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"
38 #include "main/imports.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_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_PSP 0x800
134 #define BRW_NEW_FENCE 0x2000
135 #define BRW_NEW_INDICES 0x4000
136 #define BRW_NEW_VERTICES 0x8000
138 * Used for any batch entry with a relocated pointer that will be used
139 * by any 3D rendering.
141 #define BRW_NEW_BATCH 0x10000
142 /** brw->depth_region updated */
143 #define BRW_NEW_DEPTH_BUFFER 0x20000
144 #define BRW_NEW_NR_WM_SURFACES 0x40000
145 #define BRW_NEW_NR_VS_SURFACES 0x80000
147 struct brw_state_flags
{
148 /** State update flags signalled by mesa internals */
151 * State update flags signalled as the result of brw_tracked_state updates
154 /** State update flags signalled by brw_state_cache.c searches */
159 /** Subclass of Mesa vertex program */
160 struct brw_vertex_program
{
161 struct gl_vertex_program program
;
163 dri_bo
*const_buffer
; /** Program constant buffer/surface */
167 /** Subclass of Mesa fragment program */
168 struct brw_fragment_program
{
169 struct gl_fragment_program program
;
170 GLuint id
; /**< serial no. to identify frag progs, never re-used */
171 GLboolean isGLSL
; /**< really, any IF/LOOP/CONT/BREAK instructions */
173 dri_bo
*const_buffer
; /** Program constant buffer/surface */
177 /* Data about a particular attempt to compile a program. Note that
178 * there can be many of these, each in a different GL state
179 * corresponding to a different brw_wm_prog_key struct, with different
182 struct brw_wm_prog_data
{
183 GLuint curb_read_length
;
184 GLuint urb_read_length
;
186 GLuint first_curbe_grf
;
188 GLuint total_scratch
;
190 GLuint nr_params
; /**< number of float params/constants */
193 /* Pointer to tracked values (only valid once
194 * _mesa_load_state_parameters has been called at runtime).
196 const GLfloat
*param
[BRW_MAX_CURBE
];
199 struct brw_sf_prog_data
{
200 GLuint urb_read_length
;
203 /* Each vertex may have upto 12 attributes, 4 components each,
204 * except WPOS which requires only 2. (11*4 + 2) == 44 ==> 11
207 * Actually we use 4 for each, so call it 12 rows.
209 GLuint urb_entry_size
;
212 struct brw_clip_prog_data
{
213 GLuint curb_read_length
; /* user planes? */
215 GLuint urb_read_length
;
219 struct brw_gs_prog_data
{
220 GLuint urb_read_length
;
224 struct brw_vs_prog_data
{
225 GLuint curb_read_length
;
226 GLuint urb_read_length
;
228 GLuint outputs_written
;
229 GLuint nr_params
; /**< number of float params/constants */
233 /* Used for calculating urb partitions:
235 GLuint urb_entry_size
;
239 /* Size == 0 if output either not written, or always [0,0,0,1]
241 struct brw_vs_ouput_sizes
{
242 GLubyte output_size
[VERT_RESULT_MAX
];
246 /** Number of texture sampler units */
247 #define BRW_MAX_TEX_UNIT 16
250 * Size of our surface binding table for the WM.
251 * This contains pointers to the drawing surfaces and current texture
252 * objects and shader constant buffers (+2).
254 #define BRW_WM_MAX_SURF (MAX_DRAW_BUFFERS + BRW_MAX_TEX_UNIT + 1)
257 * Helpers to convert drawing buffers, textures and constant buffers
258 * to surface binding table indexes, for WM.
260 #define SURF_INDEX_DRAW(d) (d)
261 #define SURF_INDEX_FRAG_CONST_BUFFER (MAX_DRAW_BUFFERS)
262 #define SURF_INDEX_TEXTURE(t) (MAX_DRAW_BUFFERS + 1 + (t))
265 * Size of surface binding table for the VS.
266 * Only one constant buffer for now.
268 #define BRW_VS_MAX_SURF 1
271 * Only a VS constant buffer
273 #define SURF_INDEX_VERT_CONST_BUFFER 0
280 BRW_SAMPLER_DEFAULT_COLOR
,
299 struct brw_cache_item
{
301 * Effectively part of the key, cache_id identifies what kind of state
302 * buffer is involved, and also which brw->state.dirty.cache flag should
303 * be set when this cache item is chosen.
305 enum brw_cache_id cache_id
;
306 /** 32-bit hash of the key data */
308 GLuint key_size
; /* for variable-sized keys */
311 GLuint nr_reloc_bufs
;
316 struct brw_cache_item
*next
;
322 struct brw_context
*brw
;
324 struct brw_cache_item
**items
;
325 GLuint size
, n_items
;
327 GLuint key_size
[BRW_MAX_CACHE
]; /* for fixed-size keys */
328 GLuint aux_size
[BRW_MAX_CACHE
];
329 char *name
[BRW_MAX_CACHE
];
331 /* Record of the last BOs chosen for each cache_id. Used to set
332 * brw->state.dirty.cache when a new cache item is chosen.
334 dri_bo
*last_bo
[BRW_MAX_CACHE
];
338 /* Considered adding a member to this struct to document which flags
339 * an update might raise so that ordering of the state atoms can be
340 * checked or derived at runtime. Dropped the idea in favor of having
341 * a debug mode where the state is monitored for flags which are
342 * raised that have already been tested against.
344 struct brw_tracked_state
{
345 struct brw_state_flags dirty
;
346 void (*prepare
)( struct brw_context
*brw
);
347 void (*emit
)( struct brw_context
*brw
);
350 /* Flags for brw->state.cache.
352 #define CACHE_NEW_CC_VP (1<<BRW_CC_VP)
353 #define CACHE_NEW_CC_UNIT (1<<BRW_CC_UNIT)
354 #define CACHE_NEW_WM_PROG (1<<BRW_WM_PROG)
355 #define CACHE_NEW_SAMPLER_DEFAULT_COLOR (1<<BRW_SAMPLER_DEFAULT_COLOR)
356 #define CACHE_NEW_SAMPLER (1<<BRW_SAMPLER)
357 #define CACHE_NEW_WM_UNIT (1<<BRW_WM_UNIT)
358 #define CACHE_NEW_SF_PROG (1<<BRW_SF_PROG)
359 #define CACHE_NEW_SF_VP (1<<BRW_SF_VP)
360 #define CACHE_NEW_SF_UNIT (1<<BRW_SF_UNIT)
361 #define CACHE_NEW_VS_UNIT (1<<BRW_VS_UNIT)
362 #define CACHE_NEW_VS_PROG (1<<BRW_VS_PROG)
363 #define CACHE_NEW_GS_UNIT (1<<BRW_GS_UNIT)
364 #define CACHE_NEW_GS_PROG (1<<BRW_GS_PROG)
365 #define CACHE_NEW_CLIP_VP (1<<BRW_CLIP_VP)
366 #define CACHE_NEW_CLIP_UNIT (1<<BRW_CLIP_UNIT)
367 #define CACHE_NEW_CLIP_PROG (1<<BRW_CLIP_PROG)
368 #define CACHE_NEW_SURFACE (1<<BRW_SS_SURFACE)
369 #define CACHE_NEW_SURF_BIND (1<<BRW_SS_SURF_BIND)
371 struct brw_cached_batch_item
{
372 struct header
*header
;
374 struct brw_cached_batch_item
*next
;
379 /* Protect against a future where VERT_ATTRIB_MAX > 32. Wouldn't life
380 * be easier if C allowed arrays of packed elements?
382 #define ATTRIB_BIT_DWORDS ((VERT_ATTRIB_MAX+31)/32)
384 struct brw_vertex_element
{
385 const struct gl_client_array
*glarray
;
387 /** Size of a complete element */
389 /** Number of uploaded elements for this input. */
391 /** Byte stride between elements in the uploaded array */
393 /** Offset of the first element within the buffer object */
395 /** Buffer object containing the uploaded vertex data */
401 struct brw_vertex_info
{
402 GLuint varying
; /* varying:1[VERT_ATTRIB_MAX] */
403 GLuint sizes
[ATTRIB_BIT_DWORDS
* 2]; /* sizes:2[VERT_ATTRIB_MAX] */
409 /* Cache for TNL programs.
411 struct brw_tnl_cache_item
{
415 struct brw_tnl_cache_item
*next
;
418 struct brw_tnl_cache
{
419 struct brw_tnl_cache_item
**items
;
420 GLuint size
, n_items
;
423 struct brw_query_object
{
424 struct gl_query_object Base
;
426 /** Doubly linked list of active query objects in the context. */
427 struct brw_query_object
*prev
, *next
;
429 /** Last query BO associated with this query. */
431 /** First index in bo with query data for this object. */
433 /** Last index in bo with query data for this object. */
436 /* Total count of pixels from previous BOs */
442 struct intel_context intel
;
445 GLboolean emit_state_always
;
446 GLboolean tmp_fallback
;
447 GLboolean no_batch_wrap
;
450 struct brw_state_flags dirty
;
451 struct brw_tracked_state
**atoms
;
454 GLuint nr_color_regions
;
455 struct intel_region
*color_regions
[MAX_DRAW_BUFFERS
];
456 struct intel_region
*depth_region
;
459 * List of buffers accumulated in brw_validate_state to receive
460 * dri_bo_check_aperture treatment before exec, so we can know if we
461 * should flush the batch and try again before emitting primitives.
463 * This can be a fixed number as we only have a limited number of
464 * objects referenced from the batchbuffer in a primitive emit,
465 * consisting of the vertex buffers, pipelined state pointers,
466 * the CURBE, the depth buffer, and a query BO.
468 dri_bo
*validated_bos
[VERT_ATTRIB_MAX
+ 16];
469 int validated_bo_count
;
472 struct brw_cache cache
;
473 struct brw_cached_batch_item
*cached_batch_items
;
476 struct brw_vertex_element inputs
[VERT_ATTRIB_MAX
];
478 #define BRW_NR_UPLOAD_BUFS 17
479 #define BRW_UPLOAD_INIT_SIZE (128*1024)
486 /* Summary of size and varying of active arrays, so we can check
487 * for changes to this state:
489 struct brw_vertex_info info
;
490 unsigned int min_index
, max_index
;
495 * Index buffer for this draw_prims call.
497 * Updates are signaled by BRW_NEW_INDICES.
499 const struct _mesa_index_buffer
*ib
;
505 /* Active vertex program:
507 const struct gl_vertex_program
*vertex_program
;
508 const struct gl_fragment_program
*fragment_program
;
511 /* For populating the gtt:
513 GLuint next_free_page
;
516 /* BRW_NEW_URB_ALLOCATIONS:
519 GLuint vsize
; /* vertex size plus header in urb registers */
520 GLuint csize
; /* constant buffer size in urb registers */
521 GLuint sfsize
; /* setup data size in urb registers */
523 GLboolean constrained
;
525 GLuint nr_vs_entries
;
526 GLuint nr_gs_entries
;
527 GLuint nr_clip_entries
;
528 GLuint nr_sf_entries
;
529 GLuint nr_cs_entries
;
531 /* GLuint vs_size; */
532 /* GLuint gs_size; */
533 /* GLuint clip_size; */
534 /* GLuint sf_size; */
535 /* GLuint cs_size; */
545 /* BRW_NEW_CURBE_OFFSETS:
548 GLuint wm_start
; /**< pos of first wm const in CURBE buffer */
549 GLuint wm_size
; /**< number of float[4] consts, multiple of 16 */
556 /* Dynamic tracker which changes to reflect the state referenced
557 * by active fp and vp program parameters:
559 struct brw_tracked_state tracked_state
;
562 /** Offset within curbe_bo of space for current curbe entry */
564 /** Offset within curbe_bo of space for next curbe entry */
565 GLuint curbe_next_offset
;
570 * Whether we should create a new bo instead of reusing the old one
571 * (if we just dispatch the batch pointing at the old one.
573 GLboolean need_new_bo
;
577 struct brw_vs_prog_data
*prog_data
;
582 /** Binding table of pointers to surf_bo entries */
584 dri_bo
*surf_bo
[BRW_VS_MAX_SURF
];
589 struct brw_gs_prog_data
*prog_data
;
591 GLboolean prog_active
;
597 struct brw_clip_prog_data
*prog_data
;
606 struct brw_sf_prog_data
*prog_data
;
614 struct brw_wm_prog_data
*prog_data
;
615 struct brw_wm_compile
*compile_data
;
617 /* Input sizes, calculated from active vertex program:
619 GLuint input_size_masks
[4];
621 /** Array of surface default colors (texture border color) */
622 dri_bo
*sdc_bo
[BRW_MAX_TEX_UNIT
];
630 GLuint sampler_count
;
633 /** Binding table of pointers to surf_bo entries */
635 dri_bo
*surf_bo
[BRW_WM_MAX_SURF
];
649 struct brw_query_object active_head
;
654 /* Used to give every program string a unique id
660 #define BRW_PACKCOLOR8888(r,g,b,a) ((r<<24) | (g<<16) | (b<<8) | a)
664 /*======================================================================
667 void brwInitVtbl( struct brw_context
*brw
);
669 /*======================================================================
672 GLboolean
brwCreateContext( const __GLcontextModes
*mesaVis
,
673 __DRIcontextPrivate
*driContextPriv
,
674 void *sharedContextPrivate
);
676 /*======================================================================
679 void brw_init_queryobj_functions(struct dd_function_table
*functions
);
680 void brw_prepare_query_begin(struct brw_context
*brw
);
681 void brw_emit_query_begin(struct brw_context
*brw
);
682 void brw_emit_query_end(struct brw_context
*brw
);
684 /*======================================================================
687 void brw_debug_batch(struct intel_context
*intel
);
689 /*======================================================================
692 void brwUpdateTextureState( struct intel_context
*intel
);
693 void brw_FrameBufferTexInit( struct brw_context
*brw
,
694 struct intel_region
*region
);
695 void brw_FrameBufferTexDestroy( struct brw_context
*brw
);
696 void brw_validate_textures( struct brw_context
*brw
);
699 /*======================================================================
702 void brwInitFragProgFuncs( struct dd_function_table
*functions
);
707 void brw_upload_urb_fence(struct brw_context
*brw
);
711 void brw_upload_cs_urb_state(struct brw_context
*brw
);
714 /*======================================================================
715 * Inline conversion functions. These are better-typed than the
716 * macros used previously:
718 static INLINE
struct brw_context
*
719 brw_context( GLcontext
*ctx
)
721 return (struct brw_context
*)ctx
;
724 static INLINE
struct brw_vertex_program
*
725 brw_vertex_program(struct gl_vertex_program
*p
)
727 return (struct brw_vertex_program
*) p
;
730 static INLINE
const struct brw_vertex_program
*
731 brw_vertex_program_const(const struct gl_vertex_program
*p
)
733 return (const struct brw_vertex_program
*) p
;
736 static INLINE
struct brw_fragment_program
*
737 brw_fragment_program(struct gl_fragment_program
*p
)
739 return (struct brw_fragment_program
*) p
;
742 static INLINE
const struct brw_fragment_program
*
743 brw_fragment_program_const(const struct gl_fragment_program
*p
)
745 return (const struct brw_fragment_program
*) p
;
750 #define DO_SETUP_BITS ((1<<(FRAG_ATTRIB_MAX)) - 1)