radeonsi: import r600_streamout from drivers/radeon
[mesa.git] / src / gallium / drivers / radeonsi / si_pipe.h
1 /*
2 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
22 *
23 * Authors:
24 * Jerome Glisse
25 */
26 #ifndef SI_PIPE_H
27 #define SI_PIPE_H
28
29 #include "si_shader.h"
30
31 #include "util/u_dynarray.h"
32 #include "util/u_idalloc.h"
33
34 #ifdef PIPE_ARCH_BIG_ENDIAN
35 #define SI_BIG_ENDIAN 1
36 #else
37 #define SI_BIG_ENDIAN 0
38 #endif
39
40 /* The base vertex and primitive restart can be any number, but we must pick
41 * one which will mean "unknown" for the purpose of state tracking and
42 * the number shouldn't be a commonly-used one. */
43 #define SI_BASE_VERTEX_UNKNOWN INT_MIN
44 #define SI_RESTART_INDEX_UNKNOWN INT_MIN
45 #define SI_NUM_SMOOTH_AA_SAMPLES 8
46 #define SI_GS_PER_ES 128
47 /* Alignment for optimal CP DMA performance. */
48 #define SI_CPDMA_ALIGNMENT 32
49
50 /* Instruction cache. */
51 #define SI_CONTEXT_INV_ICACHE (R600_CONTEXT_PRIVATE_FLAG << 0)
52 /* SMEM L1, other names: KCACHE, constant cache, DCACHE, data cache */
53 #define SI_CONTEXT_INV_SMEM_L1 (R600_CONTEXT_PRIVATE_FLAG << 1)
54 /* VMEM L1 can optionally be bypassed (GLC=1). Other names: TC L1 */
55 #define SI_CONTEXT_INV_VMEM_L1 (R600_CONTEXT_PRIVATE_FLAG << 2)
56 /* Used by everything except CB/DB, can be bypassed (SLC=1). Other names: TC L2 */
57 #define SI_CONTEXT_INV_GLOBAL_L2 (R600_CONTEXT_PRIVATE_FLAG << 3)
58 /* Write dirty L2 lines back to memory (shader and CP DMA stores), but don't
59 * invalidate L2. SI-CIK can't do it, so they will do complete invalidation. */
60 #define SI_CONTEXT_WRITEBACK_GLOBAL_L2 (R600_CONTEXT_PRIVATE_FLAG << 4)
61 /* Writeback & invalidate the L2 metadata cache. It can only be coupled with
62 * a CB or DB flush. */
63 #define SI_CONTEXT_INV_L2_METADATA (R600_CONTEXT_PRIVATE_FLAG << 5)
64 /* Framebuffer caches. */
65 #define SI_CONTEXT_FLUSH_AND_INV_DB (R600_CONTEXT_PRIVATE_FLAG << 6)
66 #define SI_CONTEXT_FLUSH_AND_INV_DB_META (R600_CONTEXT_PRIVATE_FLAG << 7)
67 #define SI_CONTEXT_FLUSH_AND_INV_CB (R600_CONTEXT_PRIVATE_FLAG << 8)
68 /* Engine synchronization. */
69 #define SI_CONTEXT_VS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 9)
70 #define SI_CONTEXT_PS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 10)
71 #define SI_CONTEXT_CS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 11)
72 #define SI_CONTEXT_VGT_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 12)
73 #define SI_CONTEXT_VGT_STREAMOUT_SYNC (R600_CONTEXT_PRIVATE_FLAG << 13)
74
75 #define SI_PREFETCH_VBO_DESCRIPTORS (1 << 0)
76 #define SI_PREFETCH_LS (1 << 1)
77 #define SI_PREFETCH_HS (1 << 2)
78 #define SI_PREFETCH_ES (1 << 3)
79 #define SI_PREFETCH_GS (1 << 4)
80 #define SI_PREFETCH_VS (1 << 5)
81 #define SI_PREFETCH_PS (1 << 6)
82
83 #define SI_MAX_BORDER_COLORS 4096
84 #define SI_MAX_VIEWPORTS 16
85 #define SIX_BITS 0x3F
86
87 struct si_compute;
88 struct hash_table;
89 struct u_suballocator;
90
91 struct si_screen {
92 struct r600_common_screen b;
93 unsigned gs_table_depth;
94 unsigned tess_offchip_block_dw_size;
95 bool has_clear_state;
96 bool has_distributed_tess;
97 bool has_draw_indirect_multi;
98 bool has_out_of_order_rast;
99 bool assume_no_z_fights;
100 bool commutative_blend_add;
101 bool clear_db_meta_before_clear;
102 bool has_msaa_sample_loc_bug;
103 bool dpbb_allowed;
104 bool dfsm_allowed;
105 bool llvm_has_working_vgpr_indexing;
106
107 /* Whether shaders are monolithic (1-part) or separate (3-part). */
108 bool use_monolithic_shaders;
109 bool record_llvm_ir;
110
111 mtx_t shader_parts_mutex;
112 struct si_shader_part *vs_prologs;
113 struct si_shader_part *tcs_epilogs;
114 struct si_shader_part *gs_prologs;
115 struct si_shader_part *ps_prologs;
116 struct si_shader_part *ps_epilogs;
117
118 /* Shader cache in memory.
119 *
120 * Design & limitations:
121 * - The shader cache is per screen (= per process), never saved to
122 * disk, and skips redundant shader compilations from TGSI to bytecode.
123 * - It can only be used with one-variant-per-shader support, in which
124 * case only the main (typically middle) part of shaders is cached.
125 * - Only VS, TCS, TES, PS are cached, out of which only the hw VS
126 * variants of VS and TES are cached, so LS and ES aren't.
127 * - GS and CS aren't cached, but it's certainly possible to cache
128 * those as well.
129 */
130 mtx_t shader_cache_mutex;
131 struct hash_table *shader_cache;
132
133 /* Shader compiler queue for multithreaded compilation. */
134 struct util_queue shader_compiler_queue;
135 /* Use at most 3 normal compiler threads on quadcore and better.
136 * Hyperthreaded CPUs report the number of threads, but we want
137 * the number of cores. */
138 LLVMTargetMachineRef tm[3]; /* used by the queue only */
139
140 struct util_queue shader_compiler_queue_low_priority;
141 /* Use at most 2 low priority threads on quadcore and better.
142 * We want to minimize the impact on multithreaded Mesa. */
143 LLVMTargetMachineRef tm_low_priority[2]; /* at most 2 threads */
144 };
145
146 struct si_blend_color {
147 struct r600_atom atom;
148 struct pipe_blend_color state;
149 bool any_nonzeros;
150 };
151
152 struct si_sampler_view {
153 struct pipe_sampler_view base;
154 /* [0..7] = image descriptor
155 * [4..7] = buffer descriptor */
156 uint32_t state[8];
157 uint32_t fmask_state[8];
158 const struct legacy_surf_level *base_level_info;
159 ubyte base_level;
160 ubyte block_width;
161 bool is_stencil_sampler;
162 bool is_integer;
163 bool dcc_incompatible;
164 };
165
166 #define SI_SAMPLER_STATE_MAGIC 0x34f1c35a
167
168 struct si_sampler_state {
169 #ifdef DEBUG
170 unsigned magic;
171 #endif
172 uint32_t val[4];
173 uint32_t integer_val[4];
174 uint32_t upgraded_depth_val[4];
175 };
176
177 struct si_cs_shader_state {
178 struct si_compute *program;
179 struct si_compute *emitted_program;
180 unsigned offset;
181 bool initialized;
182 bool uses_scratch;
183 };
184
185 struct si_samplers {
186 struct pipe_sampler_view *views[SI_NUM_SAMPLERS];
187 struct si_sampler_state *sampler_states[SI_NUM_SAMPLERS];
188
189 /* The i-th bit is set if that element is enabled (non-NULL resource). */
190 unsigned enabled_mask;
191 uint32_t needs_depth_decompress_mask;
192 uint32_t needs_color_decompress_mask;
193 };
194
195 struct si_images {
196 struct pipe_image_view views[SI_NUM_IMAGES];
197 uint32_t needs_color_decompress_mask;
198 unsigned enabled_mask;
199 };
200
201 struct si_framebuffer {
202 struct r600_atom atom;
203 struct pipe_framebuffer_state state;
204 unsigned colorbuf_enabled_4bit;
205 unsigned spi_shader_col_format;
206 unsigned spi_shader_col_format_alpha;
207 unsigned spi_shader_col_format_blend;
208 unsigned spi_shader_col_format_blend_alpha;
209 ubyte nr_samples:5; /* at most 16xAA */
210 ubyte log_samples:3; /* at most 4 = 16xAA */
211 ubyte compressed_cb_mask;
212 ubyte color_is_int8;
213 ubyte color_is_int10;
214 ubyte dirty_cbufs;
215 bool dirty_zsbuf;
216 bool any_dst_linear;
217 bool CB_has_shader_readable_metadata;
218 bool DB_has_shader_readable_metadata;
219 };
220
221 struct si_signed_scissor {
222 int minx;
223 int miny;
224 int maxx;
225 int maxy;
226 };
227
228 struct si_scissors {
229 struct r600_atom atom;
230 unsigned dirty_mask;
231 struct pipe_scissor_state states[SI_MAX_VIEWPORTS];
232 };
233
234 struct si_viewports {
235 struct r600_atom atom;
236 unsigned dirty_mask;
237 unsigned depth_range_dirty_mask;
238 struct pipe_viewport_state states[SI_MAX_VIEWPORTS];
239 struct si_signed_scissor as_scissor[SI_MAX_VIEWPORTS];
240 };
241
242 struct si_clip_state {
243 struct r600_atom atom;
244 struct pipe_clip_state state;
245 bool any_nonzeros;
246 };
247
248 struct si_sample_locs {
249 struct r600_atom atom;
250 unsigned nr_samples;
251 };
252
253 struct si_sample_mask {
254 struct r600_atom atom;
255 uint16_t sample_mask;
256 };
257
258 struct si_streamout_target {
259 struct pipe_stream_output_target b;
260
261 /* The buffer where BUFFER_FILLED_SIZE is stored. */
262 struct r600_resource *buf_filled_size;
263 unsigned buf_filled_size_offset;
264 bool buf_filled_size_valid;
265
266 unsigned stride_in_dw;
267 };
268
269 struct si_streamout {
270 struct r600_atom begin_atom;
271 bool begin_emitted;
272
273 unsigned enabled_mask;
274 unsigned num_targets;
275 struct si_streamout_target *targets[PIPE_MAX_SO_BUFFERS];
276
277 unsigned append_bitmask;
278 bool suspended;
279
280 /* External state which comes from the vertex shader,
281 * it must be set explicitly when binding a shader. */
282 uint16_t *stride_in_dw;
283 unsigned enabled_stream_buffers_mask; /* stream0 buffers0-3 in 4 LSB */
284
285 /* The state of VGT_STRMOUT_BUFFER_(CONFIG|EN). */
286 unsigned hw_enabled_mask;
287
288 /* The state of VGT_STRMOUT_(CONFIG|EN). */
289 struct r600_atom enable_atom;
290 bool streamout_enabled;
291 bool prims_gen_query_enabled;
292 int num_prims_gen_queries;
293 };
294
295 /* A shader state consists of the shader selector, which is a constant state
296 * object shared by multiple contexts and shouldn't be modified, and
297 * the current shader variant selected for this context.
298 */
299 struct si_shader_ctx_state {
300 struct si_shader_selector *cso;
301 struct si_shader *current;
302 };
303
304 #define SI_NUM_VGT_PARAM_KEY_BITS 12
305 #define SI_NUM_VGT_PARAM_STATES (1 << SI_NUM_VGT_PARAM_KEY_BITS)
306
307 /* The IA_MULTI_VGT_PARAM key used to index the table of precomputed values.
308 * Some fields are set by state-change calls, most are set by draw_vbo.
309 */
310 union si_vgt_param_key {
311 struct {
312 unsigned prim:4;
313 unsigned uses_instancing:1;
314 unsigned multi_instances_smaller_than_primgroup:1;
315 unsigned primitive_restart:1;
316 unsigned count_from_stream_output:1;
317 unsigned line_stipple_enabled:1;
318 unsigned uses_tess:1;
319 unsigned tess_uses_prim_id:1;
320 unsigned uses_gs:1;
321 unsigned _pad:32 - SI_NUM_VGT_PARAM_KEY_BITS;
322 } u;
323 uint32_t index;
324 };
325
326 struct si_texture_handle
327 {
328 unsigned desc_slot;
329 bool desc_dirty;
330 struct pipe_sampler_view *view;
331 struct si_sampler_state sstate;
332 };
333
334 struct si_image_handle
335 {
336 unsigned desc_slot;
337 bool desc_dirty;
338 struct pipe_image_view view;
339 };
340
341 struct si_saved_cs {
342 struct pipe_reference reference;
343 struct si_context *ctx;
344 struct radeon_saved_cs gfx;
345 struct r600_resource *trace_buf;
346 unsigned trace_id;
347
348 unsigned gfx_last_dw;
349 bool flushed;
350 };
351
352 struct si_context {
353 struct r600_common_context b;
354 struct blitter_context *blitter;
355 void *custom_dsa_flush;
356 void *custom_blend_resolve;
357 void *custom_blend_fmask_decompress;
358 void *custom_blend_eliminate_fastclear;
359 void *custom_blend_dcc_decompress;
360 void *vs_blit_pos;
361 void *vs_blit_pos_layered;
362 void *vs_blit_color;
363 void *vs_blit_color_layered;
364 void *vs_blit_texcoord;
365 struct si_screen *screen;
366 LLVMTargetMachineRef tm; /* only non-threaded compilation */
367 struct si_shader_ctx_state fixed_func_tcs_shader;
368 struct r600_resource *wait_mem_scratch;
369 unsigned wait_mem_number;
370 uint16_t prefetch_L2_mask;
371
372 bool gfx_flush_in_progress:1;
373 bool compute_is_busy:1;
374
375 /* Atoms (direct states). */
376 union si_state_atoms atoms;
377 unsigned dirty_atoms; /* mask */
378 /* PM4 states (precomputed immutable states) */
379 unsigned dirty_states;
380 union si_state queued;
381 union si_state emitted;
382
383 /* Atom declarations. */
384 struct si_framebuffer framebuffer;
385 struct si_sample_locs msaa_sample_locs;
386 struct r600_atom db_render_state;
387 struct r600_atom dpbb_state;
388 struct r600_atom msaa_config;
389 struct si_sample_mask sample_mask;
390 struct r600_atom cb_render_state;
391 unsigned last_cb_target_mask;
392 struct si_blend_color blend_color;
393 struct r600_atom clip_regs;
394 struct si_clip_state clip_state;
395 struct si_shader_data shader_pointers;
396 struct si_stencil_ref stencil_ref;
397 struct r600_atom spi_map;
398 struct si_scissors scissors;
399 struct si_streamout streamout;
400 struct si_viewports viewports;
401
402 /* Precomputed states. */
403 struct si_pm4_state *init_config;
404 struct si_pm4_state *init_config_gs_rings;
405 bool init_config_has_vgt_flush;
406 struct si_pm4_state *vgt_shader_config[4];
407
408 /* shaders */
409 struct si_shader_ctx_state ps_shader;
410 struct si_shader_ctx_state gs_shader;
411 struct si_shader_ctx_state vs_shader;
412 struct si_shader_ctx_state tcs_shader;
413 struct si_shader_ctx_state tes_shader;
414 struct si_cs_shader_state cs_shader_state;
415
416 /* shader information */
417 struct si_vertex_elements *vertex_elements;
418 unsigned sprite_coord_enable;
419 bool flatshade;
420 bool do_update_shaders;
421
422 /* shader descriptors */
423 struct si_descriptors vertex_buffers;
424 struct si_descriptors descriptors[SI_NUM_DESCS];
425 unsigned descriptors_dirty;
426 unsigned shader_pointers_dirty;
427 unsigned shader_needs_decompress_mask;
428 struct si_buffer_resources rw_buffers;
429 struct si_buffer_resources const_and_shader_buffers[SI_NUM_SHADERS];
430 struct si_samplers samplers[SI_NUM_SHADERS];
431 struct si_images images[SI_NUM_SHADERS];
432
433 /* other shader resources */
434 struct pipe_constant_buffer null_const_buf; /* used for set_constant_buffer(NULL) on CIK */
435 struct pipe_resource *esgs_ring;
436 struct pipe_resource *gsvs_ring;
437 struct pipe_resource *tf_ring;
438 struct pipe_resource *tess_offchip_ring;
439 union pipe_color_union *border_color_table; /* in CPU memory, any endian */
440 struct r600_resource *border_color_buffer;
441 union pipe_color_union *border_color_map; /* in VRAM (slow access), little endian */
442 unsigned border_color_count;
443 unsigned num_vs_blit_sgprs;
444 uint32_t vs_blit_sh_data[SI_VS_BLIT_SGPRS_POS_TEXCOORD];
445
446 /* Vertex and index buffers. */
447 bool vertex_buffers_dirty;
448 bool vertex_buffer_pointer_dirty;
449 struct pipe_vertex_buffer vertex_buffer[SI_NUM_VERTEX_BUFFERS];
450
451 /* MSAA config state. */
452 int ps_iter_samples;
453 bool smoothing_enabled;
454
455 /* DB render state. */
456 unsigned ps_db_shader_control;
457 unsigned dbcb_copy_sample;
458 bool dbcb_depth_copy_enabled:1;
459 bool dbcb_stencil_copy_enabled:1;
460 bool db_flush_depth_inplace:1;
461 bool db_flush_stencil_inplace:1;
462 bool db_depth_clear:1;
463 bool db_depth_disable_expclear:1;
464 bool db_stencil_clear:1;
465 bool db_stencil_disable_expclear:1;
466 bool occlusion_queries_disabled:1;
467 bool generate_mipmap_for_depth:1;
468
469 /* Emitted draw state. */
470 bool gs_tri_strip_adj_fix:1;
471 bool ls_vgpr_fix:1;
472 int last_index_size;
473 int last_base_vertex;
474 int last_start_instance;
475 int last_drawid;
476 int last_sh_base_reg;
477 int last_primitive_restart_en;
478 int last_restart_index;
479 int last_gs_out_prim;
480 int last_prim;
481 int last_multi_vgt_param;
482 int last_rast_prim;
483 unsigned last_sc_line_stipple;
484 unsigned current_vs_state;
485 unsigned last_vs_state;
486 enum pipe_prim_type current_rast_prim; /* primitive type after TES, GS */
487
488 /* Scratch buffer */
489 struct r600_atom scratch_state;
490 struct r600_resource *scratch_buffer;
491 unsigned scratch_waves;
492 unsigned spi_tmpring_size;
493
494 struct r600_resource *compute_scratch_buffer;
495
496 /* Emitted derived tessellation state. */
497 /* Local shader (VS), or HS if LS-HS are merged. */
498 struct si_shader *last_ls;
499 struct si_shader_selector *last_tcs;
500 int last_num_tcs_input_cp;
501 int last_tes_sh_base;
502 bool last_tess_uses_primid;
503 unsigned last_num_patches;
504
505 /* Debug state. */
506 bool is_debug;
507 struct si_saved_cs *current_saved_cs;
508 uint64_t dmesg_timestamp;
509 unsigned apitrace_call_number;
510
511 /* Other state */
512 bool need_check_render_feedback;
513 bool decompression_enabled;
514
515 bool vs_writes_viewport_index;
516 bool vs_disables_clipping_viewport;
517
518 /* Precomputed IA_MULTI_VGT_PARAM */
519 union si_vgt_param_key ia_multi_vgt_param_key;
520 unsigned ia_multi_vgt_param[SI_NUM_VGT_PARAM_STATES];
521
522 /* Bindless descriptors. */
523 struct si_descriptors bindless_descriptors;
524 struct util_idalloc bindless_used_slots;
525 unsigned num_bindless_descriptors;
526 bool bindless_descriptors_dirty;
527 bool graphics_bindless_pointer_dirty;
528 bool compute_bindless_pointer_dirty;
529
530 /* Allocated bindless handles */
531 struct hash_table *tex_handles;
532 struct hash_table *img_handles;
533
534 /* Resident bindless handles */
535 struct util_dynarray resident_tex_handles;
536 struct util_dynarray resident_img_handles;
537
538 /* Resident bindless handles which need decompression */
539 struct util_dynarray resident_tex_needs_color_decompress;
540 struct util_dynarray resident_img_needs_color_decompress;
541 struct util_dynarray resident_tex_needs_depth_decompress;
542
543 /* Bindless state */
544 bool uses_bindless_samplers;
545 bool uses_bindless_images;
546 };
547
548 /* cik_sdma.c */
549 void cik_init_sdma_functions(struct si_context *sctx);
550
551 /* si_blit.c */
552 void si_init_blit_functions(struct si_context *sctx);
553 void si_decompress_textures(struct si_context *sctx, unsigned shader_mask);
554 void si_resource_copy_region(struct pipe_context *ctx,
555 struct pipe_resource *dst,
556 unsigned dst_level,
557 unsigned dstx, unsigned dsty, unsigned dstz,
558 struct pipe_resource *src,
559 unsigned src_level,
560 const struct pipe_box *src_box);
561
562 /* si_cp_dma.c */
563 #define SI_CPDMA_SKIP_CHECK_CS_SPACE (1 << 0) /* don't call need_cs_space */
564 #define SI_CPDMA_SKIP_SYNC_AFTER (1 << 1) /* don't wait for DMA after the copy */
565 #define SI_CPDMA_SKIP_SYNC_BEFORE (1 << 2) /* don't wait for DMA before the copy (RAW hazards) */
566 #define SI_CPDMA_SKIP_GFX_SYNC (1 << 3) /* don't flush caches and don't wait for PS/CS */
567 #define SI_CPDMA_SKIP_BO_LIST_UPDATE (1 << 4) /* don't update the BO list */
568 #define SI_CPDMA_SKIP_ALL (SI_CPDMA_SKIP_CHECK_CS_SPACE | \
569 SI_CPDMA_SKIP_SYNC_AFTER | \
570 SI_CPDMA_SKIP_SYNC_BEFORE | \
571 SI_CPDMA_SKIP_GFX_SYNC | \
572 SI_CPDMA_SKIP_BO_LIST_UPDATE)
573
574 void si_copy_buffer(struct si_context *sctx,
575 struct pipe_resource *dst, struct pipe_resource *src,
576 uint64_t dst_offset, uint64_t src_offset, unsigned size,
577 unsigned user_flags);
578 void cik_prefetch_TC_L2_async(struct si_context *sctx, struct pipe_resource *buf,
579 uint64_t offset, unsigned size);
580 void cik_emit_prefetch_L2(struct si_context *sctx);
581 void si_init_cp_dma_functions(struct si_context *sctx);
582
583 /* si_debug.c */
584 void si_auto_log_cs(void *data, struct u_log_context *log);
585 void si_log_hw_flush(struct si_context *sctx);
586 void si_log_draw_state(struct si_context *sctx, struct u_log_context *log);
587 void si_log_compute_state(struct si_context *sctx, struct u_log_context *log);
588 void si_init_debug_functions(struct si_context *sctx);
589 void si_check_vm_faults(struct r600_common_context *ctx,
590 struct radeon_saved_cs *saved, enum ring_type ring);
591 bool si_replace_shader(unsigned num, struct ac_shader_binary *binary);
592
593 /* si_dma.c */
594 void si_init_dma_functions(struct si_context *sctx);
595
596 /* si_hw_context.c */
597 void si_destroy_saved_cs(struct si_saved_cs *scs);
598 void si_context_gfx_flush(void *context, unsigned flags,
599 struct pipe_fence_handle **fence);
600 void si_begin_new_cs(struct si_context *ctx);
601 void si_need_cs_space(struct si_context *ctx);
602
603 /* si_compute.c */
604 void si_init_compute_functions(struct si_context *sctx);
605
606 /* si_perfcounters.c */
607 void si_init_perfcounters(struct si_screen *screen);
608
609 /* si_uvd.c */
610 struct pipe_video_codec *si_uvd_create_decoder(struct pipe_context *context,
611 const struct pipe_video_codec *templ);
612
613 struct pipe_video_buffer *si_video_buffer_create(struct pipe_context *pipe,
614 const struct pipe_video_buffer *tmpl);
615
616 /* si_viewport.c */
617 void si_update_vs_viewport_state(struct si_context *ctx);
618 void si_init_viewport_functions(struct si_context *ctx);
619
620
621 /*
622 * common helpers
623 */
624
625 static inline void
626 si_invalidate_draw_sh_constants(struct si_context *sctx)
627 {
628 sctx->last_base_vertex = SI_BASE_VERTEX_UNKNOWN;
629 }
630
631 static inline void
632 si_set_atom_dirty(struct si_context *sctx,
633 struct r600_atom *atom, bool dirty)
634 {
635 unsigned bit = 1 << atom->id;
636
637 if (dirty)
638 sctx->dirty_atoms |= bit;
639 else
640 sctx->dirty_atoms &= ~bit;
641 }
642
643 static inline bool
644 si_is_atom_dirty(struct si_context *sctx,
645 struct r600_atom *atom)
646 {
647 unsigned bit = 1 << atom->id;
648
649 return sctx->dirty_atoms & bit;
650 }
651
652 static inline void
653 si_mark_atom_dirty(struct si_context *sctx,
654 struct r600_atom *atom)
655 {
656 si_set_atom_dirty(sctx, atom, true);
657 }
658
659 static inline struct si_shader_ctx_state *si_get_vs(struct si_context *sctx)
660 {
661 if (sctx->gs_shader.cso)
662 return &sctx->gs_shader;
663 if (sctx->tes_shader.cso)
664 return &sctx->tes_shader;
665
666 return &sctx->vs_shader;
667 }
668
669 static inline struct tgsi_shader_info *si_get_vs_info(struct si_context *sctx)
670 {
671 struct si_shader_ctx_state *vs = si_get_vs(sctx);
672
673 return vs->cso ? &vs->cso->info : NULL;
674 }
675
676 static inline struct si_shader* si_get_vs_state(struct si_context *sctx)
677 {
678 if (sctx->gs_shader.cso)
679 return sctx->gs_shader.cso->gs_copy_shader;
680
681 struct si_shader_ctx_state *vs = si_get_vs(sctx);
682 return vs->current ? vs->current : NULL;
683 }
684
685 static inline bool si_get_strmout_en(struct si_context *sctx)
686 {
687 return sctx->streamout.streamout_enabled ||
688 sctx->streamout.prims_gen_query_enabled;
689 }
690
691 static inline unsigned
692 si_optimal_tcc_alignment(struct si_context *sctx, unsigned upload_size)
693 {
694 unsigned alignment, tcc_cache_line_size;
695
696 /* If the upload size is less than the cache line size (e.g. 16, 32),
697 * the whole thing will fit into a cache line if we align it to its size.
698 * The idea is that multiple small uploads can share a cache line.
699 * If the upload size is greater, align it to the cache line size.
700 */
701 alignment = util_next_power_of_two(upload_size);
702 tcc_cache_line_size = sctx->screen->b.info.tcc_cache_line_size;
703 return MIN2(alignment, tcc_cache_line_size);
704 }
705
706 static inline void
707 si_saved_cs_reference(struct si_saved_cs **dst, struct si_saved_cs *src)
708 {
709 if (pipe_reference(&(*dst)->reference, &src->reference))
710 si_destroy_saved_cs(*dst);
711
712 *dst = src;
713 }
714
715 static inline void
716 si_make_CB_shader_coherent(struct si_context *sctx, unsigned num_samples,
717 bool shaders_read_metadata)
718 {
719 sctx->b.flags |= SI_CONTEXT_FLUSH_AND_INV_CB |
720 SI_CONTEXT_INV_VMEM_L1;
721
722 if (sctx->b.chip_class >= GFX9) {
723 /* Single-sample color is coherent with shaders on GFX9, but
724 * L2 metadata must be flushed if shaders read metadata.
725 * (DCC, CMASK).
726 */
727 if (num_samples >= 2)
728 sctx->b.flags |= SI_CONTEXT_INV_GLOBAL_L2;
729 else if (shaders_read_metadata)
730 sctx->b.flags |= SI_CONTEXT_INV_L2_METADATA;
731 } else {
732 /* SI-CI-VI */
733 sctx->b.flags |= SI_CONTEXT_INV_GLOBAL_L2;
734 }
735 }
736
737 static inline void
738 si_make_DB_shader_coherent(struct si_context *sctx, unsigned num_samples,
739 bool include_stencil, bool shaders_read_metadata)
740 {
741 sctx->b.flags |= SI_CONTEXT_FLUSH_AND_INV_DB |
742 SI_CONTEXT_INV_VMEM_L1;
743
744 if (sctx->b.chip_class >= GFX9) {
745 /* Single-sample depth (not stencil) is coherent with shaders
746 * on GFX9, but L2 metadata must be flushed if shaders read
747 * metadata.
748 */
749 if (num_samples >= 2 || include_stencil)
750 sctx->b.flags |= SI_CONTEXT_INV_GLOBAL_L2;
751 else if (shaders_read_metadata)
752 sctx->b.flags |= SI_CONTEXT_INV_L2_METADATA;
753 } else {
754 /* SI-CI-VI */
755 sctx->b.flags |= SI_CONTEXT_INV_GLOBAL_L2;
756 }
757 }
758
759 #endif