src/mesa/drivers/dri/r300/r300_context.h

   1 /*
   2 Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
   3
   4 The Weather Channel (TM) funded Tungsten Graphics to develop the
   5 initial release of the Radeon 8500 driver under the XFree86 license.
   6 This notice must be preserved.
   7
   8 Permission is hereby granted, free of charge, to any person obtaining
   9 a copy of this software and associated documentation files (the
  10 "Software"), to deal in the Software without restriction, including
  11 without limitation the rights to use, copy, modify, merge, publish,
  12 distribute, sublicense, and/or sell copies of the Software, and to
  13 permit persons to whom the Software is furnished to do so, subject to
  14 the following conditions:
  15
  16 The above copyright notice and this permission notice (including the
  17 next paragraph) shall be included in all copies or substantial
  18 portions of the Software.
  19
  20 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  21 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  22 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  23 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  24 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  25 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  26 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  27
  28 **************************************************************************/
  29
  30 /*
  31  * Authors:
  32  *   Keith Whitwell <keith@tungstengraphics.com>
  33  *   Nicolai Haehnle <prefect_@gmx.net>
  34  */
  35
  36 #ifndef __R300_CONTEXT_H__
  37 #define __R300_CONTEXT_H__
  38
  39 #include "tnl/t_vertex.h"
  40 #include "drm.h"
  41 #include "radeon_drm.h"
  42 #include "dri_util.h"
  43 #include "texmem.h"
  44
  45 #include "macros.h"
  46 #include "mtypes.h"
  47 #include "colormac.h"
  48 #include "radeon_context.h"
  49
  50 #define USER_BUFFERS
  51 /* KW: Disable this code.  Driver should hook into vbo module
  52  * directly, see i965 driver for example.
  53  */
  54 /* #define RADEON_VTXFMT_A */
  55 #define HW_VBOS
  56
  57 /* We don't handle 16 bits elts swapping yet */
  58 #ifdef MESA_BIG_ENDIAN
  59 #define FORCE_32BITS_ELTS
  60 #endif
  61
  62 //#define OPTIMIZE_ELTS
  63
  64 struct r300_context;
  65 typedef struct r300_context r300ContextRec;
  66 typedef struct r300_context *r300ContextPtr;
  67
  68 #include "radeon_lock.h"
  69 #include "mm.h"
  70
  71 /* Checkpoint.. for convenience */
  72 #define CPT     { fprintf(stderr, "%s:%s line %d\n", __FILE__, __FUNCTION__, __LINE__); }
  73 /* From http://gcc.gnu.org/onlinedocs/gcc-3.2.3/gcc/Variadic-Macros.html .
  74    I suppose we could inline this and use macro to fetch out __LINE__ and stuff in case we run into trouble
  75    with other compilers ... GLUE!
  76 */
  77 #if 1
  78 #define WARN_ONCE(a, ...)       { \
  79         static int warn##__LINE__=1; \
  80         if(warn##__LINE__){ \
  81                 fprintf(stderr, "*********************************WARN_ONCE*********************************\n"); \
  82                 fprintf(stderr, "File %s function %s line %d\n", \
  83                         __FILE__, __FUNCTION__, __LINE__); \
  84                 fprintf(stderr,  a, ## __VA_ARGS__);\
  85                 fprintf(stderr, "***************************************************************************\n"); \
  86                 warn##__LINE__=0;\
  87                 } \
  88         }
  89 #else
  90 #define WARN_ONCE(a, ...) {}
  91 #endif
  92
  93   /* We should probably change types within vertex_shader
  94       and pixel_shader structure later on */
  95 #define CARD32 GLuint
  96 #include "vertex_shader.h"
  97 #include "r300_fragprog.h"
  98 #undef CARD32
  99
 100 static __inline__ uint32_t r300PackFloat32(float fl)
 101 {
 102         union { float fl; uint32_t u; } u;
 103
 104         u.fl = fl;
 105         return u.u;
 106 }
 107
 108
 109 /************ DMA BUFFERS **************/
 110
 111 /* Need refcounting on dma buffers:
 112  */
 113 struct r300_dma_buffer {
 114         int refcount;           /* the number of retained regions in buf */
 115         drmBufPtr buf;
 116         int id;
 117 };
 118 #undef GET_START
 119 #ifdef USER_BUFFERS
 120 #define GET_START(rvb) (r300GartOffsetFromVirtual(rmesa, (rvb)->address+(rvb)->start))
 121 #else
 122 #define GET_START(rvb) (rmesa->radeon.radeonScreen->gart_buffer_offset +                \
 123                         (rvb)->address - rmesa->dma.buf0_address +      \
 124                         (rvb)->start)
 125 #endif
 126 /* A retained region, eg vertices for indexed vertices.
 127  */
 128 struct r300_dma_region {
 129         struct r300_dma_buffer *buf;
 130         char *address;          /* == buf->address */
 131         int start, end, ptr;    /* offsets from start of buf */
 132
 133     int aos_offset;     /* address in GART memory */
 134     int aos_stride;     /* distance between elements, in dwords */
 135     int aos_size;       /* number of components (1-4) */
 136     int aos_reg;        /* VAP register assignment */
 137 };
 138
 139 struct r300_dma {
 140         /* Active dma region.  Allocations for vertices and retained
 141          * regions come from here.  Also used for emitting random vertices,
 142          * these may be flushed by calling flush_current();
 143          */
 144         struct r300_dma_region current;
 145
 146         void (*flush) (r300ContextPtr);
 147
 148         char *buf0_address;     /* start of buf[0], for index calcs */
 149
 150         /* Number of "in-flight" DMA buffers, i.e. the number of buffers
 151          * for which a DISCARD command is currently queued in the command buffer.
 152          */
 153         GLuint nr_released_bufs;
 154 };
 155
 156        /* Texture related */
 157
 158 typedef struct r300_tex_obj r300TexObj, *r300TexObjPtr;
 159
 160 /* Texture object in locally shared texture space.
 161  */
 162 struct r300_tex_obj {
 163         driTextureObject base;
 164
 165         GLuint bufAddr;         /* Offset to start of locally
 166                                    shared texture block */
 167
 168         GLuint dirty_state;     /* Flags (1 per texunit) for
 169                                    whether or not this texobj
 170                                    has dirty hardware state
 171                                    (pp_*) that needs to be
 172                                    brought into the
 173                                    texunit. */
 174
 175         drm_radeon_tex_image_t image[6][RADEON_MAX_TEXTURE_LEVELS];
 176         /* Six, for the cube faces */
 177
 178
 179         GLuint pitch; /* this isn't sent to hardware just used in calculations */
 180         /* hardware register values */
 181         /* Note that R200 has 8 registers per texture and R300 only 7 */
 182         GLuint filter;
 183         GLuint filter_1;
 184         GLuint pitch_reg;
 185         GLuint size;    /* npot only */
 186         GLuint format;
 187         GLuint offset;  /* Image location in the card's address space.
 188                                    All cube faces follow. */
 189         GLuint unknown4;
 190         GLuint unknown5;
 191         /* end hardware registers */
 192
 193         /* registers computed by r200 code - keep them here to
 194            compare against what is actually written.
 195
 196            to be removed later.. */
 197         GLuint pp_border_color;
 198         GLuint pp_cubic_faces;  /* cube face 1,2,3,4 log2 sizes */
 199         GLuint format_x;
 200
 201
 202         GLboolean border_fallback;
 203
 204         GLuint tile_bits; /* hw texture tile bits used on this texture */
 205 };
 206
 207 struct r300_texture_env_state {
 208         r300TexObjPtr texobj;
 209         GLenum format;
 210         GLenum envMode;
 211 };
 212
 213
 214 /* The blit width for texture uploads
 215  */
 216 #define R300_BLIT_WIDTH_BYTES 1024
 217 #define R300_MAX_TEXTURE_UNITS 8
 218
 219 struct r300_texture_state {
 220         struct r300_texture_env_state unit[R300_MAX_TEXTURE_UNITS];
 221         int tc_count; /* number of incoming texture coordinates from VAP */
 222 };
 223
 224 /**
 225  * A block of hardware state.
 226  *
 227  * When check returns non-zero, the returned number of dwords must be
 228  * copied verbatim into the command buffer in order to update a state atom
 229  * when it is dirty.
 230  */
 231 struct r300_state_atom {
 232         struct r300_state_atom *next, *prev;
 233         const char* name;       /* for debug */
 234         int cmd_size;           /* maximum size in dwords */
 235         GLuint idx;             /* index in an array (e.g. textures) */
 236         uint32_t* cmd;
 237         GLboolean dirty;
 238
 239         int (*check)(r300ContextPtr, struct r300_state_atom* atom);
 240 };
 241
 242
 243 #define R300_VPT_CMD_0          0
 244 #define R300_VPT_XSCALE         1
 245 #define R300_VPT_XOFFSET        2
 246 #define R300_VPT_YSCALE         3
 247 #define R300_VPT_YOFFSET        4
 248 #define R300_VPT_ZSCALE         5
 249 #define R300_VPT_ZOFFSET        6
 250 #define R300_VPT_CMDSIZE        7
 251
 252 #define R300_VIR_CMD_0          0 /* vir is variable size (at least 1) */
 253 #define R300_VIR_CNTL_0         1
 254 #define R300_VIR_CNTL_1         2
 255 #define R300_VIR_CNTL_2         3
 256 #define R300_VIR_CNTL_3         4
 257 #define R300_VIR_CNTL_4         5
 258 #define R300_VIR_CNTL_5         6
 259 #define R300_VIR_CNTL_6         7
 260 #define R300_VIR_CNTL_7         8
 261 #define R300_VIR_CMDSIZE        9
 262
 263 #define R300_VIC_CMD_0          0
 264 #define R300_VIC_CNTL_0         1
 265 #define R300_VIC_CNTL_1         2
 266 #define R300_VIC_CMDSIZE        3
 267
 268 #define R300_VOF_CMD_0          0
 269 #define R300_VOF_CNTL_0         1
 270 #define R300_VOF_CNTL_1         2
 271 #define R300_VOF_CMDSIZE        3
 272
 273
 274 #define R300_PVS_CMD_0          0
 275 #define R300_PVS_CNTL_1         1
 276 #define R300_PVS_CNTL_2         2
 277 #define R300_PVS_CNTL_3         3
 278 #define R300_PVS_CMDSIZE        4
 279
 280 #define R300_GB_MISC_CMD_0              0
 281 #define R300_GB_MISC_MSPOS_0            1
 282 #define R300_GB_MISC_MSPOS_1            2
 283 #define R300_GB_MISC_TILE_CONFIG        3
 284 #define R300_GB_MISC_SELECT             4
 285 #define R300_GB_MISC_AA_CONFIG          5
 286 #define R300_GB_MISC_CMDSIZE            6
 287
 288 #define R300_TXE_CMD_0          0
 289 #define R300_TXE_ENABLE         1
 290 #define R300_TXE_CMDSIZE        2
 291
 292 #define R300_PS_CMD_0           0
 293 #define R300_PS_POINTSIZE       1
 294 #define R300_PS_CMDSIZE         2
 295
 296 #define R300_ZBS_CMD_0          0
 297 #define R300_ZBS_T_FACTOR       1
 298 #define R300_ZBS_T_CONSTANT     2
 299 #define R300_ZBS_W_FACTOR       3
 300 #define R300_ZBS_W_CONSTANT     4
 301 #define R300_ZBS_CMDSIZE        5
 302
 303 #define R300_CUL_CMD_0          0
 304 #define R300_CUL_CULL           1
 305 #define R300_CUL_CMDSIZE        2
 306
 307 #define R300_RC_CMD_0           0
 308 #define R300_RC_CNTL_0          1
 309 #define R300_RC_CNTL_1          2
 310 #define R300_RC_CMDSIZE         3
 311
 312 #define R300_RI_CMD_0           0
 313 #define R300_RI_INTERP_0        1
 314 #define R300_RI_INTERP_1        2
 315 #define R300_RI_INTERP_2        3
 316 #define R300_RI_INTERP_3        4
 317 #define R300_RI_INTERP_4        5
 318 #define R300_RI_INTERP_5        6
 319 #define R300_RI_INTERP_6        7
 320 #define R300_RI_INTERP_7        8
 321 #define R300_RI_CMDSIZE         9
 322
 323 #define R300_RR_CMD_0           0 /* rr is variable size (at least 1) */
 324 #define R300_RR_ROUTE_0         1
 325 #define R300_RR_ROUTE_1         2
 326 #define R300_RR_ROUTE_2         3
 327 #define R300_RR_ROUTE_3         4
 328 #define R300_RR_ROUTE_4         5
 329 #define R300_RR_ROUTE_5         6
 330 #define R300_RR_ROUTE_6         7
 331 #define R300_RR_ROUTE_7         8
 332 #define R300_RR_CMDSIZE         9
 333
 334 #define R300_FP_CMD_0           0
 335 #define R300_FP_CNTL0           1
 336 #define R300_FP_CNTL1           2
 337 #define R300_FP_CNTL2           3
 338 #define R300_FP_CMD_1           4
 339 #define R300_FP_NODE0           5
 340 #define R300_FP_NODE1           6
 341 #define R300_FP_NODE2           7
 342 #define R300_FP_NODE3           8
 343 #define R300_FP_CMDSIZE         9
 344
 345 #define R300_FPT_CMD_0          0
 346 #define R300_FPT_INSTR_0        1
 347 #define R300_FPT_CMDSIZE        65
 348
 349 #define R300_FPI_CMD_0          0
 350 #define R300_FPI_INSTR_0        1
 351 #define R300_FPI_CMDSIZE        65
 352
 353 #define R300_FPP_CMD_0          0
 354 #define R300_FPP_PARAM_0        1
 355 #define R300_FPP_CMDSIZE        (32*4+1)
 356
 357 #define R300_FOGS_CMD_0         0
 358 #define R300_FOGS_STATE         1
 359 #define R300_FOGS_CMDSIZE       2
 360
 361 #define R300_FOGC_CMD_0         0
 362 #define R300_FOGC_R             1
 363 #define R300_FOGC_G             2
 364 #define R300_FOGC_B             3
 365 #define R300_FOGC_CMDSIZE       4
 366
 367 #define R300_FOGP_CMD_0         0
 368 #define R300_FOGP_SCALE         1
 369 #define R300_FOGP_START         2
 370 #define R300_FOGP_CMDSIZE       3
 371
 372 #define R300_AT_CMD_0           0
 373 #define R300_AT_ALPHA_TEST      1
 374 #define R300_AT_UNKNOWN         2
 375 #define R300_AT_CMDSIZE         3
 376
 377 #define R300_BLD_CMD_0          0
 378 #define R300_BLD_CBLEND         1
 379 #define R300_BLD_ABLEND         2
 380 #define R300_BLD_CMDSIZE        3
 381
 382 #define R300_CMK_CMD_0          0
 383 #define R300_CMK_COLORMASK      1
 384 #define R300_CMK_CMDSIZE        2
 385
 386 #define R300_CB_CMD_0           0
 387 #define R300_CB_OFFSET          1
 388 #define R300_CB_CMD_1           2
 389 #define R300_CB_PITCH           3
 390 #define R300_CB_CMDSIZE         4
 391
 392 #define R300_ZS_CMD_0           0
 393 #define R300_ZS_CNTL_0          1
 394 #define R300_ZS_CNTL_1          2
 395 #define R300_ZS_CNTL_2          3
 396 #define R300_ZS_CMDSIZE         4
 397
 398 #define R300_ZB_CMD_0           0
 399 #define R300_ZB_OFFSET          1
 400 #define R300_ZB_PITCH           2
 401 #define R300_ZB_CMDSIZE         3
 402
 403 #define R300_VPI_CMD_0          0
 404 #define R300_VPI_INSTR_0        1
 405 #define R300_VPI_CMDSIZE        1025 /* 256 16 byte instructions */
 406
 407 #define R300_VPP_CMD_0          0
 408 #define R300_VPP_PARAM_0        1
 409 #define R300_VPP_CMDSIZE        1025 /* 256 4-component parameters */
 410
 411 #define R300_VPS_CMD_0          0
 412 #define R300_VPS_ZERO_0         1
 413 #define R300_VPS_ZERO_1         2
 414 #define R300_VPS_POINTSIZE      3
 415 #define R300_VPS_ZERO_3         4
 416 #define R300_VPS_CMDSIZE        5
 417
 418         /* the layout is common for all fields inside tex */
 419 #define R300_TEX_CMD_0          0
 420 #define R300_TEX_VALUE_0        1
 421 /* We don't really use this, instead specify mtu+1 dynamically
 422 #define R300_TEX_CMDSIZE        (MAX_TEXTURE_UNITS+1)
 423 */
 424
 425 /**
 426  * Cache for hardware register state.
 427  */
 428 struct r300_hw_state {
 429         struct r300_state_atom atomlist;
 430
 431         GLboolean       is_dirty;
 432         GLboolean       all_dirty;
 433         int             max_state_size; /* in dwords */
 434
 435         struct r300_state_atom vpt;     /* viewport (1D98) */
 436         struct r300_state_atom vap_cntl;
 437         struct r300_state_atom vof;     /* VAP output format register 0x2090 */
 438         struct r300_state_atom vte;     /* (20B0) */
 439         struct r300_state_atom unk2134; /* (2134) */
 440         struct r300_state_atom vap_cntl_status;
 441         struct r300_state_atom vir[2];  /* vap input route (2150/21E0) */
 442         struct r300_state_atom vic;     /* vap input control (2180) */
 443         struct r300_state_atom unk21DC; /* (21DC) */
 444         struct r300_state_atom unk221C; /* (221C) */
 445         struct r300_state_atom unk2220; /* (2220) */
 446         struct r300_state_atom unk2288; /* (2288) */
 447         struct r300_state_atom pvs;     /* pvs_cntl (22D0) */
 448         struct r300_state_atom gb_enable; /* (4008) */
 449         struct r300_state_atom gb_misc; /* Multisampling position shifts ? (4010) */
 450         struct r300_state_atom unk4200; /* (4200) */
 451         struct r300_state_atom unk4214; /* (4214) */
 452         struct r300_state_atom ps;      /* pointsize (421C) */
 453         struct r300_state_atom unk4230; /* (4230) */
 454         struct r300_state_atom lcntl;   /* line control */
 455         struct r300_state_atom unk4260; /* (4260) */
 456         struct r300_state_atom shade;
 457         struct r300_state_atom polygon_mode;
 458         struct r300_state_atom fogp;    /* fog parameters (4294) */
 459         struct r300_state_atom unk429C; /* (429C) */
 460         struct r300_state_atom zbias_cntl;
 461         struct r300_state_atom zbs;     /* zbias (42A4) */
 462         struct r300_state_atom occlusion_cntl;
 463         struct r300_state_atom cul;     /* cull cntl (42B8) */
 464         struct r300_state_atom unk42C0; /* (42C0) */
 465         struct r300_state_atom rc;      /* rs control (4300) */
 466         struct r300_state_atom ri;      /* rs interpolators (4310) */
 467         struct r300_state_atom rr;      /* rs route (4330) */
 468         struct r300_state_atom unk43A4; /* (43A4) */
 469         struct r300_state_atom unk43E8; /* (43E8) */
 470         struct r300_state_atom fp;      /* fragment program cntl + nodes (4600) */
 471         struct r300_state_atom fpt;     /* texi - (4620) */
 472         struct r300_state_atom unk46A4; /* (46A4) */
 473         struct r300_state_atom fpi[4];  /* fp instructions (46C0/47C0/48C0/49C0) */
 474         struct r300_state_atom fogs;    /* fog state (4BC0) */
 475         struct r300_state_atom fogc;    /* fog color (4BC8) */
 476         struct r300_state_atom at;      /* alpha test (4BD4) */
 477         struct r300_state_atom unk4BD8; /* (4BD8) */
 478         struct r300_state_atom fpp;     /* 0x4C00 and following */
 479         struct r300_state_atom unk4E00; /* (4E00) */
 480         struct r300_state_atom bld;     /* blending (4E04) */
 481         struct r300_state_atom cmk;     /* colormask (4E0C) */
 482         struct r300_state_atom blend_color;     /* constant blend color */
 483         struct r300_state_atom cb;      /* colorbuffer (4E28) */
 484         struct r300_state_atom unk4E50; /* (4E50) */
 485         struct r300_state_atom unk4E88; /* (4E88) */
 486         struct r300_state_atom unk4EA0; /* (4E88) I saw it only written on RV350 hardware..  */
 487         struct r300_state_atom zs;      /* zstencil control (4F00) */
 488         struct r300_state_atom zstencil_format;
 489         struct r300_state_atom zb;      /* z buffer (4F20) */
 490         struct r300_state_atom unk4F28; /* (4F28) */
 491         struct r300_state_atom unk4F30; /* (4F30) */
 492         struct r300_state_atom unk4F44; /* (4F44) */
 493         struct r300_state_atom unk4F54; /* (4F54) */
 494
 495         struct r300_state_atom vpi;     /* vp instructions */
 496         struct r300_state_atom vpp;     /* vp parameters */
 497         struct r300_state_atom vps;     /* vertex point size (?) */
 498                 /* 8 texture units */
 499                 /* the state is grouped by function and not by
 500                    texture unit. This makes single unit updates
 501                    really awkward - we are much better off
 502                    updating the whole thing at once */
 503         struct {
 504                 struct r300_state_atom filter;
 505                 struct r300_state_atom filter_1;
 506                 struct r300_state_atom size;
 507                 struct r300_state_atom format;
 508                 struct r300_state_atom pitch;
 509                 struct r300_state_atom offset;
 510                 struct r300_state_atom chroma_key;
 511                 struct r300_state_atom border_color;
 512                 } tex;
 513         struct r300_state_atom txe;     /* tex enable (4104) */
 514 };
 515
 516
 517 /**
 518  * This structure holds the command buffer while it is being constructed.
 519  *
 520  * The first batch of commands in the buffer is always the state that needs
 521  * to be re-emitted when the context is lost. This batch can be skipped
 522  * otherwise.
 523  */
 524 struct r300_cmdbuf {
 525         int             size;           /* DWORDs allocated for buffer */
 526         uint32_t*       cmd_buf;
 527         int             count_used;     /* DWORDs filled so far */
 528         int             count_reemit;   /* size of re-emission batch */
 529 };
 530
 531
 532 /**
 533  * State cache
 534  */
 535
 536 struct r300_depthbuffer_state {
 537         GLfloat scale;
 538 };
 539
 540 struct r300_stencilbuffer_state {
 541         GLuint clear;
 542         GLboolean hw_stencil;
 543
 544 };
 545
 546 /* Vertex shader state */
 547
 548 /* Perhaps more if we store programs in vmem? */
 549 /* drm_r300_cmd_header_t->vpu->count is unsigned char */
 550 #define VSF_MAX_FRAGMENT_LENGTH (255*4)
 551
 552 /* Can be tested with colormat currently. */
 553 #define VSF_MAX_FRAGMENT_TEMPS (14)
 554
 555 #define STATE_R300_WINDOW_DIMENSION (STATE_INTERNAL_DRIVER+0)
 556
 557 struct r300_vertex_shader_fragment {
 558         int length;
 559         union {
 560                 GLuint d[VSF_MAX_FRAGMENT_LENGTH];
 561                 float f[VSF_MAX_FRAGMENT_LENGTH];
 562                 VERTEX_SHADER_INSTRUCTION i[VSF_MAX_FRAGMENT_LENGTH/4];
 563                 } body;
 564         };
 565
 566 #define VSF_DEST_PROGRAM        0x0
 567 #define VSF_DEST_MATRIX0        0x200
 568 #define VSF_DEST_MATRIX1        0x204
 569 #define VSF_DEST_MATRIX2        0x208
 570 #define VSF_DEST_VECTOR0        0x20c
 571 #define VSF_DEST_VECTOR1        0x20d
 572 #define VSF_DEST_UNKNOWN1       0x400
 573 #define VSF_DEST_UNKNOWN2       0x406
 574
 575 struct r300_vertex_shader_state {
 576         struct r300_vertex_shader_fragment program;
 577
 578         /* a bit of a waste - each uses only a subset of allocated space..
 579             but easier to program */
 580         struct r300_vertex_shader_fragment matrix[3];
 581         struct r300_vertex_shader_fragment vector[2];
 582
 583         struct r300_vertex_shader_fragment unknown1;
 584         struct r300_vertex_shader_fragment unknown2;
 585
 586         int program_start;
 587         int unknown_ptr1;  /* pointer within program space */
 588         int program_end;
 589
 590         int param_offset;
 591         int param_count;
 592
 593         int unknown_ptr2;  /* pointer within program space */
 594         int unknown_ptr3;  /* pointer within program space */
 595         };
 596
 597 extern int hw_tcl_on;
 598
 599 //#define CURRENT_VERTEX_SHADER(ctx) (ctx->VertexProgram._Current)
 600 #define CURRENT_VERTEX_SHADER(ctx) (R300_CONTEXT(ctx)->selected_vp)
 601
 602 /* Should but doesnt work */
 603 //#define CURRENT_VERTEX_SHADER(ctx) (R300_CONTEXT(ctx)->curr_vp)
 604
 605 //#define TMU_ENABLED(ctx, unit) (hw_tcl_on ? ctx->Texture.Unit[unit]._ReallyEnabled && (OutputsWritten & (1<<(VERT_RESULT_TEX0+(unit)))) :
 606 //      (r300->state.render_inputs & (_TNL_BIT_TEX0<<(unit))))
 607 //#define TMU_ENABLED(ctx, unit) (hw_tcl_on ? ctx->Texture.Unit[unit]._ReallyEnabled && OutputsWritten & (1<<(VERT_RESULT_TEX0+(unit))) :
 608 //      ctx->Texture.Unit[unit]._ReallyEnabled && r300->state.render_inputs & (_TNL_BIT_TEX0<<(unit)))
 609
 610 #define TMU_ENABLED(ctx, unit) (ctx->Texture.Unit[unit]._ReallyEnabled)
 611
 612 /* r300_vertex_shader_state and r300_vertex_program should probably be merged together someday.
 613  * Keeping them them seperate for now should ensure fixed pipeline keeps functioning properly.
 614  */
 615
 616 struct r300_vertex_program_key {
 617         GLuint InputsRead;
 618         GLuint OutputsWritten;
 619 };
 620
 621 struct r300_vertex_program {
 622         struct r300_vertex_program *next;
 623         struct r300_vertex_program_key key;
 624         int translated;
 625
 626         struct r300_vertex_shader_fragment program;
 627
 628         int pos_end;
 629         int num_temporaries; /* Number of temp vars used by program */
 630         int wpos_idx;
 631         int inputs[VERT_ATTRIB_MAX];
 632         int outputs[VERT_RESULT_MAX];
 633         int native;
 634         int ref_count;
 635         int use_ref_count;
 636 };
 637
 638 struct r300_vertex_program_cont {
 639         struct gl_vertex_program mesa_program; /* Must be first */
 640         struct r300_vertex_shader_fragment params;
 641         struct r300_vertex_program *progs;
 642 };
 643
 644 #define PFS_MAX_ALU_INST        64
 645 #define PFS_MAX_TEX_INST        64
 646 #define PFS_MAX_TEX_INDIRECT 4
 647 #define PFS_NUM_TEMP_REGS       32
 648 #define PFS_NUM_CONST_REGS      16
 649
 650 /* Mapping Mesa registers to R300 temporaries */
 651 struct reg_acc {
 652        int reg;        /* Assigned hw temp */
 653        unsigned int refcount; /* Number of uses by mesa program */
 654 };
 655
 656 /**
 657  * Describe the current lifetime information for an R300 temporary
 658  */
 659 struct reg_lifetime {
 660         /* Index of the first slot where this register is free in the sense
 661            that it can be used as a new destination register.
 662            This is -1 if the register has been assigned to a Mesa register
 663            and the last access to the register has not yet been emitted */
 664         int free;
 665
 666         /* Index of the first slot where this register is currently reserved.
 667            This is used to stop e.g. a scalar operation from being moved
 668            before the allocation time of a register that was first allocated
 669            for a vector operation. */
 670         int reserved;
 671
 672         /* Index of the first slot in which the register can be used as a
 673            source without losing the value that is written by the last
 674            emitted instruction that writes to the register */
 675         int vector_valid;
 676         int scalar_valid;
 677
 678         /* Index to the slot where the register was last read.
 679            This is also the first slot in which the register may be written again */
 680         int vector_lastread;
 681         int scalar_lastread;
 682 };
 683
 684
 685 /**
 686  * Store usage information about an ALU instruction slot during the
 687  * compilation of a fragment program.
 688  */
 689 #define SLOT_SRC_VECTOR  (1<<0)
 690 #define SLOT_SRC_SCALAR  (1<<3)
 691 #define SLOT_SRC_BOTH    (SLOT_SRC_VECTOR | SLOT_SRC_SCALAR)
 692 #define SLOT_OP_VECTOR   (1<<16)
 693 #define SLOT_OP_SCALAR   (1<<17)
 694 #define SLOT_OP_BOTH     (SLOT_OP_VECTOR | SLOT_OP_SCALAR)
 695
 696 struct r300_pfs_compile_slot {
 697         /* Bitmask indicating which parts of the slot are used, using SLOT_ constants
 698            defined above */
 699         unsigned int used;
 700
 701         /* Selected sources */
 702         int vsrc[3];
 703         int ssrc[3];
 704 };
 705
 706 /**
 707  * Store information during compilation of fragment programs.
 708  */
 709 struct r300_pfs_compile_state {
 710         int nrslots;       /* number of ALU slots used so far */
 711
 712         /* Track which (parts of) slots are already filled with instructions */
 713         struct r300_pfs_compile_slot slot[PFS_MAX_ALU_INST];
 714
 715         /* Track the validity of R300 temporaries */
 716         struct reg_lifetime hwtemps[PFS_NUM_TEMP_REGS];
 717
 718         /* Used to map Mesa's inputs/temps onto hardware temps */
 719         int temp_in_use;
 720         struct reg_acc temps[PFS_NUM_TEMP_REGS];
 721         struct reg_acc inputs[32]; /* don't actually need 32... */
 722
 723         /* Track usage of hardware temps, for register allocation,
 724          * indirection detection, etc. */
 725         GLuint used_in_node;
 726         GLuint dest_in_node;
 727 };
 728
 729 /**
 730  * Store everything about a fragment program that is needed
 731  * to render with that program.
 732  */
 733 struct r300_fragment_program {
 734         struct gl_fragment_program mesa_program;
 735
 736         GLcontext *ctx;
 737         GLboolean translated;
 738         GLboolean error;
 739         struct r300_pfs_compile_state *cs;
 740
 741         struct {
 742                 int length;
 743                 GLuint inst[PFS_MAX_TEX_INST];
 744         } tex;
 745
 746         struct {
 747                 struct {
 748                         GLuint inst0;
 749                         GLuint inst1;
 750                         GLuint inst2;
 751                         GLuint inst3;
 752                 } inst[PFS_MAX_ALU_INST];
 753         } alu;
 754
 755         struct {
 756                 int tex_offset;
 757                 int tex_end;
 758                 int alu_offset;
 759                 int alu_end;
 760                 int flags;
 761         } node[4];
 762         int cur_node;
 763         int first_node_has_tex;
 764
 765         int alu_offset;
 766         int alu_end;
 767         int tex_offset;
 768         int tex_end;
 769
 770         /* Hardware constants.
 771          * Contains a pointer to the value. The destination of the pointer
 772          * is supposed to be updated when GL state changes.
 773          * Typically, this is either a pointer into
 774          * gl_program_parameter_list::ParameterValues, or a pointer to a
 775          * global constant (e.g. for sin/cos-approximation)
 776          */
 777         const GLfloat* constant[PFS_NUM_CONST_REGS];
 778         int const_nr;
 779
 780         int max_temp_idx;
 781
 782         /* the index of the sin constant is stored here */
 783         GLint const_sin[2];
 784
 785         GLuint optimization;
 786 };
 787
 788 #define R300_MAX_AOS_ARRAYS             16
 789
 790 #define AOS_FORMAT_USHORT       0
 791 #define AOS_FORMAT_FLOAT        1
 792 #define AOS_FORMAT_UBYTE        2
 793 #define AOS_FORMAT_FLOAT_COLOR  3
 794
 795 #define REG_COORDS      0
 796 #define REG_COLOR0      1
 797 #define REG_TEX0        2
 798
 799 struct dt {
 800         GLint size;
 801         GLenum type;
 802         GLsizei stride;
 803         void *data;
 804 };
 805
 806 struct radeon_vertex_buffer {
 807         int Count;
 808         void *Elts;
 809         int elt_size;
 810         int elt_min, elt_max; /* debug */
 811
 812         struct dt AttribPtr[VERT_ATTRIB_MAX];
 813
 814         const struct _mesa_prim  *Primitive;
 815         GLuint      PrimitiveCount;
 816         GLint LockFirst;
 817         GLsizei LockCount;
 818         int lock_uptodate;
 819 };
 820
 821 struct r300_aos_rec {
 822         GLuint offset;
 823         int element_size; /* in dwords */
 824         int stride;       /* distance between elements, in dwords */
 825
 826         int format;
 827
 828         int ncomponents; /* number of components - between 1 and 4, inclusive */
 829
 830         int reg; /* which register they are assigned to. */
 831
 832         };
 833
 834 struct r300_state {
 835         struct r300_depthbuffer_state depth;
 836         struct r300_texture_state texture;
 837         int sw_tcl_inputs[VERT_ATTRIB_MAX];
 838         struct r300_vertex_shader_state vertex_shader;
 839         struct r300_pfs_compile_state pfs_compile;
 840         struct r300_dma_region aos[R300_MAX_AOS_ARRAYS];
 841         int aos_count;
 842         struct radeon_vertex_buffer VB;
 843
 844         GLuint *Elts;
 845         struct r300_dma_region elt_dma;
 846
 847         DECLARE_RENDERINPUTS(render_inputs_bitset); /* actual render inputs that R300 was configured for.
 848                                  They are the same as tnl->render_inputs for fixed pipeline */
 849
 850         struct {
 851                 int transform_offset;  /* Transform matrix offset, -1 if none */
 852                 } vap_param;  /* vertex processor parameter allocation - tells where to write parameters */
 853
 854         struct r300_stencilbuffer_state stencil;
 855
 856 };
 857
 858 #define R300_FALLBACK_NONE 0
 859 #define R300_FALLBACK_TCL 1
 860 #define R300_FALLBACK_RAST 2
 861
 862 /**
 863  * R300 context structure.
 864  */
 865 struct r300_context {
 866         struct radeon_context radeon; /* parent class, must be first */
 867
 868         struct r300_hw_state hw;
 869         struct r300_cmdbuf cmdbuf;
 870         struct r300_state state;
 871         struct gl_vertex_program *curr_vp;
 872         struct r300_vertex_program *selected_vp;
 873
 874         /* Vertex buffers
 875          */
 876         struct r300_dma dma;
 877         GLboolean save_on_next_unlock;
 878         GLuint NewGLState;
 879
 880         /* Texture object bookkeeping
 881          */
 882         unsigned nr_heaps;
 883         driTexHeap *texture_heaps[RADEON_NR_TEX_HEAPS];
 884         driTextureObject swapped;
 885         int texture_depth;
 886         float initialMaxAnisotropy;
 887
 888         /* Clientdata textures;
 889          */
 890         GLuint prefer_gart_client_texturing;
 891
 892 #ifdef USER_BUFFERS
 893         struct radeon_memory_manager *rmm;
 894         GLvector4f dummy_attrib[_TNL_ATTRIB_MAX];
 895         GLvector4f *temp_attrib[_TNL_ATTRIB_MAX];
 896 #endif
 897
 898         GLboolean texmicrotile;
 899         GLboolean span_dlocking;
 900         GLboolean disable_lowimpact_fallback;
 901 };
 902
 903 struct r300_buffer_object {
 904         struct gl_buffer_object mesa_obj;
 905         int id;
 906 };
 907
 908 #define R300_CONTEXT(ctx)               ((r300ContextPtr)(ctx->DriverCtx))
 909
 910 static __inline GLuint r300PackColor( GLuint cpp,
 911                                         GLubyte r, GLubyte g,
 912                                         GLubyte b, GLubyte a )
 913 {
 914    switch ( cpp ) {
 915    case 2:
 916       return PACK_COLOR_565( r, g, b );
 917    case 4:
 918       return PACK_COLOR_8888( r, g, b, a );
 919    default:
 920       return 0;
 921    }
 922 }
 923 extern void r300DestroyContext(__DRIcontextPrivate * driContextPriv);
 924 extern GLboolean r300CreateContext(const __GLcontextModes * glVisual,
 925                                    __DRIcontextPrivate * driContextPriv,
 926                                    void *sharedContextPrivate);
 927
 928 extern int r300_get_num_verts(r300ContextPtr rmesa, int num_verts, int prim);
 929
 930 extern void r300_select_vertex_shader(r300ContextPtr r300);
 931 extern void r300InitShaderFuncs(struct dd_function_table *functions);
 932 extern int r300VertexProgUpdateParams(GLcontext *ctx, struct r300_vertex_program_cont *vp, float *dst);
 933 extern int r300Fallback(GLcontext *ctx);
 934
 935 extern void radeon_vb_to_rvb(r300ContextPtr rmesa, struct radeon_vertex_buffer *rvb, struct vertex_buffer *vb);
 936 extern GLboolean r300_run_vb_render(GLcontext *ctx, struct tnl_pipeline_stage *stage);
 937
 938 #ifdef RADEON_VTXFMT_A
 939 extern void radeon_init_vtxfmt_a(r300ContextPtr rmesa);
 940 #endif
 941
 942 #ifdef HW_VBOS
 943 extern void r300_init_vbo_funcs(struct dd_function_table *functions);
 944 extern void r300_evict_vbos(GLcontext *ctx, int amount);
 945 #endif
 946
 947 #define RADEON_D_CAPTURE 0
 948 #define RADEON_D_PLAYBACK 1
 949 #define RADEON_D_PLAYBACK_RAW 2
 950 #define RADEON_D_T 3
 951
 952 #endif                          /* __R300_CONTEXT_H__ */