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