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  * \file
  32  *
  33  * \author Keith Whitwell <keith@tungstengraphics.com>
  34  * \author Nicolai Haehnle <prefect_@gmx.net>
  35  */
  36
  37 #ifndef __R300_CONTEXT_H__
  38 #define __R300_CONTEXT_H__
  39
  40 #include "tnl/t_vertex.h"
  41 #include "drm.h"
  42 #include "radeon_drm.h"
  43 #include "dri_util.h"
  44 #include "texmem.h"
  45
  46 #include "macros.h"
  47 #include "mtypes.h"
  48 #include "colormac.h"
  49
  50 #define USER_BUFFERS
  51
  52 struct r300_context;
  53 typedef struct r300_context r300ContextRec;
  54 typedef struct r300_context *r300ContextPtr;
  55
  56 #include "radeon_lock.h"
  57 #include "mm.h"
  58
  59 /* From http://gcc.gnu.org/onlinedocs/gcc-3.2.3/gcc/Variadic-Macros.html .
  60    I suppose we could inline this and use macro to fetch out __LINE__ and stuff in case we run into trouble
  61    with other compilers ... GLUE!
  62 */
  63 #define WARN_ONCE(a, ...)       { \
  64         static int warn##__LINE__=1; \
  65         if(warn##__LINE__){ \
  66                 fprintf(stderr, "*********************************WARN_ONCE*********************************\n"); \
  67                 fprintf(stderr, "File %s function %s line %d\n", \
  68                         __FILE__, __FUNCTION__, __LINE__); \
  69                 fprintf(stderr,  a, ## __VA_ARGS__);\
  70                 fprintf(stderr, "***************************************************************************\n"); \
  71                 warn##__LINE__=0;\
  72                 } \
  73         }
  74
  75 #include "r300_vertprog.h"
  76 #include "r300_fragprog.h"
  77
  78 /**
  79  * This function takes a float and packs it into a uint32_t
  80  */
  81 static inline uint32_t r300PackFloat32(float fl)
  82 {
  83         union {
  84                 float fl;
  85                 uint32_t u;
  86         } u;
  87
  88         u.fl = fl;
  89         return u.u;
  90 }
  91
  92 /* This is probably wrong for some values, I need to test this
  93  * some more.  Range checking would be a good idea also..
  94  *
  95  * But it works for most things.  I'll fix it later if someone
  96  * else with a better clue doesn't
  97  */
  98 static inline uint32_t r300PackFloat24(float f)
  99 {
 100         float mantissa;
 101         int exponent;
 102         uint32_t float24 = 0;
 103
 104         if (f == 0.0)
 105                 return 0;
 106
 107         mantissa = frexpf(f, &exponent);
 108
 109         /* Handle -ve */
 110         if (mantissa < 0) {
 111                 float24 |= (1 << 23);
 112                 mantissa = mantissa * -1.0;
 113         }
 114         /* Handle exponent, bias of 63 */
 115         exponent += 62;
 116         float24 |= (exponent << 16);
 117         /* Kill 7 LSB of mantissa */
 118         float24 |= (r300PackFloat32(mantissa) & 0x7FFFFF) >> 7;
 119
 120         return float24;
 121 }
 122
 123 /************ DMA BUFFERS **************/
 124
 125 /* Need refcounting on dma buffers:
 126  */
 127 struct r300_dma_buffer {
 128         int refcount;           /**< the number of retained regions in buf */
 129         drmBufPtr buf;
 130         int id;
 131 };
 132 #undef GET_START
 133 #ifdef USER_BUFFERS
 134 #define GET_START(rvb) (r300GartOffsetFromVirtual(rmesa, (rvb)->address+(rvb)->start))
 135 #else
 136 #define GET_START(rvb) (rmesa->radeon.radeonScreen->gart_buffer_offset +                \
 137                         (rvb)->address - rmesa->dma.buf0_address +      \
 138                         (rvb)->start)
 139 #endif
 140 /* A retained region, eg vertices for indexed vertices.
 141  */
 142 struct r300_dma_region {
 143         struct r300_dma_buffer *buf;
 144         char *address;          /* == buf->address */
 145         int start, end, ptr;    /* offsets from start of buf */
 146
 147         int aos_offset;         /* address in GART memory */
 148         int aos_stride;         /* distance between elements, in dwords */
 149         int aos_size;           /* number of components (1-4) */
 150 };
 151
 152 struct r300_dma {
 153         /* Active dma region.  Allocations for vertices and retained
 154          * regions come from here.  Also used for emitting random vertices,
 155          * these may be flushed by calling flush_current();
 156          */
 157         struct r300_dma_region current;
 158
 159         void (*flush) (r300ContextPtr);
 160
 161         char *buf0_address;     /* start of buf[0], for index calcs */
 162
 163         /* Number of "in-flight" DMA buffers, i.e. the number of buffers
 164          * for which a DISCARD command is currently queued in the command buffer.
 165          */
 166         GLuint nr_released_bufs;
 167 };
 168
 169        /* Texture related */
 170
 171 typedef struct r300_tex_obj r300TexObj, *r300TexObjPtr;
 172
 173 /* Texture object in locally shared texture space.
 174  */
 175 struct r300_tex_obj {
 176         driTextureObject base;
 177
 178         GLuint bufAddr;         /* Offset to start of locally
 179                                    shared texture block */
 180
 181         GLuint dirty_state;     /* Flags (1 per texunit) for
 182                                    whether or not this texobj
 183                                    has dirty hardware state
 184                                    (pp_*) that needs to be
 185                                    brought into the
 186                                    texunit. */
 187
 188         drm_radeon_tex_image_t image[6][RADEON_MAX_TEXTURE_LEVELS];
 189         /* Six, for the cube faces */
 190
 191         GLboolean image_override;       /* Image overridden by GLX_EXT_tfp */
 192
 193         GLuint pitch;           /* this isn't sent to hardware just used in calculations */
 194         /* hardware register values */
 195         /* Note that R200 has 8 registers per texture and R300 only 7 */
 196         GLuint filter;
 197         GLuint filter_1;
 198         GLuint pitch_reg;
 199         GLuint size;            /* npot only */
 200         GLuint format;
 201         GLuint offset;          /* Image location in the card's address space.
 202                                    All cube faces follow. */
 203         GLuint unknown4;
 204         GLuint unknown5;
 205         /* end hardware registers */
 206
 207         /* registers computed by r200 code - keep them here to
 208            compare against what is actually written.
 209
 210            to be removed later.. */
 211         GLuint pp_border_color;
 212         GLuint pp_cubic_faces;  /* cube face 1,2,3,4 log2 sizes */
 213         GLuint format_x;
 214
 215         GLboolean border_fallback;
 216
 217         GLuint tile_bits;       /* hw texture tile bits used on this texture */
 218 };
 219
 220 struct r300_texture_env_state {
 221         r300TexObjPtr texobj;
 222         GLenum format;
 223         GLenum envMode;
 224 };
 225
 226 /* The blit width for texture uploads
 227  */
 228 #define R300_BLIT_WIDTH_BYTES 1024
 229 #define R300_MAX_TEXTURE_UNITS 8
 230
 231 struct r300_texture_state {
 232         struct r300_texture_env_state unit[R300_MAX_TEXTURE_UNITS];
 233         int tc_count;           /* number of incoming texture coordinates from VAP */
 234 };
 235
 236 /**
 237  * A block of hardware state.
 238  *
 239  * When check returns non-zero, the returned number of dwords must be
 240  * copied verbatim into the command buffer in order to update a state atom
 241  * when it is dirty.
 242  */
 243 struct r300_state_atom {
 244         struct r300_state_atom *next, *prev;
 245         const char *name;       /* for debug */
 246         int cmd_size;           /* maximum size in dwords */
 247         GLuint idx;             /* index in an array (e.g. textures) */
 248         uint32_t *cmd;
 249         GLboolean dirty;
 250
 251         int (*check) (r300ContextPtr, struct r300_state_atom * atom);
 252 };
 253
 254 #define R300_VPT_CMD_0          0
 255 #define R300_VPT_XSCALE         1
 256 #define R300_VPT_XOFFSET        2
 257 #define R300_VPT_YSCALE         3
 258 #define R300_VPT_YOFFSET        4
 259 #define R300_VPT_ZSCALE         5
 260 #define R300_VPT_ZOFFSET        6
 261 #define R300_VPT_CMDSIZE        7
 262
 263 #define R300_VIR_CMD_0          0       /* vir is variable size (at least 1) */
 264 #define R300_VIR_CNTL_0         1
 265 #define R300_VIR_CNTL_1         2
 266 #define R300_VIR_CNTL_2         3
 267 #define R300_VIR_CNTL_3         4
 268 #define R300_VIR_CNTL_4         5
 269 #define R300_VIR_CNTL_5         6
 270 #define R300_VIR_CNTL_6         7
 271 #define R300_VIR_CNTL_7         8
 272 #define R300_VIR_CMDSIZE        9
 273
 274 #define R300_VIC_CMD_0          0
 275 #define R300_VIC_CNTL_0         1
 276 #define R300_VIC_CNTL_1         2
 277 #define R300_VIC_CMDSIZE        3
 278
 279 #define R300_VOF_CMD_0          0
 280 #define R300_VOF_CNTL_0         1
 281 #define R300_VOF_CNTL_1         2
 282 #define R300_VOF_CMDSIZE        3
 283
 284 #define R300_PVS_CMD_0          0
 285 #define R300_PVS_CNTL_1         1
 286 #define R300_PVS_CNTL_2         2
 287 #define R300_PVS_CNTL_3         3
 288 #define R300_PVS_CMDSIZE        4
 289
 290 #define R300_GB_MISC_CMD_0              0
 291 #define R300_GB_MISC_MSPOS_0            1
 292 #define R300_GB_MISC_MSPOS_1            2
 293 #define R300_GB_MISC_TILE_CONFIG        3
 294 #define R300_GB_MISC_SELECT             4
 295 #define R300_GB_MISC_AA_CONFIG          5
 296 #define R300_GB_MISC_CMDSIZE            6
 297
 298 #define R300_TXE_CMD_0          0
 299 #define R300_TXE_ENABLE         1
 300 #define R300_TXE_CMDSIZE        2
 301
 302 #define R300_PS_CMD_0           0
 303 #define R300_PS_POINTSIZE       1
 304 #define R300_PS_CMDSIZE         2
 305
 306 #define R300_ZBS_CMD_0          0
 307 #define R300_ZBS_T_FACTOR       1
 308 #define R300_ZBS_T_CONSTANT     2
 309 #define R300_ZBS_W_FACTOR       3
 310 #define R300_ZBS_W_CONSTANT     4
 311 #define R300_ZBS_CMDSIZE        5
 312
 313 #define R300_CUL_CMD_0          0
 314 #define R300_CUL_CULL           1
 315 #define R300_CUL_CMDSIZE        2
 316
 317 #define R300_RC_CMD_0           0
 318 #define R300_RC_CNTL_0          1
 319 #define R300_RC_CNTL_1          2
 320 #define R300_RC_CMDSIZE         3
 321
 322 #define R300_RI_CMD_0           0
 323 #define R300_RI_INTERP_0        1
 324 #define R300_RI_INTERP_1        2
 325 #define R300_RI_INTERP_2        3
 326 #define R300_RI_INTERP_3        4
 327 #define R300_RI_INTERP_4        5
 328 #define R300_RI_INTERP_5        6
 329 #define R300_RI_INTERP_6        7
 330 #define R300_RI_INTERP_7        8
 331 #define R300_RI_CMDSIZE         9
 332
 333 #define R500_RI_CMDSIZE        17
 334
 335 #define R300_RR_CMD_0           0       /* rr is variable size (at least 1) */
 336 #define R300_RR_INST_0          1
 337 #define R300_RR_INST_1          2
 338 #define R300_RR_INST_2          3
 339 #define R300_RR_INST_3          4
 340 #define R300_RR_INST_4          5
 341 #define R300_RR_INST_5          6
 342 #define R300_RR_INST_6          7
 343 #define R300_RR_INST_7          8
 344 #define R300_RR_CMDSIZE         9
 345
 346 #define R300_FP_CMD_0           0
 347 #define R300_FP_CNTL0           1
 348 #define R300_FP_CNTL1           2
 349 #define R300_FP_CNTL2           3
 350 #define R300_FP_CMD_1           4
 351 #define R300_FP_NODE0           5
 352 #define R300_FP_NODE1           6
 353 #define R300_FP_NODE2           7
 354 #define R300_FP_NODE3           8
 355 #define R300_FP_CMDSIZE         9
 356
 357 #define R300_FPT_CMD_0          0
 358 #define R300_FPT_INSTR_0        1
 359 #define R300_FPT_CMDSIZE        65
 360
 361 #define R300_FPI_CMD_0          0
 362 #define R300_FPI_INSTR_0        1
 363 #define R300_FPI_CMDSIZE        65
 364
 365 #define R300_FPP_CMD_0          0
 366 #define R300_FPP_PARAM_0        1
 367 #define R300_FPP_CMDSIZE        (32*4+1)
 368
 369 #define R300_FOGS_CMD_0         0
 370 #define R300_FOGS_STATE         1
 371 #define R300_FOGS_CMDSIZE       2
 372
 373 #define R300_FOGC_CMD_0         0
 374 #define R300_FOGC_R             1
 375 #define R300_FOGC_G             2
 376 #define R300_FOGC_B             3
 377 #define R300_FOGC_CMDSIZE       4
 378
 379 #define R300_FOGP_CMD_0         0
 380 #define R300_FOGP_SCALE         1
 381 #define R300_FOGP_START         2
 382 #define R300_FOGP_CMDSIZE       3
 383
 384 #define R300_AT_CMD_0           0
 385 #define R300_AT_ALPHA_TEST      1
 386 #define R300_AT_UNKNOWN         2
 387 #define R300_AT_CMDSIZE         3
 388
 389 #define R300_BLD_CMD_0          0
 390 #define R300_BLD_CBLEND         1
 391 #define R300_BLD_ABLEND         2
 392 #define R300_BLD_CMDSIZE        3
 393
 394 #define R300_CMK_CMD_0          0
 395 #define R300_CMK_COLORMASK      1
 396 #define R300_CMK_CMDSIZE        2
 397
 398 #define R300_CB_CMD_0           0
 399 #define R300_CB_OFFSET          1
 400 #define R300_CB_CMD_1           2
 401 #define R300_CB_PITCH           3
 402 #define R300_CB_CMDSIZE         4
 403
 404 #define R300_ZS_CMD_0           0
 405 #define R300_ZS_CNTL_0          1
 406 #define R300_ZS_CNTL_1          2
 407 #define R300_ZS_CNTL_2          3
 408 #define R300_ZS_CMDSIZE         4
 409
 410 #define R300_ZB_CMD_0           0
 411 #define R300_ZB_OFFSET          1
 412 #define R300_ZB_PITCH           2
 413 #define R300_ZB_CMDSIZE         3
 414
 415 #define R300_VPI_CMD_0          0
 416 #define R300_VPI_INSTR_0        1
 417 #define R300_VPI_CMDSIZE        1025    /* 256 16 byte instructions */
 418
 419 #define R300_VPP_CMD_0          0
 420 #define R300_VPP_PARAM_0        1
 421 #define R300_VPP_CMDSIZE        1025    /* 256 4-component parameters */
 422
 423 #define R300_VPUCP_CMD_0                0
 424 #define R300_VPUCP_X            1
 425 #define R300_VPUCP_Y            2
 426 #define R300_VPUCP_Z            3
 427 #define R300_VPUCP_W            4
 428 #define R300_VPUCP_CMDSIZE      5       /* 256 4-component parameters */
 429
 430 #define R300_VPS_CMD_0          0
 431 #define R300_VPS_ZERO_0         1
 432 #define R300_VPS_ZERO_1         2
 433 #define R300_VPS_POINTSIZE      3
 434 #define R300_VPS_ZERO_3         4
 435 #define R300_VPS_CMDSIZE        5
 436
 437         /* the layout is common for all fields inside tex */
 438 #define R300_TEX_CMD_0          0
 439 #define R300_TEX_VALUE_0        1
 440 /* We don't really use this, instead specify mtu+1 dynamically
 441 #define R300_TEX_CMDSIZE        (MAX_TEXTURE_UNITS+1)
 442 */
 443
 444 /**
 445  * Cache for hardware register state.
 446  */
 447 struct r300_hw_state {
 448         struct r300_state_atom atomlist;
 449
 450         GLboolean is_dirty;
 451         GLboolean all_dirty;
 452         int max_state_size;     /* in dwords */
 453
 454         struct r300_state_atom vpt;     /* viewport (1D98) */
 455         struct r300_state_atom vap_cntl;
 456         struct r300_state_atom vof;     /* VAP output format register 0x2090 */
 457         struct r300_state_atom vte;     /* (20B0) */
 458         struct r300_state_atom vap_vf_max_vtx_indx;     /* Maximum Vertex Indx Clamp (2134) */
 459         struct r300_state_atom vap_cntl_status;
 460         struct r300_state_atom vir[2];  /* vap input route (2150/21E0) */
 461         struct r300_state_atom vic;     /* vap input control (2180) */
 462         struct r300_state_atom vap_psc_sgn_norm_cntl; /* Programmable Stream Control Signed Normalize Control (21DC) */
 463         struct r300_state_atom vap_clip_cntl;
 464         struct r300_state_atom vap_clip;
 465         struct r300_state_atom vap_pvs_vtx_timeout_reg; /* Vertex timeout register (2288) */
 466         struct r300_state_atom pvs;     /* pvs_cntl (22D0) */
 467         struct r300_state_atom gb_enable;       /* (4008) */
 468         struct r300_state_atom gb_misc; /* Multisampling position shifts ? (4010) */
 469         struct r300_state_atom ga_point_s0;     /* S Texture Coordinate of Vertex 0 for Point texture stuffing (LLC) (4200) */
 470         struct r300_state_atom ga_triangle_stipple;     /* (4214) */
 471         struct r300_state_atom ps;      /* pointsize (421C) */
 472         struct r300_state_atom ga_point_minmax; /* (4230) */
 473         struct r300_state_atom lcntl;   /* line control */
 474         struct r300_state_atom ga_line_stipple; /* (4260) */
 475         struct r300_state_atom shade;
 476         struct r300_state_atom polygon_mode;
 477         struct r300_state_atom fogp;    /* fog parameters (4294) */
 478         struct r300_state_atom unk429C; /* (429C) */
 479         struct r300_state_atom zbias_cntl;
 480         struct r300_state_atom zbs;     /* zbias (42A4) */
 481         struct r300_state_atom occlusion_cntl;
 482         struct r300_state_atom cul;     /* cull cntl (42B8) */
 483         struct r300_state_atom su_depth_scale;  /* (42C0) */
 484         struct r300_state_atom rc;      /* rs control (4300) */
 485         struct r300_state_atom ri;      /* rs interpolators (4310) */
 486         struct r300_state_atom rr;      /* rs route (4330) */
 487         struct r300_state_atom sc_hyperz;       /* (43A4) */
 488         struct r300_state_atom sc_screendoor;   /* (43E8) */
 489         struct r300_state_atom fp;      /* fragment program cntl + nodes (4600) */
 490         struct r300_state_atom fpt;     /* texi - (4620) */
 491         struct r300_state_atom us_out_fmt;      /* (46A4) */
 492         struct r300_state_atom r500fp;  /* r500 fp instructions */
 493         struct r300_state_atom r500fp_const;    /* r500 fp constants */
 494         struct r300_state_atom fpi[4];  /* fp instructions (46C0/47C0/48C0/49C0) */
 495         struct r300_state_atom fogs;    /* fog state (4BC0) */
 496         struct r300_state_atom fogc;    /* fog color (4BC8) */
 497         struct r300_state_atom at;      /* alpha test (4BD4) */
 498         struct r300_state_atom fg_depth_src;    /* (4BD8) */
 499         struct r300_state_atom fpp;     /* 0x4C00 and following */
 500         struct r300_state_atom rb3d_cctl;       /* (4E00) */
 501         struct r300_state_atom bld;     /* blending (4E04) */
 502         struct r300_state_atom cmk;     /* colormask (4E0C) */
 503         struct r300_state_atom blend_color;     /* constant blend color */
 504         struct r300_state_atom cb;      /* colorbuffer (4E28) */
 505         struct r300_state_atom rb3d_dither_ctl; /* (4E50) */
 506         struct r300_state_atom rb3d_aaresolve_ctl;      /* (4E88) */
 507         struct r300_state_atom rb3d_discard_src_pixel_lte_threshold;    /* (4E88) I saw it only written on RV350 hardware..  */
 508         struct r300_state_atom zs;      /* zstencil control (4F00) */
 509         struct r300_state_atom zstencil_format;
 510         struct r300_state_atom zb;      /* z buffer (4F20) */
 511         struct r300_state_atom zb_depthclearvalue;      /* (4F28) */
 512         struct r300_state_atom unk4F30; /* (4F30) */
 513         struct r300_state_atom zb_hiz_offset;   /* (4F44) */
 514         struct r300_state_atom zb_hiz_pitch;    /* (4F54) */
 515
 516         struct r300_state_atom vpi;     /* vp instructions */
 517         struct r300_state_atom vpp;     /* vp parameters */
 518         struct r300_state_atom vps;     /* vertex point size (?) */
 519         struct r300_state_atom vpucp[6];        /* vp user clip plane - 6 */
 520         /* 8 texture units */
 521         /* the state is grouped by function and not by
 522            texture unit. This makes single unit updates
 523            really awkward - we are much better off
 524            updating the whole thing at once */
 525         struct {
 526                 struct r300_state_atom filter;
 527                 struct r300_state_atom filter_1;
 528                 struct r300_state_atom size;
 529                 struct r300_state_atom format;
 530                 struct r300_state_atom pitch;
 531                 struct r300_state_atom offset;
 532                 struct r300_state_atom chroma_key;
 533                 struct r300_state_atom border_color;
 534         } tex;
 535         struct r300_state_atom txe;     /* tex enable (4104) */
 536 };
 537
 538 /**
 539  * This structure holds the command buffer while it is being constructed.
 540  *
 541  * The first batch of commands in the buffer is always the state that needs
 542  * to be re-emitted when the context is lost. This batch can be skipped
 543  * otherwise.
 544  */
 545 struct r300_cmdbuf {
 546         int size;               /* DWORDs allocated for buffer */
 547         uint32_t *cmd_buf;
 548         int count_used;         /* DWORDs filled so far */
 549         int count_reemit;       /* size of re-emission batch */
 550 };
 551
 552 /**
 553  * State cache
 554  */
 555
 556 struct r300_depthbuffer_state {
 557         GLfloat scale;
 558 };
 559
 560 struct r300_stencilbuffer_state {
 561         GLuint clear;
 562         GLboolean hw_stencil;
 563
 564 };
 565
 566 /* Vertex shader state */
 567
 568 /* Perhaps more if we store programs in vmem? */
 569 /* drm_r300_cmd_header_t->vpu->count is unsigned char */
 570 #define VSF_MAX_FRAGMENT_LENGTH (255*4)
 571
 572 /* Can be tested with colormat currently. */
 573 #define VSF_MAX_FRAGMENT_TEMPS (14)
 574
 575 #define STATE_R300_WINDOW_DIMENSION (STATE_INTERNAL_DRIVER+0)
 576 #define STATE_R300_TEXRECT_FACTOR (STATE_INTERNAL_DRIVER+1)
 577
 578 struct r300_vertex_shader_fragment {
 579         int length;
 580         union {
 581                 GLuint d[VSF_MAX_FRAGMENT_LENGTH];
 582                 float f[VSF_MAX_FRAGMENT_LENGTH];
 583                 GLuint i[VSF_MAX_FRAGMENT_LENGTH];
 584         } body;
 585 };
 586
 587 struct r300_vertex_shader_state {
 588         struct r300_vertex_shader_fragment program;
 589 };
 590
 591 extern int hw_tcl_on;
 592
 593 #define COLOR_IS_RGBA
 594 #define TAG(x) r300##x
 595 #include "tnl_dd/t_dd_vertex.h"
 596 #undef TAG
 597
 598 //#define CURRENT_VERTEX_SHADER(ctx) (ctx->VertexProgram._Current)
 599 #define CURRENT_VERTEX_SHADER(ctx) (R300_CONTEXT(ctx)->selected_vp)
 600
 601 /* Should but doesnt work */
 602 //#define CURRENT_VERTEX_SHADER(ctx) (R300_CONTEXT(ctx)->curr_vp)
 603
 604 /* r300_vertex_shader_state and r300_vertex_program should probably be merged together someday.
 605  * Keeping them them seperate for now should ensure fixed pipeline keeps functioning properly.
 606  */
 607
 608 struct r300_vertex_program_key {
 609         GLuint InputsRead;
 610         GLuint OutputsWritten;
 611         GLuint OutputsAdded;
 612 };
 613
 614 struct r300_vertex_program {
 615         struct r300_vertex_program *next;
 616         struct r300_vertex_program_key key;
 617         int translated;
 618
 619         struct r300_vertex_shader_fragment program;
 620
 621         int pos_end;
 622         int num_temporaries;    /* Number of temp vars used by program */
 623         int wpos_idx;
 624         int inputs[VERT_ATTRIB_MAX];
 625         int outputs[VERT_RESULT_MAX];
 626         int native;
 627         int ref_count;
 628         int use_ref_count;
 629 };
 630
 631 struct r300_vertex_program_cont {
 632         struct gl_vertex_program mesa_program;  /* Must be first */
 633         struct r300_vertex_shader_fragment params;
 634         struct r300_vertex_program *progs;
 635 };
 636
 637 #define PFS_MAX_ALU_INST        64
 638 #define PFS_MAX_TEX_INST        64
 639 #define PFS_MAX_TEX_INDIRECT 4
 640 #define PFS_NUM_TEMP_REGS       32
 641 #define PFS_NUM_CONST_REGS      16
 642
 643 /* Mapping Mesa registers to R300 temporaries */
 644 struct reg_acc {
 645         int reg;                /* Assigned hw temp */
 646         unsigned int refcount;  /* Number of uses by mesa program */
 647 };
 648
 649 /**
 650  * Describe the current lifetime information for an R300 temporary
 651  */
 652 struct reg_lifetime {
 653         /* Index of the first slot where this register is free in the sense
 654            that it can be used as a new destination register.
 655            This is -1 if the register has been assigned to a Mesa register
 656            and the last access to the register has not yet been emitted */
 657         int free;
 658
 659         /* Index of the first slot where this register is currently reserved.
 660            This is used to stop e.g. a scalar operation from being moved
 661            before the allocation time of a register that was first allocated
 662            for a vector operation. */
 663         int reserved;
 664
 665         /* Index of the first slot in which the register can be used as a
 666            source without losing the value that is written by the last
 667            emitted instruction that writes to the register */
 668         int vector_valid;
 669         int scalar_valid;
 670
 671         /* Index to the slot where the register was last read.
 672            This is also the first slot in which the register may be written again */
 673         int vector_lastread;
 674         int scalar_lastread;
 675 };
 676
 677 /**
 678  * Store usage information about an ALU instruction slot during the
 679  * compilation of a fragment program.
 680  */
 681 #define SLOT_SRC_VECTOR  (1<<0)
 682 #define SLOT_SRC_SCALAR  (1<<3)
 683 #define SLOT_SRC_BOTH    (SLOT_SRC_VECTOR | SLOT_SRC_SCALAR)
 684 #define SLOT_OP_VECTOR   (1<<16)
 685 #define SLOT_OP_SCALAR   (1<<17)
 686 #define SLOT_OP_BOTH     (SLOT_OP_VECTOR | SLOT_OP_SCALAR)
 687
 688 struct r300_pfs_compile_slot {
 689         /* Bitmask indicating which parts of the slot are used, using SLOT_ constants
 690            defined above */
 691         unsigned int used;
 692
 693         /* Selected sources */
 694         int vsrc[3];
 695         int ssrc[3];
 696 };
 697
 698 /**
 699  * Store information during compilation of fragment programs.
 700  */
 701 struct r300_pfs_compile_state {
 702         int nrslots;            /* number of ALU slots used so far */
 703
 704         /* Track which (parts of) slots are already filled with instructions */
 705         struct r300_pfs_compile_slot slot[PFS_MAX_ALU_INST];
 706
 707         /* Track the validity of R300 temporaries */
 708         struct reg_lifetime hwtemps[PFS_NUM_TEMP_REGS];
 709
 710         /* Used to map Mesa's inputs/temps onto hardware temps */
 711         int temp_in_use;
 712         struct reg_acc temps[PFS_NUM_TEMP_REGS];
 713         struct reg_acc inputs[32];      /* don't actually need 32... */
 714
 715         /* Track usage of hardware temps, for register allocation,
 716          * indirection detection, etc. */
 717         GLuint used_in_node;
 718         GLuint dest_in_node;
 719 };
 720
 721 /**
 722  * Store everything about a fragment program that is needed
 723  * to render with that program.
 724  */
 725 struct r300_fragment_program {
 726         struct gl_fragment_program mesa_program;
 727
 728         GLcontext *ctx;
 729         GLboolean translated;
 730         GLboolean error;
 731         struct r300_pfs_compile_state *cs;
 732
 733         struct {
 734                 int length;
 735                 GLuint inst[PFS_MAX_TEX_INST];
 736         } tex;
 737
 738         struct {
 739                 struct {
 740                         GLuint inst0;
 741                         GLuint inst1;
 742                         GLuint inst2;
 743                         GLuint inst3;
 744                 } inst[PFS_MAX_ALU_INST];
 745         } alu;
 746
 747         struct {
 748                 int tex_offset;
 749                 int tex_end;
 750                 int alu_offset;
 751                 int alu_end;
 752                 int flags;
 753         } node[4];
 754         int cur_node;
 755         int first_node_has_tex;
 756
 757         int alu_offset;
 758         int alu_end;
 759         int tex_offset;
 760         int tex_end;
 761
 762         /* Hardware constants.
 763          * Contains a pointer to the value. The destination of the pointer
 764          * is supposed to be updated when GL state changes.
 765          * Typically, this is either a pointer into
 766          * gl_program_parameter_list::ParameterValues, or a pointer to a
 767          * global constant (e.g. for sin/cos-approximation)
 768          */
 769         const GLfloat *constant[PFS_NUM_CONST_REGS];
 770         int const_nr;
 771
 772         int max_temp_idx;
 773
 774         GLuint optimization;
 775 };
 776
 777 struct r500_fragment_program {
 778         struct gl_fragment_program mesa_program;
 779
 780         GLcontext *ctx;
 781         GLboolean translated;
 782         GLboolean error;
 783         struct r300_pfs_compile_state *cs;
 784
 785         struct {
 786                 GLuint inst0;
 787                 GLuint inst1;
 788                 GLuint inst2;
 789                 GLuint inst3;
 790                 GLuint inst4;
 791                 GLuint inst5;
 792         } inst[512];
 793         /* TODO: This is magic! */
 794
 795         struct {
 796                 int tex_offset;
 797                 int tex_end;
 798                 int alu_offset;
 799                 int alu_end;
 800                 int flags;
 801         } node[4];
 802         int cur_node;
 803         int first_node_has_tex;
 804
 805         int alu_offset;
 806         int alu_end;
 807         int tex_offset;
 808         int tex_end;
 809
 810         /* Hardware constants.
 811          * Contains a pointer to the value. The destination of the pointer
 812          * is supposed to be updated when GL state changes.
 813          * Typically, this is either a pointer into
 814          * gl_program_parameter_list::ParameterValues, or a pointer to a
 815          * global constant (e.g. for sin/cos-approximation)
 816          */
 817         const GLfloat *constant[PFS_NUM_CONST_REGS];
 818         int const_nr;
 819
 820         int max_temp_idx;
 821
 822         GLuint optimization;
 823 };
 824
 825 #define R300_MAX_AOS_ARRAYS             16
 826
 827 #define REG_COORDS      0
 828 #define REG_COLOR0      1
 829 #define REG_TEX0        2
 830
 831 struct r300_state {
 832         struct r300_depthbuffer_state depth;
 833         struct r300_texture_state texture;
 834         int sw_tcl_inputs[VERT_ATTRIB_MAX];
 835         struct r300_vertex_shader_state vertex_shader;
 836         struct r300_pfs_compile_state pfs_compile;
 837         struct r300_dma_region aos[R300_MAX_AOS_ARRAYS];
 838         int aos_count;
 839
 840         GLuint *Elts;
 841         struct r300_dma_region elt_dma;
 842
 843         struct r300_dma_region swtcl_dma;
 844         DECLARE_RENDERINPUTS(render_inputs_bitset);     /* actual render inputs that R300 was configured for.
 845                                                            They are the same as tnl->render_inputs for fixed pipeline */
 846
 847         struct r300_stencilbuffer_state stencil;
 848
 849 };
 850
 851 #define R300_FALLBACK_NONE 0
 852 #define R300_FALLBACK_TCL 1
 853 #define R300_FALLBACK_RAST 2
 854
 855 /* r300_swtcl.c
 856  */
 857 struct r300_swtcl_info {
 858    GLuint RenderIndex;
 859
 860    /**
 861     * Size of a hardware vertex.  This is calculated when \c ::vertex_attrs is
 862     * installed in the Mesa state vector.
 863     */
 864    GLuint vertex_size;
 865
 866    /**
 867     * Attributes instructing the Mesa TCL pipeline where / how to put vertex
 868     * data in the hardware buffer.
 869     */
 870    struct tnl_attr_map vertex_attrs[VERT_ATTRIB_MAX];
 871
 872    /**
 873     * Number of elements of \c ::vertex_attrs that are actually used.
 874     */
 875    GLuint vertex_attr_count;
 876
 877    /**
 878     * Cached pointer to the buffer where Mesa will store vertex data.
 879     */
 880    GLubyte *verts;
 881
 882    /* Fallback rasterization functions
 883     */
 884   //   r200_point_func draw_point;
 885   //   r200_line_func draw_line;
 886   //   r200_tri_func draw_tri;
 887
 888    GLuint hw_primitive;
 889    GLenum render_primitive;
 890    GLuint numverts;
 891
 892    /**
 893     * Offset of the 4UB color data within a hardware (swtcl) vertex.
 894     */
 895    GLuint coloroffset;
 896
 897    /**
 898     * Offset of the 3UB specular color data within a hardware (swtcl) vertex.
 899     */
 900    GLuint specoffset;
 901
 902    /**
 903     * Should Mesa project vertex data or will the hardware do it?
 904     */
 905    GLboolean needproj;
 906
 907    struct r300_dma_region indexed_verts;
 908 };
 909
 910
 911 /**
 912  * \brief R300 context structure.
 913  */
 914 struct r300_context {
 915         struct radeon_context radeon;   /* parent class, must be first */
 916
 917         struct r300_hw_state hw;
 918         struct r300_cmdbuf cmdbuf;
 919         struct r300_state state;
 920         struct gl_vertex_program *curr_vp;
 921         struct r300_vertex_program *selected_vp;
 922
 923         /* Vertex buffers
 924          */
 925         struct r300_dma dma;
 926         GLboolean save_on_next_unlock;
 927         GLuint NewGLState;
 928
 929         /* Texture object bookkeeping
 930          */
 931         unsigned nr_heaps;
 932         driTexHeap *texture_heaps[RADEON_NR_TEX_HEAPS];
 933         driTextureObject swapped;
 934         int texture_depth;
 935         float initialMaxAnisotropy;
 936
 937         /* Clientdata textures;
 938          */
 939         GLuint prefer_gart_client_texturing;
 940
 941 #ifdef USER_BUFFERS
 942         struct r300_memory_manager *rmm;
 943 #endif
 944
 945         GLvector4f dummy_attrib[_TNL_ATTRIB_MAX];
 946         GLvector4f *temp_attrib[_TNL_ATTRIB_MAX];
 947
 948         GLboolean disable_lowimpact_fallback;
 949
 950         DECLARE_RENDERINPUTS(tnl_index_bitset); /* index of bits for last tnl_install_attrs */
 951         struct r300_swtcl_info swtcl;
 952 };
 953
 954 struct r300_buffer_object {
 955         struct gl_buffer_object mesa_obj;
 956         int id;
 957 };
 958
 959 #define R300_CONTEXT(ctx)               ((r300ContextPtr)(ctx->DriverCtx))
 960
 961 extern void r300DestroyContext(__DRIcontextPrivate * driContextPriv);
 962 extern GLboolean r300CreateContext(const __GLcontextModes * glVisual,
 963                                    __DRIcontextPrivate * driContextPriv,
 964                                    void *sharedContextPrivate);
 965
 966 extern void r300SelectVertexShader(r300ContextPtr r300);
 967 extern void r300InitShaderFuncs(struct dd_function_table *functions);
 968 extern int r300VertexProgUpdateParams(GLcontext * ctx,
 969                                       struct r300_vertex_program_cont *vp,
 970                                       float *dst);
 971
 972 #define RADEON_D_CAPTURE 0
 973 #define RADEON_D_PLAYBACK 1
 974 #define RADEON_D_PLAYBACK_RAW 2
 975 #define RADEON_D_T 3
 976
 977 #endif                          /* __R300_CONTEXT_H__ */