src/gallium/auxiliary/tgsi/tgsi_exec.h

   1 /**************************************************************************
   2  *
   3  * Copyright 2007-2008 VMware, Inc.
   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 VMWARE 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    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     * r - 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     * derivs - explicit derivatives.
 116     * offset - texel offsets
 117     * lod - lod value, except for shadow cube arrays (compare value there).
 118     */
 119    void (*get_samples)(struct tgsi_sampler *sampler,
 120                        const unsigned sview_index,
 121                        const unsigned sampler_index,
 122                        const float s[TGSI_QUAD_SIZE],
 123                        const float t[TGSI_QUAD_SIZE],
 124                        const float r[TGSI_QUAD_SIZE],
 125                        const float c0[TGSI_QUAD_SIZE],
 126                        const float c1[TGSI_QUAD_SIZE],
 127                        float derivs[3][2][TGSI_QUAD_SIZE],
 128                        const int8_t offset[3],
 129                        enum tgsi_sampler_control control,
 130                        float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]);
 131    void (*get_dims)(struct tgsi_sampler *sampler,
 132                     const unsigned sview_index,
 133                     int level, int dims[4]);
 134    void (*get_texel)(struct tgsi_sampler *sampler,
 135                      const unsigned sview_index,
 136                      const int i[TGSI_QUAD_SIZE],
 137                      const int j[TGSI_QUAD_SIZE], const int k[TGSI_QUAD_SIZE],
 138                      const int lod[TGSI_QUAD_SIZE], const int8_t offset[3],
 139                      float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]);
 140 };
 141
 142 #define TGSI_EXEC_NUM_TEMPS       4096
 143 #define TGSI_EXEC_NUM_IMMEDIATES  256
 144
 145 /*
 146  * Locations of various utility registers (_I = Index, _C = Channel)
 147  */
 148 #define TGSI_EXEC_TEMP_00000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
 149 #define TGSI_EXEC_TEMP_00000000_C   0
 150
 151 #define TGSI_EXEC_TEMP_7FFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
 152 #define TGSI_EXEC_TEMP_7FFFFFFF_C   1
 153
 154 #define TGSI_EXEC_TEMP_80000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
 155 #define TGSI_EXEC_TEMP_80000000_C   2
 156
 157 #define TGSI_EXEC_TEMP_FFFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
 158 #define TGSI_EXEC_TEMP_FFFFFFFF_C   3
 159
 160 #define TGSI_EXEC_TEMP_ONE_I        (TGSI_EXEC_NUM_TEMPS + 1)
 161 #define TGSI_EXEC_TEMP_ONE_C        0
 162
 163 #define TGSI_EXEC_TEMP_TWO_I        (TGSI_EXEC_NUM_TEMPS + 1)
 164 #define TGSI_EXEC_TEMP_TWO_C        1
 165
 166 #define TGSI_EXEC_TEMP_128_I        (TGSI_EXEC_NUM_TEMPS + 1)
 167 #define TGSI_EXEC_TEMP_128_C        2
 168
 169 #define TGSI_EXEC_TEMP_MINUS_128_I  (TGSI_EXEC_NUM_TEMPS + 1)
 170 #define TGSI_EXEC_TEMP_MINUS_128_C  3
 171
 172 #define TGSI_EXEC_TEMP_KILMASK_I    (TGSI_EXEC_NUM_TEMPS + 2)
 173 #define TGSI_EXEC_TEMP_KILMASK_C    0
 174
 175 #define TGSI_EXEC_TEMP_OUTPUT_I     (TGSI_EXEC_NUM_TEMPS + 2)
 176 #define TGSI_EXEC_TEMP_OUTPUT_C     1
 177
 178 #define TGSI_EXEC_TEMP_PRIMITIVE_I  (TGSI_EXEC_NUM_TEMPS + 2)
 179 #define TGSI_EXEC_TEMP_PRIMITIVE_C  2
 180
 181 #define TGSI_EXEC_TEMP_THREE_I      (TGSI_EXEC_NUM_TEMPS + 2)
 182 #define TGSI_EXEC_TEMP_THREE_C      3
 183
 184 #define TGSI_EXEC_TEMP_HALF_I       (TGSI_EXEC_NUM_TEMPS + 3)
 185 #define TGSI_EXEC_TEMP_HALF_C       0
 186
 187 /* execution mask, each value is either 0 or ~0 */
 188 #define TGSI_EXEC_MASK_I            (TGSI_EXEC_NUM_TEMPS + 3)
 189 #define TGSI_EXEC_MASK_C            1
 190
 191 /* 4 register buffer for various purposes */
 192 #define TGSI_EXEC_TEMP_R0           (TGSI_EXEC_NUM_TEMPS + 4)
 193 #define TGSI_EXEC_NUM_TEMP_R        4
 194
 195 #define TGSI_EXEC_TEMP_ADDR         (TGSI_EXEC_NUM_TEMPS + 8)
 196
 197 /* predicate register */
 198 #define TGSI_EXEC_TEMP_P0           (TGSI_EXEC_NUM_TEMPS + 9)
 199 #define TGSI_EXEC_NUM_PREDS         1
 200
 201 #define TGSI_EXEC_NUM_TEMP_EXTRAS   10
 202
 203
 204
 205 #define TGSI_EXEC_MAX_NESTING  32
 206 #define TGSI_EXEC_MAX_COND_NESTING  TGSI_EXEC_MAX_NESTING
 207 #define TGSI_EXEC_MAX_LOOP_NESTING  TGSI_EXEC_MAX_NESTING
 208 #define TGSI_EXEC_MAX_SWITCH_NESTING TGSI_EXEC_MAX_NESTING
 209 #define TGSI_EXEC_MAX_CALL_NESTING  TGSI_EXEC_MAX_NESTING
 210
 211 /* The maximum number of input attributes per vertex. For 2D
 212  * input register files, this is the stride between two 1D
 213  * arrays.
 214  */
 215 #define TGSI_EXEC_MAX_INPUT_ATTRIBS PIPE_MAX_SHADER_INPUTS
 216
 217 /* The maximum number of bytes per constant buffer.
 218  */
 219 #define TGSI_EXEC_MAX_CONST_BUFFER_SIZE  (4096 * sizeof(float[4]))
 220
 221 /* The maximum number of vertices per primitive */
 222 #define TGSI_MAX_PRIM_VERTICES 6
 223
 224 /* The maximum number of primitives to be generated */
 225 #define TGSI_MAX_PRIMITIVES 64
 226
 227 /* The maximum total number of vertices */
 228 #define TGSI_MAX_TOTAL_VERTICES (TGSI_MAX_PRIM_VERTICES * TGSI_MAX_PRIMITIVES * PIPE_MAX_ATTRIBS)
 229
 230 #define TGSI_MAX_MISC_INPUTS 8
 231
 232 /** function call/activation record */
 233 struct tgsi_call_record
 234 {
 235    uint CondStackTop;
 236    uint LoopStackTop;
 237    uint ContStackTop;
 238    int SwitchStackTop;
 239    int BreakStackTop;
 240    uint ReturnAddr;
 241 };
 242
 243
 244 /* Switch-case block state. */
 245 struct tgsi_switch_record {
 246    uint mask;                          /**< execution mask */
 247    union tgsi_exec_channel selector;   /**< a value case statements are compared to */
 248    uint defaultMask;                   /**< non-execute mask for default case */
 249 };
 250
 251
 252 enum tgsi_break_type {
 253    TGSI_EXEC_BREAK_INSIDE_LOOP,
 254    TGSI_EXEC_BREAK_INSIDE_SWITCH
 255 };
 256
 257
 258 #define TGSI_EXEC_MAX_BREAK_STACK (TGSI_EXEC_MAX_LOOP_NESTING + TGSI_EXEC_MAX_SWITCH_NESTING)
 259
 260
 261 /**
 262  * Run-time virtual machine state for executing TGSI shader.
 263  */
 264 struct tgsi_exec_machine
 265 {
 266    /* Total = program temporaries + internal temporaries
 267     */
 268    struct tgsi_exec_vector       Temps[TGSI_EXEC_NUM_TEMPS +
 269                                        TGSI_EXEC_NUM_TEMP_EXTRAS];
 270
 271    float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
 272
 273    float                         ImmArray[TGSI_EXEC_NUM_IMMEDIATES][4];
 274
 275    struct tgsi_exec_vector       *Inputs;
 276    struct tgsi_exec_vector       *Outputs;
 277
 278    /* System values */
 279    unsigned                      SysSemanticToIndex[TGSI_SEMANTIC_COUNT];
 280    union tgsi_exec_channel       SystemValue[TGSI_MAX_MISC_INPUTS];
 281
 282    struct tgsi_exec_vector       *Addrs;
 283    struct tgsi_exec_vector       *Predicates;
 284
 285    struct tgsi_sampler           *Sampler;
 286
 287    unsigned                      ImmLimit;
 288
 289    const void *Consts[PIPE_MAX_CONSTANT_BUFFERS];
 290    unsigned ConstsSize[PIPE_MAX_CONSTANT_BUFFERS];
 291
 292    const struct tgsi_token       *Tokens;   /**< Declarations, instructions */
 293    unsigned                      Processor; /**< TGSI_PROCESSOR_x */
 294
 295    /* GEOMETRY processor only. */
 296    unsigned                      *Primitives;
 297    unsigned                       NumOutputs;
 298    unsigned                       MaxGeometryShaderOutputs;
 299    unsigned                       MaxOutputVertices;
 300
 301    /* FRAGMENT processor only. */
 302    const struct tgsi_interp_coef *InterpCoefs;
 303    struct tgsi_exec_vector       QuadPos;
 304    float                         Face;    /**< +1 if front facing, -1 if back facing */
 305    bool                          flatshade_color;
 306    /* Conditional execution masks */
 307    uint CondMask;  /**< For IF/ELSE/ENDIF */
 308    uint LoopMask;  /**< For BGNLOOP/ENDLOOP */
 309    uint ContMask;  /**< For loop CONT statements */
 310    uint FuncMask;  /**< For function calls */
 311    uint ExecMask;  /**< = CondMask & LoopMask */
 312
 313    /* Current switch-case state. */
 314    struct tgsi_switch_record Switch;
 315
 316    /* Current break type. */
 317    enum tgsi_break_type BreakType;
 318
 319    /** Condition mask stack (for nested conditionals) */
 320    uint CondStack[TGSI_EXEC_MAX_COND_NESTING];
 321    int CondStackTop;
 322
 323    /** Loop mask stack (for nested loops) */
 324    uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
 325    int LoopStackTop;
 326
 327    /** Loop label stack */
 328    uint LoopLabelStack[TGSI_EXEC_MAX_LOOP_NESTING];
 329    int LoopLabelStackTop;
 330
 331    /** Loop continue mask stack (see comments in tgsi_exec.c) */
 332    uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
 333    int ContStackTop;
 334
 335    /** Switch case stack */
 336    struct tgsi_switch_record SwitchStack[TGSI_EXEC_MAX_SWITCH_NESTING];
 337    int SwitchStackTop;
 338
 339    enum tgsi_break_type BreakStack[TGSI_EXEC_MAX_BREAK_STACK];
 340    int BreakStackTop;
 341
 342    /** Function execution mask stack (for executing subroutine code) */
 343    uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
 344    int FuncStackTop;
 345
 346    /** Function call stack for saving/restoring the program counter */
 347    struct tgsi_call_record CallStack[TGSI_EXEC_MAX_CALL_NESTING];
 348    int CallStackTop;
 349
 350    struct tgsi_full_instruction *Instructions;
 351    uint NumInstructions;
 352
 353    struct tgsi_full_declaration *Declarations;
 354    uint NumDeclarations;
 355
 356    struct tgsi_declaration_sampler_view
 357       SamplerViews[PIPE_MAX_SHADER_SAMPLER_VIEWS];
 358
 359    boolean UsedGeometryShader;
 360 };
 361
 362 struct tgsi_exec_machine *
 363 tgsi_exec_machine_create( void );
 364
 365 void
 366 tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach);
 367
 368
 369 void
 370 tgsi_exec_machine_bind_shader(
 371    struct tgsi_exec_machine *mach,
 372    const struct tgsi_token *tokens,
 373    struct tgsi_sampler *sampler);
 374
 375 uint
 376 tgsi_exec_machine_run(
 377    struct tgsi_exec_machine *mach );
 378
 379
 380 void
 381 tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach);
 382
 383
 384 boolean
 385 tgsi_check_soa_dependencies(const struct tgsi_full_instruction *inst);
 386
 387
 388 static INLINE void
 389 tgsi_set_kill_mask(struct tgsi_exec_machine *mach, unsigned mask)
 390 {
 391    mach->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0] =
 392       mask;
 393 }
 394
 395
 396 /** Set execution mask values prior to executing the shader */
 397 static INLINE void
 398 tgsi_set_exec_mask(struct tgsi_exec_machine *mach,
 399                    boolean ch0, boolean ch1, boolean ch2, boolean ch3)
 400 {
 401    int *mask = mach->Temps[TGSI_EXEC_MASK_I].xyzw[TGSI_EXEC_MASK_C].i;
 402    mask[0] = ch0 ? ~0 : 0;
 403    mask[1] = ch1 ? ~0 : 0;
 404    mask[2] = ch2 ? ~0 : 0;
 405    mask[3] = ch3 ? ~0 : 0;
 406 }
 407
 408
 409 extern void
 410 tgsi_exec_set_constant_buffers(struct tgsi_exec_machine *mach,
 411                                unsigned num_bufs,
 412                                const void **bufs,
 413                                const unsigned *buf_sizes);
 414
 415
 416 static INLINE int
 417 tgsi_exec_get_shader_param(enum pipe_shader_cap param)
 418 {
 419    switch(param) {
 420    case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
 421    case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
 422    case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
 423    case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
 424       return INT_MAX;
 425    case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
 426       return TGSI_EXEC_MAX_NESTING;
 427    case PIPE_SHADER_CAP_MAX_INPUTS:
 428       return TGSI_EXEC_MAX_INPUT_ATTRIBS;
 429    case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
 430       return TGSI_EXEC_MAX_CONST_BUFFER_SIZE;
 431    case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
 432       return PIPE_MAX_CONSTANT_BUFFERS;
 433    case PIPE_SHADER_CAP_MAX_TEMPS:
 434       return TGSI_EXEC_NUM_TEMPS;
 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_MAX_SAMPLER_VIEWS:
 451       return PIPE_MAX_SHADER_SAMPLER_VIEWS;
 452    case PIPE_SHADER_CAP_PREFERRED_IR:
 453       return PIPE_SHADER_IR_TGSI;
 454    case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
 455       return 1;
 456    case PIPE_SHADER_CAP_DOUBLES:
 457       return 0;
 458    }
 459    /* if we get here, we missed a shader cap above (and should have seen
 460     * a compiler warning.)
 461     */
 462    return 0;
 463 }
 464
 465 #if defined __cplusplus
 466 } /* extern "C" */
 467 #endif
 468
 469 #endif /* TGSI_EXEC_H */