[intel] Add more cliprect modes to cover other meanings for batch emits.
[mesa.git] / src / mesa / drivers / dri / i965 / brw_context.h
1 /*
2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
5
6 Permission is hereby granted, free of charge, to any person obtaining
7 a copy of this software and associated documentation files (the
8 "Software"), to deal in the Software without restriction, including
9 without limitation the rights to use, copy, modify, merge, publish,
10 distribute, sublicense, and/or sell copies of the Software, and to
11 permit persons to whom the Software is furnished to do so, subject to
12 the following conditions:
13
14 The above copyright notice and this permission notice (including the
15 next paragraph) shall be included in all copies or substantial
16 portions of the Software.
17
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25
26 **********************************************************************/
27 /*
28 * Authors:
29 * Keith Whitwell <keith@tungstengraphics.com>
30 */
31
32
33 #ifndef BRWCONTEXT_INC
34 #define BRWCONTEXT_INC
35
36 #include "intel_context.h"
37 #include "brw_structs.h"
38 #include "imports.h"
39
40
41 /* Glossary:
42 *
43 * URB - uniform resource buffer. A mid-sized buffer which is
44 * partitioned between the fixed function units and used for passing
45 * values (vertices, primitives, constants) between them.
46 *
47 * CURBE - constant URB entry. An urb region (entry) used to hold
48 * constant values which the fixed function units can be instructed to
49 * preload into the GRF when spawining a thread.
50 *
51 * VUE - vertex URB entry. An urb entry holding a vertex and usually
52 * a vertex header. The header contains control information and
53 * things like primitive type, Begin/end flags and clip codes.
54 *
55 * PUE - primitive URB entry. An urb entry produced by the setup (SF)
56 * unit holding rasterization and interpolation parameters.
57 *
58 * GRF - general register file. One of several register files
59 * addressable by programmed threads. The inputs (r0, payload, curbe,
60 * urb) of the thread are preloaded to this area before the thread is
61 * spawned. The registers are individually 8 dwords wide and suitable
62 * for general usage. Registers holding thread input values are not
63 * special and may be overwritten.
64 *
65 * MRF - message register file. Threads communicate (and terminate)
66 * by sending messages. Message parameters are placed in contigous
67 * MRF registers. All program output is via these messages. URB
68 * entries are populated by sending a message to the shared URB
69 * function containing the new data, together with a control word,
70 * often an unmodified copy of R0.
71 *
72 * R0 - GRF register 0. Typically holds control information used when
73 * sending messages to other threads.
74 *
75 * EU or GEN4 EU: The name of the programmable subsystem of the
76 * i965 hardware. Threads are executed by the EU, the registers
77 * described above are part of the EU architecture.
78 *
79 * Fixed function units:
80 *
81 * CS - Command streamer. Notional first unit, little software
82 * interaction. Holds the URB entries used for constant data, ie the
83 * CURBEs.
84 *
85 * VF/VS - Vertex Fetch / Vertex Shader. The fixed function part of
86 * this unit is responsible for pulling vertices out of vertex buffers
87 * in vram and injecting them into the processing pipe as VUEs. If
88 * enabled, it first passes them to a VS thread which is a good place
89 * for the driver to implement any active vertex shader.
90 *
91 * GS - Geometry Shader. This corresponds to a new DX10 concept. If
92 * enabled, incoming strips etc are passed to GS threads in individual
93 * line/triangle/point units. The GS thread may perform arbitary
94 * computation and emit whatever primtives with whatever vertices it
95 * chooses. This makes GS an excellent place to implement GL's
96 * unfilled polygon modes, though of course it is capable of much
97 * more. Additionally, GS is used to translate away primitives not
98 * handled by latter units, including Quads and Lineloops.
99 *
100 * CS - Clipper. Mesa's clipping algorithms are imported to run on
101 * this unit. The fixed function part performs cliptesting against
102 * the 6 fixed clipplanes and makes descisions on whether or not the
103 * incoming primitive needs to be passed to a thread for clipping.
104 * User clip planes are handled via cooperation with the VS thread.
105 *
106 * SF - Strips Fans or Setup: Triangles are prepared for
107 * rasterization. Interpolation coefficients are calculated.
108 * Flatshading and two-side lighting usually performed here.
109 *
110 * WM - Windower. Interpolation of vertex attributes performed here.
111 * Fragment shader implemented here. SIMD aspects of EU taken full
112 * advantage of, as pixels are processed in blocks of 16.
113 *
114 * CC - Color Calculator. No EU threads associated with this unit.
115 * Handles blending and (presumably) depth and stencil testing.
116 */
117
118 #define BRW_FALLBACK_TEXTURE 0x1
119 #define BRW_MAX_CURBE (32*16)
120
121 struct brw_context;
122
123 #define BRW_NEW_URB_FENCE 0x1
124 #define BRW_NEW_FRAGMENT_PROGRAM 0x2
125 #define BRW_NEW_VERTEX_PROGRAM 0x4
126 #define BRW_NEW_INPUT_DIMENSIONS 0x8
127 #define BRW_NEW_CURBE_OFFSETS 0x10
128 #define BRW_NEW_REDUCED_PRIMITIVE 0x20
129 #define BRW_NEW_PRIMITIVE 0x40
130 #define BRW_NEW_CONTEXT 0x80
131 #define BRW_NEW_WM_INPUT_DIMENSIONS 0x100
132 #define BRW_NEW_INPUT_VARYING 0x200
133 #define BRW_NEW_TNL_PROGRAM 0x400
134 #define BRW_NEW_PSP 0x800
135 #define BRW_NEW_METAOPS 0x1000
136 #define BRW_NEW_FENCE 0x2000
137 #define BRW_NEW_LOCK 0x4000
138 /**
139 * Used for any batch entry with a relocated pointer that will be used
140 * by any 3D rendering.
141 */
142 #define BRW_NEW_BATCH 0x8000
143 /** brw->depth_region updated */
144 #define BRW_NEW_DEPTH_BUFFER 0x10000
145
146 struct brw_state_flags {
147 /** State update flags signalled by mesa internals */
148 GLuint mesa;
149 /** State update flags signalled by brw_state_cache.c searches */
150 GLuint cache;
151 /**
152 * State update flags signalled as the result of brw_tracked_state updates
153 */
154 GLuint brw;
155 };
156
157 struct brw_vertex_program {
158 struct gl_vertex_program program;
159 GLuint id;
160 GLuint param_state; /* flags indicating state tracked by params */
161 };
162
163
164
165 struct brw_fragment_program {
166 struct gl_fragment_program program;
167 GLuint id;
168 GLuint param_state; /* flags indicating state tracked by params */
169 };
170
171
172
173
174 /* Data about a particular attempt to compile a program. Note that
175 * there can be many of these, each in a different GL state
176 * corresponding to a different brw_wm_prog_key struct, with different
177 * compiled programs:
178 */
179 struct brw_wm_prog_data {
180 GLuint curb_read_length;
181 GLuint urb_read_length;
182
183 GLuint first_curbe_grf;
184 GLuint total_grf;
185 GLuint total_scratch;
186
187 GLuint nr_params;
188 GLboolean error;
189
190 /* Pointer to tracked values (only valid once
191 * _mesa_load_state_parameters has been called at runtime).
192 */
193 const GLfloat *param[BRW_MAX_CURBE];
194 };
195
196 struct brw_sf_prog_data {
197 GLuint urb_read_length;
198 GLuint total_grf;
199
200 /* Each vertex may have upto 12 attributes, 4 components each,
201 * except WPOS which requires only 2. (11*4 + 2) == 44 ==> 11
202 * rows.
203 *
204 * Actually we use 4 for each, so call it 12 rows.
205 */
206 GLuint urb_entry_size;
207 };
208
209 struct brw_clip_prog_data {
210 GLuint curb_read_length; /* user planes? */
211 GLuint clip_mode;
212 GLuint urb_read_length;
213 GLuint total_grf;
214 };
215
216 struct brw_gs_prog_data {
217 GLuint urb_read_length;
218 GLuint total_grf;
219 };
220
221 struct brw_vs_prog_data {
222 GLuint curb_read_length;
223 GLuint urb_read_length;
224 GLuint total_grf;
225 GLuint outputs_written;
226
227 GLuint inputs_read;
228
229 /* Used for calculating urb partitions:
230 */
231 GLuint urb_entry_size;
232 };
233
234
235 /* Size == 0 if output either not written, or always [0,0,0,1]
236 */
237 struct brw_vs_ouput_sizes {
238 GLubyte output_size[VERT_RESULT_MAX];
239 };
240
241
242 #define BRW_MAX_TEX_UNIT 8
243 #define BRW_WM_MAX_SURF BRW_MAX_TEX_UNIT + 1
244
245 enum brw_cache_id {
246 BRW_CC_VP,
247 BRW_CC_UNIT,
248 BRW_WM_PROG,
249 BRW_SAMPLER_DEFAULT_COLOR,
250 BRW_SAMPLER,
251 BRW_WM_UNIT,
252 BRW_SF_PROG,
253 BRW_SF_VP,
254 BRW_SF_UNIT,
255 BRW_VS_UNIT,
256 BRW_VS_PROG,
257 BRW_GS_UNIT,
258 BRW_GS_PROG,
259 BRW_CLIP_VP,
260 BRW_CLIP_UNIT,
261 BRW_CLIP_PROG,
262 BRW_SS_SURFACE,
263 BRW_SS_SURF_BIND,
264
265 BRW_MAX_CACHE
266 };
267
268 struct brw_cache_item {
269 /**
270 * Effectively part of the key, cache_id identifies what kind of state
271 * buffer is involved, and also which brw->state.dirty.cache flag should
272 * be set when this cache item is chosen.
273 */
274 enum brw_cache_id cache_id;
275 /** 32-bit hash of the key data */
276 GLuint hash;
277 GLuint key_size; /* for variable-sized keys */
278 const void *key;
279 dri_bo **reloc_bufs;
280 GLuint nr_reloc_bufs;
281
282 dri_bo *bo;
283 GLuint data_size;
284
285 struct brw_cache_item *next;
286 };
287
288
289
290 struct brw_cache {
291 struct brw_context *brw;
292
293 struct brw_cache_item **items;
294 GLuint size, n_items;
295
296 GLuint key_size[BRW_MAX_CACHE]; /* for fixed-size keys */
297 GLuint aux_size[BRW_MAX_CACHE];
298 char *name[BRW_MAX_CACHE];
299
300 /* Record of the last BOs chosen for each cache_id. Used to set
301 * brw->state.dirty.cache when a new cache item is chosen.
302 */
303 dri_bo *last_bo[BRW_MAX_CACHE];
304 };
305
306
307
308 struct brw_state_pointers {
309 struct gl_colorbuffer_attrib *Color;
310 struct gl_depthbuffer_attrib *Depth;
311 struct gl_fog_attrib *Fog;
312 struct gl_hint_attrib *Hint;
313 struct gl_light_attrib *Light;
314 struct gl_line_attrib *Line;
315 struct gl_point_attrib *Point;
316 struct gl_polygon_attrib *Polygon;
317 GLuint *PolygonStipple;
318 struct gl_scissor_attrib *Scissor;
319 struct gl_stencil_attrib *Stencil;
320 struct gl_texture_attrib *Texture;
321 struct gl_transform_attrib *Transform;
322 struct gl_viewport_attrib *Viewport;
323 struct gl_vertex_program_state *VertexProgram;
324 struct gl_fragment_program_state *FragmentProgram;
325 };
326
327 /* Considered adding a member to this struct to document which flags
328 * an update might raise so that ordering of the state atoms can be
329 * checked or derived at runtime. Dropped the idea in favor of having
330 * a debug mode where the state is monitored for flags which are
331 * raised that have already been tested against.
332 */
333 struct brw_tracked_state {
334 struct brw_state_flags dirty;
335 void (*update)( struct brw_context *brw );
336 };
337
338 /* Flags for brw->state.cache.
339 */
340 #define CACHE_NEW_CC_VP (1<<BRW_CC_VP)
341 #define CACHE_NEW_CC_UNIT (1<<BRW_CC_UNIT)
342 #define CACHE_NEW_WM_PROG (1<<BRW_WM_PROG)
343 #define CACHE_NEW_SAMPLER_DEFAULT_COLOR (1<<BRW_SAMPLER_DEFAULT_COLOR)
344 #define CACHE_NEW_SAMPLER (1<<BRW_SAMPLER)
345 #define CACHE_NEW_WM_UNIT (1<<BRW_WM_UNIT)
346 #define CACHE_NEW_SF_PROG (1<<BRW_SF_PROG)
347 #define CACHE_NEW_SF_VP (1<<BRW_SF_VP)
348 #define CACHE_NEW_SF_UNIT (1<<BRW_SF_UNIT)
349 #define CACHE_NEW_VS_UNIT (1<<BRW_VS_UNIT)
350 #define CACHE_NEW_VS_PROG (1<<BRW_VS_PROG)
351 #define CACHE_NEW_GS_UNIT (1<<BRW_GS_UNIT)
352 #define CACHE_NEW_GS_PROG (1<<BRW_GS_PROG)
353 #define CACHE_NEW_CLIP_VP (1<<BRW_CLIP_VP)
354 #define CACHE_NEW_CLIP_UNIT (1<<BRW_CLIP_UNIT)
355 #define CACHE_NEW_CLIP_PROG (1<<BRW_CLIP_PROG)
356 #define CACHE_NEW_SURFACE (1<<BRW_SS_SURFACE)
357 #define CACHE_NEW_SURF_BIND (1<<BRW_SS_SURF_BIND)
358
359 struct brw_cached_batch_item {
360 struct header *header;
361 GLuint sz;
362 struct brw_cached_batch_item *next;
363 };
364
365
366
367 /* Protect against a future where VERT_ATTRIB_MAX > 32. Wouldn't life
368 * be easier if C allowed arrays of packed elements?
369 */
370 #define ATTRIB_BIT_DWORDS ((VERT_ATTRIB_MAX+31)/32)
371
372 struct brw_vertex_element {
373 const struct gl_client_array *glarray;
374
375 struct brw_vertex_element_state *vep;
376
377 GLuint index;
378 GLuint element_size;
379 GLuint count;
380 GLuint vbo_rebase_offset;
381 };
382
383
384
385 struct brw_vertex_info {
386 GLuint varying; /* varying:1[VERT_ATTRIB_MAX] */
387 GLuint sizes[ATTRIB_BIT_DWORDS * 2]; /* sizes:2[VERT_ATTRIB_MAX] */
388 };
389
390
391
392
393 /* Cache for TNL programs.
394 */
395 struct brw_tnl_cache_item {
396 GLuint hash;
397 void *key;
398 void *data;
399 struct brw_tnl_cache_item *next;
400 };
401
402 struct brw_tnl_cache {
403 struct brw_tnl_cache_item **items;
404 GLuint size, n_items;
405 };
406
407
408
409 struct brw_context
410 {
411 struct intel_context intel;
412 GLuint primitive;
413
414 GLboolean emit_state_always;
415 GLboolean wrap;
416 GLboolean tmp_fallback;
417 GLboolean no_batch_wrap;
418
419 struct {
420 struct brw_state_flags dirty;
421 struct brw_tracked_state **atoms;
422 GLuint nr_atoms;
423
424
425 struct intel_region *draw_region;
426 struct intel_region *depth_region;
427 } state;
428
429 struct brw_state_pointers attribs;
430 struct brw_cache cache;
431 struct brw_cached_batch_item *cached_batch_items;
432
433 struct {
434
435 /* Arrays with buffer objects to copy non-bufferobj arrays into
436 * for upload:
437 */
438 struct gl_client_array vbo_array[VERT_ATTRIB_MAX];
439
440 struct brw_vertex_element inputs[VERT_ATTRIB_MAX];
441
442 #define BRW_NR_UPLOAD_BUFS 17
443 #define BRW_UPLOAD_INIT_SIZE (128*1024)
444
445 struct {
446 struct gl_buffer_object *vbo[BRW_NR_UPLOAD_BUFS];
447 GLuint buf;
448 GLuint offset;
449 GLuint size;
450 GLuint wrap;
451 } upload;
452
453 /* Summary of size and varying of active arrays, so we can check
454 * for changes to this state:
455 */
456 struct brw_vertex_info info;
457 } vb;
458
459 struct {
460 /* Will be allocated on demand if needed.
461 */
462 struct brw_state_pointers attribs;
463 struct gl_vertex_program *vp;
464 struct gl_fragment_program *fp, *fp_tex;
465
466 struct gl_buffer_object *vbo;
467
468 struct intel_region *saved_draw_region;
469 struct intel_region *saved_depth_region;
470
471 GLuint restore_draw_buffers[MAX_DRAW_BUFFERS];
472 GLuint restore_num_draw_buffers;
473
474 struct gl_fragment_program *restore_fp;
475
476 GLboolean active;
477 } metaops;
478
479 /* Track fixed function t&l in a vertex program:
480 */
481 struct gl_vertex_program *tnl_program;
482 struct brw_tnl_cache tnl_program_cache;
483
484 /* Active vertex program:
485 */
486 const struct gl_vertex_program *vertex_program;
487 const struct gl_fragment_program *fragment_program;
488
489
490 /* For populating the gtt:
491 */
492 GLuint next_free_page;
493
494
495 /* BRW_NEW_URB_ALLOCATIONS:
496 */
497 struct {
498 GLuint vsize; /* vertex size plus header in urb registers */
499 GLuint csize; /* constant buffer size in urb registers */
500 GLuint sfsize; /* setup data size in urb registers */
501
502 GLboolean constrained;
503
504 GLuint nr_vs_entries;
505 GLuint nr_gs_entries;
506 GLuint nr_clip_entries;
507 GLuint nr_sf_entries;
508 GLuint nr_cs_entries;
509
510 /* GLuint vs_size; */
511 /* GLuint gs_size; */
512 /* GLuint clip_size; */
513 /* GLuint sf_size; */
514 /* GLuint cs_size; */
515
516 GLuint vs_start;
517 GLuint gs_start;
518 GLuint clip_start;
519 GLuint sf_start;
520 GLuint cs_start;
521 } urb;
522
523
524 /* BRW_NEW_CURBE_OFFSETS:
525 */
526 struct {
527 GLuint wm_start;
528 GLuint wm_size;
529 GLuint clip_start;
530 GLuint clip_size;
531 GLuint vs_start;
532 GLuint vs_size;
533 GLuint total_size;
534
535 /* Dynamic tracker which changes to reflect the state referenced
536 * by active fp and vp program parameters:
537 */
538 struct brw_tracked_state tracked_state;
539
540 dri_bo *curbe_bo;
541
542 GLfloat *last_buf;
543 GLuint last_bufsz;
544 } curbe;
545
546 struct {
547 struct brw_vs_prog_data *prog_data;
548
549 dri_bo *prog_bo;
550 dri_bo *state_bo;
551 } vs;
552
553 struct {
554 struct brw_gs_prog_data *prog_data;
555
556 GLboolean prog_active;
557 dri_bo *prog_bo;
558 dri_bo *state_bo;
559 } gs;
560
561 struct {
562 struct brw_clip_prog_data *prog_data;
563
564 dri_bo *prog_bo;
565 dri_bo *state_bo;
566 dri_bo *vp_bo;
567 } clip;
568
569
570 struct {
571 struct brw_sf_prog_data *prog_data;
572
573 dri_bo *prog_bo;
574 dri_bo *state_bo;
575 dri_bo *vp_bo;
576 } sf;
577
578 struct {
579 struct brw_wm_prog_data *prog_data;
580 struct brw_wm_compile *compile_data;
581
582 /* Input sizes, calculated from active vertex program:
583 */
584 GLuint input_size_masks[4];
585
586 /** Array of surface default colors (texture border color) */
587 dri_bo *sdc_bo[BRW_MAX_TEX_UNIT];
588
589 GLuint render_surf;
590 GLuint nr_surfaces;
591
592 GLuint max_threads;
593 dri_bo *scratch_buffer;
594
595 GLuint sampler_count;
596 dri_bo *sampler_bo;
597
598 /** Binding table of pointers to surf_bo entries */
599 dri_bo *bind_bo;
600 dri_bo *surf_bo[BRW_WM_MAX_SURF];
601
602 dri_bo *prog_bo;
603 dri_bo *state_bo;
604 } wm;
605
606
607 struct {
608 dri_bo *prog_bo;
609 dri_bo *state_bo;
610 dri_bo *vp_bo;
611 } cc;
612
613
614 /* Used to give every program string a unique id
615 */
616 GLuint program_id;
617 };
618
619
620 #define BRW_PACKCOLOR8888(r,g,b,a) ((r<<24) | (g<<16) | (b<<8) | a)
621
622
623
624 /*======================================================================
625 * brw_vtbl.c
626 */
627 void brwInitVtbl( struct brw_context *brw );
628 void brw_do_flush( struct brw_context *brw,
629 GLuint flags );
630
631 /*======================================================================
632 * brw_context.c
633 */
634 GLboolean brwCreateContext( const __GLcontextModes *mesaVis,
635 __DRIcontextPrivate *driContextPriv,
636 void *sharedContextPrivate);
637
638
639
640 /*======================================================================
641 * brw_state.c
642 */
643 void brw_validate_state( struct brw_context *brw );
644 void brw_init_state( struct brw_context *brw );
645 void brw_destroy_state( struct brw_context *brw );
646
647
648 /*======================================================================
649 * brw_state_dump.c
650 */
651 void brw_debug_batch(struct intel_context *intel);
652
653 /*======================================================================
654 * brw_tex.c
655 */
656 void brwUpdateTextureState( struct intel_context *intel );
657 void brw_FrameBufferTexInit( struct brw_context *brw );
658 void brw_FrameBufferTexDestroy( struct brw_context *brw );
659 void brw_validate_textures( struct brw_context *brw );
660
661 /*======================================================================
662 * brw_metaops.c
663 */
664
665 void brw_init_metaops( struct brw_context *brw );
666 void brw_destroy_metaops( struct brw_context *brw );
667
668
669 /*======================================================================
670 * brw_program.c
671 */
672 void brwInitFragProgFuncs( struct dd_function_table *functions );
673
674
675 /* brw_urb.c
676 */
677 void brw_upload_urb_fence(struct brw_context *brw);
678
679 void brw_upload_constant_buffer_state(struct brw_context *brw);
680
681
682 /*======================================================================
683 * Inline conversion functions. These are better-typed than the
684 * macros used previously:
685 */
686 static inline struct brw_context *
687 brw_context( GLcontext *ctx )
688 {
689 return (struct brw_context *)ctx;
690 }
691
692 #endif
693