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 #include "pipe/p_shader_tokens.h"
  35
  36 #if defined __cplusplus
  37 extern "C" {
  38 #endif
  39
  40 #define TGSI_CHAN_X 0
  41 #define TGSI_CHAN_Y 1
  42 #define TGSI_CHAN_Z 2
  43 #define TGSI_CHAN_W 3
  44
  45 #define TGSI_NUM_CHANNELS 4  /* R,G,B,A */
  46 #define TGSI_QUAD_SIZE    4  /* 4 pixel/quad */
  47
  48 #define TGSI_FOR_EACH_CHANNEL( CHAN )\
  49    for (CHAN = 0; CHAN < TGSI_NUM_CHANNELS; CHAN++)
  50
  51 #define TGSI_IS_DST0_CHANNEL_ENABLED( INST, CHAN )\
  52    ((INST)->Dst[0].Register.WriteMask & (1 << (CHAN)))
  53
  54 #define TGSI_IF_IS_DST0_CHANNEL_ENABLED( INST, CHAN )\
  55    if (TGSI_IS_DST0_CHANNEL_ENABLED( INST, CHAN ))
  56
  57 #define TGSI_FOR_EACH_DST0_ENABLED_CHANNEL( INST, CHAN )\
  58    TGSI_FOR_EACH_CHANNEL( CHAN )\
  59       TGSI_IF_IS_DST0_CHANNEL_ENABLED( INST, CHAN )
  60
  61
  62 /**
  63   * Registers may be treated as float, signed int or unsigned int.
  64   */
  65 union tgsi_exec_channel
  66 {
  67    float    f[TGSI_QUAD_SIZE];
  68    int      i[TGSI_QUAD_SIZE];
  69    unsigned u[TGSI_QUAD_SIZE];
  70 };
  71
  72 /**
  73   * A vector[RGBA] of channels[4 pixels]
  74   */
  75 struct tgsi_exec_vector
  76 {
  77    union tgsi_exec_channel xyzw[TGSI_NUM_CHANNELS];
  78 };
  79
  80 /**
  81  * For fragment programs, information for computing fragment input
  82  * values from plane equation of the triangle/line.
  83  */
  84 struct tgsi_interp_coef
  85 {
  86    float a0[TGSI_NUM_CHANNELS]; /* in an xyzw layout */
  87    float dadx[TGSI_NUM_CHANNELS];
  88    float dady[TGSI_NUM_CHANNELS];
  89 };
  90
  91 enum tgsi_sampler_control {
  92    tgsi_sampler_lod_none,
  93    tgsi_sampler_lod_bias,
  94    tgsi_sampler_lod_explicit,
  95    tgsi_sampler_lod_zero
  96    /* FIXME: tgsi_sampler_derivs_explicit */
  97 };
  98
  99 /**
 100  * Information for sampling textures, which must be implemented
 101  * by code outside the TGSI executor.
 102  */
 103 struct tgsi_sampler
 104 {
 105    /** Get samples for four fragments in a quad */
 106    /* this interface contains 5 sets of channels that vary
 107     * depending on the sampler.
 108     * s - the first texture coordinate for sampling.
 109     * t - the second texture coordinate for sampling - unused for 1D,
 110           layer for 1D arrays.
 111     * p - the third coordinate for sampling for 3D, cube, cube arrays,
 112     *     layer for 2D arrays. Compare value for 1D/2D shadows.
 113     * c0 - Compare value for shadow cube and shadow 2d arrays,
 114     *      layer for cube arrays.
 115     * lod - lod value, except for shadow cube arrays (compare value there).
 116     */
 117    void (*get_samples)(struct tgsi_sampler *sampler,
 118                        const unsigned sview_index,
 119                        const unsigned sampler_index,
 120                        const float s[TGSI_QUAD_SIZE],
 121                        const float t[TGSI_QUAD_SIZE],
 122                        const float r[TGSI_QUAD_SIZE],
 123                        const float c0[TGSI_QUAD_SIZE],
 124                        const float c1[TGSI_QUAD_SIZE],
 125                        enum tgsi_sampler_control control,
 126                        float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]);
 127    void (*get_dims)(struct tgsi_sampler *sampler,
 128                     const unsigned sview_index,
 129                     int level, int dims[4]);
 130    void (*get_texel)(struct tgsi_sampler *sampler,
 131                      const unsigned sview_index,
 132                      const int i[TGSI_QUAD_SIZE],
 133                      const int j[TGSI_QUAD_SIZE], const int k[TGSI_QUAD_SIZE],
 134                      const int lod[TGSI_QUAD_SIZE], const int8_t offset[3],
 135                      float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]);
 136 };
 137
 138 #define TGSI_EXEC_NUM_TEMPS       4096
 139 #define TGSI_EXEC_NUM_IMMEDIATES  256
 140 #define TGSI_EXEC_NUM_TEMP_ARRAYS 8
 141
 142 /*
 143  * Locations of various utility registers (_I = Index, _C = Channel)
 144  */
 145 #define TGSI_EXEC_TEMP_00000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
 146 #define TGSI_EXEC_TEMP_00000000_C   0
 147
 148 #define TGSI_EXEC_TEMP_7FFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
 149 #define TGSI_EXEC_TEMP_7FFFFFFF_C   1
 150
 151 #define TGSI_EXEC_TEMP_80000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
 152 #define TGSI_EXEC_TEMP_80000000_C   2
 153
 154 #define TGSI_EXEC_TEMP_FFFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
 155 #define TGSI_EXEC_TEMP_FFFFFFFF_C   3
 156
 157 #define TGSI_EXEC_TEMP_ONE_I        (TGSI_EXEC_NUM_TEMPS + 1)
 158 #define TGSI_EXEC_TEMP_ONE_C        0
 159
 160 #define TGSI_EXEC_TEMP_TWO_I        (TGSI_EXEC_NUM_TEMPS + 1)
 161 #define TGSI_EXEC_TEMP_TWO_C        1
 162
 163 #define TGSI_EXEC_TEMP_128_I        (TGSI_EXEC_NUM_TEMPS + 1)
 164 #define TGSI_EXEC_TEMP_128_C        2
 165
 166 #define TGSI_EXEC_TEMP_MINUS_128_I  (TGSI_EXEC_NUM_TEMPS + 1)
 167 #define TGSI_EXEC_TEMP_MINUS_128_C  3
 168
 169 #define TGSI_EXEC_TEMP_KILMASK_I    (TGSI_EXEC_NUM_TEMPS + 2)
 170 #define TGSI_EXEC_TEMP_KILMASK_C    0
 171
 172 #define TGSI_EXEC_TEMP_OUTPUT_I     (TGSI_EXEC_NUM_TEMPS + 2)
 173 #define TGSI_EXEC_TEMP_OUTPUT_C     1
 174
 175 #define TGSI_EXEC_TEMP_PRIMITIVE_I  (TGSI_EXEC_NUM_TEMPS + 2)
 176 #define TGSI_EXEC_TEMP_PRIMITIVE_C  2
 177
 178 #define TGSI_EXEC_TEMP_THREE_I      (TGSI_EXEC_NUM_TEMPS + 2)
 179 #define TGSI_EXEC_TEMP_THREE_C      3
 180
 181 #define TGSI_EXEC_TEMP_HALF_I       (TGSI_EXEC_NUM_TEMPS + 3)
 182 #define TGSI_EXEC_TEMP_HALF_C       0
 183
 184 /* execution mask, each value is either 0 or ~0 */
 185 #define TGSI_EXEC_MASK_I            (TGSI_EXEC_NUM_TEMPS + 3)
 186 #define TGSI_EXEC_MASK_C            1
 187
 188 /* 4 register buffer for various purposes */
 189 #define TGSI_EXEC_TEMP_R0           (TGSI_EXEC_NUM_TEMPS + 4)
 190 #define TGSI_EXEC_NUM_TEMP_R        4
 191
 192 #define TGSI_EXEC_TEMP_ADDR         (TGSI_EXEC_NUM_TEMPS + 8)
 193 #define TGSI_EXEC_NUM_ADDRS         1
 194
 195 /* predicate register */
 196 #define TGSI_EXEC_TEMP_P0           (TGSI_EXEC_NUM_TEMPS + 9)
 197 #define TGSI_EXEC_NUM_PREDS         1
 198
 199 #define TGSI_EXEC_NUM_TEMP_EXTRAS   10
 200
 201
 202
 203 #define TGSI_EXEC_MAX_NESTING  32
 204 #define TGSI_EXEC_MAX_COND_NESTING  TGSI_EXEC_MAX_NESTING
 205 #define TGSI_EXEC_MAX_LOOP_NESTING  TGSI_EXEC_MAX_NESTING
 206 #define TGSI_EXEC_MAX_SWITCH_NESTING TGSI_EXEC_MAX_NESTING
 207 #define TGSI_EXEC_MAX_CALL_NESTING  TGSI_EXEC_MAX_NESTING
 208
 209 /* The maximum number of input attributes per vertex. For 2D
 210  * input register files, this is the stride between two 1D
 211  * arrays.
 212  */
 213 #define TGSI_EXEC_MAX_INPUT_ATTRIBS 17
 214
 215 /* The maximum number of constant vectors per constant buffer.
 216  */
 217 #define TGSI_EXEC_MAX_CONST_BUFFER  4096
 218
 219 /* The maximum number of vertices per primitive */
 220 #define TGSI_MAX_PRIM_VERTICES 6
 221
 222 /* The maximum number of primitives to be generated */
 223 #define TGSI_MAX_PRIMITIVES 64
 224
 225 /* The maximum total number of vertices */
 226 #define TGSI_MAX_TOTAL_VERTICES (TGSI_MAX_PRIM_VERTICES * TGSI_MAX_PRIMITIVES * PIPE_MAX_ATTRIBS)
 227
 228 #define TGSI_MAX_MISC_INPUTS 8
 229
 230 /** function call/activation record */
 231 struct tgsi_call_record
 232 {
 233    uint CondStackTop;
 234    uint LoopStackTop;
 235    uint ContStackTop;
 236    int SwitchStackTop;
 237    int BreakStackTop;
 238    uint ReturnAddr;
 239 };
 240
 241
 242 /* Switch-case block state. */
 243 struct tgsi_switch_record {
 244    uint mask;                          /**< execution mask */
 245    union tgsi_exec_channel selector;   /**< a value case statements are compared to */
 246    uint defaultMask;                   /**< non-execute mask for default case */
 247 };
 248
 249
 250 enum tgsi_break_type {
 251    TGSI_EXEC_BREAK_INSIDE_LOOP,
 252    TGSI_EXEC_BREAK_INSIDE_SWITCH
 253 };
 254
 255
 256 #define TGSI_EXEC_MAX_BREAK_STACK (TGSI_EXEC_MAX_LOOP_NESTING + TGSI_EXEC_MAX_SWITCH_NESTING)
 257
 258
 259 /**
 260  * Run-time virtual machine state for executing TGSI shader.
 261  */
 262 struct tgsi_exec_machine
 263 {
 264    /* Total = program temporaries + internal temporaries
 265     */
 266    struct tgsi_exec_vector       Temps[TGSI_EXEC_NUM_TEMPS +
 267                                        TGSI_EXEC_NUM_TEMP_EXTRAS];
 268    struct tgsi_exec_vector       TempArray[TGSI_EXEC_NUM_TEMP_ARRAYS][TGSI_EXEC_NUM_TEMPS];
 269
 270    float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
 271
 272    float                         ImmArray[TGSI_EXEC_NUM_IMMEDIATES][4];
 273
 274    struct tgsi_exec_vector       *Inputs;
 275    struct tgsi_exec_vector       *Outputs;
 276
 277    /* System values */
 278    unsigned                      SysSemanticToIndex[TGSI_SEMANTIC_COUNT];
 279    union tgsi_exec_channel       SystemValue[TGSI_MAX_MISC_INPUTS];
 280
 281    struct tgsi_exec_vector       *Addrs;
 282    struct tgsi_exec_vector       *Predicates;
 283
 284    struct tgsi_sampler           *Sampler;
 285
 286    unsigned                      ImmLimit;
 287
 288    const void *Consts[PIPE_MAX_CONSTANT_BUFFERS];
 289    unsigned ConstsSize[PIPE_MAX_CONSTANT_BUFFERS];
 290
 291    const struct tgsi_token       *Tokens;   /**< Declarations, instructions */
 292    unsigned                      Processor; /**< TGSI_PROCESSOR_x */
 293
 294    /* GEOMETRY processor only. */
 295    unsigned                      *Primitives;
 296    unsigned                       NumOutputs;
 297    unsigned                       MaxGeometryShaderOutputs;
 298
 299    /* FRAGMENT processor only. */
 300    const struct tgsi_interp_coef *InterpCoefs;
 301    struct tgsi_exec_vector       QuadPos;
 302    float                         Face;    /**< +1 if front facing, -1 if back facing */
 303    bool                          flatshade_color;
 304    /* Conditional execution masks */
 305    uint CondMask;  /**< For IF/ELSE/ENDIF */
 306    uint LoopMask;  /**< For BGNLOOP/ENDLOOP */
 307    uint ContMask;  /**< For loop CONT statements */
 308    uint FuncMask;  /**< For function calls */
 309    uint ExecMask;  /**< = CondMask & LoopMask */
 310
 311    /* Current switch-case state. */
 312    struct tgsi_switch_record Switch;
 313
 314    /* Current break type. */
 315    enum tgsi_break_type BreakType;
 316
 317    /** Condition mask stack (for nested conditionals) */
 318    uint CondStack[TGSI_EXEC_MAX_COND_NESTING];
 319    int CondStackTop;
 320
 321    /** Loop mask stack (for nested loops) */
 322    uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
 323    int LoopStackTop;
 324
 325    /** Loop label stack */
 326    uint LoopLabelStack[TGSI_EXEC_MAX_LOOP_NESTING];
 327    int LoopLabelStackTop;
 328
 329    /** Loop continue mask stack (see comments in tgsi_exec.c) */
 330    uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
 331    int ContStackTop;
 332
 333    /** Switch case stack */
 334    struct tgsi_switch_record SwitchStack[TGSI_EXEC_MAX_SWITCH_NESTING];
 335    int SwitchStackTop;
 336
 337    enum tgsi_break_type BreakStack[TGSI_EXEC_MAX_BREAK_STACK];
 338    int BreakStackTop;
 339
 340    /** Function execution mask stack (for executing subroutine code) */
 341    uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
 342    int FuncStackTop;
 343
 344    /** Function call stack for saving/restoring the program counter */
 345    struct tgsi_call_record CallStack[TGSI_EXEC_MAX_CALL_NESTING];
 346    int CallStackTop;
 347
 348    struct tgsi_full_instruction *Instructions;
 349    uint NumInstructions;
 350
 351    struct tgsi_full_declaration *Declarations;
 352    uint NumDeclarations;
 353
 354    struct tgsi_declaration_sampler_view
 355       SamplerViews[PIPE_MAX_SHADER_SAMPLER_VIEWS];
 356
 357    boolean UsedGeometryShader;
 358 };
 359
 360 struct tgsi_exec_machine *
 361 tgsi_exec_machine_create( void );
 362
 363 void
 364 tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach);
 365
 366
 367 void
 368 tgsi_exec_machine_bind_shader(
 369    struct tgsi_exec_machine *mach,
 370    const struct tgsi_token *tokens,
 371    struct tgsi_sampler *sampler);
 372
 373 uint
 374 tgsi_exec_machine_run(
 375    struct tgsi_exec_machine *mach );
 376
 377
 378 void
 379 tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach);
 380
 381
 382 boolean
 383 tgsi_check_soa_dependencies(const struct tgsi_full_instruction *inst);
 384
 385
 386 static INLINE void
 387 tgsi_set_kill_mask(struct tgsi_exec_machine *mach, unsigned mask)
 388 {
 389    mach->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0] =
 390       mask;
 391 }
 392
 393
 394 /** Set execution mask values prior to executing the shader */
 395 static INLINE void
 396 tgsi_set_exec_mask(struct tgsi_exec_machine *mach,
 397                    boolean ch0, boolean ch1, boolean ch2, boolean ch3)
 398 {
 399    int *mask = mach->Temps[TGSI_EXEC_MASK_I].xyzw[TGSI_EXEC_MASK_C].i;
 400    mask[0] = ch0 ? ~0 : 0;
 401    mask[1] = ch1 ? ~0 : 0;
 402    mask[2] = ch2 ? ~0 : 0;
 403    mask[3] = ch3 ? ~0 : 0;
 404 }
 405
 406
 407 extern void
 408 tgsi_exec_set_constant_buffers(struct tgsi_exec_machine *mach,
 409                                unsigned num_bufs,
 410                                const void **bufs,
 411                                const unsigned *buf_sizes);
 412
 413
 414 static INLINE int
 415 tgsi_exec_get_shader_param(enum pipe_shader_cap param)
 416 {
 417    switch(param) {
 418    case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 419    case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 420    case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 421    case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 422       return INT_MAX;
 423    case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 424       return TGSI_EXEC_MAX_NESTING;
 425    case PIPE_SHADER_CAP_MAX_INPUTS:
 426       return TGSI_EXEC_MAX_INPUT_ATTRIBS;
 427    case PIPE_SHADER_CAP_MAX_CONSTS:
 428       return TGSI_EXEC_MAX_CONST_BUFFER;
 429    case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 430       return PIPE_MAX_CONSTANT_BUFFERS;
 431    case PIPE_SHADER_CAP_MAX_TEMPS:
 432       return TGSI_EXEC_NUM_TEMPS;
 433    case PIPE_SHADER_CAP_MAX_ADDRS:
 434       return TGSI_EXEC_NUM_ADDRS;
 435    case PIPE_SHADER_CAP_MAX_PREDS:
 436       return TGSI_EXEC_NUM_PREDS;
 437    case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 438       return 1;
 439    case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 440    case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 441    case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 442    case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 443       return 1;
 444    case PIPE_SHADER_CAP_SUBROUTINES:
 445       return 1;
 446    case PIPE_SHADER_CAP_INTEGERS:
 447       return 1;
 448    case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 449       return PIPE_MAX_SAMPLERS;
 450    case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 451       return 1;
 452    default:
 453       return 0;
 454    }
 455 }
 456
 457 #if defined __cplusplus
 458 } /* extern "C" */
 459 #endif
 460
 461 #endif /* TGSI_EXEC_H */