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 struct softpipe_tile_cache;  /**< Opaque to TGSI */
  73
  74 /**
  75  * Information for sampling textures, which must be implemented
  76  * by code outside the TGSI executor.
  77  */
  78 struct tgsi_sampler
  79 {
  80    const struct pipe_sampler_state *state;
  81    struct pipe_texture *texture;
  82    /** Get samples for four fragments in a quad */
  83    void (*get_samples)(struct tgsi_sampler *sampler,
  84                        const float s[QUAD_SIZE],
  85                        const float t[QUAD_SIZE],
  86                        const float p[QUAD_SIZE],
  87                        float lodbias,
  88                        float rgba[NUM_CHANNELS][QUAD_SIZE]);
  89    void *pipe; /*XXX temporary*/
  90    struct softpipe_tile_cache *cache;
  91 };
  92
  93 /**
  94  * For branching/calling subroutines.
  95  */
  96 struct tgsi_exec_labels
  97 {
  98    unsigned labels[MAX_LABELS][2];
  99    unsigned count;
 100 };
 101
 102
 103 #define TGSI_EXEC_NUM_TEMPS       128
 104 #define TGSI_EXEC_NUM_TEMP_EXTRAS   6
 105 #define TGSI_EXEC_NUM_IMMEDIATES  256
 106
 107 /*
 108  * Locations of various utility registers (_I = Index, _C = Channel)
 109  */
 110 #define TGSI_EXEC_TEMP_00000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
 111 #define TGSI_EXEC_TEMP_00000000_C   0
 112
 113 #define TGSI_EXEC_TEMP_7FFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
 114 #define TGSI_EXEC_TEMP_7FFFFFFF_C   1
 115
 116 #define TGSI_EXEC_TEMP_80000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
 117 #define TGSI_EXEC_TEMP_80000000_C   2
 118
 119 #define TGSI_EXEC_TEMP_FFFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
 120 #define TGSI_EXEC_TEMP_FFFFFFFF_C   3
 121
 122 #define TGSI_EXEC_TEMP_ONE_I        (TGSI_EXEC_NUM_TEMPS + 1)
 123 #define TGSI_EXEC_TEMP_ONE_C        0
 124
 125 #define TGSI_EXEC_TEMP_TWO_I        (TGSI_EXEC_NUM_TEMPS + 1)
 126 #define TGSI_EXEC_TEMP_TWO_C        1
 127
 128 #define TGSI_EXEC_TEMP_128_I        (TGSI_EXEC_NUM_TEMPS + 1)
 129 #define TGSI_EXEC_TEMP_128_C        2
 130
 131 #define TGSI_EXEC_TEMP_MINUS_128_I  (TGSI_EXEC_NUM_TEMPS + 1)
 132 #define TGSI_EXEC_TEMP_MINUS_128_C  3
 133
 134 #define TGSI_EXEC_TEMP_KILMASK_I    (TGSI_EXEC_NUM_TEMPS + 2)
 135 #define TGSI_EXEC_TEMP_KILMASK_C    0
 136
 137 #define TGSI_EXEC_TEMP_OUTPUT_I     (TGSI_EXEC_NUM_TEMPS + 2)
 138 #define TGSI_EXEC_TEMP_OUTPUT_C     1
 139
 140 #define TGSI_EXEC_TEMP_PRIMITIVE_I  (TGSI_EXEC_NUM_TEMPS + 2)
 141 #define TGSI_EXEC_TEMP_PRIMITIVE_C  2
 142
 143 /* NVIDIA condition code (CC) vector
 144  */
 145 #define TGSI_EXEC_CC_GT       0x01
 146 #define TGSI_EXEC_CC_EQ       0x02
 147 #define TGSI_EXEC_CC_LT       0x04
 148 #define TGSI_EXEC_CC_UN       0x08
 149
 150 #define TGSI_EXEC_CC_X_MASK   0x000000ff
 151 #define TGSI_EXEC_CC_X_SHIFT  0
 152 #define TGSI_EXEC_CC_Y_MASK   0x0000ff00
 153 #define TGSI_EXEC_CC_Y_SHIFT  8
 154 #define TGSI_EXEC_CC_Z_MASK   0x00ff0000
 155 #define TGSI_EXEC_CC_Z_SHIFT  16
 156 #define TGSI_EXEC_CC_W_MASK   0xff000000
 157 #define TGSI_EXEC_CC_W_SHIFT  24
 158
 159 #define TGSI_EXEC_TEMP_CC_I         (TGSI_EXEC_NUM_TEMPS + 2)
 160 #define TGSI_EXEC_TEMP_CC_C         3
 161
 162 #define TGSI_EXEC_TEMP_THREE_I      (TGSI_EXEC_NUM_TEMPS + 3)
 163 #define TGSI_EXEC_TEMP_THREE_C      0
 164
 165 #define TGSI_EXEC_TEMP_HALF_I       (TGSI_EXEC_NUM_TEMPS + 3)
 166 #define TGSI_EXEC_TEMP_HALF_C       1
 167
 168 /* execution mask, each value is either 0 or ~0 */
 169 #define TGSI_EXEC_MASK_I            (TGSI_EXEC_NUM_TEMPS + 3)
 170 #define TGSI_EXEC_MASK_C            2
 171
 172 #define TGSI_EXEC_TEMP_R0           (TGSI_EXEC_NUM_TEMPS + 4)
 173
 174 #define TGSI_EXEC_TEMP_ADDR         (TGSI_EXEC_NUM_TEMPS + 5)
 175
 176
 177 #define TGSI_EXEC_MAX_COND_NESTING  20
 178 #define TGSI_EXEC_MAX_LOOP_NESTING  20
 179 #define TGSI_EXEC_MAX_CALL_NESTING  20
 180
 181 /* The maximum number of input attributes per vertex. For 2D
 182  * input register files, this is the stride between two 1D
 183  * arrays.
 184  */
 185 #define TGSI_EXEC_MAX_INPUT_ATTRIBS 17
 186
 187 /* The maximum number of constant vectors per constant buffer.
 188  */
 189 #define TGSI_EXEC_MAX_CONST_BUFFER  4096
 190
 191 /**
 192  * Run-time virtual machine state for executing TGSI shader.
 193  */
 194 struct tgsi_exec_machine
 195 {
 196    /* Total = program temporaries + internal temporaries
 197     *         + 1 padding to align to 16 bytes
 198     */
 199    struct tgsi_exec_vector       _Temps[TGSI_EXEC_NUM_TEMPS +
 200                                         TGSI_EXEC_NUM_TEMP_EXTRAS + 1];
 201
 202    /*
 203     * This will point to _Temps after aligning to 16B boundary.
 204     */
 205    struct tgsi_exec_vector       *Temps;
 206    struct tgsi_exec_vector       *Addrs;
 207
 208    struct tgsi_sampler           *Samplers;
 209
 210    float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
 211    unsigned                      ImmLimit;
 212    const float                   (*Consts)[4];
 213    struct tgsi_exec_vector       *Inputs;
 214    struct tgsi_exec_vector       *Outputs;
 215    const struct tgsi_token       *Tokens;
 216    unsigned                      Processor;
 217
 218    /* GEOMETRY processor only. */
 219    unsigned                      *Primitives;
 220
 221    /* FRAGMENT processor only. */
 222    const struct tgsi_interp_coef *InterpCoefs;
 223    struct tgsi_exec_vector       QuadPos;
 224
 225    /* Conditional execution masks */
 226    uint CondMask;  /**< For IF/ELSE/ENDIF */
 227    uint LoopMask;  /**< For BGNLOOP/ENDLOOP */
 228    uint ContMask;  /**< For loop CONT statements */
 229    uint FuncMask;  /**< For function calls */
 230    uint ExecMask;  /**< = CondMask & LoopMask */
 231
 232    /** Condition mask stack (for nested conditionals) */
 233    uint CondStack[TGSI_EXEC_MAX_COND_NESTING];
 234    int CondStackTop;
 235
 236    /** Loop mask stack (for nested loops) */
 237    uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
 238    int LoopStackTop;
 239
 240    /** Loop continue mask stack (see comments in tgsi_exec.c) */
 241    uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
 242    int ContStackTop;
 243
 244    /** Function execution mask stack (for executing subroutine code) */
 245    uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
 246    int FuncStackTop;
 247
 248    /** Function call stack for saving/restoring the program counter */
 249    uint CallStack[TGSI_EXEC_MAX_CALL_NESTING];
 250    int CallStackTop;
 251
 252    struct tgsi_full_instruction *Instructions;
 253    uint NumInstructions;
 254
 255    struct tgsi_full_declaration *Declarations;
 256    uint NumDeclarations;
 257
 258    struct tgsi_exec_labels Labels;
 259 };
 260
 261 void
 262 tgsi_exec_machine_init(
 263    struct tgsi_exec_machine *mach );
 264
 265
 266 void
 267 tgsi_exec_machine_bind_shader(
 268    struct tgsi_exec_machine *mach,
 269    const struct tgsi_token *tokens,
 270    uint numSamplers,
 271    struct tgsi_sampler *samplers);
 272
 273 uint
 274 tgsi_exec_machine_run(
 275    struct tgsi_exec_machine *mach );
 276
 277
 278 void
 279 tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach);
 280
 281
 282 static INLINE void
 283 tgsi_set_kill_mask(struct tgsi_exec_machine *mach, unsigned mask)
 284 {
 285    mach->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0] =
 286       mask;
 287 }
 288
 289
 290 /** Set execution mask values prior to executing the shader */
 291 static INLINE void
 292 tgsi_set_exec_mask(struct tgsi_exec_machine *mach,
 293                    boolean ch0, boolean ch1, boolean ch2, boolean ch3)
 294 {
 295    int *mask = mach->Temps[TGSI_EXEC_MASK_I].xyzw[TGSI_EXEC_MASK_C].i;
 296    mask[0] = ch0 ? ~0 : 0;
 297    mask[1] = ch1 ? ~0 : 0;
 298    mask[2] = ch2 ? ~0 : 0;
 299    mask[3] = ch3 ? ~0 : 0;
 300 }
 301
 302
 303 #if defined __cplusplus
 304 } /* extern "C" */
 305 #endif
 306
 307 #endif /* TGSI_EXEC_H */