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
138 #define BRW_NEW_INDICES 0x8000
139 #define BRW_NEW_VERTICES 0x10000
141 * Used for any batch entry with a relocated pointer that will be used
142 * by any 3D rendering.
144 #define BRW_NEW_BATCH 0x8000
145 /** brw->depth_region updated */
146 #define BRW_NEW_DEPTH_BUFFER 0x10000
148 struct brw_state_flags
{
149 /** State update flags signalled by mesa internals */
152 * State update flags signalled as the result of brw_tracked_state updates
155 /** State update flags signalled by brw_state_cache.c searches */
159 struct brw_vertex_program
{
160 struct gl_vertex_program program
;
162 GLuint param_state
; /* flags indicating state tracked by params */
167 struct brw_fragment_program
{
168 struct gl_fragment_program program
;
170 GLuint param_state
; /* flags indicating state tracked by params */
176 /* Data about a particular attempt to compile a program. Note that
177 * there can be many of these, each in a different GL state
178 * corresponding to a different brw_wm_prog_key struct, with different
181 struct brw_wm_prog_data
{
182 GLuint curb_read_length
;
183 GLuint urb_read_length
;
185 GLuint first_curbe_grf
;
187 GLuint total_scratch
;
192 /* Pointer to tracked values (only valid once
193 * _mesa_load_state_parameters has been called at runtime).
195 const GLfloat
*param
[BRW_MAX_CURBE
];
198 struct brw_sf_prog_data
{
199 GLuint urb_read_length
;
202 /* Each vertex may have upto 12 attributes, 4 components each,
203 * except WPOS which requires only 2. (11*4 + 2) == 44 ==> 11
206 * Actually we use 4 for each, so call it 12 rows.
208 GLuint urb_entry_size
;
211 struct brw_clip_prog_data
{
212 GLuint curb_read_length
; /* user planes? */
214 GLuint urb_read_length
;
218 struct brw_gs_prog_data
{
219 GLuint urb_read_length
;
223 struct brw_vs_prog_data
{
224 GLuint curb_read_length
;
225 GLuint urb_read_length
;
227 GLuint outputs_written
;
231 /* Used for calculating urb partitions:
233 GLuint urb_entry_size
;
237 /* Size == 0 if output either not written, or always [0,0,0,1]
239 struct brw_vs_ouput_sizes
{
240 GLubyte output_size
[VERT_RESULT_MAX
];
244 #define BRW_MAX_TEX_UNIT 8
245 #define BRW_WM_MAX_SURF BRW_MAX_TEX_UNIT + MAX_DRAW_BUFFERS
251 BRW_SAMPLER_DEFAULT_COLOR
,
270 struct brw_cache_item
{
272 * Effectively part of the key, cache_id identifies what kind of state
273 * buffer is involved, and also which brw->state.dirty.cache flag should
274 * be set when this cache item is chosen.
276 enum brw_cache_id cache_id
;
277 /** 32-bit hash of the key data */
279 GLuint key_size
; /* for variable-sized keys */
282 GLuint nr_reloc_bufs
;
287 struct brw_cache_item
*next
;
293 struct brw_context
*brw
;
295 struct brw_cache_item
**items
;
296 GLuint size
, n_items
;
298 GLuint key_size
[BRW_MAX_CACHE
]; /* for fixed-size keys */
299 GLuint aux_size
[BRW_MAX_CACHE
];
300 char *name
[BRW_MAX_CACHE
];
302 /* Record of the last BOs chosen for each cache_id. Used to set
303 * brw->state.dirty.cache when a new cache item is chosen.
305 dri_bo
*last_bo
[BRW_MAX_CACHE
];
310 struct brw_state_pointers
{
311 struct gl_colorbuffer_attrib
*Color
;
312 struct gl_depthbuffer_attrib
*Depth
;
313 struct gl_fog_attrib
*Fog
;
314 struct gl_hint_attrib
*Hint
;
315 struct gl_light_attrib
*Light
;
316 struct gl_line_attrib
*Line
;
317 struct gl_point_attrib
*Point
;
318 struct gl_polygon_attrib
*Polygon
;
319 GLuint
*PolygonStipple
;
320 struct gl_scissor_attrib
*Scissor
;
321 struct gl_stencil_attrib
*Stencil
;
322 struct gl_texture_attrib
*Texture
;
323 struct gl_transform_attrib
*Transform
;
324 struct gl_viewport_attrib
*Viewport
;
325 struct gl_vertex_program_state
*VertexProgram
;
326 struct gl_fragment_program_state
*FragmentProgram
;
329 /* Considered adding a member to this struct to document which flags
330 * an update might raise so that ordering of the state atoms can be
331 * checked or derived at runtime. Dropped the idea in favor of having
332 * a debug mode where the state is monitored for flags which are
333 * raised that have already been tested against.
335 struct brw_tracked_state
{
336 struct brw_state_flags dirty
;
337 void (*prepare
)( struct brw_context
*brw
);
338 void (*emit
)( struct brw_context
*brw
);
341 /* Flags for brw->state.cache.
343 #define CACHE_NEW_CC_VP (1<<BRW_CC_VP)
344 #define CACHE_NEW_CC_UNIT (1<<BRW_CC_UNIT)
345 #define CACHE_NEW_WM_PROG (1<<BRW_WM_PROG)
346 #define CACHE_NEW_SAMPLER_DEFAULT_COLOR (1<<BRW_SAMPLER_DEFAULT_COLOR)
347 #define CACHE_NEW_SAMPLER (1<<BRW_SAMPLER)
348 #define CACHE_NEW_WM_UNIT (1<<BRW_WM_UNIT)
349 #define CACHE_NEW_SF_PROG (1<<BRW_SF_PROG)
350 #define CACHE_NEW_SF_VP (1<<BRW_SF_VP)
351 #define CACHE_NEW_SF_UNIT (1<<BRW_SF_UNIT)
352 #define CACHE_NEW_VS_UNIT (1<<BRW_VS_UNIT)
353 #define CACHE_NEW_VS_PROG (1<<BRW_VS_PROG)
354 #define CACHE_NEW_GS_UNIT (1<<BRW_GS_UNIT)
355 #define CACHE_NEW_GS_PROG (1<<BRW_GS_PROG)
356 #define CACHE_NEW_CLIP_VP (1<<BRW_CLIP_VP)
357 #define CACHE_NEW_CLIP_UNIT (1<<BRW_CLIP_UNIT)
358 #define CACHE_NEW_CLIP_PROG (1<<BRW_CLIP_PROG)
359 #define CACHE_NEW_SURFACE (1<<BRW_SS_SURFACE)
360 #define CACHE_NEW_SURF_BIND (1<<BRW_SS_SURF_BIND)
362 struct brw_cached_batch_item
{
363 struct header
*header
;
365 struct brw_cached_batch_item
*next
;
370 /* Protect against a future where VERT_ATTRIB_MAX > 32. Wouldn't life
371 * be easier if C allowed arrays of packed elements?
373 #define ATTRIB_BIT_DWORDS ((VERT_ATTRIB_MAX+31)/32)
375 struct brw_vertex_element
{
376 const struct gl_client_array
*glarray
;
378 /** Size of a complete element */
380 /** Number of uploaded elements for this input. */
382 /** Byte stride between elements in the uploaded array */
384 /** Offset of the first element within the buffer object */
386 /** Buffer object containing the uploaded vertex data */
392 struct brw_vertex_info
{
393 GLuint varying
; /* varying:1[VERT_ATTRIB_MAX] */
394 GLuint sizes
[ATTRIB_BIT_DWORDS
* 2]; /* sizes:2[VERT_ATTRIB_MAX] */
400 /* Cache for TNL programs.
402 struct brw_tnl_cache_item
{
406 struct brw_tnl_cache_item
*next
;
409 struct brw_tnl_cache
{
410 struct brw_tnl_cache_item
**items
;
411 GLuint size
, n_items
;
418 struct intel_context intel
;
421 GLboolean emit_state_always
;
423 GLboolean tmp_fallback
;
424 GLboolean no_batch_wrap
;
427 struct brw_state_flags dirty
;
428 struct brw_tracked_state
**atoms
;
431 GLuint nr_draw_regions
;
432 struct intel_region
*draw_regions
[MAX_DRAW_BUFFERS
];
433 struct intel_region
*depth_region
;
436 struct brw_state_pointers attribs
;
437 struct brw_cache cache
;
438 struct brw_cached_batch_item
*cached_batch_items
;
441 struct brw_vertex_element inputs
[VERT_ATTRIB_MAX
];
443 #define BRW_NR_UPLOAD_BUFS 17
444 #define BRW_UPLOAD_INIT_SIZE (128*1024)
451 /* Summary of size and varying of active arrays, so we can check
452 * for changes to this state:
454 struct brw_vertex_info info
;
455 unsigned int min_index
, max_index
;
460 * Index buffer for this draw_prims call.
462 * Updates are signaled by BRW_NEW_INDICES.
464 const struct _mesa_index_buffer
*ib
;
471 /* Will be allocated on demand if needed.
473 struct brw_state_pointers attribs
;
474 struct gl_vertex_program
*vp
;
475 struct gl_fragment_program
*fp
, *fp_tex
;
477 struct gl_buffer_object
*vbo
;
479 struct intel_region
*saved_draw_region
;
480 GLuint saved_nr_draw_regions
;
481 struct intel_region
*saved_depth_region
;
483 GLuint restore_draw_buffers
[MAX_DRAW_BUFFERS
];
484 GLuint restore_num_draw_buffers
;
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
;
553 /** Offset within curbe_bo of space for current curbe entry */
555 /** Offset within curbe_bo of space for next curbe entry */
556 GLuint curbe_next_offset
;
563 struct brw_vs_prog_data
*prog_data
;
570 struct brw_gs_prog_data
*prog_data
;
572 GLboolean prog_active
;
578 struct brw_clip_prog_data
*prog_data
;
587 struct brw_sf_prog_data
*prog_data
;
595 struct brw_wm_prog_data
*prog_data
;
596 struct brw_wm_compile
*compile_data
;
598 /* Input sizes, calculated from active vertex program:
600 GLuint input_size_masks
[4];
602 /** Array of surface default colors (texture border color) */
603 dri_bo
*sdc_bo
[BRW_MAX_TEX_UNIT
];
609 dri_bo
*scratch_buffer
;
611 GLuint sampler_count
;
614 /** Binding table of pointers to surf_bo entries */
616 dri_bo
*surf_bo
[BRW_WM_MAX_SURF
];
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
);
644 void brw_do_flush( struct brw_context
*brw
,
647 /*======================================================================
650 GLboolean
brwCreateContext( const __GLcontextModes
*mesaVis
,
651 __DRIcontextPrivate
*driContextPriv
,
652 void *sharedContextPrivate
);
656 /*======================================================================
659 void brw_validate_state( struct brw_context
*brw
);
660 void brw_init_state( struct brw_context
*brw
);
661 void brw_destroy_state( struct brw_context
*brw
);
664 /*======================================================================
667 void brw_debug_batch(struct intel_context
*intel
);
669 /*======================================================================
672 void brwUpdateTextureState( struct intel_context
*intel
);
673 void brw_FrameBufferTexInit( struct brw_context
*brw
,
674 struct intel_region
*region
);
675 void brw_FrameBufferTexDestroy( struct brw_context
*brw
);
676 void brw_validate_textures( struct brw_context
*brw
);
678 /*======================================================================
682 void brw_init_metaops( struct brw_context
*brw
);
683 void brw_destroy_metaops( struct brw_context
*brw
);
686 /*======================================================================
689 void brwInitFragProgFuncs( struct dd_function_table
*functions
);
694 void brw_upload_urb_fence(struct brw_context
*brw
);
696 void brw_upload_constant_buffer_state(struct brw_context
*brw
);
699 /*======================================================================
700 * Inline conversion functions. These are better-typed than the
701 * macros used previously:
703 static INLINE
struct brw_context
*
704 brw_context( GLcontext
*ctx
)
706 return (struct brw_context
*)ctx
;
709 #define DO_SETUP_BITS ((1<<(FRAG_ATTRIB_MAX)) - 1)