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 #define TGSI_EXEC_TEMP_THREE_I      (TGSI_EXEC_NUM_TEMPS + 2)
 135 #define TGSI_EXEC_TEMP_THREE_C      3
 136
 137 #define TGSI_EXEC_TEMP_HALF_I       (TGSI_EXEC_NUM_TEMPS + 3)
 138 #define TGSI_EXEC_TEMP_HALF_C       0
 139
 140 /* execution mask, each value is either 0 or ~0 */
 141 #define TGSI_EXEC_MASK_I            (TGSI_EXEC_NUM_TEMPS + 3)
 142 #define TGSI_EXEC_MASK_C            1
 143
 144 /* 4 register buffer for various purposes */
 145 #define TGSI_EXEC_TEMP_R0           (TGSI_EXEC_NUM_TEMPS + 4)
 146 #define TGSI_EXEC_NUM_TEMP_R        4
 147
 148 #define TGSI_EXEC_TEMP_ADDR         (TGSI_EXEC_NUM_TEMPS + 8)
 149 #define TGSI_EXEC_NUM_ADDRS         1
 150
 151 /* predicate register */
 152 #define TGSI_EXEC_TEMP_P0           (TGSI_EXEC_NUM_TEMPS + 9)
 153 #define TGSI_EXEC_NUM_PREDS         1
 154
 155 #define TGSI_EXEC_NUM_TEMP_EXTRAS   10
 156
 157
 158
 159 #define TGSI_EXEC_MAX_NESTING  32
 160 #define TGSI_EXEC_MAX_COND_NESTING  TGSI_EXEC_MAX_NESTING
 161 #define TGSI_EXEC_MAX_LOOP_NESTING  TGSI_EXEC_MAX_NESTING
 162 #define TGSI_EXEC_MAX_SWITCH_NESTING TGSI_EXEC_MAX_NESTING
 163 #define TGSI_EXEC_MAX_CALL_NESTING  TGSI_EXEC_MAX_NESTING
 164
 165 /* The maximum number of input attributes per vertex. For 2D
 166  * input register files, this is the stride between two 1D
 167  * arrays.
 168  */
 169 #define TGSI_EXEC_MAX_INPUT_ATTRIBS 17
 170
 171 /* The maximum number of constant vectors per constant buffer.
 172  */
 173 #define TGSI_EXEC_MAX_CONST_BUFFER  4096
 174
 175 /* The maximum number of vertices per primitive */
 176 #define TGSI_MAX_PRIM_VERTICES 6
 177
 178 /* The maximum number of primitives to be generated */
 179 #define TGSI_MAX_PRIMITIVES 64
 180
 181 /* The maximum total number of vertices */
 182 #define TGSI_MAX_TOTAL_VERTICES (TGSI_MAX_PRIM_VERTICES * TGSI_MAX_PRIMITIVES * PIPE_MAX_ATTRIBS)
 183
 184 /** function call/activation record */
 185 struct tgsi_call_record
 186 {
 187    uint CondStackTop;
 188    uint LoopStackTop;
 189    uint ContStackTop;
 190    int SwitchStackTop;
 191    int BreakStackTop;
 192    uint ReturnAddr;
 193 };
 194
 195
 196 /* Switch-case block state. */
 197 struct tgsi_switch_record {
 198    uint mask;                          /**< execution mask */
 199    union tgsi_exec_channel selector;   /**< a value case statements are compared to */
 200    uint defaultMask;                   /**< non-execute mask for default case */
 201 };
 202
 203
 204 enum tgsi_break_type {
 205    TGSI_EXEC_BREAK_INSIDE_LOOP,
 206    TGSI_EXEC_BREAK_INSIDE_SWITCH
 207 };
 208
 209
 210 #define TGSI_EXEC_MAX_BREAK_STACK (TGSI_EXEC_MAX_LOOP_NESTING + TGSI_EXEC_MAX_SWITCH_NESTING)
 211
 212
 213 /**
 214  * Run-time virtual machine state for executing TGSI shader.
 215  */
 216 struct tgsi_exec_machine
 217 {
 218    /* Total = program temporaries + internal temporaries
 219     */
 220    struct tgsi_exec_vector       Temps[TGSI_EXEC_NUM_TEMPS +
 221                                        TGSI_EXEC_NUM_TEMP_EXTRAS];
 222    struct tgsi_exec_vector       TempArray[TGSI_EXEC_NUM_TEMP_ARRAYS][TGSI_EXEC_NUM_TEMPS];
 223
 224    float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
 225
 226    float                         ImmArray[TGSI_EXEC_NUM_IMMEDIATES][4];
 227
 228    struct tgsi_exec_vector       Inputs[TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS];
 229    struct tgsi_exec_vector       Outputs[TGSI_MAX_TOTAL_VERTICES];
 230
 231    struct tgsi_exec_vector       *Addrs;
 232    struct tgsi_exec_vector       *Predicates;
 233
 234    struct tgsi_sampler           **Samplers;
 235
 236    unsigned                      ImmLimit;
 237
 238    const void *Consts[PIPE_MAX_CONSTANT_BUFFERS];
 239    unsigned ConstsSize[PIPE_MAX_CONSTANT_BUFFERS];
 240
 241    const struct tgsi_token       *Tokens;   /**< Declarations, instructions */
 242    unsigned                      Processor; /**< TGSI_PROCESSOR_x */
 243
 244    /* GEOMETRY processor only. */
 245    unsigned                      *Primitives;
 246    unsigned                       NumOutputs;
 247    unsigned                       MaxGeometryShaderOutputs;
 248
 249    /* FRAGMENT processor only. */
 250    const struct tgsi_interp_coef *InterpCoefs;
 251    struct tgsi_exec_vector       QuadPos;
 252    float                         Face;    /**< +1 if front facing, -1 if back facing */
 253
 254    /* Conditional execution masks */
 255    uint CondMask;  /**< For IF/ELSE/ENDIF */
 256    uint LoopMask;  /**< For BGNLOOP/ENDLOOP */
 257    uint ContMask;  /**< For loop CONT statements */
 258    uint FuncMask;  /**< For function calls */
 259    uint ExecMask;  /**< = CondMask & LoopMask */
 260
 261    /* Current switch-case state. */
 262    struct tgsi_switch_record Switch;
 263
 264    /* Current break type. */
 265    enum tgsi_break_type BreakType;
 266
 267    /** Condition mask stack (for nested conditionals) */
 268    uint CondStack[TGSI_EXEC_MAX_COND_NESTING];
 269    int CondStackTop;
 270
 271    /** Loop mask stack (for nested loops) */
 272    uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
 273    int LoopStackTop;
 274
 275    /** Loop label stack */
 276    uint LoopLabelStack[TGSI_EXEC_MAX_LOOP_NESTING];
 277    int LoopLabelStackTop;
 278
 279    /** Loop continue mask stack (see comments in tgsi_exec.c) */
 280    uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
 281    int ContStackTop;
 282
 283    /** Switch case stack */
 284    struct tgsi_switch_record SwitchStack[TGSI_EXEC_MAX_SWITCH_NESTING];
 285    int SwitchStackTop;
 286
 287    enum tgsi_break_type BreakStack[TGSI_EXEC_MAX_BREAK_STACK];
 288    int BreakStackTop;
 289
 290    /** Function execution mask stack (for executing subroutine code) */
 291    uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
 292    int FuncStackTop;
 293
 294    /** Function call stack for saving/restoring the program counter */
 295    struct tgsi_call_record CallStack[TGSI_EXEC_MAX_CALL_NESTING];
 296    int CallStackTop;
 297
 298    struct tgsi_full_instruction *Instructions;
 299    uint NumInstructions;
 300
 301    struct tgsi_full_declaration *Declarations;
 302    uint NumDeclarations;
 303
 304 };
 305
 306 struct tgsi_exec_machine *
 307 tgsi_exec_machine_create( void );
 308
 309 void
 310 tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach);
 311
 312
 313 void
 314 tgsi_exec_machine_bind_shader(
 315    struct tgsi_exec_machine *mach,
 316    const struct tgsi_token *tokens,
 317    uint numSamplers,
 318    struct tgsi_sampler **samplers);
 319
 320 uint
 321 tgsi_exec_machine_run(
 322    struct tgsi_exec_machine *mach );
 323
 324
 325 void
 326 tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach);
 327
 328
 329 boolean
 330 tgsi_check_soa_dependencies(const struct tgsi_full_instruction *inst);
 331
 332
 333 static INLINE void
 334 tgsi_set_kill_mask(struct tgsi_exec_machine *mach, unsigned mask)
 335 {
 336    mach->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0] =
 337       mask;
 338 }
 339
 340
 341 /** Set execution mask values prior to executing the shader */
 342 static INLINE void
 343 tgsi_set_exec_mask(struct tgsi_exec_machine *mach,
 344                    boolean ch0, boolean ch1, boolean ch2, boolean ch3)
 345 {
 346    int *mask = mach->Temps[TGSI_EXEC_MASK_I].xyzw[TGSI_EXEC_MASK_C].i;
 347    mask[0] = ch0 ? ~0 : 0;
 348    mask[1] = ch1 ? ~0 : 0;
 349    mask[2] = ch2 ? ~0 : 0;
 350    mask[3] = ch3 ? ~0 : 0;
 351 }
 352
 353
 354 extern void
 355 tgsi_exec_set_constant_buffers(struct tgsi_exec_machine *mach,
 356                                unsigned num_bufs,
 357                                const void **bufs,
 358                                const unsigned *buf_sizes);
 359
 360
 361 static INLINE int
 362 tgsi_exec_get_shader_param(enum pipe_shader_cap param)
 363 {
 364    switch(param) {
 365    case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 366    case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 367    case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 368    case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 369       return INT_MAX;
 370    case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 371       return TGSI_EXEC_MAX_NESTING;
 372    case PIPE_SHADER_CAP_MAX_INPUTS:
 373       return TGSI_EXEC_MAX_INPUT_ATTRIBS;
 374    case PIPE_SHADER_CAP_MAX_CONSTS:
 375       return TGSI_EXEC_MAX_CONST_BUFFER;
 376    case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 377       return PIPE_MAX_CONSTANT_BUFFERS;
 378    case PIPE_SHADER_CAP_MAX_TEMPS:
 379       return TGSI_EXEC_NUM_TEMPS;
 380    case PIPE_SHADER_CAP_MAX_ADDRS:
 381       return TGSI_EXEC_NUM_ADDRS;
 382    case PIPE_SHADER_CAP_MAX_PREDS:
 383       return TGSI_EXEC_NUM_PREDS;
 384    case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 385       return 1;
 386    case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 387    case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 388    case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 389    case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 390       return 1;
 391    case PIPE_SHADER_CAP_SUBROUTINES:
 392       return 1;
 393    default:
 394       return 0;
 395    }
 396 }
 397
 398 #if defined __cplusplus
 399 } /* extern "C" */
 400 #endif
 401
 402 #endif /* TGSI_EXEC_H */