radeonsi: atomize the scratch buffer state
[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 /* Framebuffer caches. */
57 #define SI_CONTEXT_FLUSH_AND_INV_CB_META (R600_CONTEXT_PRIVATE_FLAG << 5)
58 #define SI_CONTEXT_FLUSH_AND_INV_DB_META (R600_CONTEXT_PRIVATE_FLAG << 6)
59 #define SI_CONTEXT_FLUSH_AND_INV_DB (R600_CONTEXT_PRIVATE_FLAG << 7)
60 #define SI_CONTEXT_FLUSH_AND_INV_CB (R600_CONTEXT_PRIVATE_FLAG << 8)
61 /* Engine synchronization. */
62 #define SI_CONTEXT_VS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 9)
63 #define SI_CONTEXT_PS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 10)
64 #define SI_CONTEXT_CS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 11)
65 #define SI_CONTEXT_VGT_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 12)
66 #define SI_CONTEXT_VGT_STREAMOUT_SYNC (R600_CONTEXT_PRIVATE_FLAG << 13)
67
68 #define SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER (SI_CONTEXT_FLUSH_AND_INV_CB | \
69 SI_CONTEXT_FLUSH_AND_INV_CB_META | \
70 SI_CONTEXT_FLUSH_AND_INV_DB | \
71 SI_CONTEXT_FLUSH_AND_INV_DB_META)
72
73 #define SI_MAX_BORDER_COLORS 4096
74
75 struct si_compute;
76 struct hash_table;
77 struct u_suballocator;
78
79 struct si_screen {
80 struct r600_common_screen b;
81 unsigned gs_table_depth;
82 unsigned tess_offchip_block_dw_size;
83 bool has_distributed_tess;
84 bool has_draw_indirect_multi;
85 bool has_ds_bpermute;
86
87 /* Whether shaders are monolithic (1-part) or separate (3-part). */
88 bool use_monolithic_shaders;
89 bool record_llvm_ir;
90
91 pipe_mutex shader_parts_mutex;
92 struct si_shader_part *vs_prologs;
93 struct si_shader_part *vs_epilogs;
94 struct si_shader_part *tcs_epilogs;
95 struct si_shader_part *gs_prologs;
96 struct si_shader_part *ps_prologs;
97 struct si_shader_part *ps_epilogs;
98
99 /* Shader cache in memory.
100 *
101 * Design & limitations:
102 * - The shader cache is per screen (= per process), never saved to
103 * disk, and skips redundant shader compilations from TGSI to bytecode.
104 * - It can only be used with one-variant-per-shader support, in which
105 * case only the main (typically middle) part of shaders is cached.
106 * - Only VS, TCS, TES, PS are cached, out of which only the hw VS
107 * variants of VS and TES are cached, so LS and ES aren't.
108 * - GS and CS aren't cached, but it's certainly possible to cache
109 * those as well.
110 */
111 pipe_mutex shader_cache_mutex;
112 struct hash_table *shader_cache;
113
114 /* Shader compiler queue for multithreaded compilation. */
115 struct util_queue shader_compiler_queue;
116 LLVMTargetMachineRef tm[4]; /* used by the queue only */
117 };
118
119 struct si_blend_color {
120 struct r600_atom atom;
121 struct pipe_blend_color state;
122 };
123
124 struct si_sampler_view {
125 struct pipe_sampler_view base;
126 /* [0..7] = image descriptor
127 * [4..7] = buffer descriptor */
128 uint32_t state[8];
129 uint32_t fmask_state[8];
130 const struct radeon_surf_level *base_level_info;
131 unsigned base_level;
132 unsigned block_width;
133 bool is_stencil_sampler;
134 };
135
136 #define SI_SAMPLER_STATE_MAGIC 0x34f1c35a
137
138 struct si_sampler_state {
139 #ifdef DEBUG
140 unsigned magic;
141 #endif
142 uint32_t val[4];
143 };
144
145 struct si_cs_shader_state {
146 struct si_compute *program;
147 struct si_compute *emitted_program;
148 unsigned offset;
149 bool initialized;
150 bool uses_scratch;
151 };
152
153 struct si_textures_info {
154 struct si_sampler_views views;
155 uint32_t depth_texture_mask; /* which textures are depth */
156 uint32_t compressed_colortex_mask;
157 };
158
159 struct si_images_info {
160 struct pipe_image_view views[SI_NUM_IMAGES];
161 uint32_t compressed_colortex_mask;
162 unsigned enabled_mask;
163 };
164
165 struct si_framebuffer {
166 struct r600_atom atom;
167 struct pipe_framebuffer_state state;
168 unsigned nr_samples;
169 unsigned log_samples;
170 unsigned compressed_cb_mask;
171 unsigned colorbuf_enabled_4bit;
172 unsigned spi_shader_col_format;
173 unsigned spi_shader_col_format_alpha;
174 unsigned spi_shader_col_format_blend;
175 unsigned spi_shader_col_format_blend_alpha;
176 unsigned color_is_int8; /* bitmask */
177 unsigned dirty_cbufs;
178 bool dirty_zsbuf;
179 bool any_dst_linear;
180 bool do_update_surf_dirtiness;
181 };
182
183 struct si_clip_state {
184 struct r600_atom atom;
185 struct pipe_clip_state state;
186 };
187
188 struct si_sample_locs {
189 struct r600_atom atom;
190 unsigned nr_samples;
191 };
192
193 struct si_sample_mask {
194 struct r600_atom atom;
195 uint16_t sample_mask;
196 };
197
198 /* A shader state consists of the shader selector, which is a constant state
199 * object shared by multiple contexts and shouldn't be modified, and
200 * the current shader variant selected for this context.
201 */
202 struct si_shader_ctx_state {
203 struct si_shader_selector *cso;
204 struct si_shader *current;
205 };
206
207 #define SI_NUM_VGT_PARAM_KEY_BITS 12
208 #define SI_NUM_VGT_PARAM_STATES (1 << SI_NUM_VGT_PARAM_KEY_BITS)
209
210 /* The IA_MULTI_VGT_PARAM key used to index the table of precomputed values.
211 * Some fields are set by state-change calls, most are set by draw_vbo.
212 */
213 union si_vgt_param_key {
214 struct {
215 unsigned prim:4;
216 unsigned uses_instancing:1;
217 unsigned multi_instances_smaller_than_primgroup:1;
218 unsigned primitive_restart:1;
219 unsigned count_from_stream_output:1;
220 unsigned line_stipple_enabled:1;
221 unsigned uses_tess:1;
222 unsigned tcs_tes_uses_prim_id:1;
223 unsigned uses_gs:1;
224 unsigned _pad:32 - SI_NUM_VGT_PARAM_KEY_BITS;
225 } u;
226 uint32_t index;
227 };
228
229 struct si_context {
230 struct r600_common_context b;
231 struct blitter_context *blitter;
232 void *custom_dsa_flush;
233 void *custom_blend_resolve;
234 void *custom_blend_decompress;
235 void *custom_blend_fastclear;
236 void *custom_blend_dcc_decompress;
237 struct si_screen *screen;
238
239 struct radeon_winsys_cs *ce_ib;
240 struct radeon_winsys_cs *ce_preamble_ib;
241 bool ce_need_synchronization;
242 struct u_suballocator *ce_suballocator;
243
244 struct si_shader_ctx_state fixed_func_tcs_shader;
245 LLVMTargetMachineRef tm; /* only non-threaded compilation */
246 bool gfx_flush_in_progress;
247 bool compute_is_busy;
248
249 /* Atoms (direct states). */
250 union si_state_atoms atoms;
251 unsigned dirty_atoms; /* mask */
252 /* PM4 states (precomputed immutable states) */
253 unsigned dirty_states;
254 union si_state queued;
255 union si_state emitted;
256
257 /* Atom declarations. */
258 struct r600_atom prefetch_L2;
259 struct si_framebuffer framebuffer;
260 struct si_sample_locs msaa_sample_locs;
261 struct r600_atom db_render_state;
262 struct r600_atom msaa_config;
263 struct si_sample_mask sample_mask;
264 struct r600_atom cb_render_state;
265 struct si_blend_color blend_color;
266 struct r600_atom clip_regs;
267 struct si_clip_state clip_state;
268 struct si_shader_data shader_userdata;
269 struct si_stencil_ref stencil_ref;
270 struct r600_atom spi_map;
271
272 /* Precomputed states. */
273 struct si_pm4_state *init_config;
274 struct si_pm4_state *init_config_gs_rings;
275 bool init_config_has_vgt_flush;
276 struct si_pm4_state *vgt_shader_config[4];
277
278 /* shaders */
279 struct si_shader_ctx_state ps_shader;
280 struct si_shader_ctx_state gs_shader;
281 struct si_shader_ctx_state vs_shader;
282 struct si_shader_ctx_state tcs_shader;
283 struct si_shader_ctx_state tes_shader;
284 struct si_cs_shader_state cs_shader_state;
285
286 /* shader information */
287 struct si_vertex_element *vertex_elements;
288 unsigned sprite_coord_enable;
289 bool flatshade;
290 bool do_update_shaders;
291
292 /* shader descriptors */
293 struct si_descriptors vertex_buffers;
294 struct si_descriptors descriptors[SI_NUM_DESCS];
295 unsigned descriptors_dirty;
296 unsigned shader_pointers_dirty;
297 unsigned compressed_tex_shader_mask;
298 struct si_buffer_resources rw_buffers;
299 struct si_buffer_resources const_buffers[SI_NUM_SHADERS];
300 struct si_buffer_resources shader_buffers[SI_NUM_SHADERS];
301 struct si_textures_info samplers[SI_NUM_SHADERS];
302 struct si_images_info images[SI_NUM_SHADERS];
303
304 /* other shader resources */
305 struct pipe_constant_buffer null_const_buf; /* used for set_constant_buffer(NULL) on CIK */
306 struct pipe_resource *esgs_ring;
307 struct pipe_resource *gsvs_ring;
308 struct pipe_resource *tf_ring;
309 struct pipe_resource *tess_offchip_ring;
310 union pipe_color_union *border_color_table; /* in CPU memory, any endian */
311 struct r600_resource *border_color_buffer;
312 union pipe_color_union *border_color_map; /* in VRAM (slow access), little endian */
313 unsigned border_color_count;
314
315 /* Vertex and index buffers. */
316 bool vertex_buffers_dirty;
317 bool vertex_buffer_pointer_dirty;
318 struct pipe_index_buffer index_buffer;
319 struct pipe_vertex_buffer vertex_buffer[SI_NUM_VERTEX_BUFFERS];
320
321 /* MSAA config state. */
322 int ps_iter_samples;
323 bool smoothing_enabled;
324
325 /* DB render state. */
326 bool dbcb_depth_copy_enabled;
327 bool dbcb_stencil_copy_enabled;
328 unsigned dbcb_copy_sample;
329 bool db_flush_depth_inplace;
330 bool db_flush_stencil_inplace;
331 bool db_depth_clear;
332 bool db_depth_disable_expclear;
333 bool db_stencil_clear;
334 bool db_stencil_disable_expclear;
335 unsigned ps_db_shader_control;
336 bool occlusion_queries_disabled;
337
338 /* Emitted draw state. */
339 int last_index_size;
340 int last_base_vertex;
341 int last_start_instance;
342 int last_drawid;
343 int last_sh_base_reg;
344 int last_primitive_restart_en;
345 int last_restart_index;
346 int last_gs_out_prim;
347 int last_prim;
348 int last_multi_vgt_param;
349 int last_rast_prim;
350 unsigned last_sc_line_stipple;
351 int current_rast_prim; /* primitive type after TES, GS */
352 bool gs_tri_strip_adj_fix;
353
354 /* Scratch buffer */
355 struct r600_atom scratch_state;
356 struct r600_resource *scratch_buffer;
357 unsigned scratch_waves;
358 unsigned spi_tmpring_size;
359
360 struct r600_resource *compute_scratch_buffer;
361
362 /* Emitted derived tessellation state. */
363 struct si_shader *last_ls; /* local shader (VS) */
364 struct si_shader_selector *last_tcs;
365 int last_num_tcs_input_cp;
366 int last_tes_sh_base;
367 unsigned last_num_patches;
368
369 /* Debug state. */
370 bool is_debug;
371 struct radeon_saved_cs last_gfx;
372 struct r600_resource *last_trace_buf;
373 struct r600_resource *trace_buf;
374 unsigned trace_id;
375 uint64_t dmesg_timestamp;
376 unsigned apitrace_call_number;
377
378 /* Other state */
379 bool need_check_render_feedback;
380
381 /* Precomputed IA_MULTI_VGT_PARAM */
382 union si_vgt_param_key ia_multi_vgt_param_key;
383 unsigned ia_multi_vgt_param[SI_NUM_VGT_PARAM_STATES];
384 };
385
386 /* cik_sdma.c */
387 void cik_init_sdma_functions(struct si_context *sctx);
388
389 /* si_blit.c */
390 void si_init_blit_functions(struct si_context *sctx);
391 void si_decompress_graphics_textures(struct si_context *sctx);
392 void si_decompress_compute_textures(struct si_context *sctx);
393 void si_resource_copy_region(struct pipe_context *ctx,
394 struct pipe_resource *dst,
395 unsigned dst_level,
396 unsigned dstx, unsigned dsty, unsigned dstz,
397 struct pipe_resource *src,
398 unsigned src_level,
399 const struct pipe_box *src_box);
400
401 /* si_cp_dma.c */
402 #define SI_CPDMA_SKIP_CHECK_CS_SPACE (1 << 0) /* don't call need_cs_space */
403 #define SI_CPDMA_SKIP_SYNC_AFTER (1 << 1) /* don't wait for DMA after the copy */
404 #define SI_CPDMA_SKIP_SYNC_BEFORE (1 << 2) /* don't wait for DMA before the copy (RAW hazards) */
405 #define SI_CPDMA_SKIP_GFX_SYNC (1 << 3) /* don't flush caches and don't wait for PS/CS */
406 #define SI_CPDMA_SKIP_BO_LIST_UPDATE (1 << 4) /* don't update the BO list */
407 #define SI_CPDMA_SKIP_ALL (SI_CPDMA_SKIP_CHECK_CS_SPACE | \
408 SI_CPDMA_SKIP_SYNC_AFTER | \
409 SI_CPDMA_SKIP_SYNC_BEFORE | \
410 SI_CPDMA_SKIP_GFX_SYNC | \
411 SI_CPDMA_SKIP_BO_LIST_UPDATE)
412
413 void si_copy_buffer(struct si_context *sctx,
414 struct pipe_resource *dst, struct pipe_resource *src,
415 uint64_t dst_offset, uint64_t src_offset, unsigned size,
416 unsigned user_flags);
417 void cik_prefetch_TC_L2_async(struct si_context *sctx, struct pipe_resource *buf,
418 uint64_t offset, unsigned size);
419 void si_init_cp_dma_functions(struct si_context *sctx);
420
421 /* si_debug.c */
422 void si_init_debug_functions(struct si_context *sctx);
423 void si_check_vm_faults(struct r600_common_context *ctx,
424 struct radeon_saved_cs *saved, enum ring_type ring);
425 bool si_replace_shader(unsigned num, struct radeon_shader_binary *binary);
426
427 /* si_dma.c */
428 void si_init_dma_functions(struct si_context *sctx);
429
430 /* si_hw_context.c */
431 void si_context_gfx_flush(void *context, unsigned flags,
432 struct pipe_fence_handle **fence);
433 void si_begin_new_cs(struct si_context *ctx);
434 void si_need_cs_space(struct si_context *ctx);
435
436 /* si_compute.c */
437 void si_init_compute_functions(struct si_context *sctx);
438
439 /* si_perfcounters.c */
440 void si_init_perfcounters(struct si_screen *screen);
441
442 /* si_uvd.c */
443 struct pipe_video_codec *si_uvd_create_decoder(struct pipe_context *context,
444 const struct pipe_video_codec *templ);
445
446 struct pipe_video_buffer *si_video_buffer_create(struct pipe_context *pipe,
447 const struct pipe_video_buffer *tmpl);
448
449 /*
450 * common helpers
451 */
452
453 static inline void
454 si_invalidate_draw_sh_constants(struct si_context *sctx)
455 {
456 sctx->last_base_vertex = SI_BASE_VERTEX_UNKNOWN;
457 }
458
459 static inline void
460 si_set_atom_dirty(struct si_context *sctx,
461 struct r600_atom *atom, bool dirty)
462 {
463 unsigned bit = 1 << atom->id;
464
465 if (dirty)
466 sctx->dirty_atoms |= bit;
467 else
468 sctx->dirty_atoms &= ~bit;
469 }
470
471 static inline bool
472 si_is_atom_dirty(struct si_context *sctx,
473 struct r600_atom *atom)
474 {
475 unsigned bit = 1 << atom->id;
476
477 return sctx->dirty_atoms & bit;
478 }
479
480 static inline void
481 si_mark_atom_dirty(struct si_context *sctx,
482 struct r600_atom *atom)
483 {
484 si_set_atom_dirty(sctx, atom, true);
485 }
486
487 static inline struct tgsi_shader_info *si_get_vs_info(struct si_context *sctx)
488 {
489 if (sctx->gs_shader.cso)
490 return &sctx->gs_shader.cso->info;
491 else if (sctx->tes_shader.cso)
492 return &sctx->tes_shader.cso->info;
493 else if (sctx->vs_shader.cso)
494 return &sctx->vs_shader.cso->info;
495 else
496 return NULL;
497 }
498
499 static inline struct si_shader* si_get_vs_state(struct si_context *sctx)
500 {
501 if (sctx->gs_shader.current)
502 return sctx->gs_shader.cso->gs_copy_shader;
503 else if (sctx->tes_shader.current)
504 return sctx->tes_shader.current;
505 else
506 return sctx->vs_shader.current;
507 }
508
509 static inline bool si_vs_exports_prim_id(struct si_shader *shader)
510 {
511 if (shader->selector->type == PIPE_SHADER_VERTEX)
512 return shader->key.part.vs.epilog.export_prim_id;
513 else if (shader->selector->type == PIPE_SHADER_TESS_EVAL)
514 return shader->key.part.tes.epilog.export_prim_id;
515 else
516 return false;
517 }
518
519 #endif