src/gallium/auxiliary/tgsi/tgsi_exec.h

   1 /**************************************************************************
   2  *
   3  * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 #ifndef TGSI_EXEC_H
  29 #define TGSI_EXEC_H
  30
  31 #include "pipe/p_compiler.h"
  32
  33 #if defined __cplusplus
  34 extern "C" {
  35 #endif
  36
  37 #define MAX_LABELS 1024
  38
  39 #define NUM_CHANNELS 4  /* R,G,B,A */
  40 #define QUAD_SIZE    4  /* 4 pixel/quad */
  41
  42 /**
  43   * Registers may be treated as float, signed int or unsigned int.
  44   */
  45 union tgsi_exec_channel
  46 {
  47    float    f[QUAD_SIZE];
  48    int      i[QUAD_SIZE];
  49    unsigned u[QUAD_SIZE];
  50 };
  51
  52 /**
  53   * A vector[RGBA] of channels[4 pixels]
  54   */
  55 struct tgsi_exec_vector
  56 {
  57    union tgsi_exec_channel xyzw[NUM_CHANNELS];
  58 };
  59
  60 /**
  61  * For fragment programs, information for computing fragment input
  62  * values from plane equation of the triangle/line.
  63  */
  64 struct tgsi_interp_coef
  65 {
  66    float a0[NUM_CHANNELS];      /* in an xyzw layout */
  67    float dadx[NUM_CHANNELS];
  68    float dady[NUM_CHANNELS];
  69 };
  70
  71 /**
  72  * Information for sampling textures, which must be implemented
  73  * by code outside the TGSI executor.
  74  */
  75 struct tgsi_sampler
  76 {
  77    /** Get samples for four fragments in a quad */
  78    void (*get_samples)(struct tgsi_sampler *sampler,
  79                        const float s[QUAD_SIZE],
  80                        const float t[QUAD_SIZE],
  81                        const float p[QUAD_SIZE],
  82                        float lodbias,
  83                        float rgba[NUM_CHANNELS][QUAD_SIZE]);
  84 };
  85
  86 /**
  87  * For branching/calling subroutines.
  88  */
  89 struct tgsi_exec_labels
  90 {
  91    unsigned labels[MAX_LABELS][2];
  92    unsigned count;
  93 };
  94
  95
  96 #define TGSI_EXEC_NUM_TEMPS       128
  97 #define TGSI_EXEC_NUM_TEMP_EXTRAS   6
  98 #define TGSI_EXEC_NUM_IMMEDIATES  256
  99
 100 /*
 101  * Locations of various utility registers (_I = Index, _C = Channel)
 102  */
 103 #define TGSI_EXEC_TEMP_00000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
 104 #define TGSI_EXEC_TEMP_00000000_C   0
 105
 106 #define TGSI_EXEC_TEMP_7FFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
 107 #define TGSI_EXEC_TEMP_7FFFFFFF_C   1
 108
 109 #define TGSI_EXEC_TEMP_80000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
 110 #define TGSI_EXEC_TEMP_80000000_C   2
 111
 112 #define TGSI_EXEC_TEMP_FFFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
 113 #define TGSI_EXEC_TEMP_FFFFFFFF_C   3
 114
 115 #define TGSI_EXEC_TEMP_ONE_I        (TGSI_EXEC_NUM_TEMPS + 1)
 116 #define TGSI_EXEC_TEMP_ONE_C        0
 117
 118 #define TGSI_EXEC_TEMP_TWO_I        (TGSI_EXEC_NUM_TEMPS + 1)
 119 #define TGSI_EXEC_TEMP_TWO_C        1
 120
 121 #define TGSI_EXEC_TEMP_128_I        (TGSI_EXEC_NUM_TEMPS + 1)
 122 #define TGSI_EXEC_TEMP_128_C        2
 123
 124 #define TGSI_EXEC_TEMP_MINUS_128_I  (TGSI_EXEC_NUM_TEMPS + 1)
 125 #define TGSI_EXEC_TEMP_MINUS_128_C  3
 126
 127 #define TGSI_EXEC_TEMP_KILMASK_I    (TGSI_EXEC_NUM_TEMPS + 2)
 128 #define TGSI_EXEC_TEMP_KILMASK_C    0
 129
 130 #define TGSI_EXEC_TEMP_OUTPUT_I     (TGSI_EXEC_NUM_TEMPS + 2)
 131 #define TGSI_EXEC_TEMP_OUTPUT_C     1
 132
 133 #define TGSI_EXEC_TEMP_PRIMITIVE_I  (TGSI_EXEC_NUM_TEMPS + 2)
 134 #define TGSI_EXEC_TEMP_PRIMITIVE_C  2
 135
 136 /* NVIDIA condition code (CC) vector
 137  */
 138 #define TGSI_EXEC_CC_GT       0x01
 139 #define TGSI_EXEC_CC_EQ       0x02
 140 #define TGSI_EXEC_CC_LT       0x04
 141 #define TGSI_EXEC_CC_UN       0x08
 142
 143 #define TGSI_EXEC_CC_X_MASK   0x000000ff
 144 #define TGSI_EXEC_CC_X_SHIFT  0
 145 #define TGSI_EXEC_CC_Y_MASK   0x0000ff00
 146 #define TGSI_EXEC_CC_Y_SHIFT  8
 147 #define TGSI_EXEC_CC_Z_MASK   0x00ff0000
 148 #define TGSI_EXEC_CC_Z_SHIFT  16
 149 #define TGSI_EXEC_CC_W_MASK   0xff000000
 150 #define TGSI_EXEC_CC_W_SHIFT  24
 151
 152 #define TGSI_EXEC_TEMP_CC_I         (TGSI_EXEC_NUM_TEMPS + 2)
 153 #define TGSI_EXEC_TEMP_CC_C         3
 154
 155 #define TGSI_EXEC_TEMP_THREE_I      (TGSI_EXEC_NUM_TEMPS + 3)
 156 #define TGSI_EXEC_TEMP_THREE_C      0
 157
 158 #define TGSI_EXEC_TEMP_HALF_I       (TGSI_EXEC_NUM_TEMPS + 3)
 159 #define TGSI_EXEC_TEMP_HALF_C       1
 160
 161 /* execution mask, each value is either 0 or ~0 */
 162 #define TGSI_EXEC_MASK_I            (TGSI_EXEC_NUM_TEMPS + 3)
 163 #define TGSI_EXEC_MASK_C            2
 164
 165 #define TGSI_EXEC_TEMP_R0           (TGSI_EXEC_NUM_TEMPS + 4)
 166
 167 #define TGSI_EXEC_TEMP_ADDR         (TGSI_EXEC_NUM_TEMPS + 5)
 168
 169
 170 #define TGSI_EXEC_MAX_COND_NESTING  20
 171 #define TGSI_EXEC_MAX_LOOP_NESTING  20
 172 #define TGSI_EXEC_MAX_CALL_NESTING  20
 173
 174 /* The maximum number of input attributes per vertex. For 2D
 175  * input register files, this is the stride between two 1D
 176  * arrays.
 177  */
 178 #define TGSI_EXEC_MAX_INPUT_ATTRIBS 17
 179
 180 /* The maximum number of constant vectors per constant buffer.
 181  */
 182 #define TGSI_EXEC_MAX_CONST_BUFFER  4096
 183
 184 /**
 185  * Run-time virtual machine state for executing TGSI shader.
 186  */
 187 struct tgsi_exec_machine
 188 {
 189    /* Total = program temporaries + internal temporaries
 190     *         + 1 padding to align to 16 bytes
 191     */
 192    struct tgsi_exec_vector       _Temps[TGSI_EXEC_NUM_TEMPS +
 193                                         TGSI_EXEC_NUM_TEMP_EXTRAS + 1];
 194
 195    /*
 196     * This will point to _Temps after aligning to 16B boundary.
 197     */
 198    struct tgsi_exec_vector       *Temps;
 199    struct tgsi_exec_vector       *Addrs;
 200
 201    struct tgsi_sampler           **Samplers;
 202
 203    float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
 204    unsigned                      ImmLimit;
 205    const float                   (*Consts)[4];
 206    struct tgsi_exec_vector       *Inputs;
 207    struct tgsi_exec_vector       *Outputs;
 208    const struct tgsi_token       *Tokens;
 209    unsigned                      Processor;
 210
 211    /* GEOMETRY processor only. */
 212    unsigned                      *Primitives;
 213
 214    /* FRAGMENT processor only. */
 215    const struct tgsi_interp_coef *InterpCoefs;
 216    struct tgsi_exec_vector       QuadPos;
 217
 218    /* Conditional execution masks */
 219    uint CondMask;  /**< For IF/ELSE/ENDIF */
 220    uint LoopMask;  /**< For BGNLOOP/ENDLOOP */
 221    uint ContMask;  /**< For loop CONT statements */
 222    uint FuncMask;  /**< For function calls */
 223    uint ExecMask;  /**< = CondMask & LoopMask */
 224
 225    /** Condition mask stack (for nested conditionals) */
 226    uint CondStack[TGSI_EXEC_MAX_COND_NESTING];
 227    int CondStackTop;
 228
 229    /** Loop mask stack (for nested loops) */
 230    uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
 231    int LoopStackTop;
 232
 233    /** Loop continue mask stack (see comments in tgsi_exec.c) */
 234    uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
 235    int ContStackTop;
 236
 237    /** Function execution mask stack (for executing subroutine code) */
 238    uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
 239    int FuncStackTop;
 240
 241    /** Function call stack for saving/restoring the program counter */
 242    uint CallStack[TGSI_EXEC_MAX_CALL_NESTING];
 243    int CallStackTop;
 244
 245    struct tgsi_full_instruction *Instructions;
 246    uint NumInstructions;
 247
 248    struct tgsi_full_declaration *Declarations;
 249    uint NumDeclarations;
 250
 251    struct tgsi_exec_labels Labels;
 252 };
 253
 254 void
 255 tgsi_exec_machine_init(
 256    struct tgsi_exec_machine *mach );
 257
 258
 259 void
 260 tgsi_exec_machine_bind_shader(
 261    struct tgsi_exec_machine *mach,
 262    const struct tgsi_token *tokens,
 263    uint numSamplers,
 264    struct tgsi_sampler **samplers);
 265
 266 uint
 267 tgsi_exec_machine_run(
 268    struct tgsi_exec_machine *mach );
 269
 270
 271 void
 272 tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach);
 273
 274
 275 static INLINE void
 276 tgsi_set_kill_mask(struct tgsi_exec_machine *mach, unsigned mask)
 277 {
 278    mach->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0] =
 279       mask;
 280 }
 281
 282
 283 /** Set execution mask values prior to executing the shader */
 284 static INLINE void
 285 tgsi_set_exec_mask(struct tgsi_exec_machine *mach,
 286                    boolean ch0, boolean ch1, boolean ch2, boolean ch3)
 287 {
 288    int *mask = mach->Temps[TGSI_EXEC_MASK_I].xyzw[TGSI_EXEC_MASK_C].i;
 289    mask[0] = ch0 ? ~0 : 0;
 290    mask[1] = ch1 ? ~0 : 0;
 291    mask[2] = ch2 ? ~0 : 0;
 292    mask[3] = ch3 ? ~0 : 0;
 293 }
 294
 295
 296 #if defined __cplusplus
 297 } /* extern "C" */
 298 #endif
 299
 300 #endif /* TGSI_EXEC_H */