src/gallium/auxiliary/tgsi/tgsi_exec.h

   1 /**************************************************************************
   2  *
   3  * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
   4  * All Rights Reserved.
   5  * Copyright 2009-2010 VMware, Inc.  All rights Reserved.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a
   8  * copy of this software and associated documentation files (the
   9  * "Software"), to deal in the Software without restriction, including
  10  * without limitation the rights to use, copy, modify, merge, publish,
  11  * distribute, sub license, and/or sell copies of the Software, and to
  12  * permit persons to whom the Software is furnished to do so, subject to
  13  * the following conditions:
  14  *
  15  * The above copyright notice and this permission notice (including the
  16  * next paragraph) shall be included in all copies or substantial portions
  17  * of the Software.
  18  *
  19  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  20  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  21  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  22  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  23  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  24  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  25  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  26  *
  27  **************************************************************************/
  28
  29 #ifndef TGSI_EXEC_H
  30 #define TGSI_EXEC_H
  31
  32 #include "pipe/p_compiler.h"
  33 #include "pipe/p_state.h"
  34
  35 #if defined __cplusplus
  36 extern "C" {
  37 #endif
  38
  39
  40 #define NUM_CHANNELS 4  /* R,G,B,A */
  41 #define QUAD_SIZE    4  /* 4 pixel/quad */
  42
  43
  44 /**
  45   * Registers may be treated as float, signed int or unsigned int.
  46   */
  47 union tgsi_exec_channel
  48 {
  49    float    f[QUAD_SIZE];
  50    int      i[QUAD_SIZE];
  51    unsigned u[QUAD_SIZE];
  52 };
  53
  54 /**
  55   * A vector[RGBA] of channels[4 pixels]
  56   */
  57 struct tgsi_exec_vector
  58 {
  59    union tgsi_exec_channel xyzw[NUM_CHANNELS];
  60 };
  61
  62 /**
  63  * For fragment programs, information for computing fragment input
  64  * values from plane equation of the triangle/line.
  65  */
  66 struct tgsi_interp_coef
  67 {
  68    float a0[NUM_CHANNELS];      /* in an xyzw layout */
  69    float dadx[NUM_CHANNELS];
  70    float dady[NUM_CHANNELS];
  71 };
  72
  73 enum tgsi_sampler_control {
  74    tgsi_sampler_lod_bias,
  75    tgsi_sampler_lod_explicit
  76 };
  77
  78 /**
  79  * Information for sampling textures, which must be implemented
  80  * by code outside the TGSI executor.
  81  */
  82 struct tgsi_sampler
  83 {
  84    /** Get samples for four fragments in a quad */
  85    void (*get_samples)(struct tgsi_sampler *sampler,
  86                        const float s[QUAD_SIZE],
  87                        const float t[QUAD_SIZE],
  88                        const float p[QUAD_SIZE],
  89                        const float c0[QUAD_SIZE],
  90                        enum tgsi_sampler_control control,
  91                        float rgba[NUM_CHANNELS][QUAD_SIZE]);
  92 };
  93
  94 #define TGSI_EXEC_NUM_TEMPS       128
  95 #define TGSI_EXEC_NUM_IMMEDIATES  256
  96 #define TGSI_EXEC_NUM_TEMP_ARRAYS 8
  97
  98 /*
  99  * Locations of various utility registers (_I = Index, _C = Channel)
 100  */
 101 #define TGSI_EXEC_TEMP_00000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
 102 #define TGSI_EXEC_TEMP_00000000_C   0
 103
 104 #define TGSI_EXEC_TEMP_7FFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
 105 #define TGSI_EXEC_TEMP_7FFFFFFF_C   1
 106
 107 #define TGSI_EXEC_TEMP_80000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
 108 #define TGSI_EXEC_TEMP_80000000_C   2
 109
 110 #define TGSI_EXEC_TEMP_FFFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
 111 #define TGSI_EXEC_TEMP_FFFFFFFF_C   3
 112
 113 #define TGSI_EXEC_TEMP_ONE_I        (TGSI_EXEC_NUM_TEMPS + 1)
 114 #define TGSI_EXEC_TEMP_ONE_C        0
 115
 116 #define TGSI_EXEC_TEMP_TWO_I        (TGSI_EXEC_NUM_TEMPS + 1)
 117 #define TGSI_EXEC_TEMP_TWO_C        1
 118
 119 #define TGSI_EXEC_TEMP_128_I        (TGSI_EXEC_NUM_TEMPS + 1)
 120 #define TGSI_EXEC_TEMP_128_C        2
 121
 122 #define TGSI_EXEC_TEMP_MINUS_128_I  (TGSI_EXEC_NUM_TEMPS + 1)
 123 #define TGSI_EXEC_TEMP_MINUS_128_C  3
 124
 125 #define TGSI_EXEC_TEMP_KILMASK_I    (TGSI_EXEC_NUM_TEMPS + 2)
 126 #define TGSI_EXEC_TEMP_KILMASK_C    0
 127
 128 #define TGSI_EXEC_TEMP_OUTPUT_I     (TGSI_EXEC_NUM_TEMPS + 2)
 129 #define TGSI_EXEC_TEMP_OUTPUT_C     1
 130
 131 #define TGSI_EXEC_TEMP_PRIMITIVE_I  (TGSI_EXEC_NUM_TEMPS + 2)
 132 #define TGSI_EXEC_TEMP_PRIMITIVE_C  2
 133
 134 /* NVIDIA condition code (CC) vector
 135  */
 136 #define TGSI_EXEC_CC_GT       0x01
 137 #define TGSI_EXEC_CC_EQ       0x02
 138 #define TGSI_EXEC_CC_LT       0x04
 139 #define TGSI_EXEC_CC_UN       0x08
 140
 141 #define TGSI_EXEC_CC_X_MASK   0x000000ff
 142 #define TGSI_EXEC_CC_X_SHIFT  0
 143 #define TGSI_EXEC_CC_Y_MASK   0x0000ff00
 144 #define TGSI_EXEC_CC_Y_SHIFT  8
 145 #define TGSI_EXEC_CC_Z_MASK   0x00ff0000
 146 #define TGSI_EXEC_CC_Z_SHIFT  16
 147 #define TGSI_EXEC_CC_W_MASK   0xff000000
 148 #define TGSI_EXEC_CC_W_SHIFT  24
 149
 150 #define TGSI_EXEC_TEMP_CC_I         (TGSI_EXEC_NUM_TEMPS + 2)
 151 #define TGSI_EXEC_TEMP_CC_C         3
 152
 153 #define TGSI_EXEC_TEMP_THREE_I      (TGSI_EXEC_NUM_TEMPS + 3)
 154 #define TGSI_EXEC_TEMP_THREE_C      0
 155
 156 #define TGSI_EXEC_TEMP_HALF_I       (TGSI_EXEC_NUM_TEMPS + 3)
 157 #define TGSI_EXEC_TEMP_HALF_C       1
 158
 159 /* execution mask, each value is either 0 or ~0 */
 160 #define TGSI_EXEC_MASK_I            (TGSI_EXEC_NUM_TEMPS + 3)
 161 #define TGSI_EXEC_MASK_C            2
 162
 163 /* 4 register buffer for various purposes */
 164 #define TGSI_EXEC_TEMP_R0           (TGSI_EXEC_NUM_TEMPS + 4)
 165 #define TGSI_EXEC_NUM_TEMP_R        4
 166
 167 #define TGSI_EXEC_TEMP_ADDR         (TGSI_EXEC_NUM_TEMPS + 8)
 168 #define TGSI_EXEC_NUM_ADDRS         1
 169
 170 /* predicate register */
 171 #define TGSI_EXEC_TEMP_P0           (TGSI_EXEC_NUM_TEMPS + 9)
 172 #define TGSI_EXEC_NUM_PREDS         1
 173
 174 #define TGSI_EXEC_NUM_TEMP_EXTRAS   10
 175
 176
 177
 178 #define TGSI_EXEC_MAX_NESTING  32
 179 #define TGSI_EXEC_MAX_COND_NESTING  TGSI_EXEC_MAX_NESTING
 180 #define TGSI_EXEC_MAX_LOOP_NESTING  TGSI_EXEC_MAX_NESTING
 181 #define TGSI_EXEC_MAX_SWITCH_NESTING TGSI_EXEC_MAX_NESTING
 182 #define TGSI_EXEC_MAX_CALL_NESTING  TGSI_EXEC_MAX_NESTING
 183
 184 /* The maximum number of input attributes per vertex. For 2D
 185  * input register files, this is the stride between two 1D
 186  * arrays.
 187  */
 188 #define TGSI_EXEC_MAX_INPUT_ATTRIBS 17
 189
 190 /* The maximum number of constant vectors per constant buffer.
 191  */
 192 #define TGSI_EXEC_MAX_CONST_BUFFER  4096
 193
 194 /* The maximum number of vertices per primitive */
 195 #define TGSI_MAX_PRIM_VERTICES 6
 196
 197 /* The maximum number of primitives to be generated */
 198 #define TGSI_MAX_PRIMITIVES 64
 199
 200 /* The maximum total number of vertices */
 201 #define TGSI_MAX_TOTAL_VERTICES (TGSI_MAX_PRIM_VERTICES * TGSI_MAX_PRIMITIVES * PIPE_MAX_ATTRIBS)
 202
 203 /** function call/activation record */
 204 struct tgsi_call_record
 205 {
 206    uint CondStackTop;
 207    uint LoopStackTop;
 208    uint ContStackTop;
 209    int SwitchStackTop;
 210    int BreakStackTop;
 211    uint ReturnAddr;
 212 };
 213
 214
 215 /* Switch-case block state. */
 216 struct tgsi_switch_record {
 217    uint mask;                          /**< execution mask */
 218    union tgsi_exec_channel selector;   /**< a value case statements are compared to */
 219    uint defaultMask;                   /**< non-execute mask for default case */
 220 };
 221
 222
 223 enum tgsi_break_type {
 224    TGSI_EXEC_BREAK_INSIDE_LOOP,
 225    TGSI_EXEC_BREAK_INSIDE_SWITCH
 226 };
 227
 228
 229 #define TGSI_EXEC_MAX_BREAK_STACK (TGSI_EXEC_MAX_LOOP_NESTING + TGSI_EXEC_MAX_SWITCH_NESTING)
 230
 231
 232 /**
 233  * Run-time virtual machine state for executing TGSI shader.
 234  */
 235 struct tgsi_exec_machine
 236 {
 237    /* Total = program temporaries + internal temporaries
 238     */
 239    struct tgsi_exec_vector       Temps[TGSI_EXEC_NUM_TEMPS +
 240                                        TGSI_EXEC_NUM_TEMP_EXTRAS];
 241    struct tgsi_exec_vector       TempArray[TGSI_EXEC_NUM_TEMP_ARRAYS][TGSI_EXEC_NUM_TEMPS];
 242
 243    float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
 244
 245    float                         ImmArray[TGSI_EXEC_NUM_IMMEDIATES][4];
 246
 247    struct tgsi_exec_vector       Inputs[TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS];
 248    struct tgsi_exec_vector       Outputs[TGSI_MAX_TOTAL_VERTICES];
 249
 250    struct tgsi_exec_vector       *Addrs;
 251    struct tgsi_exec_vector       *Predicates;
 252
 253    struct tgsi_sampler           **Samplers;
 254
 255    unsigned                      ImmLimit;
 256
 257    const void *Consts[PIPE_MAX_CONSTANT_BUFFERS];
 258    unsigned ConstsSize[PIPE_MAX_CONSTANT_BUFFERS];
 259
 260    const struct tgsi_token       *Tokens;   /**< Declarations, instructions */
 261    unsigned                      Processor; /**< TGSI_PROCESSOR_x */
 262
 263    /* GEOMETRY processor only. */
 264    unsigned                      *Primitives;
 265    unsigned                       NumOutputs;
 266    unsigned                       MaxGeometryShaderOutputs;
 267
 268    /* FRAGMENT processor only. */
 269    const struct tgsi_interp_coef *InterpCoefs;
 270    struct tgsi_exec_vector       QuadPos;
 271    float                         Face;    /**< +1 if front facing, -1 if back facing */
 272
 273    /* Conditional execution masks */
 274    uint CondMask;  /**< For IF/ELSE/ENDIF */
 275    uint LoopMask;  /**< For BGNLOOP/ENDLOOP */
 276    uint ContMask;  /**< For loop CONT statements */
 277    uint FuncMask;  /**< For function calls */
 278    uint ExecMask;  /**< = CondMask & LoopMask */
 279
 280    /* Current switch-case state. */
 281    struct tgsi_switch_record Switch;
 282
 283    /* Current break type. */
 284    enum tgsi_break_type BreakType;
 285
 286    /** Condition mask stack (for nested conditionals) */
 287    uint CondStack[TGSI_EXEC_MAX_COND_NESTING];
 288    int CondStackTop;
 289
 290    /** Loop mask stack (for nested loops) */
 291    uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
 292    int LoopStackTop;
 293
 294    /** Loop label stack */
 295    uint LoopLabelStack[TGSI_EXEC_MAX_LOOP_NESTING];
 296    int LoopLabelStackTop;
 297
 298    /** Loop continue mask stack (see comments in tgsi_exec.c) */
 299    uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
 300    int ContStackTop;
 301
 302    /** Switch case stack */
 303    struct tgsi_switch_record SwitchStack[TGSI_EXEC_MAX_SWITCH_NESTING];
 304    int SwitchStackTop;
 305
 306    enum tgsi_break_type BreakStack[TGSI_EXEC_MAX_BREAK_STACK];
 307    int BreakStackTop;
 308
 309    /** Function execution mask stack (for executing subroutine code) */
 310    uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
 311    int FuncStackTop;
 312
 313    /** Function call stack for saving/restoring the program counter */
 314    struct tgsi_call_record CallStack[TGSI_EXEC_MAX_CALL_NESTING];
 315    int CallStackTop;
 316
 317    struct tgsi_full_instruction *Instructions;
 318    uint NumInstructions;
 319
 320    struct tgsi_full_declaration *Declarations;
 321    uint NumDeclarations;
 322
 323 };
 324
 325 struct tgsi_exec_machine *
 326 tgsi_exec_machine_create( void );
 327
 328 void
 329 tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach);
 330
 331
 332 void
 333 tgsi_exec_machine_bind_shader(
 334    struct tgsi_exec_machine *mach,
 335    const struct tgsi_token *tokens,
 336    uint numSamplers,
 337    struct tgsi_sampler **samplers);
 338
 339 uint
 340 tgsi_exec_machine_run(
 341    struct tgsi_exec_machine *mach );
 342
 343
 344 void
 345 tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach);
 346
 347
 348 boolean
 349 tgsi_check_soa_dependencies(const struct tgsi_full_instruction *inst);
 350
 351
 352 static INLINE void
 353 tgsi_set_kill_mask(struct tgsi_exec_machine *mach, unsigned mask)
 354 {
 355    mach->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0] =
 356       mask;
 357 }
 358
 359
 360 /** Set execution mask values prior to executing the shader */
 361 static INLINE void
 362 tgsi_set_exec_mask(struct tgsi_exec_machine *mach,
 363                    boolean ch0, boolean ch1, boolean ch2, boolean ch3)
 364 {
 365    int *mask = mach->Temps[TGSI_EXEC_MASK_I].xyzw[TGSI_EXEC_MASK_C].i;
 366    mask[0] = ch0 ? ~0 : 0;
 367    mask[1] = ch1 ? ~0 : 0;
 368    mask[2] = ch2 ? ~0 : 0;
 369    mask[3] = ch3 ? ~0 : 0;
 370 }
 371
 372
 373 extern void
 374 tgsi_exec_set_constant_buffers(struct tgsi_exec_machine *mach,
 375                                unsigned num_bufs,
 376                                const void **bufs,
 377                                const unsigned *buf_sizes);
 378
 379
 380 static INLINE int
 381 tgsi_exec_get_shader_param(enum pipe_shader_cap param)
 382 {
 383    switch(param) {
 384    case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 385    case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 386    case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 387    case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 388       return INT_MAX;
 389    case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 390       return TGSI_EXEC_MAX_NESTING;
 391    case PIPE_SHADER_CAP_MAX_INPUTS:
 392       return TGSI_EXEC_MAX_INPUT_ATTRIBS;
 393    case PIPE_SHADER_CAP_MAX_CONSTS:
 394       return TGSI_EXEC_MAX_CONST_BUFFER;
 395    case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 396       return PIPE_MAX_CONSTANT_BUFFERS;
 397    case PIPE_SHADER_CAP_MAX_TEMPS:
 398       return TGSI_EXEC_NUM_TEMPS;
 399    case PIPE_SHADER_CAP_MAX_ADDRS:
 400       return TGSI_EXEC_NUM_ADDRS;
 401    case PIPE_SHADER_CAP_MAX_PREDS:
 402       return TGSI_EXEC_NUM_PREDS;
 403    case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 404       return 1;
 405    default:
 406       return 0;
 407    }
 408 }
 409
 410 #if defined __cplusplus
 411 } /* extern "C" */
 412 #endif
 413
 414 #endif /* TGSI_EXEC_H */