radeonsi: add a debug flag for unsafe math LLVM optimizations
[mesa.git] / src / gallium / drivers / radeon / r600_pipe_common.h
1 /*
2 * Copyright 2013 Advanced Micro Devices, Inc.
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 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * 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 NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 * SOFTWARE.
22 *
23 * Authors: Marek Olšák <maraeo@gmail.com>
24 *
25 */
26
27 /**
28 * This file contains common screen and context structures and functions
29 * for r600g and radeonsi.
30 */
31
32 #ifndef R600_PIPE_COMMON_H
33 #define R600_PIPE_COMMON_H
34
35 #include <stdio.h>
36
37 #include "radeon/radeon_winsys.h"
38
39 #include "util/u_blitter.h"
40 #include "util/list.h"
41 #include "util/u_range.h"
42 #include "util/u_slab.h"
43 #include "util/u_suballoc.h"
44 #include "util/u_transfer.h"
45
46 #define ATI_VENDOR_ID 0x1002
47
48 #define R600_RESOURCE_FLAG_TRANSFER (PIPE_RESOURCE_FLAG_DRV_PRIV << 0)
49 #define R600_RESOURCE_FLAG_FLUSHED_DEPTH (PIPE_RESOURCE_FLAG_DRV_PRIV << 1)
50 #define R600_RESOURCE_FLAG_FORCE_TILING (PIPE_RESOURCE_FLAG_DRV_PRIV << 2)
51 #define R600_RESOURCE_FLAG_DISABLE_DCC (PIPE_RESOURCE_FLAG_DRV_PRIV << 3)
52
53 #define R600_CONTEXT_STREAMOUT_FLUSH (1u << 0)
54 /* Pipeline & streamout query controls. */
55 #define R600_CONTEXT_START_PIPELINE_STATS (1u << 1)
56 #define R600_CONTEXT_STOP_PIPELINE_STATS (1u << 2)
57 #define R600_CONTEXT_PRIVATE_FLAG (1u << 3)
58
59 /* special primitive types */
60 #define R600_PRIM_RECTANGLE_LIST PIPE_PRIM_MAX
61
62 /* Debug flags. */
63 /* logging */
64 #define DBG_TEX (1 << 0)
65 /* gap - reuse */
66 #define DBG_COMPUTE (1 << 2)
67 #define DBG_VM (1 << 3)
68 /* gap - reuse */
69 /* shader logging */
70 #define DBG_FS (1 << 5)
71 #define DBG_VS (1 << 6)
72 #define DBG_GS (1 << 7)
73 #define DBG_PS (1 << 8)
74 #define DBG_CS (1 << 9)
75 #define DBG_TCS (1 << 10)
76 #define DBG_TES (1 << 11)
77 #define DBG_NO_IR (1 << 12)
78 #define DBG_NO_TGSI (1 << 13)
79 #define DBG_NO_ASM (1 << 14)
80 #define DBG_PREOPT_IR (1 << 15)
81 /* gaps */
82 #define DBG_TEST_DMA (1 << 20)
83 /* Bits 21-31 are reserved for the r600g driver. */
84 /* features */
85 #define DBG_NO_ASYNC_DMA (1llu << 32)
86 #define DBG_NO_HYPERZ (1llu << 33)
87 #define DBG_NO_DISCARD_RANGE (1llu << 34)
88 #define DBG_NO_2D_TILING (1llu << 35)
89 #define DBG_NO_TILING (1llu << 36)
90 #define DBG_SWITCH_ON_EOP (1llu << 37)
91 #define DBG_FORCE_DMA (1llu << 38)
92 #define DBG_PRECOMPILE (1llu << 39)
93 #define DBG_INFO (1llu << 40)
94 #define DBG_NO_WC (1llu << 41)
95 #define DBG_CHECK_VM (1llu << 42)
96 #define DBG_NO_DCC (1llu << 43)
97 #define DBG_NO_DCC_CLEAR (1llu << 44)
98 #define DBG_NO_RB_PLUS (1llu << 45)
99 #define DBG_SI_SCHED (1llu << 46)
100 #define DBG_MONOLITHIC_SHADERS (1llu << 47)
101 #define DBG_NO_CE (1llu << 48)
102 #define DBG_UNSAFE_MATH (1llu << 49)
103
104 #define R600_MAP_BUFFER_ALIGNMENT 64
105 #define R600_MAX_VIEWPORTS 16
106
107 enum r600_coherency {
108 R600_COHERENCY_NONE, /* no cache flushes needed */
109 R600_COHERENCY_SHADER,
110 R600_COHERENCY_CB_META,
111 };
112
113 #ifdef PIPE_ARCH_BIG_ENDIAN
114 #define R600_BIG_ENDIAN 1
115 #else
116 #define R600_BIG_ENDIAN 0
117 #endif
118
119 struct r600_common_context;
120 struct r600_perfcounters;
121 struct tgsi_shader_info;
122
123 struct radeon_shader_reloc {
124 char name[32];
125 uint64_t offset;
126 };
127
128 struct radeon_shader_binary {
129 /** Shader code */
130 unsigned char *code;
131 unsigned code_size;
132
133 /** Config/Context register state that accompanies this shader.
134 * This is a stream of dword pairs. First dword contains the
135 * register address, the second dword contains the value.*/
136 unsigned char *config;
137 unsigned config_size;
138
139 /** The number of bytes of config information for each global symbol.
140 */
141 unsigned config_size_per_symbol;
142
143 /** Constant data accessed by the shader. This will be uploaded
144 * into a constant buffer. */
145 unsigned char *rodata;
146 unsigned rodata_size;
147
148 /** List of symbol offsets for the shader */
149 uint64_t *global_symbol_offsets;
150 unsigned global_symbol_count;
151
152 struct radeon_shader_reloc *relocs;
153 unsigned reloc_count;
154
155 /** Disassembled shader in a string. */
156 char *disasm_string;
157 };
158
159 void radeon_shader_binary_init(struct radeon_shader_binary *b);
160 void radeon_shader_binary_clean(struct radeon_shader_binary *b);
161
162 /* Only 32-bit buffer allocations are supported, gallium doesn't support more
163 * at the moment.
164 */
165 struct r600_resource {
166 struct u_resource b;
167
168 /* Winsys objects. */
169 struct pb_buffer *buf;
170 uint64_t gpu_address;
171
172 /* Resource state. */
173 enum radeon_bo_domain domains;
174
175 /* The buffer range which is initialized (with a write transfer,
176 * streamout, DMA, or as a random access target). The rest of
177 * the buffer is considered invalid and can be mapped unsynchronized.
178 *
179 * This allows unsychronized mapping of a buffer range which hasn't
180 * been used yet. It's for applications which forget to use
181 * the unsynchronized map flag and expect the driver to figure it out.
182 */
183 struct util_range valid_buffer_range;
184
185 /* For buffers only. This indicates that a write operation has been
186 * performed by TC L2, but the cache hasn't been flushed.
187 * Any hw block which doesn't use or bypasses TC L2 should check this
188 * flag and flush the cache before using the buffer.
189 *
190 * For example, TC L2 must be flushed if a buffer which has been
191 * modified by a shader store instruction is about to be used as
192 * an index buffer. The reason is that VGT DMA index fetching doesn't
193 * use TC L2.
194 */
195 bool TC_L2_dirty;
196
197 /* Whether the resource has been exported via resource_get_handle. */
198 bool is_shared;
199 unsigned external_usage; /* PIPE_HANDLE_USAGE_* */
200 };
201
202 struct r600_transfer {
203 struct pipe_transfer transfer;
204 struct r600_resource *staging;
205 unsigned offset;
206 };
207
208 struct r600_fmask_info {
209 uint64_t offset;
210 uint64_t size;
211 unsigned alignment;
212 unsigned pitch_in_pixels;
213 unsigned bank_height;
214 unsigned slice_tile_max;
215 unsigned tile_mode_index;
216 };
217
218 struct r600_cmask_info {
219 uint64_t offset;
220 uint64_t size;
221 unsigned alignment;
222 unsigned pitch;
223 unsigned height;
224 unsigned xalign;
225 unsigned yalign;
226 unsigned slice_tile_max;
227 unsigned base_address_reg;
228 };
229
230 struct r600_htile_info {
231 unsigned pitch;
232 unsigned height;
233 unsigned xalign;
234 unsigned yalign;
235 };
236
237 struct r600_texture {
238 struct r600_resource resource;
239
240 uint64_t size;
241 unsigned num_level0_transfers;
242 bool is_depth;
243 unsigned dirty_level_mask; /* each bit says if that mipmap is compressed */
244 unsigned stencil_dirty_level_mask; /* each bit says if that mipmap is compressed */
245 struct r600_texture *flushed_depth_texture;
246 boolean is_flushing_texture;
247 struct radeon_surf surface;
248
249 /* Colorbuffer compression and fast clear. */
250 struct r600_fmask_info fmask;
251 struct r600_cmask_info cmask;
252 struct r600_resource *cmask_buffer;
253 uint64_t dcc_offset; /* 0 = disabled */
254 unsigned cb_color_info; /* fast clear enable bit */
255 unsigned color_clear_value[2];
256 unsigned last_msaa_resolve_target_micro_mode;
257
258 /* Depth buffer compression and fast clear. */
259 struct r600_htile_info htile;
260 struct r600_resource *htile_buffer;
261 bool depth_cleared; /* if it was cleared at least once */
262 float depth_clear_value;
263 bool stencil_cleared; /* if it was cleared at least once */
264 uint8_t stencil_clear_value;
265
266 bool non_disp_tiling; /* R600-Cayman only */
267
268 /* Counter that should be non-zero if the texture is bound to a
269 * framebuffer. Implemented in radeonsi only.
270 */
271 uint32_t framebuffers_bound;
272 };
273
274 struct r600_surface {
275 struct pipe_surface base;
276 const struct radeon_surf_level *level_info;
277
278 bool color_initialized;
279 bool depth_initialized;
280
281 /* Misc. color flags. */
282 bool alphatest_bypass;
283 bool export_16bpc;
284 bool color_is_int8;
285
286 /* Color registers. */
287 unsigned cb_color_info;
288 unsigned cb_color_base;
289 unsigned cb_color_view;
290 unsigned cb_color_size; /* R600 only */
291 unsigned cb_color_dim; /* EG only */
292 unsigned cb_color_pitch; /* EG and later */
293 unsigned cb_color_slice; /* EG and later */
294 unsigned cb_color_attrib; /* EG and later */
295 unsigned cb_dcc_control; /* VI and later */
296 unsigned cb_color_fmask; /* CB_COLORn_FMASK (EG and later) or CB_COLORn_FRAG (r600) */
297 unsigned cb_color_fmask_slice; /* EG and later */
298 unsigned cb_color_cmask; /* CB_COLORn_TILE (r600 only) */
299 unsigned cb_color_mask; /* R600 only */
300 unsigned spi_shader_col_format; /* SI+, no blending, no alpha-to-coverage. */
301 unsigned spi_shader_col_format_alpha; /* SI+, alpha-to-coverage */
302 unsigned spi_shader_col_format_blend; /* SI+, blending without alpha. */
303 unsigned spi_shader_col_format_blend_alpha; /* SI+, blending with alpha. */
304 struct r600_resource *cb_buffer_fmask; /* Used for FMASK relocations. R600 only */
305 struct r600_resource *cb_buffer_cmask; /* Used for CMASK relocations. R600 only */
306
307 /* DB registers. */
308 unsigned db_depth_info; /* R600 only, then SI and later */
309 unsigned db_z_info; /* EG and later */
310 unsigned db_depth_base; /* DB_Z_READ/WRITE_BASE (EG and later) or DB_DEPTH_BASE (r600) */
311 unsigned db_depth_view;
312 unsigned db_depth_size;
313 unsigned db_depth_slice; /* EG and later */
314 unsigned db_stencil_base; /* EG and later */
315 unsigned db_stencil_info; /* EG and later */
316 unsigned db_prefetch_limit; /* R600 only */
317 unsigned db_htile_surface;
318 unsigned db_htile_data_base;
319 unsigned db_preload_control; /* EG and later */
320 unsigned pa_su_poly_offset_db_fmt_cntl;
321 };
322
323 struct r600_common_screen {
324 struct pipe_screen b;
325 struct radeon_winsys *ws;
326 enum radeon_family family;
327 enum chip_class chip_class;
328 struct radeon_info info;
329 uint64_t debug_flags;
330 bool has_cp_dma;
331 bool has_streamout;
332
333 /* Texture filter settings. */
334 int force_aniso; /* -1 = disabled */
335
336 /* Auxiliary context. Mainly used to initialize resources.
337 * It must be locked prior to using and flushed before unlocking. */
338 struct pipe_context *aux_context;
339 pipe_mutex aux_context_lock;
340
341 /* This must be in the screen, because UE4 uses one context for
342 * compilation and another one for rendering.
343 */
344 unsigned num_compilations;
345 /* Along with ST_DEBUG=precompile, this should show if applications
346 * are loading shaders on demand. This is a monotonic counter.
347 */
348 unsigned num_shaders_created;
349
350 /* GPU load thread. */
351 pipe_mutex gpu_load_mutex;
352 pipe_thread gpu_load_thread;
353 unsigned gpu_load_counter_busy;
354 unsigned gpu_load_counter_idle;
355 volatile unsigned gpu_load_stop_thread; /* bool */
356
357 char renderer_string[64];
358
359 /* Performance counters. */
360 struct r600_perfcounters *perfcounters;
361
362 /* If pipe_screen wants to re-emit the framebuffer state of all
363 * contexts, it should atomically increment this. Each context will
364 * compare this with its own last known value of the counter before
365 * drawing and re-emit the framebuffer state accordingly.
366 */
367 unsigned dirty_fb_counter;
368
369 /* Atomically increment this counter when an existing texture's
370 * metadata is enabled or disabled in a way that requires changing
371 * contexts' compressed texture binding masks.
372 */
373 unsigned compressed_colortex_counter;
374
375 /* Atomically increment this counter when an existing texture's
376 * backing buffer or tile mode parameters have changed that requires
377 * recomputation of shader descriptors.
378 */
379 unsigned dirty_tex_descriptor_counter;
380
381 void (*query_opaque_metadata)(struct r600_common_screen *rscreen,
382 struct r600_texture *rtex,
383 struct radeon_bo_metadata *md);
384
385 void (*apply_opaque_metadata)(struct r600_common_screen *rscreen,
386 struct r600_texture *rtex,
387 struct radeon_bo_metadata *md);
388 };
389
390 /* This encapsulates a state or an operation which can emitted into the GPU
391 * command stream. */
392 struct r600_atom {
393 void (*emit)(struct r600_common_context *ctx, struct r600_atom *state);
394 unsigned num_dw;
395 unsigned short id;
396 };
397
398 struct r600_so_target {
399 struct pipe_stream_output_target b;
400
401 /* The buffer where BUFFER_FILLED_SIZE is stored. */
402 struct r600_resource *buf_filled_size;
403 unsigned buf_filled_size_offset;
404 bool buf_filled_size_valid;
405
406 unsigned stride_in_dw;
407 };
408
409 struct r600_streamout {
410 struct r600_atom begin_atom;
411 bool begin_emitted;
412 unsigned num_dw_for_end;
413
414 unsigned enabled_mask;
415 unsigned num_targets;
416 struct r600_so_target *targets[PIPE_MAX_SO_BUFFERS];
417
418 unsigned append_bitmask;
419 bool suspended;
420
421 /* External state which comes from the vertex shader,
422 * it must be set explicitly when binding a shader. */
423 unsigned *stride_in_dw;
424 unsigned enabled_stream_buffers_mask; /* stream0 buffers0-3 in 4 LSB */
425
426 /* The state of VGT_STRMOUT_BUFFER_(CONFIG|EN). */
427 unsigned hw_enabled_mask;
428
429 /* The state of VGT_STRMOUT_(CONFIG|EN). */
430 struct r600_atom enable_atom;
431 bool streamout_enabled;
432 bool prims_gen_query_enabled;
433 int num_prims_gen_queries;
434 };
435
436 struct r600_signed_scissor {
437 int minx;
438 int miny;
439 int maxx;
440 int maxy;
441 };
442
443 struct r600_scissors {
444 struct r600_atom atom;
445 unsigned dirty_mask;
446 struct pipe_scissor_state states[R600_MAX_VIEWPORTS];
447 };
448
449 struct r600_viewports {
450 struct r600_atom atom;
451 unsigned dirty_mask;
452 struct pipe_viewport_state states[R600_MAX_VIEWPORTS];
453 struct r600_signed_scissor as_scissor[R600_MAX_VIEWPORTS];
454 };
455
456 struct r600_ring {
457 struct radeon_winsys_cs *cs;
458 void (*flush)(void *ctx, unsigned flags,
459 struct pipe_fence_handle **fence);
460 };
461
462 struct r600_common_context {
463 struct pipe_context b; /* base class */
464
465 struct r600_common_screen *screen;
466 struct radeon_winsys *ws;
467 struct radeon_winsys_ctx *ctx;
468 enum radeon_family family;
469 enum chip_class chip_class;
470 struct r600_ring gfx;
471 struct r600_ring dma;
472 struct pipe_fence_handle *last_sdma_fence;
473 unsigned initial_gfx_cs_size;
474 unsigned gpu_reset_counter;
475 unsigned last_dirty_fb_counter;
476 unsigned last_compressed_colortex_counter;
477 unsigned last_dirty_tex_descriptor_counter;
478
479 struct u_upload_mgr *uploader;
480 struct u_suballocator *allocator_zeroed_memory;
481 struct util_slab_mempool pool_transfers;
482
483 /* Current unaccounted memory usage. */
484 uint64_t vram;
485 uint64_t gtt;
486
487 /* States. */
488 struct r600_streamout streamout;
489 struct r600_scissors scissors;
490 struct r600_viewports viewports;
491 bool scissor_enabled;
492 bool vs_writes_viewport_index;
493 bool vs_disables_clipping_viewport;
494
495 /* Additional context states. */
496 unsigned flags; /* flush flags */
497
498 /* Queries. */
499 /* Maintain the list of active queries for pausing between IBs. */
500 int num_occlusion_queries;
501 int num_perfect_occlusion_queries;
502 struct list_head active_queries;
503 unsigned num_cs_dw_queries_suspend;
504 /* Additional hardware info. */
505 unsigned backend_mask;
506 unsigned max_db; /* for OQ */
507 /* Misc stats. */
508 unsigned num_draw_calls;
509 unsigned num_spill_draw_calls;
510 unsigned num_compute_calls;
511 unsigned num_spill_compute_calls;
512 unsigned num_dma_calls;
513 uint64_t num_alloc_tex_transfer_bytes;
514
515 /* Render condition. */
516 struct r600_atom render_cond_atom;
517 struct pipe_query *render_cond;
518 unsigned render_cond_mode;
519 boolean render_cond_invert;
520 bool render_cond_force_off; /* for u_blitter */
521
522 /* MSAA sample locations.
523 * The first index is the sample index.
524 * The second index is the coordinate: X, Y. */
525 float sample_locations_1x[1][2];
526 float sample_locations_2x[2][2];
527 float sample_locations_4x[4][2];
528 float sample_locations_8x[8][2];
529 float sample_locations_16x[16][2];
530
531 /* The list of all texture buffer objects in this context.
532 * This list is walked when a buffer is invalidated/reallocated and
533 * the GPU addresses are updated. */
534 struct list_head texture_buffers;
535
536 struct pipe_debug_callback debug;
537
538 /* Copy one resource to another using async DMA. */
539 void (*dma_copy)(struct pipe_context *ctx,
540 struct pipe_resource *dst,
541 unsigned dst_level,
542 unsigned dst_x, unsigned dst_y, unsigned dst_z,
543 struct pipe_resource *src,
544 unsigned src_level,
545 const struct pipe_box *src_box);
546
547 void (*clear_buffer)(struct pipe_context *ctx, struct pipe_resource *dst,
548 uint64_t offset, uint64_t size, unsigned value,
549 enum r600_coherency coher);
550
551 void (*blit_decompress_depth)(struct pipe_context *ctx,
552 struct r600_texture *texture,
553 struct r600_texture *staging,
554 unsigned first_level, unsigned last_level,
555 unsigned first_layer, unsigned last_layer,
556 unsigned first_sample, unsigned last_sample);
557
558 void (*decompress_dcc)(struct pipe_context *ctx,
559 struct r600_texture *rtex);
560
561 /* Reallocate the buffer and update all resource bindings where
562 * the buffer is bound, including all resource descriptors. */
563 void (*invalidate_buffer)(struct pipe_context *ctx, struct pipe_resource *buf);
564
565 /* Enable or disable occlusion queries. */
566 void (*set_occlusion_query_state)(struct pipe_context *ctx, bool enable);
567
568 /* This ensures there is enough space in the command stream. */
569 void (*need_gfx_cs_space)(struct pipe_context *ctx, unsigned num_dw,
570 bool include_draw_vbo);
571
572 void (*set_atom_dirty)(struct r600_common_context *ctx,
573 struct r600_atom *atom, bool dirty);
574 };
575
576 /* r600_buffer.c */
577 boolean r600_rings_is_buffer_referenced(struct r600_common_context *ctx,
578 struct pb_buffer *buf,
579 enum radeon_bo_usage usage);
580 void *r600_buffer_map_sync_with_rings(struct r600_common_context *ctx,
581 struct r600_resource *resource,
582 unsigned usage);
583 bool r600_init_resource(struct r600_common_screen *rscreen,
584 struct r600_resource *res,
585 uint64_t size, unsigned alignment);
586 struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
587 const struct pipe_resource *templ,
588 unsigned alignment);
589 struct pipe_resource * r600_aligned_buffer_create(struct pipe_screen *screen,
590 unsigned bind,
591 unsigned usage,
592 unsigned size,
593 unsigned alignment);
594 struct pipe_resource *
595 r600_buffer_from_user_memory(struct pipe_screen *screen,
596 const struct pipe_resource *templ,
597 void *user_memory);
598 void
599 r600_invalidate_resource(struct pipe_context *ctx,
600 struct pipe_resource *resource);
601
602 /* r600_common_pipe.c */
603 void r600_draw_rectangle(struct blitter_context *blitter,
604 int x1, int y1, int x2, int y2, float depth,
605 enum blitter_attrib_type type,
606 const union pipe_color_union *attrib);
607 bool r600_common_screen_init(struct r600_common_screen *rscreen,
608 struct radeon_winsys *ws);
609 void r600_destroy_common_screen(struct r600_common_screen *rscreen);
610 void r600_preflush_suspend_features(struct r600_common_context *ctx);
611 void r600_postflush_resume_features(struct r600_common_context *ctx);
612 bool r600_common_context_init(struct r600_common_context *rctx,
613 struct r600_common_screen *rscreen);
614 void r600_common_context_cleanup(struct r600_common_context *rctx);
615 void r600_context_add_resource_size(struct pipe_context *ctx, struct pipe_resource *r);
616 bool r600_can_dump_shader(struct r600_common_screen *rscreen,
617 unsigned processor);
618 void r600_screen_clear_buffer(struct r600_common_screen *rscreen, struct pipe_resource *dst,
619 uint64_t offset, uint64_t size, unsigned value,
620 enum r600_coherency coher);
621 struct pipe_resource *r600_resource_create_common(struct pipe_screen *screen,
622 const struct pipe_resource *templ);
623 const char *r600_get_llvm_processor_name(enum radeon_family family);
624 void r600_need_dma_space(struct r600_common_context *ctx, unsigned num_dw,
625 struct r600_resource *dst, struct r600_resource *src);
626 void r600_dma_emit_wait_idle(struct r600_common_context *rctx);
627
628 /* r600_gpu_load.c */
629 void r600_gpu_load_kill_thread(struct r600_common_screen *rscreen);
630 uint64_t r600_gpu_load_begin(struct r600_common_screen *rscreen);
631 unsigned r600_gpu_load_end(struct r600_common_screen *rscreen, uint64_t begin);
632
633 /* r600_perfcounters.c */
634 void r600_perfcounters_destroy(struct r600_common_screen *rscreen);
635
636 /* r600_query.c */
637 void r600_init_screen_query_functions(struct r600_common_screen *rscreen);
638 void r600_query_init(struct r600_common_context *rctx);
639 void r600_suspend_queries(struct r600_common_context *ctx);
640 void r600_resume_queries(struct r600_common_context *ctx);
641 void r600_query_init_backend_mask(struct r600_common_context *ctx);
642
643 /* r600_streamout.c */
644 void r600_streamout_buffers_dirty(struct r600_common_context *rctx);
645 void r600_set_streamout_targets(struct pipe_context *ctx,
646 unsigned num_targets,
647 struct pipe_stream_output_target **targets,
648 const unsigned *offset);
649 void r600_emit_streamout_end(struct r600_common_context *rctx);
650 void r600_update_prims_generated_query_state(struct r600_common_context *rctx,
651 unsigned type, int diff);
652 void r600_streamout_init(struct r600_common_context *rctx);
653
654 /* r600_test_dma.c */
655 void r600_test_dma(struct r600_common_screen *rscreen);
656
657 /* r600_texture.c */
658 bool r600_prepare_for_dma_blit(struct r600_common_context *rctx,
659 struct r600_texture *rdst,
660 unsigned dst_level, unsigned dstx,
661 unsigned dsty, unsigned dstz,
662 struct r600_texture *rsrc,
663 unsigned src_level,
664 const struct pipe_box *src_box);
665 void r600_texture_get_fmask_info(struct r600_common_screen *rscreen,
666 struct r600_texture *rtex,
667 unsigned nr_samples,
668 struct r600_fmask_info *out);
669 void r600_texture_get_cmask_info(struct r600_common_screen *rscreen,
670 struct r600_texture *rtex,
671 struct r600_cmask_info *out);
672 bool r600_init_flushed_depth_texture(struct pipe_context *ctx,
673 struct pipe_resource *texture,
674 struct r600_texture **staging);
675 void r600_print_texture_info(struct r600_texture *rtex, FILE *f);
676 struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
677 const struct pipe_resource *templ);
678 struct pipe_surface *r600_create_surface_custom(struct pipe_context *pipe,
679 struct pipe_resource *texture,
680 const struct pipe_surface *templ,
681 unsigned width, unsigned height);
682 unsigned r600_translate_colorswap(enum pipe_format format, bool do_endian_swap);
683 void vi_dcc_clear_level(struct r600_common_context *rctx,
684 struct r600_texture *rtex,
685 unsigned level, unsigned clear_value);
686 void evergreen_do_fast_color_clear(struct r600_common_context *rctx,
687 struct pipe_framebuffer_state *fb,
688 struct r600_atom *fb_state,
689 unsigned *buffers, unsigned *dirty_cbufs,
690 const union pipe_color_union *color);
691 bool r600_texture_disable_dcc(struct r600_common_screen *rscreen,
692 struct r600_texture *rtex);
693 void r600_init_screen_texture_functions(struct r600_common_screen *rscreen);
694 void r600_init_context_texture_functions(struct r600_common_context *rctx);
695
696 /* r600_viewport.c */
697 void evergreen_apply_scissor_bug_workaround(struct r600_common_context *rctx,
698 struct pipe_scissor_state *scissor);
699 void r600_set_scissor_enable(struct r600_common_context *rctx, bool enable);
700 void r600_update_vs_writes_viewport_index(struct r600_common_context *rctx,
701 struct tgsi_shader_info *info);
702 void r600_init_viewport_functions(struct r600_common_context *rctx);
703
704 /* cayman_msaa.c */
705 extern const uint32_t eg_sample_locs_2x[4];
706 extern const unsigned eg_max_dist_2x;
707 extern const uint32_t eg_sample_locs_4x[4];
708 extern const unsigned eg_max_dist_4x;
709 void cayman_get_sample_position(struct pipe_context *ctx, unsigned sample_count,
710 unsigned sample_index, float *out_value);
711 void cayman_init_msaa(struct pipe_context *ctx);
712 void cayman_emit_msaa_sample_locs(struct radeon_winsys_cs *cs, int nr_samples);
713 void cayman_emit_msaa_config(struct radeon_winsys_cs *cs, int nr_samples,
714 int ps_iter_samples, int overrast_samples);
715
716
717 /* Inline helpers. */
718
719 static inline struct r600_resource *r600_resource(struct pipe_resource *r)
720 {
721 return (struct r600_resource*)r;
722 }
723
724 static inline void
725 r600_resource_reference(struct r600_resource **ptr, struct r600_resource *res)
726 {
727 pipe_resource_reference((struct pipe_resource **)ptr,
728 (struct pipe_resource *)res);
729 }
730
731 static inline bool r600_get_strmout_en(struct r600_common_context *rctx)
732 {
733 return rctx->streamout.streamout_enabled ||
734 rctx->streamout.prims_gen_query_enabled;
735 }
736
737 #define SQ_TEX_XY_FILTER_POINT 0x00
738 #define SQ_TEX_XY_FILTER_BILINEAR 0x01
739 #define SQ_TEX_XY_FILTER_ANISO_POINT 0x02
740 #define SQ_TEX_XY_FILTER_ANISO_BILINEAR 0x03
741
742 static inline unsigned eg_tex_filter(unsigned filter, unsigned max_aniso)
743 {
744 if (filter == PIPE_TEX_FILTER_LINEAR)
745 return max_aniso > 1 ? SQ_TEX_XY_FILTER_ANISO_BILINEAR
746 : SQ_TEX_XY_FILTER_BILINEAR;
747 else
748 return max_aniso > 1 ? SQ_TEX_XY_FILTER_ANISO_POINT
749 : SQ_TEX_XY_FILTER_POINT;
750 }
751
752 static inline unsigned r600_tex_aniso_filter(unsigned filter)
753 {
754 if (filter < 2)
755 return 0;
756 if (filter < 4)
757 return 1;
758 if (filter < 8)
759 return 2;
760 if (filter < 16)
761 return 3;
762 return 4;
763 }
764
765 static inline unsigned r600_wavefront_size(enum radeon_family family)
766 {
767 switch (family) {
768 case CHIP_RV610:
769 case CHIP_RS780:
770 case CHIP_RV620:
771 case CHIP_RS880:
772 return 16;
773 case CHIP_RV630:
774 case CHIP_RV635:
775 case CHIP_RV730:
776 case CHIP_RV710:
777 case CHIP_PALM:
778 case CHIP_CEDAR:
779 return 32;
780 default:
781 return 64;
782 }
783 }
784
785 static inline enum radeon_bo_priority
786 r600_get_sampler_view_priority(struct r600_resource *res)
787 {
788 if (res->b.b.target == PIPE_BUFFER)
789 return RADEON_PRIO_SAMPLER_BUFFER;
790
791 if (res->b.b.nr_samples > 1)
792 return RADEON_PRIO_SAMPLER_TEXTURE_MSAA;
793
794 return RADEON_PRIO_SAMPLER_TEXTURE;
795 }
796
797 #define COMPUTE_DBG(rscreen, fmt, args...) \
798 do { \
799 if ((rscreen->b.debug_flags & DBG_COMPUTE)) fprintf(stderr, fmt, ##args); \
800 } while (0);
801
802 #define R600_ERR(fmt, args...) \
803 fprintf(stderr, "EE %s:%d %s - " fmt, __FILE__, __LINE__, __func__, ##args)
804
805 /* For MSAA sample positions. */
806 #define FILL_SREG(s0x, s0y, s1x, s1y, s2x, s2y, s3x, s3y) \
807 (((s0x) & 0xf) | (((unsigned)(s0y) & 0xf) << 4) | \
808 (((unsigned)(s1x) & 0xf) << 8) | (((unsigned)(s1y) & 0xf) << 12) | \
809 (((unsigned)(s2x) & 0xf) << 16) | (((unsigned)(s2y) & 0xf) << 20) | \
810 (((unsigned)(s3x) & 0xf) << 24) | (((unsigned)(s3y) & 0xf) << 28))
811
812 #endif