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_BINDING_TABLE 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
146 #define BRW_NEW_INDEX_BUFFER 0x100000
147 #define BRW_NEW_VS_CONSTBUF 0x200000
148 #define BRW_NEW_WM_CONSTBUF 0x200000
150 struct brw_state_flags
{
151 /** State update flags signalled by mesa internals */
154 * State update flags signalled as the result of brw_tracked_state updates
157 /** State update flags signalled by brw_state_cache.c searches */
162 /** Subclass of Mesa vertex program */
163 struct brw_vertex_program
{
164 struct gl_vertex_program program
;
166 GLboolean use_const_buffer
;
170 /** Subclass of Mesa fragment program */
171 struct brw_fragment_program
{
172 struct gl_fragment_program program
;
173 GLuint id
; /**< serial no. to identify frag progs, never re-used */
175 /** for debugging, which texture units are referenced */
176 GLbitfield tex_units_used
;
180 struct gl_shader base
;
182 /** Shader IR transformed for native compile, at link time. */
183 struct exec_list
*ir
;
186 struct brw_shader_program
{
187 struct gl_shader_program base
;
190 enum param_conversion
{
197 /* Data about a particular attempt to compile a program. Note that
198 * there can be many of these, each in a different GL state
199 * corresponding to a different brw_wm_prog_key struct, with different
202 struct brw_wm_prog_data
{
203 GLuint curb_read_length
;
204 GLuint urb_read_length
;
206 GLuint first_curbe_grf
;
208 GLuint total_scratch
;
210 GLuint nr_params
; /**< number of float params/constants */
211 GLuint nr_pull_params
;
215 /* Pointer to tracked values (only valid once
216 * _mesa_load_state_parameters has been called at runtime).
218 const float *param
[MAX_UNIFORMS
* 4]; /* should be: BRW_MAX_CURBE */
219 enum param_conversion param_convert
[MAX_UNIFORMS
* 4];
220 const float *pull_param
[MAX_UNIFORMS
* 4];
221 enum param_conversion pull_param_convert
[MAX_UNIFORMS
* 4];
224 struct brw_sf_prog_data
{
225 GLuint urb_read_length
;
228 /* Each vertex may have upto 12 attributes, 4 components each,
229 * except WPOS which requires only 2. (11*4 + 2) == 44 ==> 11
232 * Actually we use 4 for each, so call it 12 rows.
234 GLuint urb_entry_size
;
237 struct brw_clip_prog_data
{
238 GLuint curb_read_length
; /* user planes? */
240 GLuint urb_read_length
;
244 struct brw_gs_prog_data
{
245 GLuint urb_read_length
;
249 struct brw_vs_prog_data
{
250 GLuint curb_read_length
;
251 GLuint urb_read_length
;
253 GLbitfield64 outputs_written
;
254 GLuint nr_params
; /**< number of float params/constants */
258 /* Used for calculating urb partitions:
260 GLuint urb_entry_size
;
264 /* Size == 0 if output either not written, or always [0,0,0,1]
266 struct brw_vs_ouput_sizes
{
267 GLubyte output_size
[VERT_RESULT_MAX
];
271 /** Number of texture sampler units */
272 #define BRW_MAX_TEX_UNIT 16
274 /** Max number of render targets in a shader */
275 #define BRW_MAX_DRAW_BUFFERS 8
278 * Size of our surface binding table for the WM.
279 * This contains pointers to the drawing surfaces and current texture
280 * objects and shader constant buffers (+2).
282 #define BRW_WM_MAX_SURF (BRW_MAX_DRAW_BUFFERS + BRW_MAX_TEX_UNIT + 1)
285 * Helpers to convert drawing buffers, textures and constant buffers
286 * to surface binding table indexes, for WM.
288 #define SURF_INDEX_DRAW(d) (d)
289 #define SURF_INDEX_FRAG_CONST_BUFFER (BRW_MAX_DRAW_BUFFERS)
290 #define SURF_INDEX_TEXTURE(t) (BRW_MAX_DRAW_BUFFERS + 1 + (t))
293 * Size of surface binding table for the VS.
294 * Only one constant buffer for now.
296 #define BRW_VS_MAX_SURF 1
299 * Only a VS constant buffer
301 #define SURF_INDEX_VERT_CONST_BUFFER 0
306 BRW_DEPTH_STENCIL_STATE
,
307 BRW_COLOR_CALC_STATE
,
311 BRW_SAMPLER_DEFAULT_COLOR
,
316 BRW_SF_UNIT
, /* scissor state on gen6 */
328 struct brw_cache_item
{
330 * Effectively part of the key, cache_id identifies what kind of state
331 * buffer is involved, and also which brw->state.dirty.cache flag should
332 * be set when this cache item is chosen.
334 enum brw_cache_id cache_id
;
335 /** 32-bit hash of the key data */
337 GLuint key_size
; /* for variable-sized keys */
339 drm_intel_bo
**reloc_bufs
;
340 GLuint nr_reloc_bufs
;
344 struct brw_cache_item
*next
;
350 struct brw_context
*brw
;
352 struct brw_cache_item
**items
;
353 GLuint size
, n_items
;
355 char *name
[BRW_MAX_CACHE
];
357 /* Record of the last BOs chosen for each cache_id. Used to set
358 * brw->state.dirty.cache when a new cache item is chosen.
360 drm_intel_bo
*last_bo
[BRW_MAX_CACHE
];
364 /* Considered adding a member to this struct to document which flags
365 * an update might raise so that ordering of the state atoms can be
366 * checked or derived at runtime. Dropped the idea in favor of having
367 * a debug mode where the state is monitored for flags which are
368 * raised that have already been tested against.
370 struct brw_tracked_state
{
371 struct brw_state_flags dirty
;
372 void (*prepare
)( struct brw_context
*brw
);
373 void (*emit
)( struct brw_context
*brw
);
376 /* Flags for brw->state.cache.
378 #define CACHE_NEW_BLEND_STATE (1<<BRW_BLEND_STATE)
379 #define CACHE_NEW_DEPTH_STENCIL_STATE (1<<BRW_DEPTH_STENCIL_STATE)
380 #define CACHE_NEW_COLOR_CALC_STATE (1<<BRW_COLOR_CALC_STATE)
381 #define CACHE_NEW_CC_VP (1<<BRW_CC_VP)
382 #define CACHE_NEW_CC_UNIT (1<<BRW_CC_UNIT)
383 #define CACHE_NEW_WM_PROG (1<<BRW_WM_PROG)
384 #define CACHE_NEW_SAMPLER_DEFAULT_COLOR (1<<BRW_SAMPLER_DEFAULT_COLOR)
385 #define CACHE_NEW_SAMPLER (1<<BRW_SAMPLER)
386 #define CACHE_NEW_WM_UNIT (1<<BRW_WM_UNIT)
387 #define CACHE_NEW_SF_PROG (1<<BRW_SF_PROG)
388 #define CACHE_NEW_SF_VP (1<<BRW_SF_VP)
389 #define CACHE_NEW_SF_UNIT (1<<BRW_SF_UNIT)
390 #define CACHE_NEW_VS_UNIT (1<<BRW_VS_UNIT)
391 #define CACHE_NEW_VS_PROG (1<<BRW_VS_PROG)
392 #define CACHE_NEW_GS_UNIT (1<<BRW_GS_UNIT)
393 #define CACHE_NEW_GS_PROG (1<<BRW_GS_PROG)
394 #define CACHE_NEW_CLIP_VP (1<<BRW_CLIP_VP)
395 #define CACHE_NEW_CLIP_UNIT (1<<BRW_CLIP_UNIT)
396 #define CACHE_NEW_CLIP_PROG (1<<BRW_CLIP_PROG)
398 struct brw_cached_batch_item
{
399 struct header
*header
;
401 struct brw_cached_batch_item
*next
;
406 /* Protect against a future where VERT_ATTRIB_MAX > 32. Wouldn't life
407 * be easier if C allowed arrays of packed elements?
409 #define ATTRIB_BIT_DWORDS ((VERT_ATTRIB_MAX+31)/32)
411 struct brw_vertex_element
{
412 const struct gl_client_array
*glarray
;
414 /** The corresponding Mesa vertex attribute */
415 gl_vert_attrib attrib
;
416 /** Size of a complete element */
418 /** Number of uploaded elements for this input. */
420 /** Byte stride between elements in the uploaded array */
422 /** Offset of the first element within the buffer object */
424 /** Buffer object containing the uploaded vertex data */
430 struct brw_vertex_info
{
431 GLuint sizes
[ATTRIB_BIT_DWORDS
* 2]; /* sizes:2[VERT_ATTRIB_MAX] */
434 struct brw_query_object
{
435 struct gl_query_object Base
;
437 /** Last query BO associated with this query. */
439 /** First index in bo with query data for this object. */
441 /** Last index in bo with query data for this object. */
447 * brw_context is derived from intel_context.
451 struct intel_context intel
; /**< base class, must be first field */
454 GLboolean emit_state_always
;
455 GLboolean has_surface_tile_offset
;
456 GLboolean has_compr4
;
457 GLboolean has_negative_rhw_bug
;
458 GLboolean has_aa_line_parameters
;
462 struct brw_state_flags dirty
;
464 GLuint nr_color_regions
;
465 struct intel_region
*color_regions
[MAX_DRAW_BUFFERS
];
466 struct intel_region
*depth_region
;
469 * List of buffers accumulated in brw_validate_state to receive
470 * drm_intel_bo_check_aperture treatment before exec, so we can
471 * know if we should flush the batch and try again before
472 * emitting primitives.
474 * This can be a fixed number as we only have a limited number of
475 * objects referenced from the batchbuffer in a primitive emit,
476 * consisting of the vertex buffers, pipelined state pointers,
477 * the CURBE, the depth buffer, and a query BO.
479 drm_intel_bo
*validated_bos
[VERT_ATTRIB_MAX
+ BRW_WM_MAX_SURF
+ 16];
480 int validated_bo_count
;
483 struct brw_cache cache
;
484 struct brw_cached_batch_item
*cached_batch_items
;
487 struct brw_vertex_element inputs
[VERT_ATTRIB_MAX
];
489 struct brw_vertex_element
*enabled
[VERT_ATTRIB_MAX
];
492 #define BRW_NR_UPLOAD_BUFS 17
493 #define BRW_UPLOAD_INIT_SIZE (128*1024)
500 /* Summary of size and varying of active arrays, so we can check
501 * for changes to this state:
503 struct brw_vertex_info info
;
504 unsigned int min_index
, max_index
;
509 * Index buffer for this draw_prims call.
511 * Updates are signaled by BRW_NEW_INDICES.
513 const struct _mesa_index_buffer
*ib
;
515 /* Updates to these fields are signaled by BRW_NEW_INDEX_BUFFER. */
519 /* Offset to index buffer index to use in CMD_3D_PRIM so that we can
520 * avoid re-uploading the IB packet over and over if we're actually
521 * referencing the same index buffer.
523 unsigned int start_vertex_offset
;
526 /* Active vertex program:
528 const struct gl_vertex_program
*vertex_program
;
529 const struct gl_fragment_program
*fragment_program
;
532 /* For populating the gtt:
534 GLuint next_free_page
;
536 /* hw-dependent 3DSTATE_VF_STATISTICS opcode */
537 uint32_t CMD_VF_STATISTICS
;
538 /* hw-dependent 3DSTATE_PIPELINE_SELECT opcode */
539 uint32_t CMD_PIPELINE_SELECT
;
543 /* BRW_NEW_URB_ALLOCATIONS:
546 GLuint vsize
; /* vertex size plus header in urb registers */
547 GLuint csize
; /* constant buffer size in urb registers */
548 GLuint sfsize
; /* setup data size in urb registers */
550 GLboolean constrained
;
552 GLuint nr_vs_entries
;
553 GLuint nr_gs_entries
;
554 GLuint nr_clip_entries
;
555 GLuint nr_sf_entries
;
556 GLuint nr_cs_entries
;
560 /* GLuint gs_size; */
561 /* GLuint clip_size; */
562 /* GLuint sf_size; */
563 /* GLuint cs_size; */
570 GLuint size
; /* Hardware URB size, in KB. */
574 /* BRW_NEW_CURBE_OFFSETS:
577 GLuint wm_start
; /**< pos of first wm const in CURBE buffer */
578 GLuint wm_size
; /**< number of float[4] consts, multiple of 16 */
585 drm_intel_bo
*curbe_bo
;
586 /** Offset within curbe_bo of space for current curbe entry */
588 /** Offset within curbe_bo of space for next curbe entry */
589 GLuint curbe_next_offset
;
592 * Copy of the last set of CURBEs uploaded. Frequently we'll end up
593 * in brw_curbe.c with the same set of constant data to be uploaded,
594 * so we'd rather not upload new constants in that case (it can cause
595 * a pipeline bubble since only up to 4 can be pipelined at a time).
599 * Allocation for where to calculate the next set of CURBEs.
600 * It's a hot enough path that malloc/free of that data matters.
607 struct brw_vs_prog_data
*prog_data
;
608 int8_t *constant_map
; /* variable array following prog_data */
610 drm_intel_bo
*prog_bo
;
611 drm_intel_bo
*state_bo
;
612 drm_intel_bo
*const_bo
;
614 /** Binding table of pointers to surf_bo entries */
615 drm_intel_bo
*bind_bo
;
616 uint32_t bind_bo_offset
;
617 drm_intel_bo
*surf_bo
[BRW_VS_MAX_SURF
];
618 uint32_t surf_offset
[BRW_VS_MAX_SURF
];
623 struct brw_gs_prog_data
*prog_data
;
625 GLboolean prog_active
;
626 drm_intel_bo
*prog_bo
;
627 drm_intel_bo
*state_bo
;
631 struct brw_clip_prog_data
*prog_data
;
633 drm_intel_bo
*prog_bo
;
634 drm_intel_bo
*state_bo
;
640 struct brw_sf_prog_data
*prog_data
;
642 drm_intel_bo
*prog_bo
;
643 drm_intel_bo
*state_bo
;
648 struct brw_wm_prog_data
*prog_data
;
649 struct brw_wm_compile
*compile_data
;
651 /** Input sizes, calculated from active vertex program.
652 * One bit per fragment program input attribute.
654 GLbitfield input_size_masks
[4];
656 /** Array of surface default colors (texture border color) */
657 drm_intel_bo
*sdc_bo
[BRW_MAX_TEX_UNIT
];
663 drm_intel_bo
*scratch_bo
;
665 GLuint sampler_count
;
666 drm_intel_bo
*sampler_bo
;
668 /** Binding table of pointers to surf_bo entries */
669 drm_intel_bo
*bind_bo
;
670 uint32_t bind_bo_offset
;
671 drm_intel_bo
*surf_bo
[BRW_WM_MAX_SURF
];
672 uint32_t surf_offset
[BRW_WM_MAX_SURF
];
674 drm_intel_bo
*prog_bo
;
675 drm_intel_bo
*state_bo
;
676 drm_intel_bo
*const_bo
; /* pull constant buffer. */
678 * This is the push constant BO on gen6.
680 * Pre-gen6, push constants live in the CURBE.
682 drm_intel_bo
*push_const_bo
;
688 drm_intel_bo
*prog_bo
;
692 drm_intel_bo
*blend_state_bo
;
693 drm_intel_bo
*depth_stencil_state_bo
;
694 drm_intel_bo
*color_calc_state_bo
;
696 drm_intel_bo
*state_bo
;
697 uint32_t state_offset
;
701 struct brw_query_object
*obj
;
706 /* Used to give every program string a unique id
712 #define BRW_PACKCOLOR8888(r,g,b,a) ((r<<24) | (g<<16) | (b<<8) | a)
714 struct brw_instruction_info
{
720 extern const struct brw_instruction_info brw_opcodes
[128];
722 /*======================================================================
725 void brwInitVtbl( struct brw_context
*brw
);
727 /*======================================================================
730 GLboolean
brwCreateContext( int api
,
731 const struct gl_config
*mesaVis
,
732 __DRIcontext
*driContextPriv
,
733 void *sharedContextPrivate
);
735 /*======================================================================
738 void brw_init_queryobj_functions(struct dd_function_table
*functions
);
739 void brw_prepare_query_begin(struct brw_context
*brw
);
740 void brw_emit_query_begin(struct brw_context
*brw
);
741 void brw_emit_query_end(struct brw_context
*brw
);
743 /*======================================================================
746 void brw_debug_batch(struct intel_context
*intel
);
748 /*======================================================================
751 void brw_validate_textures( struct brw_context
*brw
);
754 /*======================================================================
757 void brwInitFragProgFuncs( struct dd_function_table
*functions
);
762 void brw_upload_urb_fence(struct brw_context
*brw
);
765 void brw_update_cc_vp(struct brw_context
*brw
);
769 void brw_upload_cs_urb_state(struct brw_context
*brw
);
772 int brw_disasm (FILE *file
, struct brw_instruction
*inst
, int gen
);
775 void brw_enable(struct gl_context
* ctx
, GLenum cap
, GLboolean state
);
776 void brw_depth_range(struct gl_context
*ctx
, GLclampd nearval
, GLclampd farval
);
778 /*======================================================================
779 * Inline conversion functions. These are better-typed than the
780 * macros used previously:
782 static INLINE
struct brw_context
*
783 brw_context( struct gl_context
*ctx
)
785 return (struct brw_context
*)ctx
;
788 static INLINE
struct brw_vertex_program
*
789 brw_vertex_program(struct gl_vertex_program
*p
)
791 return (struct brw_vertex_program
*) p
;
794 static INLINE
const struct brw_vertex_program
*
795 brw_vertex_program_const(const struct gl_vertex_program
*p
)
797 return (const struct brw_vertex_program
*) p
;
800 static INLINE
struct brw_fragment_program
*
801 brw_fragment_program(struct gl_fragment_program
*p
)
803 return (struct brw_fragment_program
*) p
;
806 static INLINE
const struct brw_fragment_program
*
807 brw_fragment_program_const(const struct gl_fragment_program
*p
)
809 return (const struct brw_fragment_program
*) p
;
813 float convert_param(enum param_conversion conversion
, float param
)
821 switch (conversion
) {
822 case PARAM_NO_CONVERT
:
824 case PARAM_CONVERT_F2I
:
827 case PARAM_CONVERT_F2U
:
830 case PARAM_CONVERT_F2B
:
841 GLboolean
brw_do_cubemap_normalize(struct exec_list
*instructions
);