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 float s[TGSI_QUAD_SIZE],
 119                        const float t[TGSI_QUAD_SIZE],
 120                        const float r[TGSI_QUAD_SIZE],
 121                        const float c0[TGSI_QUAD_SIZE],
 122                        const float c1[TGSI_QUAD_SIZE],
 123                        enum tgsi_sampler_control control,
 124                        float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]);
 125    void (*get_dims)(struct tgsi_sampler *sampler, int level,
 126                     int dims[4]);
 127    void (*get_texel)(struct tgsi_sampler *sampler, const int i[TGSI_QUAD_SIZE],
 128                      const int j[TGSI_QUAD_SIZE], const int k[TGSI_QUAD_SIZE],
 129                      const int lod[TGSI_QUAD_SIZE], const int8_t offset[3],
 130                      float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]);
 131 };
 132
 133 #define TGSI_EXEC_NUM_TEMPS       4096
 134 #define TGSI_EXEC_NUM_IMMEDIATES  256
 135 #define TGSI_EXEC_NUM_TEMP_ARRAYS 8
 136
 137 /*
 138  * Locations of various utility registers (_I = Index, _C = Channel)
 139  */
 140 #define TGSI_EXEC_TEMP_00000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
 141 #define TGSI_EXEC_TEMP_00000000_C   0
 142
 143 #define TGSI_EXEC_TEMP_7FFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
 144 #define TGSI_EXEC_TEMP_7FFFFFFF_C   1
 145
 146 #define TGSI_EXEC_TEMP_80000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
 147 #define TGSI_EXEC_TEMP_80000000_C   2
 148
 149 #define TGSI_EXEC_TEMP_FFFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
 150 #define TGSI_EXEC_TEMP_FFFFFFFF_C   3
 151
 152 #define TGSI_EXEC_TEMP_ONE_I        (TGSI_EXEC_NUM_TEMPS + 1)
 153 #define TGSI_EXEC_TEMP_ONE_C        0
 154
 155 #define TGSI_EXEC_TEMP_TWO_I        (TGSI_EXEC_NUM_TEMPS + 1)
 156 #define TGSI_EXEC_TEMP_TWO_C        1
 157
 158 #define TGSI_EXEC_TEMP_128_I        (TGSI_EXEC_NUM_TEMPS + 1)
 159 #define TGSI_EXEC_TEMP_128_C        2
 160
 161 #define TGSI_EXEC_TEMP_MINUS_128_I  (TGSI_EXEC_NUM_TEMPS + 1)
 162 #define TGSI_EXEC_TEMP_MINUS_128_C  3
 163
 164 #define TGSI_EXEC_TEMP_KILMASK_I    (TGSI_EXEC_NUM_TEMPS + 2)
 165 #define TGSI_EXEC_TEMP_KILMASK_C    0
 166
 167 #define TGSI_EXEC_TEMP_OUTPUT_I     (TGSI_EXEC_NUM_TEMPS + 2)
 168 #define TGSI_EXEC_TEMP_OUTPUT_C     1
 169
 170 #define TGSI_EXEC_TEMP_PRIMITIVE_I  (TGSI_EXEC_NUM_TEMPS + 2)
 171 #define TGSI_EXEC_TEMP_PRIMITIVE_C  2
 172
 173 #define TGSI_EXEC_TEMP_THREE_I      (TGSI_EXEC_NUM_TEMPS + 2)
 174 #define TGSI_EXEC_TEMP_THREE_C      3
 175
 176 #define TGSI_EXEC_TEMP_HALF_I       (TGSI_EXEC_NUM_TEMPS + 3)
 177 #define TGSI_EXEC_TEMP_HALF_C       0
 178
 179 /* execution mask, each value is either 0 or ~0 */
 180 #define TGSI_EXEC_MASK_I            (TGSI_EXEC_NUM_TEMPS + 3)
 181 #define TGSI_EXEC_MASK_C            1
 182
 183 /* 4 register buffer for various purposes */
 184 #define TGSI_EXEC_TEMP_R0           (TGSI_EXEC_NUM_TEMPS + 4)
 185 #define TGSI_EXEC_NUM_TEMP_R        4
 186
 187 #define TGSI_EXEC_TEMP_ADDR         (TGSI_EXEC_NUM_TEMPS + 8)
 188 #define TGSI_EXEC_NUM_ADDRS         1
 189
 190 /* predicate register */
 191 #define TGSI_EXEC_TEMP_P0           (TGSI_EXEC_NUM_TEMPS + 9)
 192 #define TGSI_EXEC_NUM_PREDS         1
 193
 194 #define TGSI_EXEC_NUM_TEMP_EXTRAS   10
 195
 196
 197
 198 #define TGSI_EXEC_MAX_NESTING  32
 199 #define TGSI_EXEC_MAX_COND_NESTING  TGSI_EXEC_MAX_NESTING
 200 #define TGSI_EXEC_MAX_LOOP_NESTING  TGSI_EXEC_MAX_NESTING
 201 #define TGSI_EXEC_MAX_SWITCH_NESTING TGSI_EXEC_MAX_NESTING
 202 #define TGSI_EXEC_MAX_CALL_NESTING  TGSI_EXEC_MAX_NESTING
 203
 204 /* The maximum number of input attributes per vertex. For 2D
 205  * input register files, this is the stride between two 1D
 206  * arrays.
 207  */
 208 #define TGSI_EXEC_MAX_INPUT_ATTRIBS 17
 209
 210 /* The maximum number of constant vectors per constant buffer.
 211  */
 212 #define TGSI_EXEC_MAX_CONST_BUFFER  4096
 213
 214 /* The maximum number of vertices per primitive */
 215 #define TGSI_MAX_PRIM_VERTICES 6
 216
 217 /* The maximum number of primitives to be generated */
 218 #define TGSI_MAX_PRIMITIVES 64
 219
 220 /* The maximum total number of vertices */
 221 #define TGSI_MAX_TOTAL_VERTICES (TGSI_MAX_PRIM_VERTICES * TGSI_MAX_PRIMITIVES * PIPE_MAX_ATTRIBS)
 222
 223 #define TGSI_MAX_MISC_INPUTS 8
 224
 225 /** function call/activation record */
 226 struct tgsi_call_record
 227 {
 228    uint CondStackTop;
 229    uint LoopStackTop;
 230    uint ContStackTop;
 231    int SwitchStackTop;
 232    int BreakStackTop;
 233    uint ReturnAddr;
 234 };
 235
 236
 237 /* Switch-case block state. */
 238 struct tgsi_switch_record {
 239    uint mask;                          /**< execution mask */
 240    union tgsi_exec_channel selector;   /**< a value case statements are compared to */
 241    uint defaultMask;                   /**< non-execute mask for default case */
 242 };
 243
 244
 245 enum tgsi_break_type {
 246    TGSI_EXEC_BREAK_INSIDE_LOOP,
 247    TGSI_EXEC_BREAK_INSIDE_SWITCH
 248 };
 249
 250
 251 #define TGSI_EXEC_MAX_BREAK_STACK (TGSI_EXEC_MAX_LOOP_NESTING + TGSI_EXEC_MAX_SWITCH_NESTING)
 252
 253
 254 /**
 255  * Run-time virtual machine state for executing TGSI shader.
 256  */
 257 struct tgsi_exec_machine
 258 {
 259    /* Total = program temporaries + internal temporaries
 260     */
 261    struct tgsi_exec_vector       Temps[TGSI_EXEC_NUM_TEMPS +
 262                                        TGSI_EXEC_NUM_TEMP_EXTRAS];
 263    struct tgsi_exec_vector       TempArray[TGSI_EXEC_NUM_TEMP_ARRAYS][TGSI_EXEC_NUM_TEMPS];
 264
 265    float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
 266
 267    float                         ImmArray[TGSI_EXEC_NUM_IMMEDIATES][4];
 268
 269    struct tgsi_exec_vector       *Inputs;
 270    struct tgsi_exec_vector       *Outputs;
 271
 272    /* System values */
 273    unsigned                      SysSemanticToIndex[TGSI_SEMANTIC_COUNT];
 274    union tgsi_exec_channel       SystemValue[TGSI_MAX_MISC_INPUTS];
 275
 276    struct tgsi_exec_vector       *Addrs;
 277    struct tgsi_exec_vector       *Predicates;
 278
 279    struct tgsi_sampler           **Samplers;
 280
 281    unsigned                      ImmLimit;
 282
 283    const void *Consts[PIPE_MAX_CONSTANT_BUFFERS];
 284    unsigned ConstsSize[PIPE_MAX_CONSTANT_BUFFERS];
 285
 286    const struct tgsi_token       *Tokens;   /**< Declarations, instructions */
 287    unsigned                      Processor; /**< TGSI_PROCESSOR_x */
 288
 289    /* GEOMETRY processor only. */
 290    unsigned                      *Primitives;
 291    unsigned                       NumOutputs;
 292    unsigned                       MaxGeometryShaderOutputs;
 293
 294    /* FRAGMENT processor only. */
 295    const struct tgsi_interp_coef *InterpCoefs;
 296    struct tgsi_exec_vector       QuadPos;
 297    float                         Face;    /**< +1 if front facing, -1 if back facing */
 298    bool                          flatshade_color;
 299    /* Conditional execution masks */
 300    uint CondMask;  /**< For IF/ELSE/ENDIF */
 301    uint LoopMask;  /**< For BGNLOOP/ENDLOOP */
 302    uint ContMask;  /**< For loop CONT statements */
 303    uint FuncMask;  /**< For function calls */
 304    uint ExecMask;  /**< = CondMask & LoopMask */
 305
 306    /* Current switch-case state. */
 307    struct tgsi_switch_record Switch;
 308
 309    /* Current break type. */
 310    enum tgsi_break_type BreakType;
 311
 312    /** Condition mask stack (for nested conditionals) */
 313    uint CondStack[TGSI_EXEC_MAX_COND_NESTING];
 314    int CondStackTop;
 315
 316    /** Loop mask stack (for nested loops) */
 317    uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
 318    int LoopStackTop;
 319
 320    /** Loop label stack */
 321    uint LoopLabelStack[TGSI_EXEC_MAX_LOOP_NESTING];
 322    int LoopLabelStackTop;
 323
 324    /** Loop continue mask stack (see comments in tgsi_exec.c) */
 325    uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
 326    int ContStackTop;
 327
 328    /** Switch case stack */
 329    struct tgsi_switch_record SwitchStack[TGSI_EXEC_MAX_SWITCH_NESTING];
 330    int SwitchStackTop;
 331
 332    enum tgsi_break_type BreakStack[TGSI_EXEC_MAX_BREAK_STACK];
 333    int BreakStackTop;
 334
 335    /** Function execution mask stack (for executing subroutine code) */
 336    uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
 337    int FuncStackTop;
 338
 339    /** Function call stack for saving/restoring the program counter */
 340    struct tgsi_call_record CallStack[TGSI_EXEC_MAX_CALL_NESTING];
 341    int CallStackTop;
 342
 343    struct tgsi_full_instruction *Instructions;
 344    uint NumInstructions;
 345
 346    struct tgsi_full_declaration *Declarations;
 347    uint NumDeclarations;
 348
 349    struct tgsi_declaration_sampler_view
 350       SamplerViews[PIPE_MAX_SHADER_SAMPLER_VIEWS];
 351
 352    boolean UsedGeometryShader;
 353 };
 354
 355 struct tgsi_exec_machine *
 356 tgsi_exec_machine_create( void );
 357
 358 void
 359 tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach);
 360
 361
 362 void
 363 tgsi_exec_machine_bind_shader(
 364    struct tgsi_exec_machine *mach,
 365    const struct tgsi_token *tokens,
 366    uint numSamplers,
 367    struct tgsi_sampler **samplers);
 368
 369 uint
 370 tgsi_exec_machine_run(
 371    struct tgsi_exec_machine *mach );
 372
 373
 374 void
 375 tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach);
 376
 377
 378 boolean
 379 tgsi_check_soa_dependencies(const struct tgsi_full_instruction *inst);
 380
 381
 382 static INLINE void
 383 tgsi_set_kill_mask(struct tgsi_exec_machine *mach, unsigned mask)
 384 {
 385    mach->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0] =
 386       mask;
 387 }
 388
 389
 390 /** Set execution mask values prior to executing the shader */
 391 static INLINE void
 392 tgsi_set_exec_mask(struct tgsi_exec_machine *mach,
 393                    boolean ch0, boolean ch1, boolean ch2, boolean ch3)
 394 {
 395    int *mask = mach->Temps[TGSI_EXEC_MASK_I].xyzw[TGSI_EXEC_MASK_C].i;
 396    mask[0] = ch0 ? ~0 : 0;
 397    mask[1] = ch1 ? ~0 : 0;
 398    mask[2] = ch2 ? ~0 : 0;
 399    mask[3] = ch3 ? ~0 : 0;
 400 }
 401
 402
 403 extern void
 404 tgsi_exec_set_constant_buffers(struct tgsi_exec_machine *mach,
 405                                unsigned num_bufs,
 406                                const void **bufs,
 407                                const unsigned *buf_sizes);
 408
 409
 410 static INLINE int
 411 tgsi_exec_get_shader_param(enum pipe_shader_cap param)
 412 {
 413    switch(param) {
 414    case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 415    case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 416    case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 417    case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 418       return INT_MAX;
 419    case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 420       return TGSI_EXEC_MAX_NESTING;
 421    case PIPE_SHADER_CAP_MAX_INPUTS:
 422       return TGSI_EXEC_MAX_INPUT_ATTRIBS;
 423    case PIPE_SHADER_CAP_MAX_CONSTS:
 424       return TGSI_EXEC_MAX_CONST_BUFFER;
 425    case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 426       return PIPE_MAX_CONSTANT_BUFFERS;
 427    case PIPE_SHADER_CAP_MAX_TEMPS:
 428       return TGSI_EXEC_NUM_TEMPS;
 429    case PIPE_SHADER_CAP_MAX_ADDRS:
 430       return TGSI_EXEC_NUM_ADDRS;
 431    case PIPE_SHADER_CAP_MAX_PREDS:
 432       return TGSI_EXEC_NUM_PREDS;
 433    case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
 434       return 1;
 435    case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
 436    case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
 437    case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
 438    case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
 439       return 1;
 440    case PIPE_SHADER_CAP_SUBROUTINES:
 441       return 1;
 442    case PIPE_SHADER_CAP_INTEGERS:
 443       return 1;
 444    case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
 445       return PIPE_MAX_SAMPLERS;
 446    case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 447       return 1;
 448    default:
 449       return 0;
 450    }
 451 }
 452
 453 #if defined __cplusplus
 454 } /* extern "C" */
 455 #endif
 456
 457 #endif /* TGSI_EXEC_H */