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
  97 /*
  98  * Locations of various utility registers (_I = Index, _C = Channel)
  99  */
 100 #define TGSI_EXEC_TEMP_00000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
 101 #define TGSI_EXEC_TEMP_00000000_C   0
 102
 103 #define TGSI_EXEC_TEMP_7FFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
 104 #define TGSI_EXEC_TEMP_7FFFFFFF_C   1
 105
 106 #define TGSI_EXEC_TEMP_80000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
 107 #define TGSI_EXEC_TEMP_80000000_C   2
 108
 109 #define TGSI_EXEC_TEMP_FFFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
 110 #define TGSI_EXEC_TEMP_FFFFFFFF_C   3
 111
 112 #define TGSI_EXEC_TEMP_ONE_I        (TGSI_EXEC_NUM_TEMPS + 1)
 113 #define TGSI_EXEC_TEMP_ONE_C        0
 114
 115 #define TGSI_EXEC_TEMP_TWO_I        (TGSI_EXEC_NUM_TEMPS + 1)
 116 #define TGSI_EXEC_TEMP_TWO_C        1
 117
 118 #define TGSI_EXEC_TEMP_128_I        (TGSI_EXEC_NUM_TEMPS + 1)
 119 #define TGSI_EXEC_TEMP_128_C        2
 120
 121 #define TGSI_EXEC_TEMP_MINUS_128_I  (TGSI_EXEC_NUM_TEMPS + 1)
 122 #define TGSI_EXEC_TEMP_MINUS_128_C  3
 123
 124 #define TGSI_EXEC_TEMP_KILMASK_I    (TGSI_EXEC_NUM_TEMPS + 2)
 125 #define TGSI_EXEC_TEMP_KILMASK_C    0
 126
 127 #define TGSI_EXEC_TEMP_OUTPUT_I     (TGSI_EXEC_NUM_TEMPS + 2)
 128 #define TGSI_EXEC_TEMP_OUTPUT_C     1
 129
 130 #define TGSI_EXEC_TEMP_PRIMITIVE_I  (TGSI_EXEC_NUM_TEMPS + 2)
 131 #define TGSI_EXEC_TEMP_PRIMITIVE_C  2
 132
 133 /* NVIDIA condition code (CC) vector
 134  */
 135 #define TGSI_EXEC_CC_GT       0x01
 136 #define TGSI_EXEC_CC_EQ       0x02
 137 #define TGSI_EXEC_CC_LT       0x04
 138 #define TGSI_EXEC_CC_UN       0x08
 139
 140 #define TGSI_EXEC_CC_X_MASK   0x000000ff
 141 #define TGSI_EXEC_CC_X_SHIFT  0
 142 #define TGSI_EXEC_CC_Y_MASK   0x0000ff00
 143 #define TGSI_EXEC_CC_Y_SHIFT  8
 144 #define TGSI_EXEC_CC_Z_MASK   0x00ff0000
 145 #define TGSI_EXEC_CC_Z_SHIFT  16
 146 #define TGSI_EXEC_CC_W_MASK   0xff000000
 147 #define TGSI_EXEC_CC_W_SHIFT  24
 148
 149 #define TGSI_EXEC_TEMP_CC_I         (TGSI_EXEC_NUM_TEMPS + 2)
 150 #define TGSI_EXEC_TEMP_CC_C         3
 151
 152 #define TGSI_EXEC_TEMP_THREE_I      (TGSI_EXEC_NUM_TEMPS + 3)
 153 #define TGSI_EXEC_TEMP_THREE_C      0
 154
 155 #define TGSI_EXEC_TEMP_HALF_I       (TGSI_EXEC_NUM_TEMPS + 3)
 156 #define TGSI_EXEC_TEMP_HALF_C       1
 157
 158 /* execution mask, each value is either 0 or ~0 */
 159 #define TGSI_EXEC_MASK_I            (TGSI_EXEC_NUM_TEMPS + 3)
 160 #define TGSI_EXEC_MASK_C            2
 161
 162 /* 4 register buffer for various purposes */
 163 #define TGSI_EXEC_TEMP_R0           (TGSI_EXEC_NUM_TEMPS + 4)
 164 #define TGSI_EXEC_NUM_TEMP_R        4
 165
 166 #define TGSI_EXEC_TEMP_ADDR         (TGSI_EXEC_NUM_TEMPS + 8)
 167 #define TGSI_EXEC_NUM_ADDRS         1
 168
 169 /* predicate register */
 170 #define TGSI_EXEC_TEMP_P0           (TGSI_EXEC_NUM_TEMPS + 9)
 171 #define TGSI_EXEC_NUM_PREDS         1
 172
 173 #define TGSI_EXEC_NUM_TEMP_EXTRAS   10
 174
 175
 176
 177 #define TGSI_EXEC_MAX_NESTING  32
 178 #define TGSI_EXEC_MAX_COND_NESTING  TGSI_EXEC_MAX_NESTING
 179 #define TGSI_EXEC_MAX_LOOP_NESTING  TGSI_EXEC_MAX_NESTING
 180 #define TGSI_EXEC_MAX_SWITCH_NESTING TGSI_EXEC_MAX_NESTING
 181 #define TGSI_EXEC_MAX_CALL_NESTING  TGSI_EXEC_MAX_NESTING
 182
 183 /* The maximum number of input attributes per vertex. For 2D
 184  * input register files, this is the stride between two 1D
 185  * arrays.
 186  */
 187 #define TGSI_EXEC_MAX_INPUT_ATTRIBS 17
 188
 189 /* The maximum number of constant vectors per constant buffer.
 190  */
 191 #define TGSI_EXEC_MAX_CONST_BUFFER  4096
 192
 193 /* The maximum number of vertices per primitive */
 194 #define TGSI_MAX_PRIM_VERTICES 6
 195
 196 /* The maximum number of primitives to be generated */
 197 #define TGSI_MAX_PRIMITIVES 64
 198
 199 /* The maximum total number of vertices */
 200 #define TGSI_MAX_TOTAL_VERTICES (TGSI_MAX_PRIM_VERTICES * TGSI_MAX_PRIMITIVES * PIPE_MAX_ATTRIBS)
 201
 202 /** function call/activation record */
 203 struct tgsi_call_record
 204 {
 205    uint CondStackTop;
 206    uint LoopStackTop;
 207    uint ContStackTop;
 208    int SwitchStackTop;
 209    int BreakStackTop;
 210    uint ReturnAddr;
 211 };
 212
 213
 214 /* Switch-case block state. */
 215 struct tgsi_switch_record {
 216    uint mask;                          /**< execution mask */
 217    union tgsi_exec_channel selector;   /**< a value case statements are compared to */
 218    uint defaultMask;                   /**< non-execute mask for default case */
 219 };
 220
 221
 222 enum tgsi_break_type {
 223    TGSI_EXEC_BREAK_INSIDE_LOOP,
 224    TGSI_EXEC_BREAK_INSIDE_SWITCH
 225 };
 226
 227
 228 #define TGSI_EXEC_MAX_BREAK_STACK (TGSI_EXEC_MAX_LOOP_NESTING + TGSI_EXEC_MAX_SWITCH_NESTING)
 229
 230
 231 /**
 232  * Run-time virtual machine state for executing TGSI shader.
 233  */
 234 struct tgsi_exec_machine
 235 {
 236    /* Total = program temporaries + internal temporaries
 237     */
 238    struct tgsi_exec_vector       Temps[TGSI_EXEC_NUM_TEMPS +
 239                                        TGSI_EXEC_NUM_TEMP_EXTRAS];
 240
 241    float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
 242
 243    struct tgsi_exec_vector       Inputs[TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS];
 244    struct tgsi_exec_vector       Outputs[TGSI_MAX_TOTAL_VERTICES];
 245
 246    struct tgsi_exec_vector       *Addrs;
 247    struct tgsi_exec_vector       *Predicates;
 248
 249    struct tgsi_sampler           **Samplers;
 250
 251    unsigned                      ImmLimit;
 252    const void *Consts[PIPE_MAX_CONSTANT_BUFFERS];
 253    const struct tgsi_token       *Tokens;   /**< Declarations, instructions */
 254    unsigned                      Processor; /**< TGSI_PROCESSOR_x */
 255
 256    /* GEOMETRY processor only. */
 257    unsigned                      *Primitives;
 258    unsigned                       NumOutputs;
 259    unsigned                       MaxGeometryShaderOutputs;
 260
 261    /* FRAGMENT processor only. */
 262    const struct tgsi_interp_coef *InterpCoefs;
 263    struct tgsi_exec_vector       QuadPos;
 264    float                         Face;    /**< +1 if front facing, -1 if back facing */
 265
 266    /* Conditional execution masks */
 267    uint CondMask;  /**< For IF/ELSE/ENDIF */
 268    uint LoopMask;  /**< For BGNLOOP/ENDLOOP */
 269    uint ContMask;  /**< For loop CONT statements */
 270    uint FuncMask;  /**< For function calls */
 271    uint ExecMask;  /**< = CondMask & LoopMask */
 272
 273    /* Current switch-case state. */
 274    struct tgsi_switch_record Switch;
 275
 276    /* Current break type. */
 277    enum tgsi_break_type BreakType;
 278
 279    /** Condition mask stack (for nested conditionals) */
 280    uint CondStack[TGSI_EXEC_MAX_COND_NESTING];
 281    int CondStackTop;
 282
 283    /** Loop mask stack (for nested loops) */
 284    uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
 285    int LoopStackTop;
 286
 287    /** Loop label stack */
 288    uint LoopLabelStack[TGSI_EXEC_MAX_LOOP_NESTING];
 289    int LoopLabelStackTop;
 290
 291    /** Loop continue mask stack (see comments in tgsi_exec.c) */
 292    uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
 293    int ContStackTop;
 294
 295    /** Switch case stack */
 296    struct tgsi_switch_record SwitchStack[TGSI_EXEC_MAX_SWITCH_NESTING];
 297    int SwitchStackTop;
 298
 299    enum tgsi_break_type BreakStack[TGSI_EXEC_MAX_BREAK_STACK];
 300    int BreakStackTop;
 301
 302    /** Function execution mask stack (for executing subroutine code) */
 303    uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
 304    int FuncStackTop;
 305
 306    /** Function call stack for saving/restoring the program counter */
 307    struct tgsi_call_record CallStack[TGSI_EXEC_MAX_CALL_NESTING];
 308    int CallStackTop;
 309
 310    struct tgsi_full_instruction *Instructions;
 311    uint NumInstructions;
 312
 313    struct tgsi_full_declaration *Declarations;
 314    uint NumDeclarations;
 315
 316 };
 317
 318 struct tgsi_exec_machine *
 319 tgsi_exec_machine_create( void );
 320
 321 void
 322 tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach);
 323
 324
 325 void
 326 tgsi_exec_machine_bind_shader(
 327    struct tgsi_exec_machine *mach,
 328    const struct tgsi_token *tokens,
 329    uint numSamplers,
 330    struct tgsi_sampler **samplers);
 331
 332 uint
 333 tgsi_exec_machine_run(
 334    struct tgsi_exec_machine *mach );
 335
 336
 337 void
 338 tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach);
 339
 340
 341 boolean
 342 tgsi_check_soa_dependencies(const struct tgsi_full_instruction *inst);
 343
 344
 345 static INLINE void
 346 tgsi_set_kill_mask(struct tgsi_exec_machine *mach, unsigned mask)
 347 {
 348    mach->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0] =
 349       mask;
 350 }
 351
 352
 353 /** Set execution mask values prior to executing the shader */
 354 static INLINE void
 355 tgsi_set_exec_mask(struct tgsi_exec_machine *mach,
 356                    boolean ch0, boolean ch1, boolean ch2, boolean ch3)
 357 {
 358    int *mask = mach->Temps[TGSI_EXEC_MASK_I].xyzw[TGSI_EXEC_MASK_C].i;
 359    mask[0] = ch0 ? ~0 : 0;
 360    mask[1] = ch1 ? ~0 : 0;
 361    mask[2] = ch2 ? ~0 : 0;
 362    mask[3] = ch3 ? ~0 : 0;
 363 }
 364
 365
 366 #if defined __cplusplus
 367 } /* extern "C" */
 368 #endif
 369
 370 #endif /* TGSI_EXEC_H */