src/gallium/drivers/svga/svga_tgsi_emit.h

   1 /**********************************************************
   2  * Copyright 2008-2009 VMware, Inc.  All rights reserved.
   3  *
   4  * Permission is hereby granted, free of charge, to any person
   5  * obtaining a copy of this software and associated documentation
   6  * files (the "Software"), to deal in the Software without
   7  * restriction, including without limitation the rights to use, copy,
   8  * modify, merge, publish, distribute, sublicense, and/or sell copies
   9  * of the Software, and to permit persons to whom the Software is
  10  * furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice shall be
  13  * included in all copies or substantial portions of the Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  16  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  17  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  18  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  19  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  20  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  21  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  22  * SOFTWARE.
  23  *
  24  **********************************************************/
  25
  26 #ifndef SVGA_TGSI_EMIT_H
  27 #define SVGA_TGSI_EMIT_H
  28
  29 #include "tgsi/tgsi_scan.h"
  30 #include "svga_hw_reg.h"
  31 #include "svga_tgsi.h"
  32 #include "svga3d_shaderdefs.h"
  33
  34 struct src_register
  35 {
  36    SVGA3dShaderSrcToken base;
  37    SVGA3dShaderSrcToken indirect;
  38 };
  39
  40
  41 struct svga_arl_consts {
  42    int number;
  43    int idx;
  44    int swizzle;
  45    int arl_num;
  46 };
  47
  48 /* Internal functions:
  49  */
  50
  51 struct svga_shader_emitter
  52 {
  53    boolean use_sm30;
  54
  55    unsigned size;
  56    char *buf;
  57    char *ptr;
  58
  59    union svga_compile_key key;
  60    struct tgsi_shader_info info;
  61    int unit;
  62
  63    int imm_start;
  64
  65    int nr_hw_float_const;
  66    int nr_hw_int_const;
  67    int nr_hw_temp;
  68
  69    int insn_offset;
  70
  71    int internal_temp_count;
  72    int internal_imm_count;
  73
  74    int internal_color_idx[2]; /* diffuse, specular */
  75    int internal_color_count;
  76
  77    boolean emitted_vface;
  78    boolean emit_frontface;
  79    int internal_frontface_idx;
  80
  81    int ps30_input_count;
  82    int vs30_output_count;
  83
  84    int dynamic_branching_level;
  85
  86    boolean in_main_func;
  87
  88    boolean created_zero_immediate;
  89    int zero_immediate_idx;
  90
  91    boolean created_loop_const;
  92    int loop_const_idx;
  93
  94    boolean created_sincos_consts;
  95    int sincos_consts_idx;
  96
  97    unsigned inverted_texcoords;  /**< bitmask of which texcoords are flipped */
  98    struct src_register ps_true_texcoord[PIPE_MAX_ATTRIBS];
  99    struct src_register ps_inverted_texcoord[PIPE_MAX_ATTRIBS];
 100    unsigned ps_inverted_texcoord_input[PIPE_MAX_ATTRIBS];
 101
 102    unsigned label[32];
 103    unsigned nr_labels;
 104
 105    struct src_register input_map[PIPE_MAX_ATTRIBS];
 106    SVGA3dShaderDestToken output_map[PIPE_MAX_ATTRIBS];
 107
 108    boolean ps_reads_pos;
 109    boolean emitted_depth_fog;
 110    struct src_register ps_true_pos;
 111    struct src_register ps_depth_pos;
 112    SVGA3dShaderDestToken ps_temp_pos;
 113
 114    /* shared input for depth and fog */
 115    struct src_register ps_depth_fog;
 116
 117    struct src_register imm_0055;
 118    SVGA3dShaderDestToken temp_pos;
 119    SVGA3dShaderDestToken true_pos;
 120    SVGA3dShaderDestToken depth_pos;
 121
 122    /* shared output for depth and fog */
 123    SVGA3dShaderDestToken vs_depth_fog;
 124
 125    SVGA3dShaderDestToken temp_col[PIPE_MAX_COLOR_BUFS];
 126    SVGA3dShaderDestToken true_col[PIPE_MAX_COLOR_BUFS];
 127
 128    SVGA3dShaderDestToken temp_psiz;
 129    SVGA3dShaderDestToken true_psiz;
 130
 131    struct svga_arl_consts arl_consts[12];
 132    int num_arl_consts;
 133    int current_arl;
 134 };
 135
 136
 137 boolean svga_shader_emit_dword( struct svga_shader_emitter *emit,
 138                                 unsigned dword );
 139
 140 boolean svga_shader_emit_dwords( struct svga_shader_emitter *emit,
 141                                  const unsigned *dwords,
 142                                  unsigned nr );
 143
 144 boolean svga_shader_emit_opcode( struct svga_shader_emitter *emit,
 145                                  unsigned opcode );
 146
 147 boolean svga_shader_emit_instructions( struct svga_shader_emitter *emit,
 148                                        const struct tgsi_token *tokens );
 149
 150 boolean svga_translate_decl_sm20( struct svga_shader_emitter *emit,
 151                                const struct tgsi_full_declaration *decl );
 152
 153 boolean svga_translate_decl_sm30( struct svga_shader_emitter *emit,
 154                                const struct tgsi_full_declaration *decl );
 155
 156
 157 static INLINE boolean emit_dst( struct svga_shader_emitter *emit,
 158                          SVGA3dShaderDestToken dest )
 159 {
 160    assert(dest.reserved0);
 161    assert(dest.mask);
 162    return svga_shader_emit_dword( emit, dest.value );
 163 }
 164
 165 static INLINE boolean emit_src( struct svga_shader_emitter *emit,
 166                          const struct src_register src )
 167 {
 168    if (src.base.relAddr) {
 169       assert(src.base.reserved0);
 170       assert(src.indirect.reserved0);
 171       return (svga_shader_emit_dword( emit, src.base.value ) &&
 172               svga_shader_emit_dword( emit, src.indirect.value ));
 173    }
 174    else {
 175       assert(src.base.reserved0);
 176       return svga_shader_emit_dword( emit, src.base.value );
 177    }
 178 }
 179
 180
 181 static INLINE boolean emit_instruction( struct svga_shader_emitter *emit,
 182                                  SVGA3dShaderInstToken opcode )
 183 {
 184    return svga_shader_emit_opcode( emit, opcode.value );
 185 }
 186
 187
 188 static INLINE boolean emit_op1( struct svga_shader_emitter *emit,
 189                          SVGA3dShaderInstToken inst,
 190                          SVGA3dShaderDestToken dest,
 191                          struct src_register src0 )
 192 {
 193    return (emit_instruction( emit, inst ) &&
 194            emit_dst( emit, dest ) &&
 195            emit_src( emit, src0 ));
 196 }
 197
 198 static INLINE boolean emit_op2( struct svga_shader_emitter *emit,
 199                      SVGA3dShaderInstToken inst,
 200                      SVGA3dShaderDestToken dest,
 201                      struct src_register src0,
 202                      struct src_register src1 )
 203 {
 204    return (emit_instruction( emit, inst ) &&
 205            emit_dst( emit, dest ) &&
 206            emit_src( emit, src0 ) &&
 207            emit_src( emit, src1 ));
 208 }
 209
 210 static INLINE boolean emit_op3( struct svga_shader_emitter *emit,
 211                          SVGA3dShaderInstToken inst,
 212                          SVGA3dShaderDestToken dest,
 213                          struct src_register src0,
 214                          struct src_register src1,
 215                          struct src_register src2 )
 216 {
 217    return (emit_instruction( emit, inst ) &&
 218            emit_dst( emit, dest ) &&
 219            emit_src( emit, src0 ) &&
 220            emit_src( emit, src1 ) &&
 221            emit_src( emit, src2 ));
 222 }
 223
 224
 225 static INLINE boolean emit_op4( struct svga_shader_emitter *emit,
 226                                 SVGA3dShaderInstToken inst,
 227                                 SVGA3dShaderDestToken dest,
 228                                 struct src_register src0,
 229                                 struct src_register src1,
 230                                 struct src_register src2,
 231                                 struct src_register src3)
 232 {
 233    return (emit_instruction( emit, inst ) &&
 234            emit_dst( emit, dest ) &&
 235            emit_src( emit, src0 ) &&
 236            emit_src( emit, src1 ) &&
 237            emit_src( emit, src2 ) &&
 238            emit_src( emit, src3 ));
 239 }
 240
 241
 242 #define TRANSLATE_SWIZZLE(x,y,z,w)  ((x) | ((y) << 2) | ((z) << 4) | ((w) << 6))
 243 #define SWIZZLE_XYZW  \
 244  TRANSLATE_SWIZZLE(TGSI_SWIZZLE_X,TGSI_SWIZZLE_Y,TGSI_SWIZZLE_Z,TGSI_SWIZZLE_W)
 245 #define SWIZZLE_XXXX  \
 246  TRANSLATE_SWIZZLE(TGSI_SWIZZLE_X,TGSI_SWIZZLE_X,TGSI_SWIZZLE_X,TGSI_SWIZZLE_X)
 247 #define SWIZZLE_YYYY  \
 248  TRANSLATE_SWIZZLE(TGSI_SWIZZLE_Y,TGSI_SWIZZLE_Y,TGSI_SWIZZLE_Y,TGSI_SWIZZLE_Y)
 249 #define SWIZZLE_ZZZZ  \
 250  TRANSLATE_SWIZZLE(TGSI_SWIZZLE_Z,TGSI_SWIZZLE_Z,TGSI_SWIZZLE_Z,TGSI_SWIZZLE_Z)
 251 #define SWIZZLE_WWWW  \
 252  TRANSLATE_SWIZZLE(TGSI_SWIZZLE_W,TGSI_SWIZZLE_W,TGSI_SWIZZLE_W,TGSI_SWIZZLE_W)
 253
 254
 255
 256 static INLINE SVGA3dShaderInstToken
 257 inst_token( unsigned opcode )
 258 {
 259    SVGA3dShaderInstToken inst;
 260
 261    inst.value = 0;
 262    inst.op = opcode;
 263
 264    return inst;
 265 }
 266
 267 static INLINE SVGA3dShaderDestToken
 268 dst_register( unsigned file,
 269               int number )
 270 {
 271    SVGA3dShaderDestToken dest;
 272
 273    dest.value = 0;
 274    dest.num = number;
 275    dest.type_upper = file >> 3;
 276    dest.relAddr = 0;
 277    dest.reserved1 = 0;
 278    dest.mask = 0xf;
 279    dest.dstMod = 0;
 280    dest.shfScale = 0;
 281    dest.type_lower = file & 0x7;
 282    dest.reserved0 = 1;          /* is_reg */
 283
 284    return dest;
 285 }
 286
 287 static INLINE SVGA3dShaderDestToken
 288 writemask( SVGA3dShaderDestToken dest,
 289            unsigned mask )
 290 {
 291    assert(dest.mask & mask);
 292    dest.mask &= mask;
 293    return dest;
 294 }
 295
 296
 297 static INLINE SVGA3dShaderSrcToken
 298 src_token( unsigned file, int number )
 299 {
 300    SVGA3dShaderSrcToken src;
 301
 302    src.value = 0;
 303    src.num = number;
 304    src.type_upper = file >> 3;
 305    src.relAddr = 0;
 306    src.reserved1 = 0;
 307    src.swizzle = SWIZZLE_XYZW;
 308    src.srcMod = 0;
 309    src.type_lower = file & 0x7;
 310    src.reserved0 = 1;           /* is_reg */
 311
 312    return src;
 313 }
 314
 315
 316 static INLINE struct src_register
 317 absolute( struct src_register src )
 318 {
 319    src.base.srcMod = SVGA3DSRCMOD_ABS;
 320
 321    return src;
 322 }
 323
 324
 325 static INLINE struct src_register
 326 negate( struct src_register src )
 327 {
 328    switch (src.base.srcMod) {
 329    case SVGA3DSRCMOD_ABS:
 330       src.base.srcMod = SVGA3DSRCMOD_ABSNEG;
 331       break;
 332    case SVGA3DSRCMOD_ABSNEG:
 333       src.base.srcMod = SVGA3DSRCMOD_ABS;
 334       break;
 335    case SVGA3DSRCMOD_NEG:
 336       src.base.srcMod = SVGA3DSRCMOD_NONE;
 337       break;
 338    case SVGA3DSRCMOD_NONE:
 339       src.base.srcMod = SVGA3DSRCMOD_NEG;
 340       break;
 341    }
 342    return src;
 343 }
 344
 345
 346 static INLINE struct src_register
 347 src_register( unsigned file, int number )
 348 {
 349    struct src_register src;
 350
 351    src.base = src_token( file, number );
 352    src.indirect.value = 0;
 353
 354    return src;
 355 }
 356
 357 static INLINE SVGA3dShaderDestToken dst( struct src_register src )
 358 {
 359    return dst_register( SVGA3dShaderGetRegType( src.base.value ),
 360                         src.base.num );
 361 }
 362
 363 static INLINE struct src_register src( SVGA3dShaderDestToken dst )
 364 {
 365    return src_register( SVGA3dShaderGetRegType( dst.value ),
 366                         dst.num );
 367 }
 368
 369 static INLINE ubyte svga_tgsi_sampler_type( struct svga_shader_emitter *emit,
 370                                             int idx )
 371 {
 372    switch (emit->key.fkey.tex[idx].texture_target) {
 373    case PIPE_TEXTURE_1D:
 374       return SVGA3DSAMP_2D;
 375    case PIPE_TEXTURE_2D:
 376    case PIPE_TEXTURE_RECT:
 377       return SVGA3DSAMP_2D;
 378    case PIPE_TEXTURE_3D:
 379       return SVGA3DSAMP_VOLUME;
 380    case PIPE_TEXTURE_CUBE:
 381       return SVGA3DSAMP_CUBE;
 382    }
 383
 384    return SVGA3DSAMP_UNKNOWN;
 385 }
 386
 387 #endif