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.
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_PSP 0x800
133 #define BRW_NEW_WM_SURFACES 0x1000
134 #define BRW_NEW_INDICES 0x4000
135 #define BRW_NEW_VERTICES 0x8000
137 * Used for any batch entry with a relocated pointer that will be used
138 * by any 3D rendering.
140 #define BRW_NEW_BATCH 0x10000
141 /** brw->depth_region updated */
142 #define BRW_NEW_DEPTH_BUFFER 0x20000
143 #define BRW_NEW_NR_WM_SURFACES 0x40000
144 #define BRW_NEW_NR_VS_SURFACES 0x80000
145 #define BRW_NEW_INDEX_BUFFER 0x100000
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 */
164 GLboolean use_const_buffer
;
168 /** Subclass of Mesa fragment program */
169 struct brw_fragment_program
{
170 struct gl_fragment_program program
;
171 GLuint id
; /**< serial no. to identify frag progs, never re-used */
172 GLboolean isGLSL
; /**< really, any IF/LOOP/CONT/BREAK instructions */
174 GLboolean use_const_buffer
;
175 dri_bo
*const_buffer
; /** Program constant buffer/surface */
177 /** for debugging, which texture units are referenced */
178 GLbitfield tex_units_used
;
182 /* Data about a particular attempt to compile a program. Note that
183 * there can be many of these, each in a different GL state
184 * corresponding to a different brw_wm_prog_key struct, with different
187 struct brw_wm_prog_data
{
188 GLuint curb_read_length
;
189 GLuint urb_read_length
;
191 GLuint first_curbe_grf
;
193 GLuint total_scratch
;
195 GLuint nr_params
; /**< number of float params/constants */
198 /* Pointer to tracked values (only valid once
199 * _mesa_load_state_parameters has been called at runtime).
201 const GLfloat
*param
[BRW_MAX_CURBE
];
204 struct brw_sf_prog_data
{
205 GLuint urb_read_length
;
208 /* Each vertex may have upto 12 attributes, 4 components each,
209 * except WPOS which requires only 2. (11*4 + 2) == 44 ==> 11
212 * Actually we use 4 for each, so call it 12 rows.
214 GLuint urb_entry_size
;
217 struct brw_clip_prog_data
{
218 GLuint curb_read_length
; /* user planes? */
220 GLuint urb_read_length
;
224 struct brw_gs_prog_data
{
225 GLuint urb_read_length
;
229 struct brw_vs_prog_data
{
230 GLuint curb_read_length
;
231 GLuint urb_read_length
;
233 GLbitfield64 outputs_written
;
234 GLuint nr_params
; /**< number of float params/constants */
238 /* Used for calculating urb partitions:
240 GLuint urb_entry_size
;
244 /* Size == 0 if output either not written, or always [0,0,0,1]
246 struct brw_vs_ouput_sizes
{
247 GLubyte output_size
[VERT_RESULT_MAX
];
251 /** Number of texture sampler units */
252 #define BRW_MAX_TEX_UNIT 16
254 /** Max number of render targets in a shader */
255 #define BRW_MAX_DRAW_BUFFERS 4
258 * Size of our surface binding table for the WM.
259 * This contains pointers to the drawing surfaces and current texture
260 * objects and shader constant buffers (+2).
262 #define BRW_WM_MAX_SURF (BRW_MAX_DRAW_BUFFERS + BRW_MAX_TEX_UNIT + 1)
265 * Helpers to convert drawing buffers, textures and constant buffers
266 * to surface binding table indexes, for WM.
268 #define SURF_INDEX_DRAW(d) (d)
269 #define SURF_INDEX_FRAG_CONST_BUFFER (BRW_MAX_DRAW_BUFFERS)
270 #define SURF_INDEX_TEXTURE(t) (BRW_MAX_DRAW_BUFFERS + 1 + (t))
273 * Size of surface binding table for the VS.
274 * Only one constant buffer for now.
276 #define BRW_VS_MAX_SURF 1
279 * Only a VS constant buffer
281 #define SURF_INDEX_VERT_CONST_BUFFER 0
286 BRW_DEPTH_STENCIL_STATE
,
287 BRW_COLOR_CALC_STATE
,
291 BRW_SAMPLER_DEFAULT_COLOR
,
296 BRW_SF_UNIT
, /* scissor state on gen6 */
310 struct brw_cache_item
{
312 * Effectively part of the key, cache_id identifies what kind of state
313 * buffer is involved, and also which brw->state.dirty.cache flag should
314 * be set when this cache item is chosen.
316 enum brw_cache_id cache_id
;
317 /** 32-bit hash of the key data */
319 GLuint key_size
; /* for variable-sized keys */
322 GLuint nr_reloc_bufs
;
326 struct brw_cache_item
*next
;
332 struct brw_context
*brw
;
334 struct brw_cache_item
**items
;
335 GLuint size
, n_items
;
337 char *name
[BRW_MAX_CACHE
];
339 /* Record of the last BOs chosen for each cache_id. Used to set
340 * brw->state.dirty.cache when a new cache item is chosen.
342 dri_bo
*last_bo
[BRW_MAX_CACHE
];
346 /* Considered adding a member to this struct to document which flags
347 * an update might raise so that ordering of the state atoms can be
348 * checked or derived at runtime. Dropped the idea in favor of having
349 * a debug mode where the state is monitored for flags which are
350 * raised that have already been tested against.
352 struct brw_tracked_state
{
353 struct brw_state_flags dirty
;
354 void (*prepare
)( struct brw_context
*brw
);
355 void (*emit
)( struct brw_context
*brw
);
358 /* Flags for brw->state.cache.
360 #define CACHE_NEW_BLEND_STATE (1<<BRW_BLEND_STATE)
361 #define CACHE_NEW_DEPTH_STENCIL_STATE (1<<BRW_DEPTH_STENCIL_STATE)
362 #define CACHE_NEW_COLOR_CALC_STATE (1<<BRW_COLOR_CALC_STATE)
363 #define CACHE_NEW_CC_VP (1<<BRW_CC_VP)
364 #define CACHE_NEW_CC_UNIT (1<<BRW_CC_UNIT)
365 #define CACHE_NEW_WM_PROG (1<<BRW_WM_PROG)
366 #define CACHE_NEW_SAMPLER_DEFAULT_COLOR (1<<BRW_SAMPLER_DEFAULT_COLOR)
367 #define CACHE_NEW_SAMPLER (1<<BRW_SAMPLER)
368 #define CACHE_NEW_WM_UNIT (1<<BRW_WM_UNIT)
369 #define CACHE_NEW_SF_PROG (1<<BRW_SF_PROG)
370 #define CACHE_NEW_SF_VP (1<<BRW_SF_VP)
371 #define CACHE_NEW_SF_UNIT (1<<BRW_SF_UNIT)
372 #define CACHE_NEW_VS_UNIT (1<<BRW_VS_UNIT)
373 #define CACHE_NEW_VS_PROG (1<<BRW_VS_PROG)
374 #define CACHE_NEW_GS_UNIT (1<<BRW_GS_UNIT)
375 #define CACHE_NEW_GS_PROG (1<<BRW_GS_PROG)
376 #define CACHE_NEW_CLIP_VP (1<<BRW_CLIP_VP)
377 #define CACHE_NEW_CLIP_UNIT (1<<BRW_CLIP_UNIT)
378 #define CACHE_NEW_CLIP_PROG (1<<BRW_CLIP_PROG)
379 #define CACHE_NEW_SURFACE (1<<BRW_SS_SURFACE)
380 #define CACHE_NEW_SURF_BIND (1<<BRW_SS_SURF_BIND)
382 struct brw_cached_batch_item
{
383 struct header
*header
;
385 struct brw_cached_batch_item
*next
;
390 /* Protect against a future where VERT_ATTRIB_MAX > 32. Wouldn't life
391 * be easier if C allowed arrays of packed elements?
393 #define ATTRIB_BIT_DWORDS ((VERT_ATTRIB_MAX+31)/32)
395 struct brw_vertex_element
{
396 const struct gl_client_array
*glarray
;
398 /** The corresponding Mesa vertex attribute */
399 gl_vert_attrib attrib
;
400 /** Size of a complete element */
402 /** Number of uploaded elements for this input. */
404 /** Byte stride between elements in the uploaded array */
406 /** Offset of the first element within the buffer object */
408 /** Buffer object containing the uploaded vertex data */
414 struct brw_vertex_info
{
415 GLuint sizes
[ATTRIB_BIT_DWORDS
* 2]; /* sizes:2[VERT_ATTRIB_MAX] */
418 struct brw_query_object
{
419 struct gl_query_object Base
;
421 /** Doubly linked list of active query objects in the context. */
422 struct brw_query_object
*prev
, *next
;
424 /** Last query BO associated with this query. */
426 /** First index in bo with query data for this object. */
428 /** Last index in bo with query data for this object. */
431 /* Total count of pixels from previous BOs */
437 * brw_context is derived from intel_context.
441 struct intel_context intel
; /**< base class, must be first field */
444 GLboolean emit_state_always
;
445 GLboolean has_surface_tile_offset
;
446 GLboolean has_compr4
;
447 GLboolean has_negative_rhw_bug
;
448 GLboolean has_aa_line_parameters
;
452 struct brw_state_flags dirty
;
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
; /** non-surface items */
473 struct brw_cache surface_cache
; /* surface items */
474 struct brw_cached_batch_item
*cached_batch_items
;
477 struct brw_vertex_element inputs
[VERT_ATTRIB_MAX
];
479 struct brw_vertex_element
*enabled
[VERT_ATTRIB_MAX
];
482 #define BRW_NR_UPLOAD_BUFS 17
483 #define BRW_UPLOAD_INIT_SIZE (128*1024)
490 /* Summary of size and varying of active arrays, so we can check
491 * for changes to this state:
493 struct brw_vertex_info info
;
494 unsigned int min_index
, max_index
;
499 * Index buffer for this draw_prims call.
501 * Updates are signaled by BRW_NEW_INDICES.
503 const struct _mesa_index_buffer
*ib
;
505 /* Updates to these fields are signaled by BRW_NEW_INDEX_BUFFER. */
509 /* Offset to index buffer index to use in CMD_3D_PRIM so that we can
510 * avoid re-uploading the IB packet over and over if we're actually
511 * referencing the same index buffer.
513 unsigned int start_vertex_offset
;
516 /* Active vertex program:
518 const struct gl_vertex_program
*vertex_program
;
519 const struct gl_fragment_program
*fragment_program
;
522 /* For populating the gtt:
524 GLuint next_free_page
;
526 /* hw-dependent 3DSTATE_VF_STATISTICS opcode */
527 uint32_t CMD_VF_STATISTICS
;
528 /* hw-dependent 3DSTATE_PIPELINE_SELECT opcode */
529 uint32_t CMD_PIPELINE_SELECT
;
533 /* BRW_NEW_URB_ALLOCATIONS:
536 GLuint vsize
; /* vertex size plus header in urb registers */
537 GLuint csize
; /* constant buffer size in urb registers */
538 GLuint sfsize
; /* setup data size in urb registers */
540 GLboolean constrained
;
542 GLuint nr_vs_entries
;
543 GLuint nr_gs_entries
;
544 GLuint nr_clip_entries
;
545 GLuint nr_sf_entries
;
546 GLuint nr_cs_entries
;
550 /* GLuint gs_size; */
551 /* GLuint clip_size; */
552 /* GLuint sf_size; */
553 /* GLuint cs_size; */
560 GLuint size
; /* Hardware URB size, in KB. */
564 /* BRW_NEW_CURBE_OFFSETS:
567 GLuint wm_start
; /**< pos of first wm const in CURBE buffer */
568 GLuint wm_size
; /**< number of float[4] consts, multiple of 16 */
576 /** Offset within curbe_bo of space for current curbe entry */
578 /** Offset within curbe_bo of space for next curbe entry */
579 GLuint curbe_next_offset
;
586 struct brw_vs_prog_data
*prog_data
;
587 int8_t *constant_map
; /* variable array following prog_data */
592 /** Binding table of pointers to surf_bo entries */
594 dri_bo
*surf_bo
[BRW_VS_MAX_SURF
];
599 struct brw_gs_prog_data
*prog_data
;
601 GLboolean prog_active
;
607 struct brw_clip_prog_data
*prog_data
;
616 struct brw_sf_prog_data
*prog_data
;
624 struct brw_wm_prog_data
*prog_data
;
625 struct brw_wm_compile
*compile_data
;
627 /** Input sizes, calculated from active vertex program.
628 * One bit per fragment program input attribute.
630 GLbitfield input_size_masks
[4];
632 /** Array of surface default colors (texture border color) */
633 dri_bo
*sdc_bo
[BRW_MAX_TEX_UNIT
];
641 GLuint sampler_count
;
644 /** Binding table of pointers to surf_bo entries */
646 dri_bo
*surf_bo
[BRW_WM_MAX_SURF
];
659 dri_bo
*blend_state_bo
;
660 dri_bo
*depth_stencil_state_bo
;
661 dri_bo
*color_calc_state_bo
;
667 struct brw_query_object active_head
;
672 /* Used to give every program string a unique id
678 #define BRW_PACKCOLOR8888(r,g,b,a) ((r<<24) | (g<<16) | (b<<8) | a)
682 /*======================================================================
685 void brwInitVtbl( struct brw_context
*brw
);
687 /*======================================================================
690 GLboolean
brwCreateContext( int api
,
691 const __GLcontextModes
*mesaVis
,
692 __DRIcontext
*driContextPriv
,
693 void *sharedContextPrivate
);
695 /*======================================================================
698 void brw_init_queryobj_functions(struct dd_function_table
*functions
);
699 void brw_prepare_query_begin(struct brw_context
*brw
);
700 void brw_emit_query_begin(struct brw_context
*brw
);
701 void brw_emit_query_end(struct brw_context
*brw
);
703 /*======================================================================
706 void brw_debug_batch(struct intel_context
*intel
);
708 /*======================================================================
711 void brw_validate_textures( struct brw_context
*brw
);
714 /*======================================================================
717 void brwInitFragProgFuncs( struct dd_function_table
*functions
);
722 void brw_upload_urb_fence(struct brw_context
*brw
);
726 void brw_upload_cs_urb_state(struct brw_context
*brw
);
729 int brw_disasm (FILE *file
, struct brw_instruction
*inst
);
731 /*======================================================================
732 * Inline conversion functions. These are better-typed than the
733 * macros used previously:
735 static INLINE
struct brw_context
*
736 brw_context( GLcontext
*ctx
)
738 return (struct brw_context
*)ctx
;
741 static INLINE
struct brw_vertex_program
*
742 brw_vertex_program(struct gl_vertex_program
*p
)
744 return (struct brw_vertex_program
*) p
;
747 static INLINE
const struct brw_vertex_program
*
748 brw_vertex_program_const(const struct gl_vertex_program
*p
)
750 return (const struct brw_vertex_program
*) p
;
753 static INLINE
struct brw_fragment_program
*
754 brw_fragment_program(struct gl_fragment_program
*p
)
756 return (struct brw_fragment_program
*) p
;
759 static INLINE
const struct brw_fragment_program
*
760 brw_fragment_program_const(const struct gl_fragment_program
*p
)
762 return (const struct brw_fragment_program
*) p
;