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
  35 #if defined __cplusplus
  36 extern "C" {
  37 #endif
  38
  39
  40 #define MAX_LABELS (4 * 1024)  /**< basically, max instructions */
  41
  42 #define NUM_CHANNELS 4  /* R,G,B,A */
  43 #define QUAD_SIZE    4  /* 4 pixel/quad */
  44
  45
  46 /**
  47   * Registers may be treated as float, signed int or unsigned int.
  48   */
  49 union tgsi_exec_channel
  50 {
  51    float    f[QUAD_SIZE];
  52    int      i[QUAD_SIZE];
  53    unsigned u[QUAD_SIZE];
  54 };
  55
  56 /**
  57   * A vector[RGBA] of channels[4 pixels]
  58   */
  59 struct tgsi_exec_vector
  60 {
  61    union tgsi_exec_channel xyzw[NUM_CHANNELS];
  62 };
  63
  64 /**
  65  * For fragment programs, information for computing fragment input
  66  * values from plane equation of the triangle/line.
  67  */
  68 struct tgsi_interp_coef
  69 {
  70    float a0[NUM_CHANNELS];      /* in an xyzw layout */
  71    float dadx[NUM_CHANNELS];
  72    float dady[NUM_CHANNELS];
  73 };
  74
  75 enum tgsi_sampler_control {
  76    tgsi_sampler_lod_bias,
  77    tgsi_sampler_lod_explicit
  78 };
  79
  80 /**
  81  * Information for sampling textures, which must be implemented
  82  * by code outside the TGSI executor.
  83  */
  84 struct tgsi_sampler
  85 {
  86    /** Get samples for four fragments in a quad */
  87    void (*get_samples)(struct tgsi_sampler *sampler,
  88                        const float s[QUAD_SIZE],
  89                        const float t[QUAD_SIZE],
  90                        const float p[QUAD_SIZE],
  91                        const float c0[QUAD_SIZE],
  92                        enum tgsi_sampler_control control,
  93                        float rgba[NUM_CHANNELS][QUAD_SIZE]);
  94 };
  95
  96 /**
  97  * For branching/calling subroutines.
  98  */
  99 struct tgsi_exec_labels
 100 {
 101    unsigned labels[MAX_LABELS][2];
 102    unsigned count;
 103 };
 104
 105
 106 #define TGSI_EXEC_NUM_TEMPS       128
 107 #define TGSI_EXEC_NUM_IMMEDIATES  256
 108
 109 /*
 110  * Locations of various utility registers (_I = Index, _C = Channel)
 111  */
 112 #define TGSI_EXEC_TEMP_00000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
 113 #define TGSI_EXEC_TEMP_00000000_C   0
 114
 115 #define TGSI_EXEC_TEMP_7FFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
 116 #define TGSI_EXEC_TEMP_7FFFFFFF_C   1
 117
 118 #define TGSI_EXEC_TEMP_80000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
 119 #define TGSI_EXEC_TEMP_80000000_C   2
 120
 121 #define TGSI_EXEC_TEMP_FFFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
 122 #define TGSI_EXEC_TEMP_FFFFFFFF_C   3
 123
 124 #define TGSI_EXEC_TEMP_ONE_I        (TGSI_EXEC_NUM_TEMPS + 1)
 125 #define TGSI_EXEC_TEMP_ONE_C        0
 126
 127 #define TGSI_EXEC_TEMP_TWO_I        (TGSI_EXEC_NUM_TEMPS + 1)
 128 #define TGSI_EXEC_TEMP_TWO_C        1
 129
 130 #define TGSI_EXEC_TEMP_128_I        (TGSI_EXEC_NUM_TEMPS + 1)
 131 #define TGSI_EXEC_TEMP_128_C        2
 132
 133 #define TGSI_EXEC_TEMP_MINUS_128_I  (TGSI_EXEC_NUM_TEMPS + 1)
 134 #define TGSI_EXEC_TEMP_MINUS_128_C  3
 135
 136 #define TGSI_EXEC_TEMP_KILMASK_I    (TGSI_EXEC_NUM_TEMPS + 2)
 137 #define TGSI_EXEC_TEMP_KILMASK_C    0
 138
 139 #define TGSI_EXEC_TEMP_OUTPUT_I     (TGSI_EXEC_NUM_TEMPS + 2)
 140 #define TGSI_EXEC_TEMP_OUTPUT_C     1
 141
 142 #define TGSI_EXEC_TEMP_PRIMITIVE_I  (TGSI_EXEC_NUM_TEMPS + 2)
 143 #define TGSI_EXEC_TEMP_PRIMITIVE_C  2
 144
 145 /* NVIDIA condition code (CC) vector
 146  */
 147 #define TGSI_EXEC_CC_GT       0x01
 148 #define TGSI_EXEC_CC_EQ       0x02
 149 #define TGSI_EXEC_CC_LT       0x04
 150 #define TGSI_EXEC_CC_UN       0x08
 151
 152 #define TGSI_EXEC_CC_X_MASK   0x000000ff
 153 #define TGSI_EXEC_CC_X_SHIFT  0
 154 #define TGSI_EXEC_CC_Y_MASK   0x0000ff00
 155 #define TGSI_EXEC_CC_Y_SHIFT  8
 156 #define TGSI_EXEC_CC_Z_MASK   0x00ff0000
 157 #define TGSI_EXEC_CC_Z_SHIFT  16
 158 #define TGSI_EXEC_CC_W_MASK   0xff000000
 159 #define TGSI_EXEC_CC_W_SHIFT  24
 160
 161 #define TGSI_EXEC_TEMP_CC_I         (TGSI_EXEC_NUM_TEMPS + 2)
 162 #define TGSI_EXEC_TEMP_CC_C         3
 163
 164 #define TGSI_EXEC_TEMP_THREE_I      (TGSI_EXEC_NUM_TEMPS + 3)
 165 #define TGSI_EXEC_TEMP_THREE_C      0
 166
 167 #define TGSI_EXEC_TEMP_HALF_I       (TGSI_EXEC_NUM_TEMPS + 3)
 168 #define TGSI_EXEC_TEMP_HALF_C       1
 169
 170 /* execution mask, each value is either 0 or ~0 */
 171 #define TGSI_EXEC_MASK_I            (TGSI_EXEC_NUM_TEMPS + 3)
 172 #define TGSI_EXEC_MASK_C            2
 173
 174 /* 4 register buffer for various purposes */
 175 #define TGSI_EXEC_TEMP_R0           (TGSI_EXEC_NUM_TEMPS + 4)
 176 #define TGSI_EXEC_NUM_TEMP_R        4
 177
 178 #define TGSI_EXEC_TEMP_ADDR         (TGSI_EXEC_NUM_TEMPS + 8)
 179 #define TGSI_EXEC_NUM_ADDRS         1
 180
 181 /* predicate register */
 182 #define TGSI_EXEC_TEMP_P0           (TGSI_EXEC_NUM_TEMPS + 9)
 183 #define TGSI_EXEC_NUM_PREDS         1
 184
 185 #define TGSI_EXEC_NUM_TEMP_EXTRAS   10
 186
 187
 188
 189 #define TGSI_EXEC_MAX_COND_NESTING  32
 190 #define TGSI_EXEC_MAX_LOOP_NESTING  32
 191 #define TGSI_EXEC_MAX_SWITCH_NESTING 32
 192 #define TGSI_EXEC_MAX_CALL_NESTING  32
 193
 194 /* The maximum number of input attributes per vertex. For 2D
 195  * input register files, this is the stride between two 1D
 196  * arrays.
 197  */
 198 #define TGSI_EXEC_MAX_INPUT_ATTRIBS 17
 199
 200 /* The maximum number of constant vectors per constant buffer.
 201  */
 202 #define TGSI_EXEC_MAX_CONST_BUFFER  4096
 203
 204 /* The maximum number of vertices per primitive */
 205 #define TGSI_MAX_PRIM_VERTICES 6
 206
 207 /* The maximum number of primitives to be generated */
 208 #define TGSI_MAX_PRIMITIVES 64
 209
 210 /* The maximum total number of vertices */
 211 #define TGSI_MAX_TOTAL_VERTICES (TGSI_MAX_PRIM_VERTICES * TGSI_MAX_PRIMITIVES * PIPE_MAX_ATTRIBS)
 212
 213 /** function call/activation record */
 214 struct tgsi_call_record
 215 {
 216    uint CondStackTop;
 217    uint LoopStackTop;
 218    uint ContStackTop;
 219    int SwitchStackTop;
 220    int BreakStackTop;
 221    uint ReturnAddr;
 222 };
 223
 224
 225 /* Switch-case block state. */
 226 struct tgsi_switch_record {
 227    uint mask;                          /**< execution mask */
 228    union tgsi_exec_channel selector;   /**< a value case statements are compared to */
 229    uint defaultMask;                   /**< non-execute mask for default case */
 230 };
 231
 232
 233 enum tgsi_break_type {
 234    TGSI_EXEC_BREAK_INSIDE_LOOP,
 235    TGSI_EXEC_BREAK_INSIDE_SWITCH
 236 };
 237
 238
 239 #define TGSI_EXEC_MAX_BREAK_STACK (TGSI_EXEC_MAX_LOOP_NESTING + TGSI_EXEC_MAX_SWITCH_NESTING)
 240
 241
 242 /**
 243  * Run-time virtual machine state for executing TGSI shader.
 244  */
 245 struct tgsi_exec_machine
 246 {
 247    /* Total = program temporaries + internal temporaries
 248     */
 249    struct tgsi_exec_vector       Temps[TGSI_EXEC_NUM_TEMPS +
 250                                        TGSI_EXEC_NUM_TEMP_EXTRAS];
 251
 252    float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
 253
 254    struct tgsi_exec_vector       Inputs[TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS];
 255    struct tgsi_exec_vector       Outputs[TGSI_MAX_TOTAL_VERTICES];
 256
 257    struct tgsi_exec_vector       *Addrs;
 258    struct tgsi_exec_vector       *Predicates;
 259
 260    struct tgsi_sampler           **Samplers;
 261
 262    unsigned                      ImmLimit;
 263    const float                   (*Consts)[4];
 264    const struct tgsi_token       *Tokens;   /**< Declarations, instructions */
 265    unsigned                      Processor; /**< TGSI_PROCESSOR_x */
 266
 267    /* GEOMETRY processor only. */
 268    unsigned                      *Primitives;
 269    unsigned                       NumOutputs;
 270    unsigned                       MaxGeometryShaderOutputs;
 271
 272    /* FRAGMENT processor only. */
 273    const struct tgsi_interp_coef *InterpCoefs;
 274    struct tgsi_exec_vector       QuadPos;
 275    float                         Face;    /**< +1 if front facing, -1 if back facing */
 276
 277    /* Conditional execution masks */
 278    uint CondMask;  /**< For IF/ELSE/ENDIF */
 279    uint LoopMask;  /**< For BGNLOOP/ENDLOOP */
 280    uint ContMask;  /**< For loop CONT statements */
 281    uint FuncMask;  /**< For function calls */
 282    uint ExecMask;  /**< = CondMask & LoopMask */
 283
 284    /* Current switch-case state. */
 285    struct tgsi_switch_record Switch;
 286
 287    /* Current break type. */
 288    enum tgsi_break_type BreakType;
 289
 290    /** Condition mask stack (for nested conditionals) */
 291    uint CondStack[TGSI_EXEC_MAX_COND_NESTING];
 292    int CondStackTop;
 293
 294    /** Loop mask stack (for nested loops) */
 295    uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
 296    int LoopStackTop;
 297
 298    /** Loop label stack */
 299    uint LoopLabelStack[TGSI_EXEC_MAX_LOOP_NESTING];
 300    int LoopLabelStackTop;
 301
 302    /** Loop counter stack (x = index, y = counter, z = step) */
 303    struct tgsi_exec_vector LoopCounterStack[TGSI_EXEC_MAX_LOOP_NESTING];
 304    int LoopCounterStackTop;
 305
 306    /** Loop continue mask stack (see comments in tgsi_exec.c) */
 307    uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
 308    int ContStackTop;
 309
 310    /** Switch case stack */
 311    struct tgsi_switch_record SwitchStack[TGSI_EXEC_MAX_SWITCH_NESTING];
 312    int SwitchStackTop;
 313
 314    enum tgsi_break_type BreakStack[TGSI_EXEC_MAX_BREAK_STACK];
 315    int BreakStackTop;
 316
 317    /** Function execution mask stack (for executing subroutine code) */
 318    uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
 319    int FuncStackTop;
 320
 321    /** Function call stack for saving/restoring the program counter */
 322    struct tgsi_call_record CallStack[TGSI_EXEC_MAX_CALL_NESTING];
 323    int CallStackTop;
 324
 325    struct tgsi_full_instruction *Instructions;
 326    uint NumInstructions;
 327
 328    struct tgsi_full_declaration *Declarations;
 329    uint NumDeclarations;
 330
 331    struct tgsi_exec_labels Labels;
 332 };
 333
 334 struct tgsi_exec_machine *
 335 tgsi_exec_machine_create( void );
 336
 337 void
 338 tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach);
 339
 340
 341 void
 342 tgsi_exec_machine_bind_shader(
 343    struct tgsi_exec_machine *mach,
 344    const struct tgsi_token *tokens,
 345    uint numSamplers,
 346    struct tgsi_sampler **samplers);
 347
 348 uint
 349 tgsi_exec_machine_run(
 350    struct tgsi_exec_machine *mach );
 351
 352
 353 void
 354 tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach);
 355
 356
 357 boolean
 358 tgsi_check_soa_dependencies(const struct tgsi_full_instruction *inst);
 359
 360
 361 static INLINE void
 362 tgsi_set_kill_mask(struct tgsi_exec_machine *mach, unsigned mask)
 363 {
 364    mach->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0] =
 365       mask;
 366 }
 367
 368
 369 /** Set execution mask values prior to executing the shader */
 370 static INLINE void
 371 tgsi_set_exec_mask(struct tgsi_exec_machine *mach,
 372                    boolean ch0, boolean ch1, boolean ch2, boolean ch3)
 373 {
 374    int *mask = mach->Temps[TGSI_EXEC_MASK_I].xyzw[TGSI_EXEC_MASK_C].i;
 375    mask[0] = ch0 ? ~0 : 0;
 376    mask[1] = ch1 ? ~0 : 0;
 377    mask[2] = ch2 ? ~0 : 0;
 378    mask[3] = ch3 ? ~0 : 0;
 379 }
 380
 381
 382 #if defined __cplusplus
 383 } /* extern "C" */
 384 #endif
 385
 386 #endif /* TGSI_EXEC_H */