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 "intel_context.h"
37 #include "brw_structs.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 spawining 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 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.
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_FALLBACK_TEXTURE 0x1
119 #define BRW_MAX_CURBE (32*16)
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
141 struct brw_state_flags
{
147 struct brw_vertex_program
{
148 struct gl_vertex_program program
;
150 GLuint param_state
; /* flags indicating state tracked by params */
155 struct brw_fragment_program
{
156 struct gl_fragment_program program
;
158 GLuint param_state
; /* flags indicating state tracked by params */
164 /* Data about a particular attempt to compile a program. Note that
165 * there can be many of these, each in a different GL state
166 * corresponding to a different brw_wm_prog_key struct, with different
169 struct brw_wm_prog_data
{
170 GLuint curb_read_length
;
171 GLuint urb_read_length
;
173 GLuint first_curbe_grf
;
175 GLuint total_scratch
;
180 /* Pointer to tracked values (only valid once
181 * _mesa_load_state_parameters has been called at runtime).
183 const GLfloat
*param
[BRW_MAX_CURBE
];
186 struct brw_sf_prog_data
{
187 GLuint urb_read_length
;
190 /* Each vertex may have upto 12 attributes, 4 components each,
191 * except WPOS which requires only 2. (11*4 + 2) == 44 ==> 11
194 * Actually we use 4 for each, so call it 12 rows.
196 GLuint urb_entry_size
;
199 struct brw_clip_prog_data
{
200 GLuint curb_read_length
; /* user planes? */
202 GLuint urb_read_length
;
206 struct brw_gs_prog_data
{
207 GLuint urb_read_length
;
211 struct brw_vs_prog_data
{
212 GLuint curb_read_length
;
213 GLuint urb_read_length
;
215 GLuint outputs_written
;
219 /* Used for calculating urb partitions:
221 GLuint urb_entry_size
;
225 /* Size == 0 if output either not written, or always [0,0,0,1]
227 struct brw_vs_ouput_sizes
{
228 GLubyte output_size
[VERT_RESULT_MAX
];
232 #define BRW_MAX_TEX_UNIT 8
233 #define BRW_WM_MAX_SURF BRW_MAX_TEX_UNIT + 1
235 /* Create a fixed sized struct for caching binding tables:
237 struct brw_surface_binding_table
{
238 GLuint surf_ss_offset
[BRW_WM_MAX_SURF
];
244 struct brw_mem_pool
{
248 GLuint offset
; /* offset of first free byte */
250 struct brw_context
*brw
;
253 struct brw_cache_item
{
255 GLuint key_size
; /* for variable-sized keys */
258 GLuint offset
; /* offset within pool's buffer */
261 struct brw_cache_item
*next
;
271 struct brw_context
*brw
;
272 struct brw_mem_pool
*pool
;
274 struct brw_cache_item
**items
;
275 GLuint size
, n_items
;
277 GLuint key_size
; /* for fixed-size keys */
280 GLuint last_addr
; /* offset of active item */
285 struct brw_state_pointers
{
286 struct gl_colorbuffer_attrib
*Color
;
287 struct gl_depthbuffer_attrib
*Depth
;
288 struct gl_fog_attrib
*Fog
;
289 struct gl_hint_attrib
*Hint
;
290 struct gl_light_attrib
*Light
;
291 struct gl_line_attrib
*Line
;
292 struct gl_point_attrib
*Point
;
293 struct gl_polygon_attrib
*Polygon
;
294 GLuint
*PolygonStipple
;
295 struct gl_scissor_attrib
*Scissor
;
296 struct gl_stencil_attrib
*Stencil
;
297 struct gl_texture_attrib
*Texture
;
298 struct gl_transform_attrib
*Transform
;
299 struct gl_viewport_attrib
*Viewport
;
300 struct gl_vertex_program_state
*VertexProgram
;
301 struct gl_fragment_program_state
*FragmentProgram
;
304 /* Considered adding a member to this struct to document which flags
305 * an update might raise so that ordering of the state atoms can be
306 * checked or derived at runtime. Dropped the idea in favor of having
307 * a debug mode where the state is monitored for flags which are
308 * raised that have already been tested against.
310 struct brw_tracked_state
{
311 struct brw_state_flags dirty
;
312 void (*update
)( struct brw_context
*brw
);
320 BRW_SAMPLER_DEFAULT_COLOR
,
334 /* These two are in the SS pool:
342 /* Flags for brw->state.cache.
344 #define CACHE_NEW_CC_VP (1<<BRW_CC_VP)
345 #define CACHE_NEW_CC_UNIT (1<<BRW_CC_UNIT)
346 #define CACHE_NEW_WM_PROG (1<<BRW_WM_PROG)
347 #define CACHE_NEW_SAMPLER_DEFAULT_COLOR (1<<BRW_SAMPLER_DEFAULT_COLOR)
348 #define CACHE_NEW_SAMPLER (1<<BRW_SAMPLER)
349 #define CACHE_NEW_WM_UNIT (1<<BRW_WM_UNIT)
350 #define CACHE_NEW_SF_PROG (1<<BRW_SF_PROG)
351 #define CACHE_NEW_SF_VP (1<<BRW_SF_VP)
352 #define CACHE_NEW_SF_UNIT (1<<BRW_SF_UNIT)
353 #define CACHE_NEW_VS_UNIT (1<<BRW_VS_UNIT)
354 #define CACHE_NEW_VS_PROG (1<<BRW_VS_PROG)
355 #define CACHE_NEW_GS_UNIT (1<<BRW_GS_UNIT)
356 #define CACHE_NEW_GS_PROG (1<<BRW_GS_PROG)
357 #define CACHE_NEW_CLIP_VP (1<<BRW_CLIP_VP)
358 #define CACHE_NEW_CLIP_UNIT (1<<BRW_CLIP_UNIT)
359 #define CACHE_NEW_CLIP_PROG (1<<BRW_CLIP_PROG)
360 #define CACHE_NEW_SURFACE (1<<BRW_SS_SURFACE)
361 #define CACHE_NEW_SURF_BIND (1<<BRW_SS_SURF_BIND)
366 enum brw_mempool_id
{
373 struct brw_cached_batch_item
{
374 struct header
*header
;
376 struct brw_cached_batch_item
*next
;
381 /* Protect against a future where VERT_ATTRIB_MAX > 32. Wouldn't life
382 * be easier if C allowed arrays of packed elements?
384 #define ATTRIB_BIT_DWORDS ((VERT_ATTRIB_MAX+31)/32)
386 struct brw_vertex_element
{
387 const struct gl_client_array
*glarray
;
389 struct brw_vertex_element_state
*vep
;
394 GLuint vbo_rebase_offset
;
399 struct brw_vertex_info
{
400 GLuint varying
; /* varying:1[VERT_ATTRIB_MAX] */
401 GLuint sizes
[ATTRIB_BIT_DWORDS
* 2]; /* sizes:2[VERT_ATTRIB_MAX] */
407 /* Cache for TNL programs.
409 struct brw_tnl_cache_item
{
413 struct brw_tnl_cache_item
*next
;
416 struct brw_tnl_cache
{
417 struct brw_tnl_cache_item
**items
;
418 GLuint size
, n_items
;
425 struct intel_context intel
;
428 GLboolean emit_state_always
;
430 GLboolean tmp_fallback
;
433 struct brw_state_flags dirty
;
434 struct brw_tracked_state
**atoms
;
438 struct intel_region
*draw_region
;
439 struct intel_region
*depth_region
;
442 struct brw_state_pointers attribs
;
443 struct brw_mem_pool pool
[BRW_MAX_POOL
];
444 struct brw_cache cache
[BRW_MAX_CACHE
];
445 struct brw_cached_batch_item
*cached_batch_items
;
449 /* Arrays with buffer objects to copy non-bufferobj arrays into
452 struct gl_client_array vbo_array
[VERT_ATTRIB_MAX
];
454 struct brw_vertex_element inputs
[VERT_ATTRIB_MAX
];
456 #define BRW_NR_UPLOAD_BUFS 17
457 #define BRW_UPLOAD_INIT_SIZE (128*1024)
460 struct gl_buffer_object
*vbo
[BRW_NR_UPLOAD_BUFS
];
467 /* Summary of size and varying of active arrays, so we can check
468 * for changes to this state:
470 struct brw_vertex_info info
;
474 /* Will be allocated on demand if needed.
476 struct brw_state_pointers attribs
;
477 struct gl_vertex_program
*vp
;
478 struct gl_fragment_program
*fp
, *fp_tex
;
480 struct gl_buffer_object
*vbo
;
482 struct intel_region
*saved_draw_region
;
483 struct intel_region
*saved_depth_region
;
485 GLuint restore_draw_mask
;
486 struct gl_fragment_program
*restore_fp
;
491 /* Track fixed function t&l in a vertex program:
493 struct gl_vertex_program
*tnl_program
;
494 struct brw_tnl_cache tnl_program_cache
;
496 /* Active vertex program:
498 const struct gl_vertex_program
*vertex_program
;
499 const struct gl_fragment_program
*fragment_program
;
502 /* For populating the gtt:
504 GLuint next_free_page
;
507 /* BRW_NEW_URB_ALLOCATIONS:
510 GLuint vsize
; /* vertex size plus header in urb registers */
511 GLuint csize
; /* constant buffer size in urb registers */
512 GLuint sfsize
; /* setup data size in urb registers */
514 GLboolean constrained
;
516 GLuint nr_vs_entries
;
517 GLuint nr_gs_entries
;
518 GLuint nr_clip_entries
;
519 GLuint nr_sf_entries
;
520 GLuint nr_cs_entries
;
522 /* GLuint vs_size; */
523 /* GLuint gs_size; */
524 /* GLuint clip_size; */
525 /* GLuint sf_size; */
526 /* GLuint cs_size; */
536 /* BRW_NEW_CURBE_OFFSETS:
547 /* Dynamic tracker which changes to reflect the state referenced
548 * by active fp and vp program parameters:
550 struct brw_tracked_state tracked_state
;
559 struct brw_vs_prog_data
*prog_data
;
561 GLuint prog_gs_offset
;
562 GLuint state_gs_offset
;
566 struct brw_gs_prog_data
*prog_data
;
568 GLboolean prog_active
;
569 GLuint prog_gs_offset
;
570 GLuint state_gs_offset
;
574 struct brw_clip_prog_data
*prog_data
;
576 GLuint prog_gs_offset
;
578 GLuint state_gs_offset
;
583 struct brw_sf_prog_data
*prog_data
;
585 GLuint prog_gs_offset
;
587 GLuint state_gs_offset
;
591 struct brw_wm_prog_data
*prog_data
;
592 struct brw_wm_compile
*compile_data
;
594 /* Input sizes, calculated from active vertex program:
596 GLuint input_size_masks
[4];
601 struct brw_sampler_default_color sdc
[BRW_MAX_TEX_UNIT
];
602 struct brw_sampler_state sampler
[BRW_MAX_TEX_UNIT
];
608 dri_bo
*scratch_buffer
;
609 GLuint scratch_buffer_size
;
611 GLuint sampler_count
;
612 GLuint sampler_gs_offset
;
614 struct brw_surface_binding_table bind
;
615 GLuint bind_ss_offset
;
617 GLuint prog_gs_offset
;
618 GLuint state_gs_offset
;
624 GLuint state_gs_offset
;
628 /* Used to give every program string a unique id
634 #define BRW_PACKCOLOR8888(r,g,b,a) ((r<<24) | (g<<16) | (b<<8) | a)
638 /*======================================================================
641 void brwInitVtbl( struct brw_context
*brw
);
642 void brw_do_flush( struct brw_context
*brw
,
645 /*======================================================================
648 GLboolean
brwCreateContext( const __GLcontextModes
*mesaVis
,
649 __DRIcontextPrivate
*driContextPriv
,
650 void *sharedContextPrivate
);
654 /*======================================================================
657 void brw_validate_state( struct brw_context
*brw
);
658 void brw_init_state( struct brw_context
*brw
);
659 void brw_destroy_state( struct brw_context
*brw
);
663 /*======================================================================
666 void brwUpdateTextureState( struct intel_context
*intel
);
667 void brwInitTextureFuncs( struct dd_function_table
*functions
);
668 void brw_FrameBufferTexInit( struct brw_context
*brw
);
669 void brw_FrameBufferTexDestroy( struct brw_context
*brw
);
671 /*======================================================================
675 void brw_init_metaops( struct brw_context
*brw
);
676 void brw_destroy_metaops( struct brw_context
*brw
);
679 /*======================================================================
682 void brwInitFragProgFuncs( struct dd_function_table
*functions
);
687 void brw_upload_urb_fence(struct brw_context
*brw
);
689 void brw_upload_constant_buffer_state(struct brw_context
*brw
);
692 /*======================================================================
693 * Inline conversion functions. These are better-typed than the
694 * macros used previously:
696 static inline struct brw_context
*
697 brw_context( GLcontext
*ctx
)
699 return (struct brw_context
*)ctx
;