src/mesa/program/prog_instruction.h

   1 /*
   2  * Mesa 3-D graphics library
   3  *
   4  * Copyright (C) 1999-2008  Brian Paul   All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10  * and/or sell copies of the Software, and to permit persons to whom the
  11  * Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included
  14  * in all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  17  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
  20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  22  * OTHER DEALINGS IN THE SOFTWARE.
  23  */
  24
  25
  26 /**
  27  * \file prog_instruction.h
  28  *
  29  * Vertex/fragment program instruction datatypes and constants.
  30  *
  31  * \author Brian Paul
  32  * \author Keith Whitwell
  33  * \author Ian Romanick <idr@us.ibm.com>
  34  */
  35
  36
  37 #ifndef PROG_INSTRUCTION_H
  38 #define PROG_INSTRUCTION_H
  39
  40
  41 #include "main/glheader.h"
  42
  43
  44 /**
  45  * Swizzle indexes.
  46  * Do not change!
  47  */
  48 /*@{*/
  49 #define SWIZZLE_X    0
  50 #define SWIZZLE_Y    1
  51 #define SWIZZLE_Z    2
  52 #define SWIZZLE_W    3
  53 #define SWIZZLE_ZERO 4   /**< For SWZ instruction only */
  54 #define SWIZZLE_ONE  5   /**< For SWZ instruction only */
  55 #define SWIZZLE_NIL  7   /**< used during shader code gen (undefined value) */
  56 /*@}*/
  57
  58 #define MAKE_SWIZZLE4(a,b,c,d) (((a)<<0) | ((b)<<3) | ((c)<<6) | ((d)<<9))
  59 #define SWIZZLE_NOOP           MAKE_SWIZZLE4(0,1,2,3)
  60 #define GET_SWZ(swz, idx)      (((swz) >> ((idx)*3)) & 0x7)
  61 #define GET_BIT(msk, idx)      (((msk) >> (idx)) & 0x1)
  62 /** Determine if swz contains SWIZZLE_ZERO/ONE/NIL for any components. */
  63 #define HAS_EXTENDED_SWIZZLE(swz) (swz & 0x924)
  64
  65 #define SWIZZLE_XYZW MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W)
  66 #define SWIZZLE_XXXX MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)
  67 #define SWIZZLE_YYYY MAKE_SWIZZLE4(SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y)
  68 #define SWIZZLE_ZZZZ MAKE_SWIZZLE4(SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z)
  69 #define SWIZZLE_WWWW MAKE_SWIZZLE4(SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W)
  70
  71
  72 /**
  73  * Writemask values, 1 bit per component.
  74  */
  75 /*@{*/
  76 #define WRITEMASK_X     0x1
  77 #define WRITEMASK_Y     0x2
  78 #define WRITEMASK_XY    0x3
  79 #define WRITEMASK_Z     0x4
  80 #define WRITEMASK_XZ    0x5
  81 #define WRITEMASK_YZ    0x6
  82 #define WRITEMASK_XYZ   0x7
  83 #define WRITEMASK_W     0x8
  84 #define WRITEMASK_XW    0x9
  85 #define WRITEMASK_YW    0xa
  86 #define WRITEMASK_XYW   0xb
  87 #define WRITEMASK_ZW    0xc
  88 #define WRITEMASK_XZW   0xd
  89 #define WRITEMASK_YZW   0xe
  90 #define WRITEMASK_XYZW  0xf
  91 /*@}*/
  92
  93
  94 /**
  95  * Condition codes
  96  */
  97 /*@{*/
  98 #define COND_GT  1  /**< greater than zero */
  99 #define COND_EQ  2  /**< equal to zero */
 100 #define COND_LT  3  /**< less than zero */
 101 #define COND_UN  4  /**< unordered (NaN) */
 102 #define COND_GE  5  /**< greater than or equal to zero */
 103 #define COND_LE  6  /**< less than or equal to zero */
 104 #define COND_NE  7  /**< not equal to zero */
 105 #define COND_TR  8  /**< always true */
 106 #define COND_FL  9  /**< always false */
 107 /*@}*/
 108
 109
 110 /**
 111  * Instruction precision for GL_NV_fragment_program
 112  */
 113 /*@{*/
 114 #define FLOAT32  0x1
 115 #define FLOAT16  0x2
 116 #define FIXED12  0x4
 117 /*@}*/
 118
 119
 120 /**
 121  * Per-component negation masks
 122  */
 123 /*@{*/
 124 #define NEGATE_X    0x1
 125 #define NEGATE_Y    0x2
 126 #define NEGATE_Z    0x4
 127 #define NEGATE_W    0x8
 128 #define NEGATE_XYZ  0x7
 129 #define NEGATE_XYZW 0xf
 130 #define NEGATE_NONE 0x0
 131 /*@}*/
 132
 133
 134 /**
 135  * Program instruction opcodes for vertex, fragment and geometry programs.
 136  */
 137 typedef enum prog_opcode {
 138                      /* ARB_vp   ARB_fp   NV_vp   NV_fp     GLSL */
 139                      /*------------------------------------------*/
 140    OPCODE_NOP = 0,   /*                                      X   */
 141    OPCODE_ABS,       /*   X        X       1.1               X   */
 142    OPCODE_ADD,       /*   X        X       X       X         X   */
 143    OPCODE_ARL,       /*   X                X                 X   */
 144    OPCODE_BGNLOOP,   /*                                     opt  */
 145    OPCODE_BGNSUB,    /*                                     opt  */
 146    OPCODE_BRK,       /*                    2                opt  */
 147    OPCODE_CAL,       /*                    2       2        opt  */
 148    OPCODE_CMP,       /*            X                         X   */
 149    OPCODE_CONT,      /*                                     opt  */
 150    OPCODE_COS,       /*            X       2       X         X   */
 151    OPCODE_DDX,       /*                            X         X   */
 152    OPCODE_DDY,       /*                            X         X   */
 153    OPCODE_DP2,       /*                            2         X   */
 154    OPCODE_DP3,       /*   X        X       X       X         X   */
 155    OPCODE_DP4,       /*   X        X       X       X         X   */
 156    OPCODE_DPH,       /*   X        X       1.1                   */
 157    OPCODE_DST,       /*   X        X       X       X             */
 158    OPCODE_ELSE,      /*                                     opt  */
 159    OPCODE_END,       /*   X        X       X       X        opt  */
 160    OPCODE_ENDIF,     /*                                     opt  */
 161    OPCODE_ENDLOOP,   /*                                     opt  */
 162    OPCODE_ENDSUB,    /*                                     opt  */
 163    OPCODE_EX2,       /*   X        X       2       X         X   */
 164    OPCODE_EXP,       /*   X                X                     */
 165    OPCODE_FLR,       /*   X        X       2       X         X   */
 166    OPCODE_FRC,       /*   X        X       2       X         X   */
 167    OPCODE_IF,        /*                                     opt  */
 168    OPCODE_KIL,       /*            X                         X   */
 169    OPCODE_KIL_NV,    /*                            X         X   */
 170    OPCODE_LG2,       /*   X        X       2       X         X   */
 171    OPCODE_LIT,       /*   X        X       X       X             */
 172    OPCODE_LOG,       /*   X                X                     */
 173    OPCODE_LRP,       /*            X               X             */
 174    OPCODE_MAD,       /*   X        X       X       X         X   */
 175    OPCODE_MAX,       /*   X        X       X       X         X   */
 176    OPCODE_MIN,       /*   X        X       X       X         X   */
 177    OPCODE_MOV,       /*   X        X       X       X         X   */
 178    OPCODE_MUL,       /*   X        X       X       X         X   */
 179    OPCODE_NOISE1,    /*                                      X   */
 180    OPCODE_NOISE2,    /*                                      X   */
 181    OPCODE_NOISE3,    /*                                      X   */
 182    OPCODE_NOISE4,    /*                                      X   */
 183    OPCODE_POW,       /*   X        X               X         X   */
 184    OPCODE_RCP,       /*   X        X       X       X         X   */
 185    OPCODE_RET,       /*                    2       2        opt  */
 186    OPCODE_RSQ,       /*   X        X       X       X         X   */
 187    OPCODE_SCS,       /*            X                         X   */
 188    OPCODE_SEQ,       /*                    2       X         X   */
 189    OPCODE_SGE,       /*   X        X       X       X         X   */
 190    OPCODE_SGT,       /*                    2       X         X   */
 191    OPCODE_SIN,       /*            X       2       X         X   */
 192    OPCODE_SLE,       /*                    2       X         X   */
 193    OPCODE_SLT,       /*   X        X       X       X         X   */
 194    OPCODE_SNE,       /*                    2       X         X   */
 195    OPCODE_SSG,       /*                    2                 X   */
 196    OPCODE_SUB,       /*   X        X       1.1     X         X   */
 197    OPCODE_SWZ,       /*   X        X                         X   */
 198    OPCODE_TEX,       /*            X       3       X         X   */
 199    OPCODE_TXB,       /*            X       3                 X   */
 200    OPCODE_TXD,       /*                            X         X   */
 201    OPCODE_TXL,       /*                    3       2         X   */
 202    OPCODE_TXP,       /*            X                         X   */
 203    OPCODE_TXP_NV,    /*                    3       X             */
 204    OPCODE_TRUNC,     /*                                      X   */
 205    OPCODE_XPD,       /*   X        X                             */
 206    MAX_OPCODE
 207 } gl_inst_opcode;
 208
 209
 210 /**
 211  * Number of bits for the src/dst register Index field.
 212  * This limits the size of temp/uniform register files.
 213  */
 214 #define INST_INDEX_BITS 12
 215
 216
 217 /**
 218  * Instruction source register.
 219  */
 220 struct prog_src_register
 221 {
 222    GLuint File:4;       /**< One of the PROGRAM_* register file values. */
 223    GLint Index:(INST_INDEX_BITS+1); /**< Extra bit here for sign bit.
 224                                      * May be negative for relative addressing.
 225                                      */
 226    GLuint Swizzle:12;
 227    GLuint RelAddr:1;
 228
 229    /** Take the component-wise absolute value */
 230    GLuint Abs:1;
 231
 232    /**
 233     * Post-Abs negation.
 234     * This will either be NEGATE_NONE or NEGATE_XYZW, except for the SWZ
 235     * instruction which allows per-component negation.
 236     */
 237    GLuint Negate:4;
 238
 239    /**
 240     * Is the register two-dimensional.
 241     * Two dimensional registers are of the
 242     * REGISTER[index][index2] format.
 243     * They are used by the geometry shaders where
 244     * the first index is the index within an array
 245     * and the second index is the semantic of the
 246     * array, e.g. gl_PositionIn[index] would become
 247     * INPUT[index][gl_PositionIn]
 248     */
 249    GLuint HasIndex2:1;
 250    GLuint RelAddr2:1;
 251    GLint Index2:(INST_INDEX_BITS+1); /**< Extra bit here for sign bit.
 252                                        * May be negative for relative
 253                                        * addressing. */
 254 };
 255
 256
 257 /**
 258  * Instruction destination register.
 259  */
 260 struct prog_dst_register
 261 {
 262    GLuint File:4;      /**< One of the PROGRAM_* register file values */
 263    GLuint Index:INST_INDEX_BITS;  /**< Unsigned, never negative */
 264    GLuint WriteMask:4;
 265    GLuint RelAddr:1;
 266
 267    /**
 268     * \name Conditional destination update control.
 269     *
 270     * \since
 271     * NV_fragment_program_option, NV_vertex_program2, NV_vertex_program2_option.
 272     */
 273    /*@{*/
 274    /**
 275     * Takes one of the 9 possible condition values (EQ, FL, GT, GE, LE, LT,
 276     * NE, TR, or UN).  Dest reg is only written to if the matching
 277     * (swizzled) condition code value passes.  When a conditional update mask
 278     * is not specified, this will be \c COND_TR.
 279     */
 280    GLuint CondMask:4;
 281
 282    /**
 283     * Condition code swizzle value.
 284     */
 285    GLuint CondSwizzle:12;
 286 };
 287
 288
 289 /**
 290  * Vertex/fragment program instruction.
 291  */
 292 struct prog_instruction
 293 {
 294    gl_inst_opcode Opcode;
 295    struct prog_src_register SrcReg[3];
 296    struct prog_dst_register DstReg;
 297
 298    /**
 299     * Indicates that the instruction should update the condition code
 300     * register.
 301     *
 302     * \since
 303     * NV_fragment_program_option, NV_vertex_program2, NV_vertex_program2_option.
 304     */
 305    GLuint CondUpdate:1;
 306
 307    /**
 308     * If prog_instruction::CondUpdate is \c GL_TRUE, this value selects the
 309     * condition code register that is to be updated.
 310     *
 311     * In GL_NV_fragment_program or GL_NV_vertex_program2 mode, only condition
 312     * code register 0 is available.  In GL_NV_vertex_program3 mode, condition
 313     * code registers 0 and 1 are available.
 314     *
 315     * \since
 316     * NV_fragment_program_option, NV_vertex_program2, NV_vertex_program2_option.
 317     */
 318    GLuint CondDst:1;
 319
 320    /**
 321     * Saturate each value of the vectored result to the range [0,1].
 322     *
 323     * \since
 324     * NV_fragment_program_option, NV_vertex_program3.
 325     */
 326    GLuint Saturate:1;
 327
 328    /**
 329     * Per-instruction selectable precision: FLOAT32, FLOAT16, FIXED12.
 330     *
 331     * \since
 332     * NV_fragment_program_option.
 333     */
 334    GLuint Precision:3;
 335
 336    /**
 337     * \name Extra fields for TEX, TXB, TXD, TXL, TXP instructions.
 338     */
 339    /*@{*/
 340    /** Source texture unit. */
 341    GLuint TexSrcUnit:5;
 342
 343    /** Source texture target, one of TEXTURE_{1D,2D,3D,CUBE,RECT}_INDEX */
 344    GLuint TexSrcTarget:4;
 345
 346    /** True if tex instruction should do shadow comparison */
 347    GLuint TexShadow:1;
 348    /*@}*/
 349
 350    /**
 351     * For BRA and CAL instructions, the location to jump to.
 352     * For BGNLOOP, points to ENDLOOP (and vice-versa).
 353     * For BRK, points to ENDLOOP
 354     * For IF, points to ELSE or ENDIF.
 355     * For ELSE, points to ENDIF.
 356     */
 357    GLint BranchTarget;
 358
 359    /** for debugging purposes */
 360    const char *Comment;
 361 };
 362
 363
 364 #ifdef __cplusplus
 365 extern "C" {
 366 #endif
 367
 368 extern void
 369 _mesa_init_instructions(struct prog_instruction *inst, GLuint count);
 370
 371 extern struct prog_instruction *
 372 _mesa_alloc_instructions(GLuint numInst);
 373
 374 extern struct prog_instruction *
 375 _mesa_copy_instructions(struct prog_instruction *dest,
 376                         const struct prog_instruction *src, GLuint n);
 377
 378 extern void
 379 _mesa_free_instructions(struct prog_instruction *inst, GLuint count);
 380
 381 extern GLuint
 382 _mesa_num_inst_src_regs(gl_inst_opcode opcode);
 383
 384 extern GLuint
 385 _mesa_num_inst_dst_regs(gl_inst_opcode opcode);
 386
 387 extern GLboolean
 388 _mesa_is_tex_instruction(gl_inst_opcode opcode);
 389
 390 extern GLboolean
 391 _mesa_check_soa_dependencies(const struct prog_instruction *inst);
 392
 393 extern const char *
 394 _mesa_opcode_string(gl_inst_opcode opcode);
 395
 396
 397 #ifdef __cplusplus
 398 } /* extern "C" */
 399 #endif
 400
 401 #endif /* PROG_INSTRUCTION_H */