radeonsi: enable distributed tess on multi-SE parts only
[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
67 #define SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER (SI_CONTEXT_FLUSH_AND_INV_CB | \
68 SI_CONTEXT_FLUSH_AND_INV_CB_META | \
69 SI_CONTEXT_FLUSH_AND_INV_DB | \
70 SI_CONTEXT_FLUSH_AND_INV_DB_META)
71
72 #define SI_ENCODE_TRACE_POINT(id) (0xcafe0000 | ((id) & 0xffff))
73 #define SI_IS_TRACE_POINT(x) (((x) & 0xcafe0000) == 0xcafe0000)
74 #define SI_GET_TRACE_POINT_ID(x) ((x) & 0xffff)
75
76 #define SI_MAX_BORDER_COLORS 4096
77
78 struct si_compute;
79 struct hash_table;
80 struct u_suballocator;
81
82 struct si_screen {
83 struct r600_common_screen b;
84 unsigned gs_table_depth;
85 unsigned tess_offchip_block_dw_size;
86 bool has_distributed_tess;
87
88 /* Whether shaders are monolithic (1-part) or separate (3-part). */
89 bool use_monolithic_shaders;
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 *ps_prologs;
96 struct si_shader_part *ps_epilogs;
97
98 /* Shader cache in memory.
99 *
100 * Design & limitations:
101 * - The shader cache is per screen (= per process), never saved to
102 * disk, and skips redundant shader compilations from TGSI to bytecode.
103 * - It can only be used with one-variant-per-shader support, in which
104 * case only the main (typically middle) part of shaders is cached.
105 * - Only VS, TCS, TES, PS are cached, out of which only the hw VS
106 * variants of VS and TES are cached, so LS and ES aren't.
107 * - GS and CS aren't cached, but it's certainly possible to cache
108 * those as well.
109 */
110 pipe_mutex shader_cache_mutex;
111 struct hash_table *shader_cache;
112 };
113
114 struct si_blend_color {
115 struct r600_atom atom;
116 struct pipe_blend_color state;
117 };
118
119 struct si_sampler_view {
120 struct pipe_sampler_view base;
121 struct list_head list;
122 /* [0..7] = image descriptor
123 * [4..7] = buffer descriptor */
124 uint32_t state[8];
125 uint32_t fmask_state[8];
126 const struct radeon_surf_level *base_level_info;
127 unsigned base_level;
128 unsigned block_width;
129 bool is_stencil_sampler;
130 };
131
132 struct si_sampler_state {
133 uint32_t val[4];
134 };
135
136 struct si_cs_shader_state {
137 struct si_compute *program;
138 struct si_compute *emitted_program;
139 unsigned offset;
140 bool initialized;
141 bool uses_scratch;
142 };
143
144 struct si_textures_info {
145 struct si_sampler_views views;
146 uint32_t depth_texture_mask; /* which textures are depth */
147 uint32_t compressed_colortex_mask;
148 };
149
150 struct si_images_info {
151 struct pipe_image_view views[SI_NUM_IMAGES];
152 uint32_t compressed_colortex_mask;
153 unsigned enabled_mask;
154 };
155
156 struct si_framebuffer {
157 struct r600_atom atom;
158 struct pipe_framebuffer_state state;
159 unsigned nr_samples;
160 unsigned log_samples;
161 unsigned cb0_is_integer;
162 unsigned compressed_cb_mask;
163 unsigned spi_shader_col_format;
164 unsigned spi_shader_col_format_alpha;
165 unsigned spi_shader_col_format_blend;
166 unsigned spi_shader_col_format_blend_alpha;
167 unsigned color_is_int8; /* bitmask */
168 unsigned dirty_cbufs;
169 bool dirty_zsbuf;
170 bool any_dst_linear;
171 };
172
173 struct si_clip_state {
174 struct r600_atom atom;
175 struct pipe_clip_state state;
176 };
177
178 struct si_sample_mask {
179 struct r600_atom atom;
180 uint16_t sample_mask;
181 };
182
183 /* A shader state consists of the shader selector, which is a constant state
184 * object shared by multiple contexts and shouldn't be modified, and
185 * the current shader variant selected for this context.
186 */
187 struct si_shader_ctx_state {
188 struct si_shader_selector *cso;
189 struct si_shader *current;
190 };
191
192 struct si_context {
193 struct r600_common_context b;
194 struct blitter_context *blitter;
195 void *custom_dsa_flush;
196 void *custom_blend_resolve;
197 void *custom_blend_decompress;
198 void *custom_blend_fastclear;
199 void *custom_blend_dcc_decompress;
200 struct si_screen *screen;
201
202 struct radeon_winsys_cs *ce_ib;
203 struct radeon_winsys_cs *ce_preamble_ib;
204 bool ce_need_synchronization;
205 struct u_suballocator *ce_suballocator;
206
207 struct pipe_fence_handle *last_gfx_fence;
208 struct si_shader_ctx_state fixed_func_tcs_shader;
209 LLVMTargetMachineRef tm;
210 bool gfx_flush_in_progress;
211
212 /* Atoms (direct states). */
213 union si_state_atoms atoms;
214 unsigned dirty_atoms; /* mask */
215 /* PM4 states (precomputed immutable states) */
216 union si_state queued;
217 union si_state emitted;
218
219 /* Atom declarations. */
220 struct r600_atom cache_flush;
221 struct si_framebuffer framebuffer;
222 struct r600_atom msaa_sample_locs;
223 struct r600_atom db_render_state;
224 struct r600_atom msaa_config;
225 struct si_sample_mask sample_mask;
226 struct r600_atom cb_render_state;
227 struct si_blend_color blend_color;
228 struct r600_atom clip_regs;
229 struct si_clip_state clip_state;
230 struct si_shader_data shader_userdata;
231 struct si_stencil_ref stencil_ref;
232 struct r600_atom spi_map;
233
234 /* Precomputed states. */
235 struct si_pm4_state *init_config;
236 struct si_pm4_state *init_config_gs_rings;
237 bool init_config_has_vgt_flush;
238 struct si_pm4_state *vgt_shader_config[4];
239
240 /* shaders */
241 struct si_shader_ctx_state ps_shader;
242 struct si_shader_ctx_state gs_shader;
243 struct si_shader_ctx_state vs_shader;
244 struct si_shader_ctx_state tcs_shader;
245 struct si_shader_ctx_state tes_shader;
246 struct si_cs_shader_state cs_shader_state;
247
248 /* shader information */
249 struct si_vertex_element *vertex_elements;
250 unsigned sprite_coord_enable;
251 bool flatshade;
252
253 /* shader descriptors */
254 struct si_descriptors vertex_buffers;
255 struct si_descriptors descriptors[SI_NUM_DESCS];
256 unsigned descriptors_dirty;
257 struct si_buffer_resources rw_buffers;
258 struct si_buffer_resources const_buffers[SI_NUM_SHADERS];
259 struct si_buffer_resources shader_buffers[SI_NUM_SHADERS];
260 struct si_textures_info samplers[SI_NUM_SHADERS];
261 struct si_images_info images[SI_NUM_SHADERS];
262
263 /* other shader resources */
264 struct pipe_constant_buffer null_const_buf; /* used for set_constant_buffer(NULL) on CIK */
265 struct pipe_resource *esgs_ring;
266 struct pipe_resource *gsvs_ring;
267 struct pipe_resource *tf_ring;
268 struct pipe_resource *tess_offchip_ring;
269 union pipe_color_union *border_color_table; /* in CPU memory, any endian */
270 struct r600_resource *border_color_buffer;
271 union pipe_color_union *border_color_map; /* in VRAM (slow access), little endian */
272 unsigned border_color_count;
273
274 /* Vertex and index buffers. */
275 bool vertex_buffers_dirty;
276 struct pipe_index_buffer index_buffer;
277 struct pipe_vertex_buffer vertex_buffer[SI_NUM_VERTEX_BUFFERS];
278
279 /* MSAA config state. */
280 int ps_iter_samples;
281 bool smoothing_enabled;
282
283 /* DB render state. */
284 bool dbcb_depth_copy_enabled;
285 bool dbcb_stencil_copy_enabled;
286 unsigned dbcb_copy_sample;
287 bool db_flush_depth_inplace;
288 bool db_flush_stencil_inplace;
289 bool db_depth_clear;
290 bool db_depth_disable_expclear;
291 bool db_stencil_clear;
292 bool db_stencil_disable_expclear;
293 unsigned ps_db_shader_control;
294 bool occlusion_queries_disabled;
295
296 /* Emitted draw state. */
297 int last_base_vertex;
298 int last_start_instance;
299 int last_sh_base_reg;
300 int last_primitive_restart_en;
301 int last_restart_index;
302 int last_gs_out_prim;
303 int last_prim;
304 int last_multi_vgt_param;
305 int last_ls_hs_config;
306 int last_rast_prim;
307 unsigned last_sc_line_stipple;
308 int last_vtx_reuse_depth;
309 int current_rast_prim; /* primitive type after TES, GS */
310 unsigned last_gsvs_itemsize;
311
312 /* Scratch buffer */
313 struct r600_resource *scratch_buffer;
314 bool emit_scratch_reloc;
315 unsigned scratch_waves;
316 unsigned spi_tmpring_size;
317
318 struct r600_resource *compute_scratch_buffer;
319
320 /* Emitted derived tessellation state. */
321 struct si_shader *last_ls; /* local shader (VS) */
322 struct si_shader_selector *last_tcs;
323 int last_num_tcs_input_cp;
324 int last_tes_sh_base;
325
326 /* Debug state. */
327 bool is_debug;
328 struct radeon_saved_cs last_gfx;
329 struct r600_resource *last_trace_buf;
330 struct r600_resource *trace_buf;
331 unsigned trace_id;
332 uint64_t dmesg_timestamp;
333
334 /* Other state */
335 bool need_check_render_feedback;
336 };
337
338 /* cik_sdma.c */
339 void cik_init_sdma_functions(struct si_context *sctx);
340
341 /* si_blit.c */
342 void si_init_blit_functions(struct si_context *sctx);
343 void si_decompress_graphics_textures(struct si_context *sctx);
344 void si_decompress_compute_textures(struct si_context *sctx);
345 void si_resource_copy_region(struct pipe_context *ctx,
346 struct pipe_resource *dst,
347 unsigned dst_level,
348 unsigned dstx, unsigned dsty, unsigned dstz,
349 struct pipe_resource *src,
350 unsigned src_level,
351 const struct pipe_box *src_box);
352
353 /* si_cp_dma.c */
354 void si_copy_buffer(struct si_context *sctx,
355 struct pipe_resource *dst, struct pipe_resource *src,
356 uint64_t dst_offset, uint64_t src_offset, unsigned size);
357 void si_init_cp_dma_functions(struct si_context *sctx);
358
359 /* si_debug.c */
360 void si_init_debug_functions(struct si_context *sctx);
361 void si_check_vm_faults(struct r600_common_context *ctx,
362 struct radeon_saved_cs *saved, enum ring_type ring);
363 bool si_replace_shader(unsigned num, struct radeon_shader_binary *binary);
364
365 /* si_dma.c */
366 void si_init_dma_functions(struct si_context *sctx);
367
368 /* si_hw_context.c */
369 void si_context_gfx_flush(void *context, unsigned flags,
370 struct pipe_fence_handle **fence);
371 void si_begin_new_cs(struct si_context *ctx);
372 void si_need_cs_space(struct si_context *ctx);
373
374 /* si_compute.c */
375 void si_init_compute_functions(struct si_context *sctx);
376
377 /* si_perfcounters.c */
378 void si_init_perfcounters(struct si_screen *screen);
379
380 /* si_uvd.c */
381 struct pipe_video_codec *si_uvd_create_decoder(struct pipe_context *context,
382 const struct pipe_video_codec *templ);
383
384 struct pipe_video_buffer *si_video_buffer_create(struct pipe_context *pipe,
385 const struct pipe_video_buffer *tmpl);
386
387 /*
388 * common helpers
389 */
390
391 static inline struct r600_resource *
392 si_resource_create_custom(struct pipe_screen *screen,
393 unsigned usage, unsigned size)
394 {
395 assert(size);
396 return r600_resource(pipe_buffer_create(screen,
397 PIPE_BIND_CUSTOM, usage, size));
398 }
399
400 static inline void
401 si_invalidate_draw_sh_constants(struct si_context *sctx)
402 {
403 sctx->last_base_vertex = SI_BASE_VERTEX_UNKNOWN;
404 sctx->last_start_instance = -1; /* reset to an unknown value */
405 sctx->last_sh_base_reg = -1; /* reset to an unknown value */
406 }
407
408 static inline void
409 si_set_atom_dirty(struct si_context *sctx,
410 struct r600_atom *atom, bool dirty)
411 {
412 unsigned bit = 1 << (atom->id - 1);
413
414 if (dirty)
415 sctx->dirty_atoms |= bit;
416 else
417 sctx->dirty_atoms &= ~bit;
418 }
419
420 static inline bool
421 si_is_atom_dirty(struct si_context *sctx,
422 struct r600_atom *atom)
423 {
424 unsigned bit = 1 << (atom->id - 1);
425
426 return sctx->dirty_atoms & bit;
427 }
428
429 static inline void
430 si_mark_atom_dirty(struct si_context *sctx,
431 struct r600_atom *atom)
432 {
433 si_set_atom_dirty(sctx, atom, true);
434 }
435
436 #endif