radeonsi: update predicate condition for compute dispatches
[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_state.h"
30
31 #include <llvm-c/TargetMachine.h>
32
33 #ifdef PIPE_ARCH_BIG_ENDIAN
34 #define SI_BIG_ENDIAN 1
35 #else
36 #define SI_BIG_ENDIAN 0
37 #endif
38
39 /* The base vertex and primitive restart can be any number, but we must pick
40 * one which will mean "unknown" for the purpose of state tracking and
41 * the number shouldn't be a commonly-used one. */
42 #define SI_BASE_VERTEX_UNKNOWN INT_MIN
43 #define SI_RESTART_INDEX_UNKNOWN INT_MIN
44 #define SI_NUM_SMOOTH_AA_SAMPLES 8
45 #define SI_GS_PER_ES 128
46
47 /* Instruction cache. */
48 #define SI_CONTEXT_INV_ICACHE (R600_CONTEXT_PRIVATE_FLAG << 0)
49 /* SMEM L1, other names: KCACHE, constant cache, DCACHE, data cache */
50 #define SI_CONTEXT_INV_SMEM_L1 (R600_CONTEXT_PRIVATE_FLAG << 1)
51 /* VMEM L1 can optionally be bypassed (GLC=1). Other names: TC L1 */
52 #define SI_CONTEXT_INV_VMEM_L1 (R600_CONTEXT_PRIVATE_FLAG << 2)
53 /* Used by everything except CB/DB, can be bypassed (SLC=1). Other names: TC L2 */
54 #define SI_CONTEXT_INV_GLOBAL_L2 (R600_CONTEXT_PRIVATE_FLAG << 3)
55 /* Framebuffer caches. */
56 #define SI_CONTEXT_FLUSH_AND_INV_CB_META (R600_CONTEXT_PRIVATE_FLAG << 4)
57 #define SI_CONTEXT_FLUSH_AND_INV_DB_META (R600_CONTEXT_PRIVATE_FLAG << 5)
58 #define SI_CONTEXT_FLUSH_AND_INV_DB (R600_CONTEXT_PRIVATE_FLAG << 6)
59 #define SI_CONTEXT_FLUSH_AND_INV_CB (R600_CONTEXT_PRIVATE_FLAG << 7)
60 /* Engine synchronization. */
61 #define SI_CONTEXT_VS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 8)
62 #define SI_CONTEXT_PS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 9)
63 #define SI_CONTEXT_CS_PARTIAL_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 10)
64 #define SI_CONTEXT_VGT_FLUSH (R600_CONTEXT_PRIVATE_FLAG << 11)
65 #define SI_CONTEXT_VGT_STREAMOUT_SYNC (R600_CONTEXT_PRIVATE_FLAG << 12)
66 /* Compute only. */
67 #define SI_CONTEXT_FLUSH_WITH_INV_L2 (R600_CONTEXT_PRIVATE_FLAG << 13) /* TODO: merge with TC? */
68 #define SI_CONTEXT_FLAG_COMPUTE (R600_CONTEXT_PRIVATE_FLAG << 14)
69
70 #define SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER (SI_CONTEXT_FLUSH_AND_INV_CB | \
71 SI_CONTEXT_FLUSH_AND_INV_CB_META | \
72 SI_CONTEXT_FLUSH_AND_INV_DB | \
73 SI_CONTEXT_FLUSH_AND_INV_DB_META)
74
75 #define SI_ENCODE_TRACE_POINT(id) (0xcafe0000 | ((id) & 0xffff))
76 #define SI_IS_TRACE_POINT(x) (((x) & 0xcafe0000) == 0xcafe0000)
77 #define SI_GET_TRACE_POINT_ID(x) ((x) & 0xffff)
78
79 #define SI_MAX_BORDER_COLORS 4096
80
81 struct si_compute;
82 struct hash_table;
83 struct u_suballocator;
84
85 struct si_screen {
86 struct r600_common_screen b;
87 unsigned gs_table_depth;
88
89 /* Whether shaders are monolithic (1-part) or separate (3-part). */
90 bool use_monolithic_shaders;
91
92 pipe_mutex shader_parts_mutex;
93 struct si_shader_part *vs_prologs;
94 struct si_shader_part *vs_epilogs;
95 struct si_shader_part *tcs_epilogs;
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
115 struct si_blend_color {
116 struct r600_atom atom;
117 struct pipe_blend_color state;
118 };
119
120 struct si_sampler_view {
121 struct pipe_sampler_view base;
122 struct list_head list;
123 /* [0..7] = image descriptor
124 * [4..7] = buffer descriptor */
125 uint32_t state[8];
126 uint32_t fmask_state[8];
127 bool is_stencil_sampler;
128 };
129
130 struct si_sampler_state {
131 uint32_t val[4];
132 };
133
134 struct si_cs_shader_state {
135 struct si_compute *program;
136 struct si_compute *emitted_program;
137 unsigned offset;
138 bool initialized;
139 };
140
141 struct si_textures_info {
142 struct si_sampler_views views;
143 uint64_t depth_texture_mask; /* which textures are depth */
144 uint64_t compressed_colortex_mask;
145 };
146
147 struct si_images_info {
148 struct si_descriptors desc;
149 struct pipe_image_view views[SI_NUM_IMAGES];
150 uint32_t compressed_colortex_mask;
151 };
152
153 struct si_framebuffer {
154 struct r600_atom atom;
155 struct pipe_framebuffer_state state;
156 unsigned nr_samples;
157 unsigned log_samples;
158 unsigned cb0_is_integer;
159 unsigned compressed_cb_mask;
160 unsigned spi_shader_col_format;
161 unsigned spi_shader_col_format_alpha;
162 unsigned spi_shader_col_format_blend;
163 unsigned spi_shader_col_format_blend_alpha;
164 unsigned color_is_int8; /* bitmask */
165 unsigned dirty_cbufs;
166 bool dirty_zsbuf;
167 };
168
169 struct si_clip_state {
170 struct r600_atom atom;
171 struct pipe_clip_state state;
172 };
173
174 struct si_sample_mask {
175 struct r600_atom atom;
176 uint16_t sample_mask;
177 };
178
179 /* A shader state consists of the shader selector, which is a constant state
180 * object shared by multiple contexts and shouldn't be modified, and
181 * the current shader variant selected for this context.
182 */
183 struct si_shader_ctx_state {
184 struct si_shader_selector *cso;
185 struct si_shader *current;
186 };
187
188 struct si_context {
189 struct r600_common_context b;
190 struct blitter_context *blitter;
191 void *custom_dsa_flush;
192 void *custom_blend_resolve;
193 void *custom_blend_decompress;
194 void *custom_blend_fastclear;
195 void *custom_blend_dcc_decompress;
196 void *pstipple_sampler_state;
197 struct si_screen *screen;
198
199 struct radeon_winsys_cs *ce_ib;
200 struct radeon_winsys_cs *ce_preamble_ib;
201 bool ce_need_synchronization;
202 struct u_suballocator *ce_suballocator;
203
204 struct pipe_fence_handle *last_gfx_fence;
205 struct si_shader_ctx_state fixed_func_tcs_shader;
206 LLVMTargetMachineRef tm;
207 bool gfx_flush_in_progress;
208
209 /* Atoms (direct states). */
210 union si_state_atoms atoms;
211 unsigned dirty_atoms; /* mask */
212 /* PM4 states (precomputed immutable states) */
213 union si_state queued;
214 union si_state emitted;
215
216 /* Atom declarations. */
217 struct r600_atom cache_flush;
218 struct si_framebuffer framebuffer;
219 struct r600_atom msaa_sample_locs;
220 struct r600_atom db_render_state;
221 struct r600_atom msaa_config;
222 struct si_sample_mask sample_mask;
223 struct r600_atom cb_render_state;
224 struct si_blend_color blend_color;
225 struct r600_atom clip_regs;
226 struct si_clip_state clip_state;
227 struct si_shader_data shader_userdata;
228 struct si_stencil_ref stencil_ref;
229 struct r600_atom spi_map;
230
231 /* Precomputed states. */
232 struct si_pm4_state *init_config;
233 struct si_pm4_state *init_config_gs_rings;
234 bool init_config_has_vgt_flush;
235 struct si_pm4_state *vgt_shader_config[4];
236
237 /* shaders */
238 struct si_shader_ctx_state ps_shader;
239 struct si_shader_ctx_state gs_shader;
240 struct si_shader_ctx_state vs_shader;
241 struct si_shader_ctx_state tcs_shader;
242 struct si_shader_ctx_state tes_shader;
243 struct si_cs_shader_state cs_shader_state;
244
245 /* shader information */
246 struct si_vertex_element *vertex_elements;
247 unsigned sprite_coord_enable;
248 bool flatshade;
249
250 /* shader descriptors */
251 struct si_descriptors vertex_buffers;
252 struct si_buffer_resources const_buffers[SI_NUM_SHADERS];
253 struct si_buffer_resources rw_buffers[SI_NUM_SHADERS];
254 struct si_buffer_resources shader_buffers[SI_NUM_SHADERS];
255 struct si_textures_info samplers[SI_NUM_SHADERS];
256 struct si_images_info images[SI_NUM_SHADERS];
257
258 /* other shader resources */
259 struct pipe_constant_buffer null_const_buf; /* used for set_constant_buffer(NULL) on CIK */
260 struct pipe_resource *esgs_ring;
261 struct pipe_resource *gsvs_ring;
262 struct pipe_resource *tf_ring;
263 union pipe_color_union *border_color_table; /* in CPU memory, any endian */
264 struct r600_resource *border_color_buffer;
265 union pipe_color_union *border_color_map; /* in VRAM (slow access), little endian */
266 unsigned border_color_count;
267
268 /* Vertex and index buffers. */
269 bool vertex_buffers_dirty;
270 struct pipe_index_buffer index_buffer;
271 struct pipe_vertex_buffer vertex_buffer[SI_NUM_VERTEX_BUFFERS];
272
273 /* MSAA config state. */
274 int ps_iter_samples;
275 bool smoothing_enabled;
276
277 /* DB render state. */
278 bool dbcb_depth_copy_enabled;
279 bool dbcb_stencil_copy_enabled;
280 unsigned dbcb_copy_sample;
281 bool db_flush_depth_inplace;
282 bool db_flush_stencil_inplace;
283 bool db_depth_clear;
284 bool db_depth_disable_expclear;
285 bool db_stencil_clear;
286 bool db_stencil_disable_expclear;
287 unsigned ps_db_shader_control;
288 bool occlusion_queries_disabled;
289
290 /* Emitted draw state. */
291 int last_base_vertex;
292 int last_start_instance;
293 int last_sh_base_reg;
294 int last_primitive_restart_en;
295 int last_restart_index;
296 int last_gs_out_prim;
297 int last_prim;
298 int last_multi_vgt_param;
299 int last_ls_hs_config;
300 int last_rast_prim;
301 unsigned last_sc_line_stipple;
302 int current_rast_prim; /* primitive type after TES, GS */
303 unsigned last_gsvs_itemsize;
304
305 /* Scratch buffer */
306 struct r600_resource *scratch_buffer;
307 boolean emit_scratch_reloc;
308 unsigned scratch_waves;
309 unsigned spi_tmpring_size;
310
311 struct r600_resource *compute_scratch_buffer;
312
313 /* Emitted derived tessellation state. */
314 struct si_shader *last_ls; /* local shader (VS) */
315 struct si_shader_selector *last_tcs;
316 int last_num_tcs_input_cp;
317 int last_tes_sh_base;
318
319 /* Debug state. */
320 bool is_debug;
321 uint32_t *last_ib;
322 unsigned last_ib_dw_size;
323 struct r600_resource *last_trace_buf;
324 struct r600_resource *trace_buf;
325 unsigned trace_id;
326 uint64_t dmesg_timestamp;
327 unsigned last_bo_count;
328 struct radeon_bo_list_item *last_bo_list;
329 };
330
331 /* cik_sdma.c */
332 void cik_sdma_copy(struct pipe_context *ctx,
333 struct pipe_resource *dst,
334 unsigned dst_level,
335 unsigned dstx, unsigned dsty, unsigned dstz,
336 struct pipe_resource *src,
337 unsigned src_level,
338 const struct pipe_box *src_box);
339
340 /* si_blit.c */
341 void si_init_blit_functions(struct si_context *sctx);
342 void si_decompress_textures(struct si_context *sctx);
343 void si_resource_copy_region(struct pipe_context *ctx,
344 struct pipe_resource *dst,
345 unsigned dst_level,
346 unsigned dstx, unsigned dsty, unsigned dstz,
347 struct pipe_resource *src,
348 unsigned src_level,
349 const struct pipe_box *src_box);
350
351 /* si_cp_dma.c */
352 void si_copy_buffer(struct si_context *sctx,
353 struct pipe_resource *dst, struct pipe_resource *src,
354 uint64_t dst_offset, uint64_t src_offset, unsigned size,
355 bool is_framebuffer);
356 void si_init_cp_dma_functions(struct si_context *sctx);
357
358 /* si_debug.c */
359 void si_init_debug_functions(struct si_context *sctx);
360 void si_check_vm_faults(struct si_context *sctx);
361 bool si_replace_shader(unsigned num, struct radeon_shader_binary *binary);
362
363 /* si_dma.c */
364 void si_dma_copy(struct pipe_context *ctx,
365 struct pipe_resource *dst,
366 unsigned dst_level,
367 unsigned dstx, unsigned dsty, unsigned dstz,
368 struct pipe_resource *src,
369 unsigned src_level,
370 const struct pipe_box *src_box);
371
372 /* si_hw_context.c */
373 void si_context_gfx_flush(void *context, unsigned flags,
374 struct pipe_fence_handle **fence);
375 void si_begin_new_cs(struct si_context *ctx);
376 void si_need_cs_space(struct si_context *ctx);
377
378 /* si_compute.c */
379 void si_init_compute_functions(struct si_context *sctx);
380
381 /* si_perfcounters.c */
382 void si_init_perfcounters(struct si_screen *screen);
383
384 /* si_uvd.c */
385 struct pipe_video_codec *si_uvd_create_decoder(struct pipe_context *context,
386 const struct pipe_video_codec *templ);
387
388 struct pipe_video_buffer *si_video_buffer_create(struct pipe_context *pipe,
389 const struct pipe_video_buffer *tmpl);
390
391 /*
392 * common helpers
393 */
394
395 static inline struct r600_resource *
396 si_resource_create_custom(struct pipe_screen *screen,
397 unsigned usage, unsigned size)
398 {
399 assert(size);
400 return r600_resource(pipe_buffer_create(screen,
401 PIPE_BIND_CUSTOM, usage, size));
402 }
403
404 static inline void
405 si_invalidate_draw_sh_constants(struct si_context *sctx)
406 {
407 sctx->last_base_vertex = SI_BASE_VERTEX_UNKNOWN;
408 sctx->last_start_instance = -1; /* reset to an unknown value */
409 sctx->last_sh_base_reg = -1; /* reset to an unknown value */
410 }
411
412 static inline void
413 si_set_atom_dirty(struct si_context *sctx,
414 struct r600_atom *atom, bool dirty)
415 {
416 unsigned bit = 1 << (atom->id - 1);
417
418 if (dirty)
419 sctx->dirty_atoms |= bit;
420 else
421 sctx->dirty_atoms &= ~bit;
422 }
423
424 static inline bool
425 si_is_atom_dirty(struct si_context *sctx,
426 struct r600_atom *atom)
427 {
428 unsigned bit = 1 << (atom->id - 1);
429
430 return sctx->dirty_atoms & bit;
431 }
432
433 static inline void
434 si_mark_atom_dirty(struct si_context *sctx,
435 struct r600_atom *atom)
436 {
437 si_set_atom_dirty(sctx, atom, true);
438 }
439
440 #endif