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