmd_buffer
->device
;
280 struct radeon_winsys_cs
*cs
= cmd_buffer
->cs
;
283 if (!device
->trace_bo
)
286 va
= device
->ws
->buffer_get_va(device
->trace_bo
);
288 MAYBE_UNUSED
unsigned cdw_max
= radeon_check_space(cmd_buffer
->device
->ws
, cmd_buffer
->cs
, 7);
290 ++cmd_buffer
->state
.trace_id
;
291 device
->ws
->cs_add_buffer(cs
, device
->trace_bo
, 8);
292 radeon_emit(cs
, PKT3(PKT3_WRITE_DATA
, 3, 0));
293 radeon_emit(cs
, S_370_DST_SEL(V_370_MEM_ASYNC
) |
294 S_370_WR_CONFIRM(1) |
295 S_370_ENGINE_SEL(V_370_ME
));
297 radeon_emit(cs
, va
>> 32);
298 radeon_emit(cs
, cmd_buffer
->state
.trace_id
);
299 radeon_emit(cs
, PKT3(PKT3_NOP
, 0, 0));
300 radeon_emit(cs
, AC_ENCODE_TRACE_POINT(cmd_buffer
->state
.trace_id
));
304 radv_emit_graphics_blend_state(struct radv_cmd_buffer
*cmd_buffer
,
305 struct radv_pipeline
*pipeline
)
307 radeon_set_context_reg_seq(cmd_buffer
->cs
, R_028780_CB_BLEND0_CONTROL
, 8);
308 radeon_emit_array(cmd_buffer
->cs
, pipeline
->graphics
.blend
.cb_blend_control
,
310 radeon_set_context_reg(cmd_buffer
->cs
, R_028808_CB_COLOR_CONTROL
, pipeline
->graphics
.blend
.cb_color_control
);
311 radeon_set_context_reg(cmd_buffer
->cs
, R_028B70_DB_ALPHA_TO_MASK
, pipeline
->graphics
.blend
.db_alpha_to_mask
);
315 radv_emit_graphics_depth_stencil_state(struct radv_cmd_buffer
*cmd_buffer
,
316 struct radv_pipeline
*pipeline
)
318 struct radv_depth_stencil_state
*ds
= &pipeline
->graphics
.ds
;
319 radeon_set_context_reg(cmd_buffer
->cs
, R_028800_DB_DEPTH_CONTROL
, ds
->db_depth_control
);
320 radeon_set_context_reg(cmd_buffer
->cs
, R_02842C_DB_STENCIL_CONTROL
, ds
->db_stencil_control
);
322 radeon_set_context_reg(cmd_buffer
->cs
, R_028000_DB_RENDER_CONTROL
, ds
->db_render_control
);
323 radeon_set_context_reg(cmd_buffer
->cs
, R_028010_DB_RENDER_OVERRIDE2
, ds
->db_render_override2
);
326 /* 12.4 fixed-point */
327 static unsigned radv_pack_float_12p4(float x
)
330 x
>= 4096 ? 0xffff : x
* 16;
334 shader_stage_to_user_data_0(gl_shader_stage stage
, bool has_gs
)
337 case MESA_SHADER_FRAGMENT
:
338 return R_00B030_SPI_SHADER_USER_DATA_PS_0
;
339 case MESA_SHADER_VERTEX
:
340 return has_gs
? R_00B330_SPI_SHADER_USER_DATA_ES_0
: R_00B130_SPI_SHADER_USER_DATA_VS_0
;
341 case MESA_SHADER_GEOMETRY
:
342 return R_00B230_SPI_SHADER_USER_DATA_GS_0
;
343 case MESA_SHADER_COMPUTE
:
344 return R_00B900_COMPUTE_USER_DATA_0
;
346 unreachable("unknown shader");
350 static struct ac_userdata_info
*
351 radv_lookup_user_sgpr(struct radv_pipeline
*pipeline
,
352 gl_shader_stage stage
,
355 return &pipeline
->shaders
[stage
]->info
.user_sgprs_locs
.shader_data
[idx
];
359 radv_emit_userdata_address(struct radv_cmd_buffer
*cmd_buffer
,
360 struct radv_pipeline
*pipeline
,
361 gl_shader_stage stage
,
362 int idx
, uint64_t va
)
364 struct ac_userdata_info
*loc
= radv_lookup_user_sgpr(pipeline
, stage
, idx
);
365 uint32_t base_reg
= shader_stage_to_user_data_0(stage
, radv_pipeline_has_gs(pipeline
));
366 if (loc
->sgpr_idx
== -1)
368 assert(loc
->num_sgprs
== 2);
369 assert(!loc
->indirect
);
370 radeon_set_sh_reg_seq(cmd_buffer
->cs
, base_reg
+ loc
->sgpr_idx
* 4, 2);
371 radeon_emit(cmd_buffer
->cs
, va
);
372 radeon_emit(cmd_buffer
->cs
, va
>> 32);
376 radv_update_multisample_state(struct radv_cmd_buffer
*cmd_buffer
,
377 struct radv_pipeline
*pipeline
)
379 int num_samples
= pipeline
->graphics
.ms
.num_samples
;
380 struct radv_multisample_state
*ms
= &pipeline
->graphics
.ms
;
381 struct radv_pipeline
*old_pipeline
= cmd_buffer
->state
.emitted_pipeline
;
383 radeon_set_context_reg_seq(cmd_buffer
->cs
, R_028C38_PA_SC_AA_MASK_X0Y0_X1Y0
, 2);
384 radeon_emit(cmd_buffer
->cs
, ms
->pa_sc_aa_mask
[0]);
385 radeon_emit(cmd_buffer
->cs
, ms
->pa_sc_aa_mask
[1]);
387 radeon_set_context_reg(cmd_buffer
->cs
, CM_R_028804_DB_EQAA
, ms
->db_eqaa
);
388 radeon_set_context_reg(cmd_buffer
->cs
, EG_R_028A4C_PA_SC_MODE_CNTL_1
, ms
->pa_sc_mode_cntl_1
);
390 if (old_pipeline
&& num_samples
== old_pipeline
->graphics
.ms
.num_samples
)
393 radeon_set_context_reg_seq(cmd_buffer
->cs
, CM_R_028BDC_PA_SC_LINE_CNTL
, 2);
394 radeon_emit(cmd_buffer
->cs
, ms
->pa_sc_line_cntl
);
395 radeon_emit(cmd_buffer
->cs
, ms
->pa_sc_aa_config
);
397 radv_cayman_emit_msaa_sample_locs(cmd_buffer
->cs
, num_samples
);
399 uint32_t samples_offset
;
402 radv_cmd_buffer_upload_alloc(cmd_buffer
, num_samples
* 4 * 2, 256, &samples_offset
,
404 switch (num_samples
) {
406 src
= cmd_buffer
->device
->sample_locations_1x
;
409 src
= cmd_buffer
->device
->sample_locations_2x
;
412 src
= cmd_buffer
->device
->sample_locations_4x
;
415 src
= cmd_buffer
->device
->sample_locations_8x
;
418 src
= cmd_buffer
->device
->sample_locations_16x
;
421 unreachable("unknown number of samples");
423 memcpy(samples_ptr
, src
, num_samples
* 4 * 2);
425 uint64_t va
= cmd_buffer
->device
->ws
->buffer_get_va(cmd_buffer
->upload
.upload_bo
);
426 va
+= samples_offset
;
428 radv_emit_userdata_address(cmd_buffer
, pipeline
, MESA_SHADER_FRAGMENT
,
429 AC_UD_PS_SAMPLE_POS
, va
);
433 radv_emit_graphics_raster_state(struct radv_cmd_buffer
*cmd_buffer
,
434 struct radv_pipeline
*pipeline
)
436 struct radv_raster_state
*raster
= &pipeline
->graphics
.raster
;
438 radeon_set_context_reg(cmd_buffer
->cs
, R_028810_PA_CL_CLIP_CNTL
,
439 raster
->pa_cl_clip_cntl
);
441 radeon_set_context_reg(cmd_buffer
->cs
, R_0286D4_SPI_INTERP_CONTROL_0
,
442 raster
->spi_interp_control
);
444 radeon_set_context_reg_seq(cmd_buffer
->cs
, R_028A00_PA_SU_POINT_SIZE
, 2);
445 unsigned tmp
= (unsigned)(1.0 * 8.0);
446 radeon_emit(cmd_buffer
->cs
, S_028A00_HEIGHT(tmp
) | S_028A00_WIDTH(tmp
));
447 radeon_emit(cmd_buffer
->cs
, S_028A04_MIN_SIZE(radv_pack_float_12p4(0)) |
448 S_028A04_MAX_SIZE(radv_pack_float_12p4(8192/2))); /* R_028A04_PA_SU_POINT_MINMAX */
450 radeon_set_context_reg(cmd_buffer
->cs
, R_028BE4_PA_SU_VTX_CNTL
,
451 raster
->pa_su_vtx_cntl
);
453 radeon_set_context_reg(cmd_buffer
->cs
, R_028814_PA_SU_SC_MODE_CNTL
,
454 raster
->pa_su_sc_mode_cntl
);
458 radv_emit_hw_vs(struct radv_cmd_buffer
*cmd_buffer
,
459 struct radv_pipeline
*pipeline
,
460 struct radv_shader_variant
*shader
)
462 struct radeon_winsys
*ws
= cmd_buffer
->device
->ws
;
463 uint64_t va
= ws
->buffer_get_va(shader
->bo
);
464 unsigned export_count
;
466 ws
->cs_add_buffer(cmd_buffer
->cs
, shader
->bo
, 8);
468 export_count
= MAX2(1, shader
->info
.vs
.param_exports
);
469 radeon_set_context_reg(cmd_buffer
->cs
, R_0286C4_SPI_VS_OUT_CONFIG
,
470 S_0286C4_VS_EXPORT_COUNT(export_count
- 1));
472 radeon_set_context_reg(cmd_buffer
->cs
, R_02870C_SPI_SHADER_POS_FORMAT
,
473 S_02870C_POS0_EXPORT_FORMAT(V_02870C_SPI_SHADER_4COMP
) |
474 S_02870C_POS1_EXPORT_FORMAT(shader
->info
.vs
.pos_exports
> 1 ?
475 V_02870C_SPI_SHADER_4COMP
:
476 V_02870C_SPI_SHADER_NONE
) |
477 S_02870C_POS2_EXPORT_FORMAT(shader
->info
.vs
.pos_exports
> 2 ?
478 V_02870C_SPI_SHADER_4COMP
:
479 V_02870C_SPI_SHADER_NONE
) |
480 S_02870C_POS3_EXPORT_FORMAT(shader
->info
.vs
.pos_exports
> 3 ?
481 V_02870C_SPI_SHADER_4COMP
:
482 V_02870C_SPI_SHADER_NONE
));
485 radeon_set_sh_reg_seq(cmd_buffer
->cs
, R_00B120_SPI_SHADER_PGM_LO_VS
, 4);
486 radeon_emit(cmd_buffer
->cs
, va
>> 8);
487 radeon_emit(cmd_buffer
->cs
, va
>> 40);
488 radeon_emit(cmd_buffer
->cs
, shader
->rsrc1
);
489 radeon_emit(cmd_buffer
->cs
, shader
->rsrc2
);
491 radeon_set_context_reg(cmd_buffer
->cs
, R_028818_PA_CL_VTE_CNTL
,
492 S_028818_VTX_W0_FMT(1) |
493 S_028818_VPORT_X_SCALE_ENA(1) | S_028818_VPORT_X_OFFSET_ENA(1) |
494 S_028818_VPORT_Y_SCALE_ENA(1) | S_028818_VPORT_Y_OFFSET_ENA(1) |
495 S_028818_VPORT_Z_SCALE_ENA(1) | S_028818_VPORT_Z_OFFSET_ENA(1));
497 unsigned clip_dist_mask
, cull_dist_mask
, total_mask
;
498 clip_dist_mask
= shader
->info
.vs
.clip_dist_mask
;
499 cull_dist_mask
= shader
->info
.vs
.cull_dist_mask
;
500 total_mask
= clip_dist_mask
| cull_dist_mask
;
502 radeon_set_context_reg(cmd_buffer
->cs
, R_02881C_PA_CL_VS_OUT_CNTL
,
503 S_02881C_USE_VTX_POINT_SIZE(shader
->info
.vs
.writes_pointsize
) |
504 S_02881C_USE_VTX_RENDER_TARGET_INDX(shader
->info
.vs
.writes_layer
) |
505 S_02881C_USE_VTX_VIEWPORT_INDX(shader
->info
.vs
.writes_viewport_index
) |
506 S_02881C_VS_OUT_MISC_VEC_ENA(shader
->info
.vs
.writes_pointsize
||
507 shader
->info
.vs
.writes_layer
||
508 shader
->info
.vs
.writes_viewport_index
) |
509 S_02881C_VS_OUT_CCDIST0_VEC_ENA((total_mask
& 0x0f) != 0) |
510 S_02881C_VS_OUT_CCDIST1_VEC_ENA((total_mask
& 0xf0) != 0) |
511 pipeline
->graphics
.raster
.pa_cl_vs_out_cntl
|
512 cull_dist_mask
<< 8 |
515 radeon_set_context_reg(cmd_buffer
->cs
, R_028AB4_VGT_REUSE_OFF
,
516 S_028AB4_REUSE_OFF(shader
->info
.vs
.writes_viewport_index
));
520 radv_emit_hw_es(struct radv_cmd_buffer
*cmd_buffer
,
521 struct radv_shader_variant
*shader
)
523 struct radeon_winsys
*ws
= cmd_buffer
->device
->ws
;
524 uint64_t va
= ws
->buffer_get_va(shader
->bo
);
526 ws
->cs_add_buffer(cmd_buffer
->cs
, shader
->bo
, 8);
528 radeon_set_context_reg(cmd_buffer
->cs
, R_028AAC_VGT_ESGS_RING_ITEMSIZE
,
529 shader
->info
.vs
.esgs_itemsize
/ 4);
530 radeon_set_sh_reg_seq(cmd_buffer
->cs
, R_00B320_SPI_SHADER_PGM_LO_ES
, 4);
531 radeon_emit(cmd_buffer
->cs
, va
>> 8);
532 radeon_emit(cmd_buffer
->cs
, va
>> 40);
533 radeon_emit(cmd_buffer
->cs
, shader
->rsrc1
);
534 radeon_emit(cmd_buffer
->cs
, shader
->rsrc2
);
538 radv_emit_vertex_shader(struct radv_cmd_buffer
*cmd_buffer
,
539 struct radv_pipeline
*pipeline
)
541 struct radv_shader_variant
*vs
;
543 assert (pipeline
->shaders
[MESA_SHADER_VERTEX
]);
545 vs
= pipeline
->shaders
[MESA_SHADER_VERTEX
];
547 if (vs
->info
.vs
.as_es
)
548 radv_emit_hw_es(cmd_buffer
, vs
);
550 radv_emit_hw_vs(cmd_buffer
, pipeline
, vs
);
552 radeon_set_context_reg(cmd_buffer
->cs
, R_028A84_VGT_PRIMITIVEID_EN
, 0);
555 static uint32_t si_vgt_gs_mode(struct radv_shader_variant
*gs
)
557 unsigned gs_max_vert_out
= gs
->info
.gs
.vertices_out
;
560 if (gs_max_vert_out
<= 128) {
561 cut_mode
= V_028A40_GS_CUT_128
;
562 } else if (gs_max_vert_out
<= 256) {
563 cut_mode
= V_028A40_GS_CUT_256
;
564 } else if (gs_max_vert_out
<= 512) {
565 cut_mode
= V_028A40_GS_CUT_512
;
567 assert(gs_max_vert_out
<= 1024);
568 cut_mode
= V_028A40_GS_CUT_1024
;
571 return S_028A40_MODE(V_028A40_GS_SCENARIO_G
) |
572 S_028A40_CUT_MODE(cut_mode
)|
573 S_028A40_ES_WRITE_OPTIMIZE(1) |
574 S_028A40_GS_WRITE_OPTIMIZE(1);
578 radv_emit_geometry_shader(struct radv_cmd_buffer
*cmd_buffer
,
579 struct radv_pipeline
*pipeline
)
581 struct radeon_winsys
*ws
= cmd_buffer
->device
->ws
;
582 struct radv_shader_variant
*gs
;
585 gs
= pipeline
->shaders
[MESA_SHADER_GEOMETRY
];
587 radeon_set_context_reg(cmd_buffer
->cs
, R_028A40_VGT_GS_MODE
, 0);
591 radeon_set_context_reg(cmd_buffer
->cs
, R_028A40_VGT_GS_MODE
, si_vgt_gs_mode(gs
));
593 uint32_t gsvs_itemsize
= gs
->info
.gs
.max_gsvs_emit_size
>> 2;
595 radeon_set_context_reg_seq(cmd_buffer
->cs
, R_028A60_VGT_GSVS_RING_OFFSET_1
, 3);
596 radeon_emit(cmd_buffer
->cs
, gsvs_itemsize
);
597 radeon_emit(cmd_buffer
->cs
, gsvs_itemsize
);
598 radeon_emit(cmd_buffer
->cs
, gsvs_itemsize
);
600 radeon_set_context_reg(cmd_buffer
->cs
, R_028AB0_VGT_GSVS_RING_ITEMSIZE
, gsvs_itemsize
);
602 radeon_set_context_reg(cmd_buffer
->cs
, R_028B38_VGT_GS_MAX_VERT_OUT
, gs
->info
.gs
.vertices_out
);
604 uint32_t gs_vert_itemsize
= gs
->info
.gs
.gsvs_vertex_size
;
605 radeon_set_context_reg_seq(cmd_buffer
->cs
, R_028B5C_VGT_GS_VERT_ITEMSIZE
, 4);
606 radeon_emit(cmd_buffer
->cs
, gs_vert_itemsize
>> 2);
607 radeon_emit(cmd_buffer
->cs
, 0);
608 radeon_emit(cmd_buffer
->cs
, 0);
609 radeon_emit(cmd_buffer
->cs
, 0);
611 uint32_t gs_num_invocations
= gs
->info
.gs
.invocations
;
612 radeon_set_context_reg(cmd_buffer
->cs
, R_028B90_VGT_GS_INSTANCE_CNT
,
613 S_028B90_CNT(MIN2(gs_num_invocations
, 127)) |
614 S_028B90_ENABLE(gs_num_invocations
> 0));
616 va
= ws
->buffer_get_va(gs
->bo
);
617 ws
->cs_add_buffer(cmd_buffer
->cs
, gs
->bo
, 8);
618 radeon_set_sh_reg_seq(cmd_buffer
->cs
, R_00B220_SPI_SHADER_PGM_LO_GS
, 4);
619 radeon_emit(cmd_buffer
->cs
, va
>> 8);
620 radeon_emit(cmd_buffer
->cs
, va
>> 40);
621 radeon_emit(cmd_buffer
->cs
, gs
->rsrc1
);
622 radeon_emit(cmd_buffer
->cs
, gs
->rsrc2
);
624 radv_emit_hw_vs(cmd_buffer
, pipeline
, pipeline
->gs_copy_shader
);
626 struct ac_userdata_info
*loc
= radv_lookup_user_sgpr(cmd_buffer
->state
.pipeline
, MESA_SHADER_GEOMETRY
,
627 AC_UD_GS_VS_RING_STRIDE_ENTRIES
);
628 if (loc
->sgpr_idx
!= -1) {
629 uint32_t stride
= gs
->info
.gs
.max_gsvs_emit_size
;
630 uint32_t num_entries
= 64;
631 bool is_vi
= cmd_buffer
->device
->physical_device
->rad_info
.chip_class
>= VI
;
634 num_entries
*= stride
;
636 stride
= S_008F04_STRIDE(stride
);
637 radeon_set_sh_reg_seq(cmd_buffer
->cs
, R_00B230_SPI_SHADER_USER_DATA_GS_0
+ loc
->sgpr_idx
* 4, 2);
638 radeon_emit(cmd_buffer
->cs
, stride
);
639 radeon_emit(cmd_buffer
->cs
, num_entries
);
644 radv_emit_fragment_shader(struct radv_cmd_buffer
*cmd_buffer
,
645 struct radv_pipeline
*pipeline
)
647 struct radeon_winsys
*ws
= cmd_buffer
->device
->ws
;
648 struct radv_shader_variant
*ps
, *vs
;
650 unsigned spi_baryc_cntl
= S_0286E0_FRONT_FACE_ALL_BITS(1);
651 struct radv_blend_state
*blend
= &pipeline
->graphics
.blend
;
652 unsigned ps_offset
= 0;
654 assert (pipeline
->shaders
[MESA_SHADER_FRAGMENT
]);
656 ps
= pipeline
->shaders
[MESA_SHADER_FRAGMENT
];
657 vs
= radv_pipeline_has_gs(pipeline
) ? pipeline
->gs_copy_shader
: pipeline
->shaders
[MESA_SHADER_VERTEX
];
658 va
= ws
->buffer_get_va(ps
->bo
);
659 ws
->cs_add_buffer(cmd_buffer
->cs
, ps
->bo
, 8);
661 radeon_set_sh_reg_seq(cmd_buffer
->cs
, R_00B020_SPI_SHADER_PGM_LO_PS
, 4);
662 radeon_emit(cmd_buffer
->cs
, va
>> 8);
663 radeon_emit(cmd_buffer
->cs
, va
>> 40);
664 radeon_emit(cmd_buffer
->cs
, ps
->rsrc1
);
665 radeon_emit(cmd_buffer
->cs
, ps
->rsrc2
);
667 if (ps
->info
.fs
.early_fragment_test
|| !ps
->info
.fs
.writes_memory
)
668 z_order
= V_02880C_EARLY_Z_THEN_LATE_Z
;
670 z_order
= V_02880C_LATE_Z
;
673 radeon_set_context_reg(cmd_buffer
->cs
, R_02880C_DB_SHADER_CONTROL
,
674 S_02880C_Z_EXPORT_ENABLE(ps
->info
.fs
.writes_z
) |
675 S_02880C_STENCIL_TEST_VAL_EXPORT_ENABLE(ps
->info
.fs
.writes_stencil
) |
676 S_02880C_KILL_ENABLE(!!ps
->info
.fs
.can_discard
) |
677 S_02880C_Z_ORDER(z_order
) |
678 S_02880C_DEPTH_BEFORE_SHADER(ps
->info
.fs
.early_fragment_test
) |
679 S_02880C_EXEC_ON_HIER_FAIL(ps
->info
.fs
.writes_memory
) |
680 S_02880C_EXEC_ON_NOOP(ps
->info
.fs
.writes_memory
));
682 radeon_set_context_reg(cmd_buffer
->cs
, R_0286CC_SPI_PS_INPUT_ENA
,
683 ps
->config
.spi_ps_input_ena
);
685 radeon_set_context_reg(cmd_buffer
->cs
, R_0286D0_SPI_PS_INPUT_ADDR
,
686 ps
->config
.spi_ps_input_addr
);
688 spi_baryc_cntl
|= S_0286E0_POS_FLOAT_LOCATION(0);
689 radeon_set_context_reg(cmd_buffer
->cs
, R_0286D8_SPI_PS_IN_CONTROL
,
690 S_0286D8_NUM_INTERP(ps
->info
.fs
.num_interp
));
692 radeon_set_context_reg(cmd_buffer
->cs
, R_0286E0_SPI_BARYC_CNTL
, spi_baryc_cntl
);
694 radeon_set_context_reg(cmd_buffer
->cs
, R_028710_SPI_SHADER_Z_FORMAT
,
695 ps
->info
.fs
.writes_stencil
? V_028710_SPI_SHADER_32_GR
:
696 ps
->info
.fs
.writes_z
? V_028710_SPI_SHADER_32_R
:
697 V_028710_SPI_SHADER_ZERO
);
699 radeon_set_context_reg(cmd_buffer
->cs
, R_028714_SPI_SHADER_COL_FORMAT
, blend
->spi_shader_col_format
);
701 radeon_set_context_reg(cmd_buffer
->cs
, R_028238_CB_TARGET_MASK
, blend
->cb_target_mask
);
702 radeon_set_context_reg(cmd_buffer
->cs
, R_02823C_CB_SHADER_MASK
, blend
->cb_shader_mask
);
704 if (ps
->info
.fs
.has_pcoord
) {
706 val
= S_028644_PT_SPRITE_TEX(1) | S_028644_OFFSET(0x20);
707 radeon_set_context_reg(cmd_buffer
->cs
, R_028644_SPI_PS_INPUT_CNTL_0
+ 4 * ps_offset
, val
);
711 if (ps
->info
.fs
.prim_id_input
&& (vs
->info
.vs
.prim_id_output
!= 0xffffffff)) {
712 unsigned vs_offset
, flat_shade
;
714 vs_offset
= vs
->info
.vs
.prim_id_output
;
716 val
= S_028644_OFFSET(vs_offset
) | S_028644_FLAT_SHADE(flat_shade
);
717 radeon_set_context_reg(cmd_buffer
->cs
, R_028644_SPI_PS_INPUT_CNTL_0
+ 4 * ps_offset
, val
);
721 if (ps
->info
.fs
.layer_input
&& (vs
->info
.vs
.layer_output
!= 0xffffffff)) {
722 unsigned vs_offset
, flat_shade
;
724 vs_offset
= vs
->info
.vs
.layer_output
;
726 val
= S_028644_OFFSET(vs_offset
) | S_028644_FLAT_SHADE(flat_shade
);
727 radeon_set_context_reg(cmd_buffer
->cs
, R_028644_SPI_PS_INPUT_CNTL_0
+ 4 * ps_offset
, val
);
731 for (unsigned i
= 0; i
< 32 && (1u << i
) <= ps
->info
.fs
.input_mask
; ++i
) {
732 unsigned vs_offset
, flat_shade
;
735 if (!(ps
->info
.fs
.input_mask
& (1u << i
)))
739 if (!(vs
->info
.vs
.export_mask
& (1u << i
))) {
740 radeon_set_context_reg(cmd_buffer
->cs
, R_028644_SPI_PS_INPUT_CNTL_0
+ 4 * ps_offset
,
741 S_028644_OFFSET(0x20));
746 vs_offset
= util_bitcount(vs
->info
.vs
.export_mask
& ((1u << i
) - 1));
747 if (vs
->info
.vs
.prim_id_output
!= 0xffffffff) {
748 if (vs_offset
>= vs
->info
.vs
.prim_id_output
)
751 if (vs
->info
.vs
.layer_output
!= 0xffffffff) {
752 if (vs_offset
>= vs
->info
.vs
.layer_output
)
755 flat_shade
= !!(ps
->info
.fs
.flat_shaded_mask
& (1u << ps_offset
));
757 val
= S_028644_OFFSET(vs_offset
) | S_028644_FLAT_SHADE(flat_shade
);
758 radeon_set_context_reg(cmd_buffer
->cs
, R_028644_SPI_PS_INPUT_CNTL_0
+ 4 * ps_offset
, val
);
764 radv_emit_graphics_pipeline(struct radv_cmd_buffer
*cmd_buffer
,
765 struct radv_pipeline
*pipeline
)
767 if (!pipeline
|| cmd_buffer
->state
.emitted_pipeline
== pipeline
)
770 radv_emit_graphics_depth_stencil_state(cmd_buffer
, pipeline
);
771 radv_emit_graphics_blend_state(cmd_buffer
, pipeline
);
772 radv_emit_graphics_raster_state(cmd_buffer
, pipeline
);
773 radv_update_multisample_state(cmd_buffer
, pipeline
);
774 radv_emit_vertex_shader(cmd_buffer
, pipeline
);
775 radv_emit_geometry_shader(cmd_buffer
, pipeline
);
776 radv_emit_fragment_shader(cmd_buffer
, pipeline
);
778 radeon_set_context_reg(cmd_buffer
->cs
, R_028A94_VGT_MULTI_PRIM_IB_RESET_EN
,
779 pipeline
->graphics
.prim_restart_enable
);
781 cmd_buffer
->scratch_size_needed
=
782 MAX2(cmd_buffer
->scratch_size_needed
,
783 pipeline
->max_waves
* pipeline
->scratch_bytes_per_wave
);
785 radeon_set_context_reg(cmd_buffer
->cs
, R_0286E8_SPI_TMPRING_SIZE
,
786 S_0286E8_WAVES(pipeline
->max_waves
) |
787 S_0286E8_WAVESIZE(pipeline
->scratch_bytes_per_wave
>> 10));
788 cmd_buffer
->state
.emitted_pipeline
= pipeline
;
792 radv_emit_viewport(struct radv_cmd_buffer
*cmd_buffer
)
794 si_write_viewport(cmd_buffer
->cs
, 0, cmd_buffer
->state
.dynamic
.viewport
.count
,
795 cmd_buffer
->state
.dynamic
.viewport
.viewports
);
799 radv_emit_scissor(struct radv_cmd_buffer
*cmd_buffer
)
801 uint32_t count
= cmd_buffer
->state
.dynamic
.scissor
.count
;
802 si_write_scissors(cmd_buffer
->cs
, 0, count
,
803 cmd_buffer
->state
.dynamic
.scissor
.scissors
);
804 radeon_set_context_reg(cmd_buffer
->cs
, R_028A48_PA_SC_MODE_CNTL_0
,
805 cmd_buffer
->state
.pipeline
->graphics
.ms
.pa_sc_mode_cntl_0
| S_028A48_VPORT_SCISSOR_ENABLE(count
? 1 : 0));
809 radv_emit_fb_color_state(struct radv_cmd_buffer
*cmd_buffer
,
811 struct radv_color_buffer_info
*cb
)
813 bool is_vi
= cmd_buffer
->device
->physical_device
->rad_info
.chip_class
>= VI
;
814 radeon_set_context_reg_seq(cmd_buffer
->cs
, R_028C60_CB_COLOR0_BASE
+ index
* 0x3c, 11);
815 radeon_emit(cmd_buffer
->cs
, cb
->cb_color_base
);
816 radeon_emit(cmd_buffer
->cs
, cb
->cb_color_pitch
);
817 radeon_emit(cmd_buffer
->cs
, cb
->cb_color_slice
);
818 radeon_emit(cmd_buffer
->cs
, cb
->cb_color_view
);
819 radeon_emit(cmd_buffer
->cs
, cb
->cb_color_info
);
820 radeon_emit(cmd_buffer
->cs
, cb
->cb_color_attrib
);
821 radeon_emit(cmd_buffer
->cs
, cb
->cb_dcc_control
);
822 radeon_emit(cmd_buffer
->cs
, cb
->cb_color_cmask
);
823 radeon_emit(cmd_buffer
->cs
, cb
->cb_color_cmask_slice
);
824 radeon_emit(cmd_buffer
->cs
, cb
->cb_color_fmask
);
825 radeon_emit(cmd_buffer
->cs
, cb
->cb_color_fmask_slice
);
827 if (is_vi
) { /* DCC BASE */
828 radeon_set_context_reg(cmd_buffer
->cs
, R_028C94_CB_COLOR0_DCC_BASE
+ index
* 0x3c, cb
->cb_dcc_base
);
833 radv_emit_fb_ds_state(struct radv_cmd_buffer
*cmd_buffer
,
834 struct radv_ds_buffer_info
*ds
,
835 struct radv_image
*image
,
836 VkImageLayout layout
)
838 uint32_t db_z_info
= ds
->db_z_info
;
840 if (!radv_layout_has_htile(image
, layout
))
841 db_z_info
&= C_028040_TILE_SURFACE_ENABLE
;
843 if (!radv_layout_can_expclear(image
, layout
))
844 db_z_info
&= C_028040_ALLOW_EXPCLEAR
& C_028044_ALLOW_EXPCLEAR
;
846 radeon_set_context_reg(cmd_buffer
->cs
, R_028008_DB_DEPTH_VIEW
, ds
->db_depth_view
);
847 radeon_set_context_reg(cmd_buffer
->cs
, R_028014_DB_HTILE_DATA_BASE
, ds
->db_htile_data_base
);
849 radeon_set_context_reg_seq(cmd_buffer
->cs
, R_02803C_DB_DEPTH_INFO
, 9);
850 radeon_emit(cmd_buffer
->cs
, ds
->db_depth_info
); /* R_02803C_DB_DEPTH_INFO */
851 radeon_emit(cmd_buffer
->cs
, db_z_info
); /* R_028040_DB_Z_INFO */
852 radeon_emit(cmd_buffer
->cs
, ds
->db_stencil_info
); /* R_028044_DB_STENCIL_INFO */
853 radeon_emit(cmd_buffer
->cs
, ds
->db_z_read_base
); /* R_028048_DB_Z_READ_BASE */
854 radeon_emit(cmd_buffer
->cs
, ds
->db_stencil_read_base
); /* R_02804C_DB_STENCIL_READ_BASE */
855 radeon_emit(cmd_buffer
->cs
, ds
->db_z_write_base
); /* R_028050_DB_Z_WRITE_BASE */
856 radeon_emit(cmd_buffer
->cs
, ds
->db_stencil_write_base
); /* R_028054_DB_STENCIL_WRITE_BASE */
857 radeon_emit(cmd_buffer
->cs
, ds
->db_depth_size
); /* R_028058_DB_DEPTH_SIZE */
858 radeon_emit(cmd_buffer
->cs
, ds
->db_depth_slice
); /* R_02805C_DB_DEPTH_SLICE */
860 radeon_set_context_reg(cmd_buffer
->cs
, R_028ABC_DB_HTILE_SURFACE
, ds
->db_htile_surface
);
861 radeon_set_context_reg(cmd_buffer
->cs
, R_028B78_PA_SU_POLY_OFFSET_DB_FMT_CNTL
,
862 ds
->pa_su_poly_offset_db_fmt_cntl
);
866 * To hw resolve multisample images both src and dst need to have the same
867 * micro tiling mode. However we don't always know in advance when creating
868 * the images. This function gets called if we have a resolve attachment,
869 * and tests if the attachment image has the same tiling mode, then it
870 * checks if the generated framebuffer data has the same tiling mode, and
873 static void radv_set_optimal_micro_tile_mode(struct radv_device
*device
,
874 struct radv_attachment_info
*att
,
875 uint32_t micro_tile_mode
)
877 struct radv_image
*image
= att
->attachment
->image
;
878 uint32_t tile_mode_index
;
879 if (image
->surface
.nsamples
<= 1)
882 if (image
->surface
.micro_tile_mode
!= micro_tile_mode
) {
883 radv_image_set_optimal_micro_tile_mode(device
, image
, micro_tile_mode
);
886 if (att
->cb
.micro_tile_mode
!= micro_tile_mode
) {
887 tile_mode_index
= image
->surface
.tiling_index
[0];
889 att
->cb
.cb_color_attrib
&= C_028C74_TILE_MODE_INDEX
;
890 att
->cb
.cb_color_attrib
|= S_028C74_TILE_MODE_INDEX(tile_mode_index
);
891 att
->cb
.micro_tile_mode
= micro_tile_mode
;
896 radv_set_depth_clear_regs(struct radv_cmd_buffer
*cmd_buffer
,
897 struct radv_image
*image
,
898 VkClearDepthStencilValue ds_clear_value
,
899 VkImageAspectFlags aspects
)
901 uint64_t va
= cmd_buffer
->device
->ws
->buffer_get_va(image
->bo
);
902 va
+= image
->offset
+ image
->clear_value_offset
;
903 unsigned reg_offset
= 0, reg_count
= 0;
905 if (!image
->htile
.size
|| !aspects
)
908 if (aspects
& VK_IMAGE_ASPECT_STENCIL_BIT
) {
914 if (aspects
& VK_IMAGE_ASPECT_DEPTH_BIT
)
917 cmd_buffer
->device
->ws
->cs_add_buffer(cmd_buffer
->cs
, image
->bo
, 8);
919 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_WRITE_DATA
, 2 + reg_count
, 0));
920 radeon_emit(cmd_buffer
->cs
, S_370_DST_SEL(V_370_MEM_ASYNC
) |
921 S_370_WR_CONFIRM(1) |
922 S_370_ENGINE_SEL(V_370_PFP
));
923 radeon_emit(cmd_buffer
->cs
, va
);
924 radeon_emit(cmd_buffer
->cs
, va
>> 32);
925 if (aspects
& VK_IMAGE_ASPECT_STENCIL_BIT
)
926 radeon_emit(cmd_buffer
->cs
, ds_clear_value
.stencil
);
927 if (aspects
& VK_IMAGE_ASPECT_DEPTH_BIT
)
928 radeon_emit(cmd_buffer
->cs
, fui(ds_clear_value
.depth
));
930 radeon_set_context_reg_seq(cmd_buffer
->cs
, R_028028_DB_STENCIL_CLEAR
+ 4 * reg_offset
, reg_count
);
931 if (aspects
& VK_IMAGE_ASPECT_STENCIL_BIT
)
932 radeon_emit(cmd_buffer
->cs
, ds_clear_value
.stencil
); /* R_028028_DB_STENCIL_CLEAR */
933 if (aspects
& VK_IMAGE_ASPECT_DEPTH_BIT
)
934 radeon_emit(cmd_buffer
->cs
, fui(ds_clear_value
.depth
)); /* R_02802C_DB_DEPTH_CLEAR */
938 radv_load_depth_clear_regs(struct radv_cmd_buffer
*cmd_buffer
,
939 struct radv_image
*image
)
941 uint64_t va
= cmd_buffer
->device
->ws
->buffer_get_va(image
->bo
);
942 va
+= image
->offset
+ image
->clear_value_offset
;
944 if (!image
->htile
.size
)
947 cmd_buffer
->device
->ws
->cs_add_buffer(cmd_buffer
->cs
, image
->bo
, 8);
949 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_COPY_DATA
, 4, 0));
950 radeon_emit(cmd_buffer
->cs
, COPY_DATA_SRC_SEL(COPY_DATA_MEM
) |
951 COPY_DATA_DST_SEL(COPY_DATA_REG
) |
952 COPY_DATA_COUNT_SEL
);
953 radeon_emit(cmd_buffer
->cs
, va
);
954 radeon_emit(cmd_buffer
->cs
, va
>> 32);
955 radeon_emit(cmd_buffer
->cs
, R_028028_DB_STENCIL_CLEAR
>> 2);
956 radeon_emit(cmd_buffer
->cs
, 0);
958 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_PFP_SYNC_ME
, 0, 0));
959 radeon_emit(cmd_buffer
->cs
, 0);
963 radv_set_color_clear_regs(struct radv_cmd_buffer
*cmd_buffer
,
964 struct radv_image
*image
,
966 uint32_t color_values
[2])
968 uint64_t va
= cmd_buffer
->device
->ws
->buffer_get_va(image
->bo
);
969 va
+= image
->offset
+ image
->clear_value_offset
;
971 if (!image
->cmask
.size
&& !image
->surface
.dcc_size
)
974 cmd_buffer
->device
->ws
->cs_add_buffer(cmd_buffer
->cs
, image
->bo
, 8);
976 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_WRITE_DATA
, 4, 0));
977 radeon_emit(cmd_buffer
->cs
, S_370_DST_SEL(V_370_MEM_ASYNC
) |
978 S_370_WR_CONFIRM(1) |
979 S_370_ENGINE_SEL(V_370_PFP
));
980 radeon_emit(cmd_buffer
->cs
, va
);
981 radeon_emit(cmd_buffer
->cs
, va
>> 32);
982 radeon_emit(cmd_buffer
->cs
, color_values
[0]);
983 radeon_emit(cmd_buffer
->cs
, color_values
[1]);
985 radeon_set_context_reg_seq(cmd_buffer
->cs
, R_028C8C_CB_COLOR0_CLEAR_WORD0
+ idx
* 0x3c, 2);
986 radeon_emit(cmd_buffer
->cs
, color_values
[0]);
987 radeon_emit(cmd_buffer
->cs
, color_values
[1]);
991 radv_load_color_clear_regs(struct radv_cmd_buffer
*cmd_buffer
,
992 struct radv_image
*image
,
995 uint64_t va
= cmd_buffer
->device
->ws
->buffer_get_va(image
->bo
);
996 va
+= image
->offset
+ image
->clear_value_offset
;
998 if (!image
->cmask
.size
&& !image
->surface
.dcc_size
)
1001 uint32_t reg
= R_028C8C_CB_COLOR0_CLEAR_WORD0
+ idx
* 0x3c;
1002 cmd_buffer
->device
->ws
->cs_add_buffer(cmd_buffer
->cs
, image
->bo
, 8);
1004 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_COPY_DATA
, 4, 0));
1005 radeon_emit(cmd_buffer
->cs
, COPY_DATA_SRC_SEL(COPY_DATA_MEM
) |
1006 COPY_DATA_DST_SEL(COPY_DATA_REG
) |
1007 COPY_DATA_COUNT_SEL
);
1008 radeon_emit(cmd_buffer
->cs
, va
);
1009 radeon_emit(cmd_buffer
->cs
, va
>> 32);
1010 radeon_emit(cmd_buffer
->cs
, reg
>> 2);
1011 radeon_emit(cmd_buffer
->cs
, 0);
1013 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_PFP_SYNC_ME
, 0, 0));
1014 radeon_emit(cmd_buffer
->cs
, 0);
1018 radv_emit_framebuffer_state(struct radv_cmd_buffer
*cmd_buffer
)
1021 struct radv_framebuffer
*framebuffer
= cmd_buffer
->state
.framebuffer
;
1022 const struct radv_subpass
*subpass
= cmd_buffer
->state
.subpass
;
1023 int dst_resolve_micro_tile_mode
= -1;
1025 if (subpass
->has_resolve
) {
1026 uint32_t a
= subpass
->resolve_attachments
[0].attachment
;
1027 const struct radv_image
*image
= framebuffer
->attachments
[a
].attachment
->image
;
1028 dst_resolve_micro_tile_mode
= image
->surface
.micro_tile_mode
;
1030 for (i
= 0; i
< subpass
->color_count
; ++i
) {
1031 int idx
= subpass
->color_attachments
[i
].attachment
;
1032 struct radv_attachment_info
*att
= &framebuffer
->attachments
[idx
];
1034 if (dst_resolve_micro_tile_mode
!= -1) {
1035 radv_set_optimal_micro_tile_mode(cmd_buffer
->device
,
1036 att
, dst_resolve_micro_tile_mode
);
1038 cmd_buffer
->device
->ws
->cs_add_buffer(cmd_buffer
->cs
, att
->attachment
->bo
, 8);
1040 assert(att
->attachment
->aspect_mask
& VK_IMAGE_ASPECT_COLOR_BIT
);
1041 radv_emit_fb_color_state(cmd_buffer
, i
, &att
->cb
);
1043 radv_load_color_clear_regs(cmd_buffer
, att
->attachment
->image
, i
);
1046 for (i
= subpass
->color_count
; i
< 8; i
++)
1047 radeon_set_context_reg(cmd_buffer
->cs
, R_028C70_CB_COLOR0_INFO
+ i
* 0x3C,
1048 S_028C70_FORMAT(V_028C70_COLOR_INVALID
));
1050 if(subpass
->depth_stencil_attachment
.attachment
!= VK_ATTACHMENT_UNUSED
) {
1051 int idx
= subpass
->depth_stencil_attachment
.attachment
;
1052 VkImageLayout layout
= subpass
->depth_stencil_attachment
.layout
;
1053 struct radv_attachment_info
*att
= &framebuffer
->attachments
[idx
];
1054 struct radv_image
*image
= att
->attachment
->image
;
1055 cmd_buffer
->device
->ws
->cs_add_buffer(cmd_buffer
->cs
, att
->attachment
->bo
, 8);
1057 radv_emit_fb_ds_state(cmd_buffer
, &att
->ds
, image
, layout
);
1059 if (att
->ds
.offset_scale
!= cmd_buffer
->state
.offset_scale
) {
1060 cmd_buffer
->state
.dirty
|= RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS
;
1061 cmd_buffer
->state
.offset_scale
= att
->ds
.offset_scale
;
1063 radv_load_depth_clear_regs(cmd_buffer
, image
);
1065 radeon_set_context_reg_seq(cmd_buffer
->cs
, R_028040_DB_Z_INFO
, 2);
1066 radeon_emit(cmd_buffer
->cs
, S_028040_FORMAT(V_028040_Z_INVALID
)); /* R_028040_DB_Z_INFO */
1067 radeon_emit(cmd_buffer
->cs
, S_028044_FORMAT(V_028044_STENCIL_INVALID
)); /* R_028044_DB_STENCIL_INFO */
1069 radeon_set_context_reg(cmd_buffer
->cs
, R_028208_PA_SC_WINDOW_SCISSOR_BR
,
1070 S_028208_BR_X(framebuffer
->width
) |
1071 S_028208_BR_Y(framebuffer
->height
));
1074 void radv_set_db_count_control(struct radv_cmd_buffer
*cmd_buffer
)
1076 uint32_t db_count_control
;
1078 if(!cmd_buffer
->state
.active_occlusion_queries
) {
1079 if (cmd_buffer
->device
->physical_device
->rad_info
.chip_class
>= CIK
) {
1080 db_count_control
= 0;
1082 db_count_control
= S_028004_ZPASS_INCREMENT_DISABLE(1);
1085 if (cmd_buffer
->device
->physical_device
->rad_info
.chip_class
>= CIK
) {
1086 db_count_control
= S_028004_PERFECT_ZPASS_COUNTS(1) |
1087 S_028004_SAMPLE_RATE(0) | /* TODO: set this to the number of samples of the current framebuffer */
1088 S_028004_ZPASS_ENABLE(1) |
1089 S_028004_SLICE_EVEN_ENABLE(1) |
1090 S_028004_SLICE_ODD_ENABLE(1);
1092 db_count_control
= S_028004_PERFECT_ZPASS_COUNTS(1) |
1093 S_028004_SAMPLE_RATE(0); /* TODO: set this to the number of samples of the current framebuffer */
1097 radeon_set_context_reg(cmd_buffer
->cs
, R_028004_DB_COUNT_CONTROL
, db_count_control
);
1101 radv_cmd_buffer_flush_dynamic_state(struct radv_cmd_buffer
*cmd_buffer
)
1103 struct radv_dynamic_state
*d
= &cmd_buffer
->state
.dynamic
;
1105 if (cmd_buffer
->state
.dirty
& RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH
) {
1106 unsigned width
= cmd_buffer
->state
.dynamic
.line_width
* 8;
1107 radeon_set_context_reg(cmd_buffer
->cs
, R_028A08_PA_SU_LINE_CNTL
,
1108 S_028A08_WIDTH(CLAMP(width
, 0, 0xFFF)));
1111 if (cmd_buffer
->state
.dirty
& RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS
) {
1112 radeon_set_context_reg_seq(cmd_buffer
->cs
, R_028414_CB_BLEND_RED
, 4);
1113 radeon_emit_array(cmd_buffer
->cs
, (uint32_t*)d
->blend_constants
, 4);
1116 if (cmd_buffer
->state
.dirty
& (RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE
|
1117 RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK
|
1118 RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK
)) {
1119 radeon_set_context_reg_seq(cmd_buffer
->cs
, R_028430_DB_STENCILREFMASK
, 2);
1120 radeon_emit(cmd_buffer
->cs
, S_028430_STENCILTESTVAL(d
->stencil_reference
.front
) |
1121 S_028430_STENCILMASK(d
->stencil_compare_mask
.front
) |
1122 S_028430_STENCILWRITEMASK(d
->stencil_write_mask
.front
) |
1123 S_028430_STENCILOPVAL(1));
1124 radeon_emit(cmd_buffer
->cs
, S_028434_STENCILTESTVAL_BF(d
->stencil_reference
.back
) |
1125 S_028434_STENCILMASK_BF(d
->stencil_compare_mask
.back
) |
1126 S_028434_STENCILWRITEMASK_BF(d
->stencil_write_mask
.back
) |
1127 S_028434_STENCILOPVAL_BF(1));
1130 if (cmd_buffer
->state
.dirty
& (RADV_CMD_DIRTY_PIPELINE
|
1131 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS
)) {
1132 radeon_set_context_reg(cmd_buffer
->cs
, R_028020_DB_DEPTH_BOUNDS_MIN
, fui(d
->depth_bounds
.min
));
1133 radeon_set_context_reg(cmd_buffer
->cs
, R_028024_DB_DEPTH_BOUNDS_MAX
, fui(d
->depth_bounds
.max
));
1136 if (cmd_buffer
->state
.dirty
& (RADV_CMD_DIRTY_PIPELINE
|
1137 RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS
)) {
1138 struct radv_raster_state
*raster
= &cmd_buffer
->state
.pipeline
->graphics
.raster
;
1139 unsigned slope
= fui(d
->depth_bias
.slope
* 16.0f
);
1140 unsigned bias
= fui(d
->depth_bias
.bias
* cmd_buffer
->state
.offset_scale
);
1142 if (G_028814_POLY_OFFSET_FRONT_ENABLE(raster
->pa_su_sc_mode_cntl
)) {
1143 radeon_set_context_reg_seq(cmd_buffer
->cs
, R_028B7C_PA_SU_POLY_OFFSET_CLAMP
, 5);
1144 radeon_emit(cmd_buffer
->cs
, fui(d
->depth_bias
.clamp
)); /* CLAMP */
1145 radeon_emit(cmd_buffer
->cs
, slope
); /* FRONT SCALE */
1146 radeon_emit(cmd_buffer
->cs
, bias
); /* FRONT OFFSET */
1147 radeon_emit(cmd_buffer
->cs
, slope
); /* BACK SCALE */
1148 radeon_emit(cmd_buffer
->cs
, bias
); /* BACK OFFSET */
1152 cmd_buffer
->state
.dirty
= 0;
1156 emit_stage_descriptor_set_userdata(struct radv_cmd_buffer
*cmd_buffer
,
1157 struct radv_pipeline
*pipeline
,
1160 gl_shader_stage stage
)
1162 struct ac_userdata_info
*desc_set_loc
= &pipeline
->shaders
[stage
]->info
.user_sgprs_locs
.descriptor_sets
[idx
];
1163 uint32_t base_reg
= shader_stage_to_user_data_0(stage
, radv_pipeline_has_gs(pipeline
));
1165 if (desc_set_loc
->sgpr_idx
== -1)
1168 assert(!desc_set_loc
->indirect
);
1169 assert(desc_set_loc
->num_sgprs
== 2);
1170 radeon_set_sh_reg_seq(cmd_buffer
->cs
,
1171 base_reg
+ desc_set_loc
->sgpr_idx
* 4, 2);
1172 radeon_emit(cmd_buffer
->cs
, va
);
1173 radeon_emit(cmd_buffer
->cs
, va
>> 32);
1177 radv_emit_descriptor_set_userdata(struct radv_cmd_buffer
*cmd_buffer
,
1178 struct radv_pipeline
*pipeline
,
1179 VkShaderStageFlags stages
,
1180 struct radv_descriptor_set
*set
,
1183 if (stages
& VK_SHADER_STAGE_FRAGMENT_BIT
)
1184 emit_stage_descriptor_set_userdata(cmd_buffer
, pipeline
,
1186 MESA_SHADER_FRAGMENT
);
1188 if (stages
& VK_SHADER_STAGE_VERTEX_BIT
)
1189 emit_stage_descriptor_set_userdata(cmd_buffer
, pipeline
,
1191 MESA_SHADER_VERTEX
);
1193 if ((stages
& VK_SHADER_STAGE_GEOMETRY_BIT
) && radv_pipeline_has_gs(pipeline
))
1194 emit_stage_descriptor_set_userdata(cmd_buffer
, pipeline
,
1196 MESA_SHADER_GEOMETRY
);
1198 if (stages
& VK_SHADER_STAGE_COMPUTE_BIT
)
1199 emit_stage_descriptor_set_userdata(cmd_buffer
, pipeline
,
1201 MESA_SHADER_COMPUTE
);
1205 radv_flush_descriptors(struct radv_cmd_buffer
*cmd_buffer
,
1206 struct radv_pipeline
*pipeline
,
1207 VkShaderStageFlags stages
)
1210 if (!cmd_buffer
->state
.descriptors_dirty
)
1213 for (i
= 0; i
< MAX_SETS
; i
++) {
1214 if (!(cmd_buffer
->state
.descriptors_dirty
& (1 << i
)))
1216 struct radv_descriptor_set
*set
= cmd_buffer
->state
.descriptors
[i
];
1220 radv_emit_descriptor_set_userdata(cmd_buffer
, pipeline
, stages
, set
, i
);
1222 cmd_buffer
->state
.descriptors_dirty
= 0;
1226 radv_flush_constants(struct radv_cmd_buffer
*cmd_buffer
,
1227 struct radv_pipeline
*pipeline
,
1228 VkShaderStageFlags stages
)
1230 struct radv_pipeline_layout
*layout
= pipeline
->layout
;
1235 stages
&= cmd_buffer
->push_constant_stages
;
1236 if (!stages
|| !layout
|| (!layout
->push_constant_size
&& !layout
->dynamic_offset_count
))
1239 radv_cmd_buffer_upload_alloc(cmd_buffer
, layout
->push_constant_size
+
1240 16 * layout
->dynamic_offset_count
,
1241 256, &offset
, &ptr
);
1243 memcpy(ptr
, cmd_buffer
->push_constants
, layout
->push_constant_size
);
1244 memcpy((char*)ptr
+ layout
->push_constant_size
, cmd_buffer
->dynamic_buffers
,
1245 16 * layout
->dynamic_offset_count
);
1247 va
= cmd_buffer
->device
->ws
->buffer_get_va(cmd_buffer
->upload
.upload_bo
);
1250 if (stages
& VK_SHADER_STAGE_VERTEX_BIT
)
1251 radv_emit_userdata_address(cmd_buffer
, pipeline
, MESA_SHADER_VERTEX
,
1252 AC_UD_PUSH_CONSTANTS
, va
);
1254 if (stages
& VK_SHADER_STAGE_FRAGMENT_BIT
)
1255 radv_emit_userdata_address(cmd_buffer
, pipeline
, MESA_SHADER_FRAGMENT
,
1256 AC_UD_PUSH_CONSTANTS
, va
);
1258 if ((stages
& VK_SHADER_STAGE_GEOMETRY_BIT
) && radv_pipeline_has_gs(pipeline
))
1259 radv_emit_userdata_address(cmd_buffer
, pipeline
, MESA_SHADER_GEOMETRY
,
1260 AC_UD_PUSH_CONSTANTS
, va
);
1262 if (stages
& VK_SHADER_STAGE_COMPUTE_BIT
)
1263 radv_emit_userdata_address(cmd_buffer
, pipeline
, MESA_SHADER_COMPUTE
,
1264 AC_UD_PUSH_CONSTANTS
, va
);
1266 cmd_buffer
->push_constant_stages
&= ~stages
;
1270 radv_cmd_buffer_flush_state(struct radv_cmd_buffer
*cmd_buffer
, bool instanced_or_indirect_draw
,
1271 uint32_t draw_vertex_count
)
1273 struct radv_pipeline
*pipeline
= cmd_buffer
->state
.pipeline
;
1274 struct radv_device
*device
= cmd_buffer
->device
;
1275 uint32_t ia_multi_vgt_param
;
1276 uint32_t ls_hs_config
= 0;
1278 MAYBE_UNUSED
unsigned cdw_max
= radeon_check_space(cmd_buffer
->device
->ws
,
1279 cmd_buffer
->cs
, 4096);
1281 if ((cmd_buffer
->state
.vertex_descriptors_dirty
|| cmd_buffer
->state
.vb_dirty
) &&
1282 cmd_buffer
->state
.pipeline
->num_vertex_attribs
) {
1286 uint32_t num_attribs
= cmd_buffer
->state
.pipeline
->num_vertex_attribs
;
1289 /* allocate some descriptor state for vertex buffers */
1290 radv_cmd_buffer_upload_alloc(cmd_buffer
, num_attribs
* 16, 256,
1291 &vb_offset
, &vb_ptr
);
1293 for (i
= 0; i
< num_attribs
; i
++) {
1294 uint32_t *desc
= &((uint32_t *)vb_ptr
)[i
* 4];
1296 int vb
= cmd_buffer
->state
.pipeline
->va_binding
[i
];
1297 struct radv_buffer
*buffer
= cmd_buffer
->state
.vertex_bindings
[vb
].buffer
;
1298 uint32_t stride
= cmd_buffer
->state
.pipeline
->binding_stride
[vb
];
1300 device
->ws
->cs_add_buffer(cmd_buffer
->cs
, buffer
->bo
, 8);
1301 va
= device
->ws
->buffer_get_va(buffer
->bo
);
1303 offset
= cmd_buffer
->state
.vertex_bindings
[vb
].offset
+ cmd_buffer
->state
.pipeline
->va_offset
[i
];
1304 va
+= offset
+ buffer
->offset
;
1306 desc
[1] = S_008F04_BASE_ADDRESS_HI(va
>> 32) | S_008F04_STRIDE(stride
);
1307 if (cmd_buffer
->device
->physical_device
->rad_info
.chip_class
<= CIK
&& stride
)
1308 desc
[2] = (buffer
->size
- offset
- cmd_buffer
->state
.pipeline
->va_format_size
[i
]) / stride
+ 1;
1310 desc
[2] = buffer
->size
- offset
;
1311 desc
[3] = cmd_buffer
->state
.pipeline
->va_rsrc_word3
[i
];
1314 va
= device
->ws
->buffer_get_va(cmd_buffer
->upload
.upload_bo
);
1317 radv_emit_userdata_address(cmd_buffer
, pipeline
, MESA_SHADER_VERTEX
,
1318 AC_UD_VS_VERTEX_BUFFERS
, va
);
1321 cmd_buffer
->state
.vertex_descriptors_dirty
= false;
1322 cmd_buffer
->state
.vb_dirty
= 0;
1323 if (cmd_buffer
->state
.dirty
& RADV_CMD_DIRTY_PIPELINE
)
1324 radv_emit_graphics_pipeline(cmd_buffer
, pipeline
);
1326 if (cmd_buffer
->state
.dirty
& RADV_CMD_DIRTY_RENDER_TARGETS
)
1327 radv_emit_framebuffer_state(cmd_buffer
);
1329 if (cmd_buffer
->state
.dirty
& (RADV_CMD_DIRTY_DYNAMIC_VIEWPORT
))
1330 radv_emit_viewport(cmd_buffer
);
1332 if (cmd_buffer
->state
.dirty
& (RADV_CMD_DIRTY_DYNAMIC_SCISSOR
))
1333 radv_emit_scissor(cmd_buffer
);
1335 ia_multi_vgt_param
= si_get_ia_multi_vgt_param(cmd_buffer
, instanced_or_indirect_draw
, draw_vertex_count
);
1336 if (cmd_buffer
->state
.last_ia_multi_vgt_param
!= ia_multi_vgt_param
) {
1337 if (cmd_buffer
->device
->physical_device
->rad_info
.chip_class
>= CIK
)
1338 radeon_set_context_reg_idx(cmd_buffer
->cs
, R_028AA8_IA_MULTI_VGT_PARAM
, 1, ia_multi_vgt_param
);
1340 radeon_set_context_reg(cmd_buffer
->cs
, R_028AA8_IA_MULTI_VGT_PARAM
, ia_multi_vgt_param
);
1341 cmd_buffer
->state
.last_ia_multi_vgt_param
= ia_multi_vgt_param
;
1344 if (cmd_buffer
->state
.dirty
& RADV_CMD_DIRTY_PIPELINE
) {
1345 uint32_t stages
= 0;
1347 if (radv_pipeline_has_gs(cmd_buffer
->state
.pipeline
))
1348 stages
|= S_028B54_ES_EN(V_028B54_ES_STAGE_REAL
) |
1350 S_028B54_VS_EN(V_028B54_VS_STAGE_COPY_SHADER
);
1352 radeon_set_context_reg(cmd_buffer
->cs
, R_028B54_VGT_SHADER_STAGES_EN
, stages
);
1354 if (cmd_buffer
->device
->physical_device
->rad_info
.chip_class
>= CIK
) {
1355 radeon_set_context_reg_idx(cmd_buffer
->cs
, R_028B58_VGT_LS_HS_CONFIG
, 2, ls_hs_config
);
1356 radeon_set_uconfig_reg_idx(cmd_buffer
->cs
, R_030908_VGT_PRIMITIVE_TYPE
, 1, cmd_buffer
->state
.pipeline
->graphics
.prim
);
1358 radeon_set_config_reg(cmd_buffer
->cs
, R_008958_VGT_PRIMITIVE_TYPE
, cmd_buffer
->state
.pipeline
->graphics
.prim
);
1359 radeon_set_context_reg(cmd_buffer
->cs
, R_028B58_VGT_LS_HS_CONFIG
, ls_hs_config
);
1361 radeon_set_context_reg(cmd_buffer
->cs
, R_028A6C_VGT_GS_OUT_PRIM_TYPE
, cmd_buffer
->state
.pipeline
->graphics
.gs_out
);
1364 radv_cmd_buffer_flush_dynamic_state(cmd_buffer
);
1366 radv_flush_descriptors(cmd_buffer
, cmd_buffer
->state
.pipeline
,
1367 VK_SHADER_STAGE_ALL_GRAPHICS
);
1368 radv_flush_constants(cmd_buffer
, cmd_buffer
->state
.pipeline
,
1369 VK_SHADER_STAGE_ALL_GRAPHICS
);
1371 assert(cmd_buffer
->cs
->cdw
<= cdw_max
);
1373 si_emit_cache_flush(cmd_buffer
);
1376 static void radv_stage_flush(struct radv_cmd_buffer
*cmd_buffer
,
1377 VkPipelineStageFlags src_stage_mask
)
1379 if (src_stage_mask
& (VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT
|
1380 VK_PIPELINE_STAGE_TRANSFER_BIT
|
1381 VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT
|
1382 VK_PIPELINE_STAGE_ALL_COMMANDS_BIT
)) {
1383 cmd_buffer
->state
.flush_bits
|= RADV_CMD_FLAG_CS_PARTIAL_FLUSH
;
1386 if (src_stage_mask
& (VK_PIPELINE_STAGE_TESSELLATION_CONTROL_SHADER_BIT
|
1387 VK_PIPELINE_STAGE_TESSELLATION_EVALUATION_SHADER_BIT
|
1388 VK_PIPELINE_STAGE_GEOMETRY_SHADER_BIT
|
1389 VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT
|
1390 VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT
|
1391 VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT
|
1392 VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT
|
1393 VK_PIPELINE_STAGE_TRANSFER_BIT
|
1394 VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT
|
1395 VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT
|
1396 VK_PIPELINE_STAGE_ALL_COMMANDS_BIT
)) {
1397 cmd_buffer
->state
.flush_bits
|= RADV_CMD_FLAG_PS_PARTIAL_FLUSH
;
1398 } else if (src_stage_mask
& (VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT
|
1399 VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT
|
1400 VK_PIPELINE_STAGE_VERTEX_INPUT_BIT
|
1401 VK_PIPELINE_STAGE_VERTEX_SHADER_BIT
)) {
1402 cmd_buffer
->state
.flush_bits
|= RADV_CMD_FLAG_VS_PARTIAL_FLUSH
;
1406 static void radv_src_access_flush(struct radv_cmd_buffer
*cmd_buffer
,
1407 VkAccessFlags src_flags
)
1409 enum radv_cmd_flush_bits flush_bits
= 0;
1411 for_each_bit(b
, src_flags
) {
1412 switch ((VkAccessFlagBits
)(1 << b
)) {
1413 case VK_ACCESS_SHADER_WRITE_BIT
:
1414 flush_bits
|= RADV_CMD_FLAG_INV_GLOBAL_L2
;
1416 case VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT
:
1417 flush_bits
|= RADV_CMD_FLAG_FLUSH_AND_INV_CB
;
1419 case VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT
:
1420 flush_bits
|= RADV_CMD_FLAG_FLUSH_AND_INV_DB
;
1422 case VK_ACCESS_TRANSFER_WRITE_BIT
:
1423 flush_bits
|= RADV_CMD_FLAG_FLUSH_AND_INV_CB
;
1429 cmd_buffer
->state
.flush_bits
|= flush_bits
;
1432 static void radv_dst_access_flush(struct radv_cmd_buffer
*cmd_buffer
,
1433 VkAccessFlags dst_flags
)
1435 enum radv_cmd_flush_bits flush_bits
= 0;
1437 for_each_bit(b
, dst_flags
) {
1438 switch ((VkAccessFlagBits
)(1 << b
)) {
1439 case VK_ACCESS_INDIRECT_COMMAND_READ_BIT
:
1440 case VK_ACCESS_INDEX_READ_BIT
:
1441 case VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT
:
1442 flush_bits
|= RADV_CMD_FLAG_INV_VMEM_L1
;
1444 case VK_ACCESS_UNIFORM_READ_BIT
:
1445 flush_bits
|= RADV_CMD_FLAG_INV_VMEM_L1
| RADV_CMD_FLAG_INV_SMEM_L1
;
1447 case VK_ACCESS_SHADER_READ_BIT
:
1448 flush_bits
|= RADV_CMD_FLAG_INV_GLOBAL_L2
;
1450 case VK_ACCESS_COLOR_ATTACHMENT_READ_BIT
:
1451 case VK_ACCESS_TRANSFER_READ_BIT
:
1452 case VK_ACCESS_TRANSFER_WRITE_BIT
:
1453 case VK_ACCESS_INPUT_ATTACHMENT_READ_BIT
:
1454 flush_bits
|= RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER
| RADV_CMD_FLAG_INV_GLOBAL_L2
;
1459 cmd_buffer
->state
.flush_bits
|= flush_bits
;
1462 static void radv_subpass_barrier(struct radv_cmd_buffer
*cmd_buffer
, const struct radv_subpass_barrier
*barrier
)
1464 radv_src_access_flush(cmd_buffer
, barrier
->src_access_mask
);
1465 radv_stage_flush(cmd_buffer
, barrier
->src_stage_mask
);
1466 radv_dst_access_flush(cmd_buffer
, barrier
->dst_access_mask
);
1469 static void radv_handle_subpass_image_transition(struct radv_cmd_buffer
*cmd_buffer
,
1470 VkAttachmentReference att
)
1472 unsigned idx
= att
.attachment
;
1473 struct radv_image_view
*view
= cmd_buffer
->state
.framebuffer
->attachments
[idx
].attachment
;
1474 VkImageSubresourceRange range
;
1475 range
.aspectMask
= 0;
1476 range
.baseMipLevel
= view
->base_mip
;
1477 range
.levelCount
= 1;
1478 range
.baseArrayLayer
= view
->base_layer
;
1479 range
.layerCount
= cmd_buffer
->state
.framebuffer
->layers
;
1481 radv_handle_image_transition(cmd_buffer
,
1483 cmd_buffer
->state
.attachments
[idx
].current_layout
,
1484 att
.layout
, 0, 0, &range
,
1485 cmd_buffer
->state
.attachments
[idx
].pending_clear_aspects
);
1487 cmd_buffer
->state
.attachments
[idx
].current_layout
= att
.layout
;
1493 radv_cmd_buffer_set_subpass(struct radv_cmd_buffer
*cmd_buffer
,
1494 const struct radv_subpass
*subpass
, bool transitions
)
1497 radv_subpass_barrier(cmd_buffer
, &subpass
->start_barrier
);
1499 for (unsigned i
= 0; i
< subpass
->color_count
; ++i
) {
1500 radv_handle_subpass_image_transition(cmd_buffer
,
1501 subpass
->color_attachments
[i
]);
1504 for (unsigned i
= 0; i
< subpass
->input_count
; ++i
) {
1505 radv_handle_subpass_image_transition(cmd_buffer
,
1506 subpass
->input_attachments
[i
]);
1509 if (subpass
->depth_stencil_attachment
.attachment
!= VK_ATTACHMENT_UNUSED
) {
1510 radv_handle_subpass_image_transition(cmd_buffer
,
1511 subpass
->depth_stencil_attachment
);
1515 cmd_buffer
->state
.subpass
= subpass
;
1517 cmd_buffer
->state
.dirty
|= RADV_CMD_DIRTY_RENDER_TARGETS
;
1521 radv_cmd_state_setup_attachments(struct radv_cmd_buffer
*cmd_buffer
,
1522 struct radv_render_pass
*pass
,
1523 const VkRenderPassBeginInfo
*info
)
1525 struct radv_cmd_state
*state
= &cmd_buffer
->state
;
1527 if (pass
->attachment_count
== 0) {
1528 state
->attachments
= NULL
;
1532 state
->attachments
= vk_alloc(&cmd_buffer
->pool
->alloc
,
1533 pass
->attachment_count
*
1534 sizeof(state
->attachments
[0]),
1535 8, VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
1536 if (state
->attachments
== NULL
) {
1537 /* FIXME: Propagate VK_ERROR_OUT_OF_HOST_MEMORY to vkEndCommandBuffer */
1541 for (uint32_t i
= 0; i
< pass
->attachment_count
; ++i
) {
1542 struct radv_render_pass_attachment
*att
= &pass
->attachments
[i
];
1543 VkImageAspectFlags att_aspects
= vk_format_aspects(att
->format
);
1544 VkImageAspectFlags clear_aspects
= 0;
1546 if (att_aspects
== VK_IMAGE_ASPECT_COLOR_BIT
) {
1547 /* color attachment */
1548 if (att
->load_op
== VK_ATTACHMENT_LOAD_OP_CLEAR
) {
1549 clear_aspects
|= VK_IMAGE_ASPECT_COLOR_BIT
;
1552 /* depthstencil attachment */
1553 if ((att_aspects
& VK_IMAGE_ASPECT_DEPTH_BIT
) &&
1554 att
->load_op
== VK_ATTACHMENT_LOAD_OP_CLEAR
) {
1555 clear_aspects
|= VK_IMAGE_ASPECT_DEPTH_BIT
;
1557 if ((att_aspects
& VK_IMAGE_ASPECT_STENCIL_BIT
) &&
1558 att
->stencil_load_op
== VK_ATTACHMENT_LOAD_OP_CLEAR
) {
1559 clear_aspects
|= VK_IMAGE_ASPECT_STENCIL_BIT
;
1563 state
->attachments
[i
].pending_clear_aspects
= clear_aspects
;
1564 if (clear_aspects
&& info
) {
1565 assert(info
->clearValueCount
> i
);
1566 state
->attachments
[i
].clear_value
= info
->pClearValues
[i
];
1569 state
->attachments
[i
].current_layout
= att
->initial_layout
;
1573 VkResult
radv_AllocateCommandBuffers(
1575 const VkCommandBufferAllocateInfo
*pAllocateInfo
,
1576 VkCommandBuffer
*pCommandBuffers
)
1578 RADV_FROM_HANDLE(radv_device
, device
, _device
);
1579 RADV_FROM_HANDLE(radv_cmd_pool
, pool
, pAllocateInfo
->commandPool
);
1581 VkResult result
= VK_SUCCESS
;
1584 memset(pCommandBuffers
, 0,
1585 sizeof(*pCommandBuffers
)*pAllocateInfo
->commandBufferCount
);
1587 for (i
= 0; i
< pAllocateInfo
->commandBufferCount
; i
++) {
1588 result
= radv_create_cmd_buffer(device
, pool
, pAllocateInfo
->level
,
1589 &pCommandBuffers
[i
]);
1590 if (result
!= VK_SUCCESS
)
1594 if (result
!= VK_SUCCESS
)
1595 radv_FreeCommandBuffers(_device
, pAllocateInfo
->commandPool
,
1596 i
, pCommandBuffers
);
1602 radv_cmd_buffer_destroy(struct radv_cmd_buffer
*cmd_buffer
)
1604 list_del(&cmd_buffer
->pool_link
);
1606 list_for_each_entry_safe(struct radv_cmd_buffer_upload
, up
,
1607 &cmd_buffer
->upload
.list
, list
) {
1608 cmd_buffer
->device
->ws
->buffer_destroy(up
->upload_bo
);
1609 list_del(&up
->list
);
1613 if (cmd_buffer
->upload
.upload_bo
)
1614 cmd_buffer
->device
->ws
->buffer_destroy(cmd_buffer
->upload
.upload_bo
);
1615 cmd_buffer
->device
->ws
->cs_destroy(cmd_buffer
->cs
);
1616 vk_free(&cmd_buffer
->pool
->alloc
, cmd_buffer
);
1619 void radv_FreeCommandBuffers(
1621 VkCommandPool commandPool
,
1622 uint32_t commandBufferCount
,
1623 const VkCommandBuffer
*pCommandBuffers
)
1625 for (uint32_t i
= 0; i
< commandBufferCount
; i
++) {
1626 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, pCommandBuffers
[i
]);
1629 radv_cmd_buffer_destroy(cmd_buffer
);
1633 static void radv_reset_cmd_buffer(struct radv_cmd_buffer
*cmd_buffer
)
1636 cmd_buffer
->device
->ws
->cs_reset(cmd_buffer
->cs
);
1638 list_for_each_entry_safe(struct radv_cmd_buffer_upload
, up
,
1639 &cmd_buffer
->upload
.list
, list
) {
1640 cmd_buffer
->device
->ws
->buffer_destroy(up
->upload_bo
);
1641 list_del(&up
->list
);
1645 cmd_buffer
->scratch_size_needed
= 0;
1646 cmd_buffer
->compute_scratch_size_needed
= 0;
1647 cmd_buffer
->esgs_ring_size_needed
= 0;
1648 cmd_buffer
->gsvs_ring_size_needed
= 0;
1650 if (cmd_buffer
->upload
.upload_bo
)
1651 cmd_buffer
->device
->ws
->cs_add_buffer(cmd_buffer
->cs
,
1652 cmd_buffer
->upload
.upload_bo
, 8);
1653 cmd_buffer
->upload
.offset
= 0;
1655 cmd_buffer
->record_fail
= false;
1657 cmd_buffer
->ring_offsets_idx
= -1;
1660 VkResult
radv_ResetCommandBuffer(
1661 VkCommandBuffer commandBuffer
,
1662 VkCommandBufferResetFlags flags
)
1664 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
1665 radv_reset_cmd_buffer(cmd_buffer
);
1669 static void emit_gfx_buffer_state(struct radv_cmd_buffer
*cmd_buffer
)
1671 struct radv_device
*device
= cmd_buffer
->device
;
1672 if (device
->gfx_init
) {
1673 uint64_t va
= device
->ws
->buffer_get_va(device
->gfx_init
);
1674 device
->ws
->cs_add_buffer(cmd_buffer
->cs
, device
->gfx_init
, 8);
1675 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_INDIRECT_BUFFER_CIK
, 2, 0));
1676 radeon_emit(cmd_buffer
->cs
, va
);
1677 radeon_emit(cmd_buffer
->cs
, (va
>> 32) & 0xffff);
1678 radeon_emit(cmd_buffer
->cs
, device
->gfx_init_size_dw
& 0xffff);
1680 si_init_config(cmd_buffer
);
1683 VkResult
radv_BeginCommandBuffer(
1684 VkCommandBuffer commandBuffer
,
1685 const VkCommandBufferBeginInfo
*pBeginInfo
)
1687 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
1688 radv_reset_cmd_buffer(cmd_buffer
);
1690 memset(&cmd_buffer
->state
, 0, sizeof(cmd_buffer
->state
));
1692 /* setup initial configuration into command buffer */
1693 if (cmd_buffer
->level
== VK_COMMAND_BUFFER_LEVEL_PRIMARY
) {
1694 switch (cmd_buffer
->queue_family_index
) {
1695 case RADV_QUEUE_GENERAL
:
1696 /* Flush read caches at the beginning of CS not flushed by the kernel. */
1697 cmd_buffer
->state
.flush_bits
|= RADV_CMD_FLAG_INV_ICACHE
|
1698 RADV_CMD_FLAG_PS_PARTIAL_FLUSH
|
1699 RADV_CMD_FLAG_CS_PARTIAL_FLUSH
|
1700 RADV_CMD_FLAG_INV_VMEM_L1
|
1701 RADV_CMD_FLAG_INV_SMEM_L1
|
1702 RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER
|
1703 RADV_CMD_FLAG_INV_GLOBAL_L2
;
1704 emit_gfx_buffer_state(cmd_buffer
);
1705 radv_set_db_count_control(cmd_buffer
);
1706 si_emit_cache_flush(cmd_buffer
);
1708 case RADV_QUEUE_COMPUTE
:
1709 cmd_buffer
->state
.flush_bits
= RADV_CMD_FLAG_INV_ICACHE
|
1710 RADV_CMD_FLAG_CS_PARTIAL_FLUSH
|
1711 RADV_CMD_FLAG_INV_VMEM_L1
|
1712 RADV_CMD_FLAG_INV_SMEM_L1
|
1713 RADV_CMD_FLAG_INV_GLOBAL_L2
;
1714 si_init_compute(cmd_buffer
);
1715 si_emit_cache_flush(cmd_buffer
);
1717 case RADV_QUEUE_TRANSFER
:
1723 if (pBeginInfo
->flags
& VK_COMMAND_BUFFER_USAGE_RENDER_PASS_CONTINUE_BIT
) {
1724 cmd_buffer
->state
.framebuffer
= radv_framebuffer_from_handle(pBeginInfo
->pInheritanceInfo
->framebuffer
);
1725 cmd_buffer
->state
.pass
= radv_render_pass_from_handle(pBeginInfo
->pInheritanceInfo
->renderPass
);
1727 struct radv_subpass
*subpass
=
1728 &cmd_buffer
->state
.pass
->subpasses
[pBeginInfo
->pInheritanceInfo
->subpass
];
1730 radv_cmd_state_setup_attachments(cmd_buffer
, cmd_buffer
->state
.pass
, NULL
);
1731 radv_cmd_buffer_set_subpass(cmd_buffer
, subpass
, false);
1737 void radv_CmdBindVertexBuffers(
1738 VkCommandBuffer commandBuffer
,
1739 uint32_t firstBinding
,
1740 uint32_t bindingCount
,
1741 const VkBuffer
* pBuffers
,
1742 const VkDeviceSize
* pOffsets
)
1744 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
1745 struct radv_vertex_binding
*vb
= cmd_buffer
->state
.vertex_bindings
;
1747 /* We have to defer setting up vertex buffer since we need the buffer
1748 * stride from the pipeline. */
1750 assert(firstBinding
+ bindingCount
< MAX_VBS
);
1751 for (uint32_t i
= 0; i
< bindingCount
; i
++) {
1752 vb
[firstBinding
+ i
].buffer
= radv_buffer_from_handle(pBuffers
[i
]);
1753 vb
[firstBinding
+ i
].offset
= pOffsets
[i
];
1754 cmd_buffer
->state
.vb_dirty
|= 1 << (firstBinding
+ i
);
1758 void radv_CmdBindIndexBuffer(
1759 VkCommandBuffer commandBuffer
,
1761 VkDeviceSize offset
,
1762 VkIndexType indexType
)
1764 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
1766 cmd_buffer
->state
.index_buffer
= radv_buffer_from_handle(buffer
);
1767 cmd_buffer
->state
.index_offset
= offset
;
1768 cmd_buffer
->state
.index_type
= indexType
; /* vk matches hw */
1769 cmd_buffer
->state
.dirty
|= RADV_CMD_DIRTY_INDEX_BUFFER
;
1770 cmd_buffer
->device
->ws
->cs_add_buffer(cmd_buffer
->cs
, cmd_buffer
->state
.index_buffer
->bo
, 8);
1774 void radv_bind_descriptor_set(struct radv_cmd_buffer
*cmd_buffer
,
1775 struct radv_descriptor_set
*set
,
1778 struct radeon_winsys
*ws
= cmd_buffer
->device
->ws
;
1780 cmd_buffer
->state
.descriptors
[idx
] = set
;
1781 cmd_buffer
->state
.descriptors_dirty
|= (1 << idx
);
1785 for (unsigned j
= 0; j
< set
->layout
->buffer_count
; ++j
)
1786 if (set
->descriptors
[j
])
1787 ws
->cs_add_buffer(cmd_buffer
->cs
, set
->descriptors
[j
], 7);
1790 ws
->cs_add_buffer(cmd_buffer
->cs
, set
->bo
, 8);
1793 void radv_CmdBindDescriptorSets(
1794 VkCommandBuffer commandBuffer
,
1795 VkPipelineBindPoint pipelineBindPoint
,
1796 VkPipelineLayout _layout
,
1798 uint32_t descriptorSetCount
,
1799 const VkDescriptorSet
* pDescriptorSets
,
1800 uint32_t dynamicOffsetCount
,
1801 const uint32_t* pDynamicOffsets
)
1803 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
1804 RADV_FROM_HANDLE(radv_pipeline_layout
, layout
, _layout
);
1805 unsigned dyn_idx
= 0;
1807 MAYBE_UNUSED
unsigned cdw_max
= radeon_check_space(cmd_buffer
->device
->ws
,
1808 cmd_buffer
->cs
, MAX_SETS
* 4 * 6);
1810 for (unsigned i
= 0; i
< descriptorSetCount
; ++i
) {
1811 unsigned idx
= i
+ firstSet
;
1812 RADV_FROM_HANDLE(radv_descriptor_set
, set
, pDescriptorSets
[i
]);
1813 radv_bind_descriptor_set(cmd_buffer
, set
, idx
);
1815 for(unsigned j
= 0; j
< set
->layout
->dynamic_offset_count
; ++j
, ++dyn_idx
) {
1816 unsigned idx
= j
+ layout
->set
[i
].dynamic_offset_start
;
1817 uint32_t *dst
= cmd_buffer
->dynamic_buffers
+ idx
* 4;
1818 assert(dyn_idx
< dynamicOffsetCount
);
1820 struct radv_descriptor_range
*range
= set
->dynamic_descriptors
+ j
;
1821 uint64_t va
= range
->va
+ pDynamicOffsets
[dyn_idx
];
1823 dst
[1] = S_008F04_BASE_ADDRESS_HI(va
>> 32);
1824 dst
[2] = range
->size
;
1825 dst
[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X
) |
1826 S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y
) |
1827 S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z
) |
1828 S_008F0C_DST_SEL_W(V_008F0C_SQ_SEL_W
) |
1829 S_008F0C_NUM_FORMAT(V_008F0C_BUF_NUM_FORMAT_FLOAT
) |
1830 S_008F0C_DATA_FORMAT(V_008F0C_BUF_DATA_FORMAT_32
);
1831 cmd_buffer
->push_constant_stages
|=
1832 set
->layout
->dynamic_shader_stages
;
1836 assert(cmd_buffer
->cs
->cdw
<= cdw_max
);
1839 void radv_CmdPushConstants(VkCommandBuffer commandBuffer
,
1840 VkPipelineLayout layout
,
1841 VkShaderStageFlags stageFlags
,
1844 const void* pValues
)
1846 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
1847 memcpy(cmd_buffer
->push_constants
+ offset
, pValues
, size
);
1848 cmd_buffer
->push_constant_stages
|= stageFlags
;
1851 VkResult
radv_EndCommandBuffer(
1852 VkCommandBuffer commandBuffer
)
1854 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
1856 if (cmd_buffer
->queue_family_index
!= RADV_QUEUE_TRANSFER
)
1857 si_emit_cache_flush(cmd_buffer
);
1859 if (!cmd_buffer
->device
->ws
->cs_finalize(cmd_buffer
->cs
) ||
1860 cmd_buffer
->record_fail
)
1861 return VK_ERROR_OUT_OF_DEVICE_MEMORY
;
1866 radv_emit_compute_pipeline(struct radv_cmd_buffer
*cmd_buffer
)
1868 struct radeon_winsys
*ws
= cmd_buffer
->device
->ws
;
1869 struct radv_shader_variant
*compute_shader
;
1870 struct radv_pipeline
*pipeline
= cmd_buffer
->state
.compute_pipeline
;
1873 if (!pipeline
|| pipeline
== cmd_buffer
->state
.emitted_compute_pipeline
)
1876 cmd_buffer
->state
.emitted_compute_pipeline
= pipeline
;
1878 compute_shader
= pipeline
->shaders
[MESA_SHADER_COMPUTE
];
1879 va
= ws
->buffer_get_va(compute_shader
->bo
);
1881 ws
->cs_add_buffer(cmd_buffer
->cs
, compute_shader
->bo
, 8);
1883 MAYBE_UNUSED
unsigned cdw_max
= radeon_check_space(cmd_buffer
->device
->ws
,
1884 cmd_buffer
->cs
, 16);
1886 radeon_set_sh_reg_seq(cmd_buffer
->cs
, R_00B830_COMPUTE_PGM_LO
, 2);
1887 radeon_emit(cmd_buffer
->cs
, va
>> 8);
1888 radeon_emit(cmd_buffer
->cs
, va
>> 40);
1890 radeon_set_sh_reg_seq(cmd_buffer
->cs
, R_00B848_COMPUTE_PGM_RSRC1
, 2);
1891 radeon_emit(cmd_buffer
->cs
, compute_shader
->rsrc1
);
1892 radeon_emit(cmd_buffer
->cs
, compute_shader
->rsrc2
);
1895 cmd_buffer
->compute_scratch_size_needed
=
1896 MAX2(cmd_buffer
->compute_scratch_size_needed
,
1897 pipeline
->max_waves
* pipeline
->scratch_bytes_per_wave
);
1899 /* change these once we have scratch support */
1900 radeon_set_sh_reg(cmd_buffer
->cs
, R_00B860_COMPUTE_TMPRING_SIZE
,
1901 S_00B860_WAVES(pipeline
->max_waves
) |
1902 S_00B860_WAVESIZE(pipeline
->scratch_bytes_per_wave
>> 10));
1904 radeon_set_sh_reg_seq(cmd_buffer
->cs
, R_00B81C_COMPUTE_NUM_THREAD_X
, 3);
1905 radeon_emit(cmd_buffer
->cs
,
1906 S_00B81C_NUM_THREAD_FULL(compute_shader
->info
.cs
.block_size
[0]));
1907 radeon_emit(cmd_buffer
->cs
,
1908 S_00B81C_NUM_THREAD_FULL(compute_shader
->info
.cs
.block_size
[1]));
1909 radeon_emit(cmd_buffer
->cs
,
1910 S_00B81C_NUM_THREAD_FULL(compute_shader
->info
.cs
.block_size
[2]));
1912 assert(cmd_buffer
->cs
->cdw
<= cdw_max
);
1916 void radv_CmdBindPipeline(
1917 VkCommandBuffer commandBuffer
,
1918 VkPipelineBindPoint pipelineBindPoint
,
1919 VkPipeline _pipeline
)
1921 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
1922 RADV_FROM_HANDLE(radv_pipeline
, pipeline
, _pipeline
);
1924 for (unsigned i
= 0; i
< MAX_SETS
; i
++) {
1925 if (cmd_buffer
->state
.descriptors
[i
])
1926 cmd_buffer
->state
.descriptors_dirty
|= (1 << i
);
1929 switch (pipelineBindPoint
) {
1930 case VK_PIPELINE_BIND_POINT_COMPUTE
:
1931 cmd_buffer
->state
.compute_pipeline
= pipeline
;
1932 cmd_buffer
->push_constant_stages
|= VK_SHADER_STAGE_COMPUTE_BIT
;
1934 case VK_PIPELINE_BIND_POINT_GRAPHICS
:
1935 cmd_buffer
->state
.pipeline
= pipeline
;
1936 cmd_buffer
->state
.vertex_descriptors_dirty
= true;
1937 cmd_buffer
->state
.dirty
|= RADV_CMD_DIRTY_PIPELINE
;
1938 cmd_buffer
->push_constant_stages
|= pipeline
->active_stages
;
1940 /* Apply the dynamic state from the pipeline */
1941 cmd_buffer
->state
.dirty
|= pipeline
->dynamic_state_mask
;
1942 radv_dynamic_state_copy(&cmd_buffer
->state
.dynamic
,
1943 &pipeline
->dynamic_state
,
1944 pipeline
->dynamic_state_mask
);
1946 if (pipeline
->graphics
.esgs_ring_size
> cmd_buffer
->esgs_ring_size_needed
)
1947 cmd_buffer
->esgs_ring_size_needed
= pipeline
->graphics
.esgs_ring_size
;
1948 if (pipeline
->graphics
.gsvs_ring_size
> cmd_buffer
->gsvs_ring_size_needed
)
1949 cmd_buffer
->gsvs_ring_size_needed
= pipeline
->graphics
.gsvs_ring_size
;
1951 if (radv_pipeline_has_gs(pipeline
)) {
1952 struct ac_userdata_info
*loc
= radv_lookup_user_sgpr(cmd_buffer
->state
.pipeline
, MESA_SHADER_GEOMETRY
,
1953 AC_UD_SCRATCH_RING_OFFSETS
);
1954 if (cmd_buffer
->ring_offsets_idx
== -1)
1955 cmd_buffer
->ring_offsets_idx
= loc
->sgpr_idx
;
1956 else if (loc
->sgpr_idx
!= -1)
1957 assert(loc
->sgpr_idx
== cmd_buffer
->ring_offsets_idx
);
1961 assert(!"invalid bind point");
1966 void radv_CmdSetViewport(
1967 VkCommandBuffer commandBuffer
,
1968 uint32_t firstViewport
,
1969 uint32_t viewportCount
,
1970 const VkViewport
* pViewports
)
1972 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
1974 const uint32_t total_count
= firstViewport
+ viewportCount
;
1975 if (cmd_buffer
->state
.dynamic
.viewport
.count
< total_count
)
1976 cmd_buffer
->state
.dynamic
.viewport
.count
= total_count
;
1978 memcpy(cmd_buffer
->state
.dynamic
.viewport
.viewports
+ firstViewport
,
1979 pViewports
, viewportCount
* sizeof(*pViewports
));
1981 cmd_buffer
->state
.dirty
|= RADV_CMD_DIRTY_DYNAMIC_VIEWPORT
;
1984 void radv_CmdSetScissor(
1985 VkCommandBuffer commandBuffer
,
1986 uint32_t firstScissor
,
1987 uint32_t scissorCount
,
1988 const VkRect2D
* pScissors
)
1990 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
1992 const uint32_t total_count
= firstScissor
+ scissorCount
;
1993 if (cmd_buffer
->state
.dynamic
.scissor
.count
< total_count
)
1994 cmd_buffer
->state
.dynamic
.scissor
.count
= total_count
;
1996 memcpy(cmd_buffer
->state
.dynamic
.scissor
.scissors
+ firstScissor
,
1997 pScissors
, scissorCount
* sizeof(*pScissors
));
1998 cmd_buffer
->state
.dirty
|= RADV_CMD_DIRTY_DYNAMIC_SCISSOR
;
2001 void radv_CmdSetLineWidth(
2002 VkCommandBuffer commandBuffer
,
2005 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2006 cmd_buffer
->state
.dynamic
.line_width
= lineWidth
;
2007 cmd_buffer
->state
.dirty
|= RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH
;
2010 void radv_CmdSetDepthBias(
2011 VkCommandBuffer commandBuffer
,
2012 float depthBiasConstantFactor
,
2013 float depthBiasClamp
,
2014 float depthBiasSlopeFactor
)
2016 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2018 cmd_buffer
->state
.dynamic
.depth_bias
.bias
= depthBiasConstantFactor
;
2019 cmd_buffer
->state
.dynamic
.depth_bias
.clamp
= depthBiasClamp
;
2020 cmd_buffer
->state
.dynamic
.depth_bias
.slope
= depthBiasSlopeFactor
;
2022 cmd_buffer
->state
.dirty
|= RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS
;
2025 void radv_CmdSetBlendConstants(
2026 VkCommandBuffer commandBuffer
,
2027 const float blendConstants
[4])
2029 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2031 memcpy(cmd_buffer
->state
.dynamic
.blend_constants
,
2032 blendConstants
, sizeof(float) * 4);
2034 cmd_buffer
->state
.dirty
|= RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS
;
2037 void radv_CmdSetDepthBounds(
2038 VkCommandBuffer commandBuffer
,
2039 float minDepthBounds
,
2040 float maxDepthBounds
)
2042 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2044 cmd_buffer
->state
.dynamic
.depth_bounds
.min
= minDepthBounds
;
2045 cmd_buffer
->state
.dynamic
.depth_bounds
.max
= maxDepthBounds
;
2047 cmd_buffer
->state
.dirty
|= RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS
;
2050 void radv_CmdSetStencilCompareMask(
2051 VkCommandBuffer commandBuffer
,
2052 VkStencilFaceFlags faceMask
,
2053 uint32_t compareMask
)
2055 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2057 if (faceMask
& VK_STENCIL_FACE_FRONT_BIT
)
2058 cmd_buffer
->state
.dynamic
.stencil_compare_mask
.front
= compareMask
;
2059 if (faceMask
& VK_STENCIL_FACE_BACK_BIT
)
2060 cmd_buffer
->state
.dynamic
.stencil_compare_mask
.back
= compareMask
;
2062 cmd_buffer
->state
.dirty
|= RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK
;
2065 void radv_CmdSetStencilWriteMask(
2066 VkCommandBuffer commandBuffer
,
2067 VkStencilFaceFlags faceMask
,
2070 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2072 if (faceMask
& VK_STENCIL_FACE_FRONT_BIT
)
2073 cmd_buffer
->state
.dynamic
.stencil_write_mask
.front
= writeMask
;
2074 if (faceMask
& VK_STENCIL_FACE_BACK_BIT
)
2075 cmd_buffer
->state
.dynamic
.stencil_write_mask
.back
= writeMask
;
2077 cmd_buffer
->state
.dirty
|= RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK
;
2080 void radv_CmdSetStencilReference(
2081 VkCommandBuffer commandBuffer
,
2082 VkStencilFaceFlags faceMask
,
2085 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2087 if (faceMask
& VK_STENCIL_FACE_FRONT_BIT
)
2088 cmd_buffer
->state
.dynamic
.stencil_reference
.front
= reference
;
2089 if (faceMask
& VK_STENCIL_FACE_BACK_BIT
)
2090 cmd_buffer
->state
.dynamic
.stencil_reference
.back
= reference
;
2092 cmd_buffer
->state
.dirty
|= RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE
;
2096 void radv_CmdExecuteCommands(
2097 VkCommandBuffer commandBuffer
,
2098 uint32_t commandBufferCount
,
2099 const VkCommandBuffer
* pCmdBuffers
)
2101 RADV_FROM_HANDLE(radv_cmd_buffer
, primary
, commandBuffer
);
2103 for (uint32_t i
= 0; i
< commandBufferCount
; i
++) {
2104 RADV_FROM_HANDLE(radv_cmd_buffer
, secondary
, pCmdBuffers
[i
]);
2106 primary
->scratch_size_needed
= MAX2(primary
->scratch_size_needed
,
2107 secondary
->scratch_size_needed
);
2108 primary
->compute_scratch_size_needed
= MAX2(primary
->compute_scratch_size_needed
,
2109 secondary
->compute_scratch_size_needed
);
2111 if (secondary
->esgs_ring_size_needed
> primary
->esgs_ring_size_needed
)
2112 primary
->esgs_ring_size_needed
= secondary
->esgs_ring_size_needed
;
2113 if (secondary
->gsvs_ring_size_needed
> primary
->gsvs_ring_size_needed
)
2114 primary
->gsvs_ring_size_needed
= secondary
->gsvs_ring_size_needed
;
2116 if (secondary
->ring_offsets_idx
!= -1) {
2117 if (primary
->ring_offsets_idx
== -1)
2118 primary
->ring_offsets_idx
= secondary
->ring_offsets_idx
;
2120 assert(secondary
->ring_offsets_idx
== primary
->ring_offsets_idx
);
2122 primary
->device
->ws
->cs_execute_secondary(primary
->cs
, secondary
->cs
);
2125 /* if we execute secondary we need to re-emit out pipelines */
2126 if (commandBufferCount
) {
2127 primary
->state
.emitted_pipeline
= NULL
;
2128 primary
->state
.dirty
|= RADV_CMD_DIRTY_PIPELINE
;
2129 primary
->state
.dirty
|= RADV_CMD_DIRTY_DYNAMIC_ALL
;
2133 VkResult
radv_CreateCommandPool(
2135 const VkCommandPoolCreateInfo
* pCreateInfo
,
2136 const VkAllocationCallbacks
* pAllocator
,
2137 VkCommandPool
* pCmdPool
)
2139 RADV_FROM_HANDLE(radv_device
, device
, _device
);
2140 struct radv_cmd_pool
*pool
;
2142 pool
= vk_alloc2(&device
->alloc
, pAllocator
, sizeof(*pool
), 8,
2143 VK_SYSTEM_ALLOCATION_SCOPE_OBJECT
);
2145 return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY
);
2148 pool
->alloc
= *pAllocator
;
2150 pool
->alloc
= device
->alloc
;
2152 list_inithead(&pool
->cmd_buffers
);
2154 pool
->queue_family_index
= pCreateInfo
->queueFamilyIndex
;
2156 *pCmdPool
= radv_cmd_pool_to_handle(pool
);
2162 void radv_DestroyCommandPool(
2164 VkCommandPool commandPool
,
2165 const VkAllocationCallbacks
* pAllocator
)
2167 RADV_FROM_HANDLE(radv_device
, device
, _device
);
2168 RADV_FROM_HANDLE(radv_cmd_pool
, pool
, commandPool
);
2173 list_for_each_entry_safe(struct radv_cmd_buffer
, cmd_buffer
,
2174 &pool
->cmd_buffers
, pool_link
) {
2175 radv_cmd_buffer_destroy(cmd_buffer
);
2178 vk_free2(&device
->alloc
, pAllocator
, pool
);
2181 VkResult
radv_ResetCommandPool(
2183 VkCommandPool commandPool
,
2184 VkCommandPoolResetFlags flags
)
2186 RADV_FROM_HANDLE(radv_cmd_pool
, pool
, commandPool
);
2188 list_for_each_entry(struct radv_cmd_buffer
, cmd_buffer
,
2189 &pool
->cmd_buffers
, pool_link
) {
2190 radv_reset_cmd_buffer(cmd_buffer
);
2196 void radv_TrimCommandPoolKHR(
2198 VkCommandPool commandPool
,
2199 VkCommandPoolTrimFlagsKHR flags
)
2203 void radv_CmdBeginRenderPass(
2204 VkCommandBuffer commandBuffer
,
2205 const VkRenderPassBeginInfo
* pRenderPassBegin
,
2206 VkSubpassContents contents
)
2208 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2209 RADV_FROM_HANDLE(radv_render_pass
, pass
, pRenderPassBegin
->renderPass
);
2210 RADV_FROM_HANDLE(radv_framebuffer
, framebuffer
, pRenderPassBegin
->framebuffer
);
2212 MAYBE_UNUSED
unsigned cdw_max
= radeon_check_space(cmd_buffer
->device
->ws
,
2213 cmd_buffer
->cs
, 2048);
2215 cmd_buffer
->state
.framebuffer
= framebuffer
;
2216 cmd_buffer
->state
.pass
= pass
;
2217 cmd_buffer
->state
.render_area
= pRenderPassBegin
->renderArea
;
2218 radv_cmd_state_setup_attachments(cmd_buffer
, pass
, pRenderPassBegin
);
2220 si_emit_cache_flush(cmd_buffer
);
2222 radv_cmd_buffer_set_subpass(cmd_buffer
, pass
->subpasses
, true);
2223 assert(cmd_buffer
->cs
->cdw
<= cdw_max
);
2225 radv_cmd_buffer_clear_subpass(cmd_buffer
);
2228 void radv_CmdNextSubpass(
2229 VkCommandBuffer commandBuffer
,
2230 VkSubpassContents contents
)
2232 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2234 si_emit_cache_flush(cmd_buffer
);
2235 radv_cmd_buffer_resolve_subpass(cmd_buffer
);
2237 radeon_check_space(cmd_buffer
->device
->ws
, cmd_buffer
->cs
,
2240 radv_cmd_buffer_set_subpass(cmd_buffer
, cmd_buffer
->state
.subpass
+ 1, true);
2241 radv_cmd_buffer_clear_subpass(cmd_buffer
);
2245 VkCommandBuffer commandBuffer
,
2246 uint32_t vertexCount
,
2247 uint32_t instanceCount
,
2248 uint32_t firstVertex
,
2249 uint32_t firstInstance
)
2251 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2253 radv_cmd_buffer_flush_state(cmd_buffer
, (instanceCount
> 1), vertexCount
);
2255 MAYBE_UNUSED
unsigned cdw_max
= radeon_check_space(cmd_buffer
->device
->ws
, cmd_buffer
->cs
, 10);
2257 struct ac_userdata_info
*loc
= radv_lookup_user_sgpr(cmd_buffer
->state
.pipeline
, MESA_SHADER_VERTEX
,
2258 AC_UD_VS_BASE_VERTEX_START_INSTANCE
);
2259 if (loc
->sgpr_idx
!= -1) {
2260 uint32_t base_reg
= shader_stage_to_user_data_0(MESA_SHADER_VERTEX
, radv_pipeline_has_gs(cmd_buffer
->state
.pipeline
));
2261 radeon_set_sh_reg_seq(cmd_buffer
->cs
, base_reg
+ loc
->sgpr_idx
* 4, 3);
2262 radeon_emit(cmd_buffer
->cs
, firstVertex
);
2263 radeon_emit(cmd_buffer
->cs
, firstInstance
);
2264 radeon_emit(cmd_buffer
->cs
, 0);
2266 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_NUM_INSTANCES
, 0, 0));
2267 radeon_emit(cmd_buffer
->cs
, instanceCount
);
2269 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_DRAW_INDEX_AUTO
, 1, 0));
2270 radeon_emit(cmd_buffer
->cs
, vertexCount
);
2271 radeon_emit(cmd_buffer
->cs
, V_0287F0_DI_SRC_SEL_AUTO_INDEX
|
2272 S_0287F0_USE_OPAQUE(0));
2274 assert(cmd_buffer
->cs
->cdw
<= cdw_max
);
2276 radv_cmd_buffer_trace_emit(cmd_buffer
);
2279 static void radv_emit_primitive_reset_index(struct radv_cmd_buffer
*cmd_buffer
)
2281 uint32_t primitive_reset_index
= cmd_buffer
->state
.last_primitive_reset_index
? 0xffffffffu
: 0xffffu
;
2283 if (cmd_buffer
->state
.pipeline
->graphics
.prim_restart_enable
&&
2284 primitive_reset_index
!= cmd_buffer
->state
.last_primitive_reset_index
) {
2285 cmd_buffer
->state
.last_primitive_reset_index
= primitive_reset_index
;
2286 radeon_set_context_reg(cmd_buffer
->cs
, R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX
,
2287 primitive_reset_index
);
2291 void radv_CmdDrawIndexed(
2292 VkCommandBuffer commandBuffer
,
2293 uint32_t indexCount
,
2294 uint32_t instanceCount
,
2295 uint32_t firstIndex
,
2296 int32_t vertexOffset
,
2297 uint32_t firstInstance
)
2299 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2300 int index_size
= cmd_buffer
->state
.index_type
? 4 : 2;
2301 uint32_t index_max_size
= (cmd_buffer
->state
.index_buffer
->size
- cmd_buffer
->state
.index_offset
) / index_size
;
2304 radv_cmd_buffer_flush_state(cmd_buffer
, (instanceCount
> 1), indexCount
);
2305 radv_emit_primitive_reset_index(cmd_buffer
);
2307 MAYBE_UNUSED
unsigned cdw_max
= radeon_check_space(cmd_buffer
->device
->ws
, cmd_buffer
->cs
, 15);
2309 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_INDEX_TYPE
, 0, 0));
2310 radeon_emit(cmd_buffer
->cs
, cmd_buffer
->state
.index_type
);
2312 struct ac_userdata_info
*loc
= radv_lookup_user_sgpr(cmd_buffer
->state
.pipeline
, MESA_SHADER_VERTEX
,
2313 AC_UD_VS_BASE_VERTEX_START_INSTANCE
);
2314 if (loc
->sgpr_idx
!= -1) {
2315 uint32_t base_reg
= shader_stage_to_user_data_0(MESA_SHADER_VERTEX
, radv_pipeline_has_gs(cmd_buffer
->state
.pipeline
));
2316 radeon_set_sh_reg_seq(cmd_buffer
->cs
, base_reg
+ loc
->sgpr_idx
* 4, 3);
2317 radeon_emit(cmd_buffer
->cs
, vertexOffset
);
2318 radeon_emit(cmd_buffer
->cs
, firstInstance
);
2319 radeon_emit(cmd_buffer
->cs
, 0);
2321 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_NUM_INSTANCES
, 0, 0));
2322 radeon_emit(cmd_buffer
->cs
, instanceCount
);
2324 index_va
= cmd_buffer
->device
->ws
->buffer_get_va(cmd_buffer
->state
.index_buffer
->bo
);
2325 index_va
+= firstIndex
* index_size
+ cmd_buffer
->state
.index_buffer
->offset
+ cmd_buffer
->state
.index_offset
;
2326 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_DRAW_INDEX_2
, 4, false));
2327 radeon_emit(cmd_buffer
->cs
, index_max_size
);
2328 radeon_emit(cmd_buffer
->cs
, index_va
);
2329 radeon_emit(cmd_buffer
->cs
, (index_va
>> 32UL) & 0xFF);
2330 radeon_emit(cmd_buffer
->cs
, indexCount
);
2331 radeon_emit(cmd_buffer
->cs
, V_0287F0_DI_SRC_SEL_DMA
);
2333 assert(cmd_buffer
->cs
->cdw
<= cdw_max
);
2334 radv_cmd_buffer_trace_emit(cmd_buffer
);
2338 radv_emit_indirect_draw(struct radv_cmd_buffer
*cmd_buffer
,
2340 VkDeviceSize offset
,
2341 VkBuffer _count_buffer
,
2342 VkDeviceSize count_offset
,
2343 uint32_t draw_count
,
2347 RADV_FROM_HANDLE(radv_buffer
, buffer
, _buffer
);
2348 RADV_FROM_HANDLE(radv_buffer
, count_buffer
, _count_buffer
);
2349 struct radeon_winsys_cs
*cs
= cmd_buffer
->cs
;
2350 unsigned di_src_sel
= indexed
? V_0287F0_DI_SRC_SEL_DMA
2351 : V_0287F0_DI_SRC_SEL_AUTO_INDEX
;
2352 uint64_t indirect_va
= cmd_buffer
->device
->ws
->buffer_get_va(buffer
->bo
);
2353 indirect_va
+= offset
+ buffer
->offset
;
2354 uint64_t count_va
= 0;
2357 count_va
= cmd_buffer
->device
->ws
->buffer_get_va(count_buffer
->bo
);
2358 count_va
+= count_offset
+ count_buffer
->offset
;
2364 cmd_buffer
->device
->ws
->cs_add_buffer(cs
, buffer
->bo
, 8);
2366 struct ac_userdata_info
*loc
= radv_lookup_user_sgpr(cmd_buffer
->state
.pipeline
, MESA_SHADER_VERTEX
,
2367 AC_UD_VS_BASE_VERTEX_START_INSTANCE
);
2368 uint32_t base_reg
= shader_stage_to_user_data_0(MESA_SHADER_VERTEX
, radv_pipeline_has_gs(cmd_buffer
->state
.pipeline
));
2369 assert(loc
->sgpr_idx
!= -1);
2370 radeon_emit(cs
, PKT3(PKT3_SET_BASE
, 2, 0));
2372 radeon_emit(cs
, indirect_va
);
2373 radeon_emit(cs
, indirect_va
>> 32);
2375 radeon_emit(cs
, PKT3(indexed
? PKT3_DRAW_INDEX_INDIRECT_MULTI
:
2376 PKT3_DRAW_INDIRECT_MULTI
,
2379 radeon_emit(cs
, ((base_reg
+ loc
->sgpr_idx
* 4) - SI_SH_REG_OFFSET
) >> 2);
2380 radeon_emit(cs
, ((base_reg
+ (loc
->sgpr_idx
+ 1) * 4) - SI_SH_REG_OFFSET
) >> 2);
2381 radeon_emit(cs
, (((base_reg
+ (loc
->sgpr_idx
+ 2) * 4) - SI_SH_REG_OFFSET
) >> 2) |
2382 S_2C3_DRAW_INDEX_ENABLE(1) |
2383 S_2C3_COUNT_INDIRECT_ENABLE(!!count_va
));
2384 radeon_emit(cs
, draw_count
); /* count */
2385 radeon_emit(cs
, count_va
); /* count_addr */
2386 radeon_emit(cs
, count_va
>> 32);
2387 radeon_emit(cs
, stride
); /* stride */
2388 radeon_emit(cs
, di_src_sel
);
2389 radv_cmd_buffer_trace_emit(cmd_buffer
);
2393 radv_cmd_draw_indirect_count(VkCommandBuffer commandBuffer
,
2395 VkDeviceSize offset
,
2396 VkBuffer countBuffer
,
2397 VkDeviceSize countBufferOffset
,
2398 uint32_t maxDrawCount
,
2401 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2402 radv_cmd_buffer_flush_state(cmd_buffer
, true, 0);
2404 MAYBE_UNUSED
unsigned cdw_max
= radeon_check_space(cmd_buffer
->device
->ws
,
2405 cmd_buffer
->cs
, 14);
2407 radv_emit_indirect_draw(cmd_buffer
, buffer
, offset
,
2408 countBuffer
, countBufferOffset
, maxDrawCount
, stride
, false);
2410 assert(cmd_buffer
->cs
->cdw
<= cdw_max
);
2414 radv_cmd_draw_indexed_indirect_count(
2415 VkCommandBuffer commandBuffer
,
2417 VkDeviceSize offset
,
2418 VkBuffer countBuffer
,
2419 VkDeviceSize countBufferOffset
,
2420 uint32_t maxDrawCount
,
2423 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2424 int index_size
= cmd_buffer
->state
.index_type
? 4 : 2;
2425 uint32_t index_max_size
= (cmd_buffer
->state
.index_buffer
->size
- cmd_buffer
->state
.index_offset
) / index_size
;
2427 radv_cmd_buffer_flush_state(cmd_buffer
, true, 0);
2428 radv_emit_primitive_reset_index(cmd_buffer
);
2430 index_va
= cmd_buffer
->device
->ws
->buffer_get_va(cmd_buffer
->state
.index_buffer
->bo
);
2431 index_va
+= cmd_buffer
->state
.index_buffer
->offset
+ cmd_buffer
->state
.index_offset
;
2433 MAYBE_UNUSED
unsigned cdw_max
= radeon_check_space(cmd_buffer
->device
->ws
, cmd_buffer
->cs
, 21);
2435 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_INDEX_TYPE
, 0, 0));
2436 radeon_emit(cmd_buffer
->cs
, cmd_buffer
->state
.index_type
);
2438 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_INDEX_BASE
, 1, 0));
2439 radeon_emit(cmd_buffer
->cs
, index_va
);
2440 radeon_emit(cmd_buffer
->cs
, index_va
>> 32);
2442 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_INDEX_BUFFER_SIZE
, 0, 0));
2443 radeon_emit(cmd_buffer
->cs
, index_max_size
);
2445 radv_emit_indirect_draw(cmd_buffer
, buffer
, offset
,
2446 countBuffer
, countBufferOffset
, maxDrawCount
, stride
, true);
2448 assert(cmd_buffer
->cs
->cdw
<= cdw_max
);
2451 void radv_CmdDrawIndirect(
2452 VkCommandBuffer commandBuffer
,
2454 VkDeviceSize offset
,
2458 radv_cmd_draw_indirect_count(commandBuffer
, buffer
, offset
,
2459 VK_NULL_HANDLE
, 0, drawCount
, stride
);
2462 void radv_CmdDrawIndexedIndirect(
2463 VkCommandBuffer commandBuffer
,
2465 VkDeviceSize offset
,
2469 radv_cmd_draw_indexed_indirect_count(commandBuffer
, buffer
, offset
,
2470 VK_NULL_HANDLE
, 0, drawCount
, stride
);
2473 void radv_CmdDrawIndirectCountAMD(
2474 VkCommandBuffer commandBuffer
,
2476 VkDeviceSize offset
,
2477 VkBuffer countBuffer
,
2478 VkDeviceSize countBufferOffset
,
2479 uint32_t maxDrawCount
,
2482 radv_cmd_draw_indirect_count(commandBuffer
, buffer
, offset
,
2483 countBuffer
, countBufferOffset
,
2484 maxDrawCount
, stride
);
2487 void radv_CmdDrawIndexedIndirectCountAMD(
2488 VkCommandBuffer commandBuffer
,
2490 VkDeviceSize offset
,
2491 VkBuffer countBuffer
,
2492 VkDeviceSize countBufferOffset
,
2493 uint32_t maxDrawCount
,
2496 radv_cmd_draw_indexed_indirect_count(commandBuffer
, buffer
, offset
,
2497 countBuffer
, countBufferOffset
,
2498 maxDrawCount
, stride
);
2502 radv_flush_compute_state(struct radv_cmd_buffer
*cmd_buffer
)
2504 radv_emit_compute_pipeline(cmd_buffer
);
2505 radv_flush_descriptors(cmd_buffer
, cmd_buffer
->state
.compute_pipeline
,
2506 VK_SHADER_STAGE_COMPUTE_BIT
);
2507 radv_flush_constants(cmd_buffer
, cmd_buffer
->state
.compute_pipeline
,
2508 VK_SHADER_STAGE_COMPUTE_BIT
);
2509 si_emit_cache_flush(cmd_buffer
);
2512 void radv_CmdDispatch(
2513 VkCommandBuffer commandBuffer
,
2518 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2520 radv_flush_compute_state(cmd_buffer
);
2522 MAYBE_UNUSED
unsigned cdw_max
= radeon_check_space(cmd_buffer
->device
->ws
, cmd_buffer
->cs
, 10);
2524 struct ac_userdata_info
*loc
= radv_lookup_user_sgpr(cmd_buffer
->state
.compute_pipeline
,
2525 MESA_SHADER_COMPUTE
, AC_UD_CS_GRID_SIZE
);
2526 if (loc
->sgpr_idx
!= -1) {
2527 assert(!loc
->indirect
);
2528 assert(loc
->num_sgprs
== 3);
2529 radeon_set_sh_reg_seq(cmd_buffer
->cs
, R_00B900_COMPUTE_USER_DATA_0
+ loc
->sgpr_idx
* 4, 3);
2530 radeon_emit(cmd_buffer
->cs
, x
);
2531 radeon_emit(cmd_buffer
->cs
, y
);
2532 radeon_emit(cmd_buffer
->cs
, z
);
2535 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_DISPATCH_DIRECT
, 3, 0) |
2536 PKT3_SHADER_TYPE_S(1));
2537 radeon_emit(cmd_buffer
->cs
, x
);
2538 radeon_emit(cmd_buffer
->cs
, y
);
2539 radeon_emit(cmd_buffer
->cs
, z
);
2540 radeon_emit(cmd_buffer
->cs
, 1);
2542 assert(cmd_buffer
->cs
->cdw
<= cdw_max
);
2543 radv_cmd_buffer_trace_emit(cmd_buffer
);
2546 void radv_CmdDispatchIndirect(
2547 VkCommandBuffer commandBuffer
,
2549 VkDeviceSize offset
)
2551 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2552 RADV_FROM_HANDLE(radv_buffer
, buffer
, _buffer
);
2553 uint64_t va
= cmd_buffer
->device
->ws
->buffer_get_va(buffer
->bo
);
2554 va
+= buffer
->offset
+ offset
;
2556 cmd_buffer
->device
->ws
->cs_add_buffer(cmd_buffer
->cs
, buffer
->bo
, 8);
2558 radv_flush_compute_state(cmd_buffer
);
2560 MAYBE_UNUSED
unsigned cdw_max
= radeon_check_space(cmd_buffer
->device
->ws
, cmd_buffer
->cs
, 25);
2561 struct ac_userdata_info
*loc
= radv_lookup_user_sgpr(cmd_buffer
->state
.compute_pipeline
,
2562 MESA_SHADER_COMPUTE
, AC_UD_CS_GRID_SIZE
);
2563 if (loc
->sgpr_idx
!= -1) {
2564 for (unsigned i
= 0; i
< 3; ++i
) {
2565 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_COPY_DATA
, 4, 0));
2566 radeon_emit(cmd_buffer
->cs
, COPY_DATA_SRC_SEL(COPY_DATA_MEM
) |
2567 COPY_DATA_DST_SEL(COPY_DATA_REG
));
2568 radeon_emit(cmd_buffer
->cs
, (va
+ 4 * i
));
2569 radeon_emit(cmd_buffer
->cs
, (va
+ 4 * i
) >> 32);
2570 radeon_emit(cmd_buffer
->cs
, ((R_00B900_COMPUTE_USER_DATA_0
+ loc
->sgpr_idx
* 4) >> 2) + i
);
2571 radeon_emit(cmd_buffer
->cs
, 0);
2575 if (radv_cmd_buffer_uses_mec(cmd_buffer
)) {
2576 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_DISPATCH_INDIRECT
, 2, 0) |
2577 PKT3_SHADER_TYPE_S(1));
2578 radeon_emit(cmd_buffer
->cs
, va
);
2579 radeon_emit(cmd_buffer
->cs
, va
>> 32);
2580 radeon_emit(cmd_buffer
->cs
, 1);
2582 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_SET_BASE
, 2, 0) |
2583 PKT3_SHADER_TYPE_S(1));
2584 radeon_emit(cmd_buffer
->cs
, 1);
2585 radeon_emit(cmd_buffer
->cs
, va
);
2586 radeon_emit(cmd_buffer
->cs
, va
>> 32);
2588 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_DISPATCH_INDIRECT
, 1, 0) |
2589 PKT3_SHADER_TYPE_S(1));
2590 radeon_emit(cmd_buffer
->cs
, 0);
2591 radeon_emit(cmd_buffer
->cs
, 1);
2594 assert(cmd_buffer
->cs
->cdw
<= cdw_max
);
2595 radv_cmd_buffer_trace_emit(cmd_buffer
);
2598 void radv_unaligned_dispatch(
2599 struct radv_cmd_buffer
*cmd_buffer
,
2604 struct radv_pipeline
*pipeline
= cmd_buffer
->state
.compute_pipeline
;
2605 struct radv_shader_variant
*compute_shader
= pipeline
->shaders
[MESA_SHADER_COMPUTE
];
2606 uint32_t blocks
[3], remainder
[3];
2608 blocks
[0] = round_up_u32(x
, compute_shader
->info
.cs
.block_size
[0]);
2609 blocks
[1] = round_up_u32(y
, compute_shader
->info
.cs
.block_size
[1]);
2610 blocks
[2] = round_up_u32(z
, compute_shader
->info
.cs
.block_size
[2]);
2612 /* If aligned, these should be an entire block size, not 0 */
2613 remainder
[0] = x
+ compute_shader
->info
.cs
.block_size
[0] - align_u32_npot(x
, compute_shader
->info
.cs
.block_size
[0]);
2614 remainder
[1] = y
+ compute_shader
->info
.cs
.block_size
[1] - align_u32_npot(y
, compute_shader
->info
.cs
.block_size
[1]);
2615 remainder
[2] = z
+ compute_shader
->info
.cs
.block_size
[2] - align_u32_npot(z
, compute_shader
->info
.cs
.block_size
[2]);
2617 radv_flush_compute_state(cmd_buffer
);
2619 MAYBE_UNUSED
unsigned cdw_max
= radeon_check_space(cmd_buffer
->device
->ws
, cmd_buffer
->cs
, 15);
2621 radeon_set_sh_reg_seq(cmd_buffer
->cs
, R_00B81C_COMPUTE_NUM_THREAD_X
, 3);
2622 radeon_emit(cmd_buffer
->cs
,
2623 S_00B81C_NUM_THREAD_FULL(compute_shader
->info
.cs
.block_size
[0]) |
2624 S_00B81C_NUM_THREAD_PARTIAL(remainder
[0]));
2625 radeon_emit(cmd_buffer
->cs
,
2626 S_00B81C_NUM_THREAD_FULL(compute_shader
->info
.cs
.block_size
[1]) |
2627 S_00B81C_NUM_THREAD_PARTIAL(remainder
[1]));
2628 radeon_emit(cmd_buffer
->cs
,
2629 S_00B81C_NUM_THREAD_FULL(compute_shader
->info
.cs
.block_size
[2]) |
2630 S_00B81C_NUM_THREAD_PARTIAL(remainder
[2]));
2632 struct ac_userdata_info
*loc
= radv_lookup_user_sgpr(cmd_buffer
->state
.compute_pipeline
,
2633 MESA_SHADER_COMPUTE
, AC_UD_CS_GRID_SIZE
);
2634 if (loc
->sgpr_idx
!= -1) {
2635 radeon_set_sh_reg_seq(cmd_buffer
->cs
, R_00B900_COMPUTE_USER_DATA_0
+ loc
->sgpr_idx
* 4, 3);
2636 radeon_emit(cmd_buffer
->cs
, blocks
[0]);
2637 radeon_emit(cmd_buffer
->cs
, blocks
[1]);
2638 radeon_emit(cmd_buffer
->cs
, blocks
[2]);
2640 radeon_emit(cmd_buffer
->cs
, PKT3(PKT3_DISPATCH_DIRECT
, 3, 0) |
2641 PKT3_SHADER_TYPE_S(1));
2642 radeon_emit(cmd_buffer
->cs
, blocks
[0]);
2643 radeon_emit(cmd_buffer
->cs
, blocks
[1]);
2644 radeon_emit(cmd_buffer
->cs
, blocks
[2]);
2645 radeon_emit(cmd_buffer
->cs
, S_00B800_COMPUTE_SHADER_EN(1) |
2646 S_00B800_PARTIAL_TG_EN(1));
2648 assert(cmd_buffer
->cs
->cdw
<= cdw_max
);
2649 radv_cmd_buffer_trace_emit(cmd_buffer
);
2652 void radv_CmdEndRenderPass(
2653 VkCommandBuffer commandBuffer
)
2655 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2657 radv_subpass_barrier(cmd_buffer
, &cmd_buffer
->state
.pass
->end_barrier
);
2659 si_emit_cache_flush(cmd_buffer
);
2660 radv_cmd_buffer_resolve_subpass(cmd_buffer
);
2662 for (unsigned i
= 0; i
< cmd_buffer
->state
.framebuffer
->attachment_count
; ++i
) {
2663 VkImageLayout layout
= cmd_buffer
->state
.pass
->attachments
[i
].final_layout
;
2664 radv_handle_subpass_image_transition(cmd_buffer
,
2665 (VkAttachmentReference
){i
, layout
});
2668 vk_free(&cmd_buffer
->pool
->alloc
, cmd_buffer
->state
.attachments
);
2670 cmd_buffer
->state
.pass
= NULL
;
2671 cmd_buffer
->state
.subpass
= NULL
;
2672 cmd_buffer
->state
.attachments
= NULL
;
2673 cmd_buffer
->state
.framebuffer
= NULL
;
2677 static void radv_initialize_htile(struct radv_cmd_buffer
*cmd_buffer
,
2678 struct radv_image
*image
)
2681 cmd_buffer
->state
.flush_bits
|= RADV_CMD_FLAG_FLUSH_AND_INV_DB
|
2682 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
;
2684 radv_fill_buffer(cmd_buffer
, image
->bo
, image
->offset
+ image
->htile
.offset
,
2685 image
->htile
.size
, 0xffffffff);
2687 cmd_buffer
->state
.flush_bits
|= RADV_CMD_FLAG_FLUSH_AND_INV_DB_META
|
2688 RADV_CMD_FLAG_CS_PARTIAL_FLUSH
|
2689 RADV_CMD_FLAG_INV_VMEM_L1
|
2690 RADV_CMD_FLAG_INV_GLOBAL_L2
;
2693 static void radv_handle_depth_image_transition(struct radv_cmd_buffer
*cmd_buffer
,
2694 struct radv_image
*image
,
2695 VkImageLayout src_layout
,
2696 VkImageLayout dst_layout
,
2697 const VkImageSubresourceRange
*range
,
2698 VkImageAspectFlags pending_clears
)
2700 if (dst_layout
== VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL
&&
2701 (pending_clears
& vk_format_aspects(image
->vk_format
)) == vk_format_aspects(image
->vk_format
) &&
2702 cmd_buffer
->state
.render_area
.offset
.x
== 0 && cmd_buffer
->state
.render_area
.offset
.y
== 0 &&
2703 cmd_buffer
->state
.render_area
.extent
.width
== image
->extent
.width
&&
2704 cmd_buffer
->state
.render_area
.extent
.height
== image
->extent
.height
) {
2705 /* The clear will initialize htile. */
2707 } else if (src_layout
== VK_IMAGE_LAYOUT_UNDEFINED
&&
2708 radv_layout_has_htile(image
, dst_layout
)) {
2709 /* TODO: merge with the clear if applicable */
2710 radv_initialize_htile(cmd_buffer
, image
);
2711 } else if (!radv_layout_has_htile(image
, src_layout
) &&
2712 radv_layout_has_htile(image
, dst_layout
)) {
2713 radv_initialize_htile(cmd_buffer
, image
);
2714 } else if ((radv_layout_has_htile(image
, src_layout
) &&
2715 !radv_layout_has_htile(image
, dst_layout
)) ||
2716 (radv_layout_is_htile_compressed(image
, src_layout
) &&
2717 !radv_layout_is_htile_compressed(image
, dst_layout
))) {
2718 VkImageSubresourceRange local_range
= *range
;
2719 local_range
.aspectMask
= VK_IMAGE_ASPECT_DEPTH_BIT
;
2720 local_range
.baseMipLevel
= 0;
2721 local_range
.levelCount
= 1;
2723 radv_decompress_depth_image_inplace(cmd_buffer
, image
, &local_range
);
2727 void radv_initialise_cmask(struct radv_cmd_buffer
*cmd_buffer
,
2728 struct radv_image
*image
, uint32_t value
)
2730 cmd_buffer
->state
.flush_bits
|= RADV_CMD_FLAG_FLUSH_AND_INV_CB
|
2731 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
;
2733 radv_fill_buffer(cmd_buffer
, image
->bo
, image
->offset
+ image
->cmask
.offset
,
2734 image
->cmask
.size
, value
);
2736 cmd_buffer
->state
.flush_bits
|= RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
|
2737 RADV_CMD_FLAG_CS_PARTIAL_FLUSH
|
2738 RADV_CMD_FLAG_INV_VMEM_L1
|
2739 RADV_CMD_FLAG_INV_GLOBAL_L2
;
2742 static void radv_handle_cmask_image_transition(struct radv_cmd_buffer
*cmd_buffer
,
2743 struct radv_image
*image
,
2744 VkImageLayout src_layout
,
2745 VkImageLayout dst_layout
,
2746 unsigned src_queue_mask
,
2747 unsigned dst_queue_mask
,
2748 const VkImageSubresourceRange
*range
,
2749 VkImageAspectFlags pending_clears
)
2751 if (src_layout
== VK_IMAGE_LAYOUT_UNDEFINED
) {
2752 if (image
->fmask
.size
)
2753 radv_initialise_cmask(cmd_buffer
, image
, 0xccccccccu
);
2755 radv_initialise_cmask(cmd_buffer
, image
, 0xffffffffu
);
2756 } else if (radv_layout_can_fast_clear(image
, src_layout
, src_queue_mask
) &&
2757 !radv_layout_can_fast_clear(image
, dst_layout
, dst_queue_mask
)) {
2758 radv_fast_clear_flush_image_inplace(cmd_buffer
, image
, range
);
2762 void radv_initialize_dcc(struct radv_cmd_buffer
*cmd_buffer
,
2763 struct radv_image
*image
, uint32_t value
)
2766 cmd_buffer
->state
.flush_bits
|= RADV_CMD_FLAG_FLUSH_AND_INV_CB
|
2767 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
;
2769 radv_fill_buffer(cmd_buffer
, image
->bo
, image
->offset
+ image
->dcc_offset
,
2770 image
->surface
.dcc_size
, value
);
2772 cmd_buffer
->state
.flush_bits
|= RADV_CMD_FLAG_FLUSH_AND_INV_CB
|
2773 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META
|
2774 RADV_CMD_FLAG_CS_PARTIAL_FLUSH
|
2775 RADV_CMD_FLAG_INV_VMEM_L1
|
2776 RADV_CMD_FLAG_INV_GLOBAL_L2
;
2779 static void radv_handle_dcc_image_transition(struct radv_cmd_buffer
*cmd_buffer
,
2780 struct radv_image
*image
,
2781 VkImageLayout src_layout
,
2782 VkImageLayout dst_layout
,
2783 unsigned src_queue_mask
,
2784 unsigned dst_queue_mask
,
2785 const VkImageSubresourceRange
*range
,
2786 VkImageAspectFlags pending_clears
)
2788 if (src_layout
== VK_IMAGE_LAYOUT_UNDEFINED
) {
2789 radv_initialize_dcc(cmd_buffer
, image
, 0x20202020u
);
2790 } else if (radv_layout_can_fast_clear(image
, src_layout
, src_queue_mask
) &&
2791 !radv_layout_can_fast_clear(image
, dst_layout
, dst_queue_mask
)) {
2792 radv_fast_clear_flush_image_inplace(cmd_buffer
, image
, range
);
2796 static void radv_handle_image_transition(struct radv_cmd_buffer
*cmd_buffer
,
2797 struct radv_image
*image
,
2798 VkImageLayout src_layout
,
2799 VkImageLayout dst_layout
,
2800 uint32_t src_family
,
2801 uint32_t dst_family
,
2802 const VkImageSubresourceRange
*range
,
2803 VkImageAspectFlags pending_clears
)
2805 if (image
->exclusive
&& src_family
!= dst_family
) {
2806 /* This is an acquire or a release operation and there will be
2807 * a corresponding release/acquire. Do the transition in the
2808 * most flexible queue. */
2810 assert(src_family
== cmd_buffer
->queue_family_index
||
2811 dst_family
== cmd_buffer
->queue_family_index
);
2813 if (cmd_buffer
->queue_family_index
== RADV_QUEUE_TRANSFER
)
2816 if (cmd_buffer
->queue_family_index
== RADV_QUEUE_COMPUTE
&&
2817 (src_family
== RADV_QUEUE_GENERAL
||
2818 dst_family
== RADV_QUEUE_GENERAL
))
2822 unsigned src_queue_mask
= radv_image_queue_family_mask(image
, src_family
, cmd_buffer
->queue_family_index
);
2823 unsigned dst_queue_mask
= radv_image_queue_family_mask(image
, dst_family
, cmd_buffer
->queue_family_index
);
2825 if (image
->htile
.size
)
2826 radv_handle_depth_image_transition(cmd_buffer
, image
, src_layout
,
2827 dst_layout
, range
, pending_clears
);
2829 if (image
->cmask
.size
)
2830 radv_handle_cmask_image_transition(cmd_buffer
, image
, src_layout
,
2831 dst_layout
, src_queue_mask
,
2832 dst_queue_mask
, range
,
2835 if (image
->surface
.dcc_size
)
2836 radv_handle_dcc_image_transition(cmd_buffer
, image
, src_layout
,
2837 dst_layout
, src_queue_mask
,
2838 dst_queue_mask
, range
,
2842 void radv_CmdPipelineBarrier(
2843 VkCommandBuffer commandBuffer
,
2844 VkPipelineStageFlags srcStageMask
,
2845 VkPipelineStageFlags destStageMask
,
2847 uint32_t memoryBarrierCount
,
2848 const VkMemoryBarrier
* pMemoryBarriers
,
2849 uint32_t bufferMemoryBarrierCount
,
2850 const VkBufferMemoryBarrier
* pBufferMemoryBarriers
,
2851 uint32_t imageMemoryBarrierCount
,
2852 const VkImageMemoryBarrier
* pImageMemoryBarriers
)
2854 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2855 VkAccessFlags src_flags
= 0;
2856 VkAccessFlags dst_flags
= 0;
2858 for (uint32_t i
= 0; i
< memoryBarrierCount
; i
++) {
2859 src_flags
|= pMemoryBarriers
[i
].srcAccessMask
;
2860 dst_flags
|= pMemoryBarriers
[i
].dstAccessMask
;
2863 for (uint32_t i
= 0; i
< bufferMemoryBarrierCount
; i
++) {
2864 src_flags
|= pBufferMemoryBarriers
[i
].srcAccessMask
;
2865 dst_flags
|= pBufferMemoryBarriers
[i
].dstAccessMask
;
2868 for (uint32_t i
= 0; i
< imageMemoryBarrierCount
; i
++) {
2869 src_flags
|= pImageMemoryBarriers
[i
].srcAccessMask
;
2870 dst_flags
|= pImageMemoryBarriers
[i
].dstAccessMask
;
2873 radv_src_access_flush(cmd_buffer
, src_flags
);
2875 for (uint32_t i
= 0; i
< imageMemoryBarrierCount
; i
++) {
2876 RADV_FROM_HANDLE(radv_image
, image
, pImageMemoryBarriers
[i
].image
);
2877 radv_handle_image_transition(cmd_buffer
, image
,
2878 pImageMemoryBarriers
[i
].oldLayout
,
2879 pImageMemoryBarriers
[i
].newLayout
,
2880 pImageMemoryBarriers
[i
].srcQueueFamilyIndex
,
2881 pImageMemoryBarriers
[i
].dstQueueFamilyIndex
,
2882 &pImageMemoryBarriers
[i
].subresourceRange
,
2886 radv_dst_access_flush(cmd_buffer
, src_flags
);
2888 /* TODO reduce this */
2889 enum radv_cmd_flush_bits flush_bits
= RADV_CMD_FLAG_CS_PARTIAL_FLUSH
|
2890 RADV_CMD_FLAG_PS_PARTIAL_FLUSH
;
2892 cmd_buffer
->state
.flush_bits
|= flush_bits
;
2896 static void write_event(struct radv_cmd_buffer
*cmd_buffer
,
2897 struct radv_event
*event
,
2898 VkPipelineStageFlags stageMask
,
2901 struct radeon_winsys_cs
*cs
= cmd_buffer
->cs
;
2902 uint64_t va
= cmd_buffer
->device
->ws
->buffer_get_va(event
->bo
);
2904 cmd_buffer
->device
->ws
->cs_add_buffer(cs
, event
->bo
, 8);
2906 MAYBE_UNUSED
unsigned cdw_max
= radeon_check_space(cmd_buffer
->device
->ws
, cs
, 12);
2908 /* TODO: this is overkill. Probably should figure something out from
2909 * the stage mask. */
2911 if (cmd_buffer
->device
->physical_device
->rad_info
.chip_class
== CIK
) {
2912 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE_EOP
, 4, 0));
2913 radeon_emit(cs
, EVENT_TYPE(EVENT_TYPE_BOTTOM_OF_PIPE_TS
) |
2915 radeon_emit(cs
, va
);
2916 radeon_emit(cs
, (va
>> 32) | EOP_DATA_SEL(1));
2921 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE_EOP
, 4, 0));
2922 radeon_emit(cs
, EVENT_TYPE(EVENT_TYPE_BOTTOM_OF_PIPE_TS
) |
2924 radeon_emit(cs
, va
);
2925 radeon_emit(cs
, (va
>> 32) | EOP_DATA_SEL(1));
2926 radeon_emit(cs
, value
);
2929 assert(cmd_buffer
->cs
->cdw
<= cdw_max
);
2932 void radv_CmdSetEvent(VkCommandBuffer commandBuffer
,
2934 VkPipelineStageFlags stageMask
)
2936 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2937 RADV_FROM_HANDLE(radv_event
, event
, _event
);
2939 write_event(cmd_buffer
, event
, stageMask
, 1);
2942 void radv_CmdResetEvent(VkCommandBuffer commandBuffer
,
2944 VkPipelineStageFlags stageMask
)
2946 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2947 RADV_FROM_HANDLE(radv_event
, event
, _event
);
2949 write_event(cmd_buffer
, event
, stageMask
, 0);
2952 void radv_CmdWaitEvents(VkCommandBuffer commandBuffer
,
2953 uint32_t eventCount
,
2954 const VkEvent
* pEvents
,
2955 VkPipelineStageFlags srcStageMask
,
2956 VkPipelineStageFlags dstStageMask
,
2957 uint32_t memoryBarrierCount
,
2958 const VkMemoryBarrier
* pMemoryBarriers
,
2959 uint32_t bufferMemoryBarrierCount
,
2960 const VkBufferMemoryBarrier
* pBufferMemoryBarriers
,
2961 uint32_t imageMemoryBarrierCount
,
2962 const VkImageMemoryBarrier
* pImageMemoryBarriers
)
2964 RADV_FROM_HANDLE(radv_cmd_buffer
, cmd_buffer
, commandBuffer
);
2965 struct radeon_winsys_cs
*cs
= cmd_buffer
->cs
;
2967 for (unsigned i
= 0; i
< eventCount
; ++i
) {
2968 RADV_FROM_HANDLE(radv_event
, event
, pEvents
[i
]);
2969 uint64_t va
= cmd_buffer
->device
->ws
->buffer_get_va(event
->bo
);
2971 cmd_buffer
->device
->ws
->cs_add_buffer(cs
, event
->bo
, 8);
2973 MAYBE_UNUSED
unsigned cdw_max
= radeon_check_space(cmd_buffer
->device
->ws
, cs
, 7);
2975 radeon_emit(cs
, PKT3(PKT3_WAIT_REG_MEM
, 5, 0));
2976 radeon_emit(cs
, WAIT_REG_MEM_EQUAL
| WAIT_REG_MEM_MEM_SPACE(1));
2977 radeon_emit(cs
, va
);
2978 radeon_emit(cs
, va
>> 32);
2979 radeon_emit(cs
, 1); /* reference value */
2980 radeon_emit(cs
, 0xffffffff); /* mask */
2981 radeon_emit(cs
, 4); /* poll interval */
2983 assert(cmd_buffer
->cs
->cdw
<= cdw_max
);
2987 for (uint32_t i
= 0; i
< imageMemoryBarrierCount
; i
++) {
2988 RADV_FROM_HANDLE(radv_image
, image
, pImageMemoryBarriers
[i
].image
);
2990 radv_handle_image_transition(cmd_buffer
, image
,
2991 pImageMemoryBarriers
[i
].oldLayout
,
2992 pImageMemoryBarriers
[i
].newLayout
,
2993 pImageMemoryBarriers
[i
].srcQueueFamilyIndex
,
2994 pImageMemoryBarriers
[i
].dstQueueFamilyIndex
,
2995 &pImageMemoryBarriers
[i
].subresourceRange
,
2999 /* TODO: figure out how to do memory barriers without waiting */
3000 cmd_buffer
->state
.flush_bits
|= RADV_CMD_FLUSH_AND_INV_FRAMEBUFFER
|
3001 RADV_CMD_FLAG_INV_GLOBAL_L2
|
3002 RADV_CMD_FLAG_INV_VMEM_L1
|
3003 RADV_CMD_FLAG_INV_SMEM_L1
;