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)
122 struct brw_instruction
;
123 struct brw_vs_prog_key
;
124 struct brw_wm_prog_key
;
125 struct brw_wm_prog_data
;
129 BRW_STATE_FRAGMENT_PROGRAM
,
130 BRW_STATE_VERTEX_PROGRAM
,
131 BRW_STATE_INPUT_DIMENSIONS
,
132 BRW_STATE_CURBE_OFFSETS
,
133 BRW_STATE_REDUCED_PRIMITIVE
,
136 BRW_STATE_WM_INPUT_DIMENSIONS
,
139 BRW_STATE_VS_BINDING_TABLE
,
140 BRW_STATE_GS_BINDING_TABLE
,
141 BRW_STATE_PS_BINDING_TABLE
,
145 BRW_STATE_NR_WM_SURFACES
,
146 BRW_STATE_NR_VS_SURFACES
,
147 BRW_STATE_INDEX_BUFFER
,
148 BRW_STATE_VS_CONSTBUF
,
149 BRW_STATE_PROGRAM_CACHE
,
150 BRW_STATE_STATE_BASE_ADDRESS
,
151 BRW_STATE_SOL_INDICES
,
154 #define BRW_NEW_URB_FENCE (1 << BRW_STATE_URB_FENCE)
155 #define BRW_NEW_FRAGMENT_PROGRAM (1 << BRW_STATE_FRAGMENT_PROGRAM)
156 #define BRW_NEW_VERTEX_PROGRAM (1 << BRW_STATE_VERTEX_PROGRAM)
157 #define BRW_NEW_INPUT_DIMENSIONS (1 << BRW_STATE_INPUT_DIMENSIONS)
158 #define BRW_NEW_CURBE_OFFSETS (1 << BRW_STATE_CURBE_OFFSETS)
159 #define BRW_NEW_REDUCED_PRIMITIVE (1 << BRW_STATE_REDUCED_PRIMITIVE)
160 #define BRW_NEW_PRIMITIVE (1 << BRW_STATE_PRIMITIVE)
161 #define BRW_NEW_CONTEXT (1 << BRW_STATE_CONTEXT)
162 #define BRW_NEW_WM_INPUT_DIMENSIONS (1 << BRW_STATE_WM_INPUT_DIMENSIONS)
163 #define BRW_NEW_PSP (1 << BRW_STATE_PSP)
164 #define BRW_NEW_SURFACES (1 << BRW_STATE_SURFACES)
165 #define BRW_NEW_VS_BINDING_TABLE (1 << BRW_STATE_VS_BINDING_TABLE)
166 #define BRW_NEW_GS_BINDING_TABLE (1 << BRW_STATE_GS_BINDING_TABLE)
167 #define BRW_NEW_PS_BINDING_TABLE (1 << BRW_STATE_PS_BINDING_TABLE)
168 #define BRW_NEW_INDICES (1 << BRW_STATE_INDICES)
169 #define BRW_NEW_VERTICES (1 << BRW_STATE_VERTICES)
171 * Used for any batch entry with a relocated pointer that will be used
172 * by any 3D rendering.
174 #define BRW_NEW_BATCH (1 << BRW_STATE_BATCH)
175 /** \see brw.state.depth_region */
176 #define BRW_NEW_INDEX_BUFFER (1 << BRW_STATE_INDEX_BUFFER)
177 #define BRW_NEW_VS_CONSTBUF (1 << BRW_STATE_VS_CONSTBUF)
178 #define BRW_NEW_PROGRAM_CACHE (1 << BRW_STATE_PROGRAM_CACHE)
179 #define BRW_NEW_STATE_BASE_ADDRESS (1 << BRW_STATE_STATE_BASE_ADDRESS)
180 #define BRW_NEW_SOL_INDICES (1 << BRW_STATE_SOL_INDICES)
182 struct brw_state_flags
{
183 /** State update flags signalled by mesa internals */
186 * State update flags signalled as the result of brw_tracked_state updates
189 /** State update flags signalled by brw_state_cache.c searches */
193 enum state_struct_type
{
194 AUB_TRACE_VS_STATE
= 1,
195 AUB_TRACE_GS_STATE
= 2,
196 AUB_TRACE_CLIP_STATE
= 3,
197 AUB_TRACE_SF_STATE
= 4,
198 AUB_TRACE_WM_STATE
= 5,
199 AUB_TRACE_CC_STATE
= 6,
200 AUB_TRACE_CLIP_VP_STATE
= 7,
201 AUB_TRACE_SF_VP_STATE
= 8,
202 AUB_TRACE_CC_VP_STATE
= 0x9,
203 AUB_TRACE_SAMPLER_STATE
= 0xa,
204 AUB_TRACE_KERNEL_INSTRUCTIONS
= 0xb,
205 AUB_TRACE_SCRATCH_SPACE
= 0xc,
206 AUB_TRACE_SAMPLER_DEFAULT_COLOR
= 0xd,
208 AUB_TRACE_SCISSOR_STATE
= 0x15,
209 AUB_TRACE_BLEND_STATE
= 0x16,
210 AUB_TRACE_DEPTH_STENCIL_STATE
= 0x17,
212 /* Not written to .aub files the same way the structures above are. */
213 AUB_TRACE_NO_TYPE
= 0x100,
214 AUB_TRACE_BINDING_TABLE
= 0x101,
215 AUB_TRACE_SURFACE_STATE
= 0x102,
216 AUB_TRACE_VS_CONSTANTS
= 0x103,
217 AUB_TRACE_WM_CONSTANTS
= 0x104,
220 /** Subclass of Mesa vertex program */
221 struct brw_vertex_program
{
222 struct gl_vertex_program program
;
224 bool use_const_buffer
;
228 /** Subclass of Mesa fragment program */
229 struct brw_fragment_program
{
230 struct gl_fragment_program program
;
231 GLuint id
; /**< serial no. to identify frag progs, never re-used */
235 struct gl_shader base
;
237 /** Shader IR transformed for native compile, at link time. */
238 struct exec_list
*ir
;
241 struct brw_shader_program
{
242 struct gl_shader_program base
;
245 enum param_conversion
{
253 /* Data about a particular attempt to compile a program. Note that
254 * there can be many of these, each in a different GL state
255 * corresponding to a different brw_wm_prog_key struct, with different
258 struct brw_wm_prog_data
{
259 GLuint curb_read_length
;
260 GLuint urb_read_length
;
262 GLuint first_curbe_grf
;
263 GLuint first_curbe_grf_16
;
265 GLuint reg_blocks_16
;
266 GLuint total_scratch
;
268 GLuint nr_params
; /**< number of float params/constants */
269 GLuint nr_pull_params
;
272 uint32_t prog_offset_16
;
274 /* Pointer to tracked values (only valid once
275 * _mesa_load_state_parameters has been called at runtime).
277 const float *param
[MAX_UNIFORMS
* 4]; /* should be: BRW_MAX_CURBE */
278 enum param_conversion param_convert
[MAX_UNIFORMS
* 4];
279 const float *pull_param
[MAX_UNIFORMS
* 4];
280 enum param_conversion pull_param_convert
[MAX_UNIFORMS
* 4];
284 * Enum representing the i965-specific vertex results that don't correspond
285 * exactly to any element of gl_vert_result. The values of this enum are
286 * assigned such that they don't conflict with gl_vert_result.
290 BRW_VERT_RESULT_NDC
= VERT_RESULT_MAX
,
291 BRW_VERT_RESULT_HPOS_DUPLICATE
,
298 * Data structure recording the relationship between the gl_vert_result enum
299 * and "slots" within the vertex URB entry (VUE). A "slot" is defined as a
300 * single octaword within the VUE (128 bits).
302 * Note that each BRW register contains 256 bits (2 octawords), so when
303 * accessing the VUE in URB_NOSWIZZLE mode, each register corresponds to two
304 * consecutive VUE slots. When accessing the VUE in URB_INTERLEAVED mode (as
305 * in a vertex shader), each register corresponds to a single VUE slot, since
306 * it contains data for two separate vertices.
310 * Map from gl_vert_result value to VUE slot. For gl_vert_results that are
311 * not stored in a slot (because they are not written, or because
312 * additional processing is applied before storing them in the VUE), the
315 int vert_result_to_slot
[BRW_VERT_RESULT_MAX
];
318 * Map from VUE slot to gl_vert_result value. For slots that do not
319 * directly correspond to a gl_vert_result, the value comes from
322 * For slots that are not in use, the value is BRW_VERT_RESULT_MAX (this
323 * simplifies code that uses the value stored in slot_to_vert_result to
324 * create a bit mask).
326 int slot_to_vert_result
[BRW_VERT_RESULT_MAX
];
329 * Total number of VUE slots in use
335 * Convert a VUE slot number into a byte offset within the VUE.
337 static inline GLuint
brw_vue_slot_to_offset(GLuint slot
)
343 * Convert a vert_result into a byte offset within the VUE.
345 static inline GLuint
brw_vert_result_to_offset(struct brw_vue_map
*vue_map
,
348 return brw_vue_slot_to_offset(vue_map
->vert_result_to_slot
[vert_result
]);
352 struct brw_sf_prog_data
{
353 GLuint urb_read_length
;
356 /* Each vertex may have upto 12 attributes, 4 components each,
357 * except WPOS which requires only 2. (11*4 + 2) == 44 ==> 11
360 * Actually we use 4 for each, so call it 12 rows.
362 GLuint urb_entry_size
;
365 struct brw_clip_prog_data
{
366 GLuint curb_read_length
; /* user planes? */
368 GLuint urb_read_length
;
372 struct brw_gs_prog_data
{
373 GLuint urb_read_length
;
377 * Gen6 transform feedback: Amount by which the streaming vertex buffer
378 * indices should be incremented each time the GS is invoked.
380 unsigned svbi_postincrement_value
;
383 struct brw_vs_prog_data
{
384 struct brw_vue_map vue_map
;
386 GLuint curb_read_length
;
387 GLuint urb_read_length
;
389 GLbitfield64 outputs_written
;
390 GLuint nr_params
; /**< number of float params/constants */
391 GLuint nr_pull_params
; /**< number of dwords referenced by pull_param[] */
392 GLuint total_scratch
;
394 GLbitfield64 inputs_read
;
396 /* Used for calculating urb partitions:
398 GLuint urb_entry_size
;
400 const float *param
[MAX_UNIFORMS
* 4]; /* should be: BRW_MAX_CURBE */
401 const float *pull_param
[MAX_UNIFORMS
* 4];
403 bool uses_new_param_layout
;
409 /* Size == 0 if output either not written, or always [0,0,0,1]
411 struct brw_vs_ouput_sizes
{
412 GLubyte output_size
[VERT_RESULT_MAX
];
416 /** Number of texture sampler units */
417 #define BRW_MAX_TEX_UNIT 16
419 /** Max number of render targets in a shader */
420 #define BRW_MAX_DRAW_BUFFERS 8
423 * Max number of binding table entries used for stream output.
425 * From the OpenGL 3.0 spec, table 6.44 (Transform Feedback State), the
426 * minimum value of MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS is 64.
428 * On Gen6, the size of transform feedback data is limited not by the number
429 * of components but by the number of binding table entries we set aside. We
430 * use one binding table entry for a float, one entry for a vector, and one
431 * entry per matrix column. Since the only way we can communicate our
432 * transform feedback capabilities to the client is via
433 * MAX_TRANSFORM_FEEDBACK_INTERLEAVED_COMPONENTS, we need to plan for the
434 * worst case, in which all the varyings are floats, so we use up one binding
435 * table entry per component. Therefore we need to set aside at least 64
436 * binding table entries for use by transform feedback.
438 * Note: since we don't currently pack varyings, it is currently impossible
439 * for the client to actually use up all of these binding table entries--if
440 * all of their varyings were floats, they would run out of varying slots and
441 * fail to link. But that's a bug, so it seems prudent to go ahead and
442 * allocate the number of binding table entries we will need once the bug is
445 #define BRW_MAX_SOL_BINDINGS 64
447 /** Maximum number of actual buffers used for stream output */
448 #define BRW_MAX_SOL_BUFFERS 4
451 * Helpers to create Surface Binding Table indexes for draw buffers,
452 * textures, and constant buffers.
454 * Shader threads access surfaces via numeric handles, rather than directly
455 * using pointers. The binding table maps these numeric handles to the
456 * address of the actual buffer.
458 * For example, a shader might ask to sample from "surface 7." In this case,
459 * bind[7] would contain a pointer to a texture.
461 * Although the hardware supports separate binding tables per pipeline stage
462 * (VS, HS, DS, GS, PS), we currently share a single binding table for all of
463 * them. This is purely for convenience.
465 * Currently our binding tables are (arbitrarily) programmed as follows:
467 * +-------------------------------+
468 * | 0 | Draw buffer 0 | .
470 * | : | : | > Only relevant to the WM.
471 * | 7 | Draw buffer 7 | /
472 * |-----|-------------------------| `
473 * | 8 | VS Pull Constant Buffer |
474 * | 9 | WM Pull Constant Buffer |
475 * |-----|-------------------------|
479 * | 25 | Texture 15 |
480 * +-----|-------------------------+
481 * | 26 | SOL Binding 0 |
484 * | 89 | SOL Binding 63 |
485 * +-------------------------------+
487 * Note that nothing actually uses the SURF_INDEX_DRAW macro, so it has to be
488 * the identity function or things will break. We do want to keep draw buffers
489 * first so we can use headerless render target writes for RT 0.
491 #define SURF_INDEX_DRAW(d) (d)
492 #define SURF_INDEX_VERT_CONST_BUFFER (BRW_MAX_DRAW_BUFFERS + 0)
493 #define SURF_INDEX_FRAG_CONST_BUFFER (BRW_MAX_DRAW_BUFFERS + 1)
494 #define SURF_INDEX_TEXTURE(t) (BRW_MAX_DRAW_BUFFERS + 2 + (t))
495 #define SURF_INDEX_SOL_BINDING(t) (SURF_INDEX_TEXTURE(BRW_MAX_TEX_UNIT) + (t))
497 /** Maximum size of the binding table. */
498 #define BRW_MAX_SURFACES SURF_INDEX_SOL_BINDING(BRW_MAX_SOL_BINDINGS)
502 BRW_DEPTH_STENCIL_STATE
,
503 BRW_COLOR_CALC_STATE
,
511 BRW_SF_UNIT
, /* scissor state on gen6 */
523 struct brw_cache_item
{
525 * Effectively part of the key, cache_id identifies what kind of state
526 * buffer is involved, and also which brw->state.dirty.cache flag should
527 * be set when this cache item is chosen.
529 enum brw_cache_id cache_id
;
530 /** 32-bit hash of the key data */
532 GLuint key_size
; /* for variable-sized keys */
539 struct brw_cache_item
*next
;
545 struct brw_context
*brw
;
547 struct brw_cache_item
**items
;
549 GLuint size
, n_items
;
551 uint32_t next_offset
;
556 /* Considered adding a member to this struct to document which flags
557 * an update might raise so that ordering of the state atoms can be
558 * checked or derived at runtime. Dropped the idea in favor of having
559 * a debug mode where the state is monitored for flags which are
560 * raised that have already been tested against.
562 struct brw_tracked_state
{
563 struct brw_state_flags dirty
;
564 void (*emit
)( struct brw_context
*brw
);
567 /* Flags for brw->state.cache.
569 #define CACHE_NEW_BLEND_STATE (1<<BRW_BLEND_STATE)
570 #define CACHE_NEW_DEPTH_STENCIL_STATE (1<<BRW_DEPTH_STENCIL_STATE)
571 #define CACHE_NEW_COLOR_CALC_STATE (1<<BRW_COLOR_CALC_STATE)
572 #define CACHE_NEW_CC_VP (1<<BRW_CC_VP)
573 #define CACHE_NEW_CC_UNIT (1<<BRW_CC_UNIT)
574 #define CACHE_NEW_WM_PROG (1<<BRW_WM_PROG)
575 #define CACHE_NEW_SAMPLER (1<<BRW_SAMPLER)
576 #define CACHE_NEW_WM_UNIT (1<<BRW_WM_UNIT)
577 #define CACHE_NEW_SF_PROG (1<<BRW_SF_PROG)
578 #define CACHE_NEW_SF_VP (1<<BRW_SF_VP)
579 #define CACHE_NEW_SF_UNIT (1<<BRW_SF_UNIT)
580 #define CACHE_NEW_VS_UNIT (1<<BRW_VS_UNIT)
581 #define CACHE_NEW_VS_PROG (1<<BRW_VS_PROG)
582 #define CACHE_NEW_GS_UNIT (1<<BRW_GS_UNIT)
583 #define CACHE_NEW_GS_PROG (1<<BRW_GS_PROG)
584 #define CACHE_NEW_CLIP_VP (1<<BRW_CLIP_VP)
585 #define CACHE_NEW_CLIP_UNIT (1<<BRW_CLIP_UNIT)
586 #define CACHE_NEW_CLIP_PROG (1<<BRW_CLIP_PROG)
588 struct brw_cached_batch_item
{
589 struct header
*header
;
591 struct brw_cached_batch_item
*next
;
596 /* Protect against a future where VERT_ATTRIB_MAX > 32. Wouldn't life
597 * be easier if C allowed arrays of packed elements?
599 #define ATTRIB_BIT_DWORDS ((VERT_ATTRIB_MAX+31)/32)
601 struct brw_vertex_buffer
{
602 /** Buffer object containing the uploaded vertex data */
605 /** Byte stride between elements in the uploaded array */
608 struct brw_vertex_element
{
609 const struct gl_client_array
*glarray
;
613 /** The corresponding Mesa vertex attribute */
614 gl_vert_attrib attrib
;
615 /** Size of a complete element */
617 /** Offset of the first element within the buffer object */
623 struct brw_vertex_info
{
624 GLuint sizes
[ATTRIB_BIT_DWORDS
* 2]; /* sizes:2[VERT_ATTRIB_MAX] */
627 struct brw_query_object
{
628 struct gl_query_object Base
;
630 /** Last query BO associated with this query. */
632 /** First index in bo with query data for this object. */
634 /** Last index in bo with query data for this object. */
640 * brw_context is derived from intel_context.
644 struct intel_context intel
; /**< base class, must be first field */
645 GLuint primitive
; /**< Hardware primitive, such as _3DPRIM_TRILIST. */
647 bool emit_state_always
;
648 bool has_surface_tile_offset
;
650 bool has_negative_rhw_bug
;
651 bool has_aa_line_parameters
;
656 struct brw_state_flags dirty
;
659 struct brw_cache cache
;
660 struct brw_cached_batch_item
*cached_batch_items
;
663 struct brw_vertex_element inputs
[VERT_ATTRIB_MAX
];
664 struct brw_vertex_buffer buffers
[VERT_ATTRIB_MAX
];
669 } current_buffers
[VERT_ATTRIB_MAX
];
671 struct brw_vertex_element
*enabled
[VERT_ATTRIB_MAX
];
673 GLuint nr_buffers
, nr_current_buffers
;
675 /* Summary of size and varying of active arrays, so we can check
676 * for changes to this state:
678 struct brw_vertex_info info
;
679 unsigned int min_index
, max_index
;
681 /* Offset from start of vertex buffer so we can avoid redefining
682 * the same VB packed over and over again.
684 unsigned int start_vertex_bias
;
689 * Index buffer for this draw_prims call.
691 * Updates are signaled by BRW_NEW_INDICES.
693 const struct _mesa_index_buffer
*ib
;
695 /* Updates are signaled by BRW_NEW_INDEX_BUFFER. */
699 /* Offset to index buffer index to use in CMD_3D_PRIM so that we can
700 * avoid re-uploading the IB packet over and over if we're actually
701 * referencing the same index buffer.
703 unsigned int start_vertex_offset
;
706 /* Active vertex program:
708 const struct gl_vertex_program
*vertex_program
;
709 const struct gl_fragment_program
*fragment_program
;
711 /* hw-dependent 3DSTATE_VF_STATISTICS opcode */
712 uint32_t CMD_VF_STATISTICS
;
713 /* hw-dependent 3DSTATE_PIPELINE_SELECT opcode */
714 uint32_t CMD_PIPELINE_SELECT
;
717 * Platform specific constants containing the maximum number of threads
718 * for each pipeline stage.
724 /* BRW_NEW_URB_ALLOCATIONS:
727 GLuint vsize
; /* vertex size plus header in urb registers */
728 GLuint csize
; /* constant buffer size in urb registers */
729 GLuint sfsize
; /* setup data size in urb registers */
733 GLuint max_vs_entries
; /* Maximum number of VS entries */
734 GLuint max_gs_entries
; /* Maximum number of GS entries */
736 GLuint nr_vs_entries
;
737 GLuint nr_gs_entries
;
738 GLuint nr_clip_entries
;
739 GLuint nr_sf_entries
;
740 GLuint nr_cs_entries
;
743 * The length of each URB entry owned by the VS (or GS), as
744 * a number of 1024-bit (128-byte) rows. Should be >= 1.
746 * gen7: Same meaning, but in 512-bit (64-byte) rows.
756 GLuint size
; /* Hardware URB size, in KB. */
758 /* gen6: True if the most recently sent _3DSTATE_URB message allocated
759 * URB space for the GS.
761 bool gen6_gs_previously_active
;
765 /* BRW_NEW_CURBE_OFFSETS:
768 GLuint wm_start
; /**< pos of first wm const in CURBE buffer */
769 GLuint wm_size
; /**< number of float[4] consts, multiple of 16 */
776 drm_intel_bo
*curbe_bo
;
777 /** Offset within curbe_bo of space for current curbe entry */
779 /** Offset within curbe_bo of space for next curbe entry */
780 GLuint curbe_next_offset
;
783 * Copy of the last set of CURBEs uploaded. Frequently we'll end up
784 * in brw_curbe.c with the same set of constant data to be uploaded,
785 * so we'd rather not upload new constants in that case (it can cause
786 * a pipeline bubble since only up to 4 can be pipelined at a time).
790 * Allocation for where to calculate the next set of CURBEs.
791 * It's a hot enough path that malloc/free of that data matters.
798 /** Binding table of pointers to surf_bo entries */
800 uint32_t surf_offset
[BRW_MAX_SURFACES
];
803 /** SAMPLER_STATE count and offset */
810 struct brw_vs_prog_data
*prog_data
;
811 int8_t *constant_map
; /* variable array following prog_data */
813 drm_intel_bo
*scratch_bo
;
814 drm_intel_bo
*const_bo
;
815 /** Offset in the program cache to the VS program */
816 uint32_t prog_offset
;
817 uint32_t state_offset
;
819 uint32_t push_const_offset
; /* Offset in the batchbuffer */
820 int push_const_size
; /* in 256-bit register increments */
822 /** @{ register allocator */
824 struct ra_regs
*regs
;
827 * Array of the ra classes for the unaligned contiguous register
833 * Mapping for register-allocated objects in *regs to the first
834 * GRF for that object.
836 uint8_t *ra_reg_to_grf
;
841 struct brw_gs_prog_data
*prog_data
;
844 /** Offset in the program cache to the CLIP program pre-gen6 */
845 uint32_t prog_offset
;
846 uint32_t state_offset
;
850 struct brw_clip_prog_data
*prog_data
;
852 /** Offset in the program cache to the CLIP program pre-gen6 */
853 uint32_t prog_offset
;
855 /* Offset in the batch to the CLIP state on pre-gen6. */
856 uint32_t state_offset
;
858 /* As of gen6, this is the offset in the batch to the CLIP VP,
866 struct brw_sf_prog_data
*prog_data
;
868 /** Offset in the program cache to the CLIP program pre-gen6 */
869 uint32_t prog_offset
;
870 uint32_t state_offset
;
875 struct brw_wm_prog_data
*prog_data
;
876 struct brw_wm_compile
*compile_data
;
878 /** Input sizes, calculated from active vertex program.
879 * One bit per fragment program input attribute.
881 GLbitfield input_size_masks
[4];
883 /** offsets in the batch to sampler default colors (texture border color)
885 uint32_t sdc_offset
[BRW_MAX_TEX_UNIT
];
889 drm_intel_bo
*scratch_bo
;
891 /** Offset in the program cache to the WM program */
892 uint32_t prog_offset
;
894 uint32_t state_offset
; /* offset in batchbuffer to pre-gen6 WM state */
896 drm_intel_bo
*const_bo
; /* pull constant buffer. */
898 * This is offset in the batch to the push constants on gen6.
900 * Pre-gen6, push constants live in the CURBE.
902 uint32_t push_const_offset
;
904 /** @{ register allocator */
906 struct ra_regs
*regs
;
908 /** Array of the ra classes for the unaligned contiguous
909 * register block sizes used.
914 * Mapping for register-allocated objects in *regs to the first
915 * GRF for that object.
917 uint8_t *ra_reg_to_grf
;
920 * ra class for the aligned pairs we use for PLN, which doesn't
921 * appear in *classes.
923 int aligned_pairs_class
;
930 uint32_t state_offset
;
931 uint32_t blend_state_offset
;
932 uint32_t depth_stencil_state_offset
;
937 struct brw_query_object
*obj
;
942 /* Used to give every program string a unique id
947 const struct brw_tracked_state
**atoms
;
949 /* If (INTEL_DEBUG & DEBUG_BATCH) */
953 enum state_struct_type type
;
955 int state_batch_count
;
958 * \brief State needed to execute HiZ ops.
960 * \see gen6_hiz_init()
961 * \see gen6_hiz_exec()
963 struct brw_hiz_state
{
964 /** \brief VBO for rectangle primitive.
966 * Rather than using glGenBuffers(), we allocate the VBO directly
969 drm_intel_bo
*vertex_bo
;
972 struct brw_sol_state
{
973 uint32_t svbi_0_starting_index
;
974 uint32_t svbi_0_max_index
;
975 uint32_t offset_0_batch_start
;
976 uint32_t primitives_generated
;
977 uint32_t primitives_written
;
980 uint32_t render_target_format
[MESA_FORMAT_COUNT
];
981 bool format_supported_as_render_target
[MESA_FORMAT_COUNT
];
986 #define BRW_PACKCOLOR8888(r,g,b,a) ((r<<24) | (g<<16) | (b<<8) | a)
988 struct brw_instruction_info
{
994 extern const struct brw_instruction_info brw_opcodes
[128];
996 /*======================================================================
999 void brwInitVtbl( struct brw_context
*brw
);
1001 /*======================================================================
1004 bool brwCreateContext(int api
,
1005 const struct gl_config
*mesaVis
,
1006 __DRIcontext
*driContextPriv
,
1007 void *sharedContextPrivate
);
1009 /*======================================================================
1012 void brw_init_queryobj_functions(struct dd_function_table
*functions
);
1013 void brw_prepare_query_begin(struct brw_context
*brw
);
1014 void brw_emit_query_begin(struct brw_context
*brw
);
1015 void brw_emit_query_end(struct brw_context
*brw
);
1017 /*======================================================================
1020 void brw_debug_batch(struct intel_context
*intel
);
1022 /*======================================================================
1025 void brw_validate_textures( struct brw_context
*brw
);
1028 /*======================================================================
1031 void brwInitFragProgFuncs( struct dd_function_table
*functions
);
1033 int brw_get_scratch_size(int size
);
1034 void brw_get_scratch_bo(struct intel_context
*intel
,
1035 drm_intel_bo
**scratch_bo
, int size
);
1040 void brw_upload_urb_fence(struct brw_context
*brw
);
1044 void brw_upload_cs_urb_state(struct brw_context
*brw
);
1047 int brw_disasm (FILE *file
, struct brw_instruction
*inst
, int gen
);
1050 gl_clip_plane
*brw_select_clip_planes(struct gl_context
*ctx
);
1054 brw_compute_barycentric_interp_modes(bool shade_model_flat
,
1055 const struct gl_fragment_program
*fprog
);
1057 /* brw_wm_surface_state.c */
1058 void brw_init_surface_formats(struct brw_context
*brw
);
1060 brw_update_sol_surface(struct brw_context
*brw
,
1061 struct gl_buffer_object
*buffer_obj
,
1062 uint32_t *out_offset
, unsigned num_vector_components
,
1063 unsigned stride_dwords
, unsigned offset_dwords
);
1065 /* gen6_clip_state.c */
1067 brw_fprog_uses_noperspective(const struct gl_fragment_program
*fprog
);
1071 brw_begin_transform_feedback(struct gl_context
*ctx
, GLenum mode
,
1072 struct gl_transform_feedback_object
*obj
);
1074 brw_end_transform_feedback(struct gl_context
*ctx
,
1075 struct gl_transform_feedback_object
*obj
);
1077 /* gen7_sol_state.c */
1079 gen7_end_transform_feedback(struct gl_context
*ctx
,
1080 struct gl_transform_feedback_object
*obj
);
1084 /*======================================================================
1085 * Inline conversion functions. These are better-typed than the
1086 * macros used previously:
1088 static INLINE
struct brw_context
*
1089 brw_context( struct gl_context
*ctx
)
1091 return (struct brw_context
*)ctx
;
1094 static INLINE
struct brw_vertex_program
*
1095 brw_vertex_program(struct gl_vertex_program
*p
)
1097 return (struct brw_vertex_program
*) p
;
1100 static INLINE
const struct brw_vertex_program
*
1101 brw_vertex_program_const(const struct gl_vertex_program
*p
)
1103 return (const struct brw_vertex_program
*) p
;
1106 static INLINE
struct brw_fragment_program
*
1107 brw_fragment_program(struct gl_fragment_program
*p
)
1109 return (struct brw_fragment_program
*) p
;
1112 static INLINE
const struct brw_fragment_program
*
1113 brw_fragment_program_const(const struct gl_fragment_program
*p
)
1115 return (const struct brw_fragment_program
*) p
;
1119 float convert_param(enum param_conversion conversion
, const float *param
)
1127 switch (conversion
) {
1128 case PARAM_NO_CONVERT
:
1130 case PARAM_CONVERT_F2I
:
1133 case PARAM_CONVERT_F2U
:
1136 case PARAM_CONVERT_F2B
:
1142 case PARAM_CONVERT_ZERO
:
1150 * Pre-gen6, the register file of the EUs was shared between threads,
1151 * and each thread used some subset allocated on a 16-register block
1152 * granularity. The unit states wanted these block counts.
1155 brw_register_blocks(int reg_count
)
1157 return ALIGN(reg_count
, 16) / 16 - 1;
1160 static inline uint32_t
1161 brw_program_reloc(struct brw_context
*brw
, uint32_t state_offset
,
1162 uint32_t prog_offset
)
1164 struct intel_context
*intel
= &brw
->intel
;
1166 if (intel
->gen
>= 5) {
1167 /* Using state base address. */
1171 drm_intel_bo_emit_reloc(intel
->batch
.bo
,
1175 I915_GEM_DOMAIN_INSTRUCTION
, 0);
1177 return brw
->cache
.bo
->offset
+ prog_offset
;
1180 bool brw_do_cubemap_normalize(struct exec_list
*instructions
);