radeonsi: remove separate CB/DB_META flush flags
[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 #ifdef PIPE_ARCH_BIG_ENDIAN
32 #define SI_BIG_ENDIAN 1
33 #else
34 #define SI_BIG_ENDIAN 0
35 #endif
36
37 /* The base vertex and primitive restart can be any number, but we must pick
38 * one which will mean "unknown" for the purpose of state tracking and
39 * the number shouldn't be a commonly-used one. */
40 #define SI_BASE_VERTEX_UNKNOWN INT_MIN
41 #define SI_RESTART_INDEX_UNKNOWN INT_MIN
42 #define SI_NUM_SMOOTH_AA_SAMPLES 8
43 #define SI_GS_PER_ES 128
44
45 /* Instruction cache. */
46 #define SI_CONTEXT_INV_ICACHE (R600_CONTEXT_PRIVATE_FLAG << 0)
47 /* SMEM L1, other names: KCACHE, constant cache, DCACHE, data cache */
48 #define SI_CONTEXT_INV_SMEM_L1 (R600_CONTEXT_PRIVATE_FLAG << 1)
49 /* VMEM L1 can optionally be bypassed (GLC=1). Other names: TC L1 */
50 #define SI_CONTEXT_INV_VMEM_L1 (R600_CONTEXT_PRIVATE_FLAG << 2)
51 /* Used by everything except CB/DB, can be bypassed (SLC=1). Other names: TC L2 */
52 #define SI_CONTEXT_INV_GLOBAL_L2 (R600_CONTEXT_PRIVATE_FLAG << 3)
53 /* Write dirty L2 lines back to memory (shader and CP DMA stores), but don't
54 * invalidate L2. SI-CIK can't do it, so they will do complete invalidation. */
55 #define SI_CONTEXT_WRITEBACK_GLOBAL_L2 (R600_CONTEXT_PRIVATE_FLAG << 4)
56 /* gaps */
57 /* Framebuffer caches. */
58 #define SI_CONTEXT_FLUSH_AND_INV_DB (R600_CONTEXT_PRIVATE_FLAG << 7)
59 #define SI_CONTEXT_FLUSH_AND_INV_CB (R600_CONTEXT_PRIVATE_FLAG << 8)
60 /* Engine synchronization. */
61 #define SI_CONTEXT_VS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 9)
62 #define SI_CONTEXT_PS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 10)
63 #define SI_CONTEXT_CS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 11)
64 #define SI_CONTEXT_VGT_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 12)
65 #define SI_CONTEXT_VGT_STREAMOUT_SYNC (R600_CONTEXT_PRIVATE_FLAG << 13)
66
67 #define SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER (SI_CONTEXT_FLUSH_AND_INV_CB | \
68 SI_CONTEXT_FLUSH_AND_INV_DB)
69
70 #define SI_MAX_BORDER_COLORS 4096
71
72 struct si_compute;
73 struct hash_table;
74 struct u_suballocator;
75
76 struct si_screen {
77 struct r600_common_screen b;
78 unsigned gs_table_depth;
79 unsigned tess_offchip_block_dw_size;
80 bool has_distributed_tess;
81 bool has_draw_indirect_multi;
82 bool has_ds_bpermute;
83
84 /* Whether shaders are monolithic (1-part) or separate (3-part). */
85 bool use_monolithic_shaders;
86 bool record_llvm_ir;
87
88 pipe_mutex shader_parts_mutex;
89 struct si_shader_part *vs_prologs;
90 struct si_shader_part *vs_epilogs;
91 struct si_shader_part *tcs_epilogs;
92 struct si_shader_part *gs_prologs;
93 struct si_shader_part *ps_prologs;
94 struct si_shader_part *ps_epilogs;
95
96 /* Shader cache in memory.
97 *
98 * Design & limitations:
99 * - The shader cache is per screen (= per process), never saved to
100 * disk, and skips redundant shader compilations from TGSI to bytecode.
101 * - It can only be used with one-variant-per-shader support, in which
102 * case only the main (typically middle) part of shaders is cached.
103 * - Only VS, TCS, TES, PS are cached, out of which only the hw VS
104 * variants of VS and TES are cached, so LS and ES aren't.
105 * - GS and CS aren't cached, but it's certainly possible to cache
106 * those as well.
107 */
108 pipe_mutex shader_cache_mutex;
109 struct hash_table *shader_cache;
110
111 /* Shader compiler queue for multithreaded compilation. */
112 struct util_queue shader_compiler_queue;
113 LLVMTargetMachineRef tm[4]; /* used by the queue only */
114 };
115
116 struct si_blend_color {
117 struct r600_atom atom;
118 struct pipe_blend_color state;
119 };
120
121 struct si_sampler_view {
122 struct pipe_sampler_view base;
123 /* [0..7] = image descriptor
124 * [4..7] = buffer descriptor */
125 uint32_t state[8];
126 uint32_t fmask_state[8];
127 const struct radeon_surf_level *base_level_info;
128 unsigned base_level;
129 unsigned block_width;
130 bool is_stencil_sampler;
131 };
132
133 #define SI_SAMPLER_STATE_MAGIC 0x34f1c35a
134
135 struct si_sampler_state {
136 #ifdef DEBUG
137 unsigned magic;
138 #endif
139 uint32_t val[4];
140 };
141
142 struct si_cs_shader_state {
143 struct si_compute *program;
144 struct si_compute *emitted_program;
145 unsigned offset;
146 bool initialized;
147 bool uses_scratch;
148 };
149
150 struct si_textures_info {
151 struct si_sampler_views views;
152 uint32_t depth_texture_mask; /* which textures are depth */
153 uint32_t compressed_colortex_mask;
154 };
155
156 struct si_images_info {
157 struct pipe_image_view views[SI_NUM_IMAGES];
158 uint32_t compressed_colortex_mask;
159 unsigned enabled_mask;
160 };
161
162 struct si_framebuffer {
163 struct r600_atom atom;
164 struct pipe_framebuffer_state state;
165 unsigned nr_samples;
166 unsigned log_samples;
167 unsigned compressed_cb_mask;
168 unsigned colorbuf_enabled_4bit;
169 unsigned spi_shader_col_format;
170 unsigned spi_shader_col_format_alpha;
171 unsigned spi_shader_col_format_blend;
172 unsigned spi_shader_col_format_blend_alpha;
173 unsigned color_is_int8; /* bitmask */
174 unsigned dirty_cbufs;
175 bool dirty_zsbuf;
176 bool any_dst_linear;
177 bool do_update_surf_dirtiness;
178 };
179
180 struct si_clip_state {
181 struct r600_atom atom;
182 struct pipe_clip_state state;
183 };
184
185 struct si_sample_locs {
186 struct r600_atom atom;
187 unsigned nr_samples;
188 };
189
190 struct si_sample_mask {
191 struct r600_atom atom;
192 uint16_t sample_mask;
193 };
194
195 /* A shader state consists of the shader selector, which is a constant state
196 * object shared by multiple contexts and shouldn't be modified, and
197 * the current shader variant selected for this context.
198 */
199 struct si_shader_ctx_state {
200 struct si_shader_selector *cso;
201 struct si_shader *current;
202 };
203
204 #define SI_NUM_VGT_PARAM_KEY_BITS 12
205 #define SI_NUM_VGT_PARAM_STATES (1 << SI_NUM_VGT_PARAM_KEY_BITS)
206
207 /* The IA_MULTI_VGT_PARAM key used to index the table of precomputed values.
208 * Some fields are set by state-change calls, most are set by draw_vbo.
209 */
210 union si_vgt_param_key {
211 struct {
212 unsigned prim:4;
213 unsigned uses_instancing:1;
214 unsigned multi_instances_smaller_than_primgroup:1;
215 unsigned primitive_restart:1;
216 unsigned count_from_stream_output:1;
217 unsigned line_stipple_enabled:1;
218 unsigned uses_tess:1;
219 unsigned tcs_tes_uses_prim_id:1;
220 unsigned uses_gs:1;
221 unsigned _pad:32 - SI_NUM_VGT_PARAM_KEY_BITS;
222 } u;
223 uint32_t index;
224 };
225
226 struct si_context {
227 struct r600_common_context b;
228 struct blitter_context *blitter;
229 void *custom_dsa_flush;
230 void *custom_blend_resolve;
231 void *custom_blend_decompress;
232 void *custom_blend_fastclear;
233 void *custom_blend_dcc_decompress;
234 struct si_screen *screen;
235
236 struct radeon_winsys_cs *ce_ib;
237 struct radeon_winsys_cs *ce_preamble_ib;
238 bool ce_need_synchronization;
239 struct u_suballocator *ce_suballocator;
240
241 struct si_shader_ctx_state fixed_func_tcs_shader;
242 LLVMTargetMachineRef tm; /* only non-threaded compilation */
243 bool gfx_flush_in_progress;
244 bool compute_is_busy;
245
246 /* Atoms (direct states). */
247 union si_state_atoms atoms;
248 unsigned dirty_atoms; /* mask */
249 /* PM4 states (precomputed immutable states) */
250 unsigned dirty_states;
251 union si_state queued;
252 union si_state emitted;
253
254 /* Atom declarations. */
255 struct r600_atom prefetch_L2;
256 struct si_framebuffer framebuffer;
257 struct si_sample_locs msaa_sample_locs;
258 struct r600_atom db_render_state;
259 struct r600_atom msaa_config;
260 struct si_sample_mask sample_mask;
261 struct r600_atom cb_render_state;
262 struct si_blend_color blend_color;
263 struct r600_atom clip_regs;
264 struct si_clip_state clip_state;
265 struct si_shader_data shader_userdata;
266 struct si_stencil_ref stencil_ref;
267 struct r600_atom spi_map;
268
269 /* Precomputed states. */
270 struct si_pm4_state *init_config;
271 struct si_pm4_state *init_config_gs_rings;
272 bool init_config_has_vgt_flush;
273 struct si_pm4_state *vgt_shader_config[4];
274
275 /* shaders */
276 struct si_shader_ctx_state ps_shader;
277 struct si_shader_ctx_state gs_shader;
278 struct si_shader_ctx_state vs_shader;
279 struct si_shader_ctx_state tcs_shader;
280 struct si_shader_ctx_state tes_shader;
281 struct si_cs_shader_state cs_shader_state;
282
283 /* shader information */
284 struct si_vertex_element *vertex_elements;
285 unsigned sprite_coord_enable;
286 bool flatshade;
287 bool do_update_shaders;
288
289 /* shader descriptors */
290 struct si_descriptors vertex_buffers;
291 struct si_descriptors descriptors[SI_NUM_DESCS];
292 unsigned descriptors_dirty;
293 unsigned shader_pointers_dirty;
294 unsigned compressed_tex_shader_mask;
295 struct si_buffer_resources rw_buffers;
296 struct si_buffer_resources const_buffers[SI_NUM_SHADERS];
297 struct si_buffer_resources shader_buffers[SI_NUM_SHADERS];
298 struct si_textures_info samplers[SI_NUM_SHADERS];
299 struct si_images_info images[SI_NUM_SHADERS];
300
301 /* other shader resources */
302 struct pipe_constant_buffer null_const_buf; /* used for set_constant_buffer(NULL) on CIK */
303 struct pipe_resource *esgs_ring;
304 struct pipe_resource *gsvs_ring;
305 struct pipe_resource *tf_ring;
306 struct pipe_resource *tess_offchip_ring;
307 union pipe_color_union *border_color_table; /* in CPU memory, any endian */
308 struct r600_resource *border_color_buffer;
309 union pipe_color_union *border_color_map; /* in VRAM (slow access), little endian */
310 unsigned border_color_count;
311
312 /* Vertex and index buffers. */
313 bool vertex_buffers_dirty;
314 bool vertex_buffer_pointer_dirty;
315 struct pipe_index_buffer index_buffer;
316 struct pipe_vertex_buffer vertex_buffer[SI_NUM_VERTEX_BUFFERS];
317
318 /* MSAA config state. */
319 int ps_iter_samples;
320 bool smoothing_enabled;
321
322 /* DB render state. */
323 bool dbcb_depth_copy_enabled;
324 bool dbcb_stencil_copy_enabled;
325 unsigned dbcb_copy_sample;
326 bool db_flush_depth_inplace;
327 bool db_flush_stencil_inplace;
328 bool db_depth_clear;
329 bool db_depth_disable_expclear;
330 bool db_stencil_clear;
331 bool db_stencil_disable_expclear;
332 unsigned ps_db_shader_control;
333 bool occlusion_queries_disabled;
334
335 /* Emitted draw state. */
336 int last_index_size;
337 int last_base_vertex;
338 int last_start_instance;
339 int last_drawid;
340 int last_sh_base_reg;
341 int last_primitive_restart_en;
342 int last_restart_index;
343 int last_gs_out_prim;
344 int last_prim;
345 int last_multi_vgt_param;
346 int last_rast_prim;
347 unsigned last_sc_line_stipple;
348 int current_rast_prim; /* primitive type after TES, GS */
349 bool gs_tri_strip_adj_fix;
350
351 /* Scratch buffer */
352 struct r600_atom scratch_state;
353 struct r600_resource *scratch_buffer;
354 unsigned scratch_waves;
355 unsigned spi_tmpring_size;
356
357 struct r600_resource *compute_scratch_buffer;
358
359 /* Emitted derived tessellation state. */
360 struct si_shader *last_ls; /* local shader (VS) */
361 struct si_shader_selector *last_tcs;
362 int last_num_tcs_input_cp;
363 int last_tes_sh_base;
364 unsigned last_num_patches;
365
366 /* Debug state. */
367 bool is_debug;
368 struct radeon_saved_cs last_gfx;
369 struct r600_resource *last_trace_buf;
370 struct r600_resource *trace_buf;
371 unsigned trace_id;
372 uint64_t dmesg_timestamp;
373 unsigned apitrace_call_number;
374
375 /* Other state */
376 bool need_check_render_feedback;
377
378 /* Precomputed IA_MULTI_VGT_PARAM */
379 union si_vgt_param_key ia_multi_vgt_param_key;
380 unsigned ia_multi_vgt_param[SI_NUM_VGT_PARAM_STATES];
381 };
382
383 /* cik_sdma.c */
384 void cik_init_sdma_functions(struct si_context *sctx);
385
386 /* si_blit.c */
387 void si_init_blit_functions(struct si_context *sctx);
388 void si_decompress_graphics_textures(struct si_context *sctx);
389 void si_decompress_compute_textures(struct si_context *sctx);
390 void si_resource_copy_region(struct pipe_context *ctx,
391 struct pipe_resource *dst,
392 unsigned dst_level,
393 unsigned dstx, unsigned dsty, unsigned dstz,
394 struct pipe_resource *src,
395 unsigned src_level,
396 const struct pipe_box *src_box);
397
398 /* si_cp_dma.c */
399 #define SI_CPDMA_SKIP_CHECK_CS_SPACE (1 << 0) /* don't call need_cs_space */
400 #define SI_CPDMA_SKIP_SYNC_AFTER (1 << 1) /* don't wait for DMA after the copy */
401 #define SI_CPDMA_SKIP_SYNC_BEFORE (1 << 2) /* don't wait for DMA before the copy (RAW hazards) */
402 #define SI_CPDMA_SKIP_GFX_SYNC (1 << 3) /* don't flush caches and don't wait for PS/CS */
403 #define SI_CPDMA_SKIP_BO_LIST_UPDATE (1 << 4) /* don't update the BO list */
404 #define SI_CPDMA_SKIP_ALL (SI_CPDMA_SKIP_CHECK_CS_SPACE | \
405 SI_CPDMA_SKIP_SYNC_AFTER | \
406 SI_CPDMA_SKIP_SYNC_BEFORE | \
407 SI_CPDMA_SKIP_GFX_SYNC | \
408 SI_CPDMA_SKIP_BO_LIST_UPDATE)
409
410 void si_copy_buffer(struct si_context *sctx,
411 struct pipe_resource *dst, struct pipe_resource *src,
412 uint64_t dst_offset, uint64_t src_offset, unsigned size,
413 unsigned user_flags);
414 void cik_prefetch_TC_L2_async(struct si_context *sctx, struct pipe_resource *buf,
415 uint64_t offset, unsigned size);
416 void si_init_cp_dma_functions(struct si_context *sctx);
417
418 /* si_debug.c */
419 void si_init_debug_functions(struct si_context *sctx);
420 void si_check_vm_faults(struct r600_common_context *ctx,
421 struct radeon_saved_cs *saved, enum ring_type ring);
422 bool si_replace_shader(unsigned num, struct radeon_shader_binary *binary);
423
424 /* si_dma.c */
425 void si_init_dma_functions(struct si_context *sctx);
426
427 /* si_hw_context.c */
428 void si_context_gfx_flush(void *context, unsigned flags,
429 struct pipe_fence_handle **fence);
430 void si_begin_new_cs(struct si_context *ctx);
431 void si_need_cs_space(struct si_context *ctx);
432
433 /* si_compute.c */
434 void si_init_compute_functions(struct si_context *sctx);
435
436 /* si_perfcounters.c */
437 void si_init_perfcounters(struct si_screen *screen);
438
439 /* si_uvd.c */
440 struct pipe_video_codec *si_uvd_create_decoder(struct pipe_context *context,
441 const struct pipe_video_codec *templ);
442
443 struct pipe_video_buffer *si_video_buffer_create(struct pipe_context *pipe,
444 const struct pipe_video_buffer *tmpl);
445
446 /*
447 * common helpers
448 */
449
450 static inline void
451 si_invalidate_draw_sh_constants(struct si_context *sctx)
452 {
453 sctx->last_base_vertex = SI_BASE_VERTEX_UNKNOWN;
454 }
455
456 static inline void
457 si_set_atom_dirty(struct si_context *sctx,
458 struct r600_atom *atom, bool dirty)
459 {
460 unsigned bit = 1 << atom->id;
461
462 if (dirty)
463 sctx->dirty_atoms |= bit;
464 else
465 sctx->dirty_atoms &= ~bit;
466 }
467
468 static inline bool
469 si_is_atom_dirty(struct si_context *sctx,
470 struct r600_atom *atom)
471 {
472 unsigned bit = 1 << atom->id;
473
474 return sctx->dirty_atoms & bit;
475 }
476
477 static inline void
478 si_mark_atom_dirty(struct si_context *sctx,
479 struct r600_atom *atom)
480 {
481 si_set_atom_dirty(sctx, atom, true);
482 }
483
484 static inline struct tgsi_shader_info *si_get_vs_info(struct si_context *sctx)
485 {
486 if (sctx->gs_shader.cso)
487 return &sctx->gs_shader.cso->info;
488 else if (sctx->tes_shader.cso)
489 return &sctx->tes_shader.cso->info;
490 else if (sctx->vs_shader.cso)
491 return &sctx->vs_shader.cso->info;
492 else
493 return NULL;
494 }
495
496 static inline struct si_shader* si_get_vs_state(struct si_context *sctx)
497 {
498 if (sctx->gs_shader.current)
499 return sctx->gs_shader.cso->gs_copy_shader;
500 else if (sctx->tes_shader.current)
501 return sctx->tes_shader.current;
502 else
503 return sctx->vs_shader.current;
504 }
505
506 static inline bool si_vs_exports_prim_id(struct si_shader *shader)
507 {
508 if (shader->selector->type == PIPE_SHADER_VERTEX)
509 return shader->key.part.vs.epilog.export_prim_id;
510 else if (shader->selector->type == PIPE_SHADER_TESS_EVAL)
511 return shader->key.part.tes.epilog.export_prim_id;
512 else
513 return false;
514 }
515
516 #endif