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