i965: Delete old metaops code now that there are no remaining consumers.
[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 "main/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_PSP 0x800
134 #define BRW_NEW_FENCE 0x2000
135 #define BRW_NEW_INDICES 0x4000
136 #define BRW_NEW_VERTICES 0x8000
137 /**
138 * Used for any batch entry with a relocated pointer that will be used
139 * by any 3D rendering.
140 */
141 #define BRW_NEW_BATCH 0x10000
142 /** brw->depth_region updated */
143 #define BRW_NEW_DEPTH_BUFFER 0x20000
144 #define BRW_NEW_NR_SURFACES 0x40000
145
146 struct brw_state_flags {
147 /** State update flags signalled by mesa internals */
148 GLuint mesa;
149 /**
150 * State update flags signalled as the result of brw_tracked_state updates
151 */
152 GLuint brw;
153 /** State update flags signalled by brw_state_cache.c searches */
154 GLuint cache;
155 };
156
157 struct brw_vertex_program {
158 struct gl_vertex_program program;
159 GLuint id;
160 };
161
162
163
164 struct brw_fragment_program {
165 struct gl_fragment_program program;
166 GLuint id;
167 };
168
169
170
171
172 /* Data about a particular attempt to compile a program. Note that
173 * there can be many of these, each in a different GL state
174 * corresponding to a different brw_wm_prog_key struct, with different
175 * compiled programs:
176 */
177 struct brw_wm_prog_data {
178 GLuint curb_read_length;
179 GLuint urb_read_length;
180
181 GLuint first_curbe_grf;
182 GLuint total_grf;
183 GLuint total_scratch;
184
185 GLuint nr_params;
186 GLboolean error;
187
188 /* Pointer to tracked values (only valid once
189 * _mesa_load_state_parameters has been called at runtime).
190 */
191 const GLfloat *param[BRW_MAX_CURBE];
192 };
193
194 struct brw_sf_prog_data {
195 GLuint urb_read_length;
196 GLuint total_grf;
197
198 /* Each vertex may have upto 12 attributes, 4 components each,
199 * except WPOS which requires only 2. (11*4 + 2) == 44 ==> 11
200 * rows.
201 *
202 * Actually we use 4 for each, so call it 12 rows.
203 */
204 GLuint urb_entry_size;
205 };
206
207 struct brw_clip_prog_data {
208 GLuint curb_read_length; /* user planes? */
209 GLuint clip_mode;
210 GLuint urb_read_length;
211 GLuint total_grf;
212 };
213
214 struct brw_gs_prog_data {
215 GLuint urb_read_length;
216 GLuint total_grf;
217 };
218
219 struct brw_vs_prog_data {
220 GLuint curb_read_length;
221 GLuint urb_read_length;
222 GLuint total_grf;
223 GLuint outputs_written;
224
225 GLuint inputs_read;
226
227 /* Used for calculating urb partitions:
228 */
229 GLuint urb_entry_size;
230 };
231
232
233 /* Size == 0 if output either not written, or always [0,0,0,1]
234 */
235 struct brw_vs_ouput_sizes {
236 GLubyte output_size[VERT_RESULT_MAX];
237 };
238
239
240 #define BRW_MAX_TEX_UNIT 16
241 #define BRW_WM_MAX_SURF BRW_MAX_TEX_UNIT + MAX_DRAW_BUFFERS
242
243 enum brw_cache_id {
244 BRW_CC_VP,
245 BRW_CC_UNIT,
246 BRW_WM_PROG,
247 BRW_SAMPLER_DEFAULT_COLOR,
248 BRW_SAMPLER,
249 BRW_WM_UNIT,
250 BRW_SF_PROG,
251 BRW_SF_VP,
252 BRW_SF_UNIT,
253 BRW_VS_UNIT,
254 BRW_VS_PROG,
255 BRW_GS_UNIT,
256 BRW_GS_PROG,
257 BRW_CLIP_VP,
258 BRW_CLIP_UNIT,
259 BRW_CLIP_PROG,
260 BRW_SS_SURFACE,
261 BRW_SS_SURF_BIND,
262
263 BRW_MAX_CACHE
264 };
265
266 struct brw_cache_item {
267 /**
268 * Effectively part of the key, cache_id identifies what kind of state
269 * buffer is involved, and also which brw->state.dirty.cache flag should
270 * be set when this cache item is chosen.
271 */
272 enum brw_cache_id cache_id;
273 /** 32-bit hash of the key data */
274 GLuint hash;
275 GLuint key_size; /* for variable-sized keys */
276 const void *key;
277 dri_bo **reloc_bufs;
278 GLuint nr_reloc_bufs;
279
280 dri_bo *bo;
281 GLuint data_size;
282
283 struct brw_cache_item *next;
284 };
285
286
287
288 struct brw_cache {
289 struct brw_context *brw;
290
291 struct brw_cache_item **items;
292 GLuint size, n_items;
293
294 GLuint key_size[BRW_MAX_CACHE]; /* for fixed-size keys */
295 GLuint aux_size[BRW_MAX_CACHE];
296 char *name[BRW_MAX_CACHE];
297
298 /* Record of the last BOs chosen for each cache_id. Used to set
299 * brw->state.dirty.cache when a new cache item is chosen.
300 */
301 dri_bo *last_bo[BRW_MAX_CACHE];
302 };
303
304
305
306 struct brw_state_pointers {
307 struct gl_colorbuffer_attrib *Color;
308 struct gl_depthbuffer_attrib *Depth;
309 struct gl_fog_attrib *Fog;
310 struct gl_hint_attrib *Hint;
311 struct gl_light_attrib *Light;
312 struct gl_line_attrib *Line;
313 struct gl_point_attrib *Point;
314 struct gl_polygon_attrib *Polygon;
315 GLuint *PolygonStipple;
316 struct gl_scissor_attrib *Scissor;
317 struct gl_stencil_attrib *Stencil;
318 struct gl_texture_attrib *Texture;
319 struct gl_transform_attrib *Transform;
320 struct gl_viewport_attrib *Viewport;
321 struct gl_vertex_program_state *VertexProgram;
322 struct gl_fragment_program_state *FragmentProgram;
323 };
324
325 /* Considered adding a member to this struct to document which flags
326 * an update might raise so that ordering of the state atoms can be
327 * checked or derived at runtime. Dropped the idea in favor of having
328 * a debug mode where the state is monitored for flags which are
329 * raised that have already been tested against.
330 */
331 struct brw_tracked_state {
332 struct brw_state_flags dirty;
333 void (*prepare)( struct brw_context *brw );
334 void (*emit)( struct brw_context *brw );
335 };
336
337 /* Flags for brw->state.cache.
338 */
339 #define CACHE_NEW_CC_VP (1<<BRW_CC_VP)
340 #define CACHE_NEW_CC_UNIT (1<<BRW_CC_UNIT)
341 #define CACHE_NEW_WM_PROG (1<<BRW_WM_PROG)
342 #define CACHE_NEW_SAMPLER_DEFAULT_COLOR (1<<BRW_SAMPLER_DEFAULT_COLOR)
343 #define CACHE_NEW_SAMPLER (1<<BRW_SAMPLER)
344 #define CACHE_NEW_WM_UNIT (1<<BRW_WM_UNIT)
345 #define CACHE_NEW_SF_PROG (1<<BRW_SF_PROG)
346 #define CACHE_NEW_SF_VP (1<<BRW_SF_VP)
347 #define CACHE_NEW_SF_UNIT (1<<BRW_SF_UNIT)
348 #define CACHE_NEW_VS_UNIT (1<<BRW_VS_UNIT)
349 #define CACHE_NEW_VS_PROG (1<<BRW_VS_PROG)
350 #define CACHE_NEW_GS_UNIT (1<<BRW_GS_UNIT)
351 #define CACHE_NEW_GS_PROG (1<<BRW_GS_PROG)
352 #define CACHE_NEW_CLIP_VP (1<<BRW_CLIP_VP)
353 #define CACHE_NEW_CLIP_UNIT (1<<BRW_CLIP_UNIT)
354 #define CACHE_NEW_CLIP_PROG (1<<BRW_CLIP_PROG)
355 #define CACHE_NEW_SURFACE (1<<BRW_SS_SURFACE)
356 #define CACHE_NEW_SURF_BIND (1<<BRW_SS_SURF_BIND)
357
358 struct brw_cached_batch_item {
359 struct header *header;
360 GLuint sz;
361 struct brw_cached_batch_item *next;
362 };
363
364
365
366 /* Protect against a future where VERT_ATTRIB_MAX > 32. Wouldn't life
367 * be easier if C allowed arrays of packed elements?
368 */
369 #define ATTRIB_BIT_DWORDS ((VERT_ATTRIB_MAX+31)/32)
370
371 struct brw_vertex_element {
372 const struct gl_client_array *glarray;
373
374 /** Size of a complete element */
375 GLuint element_size;
376 /** Number of uploaded elements for this input. */
377 GLuint count;
378 /** Byte stride between elements in the uploaded array */
379 GLuint stride;
380 /** Offset of the first element within the buffer object */
381 unsigned int offset;
382 /** Buffer object containing the uploaded vertex data */
383 dri_bo *bo;
384 };
385
386
387
388 struct brw_vertex_info {
389 GLuint varying; /* varying:1[VERT_ATTRIB_MAX] */
390 GLuint sizes[ATTRIB_BIT_DWORDS * 2]; /* sizes:2[VERT_ATTRIB_MAX] */
391 };
392
393
394
395
396 /* Cache for TNL programs.
397 */
398 struct brw_tnl_cache_item {
399 GLuint hash;
400 void *key;
401 void *data;
402 struct brw_tnl_cache_item *next;
403 };
404
405 struct brw_tnl_cache {
406 struct brw_tnl_cache_item **items;
407 GLuint size, n_items;
408 };
409
410 struct brw_query_object {
411 struct gl_query_object Base;
412
413 /** Doubly linked list of active query objects in the context. */
414 struct brw_query_object *prev, *next;
415
416 /** Last query BO associated with this query. */
417 dri_bo *bo;
418 /** First index in bo with query data for this object. */
419 int first_index;
420 /** Last index in bo with query data for this object. */
421 int last_index;
422
423 /* Total count of pixels from previous BOs */
424 unsigned int count;
425 };
426
427 struct brw_context
428 {
429 struct intel_context intel;
430 GLuint primitive;
431
432 GLboolean emit_state_always;
433 GLboolean tmp_fallback;
434 GLboolean no_batch_wrap;
435
436 struct {
437 struct brw_state_flags dirty;
438 struct brw_tracked_state **atoms;
439 GLuint nr_atoms;
440
441 GLuint nr_draw_regions;
442 struct intel_region *draw_regions[MAX_DRAW_BUFFERS];
443 struct intel_region *depth_region;
444
445 /**
446 * List of buffers accumulated in brw_validate_state to receive
447 * dri_bo_check_aperture treatment before exec, so we can know if we
448 * should flush the batch and try again before emitting primitives.
449 *
450 * This can be a fixed number as we only have a limited number of
451 * objects referenced from the batchbuffer in a primitive emit,
452 * consisting of the vertex buffers, pipelined state pointers,
453 * the CURBE, the depth buffer, and a query BO.
454 */
455 dri_bo *validated_bos[VERT_ATTRIB_MAX + 16];
456 int validated_bo_count;
457 } state;
458
459 struct brw_state_pointers attribs;
460 struct brw_cache cache;
461 struct brw_cached_batch_item *cached_batch_items;
462
463 struct {
464 struct brw_vertex_element inputs[VERT_ATTRIB_MAX];
465
466 #define BRW_NR_UPLOAD_BUFS 17
467 #define BRW_UPLOAD_INIT_SIZE (128*1024)
468
469 struct {
470 dri_bo *bo;
471 GLuint offset;
472 } upload;
473
474 /* Summary of size and varying of active arrays, so we can check
475 * for changes to this state:
476 */
477 struct brw_vertex_info info;
478 unsigned int min_index, max_index;
479 } vb;
480
481 struct {
482 /**
483 * Index buffer for this draw_prims call.
484 *
485 * Updates are signaled by BRW_NEW_INDICES.
486 */
487 const struct _mesa_index_buffer *ib;
488
489 dri_bo *bo;
490 unsigned int offset;
491 } ib;
492
493 /* Active vertex program:
494 */
495 const struct gl_vertex_program *vertex_program;
496 const struct gl_fragment_program *fragment_program;
497
498
499 /* For populating the gtt:
500 */
501 GLuint next_free_page;
502
503
504 /* BRW_NEW_URB_ALLOCATIONS:
505 */
506 struct {
507 GLuint vsize; /* vertex size plus header in urb registers */
508 GLuint csize; /* constant buffer size in urb registers */
509 GLuint sfsize; /* setup data size in urb registers */
510
511 GLboolean constrained;
512
513 GLuint nr_vs_entries;
514 GLuint nr_gs_entries;
515 GLuint nr_clip_entries;
516 GLuint nr_sf_entries;
517 GLuint nr_cs_entries;
518
519 /* GLuint vs_size; */
520 /* GLuint gs_size; */
521 /* GLuint clip_size; */
522 /* GLuint sf_size; */
523 /* GLuint cs_size; */
524
525 GLuint vs_start;
526 GLuint gs_start;
527 GLuint clip_start;
528 GLuint sf_start;
529 GLuint cs_start;
530 } urb;
531
532
533 /* BRW_NEW_CURBE_OFFSETS:
534 */
535 struct {
536 GLuint wm_start;
537 GLuint wm_size;
538 GLuint clip_start;
539 GLuint clip_size;
540 GLuint vs_start;
541 GLuint vs_size;
542 GLuint total_size;
543
544 /* Dynamic tracker which changes to reflect the state referenced
545 * by active fp and vp program parameters:
546 */
547 struct brw_tracked_state tracked_state;
548
549 dri_bo *curbe_bo;
550 /** Offset within curbe_bo of space for current curbe entry */
551 GLuint curbe_offset;
552 /** Offset within curbe_bo of space for next curbe entry */
553 GLuint curbe_next_offset;
554
555 GLfloat *last_buf;
556 GLuint last_bufsz;
557 /**
558 * Whether we should create a new bo instead of reusing the old one
559 * (if we just dispatch the batch pointing at the old one.
560 */
561 GLboolean need_new_bo;
562 } curbe;
563
564 struct {
565 struct brw_vs_prog_data *prog_data;
566
567 dri_bo *prog_bo;
568 dri_bo *state_bo;
569 } vs;
570
571 struct {
572 struct brw_gs_prog_data *prog_data;
573
574 GLboolean prog_active;
575 dri_bo *prog_bo;
576 dri_bo *state_bo;
577 } gs;
578
579 struct {
580 struct brw_clip_prog_data *prog_data;
581
582 dri_bo *prog_bo;
583 dri_bo *state_bo;
584 dri_bo *vp_bo;
585 } clip;
586
587
588 struct {
589 struct brw_sf_prog_data *prog_data;
590
591 dri_bo *prog_bo;
592 dri_bo *state_bo;
593 dri_bo *vp_bo;
594 } sf;
595
596 struct {
597 struct brw_wm_prog_data *prog_data;
598 struct brw_wm_compile *compile_data;
599
600 /* Input sizes, calculated from active vertex program:
601 */
602 GLuint input_size_masks[4];
603
604 /** Array of surface default colors (texture border color) */
605 dri_bo *sdc_bo[BRW_MAX_TEX_UNIT];
606
607 GLuint render_surf;
608 GLuint nr_surfaces;
609
610 GLuint max_threads;
611 dri_bo *scratch_buffer;
612
613 GLuint sampler_count;
614 dri_bo *sampler_bo;
615
616 /** Binding table of pointers to surf_bo entries */
617 dri_bo *bind_bo;
618 dri_bo *surf_bo[BRW_WM_MAX_SURF];
619
620 dri_bo *prog_bo;
621 dri_bo *state_bo;
622 } wm;
623
624
625 struct {
626 dri_bo *prog_bo;
627 dri_bo *state_bo;
628 dri_bo *vp_bo;
629 } cc;
630
631 struct {
632 struct brw_query_object active_head;
633 dri_bo *bo;
634 int index;
635 GLboolean active;
636 } query;
637 /* Used to give every program string a unique id
638 */
639 GLuint program_id;
640 };
641
642
643 #define BRW_PACKCOLOR8888(r,g,b,a) ((r<<24) | (g<<16) | (b<<8) | a)
644
645
646
647 /*======================================================================
648 * brw_vtbl.c
649 */
650 void brwInitVtbl( struct brw_context *brw );
651 void brw_do_flush( struct brw_context *brw,
652 GLuint flags );
653
654 /*======================================================================
655 * brw_context.c
656 */
657 GLboolean brwCreateContext( const __GLcontextModes *mesaVis,
658 __DRIcontextPrivate *driContextPriv,
659 void *sharedContextPrivate);
660
661 /*======================================================================
662 * brw_queryobj.c
663 */
664 void brw_init_queryobj_functions(struct dd_function_table *functions);
665 void brw_prepare_query_begin(struct brw_context *brw);
666 void brw_emit_query_begin(struct brw_context *brw);
667 void brw_emit_query_end(struct brw_context *brw);
668
669 /*======================================================================
670 * brw_state_dump.c
671 */
672 void brw_debug_batch(struct intel_context *intel);
673
674 /*======================================================================
675 * brw_tex.c
676 */
677 void brwUpdateTextureState( struct intel_context *intel );
678 void brw_FrameBufferTexInit( struct brw_context *brw,
679 struct intel_region *region );
680 void brw_FrameBufferTexDestroy( struct brw_context *brw );
681 void brw_validate_textures( struct brw_context *brw );
682
683
684 /*======================================================================
685 * brw_program.c
686 */
687 void brwInitFragProgFuncs( struct dd_function_table *functions );
688
689
690 /* brw_urb.c
691 */
692 void brw_upload_urb_fence(struct brw_context *brw);
693
694 void brw_upload_constant_buffer_state(struct brw_context *brw);
695
696
697 /*======================================================================
698 * Inline conversion functions. These are better-typed than the
699 * macros used previously:
700 */
701 static INLINE struct brw_context *
702 brw_context( GLcontext *ctx )
703 {
704 return (struct brw_context *)ctx;
705 }
706
707 #define DO_SETUP_BITS ((1<<(FRAG_ATTRIB_MAX)) - 1)
708
709 #endif
710