src/mesa/drivers/dri/i965/brw_context.h

   1 /*
   2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
   4  develop this 3D driver.
   5
   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:
  13
  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.
  17
  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.
  25
  26  **********************************************************************/
  27  /*
  28   * Authors:
  29   *   Keith Whitwell <keith@tungstengraphics.com>
  30   */
  31
  32
  33 #ifndef BRWCONTEXT_INC
  34 #define BRWCONTEXT_INC
  35
  36 #include "intel_context.h"
  37 #include "brw_structs.h"
  38 #include "main/imports.h"
  39
  40
  41 /* Glossary:
  42  *
  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.
  46  *
  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.
  50  *
  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.
  54  *
  55  * PUE - primitive URB entry.  An urb entry produced by the setup (SF)
  56  * unit holding rasterization and interpolation parameters.
  57  *
  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.
  64  *
  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.
  71  *
  72  * R0 - GRF register 0.  Typically holds control information used when
  73  * sending messages to other threads.
  74  *
  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.
  78  *
  79  * Fixed function units:
  80  *
  81  * CS - Command streamer.  Notional first unit, little software
  82  * interaction.  Holds the URB entries used for constant data, ie the
  83  * CURBEs.
  84  *
  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.
  90  *
  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.
  99  *
 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.
 105  *
 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.
 109  *
 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.
 113  *
 114  * CC - Color Calculator.  No EU threads associated with this unit.
 115  * Handles blending and (presumably) depth and stencil testing.
 116  */
 117
 118
 119 #define BRW_MAX_CURBE                    (32*16)
 120
 121 struct brw_context;
 122
 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
 137 /**
 138  * Used for any batch entry with a relocated pointer that will be used
 139  * by any 3D rendering.
 140  */
 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
 149
 150 struct brw_state_flags {
 151    /** State update flags signalled by mesa internals */
 152    GLuint mesa;
 153    /**
 154     * State update flags signalled as the result of brw_tracked_state updates
 155     */
 156    GLuint brw;
 157    /** State update flags signalled by brw_state_cache.c searches */
 158    GLuint cache;
 159 };
 160
 161
 162 /** Subclass of Mesa vertex program */
 163 struct brw_vertex_program {
 164    struct gl_vertex_program program;
 165    GLuint id;
 166    GLboolean use_const_buffer;
 167 };
 168
 169
 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 */
 174    GLboolean isGLSL;  /**< really, any IF/LOOP/CONT/BREAK instructions */
 175
 176    GLboolean use_const_buffer;
 177
 178    /** for debugging, which texture units are referenced */
 179    GLbitfield tex_units_used;
 180 };
 181
 182 struct brw_shader {
 183    struct gl_shader base;
 184
 185    /** Shader IR transformed for native compile, at link time. */
 186    struct exec_list *ir;
 187 };
 188
 189 struct brw_shader_program {
 190    struct gl_shader_program base;
 191 };
 192
 193 /* Data about a particular attempt to compile a program.  Note that
 194  * there can be many of these, each in a different GL state
 195  * corresponding to a different brw_wm_prog_key struct, with different
 196  * compiled programs:
 197  */
 198 struct brw_wm_prog_data {
 199    GLuint curb_read_length;
 200    GLuint urb_read_length;
 201
 202    GLuint first_curbe_grf;
 203    GLuint total_grf;
 204    GLuint total_scratch;
 205
 206    GLuint nr_params;       /**< number of float params/constants */
 207    GLboolean error;
 208
 209    /* Pointer to tracked values (only valid once
 210     * _mesa_load_state_parameters has been called at runtime).
 211     */
 212    const GLfloat *param[BRW_MAX_CURBE];
 213 };
 214
 215 struct brw_sf_prog_data {
 216    GLuint urb_read_length;
 217    GLuint total_grf;
 218
 219    /* Each vertex may have upto 12 attributes, 4 components each,
 220     * except WPOS which requires only 2.  (11*4 + 2) == 44 ==> 11
 221     * rows.
 222     *
 223     * Actually we use 4 for each, so call it 12 rows.
 224     */
 225    GLuint urb_entry_size;
 226 };
 227
 228 struct brw_clip_prog_data {
 229    GLuint curb_read_length;     /* user planes? */
 230    GLuint clip_mode;
 231    GLuint urb_read_length;
 232    GLuint total_grf;
 233 };
 234
 235 struct brw_gs_prog_data {
 236    GLuint urb_read_length;
 237    GLuint total_grf;
 238 };
 239
 240 struct brw_vs_prog_data {
 241    GLuint curb_read_length;
 242    GLuint urb_read_length;
 243    GLuint total_grf;
 244    GLbitfield64 outputs_written;
 245    GLuint nr_params;       /**< number of float params/constants */
 246
 247    GLuint inputs_read;
 248
 249    /* Used for calculating urb partitions:
 250     */
 251    GLuint urb_entry_size;
 252 };
 253
 254
 255 /* Size == 0 if output either not written, or always [0,0,0,1]
 256  */
 257 struct brw_vs_ouput_sizes {
 258    GLubyte output_size[VERT_RESULT_MAX];
 259 };
 260
 261
 262 /** Number of texture sampler units */
 263 #define BRW_MAX_TEX_UNIT 16
 264
 265 /** Max number of render targets in a shader */
 266 #define BRW_MAX_DRAW_BUFFERS 8
 267
 268 /**
 269  * Size of our surface binding table for the WM.
 270  * This contains pointers to the drawing surfaces and current texture
 271  * objects and shader constant buffers (+2).
 272  */
 273 #define BRW_WM_MAX_SURF (BRW_MAX_DRAW_BUFFERS + BRW_MAX_TEX_UNIT + 1)
 274
 275 /**
 276  * Helpers to convert drawing buffers, textures and constant buffers
 277  * to surface binding table indexes, for WM.
 278  */
 279 #define SURF_INDEX_DRAW(d)           (d)
 280 #define SURF_INDEX_FRAG_CONST_BUFFER (BRW_MAX_DRAW_BUFFERS)
 281 #define SURF_INDEX_TEXTURE(t)        (BRW_MAX_DRAW_BUFFERS + 1 + (t))
 282
 283 /**
 284  * Size of surface binding table for the VS.
 285  * Only one constant buffer for now.
 286  */
 287 #define BRW_VS_MAX_SURF 1
 288
 289 /**
 290  * Only a VS constant buffer
 291  */
 292 #define SURF_INDEX_VERT_CONST_BUFFER 0
 293
 294
 295 enum brw_cache_id {
 296    BRW_BLEND_STATE,
 297    BRW_DEPTH_STENCIL_STATE,
 298    BRW_COLOR_CALC_STATE,
 299    BRW_CC_VP,
 300    BRW_CC_UNIT,
 301    BRW_WM_PROG,
 302    BRW_SAMPLER_DEFAULT_COLOR,
 303    BRW_SAMPLER,
 304    BRW_WM_UNIT,
 305    BRW_SF_PROG,
 306    BRW_SF_VP,
 307    BRW_SF_UNIT, /* scissor state on gen6 */
 308    BRW_VS_UNIT,
 309    BRW_VS_PROG,
 310    BRW_GS_UNIT,
 311    BRW_GS_PROG,
 312    BRW_CLIP_VP,
 313    BRW_CLIP_UNIT,
 314    BRW_CLIP_PROG,
 315
 316    BRW_MAX_CACHE
 317 };
 318
 319 struct brw_cache_item {
 320    /**
 321     * Effectively part of the key, cache_id identifies what kind of state
 322     * buffer is involved, and also which brw->state.dirty.cache flag should
 323     * be set when this cache item is chosen.
 324     */
 325    enum brw_cache_id cache_id;
 326    /** 32-bit hash of the key data */
 327    GLuint hash;
 328    GLuint key_size;             /* for variable-sized keys */
 329    const void *key;
 330    drm_intel_bo **reloc_bufs;
 331    GLuint nr_reloc_bufs;
 332
 333    drm_intel_bo *bo;
 334
 335    struct brw_cache_item *next;
 336 };
 337
 338
 339
 340 struct brw_cache {
 341    struct brw_context *brw;
 342
 343    struct brw_cache_item **items;
 344    GLuint size, n_items;
 345
 346    char *name[BRW_MAX_CACHE];
 347
 348    /* Record of the last BOs chosen for each cache_id.  Used to set
 349     * brw->state.dirty.cache when a new cache item is chosen.
 350     */
 351    drm_intel_bo *last_bo[BRW_MAX_CACHE];
 352 };
 353
 354
 355 /* Considered adding a member to this struct to document which flags
 356  * an update might raise so that ordering of the state atoms can be
 357  * checked or derived at runtime.  Dropped the idea in favor of having
 358  * a debug mode where the state is monitored for flags which are
 359  * raised that have already been tested against.
 360  */
 361 struct brw_tracked_state {
 362    struct brw_state_flags dirty;
 363    void (*prepare)( struct brw_context *brw );
 364    void (*emit)( struct brw_context *brw );
 365 };
 366
 367 /* Flags for brw->state.cache.
 368  */
 369 #define CACHE_NEW_BLEND_STATE            (1<<BRW_BLEND_STATE)
 370 #define CACHE_NEW_DEPTH_STENCIL_STATE    (1<<BRW_DEPTH_STENCIL_STATE)
 371 #define CACHE_NEW_COLOR_CALC_STATE       (1<<BRW_COLOR_CALC_STATE)
 372 #define CACHE_NEW_CC_VP                  (1<<BRW_CC_VP)
 373 #define CACHE_NEW_CC_UNIT                (1<<BRW_CC_UNIT)
 374 #define CACHE_NEW_WM_PROG                (1<<BRW_WM_PROG)
 375 #define CACHE_NEW_SAMPLER_DEFAULT_COLOR  (1<<BRW_SAMPLER_DEFAULT_COLOR)
 376 #define CACHE_NEW_SAMPLER                (1<<BRW_SAMPLER)
 377 #define CACHE_NEW_WM_UNIT                (1<<BRW_WM_UNIT)
 378 #define CACHE_NEW_SF_PROG                (1<<BRW_SF_PROG)
 379 #define CACHE_NEW_SF_VP                  (1<<BRW_SF_VP)
 380 #define CACHE_NEW_SF_UNIT                (1<<BRW_SF_UNIT)
 381 #define CACHE_NEW_VS_UNIT                (1<<BRW_VS_UNIT)
 382 #define CACHE_NEW_VS_PROG                (1<<BRW_VS_PROG)
 383 #define CACHE_NEW_GS_UNIT                (1<<BRW_GS_UNIT)
 384 #define CACHE_NEW_GS_PROG                (1<<BRW_GS_PROG)
 385 #define CACHE_NEW_CLIP_VP                (1<<BRW_CLIP_VP)
 386 #define CACHE_NEW_CLIP_UNIT              (1<<BRW_CLIP_UNIT)
 387 #define CACHE_NEW_CLIP_PROG              (1<<BRW_CLIP_PROG)
 388
 389 struct brw_cached_batch_item {
 390    struct header *header;
 391    GLuint sz;
 392    struct brw_cached_batch_item *next;
 393 };
 394
 395
 396
 397 /* Protect against a future where VERT_ATTRIB_MAX > 32.  Wouldn't life
 398  * be easier if C allowed arrays of packed elements?
 399  */
 400 #define ATTRIB_BIT_DWORDS  ((VERT_ATTRIB_MAX+31)/32)
 401
 402 struct brw_vertex_element {
 403    const struct gl_client_array *glarray;
 404
 405    /** The corresponding Mesa vertex attribute */
 406    gl_vert_attrib attrib;
 407    /** Size of a complete element */
 408    GLuint element_size;
 409    /** Number of uploaded elements for this input. */
 410    GLuint count;
 411    /** Byte stride between elements in the uploaded array */
 412    GLuint stride;
 413    /** Offset of the first element within the buffer object */
 414    unsigned int offset;
 415    /** Buffer object containing the uploaded vertex data */
 416    drm_intel_bo *bo;
 417 };
 418
 419
 420
 421 struct brw_vertex_info {
 422    GLuint sizes[ATTRIB_BIT_DWORDS * 2]; /* sizes:2[VERT_ATTRIB_MAX] */
 423 };
 424
 425 struct brw_query_object {
 426    struct gl_query_object Base;
 427
 428    /** Last query BO associated with this query. */
 429    drm_intel_bo *bo;
 430    /** First index in bo with query data for this object. */
 431    int first_index;
 432    /** Last index in bo with query data for this object. */
 433    int last_index;
 434 };
 435
 436
 437 /**
 438  * brw_context is derived from intel_context.
 439  */
 440 struct brw_context
 441 {
 442    struct intel_context intel;  /**< base class, must be first field */
 443    GLuint primitive;
 444
 445    GLboolean emit_state_always;
 446    GLboolean has_surface_tile_offset;
 447    GLboolean has_compr4;
 448    GLboolean has_negative_rhw_bug;
 449    GLboolean has_aa_line_parameters;
 450    GLboolean has_pln;
 451 ;
 452    struct {
 453       struct brw_state_flags dirty;
 454
 455       GLuint nr_color_regions;
 456       struct intel_region *color_regions[MAX_DRAW_BUFFERS];
 457       struct intel_region *depth_region;
 458
 459       /**
 460        * List of buffers accumulated in brw_validate_state to receive
 461        * drm_intel_bo_check_aperture treatment before exec, so we can
 462        * know if we should flush the batch and try again before
 463        * emitting primitives.
 464        *
 465        * This can be a fixed number as we only have a limited number of
 466        * objects referenced from the batchbuffer in a primitive emit,
 467        * consisting of the vertex buffers, pipelined state pointers,
 468        * the CURBE, the depth buffer, and a query BO.
 469        */
 470       drm_intel_bo *validated_bos[VERT_ATTRIB_MAX + BRW_WM_MAX_SURF + 16];
 471       int validated_bo_count;
 472    } state;
 473
 474    struct brw_cache cache;
 475    struct brw_cached_batch_item *cached_batch_items;
 476
 477    struct {
 478       struct brw_vertex_element inputs[VERT_ATTRIB_MAX];
 479
 480       struct brw_vertex_element *enabled[VERT_ATTRIB_MAX];
 481       GLuint nr_enabled;
 482
 483 #define BRW_NR_UPLOAD_BUFS 17
 484 #define BRW_UPLOAD_INIT_SIZE (128*1024)
 485
 486       struct {
 487          drm_intel_bo *bo;
 488          GLuint offset;
 489       } upload;
 490
 491       /* Summary of size and varying of active arrays, so we can check
 492        * for changes to this state:
 493        */
 494       struct brw_vertex_info info;
 495       unsigned int min_index, max_index;
 496    } vb;
 497
 498    struct {
 499       /**
 500        * Index buffer for this draw_prims call.
 501        *
 502        * Updates are signaled by BRW_NEW_INDICES.
 503        */
 504       const struct _mesa_index_buffer *ib;
 505
 506       /* Updates to these fields are signaled by BRW_NEW_INDEX_BUFFER. */
 507       drm_intel_bo *bo;
 508       unsigned int offset;
 509       unsigned int size;
 510       /* Offset to index buffer index to use in CMD_3D_PRIM so that we can
 511        * avoid re-uploading the IB packet over and over if we're actually
 512        * referencing the same index buffer.
 513        */
 514       unsigned int start_vertex_offset;
 515    } ib;
 516
 517    /* Active vertex program:
 518     */
 519    const struct gl_vertex_program *vertex_program;
 520    const struct gl_fragment_program *fragment_program;
 521
 522
 523    /* For populating the gtt:
 524     */
 525    GLuint next_free_page;
 526
 527    /* hw-dependent 3DSTATE_VF_STATISTICS opcode */
 528    uint32_t CMD_VF_STATISTICS;
 529    /* hw-dependent 3DSTATE_PIPELINE_SELECT opcode */
 530    uint32_t CMD_PIPELINE_SELECT;
 531    int vs_max_threads;
 532    int wm_max_threads;
 533
 534    /* BRW_NEW_URB_ALLOCATIONS:
 535     */
 536    struct {
 537       GLuint vsize;             /* vertex size plus header in urb registers */
 538       GLuint csize;             /* constant buffer size in urb registers */
 539       GLuint sfsize;            /* setup data size in urb registers */
 540
 541       GLboolean constrained;
 542
 543       GLuint nr_vs_entries;
 544       GLuint nr_gs_entries;
 545       GLuint nr_clip_entries;
 546       GLuint nr_sf_entries;
 547       GLuint nr_cs_entries;
 548
 549       /* gen6 */
 550       GLuint vs_size;
 551 /*       GLuint gs_size; */
 552 /*       GLuint clip_size; */
 553 /*       GLuint sf_size; */
 554 /*       GLuint cs_size; */
 555
 556       GLuint vs_start;
 557       GLuint gs_start;
 558       GLuint clip_start;
 559       GLuint sf_start;
 560       GLuint cs_start;
 561       GLuint size; /* Hardware URB size, in KB. */
 562    } urb;
 563
 564
 565    /* BRW_NEW_CURBE_OFFSETS:
 566     */
 567    struct {
 568       GLuint wm_start;  /**< pos of first wm const in CURBE buffer */
 569       GLuint wm_size;   /**< number of float[4] consts, multiple of 16 */
 570       GLuint clip_start;
 571       GLuint clip_size;
 572       GLuint vs_start;
 573       GLuint vs_size;
 574       GLuint total_size;
 575
 576       drm_intel_bo *curbe_bo;
 577       /** Offset within curbe_bo of space for current curbe entry */
 578       GLuint curbe_offset;
 579       /** Offset within curbe_bo of space for next curbe entry */
 580       GLuint curbe_next_offset;
 581
 582       /**
 583        * Copy of the last set of CURBEs uploaded.  Frequently we'll end up
 584        * in brw_curbe.c with the same set of constant data to be uploaded,
 585        * so we'd rather not upload new constants in that case (it can cause
 586        * a pipeline bubble since only up to 4 can be pipelined at a time).
 587        */
 588       GLfloat *last_buf;
 589       /**
 590        * Allocation for where to calculate the next set of CURBEs.
 591        * It's a hot enough path that malloc/free of that data matters.
 592        */
 593       GLfloat *next_buf;
 594       GLuint last_bufsz;
 595    } curbe;
 596
 597    struct {
 598       struct brw_vs_prog_data *prog_data;
 599       int8_t *constant_map; /* variable array following prog_data */
 600
 601       drm_intel_bo *prog_bo;
 602       drm_intel_bo *state_bo;
 603       drm_intel_bo *const_bo;
 604
 605       /** Binding table of pointers to surf_bo entries */
 606       drm_intel_bo *bind_bo;
 607       uint32_t bind_bo_offset;
 608       drm_intel_bo *surf_bo[BRW_VS_MAX_SURF];
 609       uint32_t surf_offset[BRW_VS_MAX_SURF];
 610       GLuint nr_surfaces;
 611    } vs;
 612
 613    struct {
 614       struct brw_gs_prog_data *prog_data;
 615
 616       GLboolean prog_active;
 617       drm_intel_bo *prog_bo;
 618       drm_intel_bo *state_bo;
 619    } gs;
 620
 621    struct {
 622       struct brw_clip_prog_data *prog_data;
 623
 624       drm_intel_bo *prog_bo;
 625       drm_intel_bo *state_bo;
 626       drm_intel_bo *vp_bo;
 627    } clip;
 628
 629
 630    struct {
 631       struct brw_sf_prog_data *prog_data;
 632
 633       drm_intel_bo *prog_bo;
 634       drm_intel_bo *state_bo;
 635       drm_intel_bo *vp_bo;
 636    } sf;
 637
 638    struct {
 639       struct brw_wm_prog_data *prog_data;
 640       struct brw_wm_compile *compile_data;
 641
 642       /** Input sizes, calculated from active vertex program.
 643        * One bit per fragment program input attribute.
 644        */
 645       GLbitfield input_size_masks[4];
 646
 647       /** Array of surface default colors (texture border color) */
 648       drm_intel_bo *sdc_bo[BRW_MAX_TEX_UNIT];
 649
 650       GLuint render_surf;
 651       GLuint nr_surfaces;
 652
 653       GLuint max_threads;
 654       drm_intel_bo *scratch_bo;
 655
 656       GLuint sampler_count;
 657       drm_intel_bo *sampler_bo;
 658
 659       /** Binding table of pointers to surf_bo entries */
 660       drm_intel_bo *bind_bo;
 661       uint32_t bind_bo_offset;
 662       drm_intel_bo *surf_bo[BRW_WM_MAX_SURF];
 663       uint32_t surf_offset[BRW_WM_MAX_SURF];
 664
 665       drm_intel_bo *prog_bo;
 666       drm_intel_bo *state_bo;
 667       drm_intel_bo *const_bo; /* pull constant buffer. */
 668       /**
 669        *  This is the push constant BO on gen6.
 670        *
 671        * Pre-gen6, push constants live in the CURBE.
 672        */
 673       drm_intel_bo *push_const_bo;
 674    } wm;
 675
 676
 677    struct {
 678       /* gen4 */
 679       drm_intel_bo *prog_bo;
 680       drm_intel_bo *vp_bo;
 681
 682       /* gen6 */
 683       drm_intel_bo *blend_state_bo;
 684       drm_intel_bo *depth_stencil_state_bo;
 685       drm_intel_bo *color_calc_state_bo;
 686
 687       drm_intel_bo *state_bo;
 688       uint32_t state_offset;
 689    } cc;
 690
 691    struct {
 692       struct brw_query_object *obj;
 693       drm_intel_bo *bo;
 694       int index;
 695       GLboolean active;
 696    } query;
 697    /* Used to give every program string a unique id
 698     */
 699    GLuint program_id;
 700 };
 701
 702
 703 #define BRW_PACKCOLOR8888(r,g,b,a)  ((r<<24) | (g<<16) | (b<<8) | a)
 704
 705 struct brw_instruction_info {
 706     char    *name;
 707     int     nsrc;
 708     int     ndst;
 709     GLboolean is_arith;
 710 };
 711 extern const struct brw_instruction_info brw_opcodes[128];
 712
 713 /*======================================================================
 714  * brw_vtbl.c
 715  */
 716 void brwInitVtbl( struct brw_context *brw );
 717
 718 /*======================================================================
 719  * brw_context.c
 720  */
 721 GLboolean brwCreateContext( int api,
 722                             const struct gl_config *mesaVis,
 723                             __DRIcontext *driContextPriv,
 724                             void *sharedContextPrivate);
 725
 726 /*======================================================================
 727  * brw_queryobj.c
 728  */
 729 void brw_init_queryobj_functions(struct dd_function_table *functions);
 730 void brw_prepare_query_begin(struct brw_context *brw);
 731 void brw_emit_query_begin(struct brw_context *brw);
 732 void brw_emit_query_end(struct brw_context *brw);
 733
 734 /*======================================================================
 735  * brw_state_dump.c
 736  */
 737 void brw_debug_batch(struct intel_context *intel);
 738
 739 /*======================================================================
 740  * brw_tex.c
 741  */
 742 void brw_validate_textures( struct brw_context *brw );
 743
 744
 745 /*======================================================================
 746  * brw_program.c
 747  */
 748 void brwInitFragProgFuncs( struct dd_function_table *functions );
 749
 750
 751 /* brw_urb.c
 752  */
 753 void brw_upload_urb_fence(struct brw_context *brw);
 754
 755 /* brw_cc.c */
 756 void brw_update_cc_vp(struct brw_context *brw);
 757
 758 /* brw_curbe.c
 759  */
 760 void brw_upload_cs_urb_state(struct brw_context *brw);
 761
 762 /* brw_disasm.c */
 763 int brw_disasm (FILE *file, struct brw_instruction *inst, int gen);
 764
 765 /* brw_state.c */
 766 void brw_enable(struct gl_context * ctx, GLenum cap, GLboolean state);
 767 void brw_depth_range(struct gl_context *ctx, GLclampd nearval, GLclampd farval);
 768
 769 /*======================================================================
 770  * Inline conversion functions.  These are better-typed than the
 771  * macros used previously:
 772  */
 773 static INLINE struct brw_context *
 774 brw_context( struct gl_context *ctx )
 775 {
 776    return (struct brw_context *)ctx;
 777 }
 778
 779 static INLINE struct brw_vertex_program *
 780 brw_vertex_program(struct gl_vertex_program *p)
 781 {
 782    return (struct brw_vertex_program *) p;
 783 }
 784
 785 static INLINE const struct brw_vertex_program *
 786 brw_vertex_program_const(const struct gl_vertex_program *p)
 787 {
 788    return (const struct brw_vertex_program *) p;
 789 }
 790
 791 static INLINE struct brw_fragment_program *
 792 brw_fragment_program(struct gl_fragment_program *p)
 793 {
 794    return (struct brw_fragment_program *) p;
 795 }
 796
 797 static INLINE const struct brw_fragment_program *
 798 brw_fragment_program_const(const struct gl_fragment_program *p)
 799 {
 800    return (const struct brw_fragment_program *) p;
 801 }
 802
 803 GLboolean brw_do_cubemap_normalize(struct exec_list *instructions);
 804
 805 #endif
 806