2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
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:
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.
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.
26 **********************************************************************/
29 * Keith Whitwell <keith@tungstengraphics.com>
33 #ifndef BRWCONTEXT_INC
34 #define BRWCONTEXT_INC
36 #include "brw_structs.h"
37 #include "brw_winsys.h"
38 #include "pipe/p_state.h"
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.
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 spawning a thread.
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.
55 * PUE - primitive URB entry. An urb entry produced by the setup (SF)
56 * unit holding rasterization and interpolation parameters.
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.
65 * MRF - message register file. Threads communicate (and terminate)
66 * by sending messages. Message parameters are placed in contiguous
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.
72 * R0 - GRF register 0. Typically holds control information used when
73 * sending messages to other threads.
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.
79 * Fixed function units:
81 * CS - Command streamer. Notional first unit, little software
82 * interaction. Holds the URB entries used for constant data, ie the
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.
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.
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.
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.
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.
114 * CC - Color Calculator. No EU threads associated with this unit.
115 * Handles blending and (presumably) depth and stencil testing.
118 #define BRW_MAX_CURBE (32*16)
122 #define PIPE_NEW_DEPTH_STENCIL_ALPHA 0x1
123 #define PIPE_NEW_RAST 0x2
124 #define PIPE_NEW_BLEND 0x2
125 #define PIPE_NEW_VIEWPORT 0x2
126 #define PIPE_NEW_FRAMEBUFFER 0x2
127 #define PIPE_NEW_VERTEX_BUFFER 0x2
128 #define PIPE_NEW_VERTEX_ELEMENT 0x2
129 #define PIPE_NEW_FRAGMENT_SHADER 0x2
130 #define PIPE_NEW_VERTEX_SHADER 0x2
131 #define PIPE_NEW_FRAGMENT_CONSTS 0x2
132 #define PIPE_NEW_VERTEX_CONSTS 0x2
135 #define BRW_NEW_URB_FENCE 0x1
136 #define BRW_NEW_FRAGMENT_PROGRAM 0x2
137 #define BRW_NEW_VERTEX_PROGRAM 0x4
138 #define BRW_NEW_INPUT_DIMENSIONS 0x8
139 #define BRW_NEW_CURBE_OFFSETS 0x10
140 #define BRW_NEW_REDUCED_PRIMITIVE 0x20
141 #define BRW_NEW_PRIMITIVE 0x40
142 #define BRW_NEW_CONTEXT 0x80
143 #define BRW_NEW_WM_INPUT_DIMENSIONS 0x100
144 #define BRW_NEW_PSP 0x800
145 #define BRW_NEW_WM_SURFACES 0x1000
146 #define BRW_NEW_FENCE 0x2000
147 #define BRW_NEW_INDICES 0x4000
148 #define BRW_NEW_VERTICES 0x8000
150 * Used for any batch entry with a relocated pointer that will be used
151 * by any 3D rendering.
153 #define BRW_NEW_BATCH 0x10000
154 /** brw->depth_region updated */
155 #define BRW_NEW_DEPTH_BUFFER 0x20000
156 #define BRW_NEW_NR_WM_SURFACES 0x40000
157 #define BRW_NEW_NR_VS_SURFACES 0x80000
158 #define BRW_NEW_INDEX_BUFFER 0x100000
160 struct brw_state_flags
{
161 /** State update flags signalled by mesa internals */
164 * State update flags signalled as the result of brw_tracked_state updates
167 /** State update flags signalled by brw_state_cache.c searches */
172 struct brw_vertex_program
{
173 const struct tgsi_token
*tokens
;
175 struct brw_winsys_buffer
*const_buffer
; /** Program constant buffer/surface */
176 GLboolean use_const_buffer
;
180 /** Subclass of Mesa fragment program */
181 struct brw_fragment_program
{
182 const struct tgsi_token
*tokens
;
184 GLuint id
; /**< serial no. to identify frag progs, never re-used */
185 GLboolean isGLSL
; /**< any IF/LOOP/CONT/BREAK instructions */
187 struct brw_winsys_buffer
*const_buffer
; /** Program constant buffer/surface */
188 GLboolean use_const_buffer
;
192 /* Data about a particular attempt to compile a program. Note that
193 * there can be many of these, each in a different GL state
194 * corresponding to a different brw_wm_prog_key struct, with different
197 struct brw_wm_prog_data
{
198 GLuint curb_read_length
;
199 GLuint urb_read_length
;
201 GLuint first_curbe_grf
;
203 GLuint total_scratch
;
205 GLuint nr_params
; /**< number of float params/constants */
208 /* Pointer to tracked values (only valid once
209 * _mesa_load_state_parameters has been called at runtime).
211 const GLfloat
*param
[BRW_MAX_CURBE
];
214 struct brw_sf_prog_data
{
215 GLuint urb_read_length
;
218 /* Each vertex may have upto 12 attributes, 4 components each,
219 * except WPOS which requires only 2. (11*4 + 2) == 44 ==> 11
222 * Actually we use 4 for each, so call it 12 rows.
224 GLuint urb_entry_size
;
227 struct brw_clip_prog_data
{
228 GLuint curb_read_length
; /* user planes? */
230 GLuint urb_read_length
;
234 struct brw_gs_prog_data
{
235 GLuint urb_read_length
;
239 struct brw_vs_prog_data
{
240 GLuint curb_read_length
;
241 GLuint urb_read_length
;
243 GLuint outputs_written
;
244 GLuint nr_params
; /**< number of float params/constants */
248 /* Used for calculating urb partitions:
250 GLuint urb_entry_size
;
254 /* Size == 0 if output either not written, or always [0,0,0,1]
256 struct brw_vs_ouput_sizes
{
257 GLubyte output_size
[PIPE_MAX_SHADER_OUTPUTS
];
261 /** Number of texture sampler units */
262 #define BRW_MAX_TEX_UNIT 16
265 * Size of our surface binding table for the WM.
266 * This contains pointers to the drawing surfaces and current texture
267 * objects and shader constant buffers (+2).
269 #define BRW_WM_MAX_SURF (MAX_DRAW_BUFFERS + BRW_MAX_TEX_UNIT + 1)
272 * Helpers to convert drawing buffers, textures and constant buffers
273 * to surface binding table indexes, for WM.
275 #define SURF_INDEX_DRAW(d) (d)
276 #define SURF_INDEX_FRAG_CONST_BUFFER (MAX_DRAW_BUFFERS)
277 #define SURF_INDEX_TEXTURE(t) (MAX_DRAW_BUFFERS + 1 + (t))
280 * Size of surface binding table for the VS.
281 * Only one constant buffer for now.
283 #define BRW_VS_MAX_SURF 1
286 * Only a VS constant buffer
288 #define SURF_INDEX_VERT_CONST_BUFFER 0
295 BRW_SAMPLER_DEFAULT_COLOR
,
314 struct brw_cache_item
{
316 * Effectively part of the key, cache_id identifies what kind of state
317 * buffer is involved, and also which brw->state.dirty.cache flag should
318 * be set when this cache item is chosen.
320 enum brw_cache_id cache_id
;
321 /** 32-bit hash of the key data */
323 GLuint key_size
; /* for variable-sized keys */
325 struct brw_winsys_buffer
**reloc_bufs
;
326 GLuint nr_reloc_bufs
;
328 struct brw_winsys_buffer
*bo
;
331 struct brw_cache_item
*next
;
337 struct brw_context
*brw
;
339 struct brw_cache_item
**items
;
340 GLuint size
, n_items
;
342 GLuint key_size
[BRW_MAX_CACHE
]; /* for fixed-size keys */
343 GLuint aux_size
[BRW_MAX_CACHE
];
344 char *name
[BRW_MAX_CACHE
];
346 /* Record of the last BOs chosen for each cache_id. Used to set
347 * brw->state.dirty.cache when a new cache item is chosen.
349 struct brw_winsys_buffer
*last_bo
[BRW_MAX_CACHE
];
353 /* Considered adding a member to this struct to document which flags
354 * an update might raise so that ordering of the state atoms can be
355 * checked or derived at runtime. Dropped the idea in favor of having
356 * a debug mode where the state is monitored for flags which are
357 * raised that have already been tested against.
359 struct brw_tracked_state
{
360 struct brw_state_flags dirty
;
361 void (*prepare
)( struct brw_context
*brw
);
362 void (*emit
)( struct brw_context
*brw
);
365 /* Flags for brw->state.cache.
367 #define CACHE_NEW_CC_VP (1<<BRW_CC_VP)
368 #define CACHE_NEW_CC_UNIT (1<<BRW_CC_UNIT)
369 #define CACHE_NEW_WM_PROG (1<<BRW_WM_PROG)
370 #define CACHE_NEW_SAMPLER_DEFAULT_COLOR (1<<BRW_SAMPLER_DEFAULT_COLOR)
371 #define CACHE_NEW_SAMPLER (1<<BRW_SAMPLER)
372 #define CACHE_NEW_WM_UNIT (1<<BRW_WM_UNIT)
373 #define CACHE_NEW_SF_PROG (1<<BRW_SF_PROG)
374 #define CACHE_NEW_SF_VP (1<<BRW_SF_VP)
375 #define CACHE_NEW_SF_UNIT (1<<BRW_SF_UNIT)
376 #define CACHE_NEW_VS_UNIT (1<<BRW_VS_UNIT)
377 #define CACHE_NEW_VS_PROG (1<<BRW_VS_PROG)
378 #define CACHE_NEW_GS_UNIT (1<<BRW_GS_UNIT)
379 #define CACHE_NEW_GS_PROG (1<<BRW_GS_PROG)
380 #define CACHE_NEW_CLIP_VP (1<<BRW_CLIP_VP)
381 #define CACHE_NEW_CLIP_UNIT (1<<BRW_CLIP_UNIT)
382 #define CACHE_NEW_CLIP_PROG (1<<BRW_CLIP_PROG)
383 #define CACHE_NEW_SURFACE (1<<BRW_SS_SURFACE)
384 #define CACHE_NEW_SURF_BIND (1<<BRW_SS_SURF_BIND)
386 struct brw_cached_batch_item
{
387 struct header
*header
;
389 struct brw_cached_batch_item
*next
;
394 /* Protect against a future where VERT_ATTRIB_MAX > 32. Wouldn't life
395 * be easier if C allowed arrays of packed elements?
397 #define VS_INPUT_BITMASK_DWORDS ((PIPE_MAX_SHADER_INPUTS+31)/32)
402 struct brw_vertex_info
{
403 GLuint sizes
[VS_INPUT_BITMASK_DWORDS
* 2]; /* sizes:2[VERT_ATTRIB_MAX] */
407 struct brw_query_object
{
408 /** Doubly linked list of active query objects in the context. */
409 struct brw_query_object
*prev
, *next
;
411 /** Last query BO associated with this query. */
412 struct brw_winsys_buffer
*bo
;
413 /** First index in bo with query data for this object. */
415 /** Last index in bo with query data for this object. */
418 /* Total count of pixels from previous BOs */
424 * brw_context is derived from pipe_context
428 struct pipe_context
*pipe
;
429 struct pipe_screen
*screen
;
431 struct brw_winsys_screen
*sws
;
435 GLboolean emit_state_always
;
436 GLboolean no_batch_wrap
;
438 /* Active vertex program:
440 const struct gl_vertex_program
*vertex_program
;
441 const struct gl_fragment_program
*fragment_program
;
442 struct pipe_framebuffer_state fb
;
445 struct brw_state_flags dirty
;
448 * List of buffers accumulated in brw_validate_state to receive
449 * dri_bo_check_aperture treatment before exec, so we can know if we
450 * should flush the batch and try again before emitting primitives.
452 * This can be a fixed number as we only have a limited number of
453 * objects referenced from the batchbuffer in a primitive emit,
454 * consisting of the vertex buffers, pipelined state pointers,
455 * the CURBE, the depth buffer, and a query BO.
457 struct brw_winsys_buffer
*validated_bos
[PIPE_MAX_SHADER_INPUTS
+ 16];
458 int validated_bo_count
;
461 struct brw_cache cache
; /** non-surface items */
462 struct brw_cache surface_cache
; /* surface items */
463 struct brw_cached_batch_item
*cached_batch_items
;
466 struct pipe_vertex_element vertex_element
[PIPE_MAX_ATTRIBS
];
467 struct pipe_vertex_buffer vertex_buffer
[PIPE_MAX_ATTRIBS
];
468 unsigned num_vertex_element
;
469 unsigned num_vertex_buffer
;
471 struct u_upload_mgr
*upload_vertex
;
472 struct u_upload_mgr
*upload_index
;
475 /* Summary of size and varying of active arrays, so we can check
476 * for changes to this state:
478 struct brw_vertex_info info
;
479 unsigned int min_index
, max_index
;
484 * Index buffer for this draw_prims call.
486 * Updates are signaled by BRW_NEW_INDICES.
488 const struct _mesa_index_buffer
*ib
;
490 /* Updates to these fields are signaled by BRW_NEW_INDEX_BUFFER. */
491 struct brw_winsys_buffer
*bo
;
494 /* Offset to index buffer index to use in CMD_3D_PRIM so that we can
495 * avoid re-uploading the IB packet over and over if we're actually
496 * referencing the same index buffer.
498 unsigned int start_vertex_offset
;
502 /* BRW_NEW_URB_ALLOCATIONS:
505 GLuint vsize
; /* vertex size plus header in urb registers */
506 GLuint csize
; /* constant buffer size in urb registers */
507 GLuint sfsize
; /* setup data size in urb registers */
509 GLboolean constrained
;
511 GLuint nr_vs_entries
;
512 GLuint nr_gs_entries
;
513 GLuint nr_clip_entries
;
514 GLuint nr_sf_entries
;
515 GLuint nr_cs_entries
;
525 /* BRW_NEW_CURBE_OFFSETS:
528 GLuint wm_start
; /**< pos of first wm const in CURBE buffer */
529 GLuint wm_size
; /**< number of float[4] consts, multiple of 16 */
536 struct brw_winsys_buffer
*curbe_bo
;
537 /** Offset within curbe_bo of space for current curbe entry */
539 /** Offset within curbe_bo of space for next curbe entry */
540 GLuint curbe_next_offset
;
545 * Whether we should create a new bo instead of reusing the old one
546 * (if we just dispatch the batch pointing at the old one.
548 GLboolean need_new_bo
;
552 struct brw_vs_prog_data
*prog_data
;
554 struct brw_winsys_buffer
*prog_bo
;
555 struct brw_winsys_buffer
*state_bo
;
557 /** Binding table of pointers to surf_bo entries */
558 struct brw_winsys_buffer
*bind_bo
;
559 struct brw_winsys_buffer
*surf_bo
[BRW_VS_MAX_SURF
];
564 struct brw_gs_prog_data
*prog_data
;
566 GLboolean prog_active
;
567 struct brw_winsys_buffer
*prog_bo
;
568 struct brw_winsys_buffer
*state_bo
;
572 struct brw_clip_prog_data
*prog_data
;
574 struct brw_winsys_buffer
*prog_bo
;
575 struct brw_winsys_buffer
*state_bo
;
576 struct brw_winsys_buffer
*vp_bo
;
581 struct brw_sf_prog_data
*prog_data
;
583 struct brw_winsys_buffer
*prog_bo
;
584 struct brw_winsys_buffer
*state_bo
;
585 struct brw_winsys_buffer
*vp_bo
;
589 struct brw_wm_prog_data
*prog_data
;
590 struct brw_wm_compile
*compile_data
;
592 /** Input sizes, calculated from active vertex program.
593 * One bit per fragment program input attribute.
595 //GLbitfield input_size_masks[4];
597 /** Array of surface default colors (texture border color) */
598 struct brw_winsys_buffer
*sdc_bo
[BRW_MAX_TEX_UNIT
];
604 struct brw_winsys_buffer
*scratch_bo
;
606 GLuint sampler_count
;
607 struct brw_winsys_buffer
*sampler_bo
;
609 /** Binding table of pointers to surf_bo entries */
610 struct brw_winsys_buffer
*bind_bo
;
611 struct brw_winsys_buffer
*surf_bo
[PIPE_MAX_COLOR_BUFS
];
613 struct brw_winsys_buffer
*prog_bo
;
614 struct brw_winsys_buffer
*state_bo
;
619 struct brw_winsys_buffer
*prog_bo
;
620 struct brw_winsys_buffer
*state_bo
;
621 struct brw_winsys_buffer
*vp_bo
;
625 struct brw_query_object active_head
;
626 struct brw_winsys_buffer
*bo
;
630 /* Used to give every program string a unique id
636 #define BRW_PACKCOLOR8888(r,g,b,a) ((r<<24) | (g<<16) | (b<<8) | a)
640 /*======================================================================
643 void brwInitVtbl( struct brw_context
*brw
);
646 /*======================================================================
649 void brw_init_query(struct brw_context
*brw
);
650 void brw_prepare_query_begin(struct brw_context
*brw
);
651 void brw_emit_query_begin(struct brw_context
*brw
);
652 void brw_emit_query_end(struct brw_context
*brw
);
654 /*======================================================================
657 void brw_debug_batch(struct brw_context
*intel
);
659 /*======================================================================
662 void brw_validate_textures( struct brw_context
*brw
);
665 /*======================================================================
668 void brw_init_shader_funcs( struct brw_context
*brw
);
673 void brw_upload_urb_fence(struct brw_context
*brw
);
677 void brw_upload_cs_urb_state(struct brw_context
*brw
);
680 int brw_disasm (FILE *file
, struct brw_instruction
*inst
);
682 /*======================================================================
683 * Inline conversion functions. These are better-typed than the
684 * macros used previously:
686 static INLINE
struct brw_context
*
687 brw_context( struct pipe_context
*ctx
)
689 return (struct brw_context
*)ctx
;
692 static INLINE
struct brw_vertex_program
*
693 brw_vertex_program(struct gl_vertex_program
*p
)
695 return (struct brw_vertex_program
*) p
;
698 static INLINE
const struct brw_vertex_program
*
699 brw_vertex_program_const(const struct gl_vertex_program
*p
)
701 return (const struct brw_vertex_program
*) p
;
704 static INLINE
struct brw_fragment_program
*
705 brw_fragment_program(struct gl_fragment_program
*p
)
707 return (struct brw_fragment_program
*) p
;
710 static INLINE
const struct brw_fragment_program
*
711 brw_fragment_program_const(const struct gl_fragment_program
*p
)
713 return (const struct brw_fragment_program
*) p
;
718 #define DO_SETUP_BITS ((1<<(FRAG_ATTRIB_MAX)) - 1)