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 #include "pipe/p_state.h"
  33
  34 #if defined __cplusplus
  35 extern "C" {
  36 #endif
  37
  38 #define MAX_LABELS (4 * 1024)  /**< basically, max instructions */
  39
  40 #define NUM_CHANNELS 4  /* R,G,B,A */
  41 #define QUAD_SIZE    4  /* 4 pixel/quad */
  42
  43 /**
  44   * Registers may be treated as float, signed int or unsigned int.
  45   */
  46 union tgsi_exec_channel
  47 {
  48    float    f[QUAD_SIZE];
  49    int      i[QUAD_SIZE];
  50    unsigned u[QUAD_SIZE];
  51 };
  52
  53 /**
  54   * A vector[RGBA] of channels[4 pixels]
  55   */
  56 struct tgsi_exec_vector
  57 {
  58    union tgsi_exec_channel xyzw[NUM_CHANNELS];
  59 };
  60
  61 /**
  62  * For fragment programs, information for computing fragment input
  63  * values from plane equation of the triangle/line.
  64  */
  65 struct tgsi_interp_coef
  66 {
  67    float a0[NUM_CHANNELS];      /* in an xyzw layout */
  68    float dadx[NUM_CHANNELS];
  69    float dady[NUM_CHANNELS];
  70 };
  71
  72 /**
  73  * Information for sampling textures, which must be implemented
  74  * by code outside the TGSI executor.
  75  */
  76 struct tgsi_sampler
  77 {
  78    /** Get samples for four fragments in a quad */
  79    void (*get_samples)(struct tgsi_sampler *sampler,
  80                        const float s[QUAD_SIZE],
  81                        const float t[QUAD_SIZE],
  82                        const float p[QUAD_SIZE],
  83                        float lodbias,
  84                        float rgba[NUM_CHANNELS][QUAD_SIZE]);
  85 };
  86
  87 /**
  88  * For branching/calling subroutines.
  89  */
  90 struct tgsi_exec_labels
  91 {
  92    unsigned labels[MAX_LABELS][2];
  93    unsigned count;
  94 };
  95
  96
  97 #define TGSI_EXEC_NUM_TEMPS       128
  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 /* 4 register buffer for various purposes */
 166 #define TGSI_EXEC_TEMP_R0           (TGSI_EXEC_NUM_TEMPS + 4)
 167 #define TGSI_EXEC_NUM_TEMP_R        4
 168
 169 #define TGSI_EXEC_TEMP_ADDR         (TGSI_EXEC_NUM_TEMPS + 8)
 170 #define TGSI_EXEC_NUM_ADDRS         1
 171 #define TGSI_EXEC_NUM_TEMP_EXTRAS   9
 172
 173
 174
 175 #define TGSI_EXEC_MAX_COND_NESTING  20
 176 #define TGSI_EXEC_MAX_LOOP_NESTING  20
 177 #define TGSI_EXEC_MAX_CALL_NESTING  20
 178
 179 /* The maximum number of input attributes per vertex. For 2D
 180  * input register files, this is the stride between two 1D
 181  * arrays.
 182  */
 183 #define TGSI_EXEC_MAX_INPUT_ATTRIBS 17
 184
 185 /* The maximum number of constant vectors per constant buffer.
 186  */
 187 #define TGSI_EXEC_MAX_CONST_BUFFER  4096
 188
 189 /**
 190  * Run-time virtual machine state for executing TGSI shader.
 191  */
 192 struct tgsi_exec_machine
 193 {
 194    /* Total = program temporaries + internal temporaries
 195     */
 196    struct tgsi_exec_vector       Temps[TGSI_EXEC_NUM_TEMPS +
 197                                        TGSI_EXEC_NUM_TEMP_EXTRAS];
 198
 199    float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
 200
 201    struct tgsi_exec_vector       Inputs[PIPE_MAX_ATTRIBS];
 202    struct tgsi_exec_vector       Outputs[PIPE_MAX_ATTRIBS];
 203
 204    struct tgsi_exec_vector       *Addrs;
 205
 206    struct tgsi_sampler           **Samplers;
 207
 208    unsigned                      ImmLimit;
 209    const float                   (*Consts)[4];
 210    const struct tgsi_token       *Tokens;   /**< Declarations, instructions */
 211    unsigned                      Processor; /**< TGSI_PROCESSOR_x */
 212
 213    /* GEOMETRY processor only. */
 214    unsigned                      *Primitives;
 215
 216    /* FRAGMENT processor only. */
 217    const struct tgsi_interp_coef *InterpCoefs;
 218    struct tgsi_exec_vector       QuadPos;
 219
 220    /* Conditional execution masks */
 221    uint CondMask;  /**< For IF/ELSE/ENDIF */
 222    uint LoopMask;  /**< For BGNLOOP/ENDLOOP */
 223    uint ContMask;  /**< For loop CONT statements */
 224    uint FuncMask;  /**< For function calls */
 225    uint ExecMask;  /**< = CondMask & LoopMask */
 226
 227    /** Condition mask stack (for nested conditionals) */
 228    uint CondStack[TGSI_EXEC_MAX_COND_NESTING];
 229    int CondStackTop;
 230
 231    /** Loop mask stack (for nested loops) */
 232    uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
 233    int LoopStackTop;
 234
 235    /** Loop label stack */
 236    uint LoopLabelStack[TGSI_EXEC_MAX_LOOP_NESTING];
 237    int LoopLabelStackTop;
 238
 239    /** Loop counter stack (x = count, y = current, z = step) */
 240    struct tgsi_exec_vector LoopCounterStack[TGSI_EXEC_MAX_LOOP_NESTING];
 241    int LoopCounterStackTop;
 242
 243    /** Loop continue mask stack (see comments in tgsi_exec.c) */
 244    uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
 245    int ContStackTop;
 246
 247    /** Function execution mask stack (for executing subroutine code) */
 248    uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
 249    int FuncStackTop;
 250
 251    /** Function call stack for saving/restoring the program counter */
 252    uint CallStack[TGSI_EXEC_MAX_CALL_NESTING];
 253    int CallStackTop;
 254
 255    struct tgsi_full_instruction *Instructions;
 256    uint NumInstructions;
 257
 258    struct tgsi_full_declaration *Declarations;
 259    uint NumDeclarations;
 260
 261    struct tgsi_exec_labels Labels;
 262 };
 263
 264 struct tgsi_exec_machine *
 265 tgsi_exec_machine_create( void );
 266
 267 void
 268 tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach);
 269
 270
 271 void
 272 tgsi_exec_machine_bind_shader(
 273    struct tgsi_exec_machine *mach,
 274    const struct tgsi_token *tokens,
 275    uint numSamplers,
 276    struct tgsi_sampler **samplers);
 277
 278 uint
 279 tgsi_exec_machine_run(
 280    struct tgsi_exec_machine *mach );
 281
 282
 283 void
 284 tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach);
 285
 286
 287 boolean
 288 tgsi_check_soa_dependencies(const struct tgsi_full_instruction *inst);
 289
 290
 291 static INLINE void
 292 tgsi_set_kill_mask(struct tgsi_exec_machine *mach, unsigned mask)
 293 {
 294    mach->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0] =
 295       mask;
 296 }
 297
 298
 299 /** Set execution mask values prior to executing the shader */
 300 static INLINE void
 301 tgsi_set_exec_mask(struct tgsi_exec_machine *mach,
 302                    boolean ch0, boolean ch1, boolean ch2, boolean ch3)
 303 {
 304    int *mask = mach->Temps[TGSI_EXEC_MASK_I].xyzw[TGSI_EXEC_MASK_C].i;
 305    mask[0] = ch0 ? ~0 : 0;
 306    mask[1] = ch1 ? ~0 : 0;
 307    mask[2] = ch2 ? ~0 : 0;
 308    mask[3] = ch3 ? ~0 : 0;
 309 }
 310
 311
 312 #if defined __cplusplus
 313 } /* extern "C" */
 314 #endif
 315
 316 #endif /* TGSI_EXEC_H */