radeon: Teach radeon_elf_read() how to parse reloc information v3
[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/drm/radeon_winsys.h"
38
39 #include "util/u_blitter.h"
40 #include "util/u_double_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 R600_RESOURCE_FLAG_TRANSFER (PIPE_RESOURCE_FLAG_DRV_PRIV << 0)
47 #define R600_RESOURCE_FLAG_FLUSHED_DEPTH (PIPE_RESOURCE_FLAG_DRV_PRIV << 1)
48 #define R600_RESOURCE_FLAG_FORCE_TILING (PIPE_RESOURCE_FLAG_DRV_PRIV << 2)
49
50 #define R600_QUERY_DRAW_CALLS (PIPE_QUERY_DRIVER_SPECIFIC + 0)
51 #define R600_QUERY_REQUESTED_VRAM (PIPE_QUERY_DRIVER_SPECIFIC + 1)
52 #define R600_QUERY_REQUESTED_GTT (PIPE_QUERY_DRIVER_SPECIFIC + 2)
53 #define R600_QUERY_BUFFER_WAIT_TIME (PIPE_QUERY_DRIVER_SPECIFIC + 3)
54 #define R600_QUERY_NUM_CS_FLUSHES (PIPE_QUERY_DRIVER_SPECIFIC + 4)
55 #define R600_QUERY_NUM_BYTES_MOVED (PIPE_QUERY_DRIVER_SPECIFIC + 5)
56 #define R600_QUERY_VRAM_USAGE (PIPE_QUERY_DRIVER_SPECIFIC + 6)
57 #define R600_QUERY_GTT_USAGE (PIPE_QUERY_DRIVER_SPECIFIC + 7)
58
59 #define R600_CONTEXT_STREAMOUT_FLUSH (1u << 0)
60 #define R600_CONTEXT_PRIVATE_FLAG (1u << 1)
61
62 /* special primitive types */
63 #define R600_PRIM_RECTANGLE_LIST PIPE_PRIM_MAX
64
65 /* Debug flags. */
66 /* logging */
67 #define DBG_TEX (1 << 0)
68 #define DBG_TEXMIP (1 << 1)
69 #define DBG_COMPUTE (1 << 2)
70 #define DBG_VM (1 << 3)
71 #define DBG_TRACE_CS (1 << 4)
72 /* shader logging */
73 #define DBG_FS (1 << 5)
74 #define DBG_VS (1 << 6)
75 #define DBG_GS (1 << 7)
76 #define DBG_PS (1 << 8)
77 #define DBG_CS (1 << 9)
78 /* features */
79 #define DBG_NO_ASYNC_DMA (1 << 10)
80 #define DBG_NO_HYPERZ (1 << 11)
81 #define DBG_NO_DISCARD_RANGE (1 << 12)
82 #define DBG_NO_2D_TILING (1 << 13)
83 #define DBG_NO_TILING (1 << 14)
84 #define DBG_SWITCH_ON_EOP (1 << 15)
85 #define DBG_FORCE_DMA (1 << 16)
86 /* The maximum allowed bit is 20. */
87
88 #define R600_MAP_BUFFER_ALIGNMENT 64
89
90 struct r600_common_context;
91
92 struct radeon_shader_reloc {
93 char *name;
94 uint64_t offset;
95 };
96
97 struct radeon_shader_binary {
98 /** Shader code */
99 unsigned char *code;
100 unsigned code_size;
101
102 /** Config/Context register state that accompanies this shader.
103 * This is a stream of dword pairs. First dword contains the
104 * register address, the second dword contains the value.*/
105 unsigned char *config;
106 unsigned config_size;
107
108 /** The number of bytes of config information for each global symbol.
109 */
110 unsigned config_size_per_symbol;
111
112 /** Constant data accessed by the shader. This will be uploaded
113 * into a constant buffer. */
114 unsigned char *rodata;
115 unsigned rodata_size;
116
117 /** List of symbol offsets for the shader */
118 uint64_t *global_symbol_offsets;
119 unsigned global_symbol_count;
120
121 struct radeon_shader_reloc *relocs;
122 unsigned reloc_count;
123
124 /** Set to 1 if the disassembly for this binary has been dumped to
125 * stderr. */
126 int disassembled;
127 };
128
129 struct r600_resource {
130 struct u_resource b;
131
132 /* Winsys objects. */
133 struct pb_buffer *buf;
134 struct radeon_winsys_cs_handle *cs_buf;
135 uint64_t gpu_address;
136
137 /* Resource state. */
138 enum radeon_bo_domain domains;
139
140 /* The buffer range which is initialized (with a write transfer,
141 * streamout, DMA, or as a random access target). The rest of
142 * the buffer is considered invalid and can be mapped unsynchronized.
143 *
144 * This allows unsychronized mapping of a buffer range which hasn't
145 * been used yet. It's for applications which forget to use
146 * the unsynchronized map flag and expect the driver to figure it out.
147 */
148 struct util_range valid_buffer_range;
149
150 /* For buffers only. This indicates that a write operation has been
151 * performed by TC L2, but the cache hasn't been flushed.
152 * Any hw block which doesn't use or bypasses TC L2 should check this
153 * flag and flush the cache before using the buffer.
154 *
155 * For example, TC L2 must be flushed if a buffer which has been
156 * modified by a shader store instruction is about to be used as
157 * an index buffer. The reason is that VGT DMA index fetching doesn't
158 * use TC L2.
159 */
160 bool TC_L2_dirty;
161 };
162
163 struct r600_transfer {
164 struct pipe_transfer transfer;
165 struct r600_resource *staging;
166 unsigned offset;
167 };
168
169 struct r600_fmask_info {
170 unsigned offset;
171 unsigned size;
172 unsigned alignment;
173 unsigned pitch;
174 unsigned bank_height;
175 unsigned slice_tile_max;
176 unsigned tile_mode_index;
177 };
178
179 struct r600_cmask_info {
180 unsigned offset;
181 unsigned size;
182 unsigned alignment;
183 unsigned slice_tile_max;
184 unsigned base_address_reg;
185 };
186
187 struct r600_texture {
188 struct r600_resource resource;
189
190 unsigned size;
191 unsigned pitch_override;
192 bool is_depth;
193 unsigned dirty_level_mask; /* each bit says if that mipmap is compressed */
194 struct r600_texture *flushed_depth_texture;
195 boolean is_flushing_texture;
196 struct radeon_surface surface;
197
198 /* Colorbuffer compression and fast clear. */
199 struct r600_fmask_info fmask;
200 struct r600_cmask_info cmask;
201 struct r600_resource *cmask_buffer;
202 unsigned cb_color_info; /* fast clear enable bit */
203 unsigned color_clear_value[2];
204
205 /* Depth buffer compression and fast clear. */
206 struct r600_resource *htile_buffer;
207 bool depth_cleared; /* if it was cleared at least once */
208 float depth_clear_value;
209
210 bool non_disp_tiling; /* R600-Cayman only */
211 unsigned mipmap_shift;
212 };
213
214 struct r600_surface {
215 struct pipe_surface base;
216
217 bool color_initialized;
218 bool depth_initialized;
219
220 /* Misc. color flags. */
221 bool alphatest_bypass;
222 bool export_16bpc;
223
224 /* Color registers. */
225 unsigned cb_color_info;
226 unsigned cb_color_base;
227 unsigned cb_color_view;
228 unsigned cb_color_size; /* R600 only */
229 unsigned cb_color_dim; /* EG only */
230 unsigned cb_color_pitch; /* EG and later */
231 unsigned cb_color_slice; /* EG and later */
232 unsigned cb_color_attrib; /* EG and later */
233 unsigned cb_color_fmask; /* CB_COLORn_FMASK (EG and later) or CB_COLORn_FRAG (r600) */
234 unsigned cb_color_fmask_slice; /* EG and later */
235 unsigned cb_color_cmask; /* CB_COLORn_TILE (r600 only) */
236 unsigned cb_color_mask; /* R600 only */
237 struct r600_resource *cb_buffer_fmask; /* Used for FMASK relocations. R600 only */
238 struct r600_resource *cb_buffer_cmask; /* Used for CMASK relocations. R600 only */
239
240 /* DB registers. */
241 unsigned db_depth_info; /* R600 only, then SI and later */
242 unsigned db_z_info; /* EG and later */
243 unsigned db_depth_base; /* DB_Z_READ/WRITE_BASE (EG and later) or DB_DEPTH_BASE (r600) */
244 unsigned db_depth_view;
245 unsigned db_depth_size;
246 unsigned db_depth_slice; /* EG and later */
247 unsigned db_stencil_base; /* EG and later */
248 unsigned db_stencil_info; /* EG and later */
249 unsigned db_prefetch_limit; /* R600 only */
250 unsigned db_htile_surface;
251 unsigned db_htile_data_base;
252 unsigned db_preload_control; /* EG and later */
253 unsigned pa_su_poly_offset_db_fmt_cntl;
254 };
255
256 struct r600_tiling_info {
257 unsigned num_channels;
258 unsigned num_banks;
259 unsigned group_bytes;
260 };
261
262 struct r600_common_screen {
263 struct pipe_screen b;
264 struct radeon_winsys *ws;
265 enum radeon_family family;
266 enum chip_class chip_class;
267 struct radeon_info info;
268 struct r600_tiling_info tiling_info;
269 unsigned debug_flags;
270 bool has_cp_dma;
271 bool has_streamout;
272
273 /* Auxiliary context. Mainly used to initialize resources.
274 * It must be locked prior to using and flushed before unlocking. */
275 struct pipe_context *aux_context;
276 pipe_mutex aux_context_lock;
277
278 struct r600_resource *trace_bo;
279 uint32_t *trace_ptr;
280 unsigned cs_count;
281 };
282
283 /* This encapsulates a state or an operation which can emitted into the GPU
284 * command stream. */
285 struct r600_atom {
286 void (*emit)(struct r600_common_context *ctx, struct r600_atom *state);
287 unsigned num_dw;
288 bool dirty;
289 };
290
291 struct r600_so_target {
292 struct pipe_stream_output_target b;
293
294 /* The buffer where BUFFER_FILLED_SIZE is stored. */
295 struct r600_resource *buf_filled_size;
296 unsigned buf_filled_size_offset;
297
298 unsigned stride_in_dw;
299 };
300
301 struct r600_streamout {
302 struct r600_atom begin_atom;
303 bool begin_emitted;
304 unsigned num_dw_for_end;
305
306 unsigned enabled_mask;
307 unsigned num_targets;
308 struct r600_so_target *targets[PIPE_MAX_SO_BUFFERS];
309
310 unsigned append_bitmask;
311 bool suspended;
312
313 /* External state which comes from the vertex shader,
314 * it must be set explicitly when binding a shader. */
315 unsigned *stride_in_dw;
316
317 /* The state of VGT_STRMOUT_(CONFIG|EN). */
318 struct r600_atom enable_atom;
319 bool streamout_enabled;
320 bool prims_gen_query_enabled;
321 int num_prims_gen_queries;
322 };
323
324 struct r600_ring {
325 struct radeon_winsys_cs *cs;
326 bool flushing;
327 void (*flush)(void *ctx, unsigned flags,
328 struct pipe_fence_handle **fence);
329 };
330
331 struct r600_rings {
332 struct r600_ring gfx;
333 struct r600_ring dma;
334 };
335
336 struct r600_common_context {
337 struct pipe_context b; /* base class */
338
339 struct r600_common_screen *screen;
340 struct radeon_winsys *ws;
341 enum radeon_family family;
342 enum chip_class chip_class;
343 struct r600_rings rings;
344 unsigned initial_gfx_cs_size;
345
346 struct u_upload_mgr *uploader;
347 struct u_suballocator *allocator_so_filled_size;
348 struct util_slab_mempool pool_transfers;
349
350 /* Current unaccounted memory usage. */
351 uint64_t vram;
352 uint64_t gtt;
353
354 /* States. */
355 struct r600_streamout streamout;
356
357 /* Additional context states. */
358 unsigned flags; /* flush flags */
359
360 /* Queries. */
361 /* The list of active queries. Only one query of each type can be active. */
362 int num_occlusion_queries;
363 int num_pipelinestat_queries;
364 /* Keep track of non-timer queries, because they should be suspended
365 * during context flushing.
366 * The timer queries (TIME_ELAPSED) shouldn't be suspended. */
367 struct list_head active_nontimer_queries;
368 unsigned num_cs_dw_nontimer_queries_suspend;
369 /* If queries have been suspended. */
370 bool nontimer_queries_suspended;
371 /* Additional hardware info. */
372 unsigned backend_mask;
373 unsigned max_db; /* for OQ */
374 /* Misc stats. */
375 unsigned num_draw_calls;
376
377 /* Render condition. */
378 struct pipe_query *current_render_cond;
379 unsigned current_render_cond_mode;
380 boolean current_render_cond_cond;
381 boolean predicate_drawing;
382 /* For context flushing. */
383 struct pipe_query *saved_render_cond;
384 boolean saved_render_cond_cond;
385 unsigned saved_render_cond_mode;
386
387 /* MSAA sample locations.
388 * The first index is the sample index.
389 * The second index is the coordinate: X, Y. */
390 float sample_locations_1x[1][2];
391 float sample_locations_2x[2][2];
392 float sample_locations_4x[4][2];
393 float sample_locations_8x[8][2];
394 float sample_locations_16x[16][2];
395
396 /* The list of all texture buffer objects in this context.
397 * This list is walked when a buffer is invalidated/reallocated and
398 * the GPU addresses are updated. */
399 struct list_head texture_buffers;
400
401 /* Copy one resource to another using async DMA. */
402 void (*dma_copy)(struct pipe_context *ctx,
403 struct pipe_resource *dst,
404 unsigned dst_level,
405 unsigned dst_x, unsigned dst_y, unsigned dst_z,
406 struct pipe_resource *src,
407 unsigned src_level,
408 const struct pipe_box *src_box);
409
410 void (*clear_buffer)(struct pipe_context *ctx, struct pipe_resource *dst,
411 unsigned offset, unsigned size, unsigned value,
412 bool is_framebuffer);
413
414 void (*blit_decompress_depth)(struct pipe_context *ctx,
415 struct r600_texture *texture,
416 struct r600_texture *staging,
417 unsigned first_level, unsigned last_level,
418 unsigned first_layer, unsigned last_layer,
419 unsigned first_sample, unsigned last_sample);
420
421 /* Reallocate the buffer and update all resource bindings where
422 * the buffer is bound, including all resource descriptors. */
423 void (*invalidate_buffer)(struct pipe_context *ctx, struct pipe_resource *buf);
424
425 /* Enable or disable occlusion queries. */
426 void (*set_occlusion_query_state)(struct pipe_context *ctx, bool enable);
427
428 /* This ensures there is enough space in the command stream. */
429 void (*need_gfx_cs_space)(struct pipe_context *ctx, unsigned num_dw,
430 bool include_draw_vbo);
431 };
432
433 /* r600_buffer.c */
434 boolean r600_rings_is_buffer_referenced(struct r600_common_context *ctx,
435 struct radeon_winsys_cs_handle *buf,
436 enum radeon_bo_usage usage);
437 void *r600_buffer_map_sync_with_rings(struct r600_common_context *ctx,
438 struct r600_resource *resource,
439 unsigned usage);
440 bool r600_init_resource(struct r600_common_screen *rscreen,
441 struct r600_resource *res,
442 unsigned size, unsigned alignment,
443 bool use_reusable_pool);
444 struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
445 const struct pipe_resource *templ,
446 unsigned alignment);
447
448 /* r600_common_pipe.c */
449 void r600_draw_rectangle(struct blitter_context *blitter,
450 int x1, int y1, int x2, int y2, float depth,
451 enum blitter_attrib_type type,
452 const union pipe_color_union *attrib);
453 bool r600_common_screen_init(struct r600_common_screen *rscreen,
454 struct radeon_winsys *ws);
455 void r600_destroy_common_screen(struct r600_common_screen *rscreen);
456 void r600_preflush_suspend_features(struct r600_common_context *ctx);
457 void r600_postflush_resume_features(struct r600_common_context *ctx);
458 bool r600_common_context_init(struct r600_common_context *rctx,
459 struct r600_common_screen *rscreen);
460 void r600_common_context_cleanup(struct r600_common_context *rctx);
461 void r600_context_add_resource_size(struct pipe_context *ctx, struct pipe_resource *r);
462 bool r600_can_dump_shader(struct r600_common_screen *rscreen,
463 const struct tgsi_token *tokens);
464 void r600_screen_clear_buffer(struct r600_common_screen *rscreen, struct pipe_resource *dst,
465 unsigned offset, unsigned size, unsigned value,
466 bool is_framebuffer);
467 struct pipe_resource *r600_resource_create_common(struct pipe_screen *screen,
468 const struct pipe_resource *templ);
469 const char *r600_get_llvm_processor_name(enum radeon_family family);
470 void r600_need_dma_space(struct r600_common_context *ctx, unsigned num_dw);
471
472 /* r600_query.c */
473 void r600_query_init(struct r600_common_context *rctx);
474 void r600_suspend_nontimer_queries(struct r600_common_context *ctx);
475 void r600_resume_nontimer_queries(struct r600_common_context *ctx);
476 void r600_query_init_backend_mask(struct r600_common_context *ctx);
477
478 /* r600_streamout.c */
479 void r600_streamout_buffers_dirty(struct r600_common_context *rctx);
480 void r600_set_streamout_targets(struct pipe_context *ctx,
481 unsigned num_targets,
482 struct pipe_stream_output_target **targets,
483 const unsigned *offset);
484 void r600_emit_streamout_end(struct r600_common_context *rctx);
485 void r600_update_prims_generated_query_state(struct r600_common_context *rctx,
486 unsigned type, int diff);
487 void r600_streamout_init(struct r600_common_context *rctx);
488
489 /* r600_texture.c */
490 void r600_texture_get_fmask_info(struct r600_common_screen *rscreen,
491 struct r600_texture *rtex,
492 unsigned nr_samples,
493 struct r600_fmask_info *out);
494 void r600_texture_get_cmask_info(struct r600_common_screen *rscreen,
495 struct r600_texture *rtex,
496 struct r600_cmask_info *out);
497 bool r600_init_flushed_depth_texture(struct pipe_context *ctx,
498 struct pipe_resource *texture,
499 struct r600_texture **staging);
500 struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
501 const struct pipe_resource *templ);
502 struct pipe_surface *r600_create_surface_custom(struct pipe_context *pipe,
503 struct pipe_resource *texture,
504 const struct pipe_surface *templ,
505 unsigned width, unsigned height);
506 unsigned r600_translate_colorswap(enum pipe_format format);
507 void evergreen_do_fast_color_clear(struct r600_common_context *rctx,
508 struct pipe_framebuffer_state *fb,
509 struct r600_atom *fb_state,
510 unsigned *buffers,
511 const union pipe_color_union *color);
512 void r600_init_screen_texture_functions(struct r600_common_screen *rscreen);
513 void r600_init_context_texture_functions(struct r600_common_context *rctx);
514
515 /* cayman_msaa.c */
516 extern const uint32_t eg_sample_locs_2x[4];
517 extern const unsigned eg_max_dist_2x;
518 extern const uint32_t eg_sample_locs_4x[4];
519 extern const unsigned eg_max_dist_4x;
520 void cayman_get_sample_position(struct pipe_context *ctx, unsigned sample_count,
521 unsigned sample_index, float *out_value);
522 void cayman_init_msaa(struct pipe_context *ctx);
523 void cayman_emit_msaa_sample_locs(struct radeon_winsys_cs *cs, int nr_samples);
524 void cayman_emit_msaa_config(struct radeon_winsys_cs *cs, int nr_samples,
525 int ps_iter_samples);
526
527
528 /* Inline helpers. */
529
530 static INLINE struct r600_resource *r600_resource(struct pipe_resource *r)
531 {
532 return (struct r600_resource*)r;
533 }
534
535 static INLINE void
536 r600_resource_reference(struct r600_resource **ptr, struct r600_resource *res)
537 {
538 pipe_resource_reference((struct pipe_resource **)ptr,
539 (struct pipe_resource *)res);
540 }
541
542 static inline unsigned r600_tex_aniso_filter(unsigned filter)
543 {
544 if (filter <= 1) return 0;
545 if (filter <= 2) return 1;
546 if (filter <= 4) return 2;
547 if (filter <= 8) return 3;
548 /* else */ return 4;
549 }
550
551 #define COMPUTE_DBG(rscreen, fmt, args...) \
552 do { \
553 if ((rscreen->b.debug_flags & DBG_COMPUTE)) fprintf(stderr, fmt, ##args); \
554 } while (0);
555
556 #define R600_ERR(fmt, args...) \
557 fprintf(stderr, "EE %s:%d %s - "fmt, __FILE__, __LINE__, __func__, ##args)
558
559 /* For MSAA sample positions. */
560 #define FILL_SREG(s0x, s0y, s1x, s1y, s2x, s2y, s3x, s3y) \
561 (((s0x) & 0xf) | (((s0y) & 0xf) << 4) | \
562 (((s1x) & 0xf) << 8) | (((s1y) & 0xf) << 12) | \
563 (((s2x) & 0xf) << 16) | (((s2y) & 0xf) << 20) | \
564 (((s3x) & 0xf) << 24) | (((s3y) & 0xf) << 28))
565
566 #endif