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 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_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_SURFACES 0x40000
146 struct brw_state_flags
{
147 /** State update flags signalled by mesa internals */
150 * State update flags signalled as the result of brw_tracked_state updates
153 /** State update flags signalled by brw_state_cache.c searches */
157 struct brw_vertex_program
{
158 struct gl_vertex_program program
;
164 struct brw_fragment_program
{
165 struct gl_fragment_program program
;
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
177 struct brw_wm_prog_data
{
178 GLuint curb_read_length
;
179 GLuint urb_read_length
;
181 GLuint first_curbe_grf
;
183 GLuint total_scratch
;
188 /* Pointer to tracked values (only valid once
189 * _mesa_load_state_parameters has been called at runtime).
191 const GLfloat
*param
[BRW_MAX_CURBE
];
194 struct brw_sf_prog_data
{
195 GLuint urb_read_length
;
198 /* Each vertex may have upto 12 attributes, 4 components each,
199 * except WPOS which requires only 2. (11*4 + 2) == 44 ==> 11
202 * Actually we use 4 for each, so call it 12 rows.
204 GLuint urb_entry_size
;
207 struct brw_clip_prog_data
{
208 GLuint curb_read_length
; /* user planes? */
210 GLuint urb_read_length
;
214 struct brw_gs_prog_data
{
215 GLuint urb_read_length
;
219 struct brw_vs_prog_data
{
220 GLuint curb_read_length
;
221 GLuint urb_read_length
;
223 GLuint outputs_written
;
227 /* Used for calculating urb partitions:
229 GLuint urb_entry_size
;
233 /* Size == 0 if output either not written, or always [0,0,0,1]
235 struct brw_vs_ouput_sizes
{
236 GLubyte output_size
[VERT_RESULT_MAX
];
240 #define BRW_MAX_TEX_UNIT 16
241 #define BRW_WM_MAX_SURF BRW_MAX_TEX_UNIT + MAX_DRAW_BUFFERS
247 BRW_SAMPLER_DEFAULT_COLOR
,
266 struct brw_cache_item
{
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.
272 enum brw_cache_id cache_id
;
273 /** 32-bit hash of the key data */
275 GLuint key_size
; /* for variable-sized keys */
278 GLuint nr_reloc_bufs
;
283 struct brw_cache_item
*next
;
289 struct brw_context
*brw
;
291 struct brw_cache_item
**items
;
292 GLuint size
, n_items
;
294 GLuint key_size
[BRW_MAX_CACHE
]; /* for fixed-size keys */
295 GLuint aux_size
[BRW_MAX_CACHE
];
296 char *name
[BRW_MAX_CACHE
];
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.
301 dri_bo
*last_bo
[BRW_MAX_CACHE
];
305 /* Considered adding a member to this struct to document which flags
306 * an update might raise so that ordering of the state atoms can be
307 * checked or derived at runtime. Dropped the idea in favor of having
308 * a debug mode where the state is monitored for flags which are
309 * raised that have already been tested against.
311 struct brw_tracked_state
{
312 struct brw_state_flags dirty
;
313 void (*prepare
)( struct brw_context
*brw
);
314 void (*emit
)( struct brw_context
*brw
);
317 /* Flags for brw->state.cache.
319 #define CACHE_NEW_CC_VP (1<<BRW_CC_VP)
320 #define CACHE_NEW_CC_UNIT (1<<BRW_CC_UNIT)
321 #define CACHE_NEW_WM_PROG (1<<BRW_WM_PROG)
322 #define CACHE_NEW_SAMPLER_DEFAULT_COLOR (1<<BRW_SAMPLER_DEFAULT_COLOR)
323 #define CACHE_NEW_SAMPLER (1<<BRW_SAMPLER)
324 #define CACHE_NEW_WM_UNIT (1<<BRW_WM_UNIT)
325 #define CACHE_NEW_SF_PROG (1<<BRW_SF_PROG)
326 #define CACHE_NEW_SF_VP (1<<BRW_SF_VP)
327 #define CACHE_NEW_SF_UNIT (1<<BRW_SF_UNIT)
328 #define CACHE_NEW_VS_UNIT (1<<BRW_VS_UNIT)
329 #define CACHE_NEW_VS_PROG (1<<BRW_VS_PROG)
330 #define CACHE_NEW_GS_UNIT (1<<BRW_GS_UNIT)
331 #define CACHE_NEW_GS_PROG (1<<BRW_GS_PROG)
332 #define CACHE_NEW_CLIP_VP (1<<BRW_CLIP_VP)
333 #define CACHE_NEW_CLIP_UNIT (1<<BRW_CLIP_UNIT)
334 #define CACHE_NEW_CLIP_PROG (1<<BRW_CLIP_PROG)
335 #define CACHE_NEW_SURFACE (1<<BRW_SS_SURFACE)
336 #define CACHE_NEW_SURF_BIND (1<<BRW_SS_SURF_BIND)
338 struct brw_cached_batch_item
{
339 struct header
*header
;
341 struct brw_cached_batch_item
*next
;
346 /* Protect against a future where VERT_ATTRIB_MAX > 32. Wouldn't life
347 * be easier if C allowed arrays of packed elements?
349 #define ATTRIB_BIT_DWORDS ((VERT_ATTRIB_MAX+31)/32)
351 struct brw_vertex_element
{
352 const struct gl_client_array
*glarray
;
354 /** Size of a complete element */
356 /** Number of uploaded elements for this input. */
358 /** Byte stride between elements in the uploaded array */
360 /** Offset of the first element within the buffer object */
362 /** Buffer object containing the uploaded vertex data */
368 struct brw_vertex_info
{
369 GLuint varying
; /* varying:1[VERT_ATTRIB_MAX] */
370 GLuint sizes
[ATTRIB_BIT_DWORDS
* 2]; /* sizes:2[VERT_ATTRIB_MAX] */
376 /* Cache for TNL programs.
378 struct brw_tnl_cache_item
{
382 struct brw_tnl_cache_item
*next
;
385 struct brw_tnl_cache
{
386 struct brw_tnl_cache_item
**items
;
387 GLuint size
, n_items
;
390 struct brw_query_object
{
391 struct gl_query_object Base
;
393 /** Doubly linked list of active query objects in the context. */
394 struct brw_query_object
*prev
, *next
;
396 /** Last query BO associated with this query. */
398 /** First index in bo with query data for this object. */
400 /** Last index in bo with query data for this object. */
403 /* Total count of pixels from previous BOs */
409 struct intel_context intel
;
412 GLboolean emit_state_always
;
413 GLboolean tmp_fallback
;
414 GLboolean no_batch_wrap
;
417 struct brw_state_flags dirty
;
418 struct brw_tracked_state
**atoms
;
421 GLuint nr_draw_regions
;
422 struct intel_region
*draw_regions
[MAX_DRAW_BUFFERS
];
423 struct intel_region
*depth_region
;
426 * List of buffers accumulated in brw_validate_state to receive
427 * dri_bo_check_aperture treatment before exec, so we can know if we
428 * should flush the batch and try again before emitting primitives.
430 * This can be a fixed number as we only have a limited number of
431 * objects referenced from the batchbuffer in a primitive emit,
432 * consisting of the vertex buffers, pipelined state pointers,
433 * the CURBE, the depth buffer, and a query BO.
435 dri_bo
*validated_bos
[VERT_ATTRIB_MAX
+ 16];
436 int validated_bo_count
;
439 struct brw_cache cache
;
440 struct brw_cached_batch_item
*cached_batch_items
;
443 struct brw_vertex_element inputs
[VERT_ATTRIB_MAX
];
445 #define BRW_NR_UPLOAD_BUFS 17
446 #define BRW_UPLOAD_INIT_SIZE (128*1024)
453 /* Summary of size and varying of active arrays, so we can check
454 * for changes to this state:
456 struct brw_vertex_info info
;
457 unsigned int min_index
, max_index
;
462 * Index buffer for this draw_prims call.
464 * Updates are signaled by BRW_NEW_INDICES.
466 const struct _mesa_index_buffer
*ib
;
472 /* Active vertex program:
474 const struct gl_vertex_program
*vertex_program
;
475 const struct gl_fragment_program
*fragment_program
;
478 /* For populating the gtt:
480 GLuint next_free_page
;
483 /* BRW_NEW_URB_ALLOCATIONS:
486 GLuint vsize
; /* vertex size plus header in urb registers */
487 GLuint csize
; /* constant buffer size in urb registers */
488 GLuint sfsize
; /* setup data size in urb registers */
490 GLboolean constrained
;
492 GLuint nr_vs_entries
;
493 GLuint nr_gs_entries
;
494 GLuint nr_clip_entries
;
495 GLuint nr_sf_entries
;
496 GLuint nr_cs_entries
;
498 /* GLuint vs_size; */
499 /* GLuint gs_size; */
500 /* GLuint clip_size; */
501 /* GLuint sf_size; */
502 /* GLuint cs_size; */
512 /* BRW_NEW_CURBE_OFFSETS:
523 /* Dynamic tracker which changes to reflect the state referenced
524 * by active fp and vp program parameters:
526 struct brw_tracked_state tracked_state
;
529 /** Offset within curbe_bo of space for current curbe entry */
531 /** Offset within curbe_bo of space for next curbe entry */
532 GLuint curbe_next_offset
;
537 * Whether we should create a new bo instead of reusing the old one
538 * (if we just dispatch the batch pointing at the old one.
540 GLboolean need_new_bo
;
544 struct brw_vs_prog_data
*prog_data
;
551 struct brw_gs_prog_data
*prog_data
;
553 GLboolean prog_active
;
559 struct brw_clip_prog_data
*prog_data
;
568 struct brw_sf_prog_data
*prog_data
;
576 struct brw_wm_prog_data
*prog_data
;
577 struct brw_wm_compile
*compile_data
;
579 /* Input sizes, calculated from active vertex program:
581 GLuint input_size_masks
[4];
583 /** Array of surface default colors (texture border color) */
584 dri_bo
*sdc_bo
[BRW_MAX_TEX_UNIT
];
590 dri_bo
*scratch_buffer
;
592 GLuint sampler_count
;
595 /** Binding table of pointers to surf_bo entries */
597 dri_bo
*surf_bo
[BRW_WM_MAX_SURF
];
611 struct brw_query_object active_head
;
616 /* Used to give every program string a unique id
622 #define BRW_PACKCOLOR8888(r,g,b,a) ((r<<24) | (g<<16) | (b<<8) | a)
626 /*======================================================================
629 void brwInitVtbl( struct brw_context
*brw
);
630 void brw_do_flush( struct brw_context
*brw
,
633 /*======================================================================
636 GLboolean
brwCreateContext( const __GLcontextModes
*mesaVis
,
637 __DRIcontextPrivate
*driContextPriv
,
638 void *sharedContextPrivate
);
640 /*======================================================================
643 void brw_init_queryobj_functions(struct dd_function_table
*functions
);
644 void brw_prepare_query_begin(struct brw_context
*brw
);
645 void brw_emit_query_begin(struct brw_context
*brw
);
646 void brw_emit_query_end(struct brw_context
*brw
);
648 /*======================================================================
651 void brw_debug_batch(struct intel_context
*intel
);
653 /*======================================================================
656 void brwUpdateTextureState( struct intel_context
*intel
);
657 void brw_FrameBufferTexInit( struct brw_context
*brw
,
658 struct intel_region
*region
);
659 void brw_FrameBufferTexDestroy( struct brw_context
*brw
);
660 void brw_validate_textures( struct brw_context
*brw
);
663 /*======================================================================
666 void brwInitFragProgFuncs( struct dd_function_table
*functions
);
671 void brw_upload_urb_fence(struct brw_context
*brw
);
673 void brw_upload_constant_buffer_state(struct brw_context
*brw
);
676 /*======================================================================
677 * Inline conversion functions. These are better-typed than the
678 * macros used previously:
680 static INLINE
struct brw_context
*
681 brw_context( GLcontext
*ctx
)
683 return (struct brw_context
*)ctx
;
686 #define DO_SETUP_BITS ((1<<(FRAG_ATTRIB_MAX)) - 1)