src/mesa/state_tracker/st_program.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2007 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   * Authors:
  29   *   Keith Whitwell <keith@tungstengraphics.com>
  30   *   Brian Paul
  31   */
  32
  33
  34 #include "main/imports.h"
  35 #include "main/mtypes.h"
  36 #include "shader/prog_print.h"
  37
  38 #include "pipe/p_context.h"
  39 #include "pipe/p_defines.h"
  40 #include "pipe/p_shader_tokens.h"
  41 #include "draw/draw_context.h"
  42 #include "tgsi/util/tgsi_dump.h"
  43
  44 #include "st_context.h"
  45 #include "st_atom.h"
  46 #include "st_program.h"
  47 #include "st_mesa_to_tgsi.h"
  48 #include "cso_cache/cso_context.h"
  49
  50
  51 #define TGSI_DEBUG 0
  52
  53
  54 /** XXX we should use the version of this from p_util.h but including
  55  * that header causes symbol collisions.
  56  */
  57 static INLINE void *
  58 mem_dup(const void *src, uint size)
  59 {
  60    void *dup = MALLOC(size);
  61    if (dup)
  62       memcpy(dup, src, size);
  63    return dup;
  64 }
  65
  66
  67
  68 /**
  69  * Translate a Mesa vertex shader into a TGSI shader.
  70  * \param outputMapping  to map vertex program output registers to TGSI
  71  *                       output slots
  72  * \param tokensOut  destination for TGSI tokens
  73  * \return  pointer to cached pipe_shader object.
  74  */
  75 void
  76 st_translate_vertex_program(struct st_context *st,
  77                             struct st_vertex_program *stvp,
  78                             const GLuint outputMapping[])
  79 {
  80    struct pipe_context *pipe = st->pipe;
  81    struct tgsi_token tokens[ST_MAX_SHADER_TOKENS];
  82    GLuint defaultOutputMapping[VERT_RESULT_MAX];
  83    struct pipe_shader_state vs;
  84    GLuint attr, i;
  85    GLuint num_generic = 0;
  86    GLuint num_tokens;
  87
  88    ubyte vs_input_semantic_name[PIPE_MAX_SHADER_INPUTS];
  89    ubyte vs_input_semantic_index[PIPE_MAX_SHADER_INPUTS];
  90    uint vs_num_inputs = 0;
  91
  92    ubyte vs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
  93    ubyte vs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
  94    uint vs_num_outputs = 0;
  95
  96    memset(&vs, 0, sizeof(vs));
  97
  98    /*
  99     * Determine number of inputs, the mappings between VERT_ATTRIB_x
 100     * and TGSI generic input indexes, plus input attrib semantic info.
 101     */
 102    for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
 103       if (stvp->Base.Base.InputsRead & (1 << attr)) {
 104          const GLuint slot = vs_num_inputs;
 105
 106          vs_num_inputs++;
 107
 108          stvp->input_to_index[attr] = slot;
 109          stvp->index_to_input[slot] = attr;
 110
 111          switch (attr) {
 112          case VERT_ATTRIB_POS:
 113             vs_input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
 114             vs_input_semantic_index[slot] = 0;
 115             break;
 116          case VERT_ATTRIB_WEIGHT:
 117             /* fall-through */
 118          case VERT_ATTRIB_NORMAL:
 119             /* just label as a generic */
 120             vs_input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 121             vs_input_semantic_index[slot] = 0;
 122             break;
 123          case VERT_ATTRIB_COLOR0:
 124             vs_input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 125             vs_input_semantic_index[slot] = 0;
 126             break;
 127          case VERT_ATTRIB_COLOR1:
 128             vs_input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 129             vs_input_semantic_index[slot] = 1;
 130             break;
 131          case VERT_ATTRIB_FOG:
 132             vs_input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
 133             vs_input_semantic_index[slot] = 0;
 134             break;
 135          case VERT_ATTRIB_TEX0:
 136          case VERT_ATTRIB_TEX1:
 137          case VERT_ATTRIB_TEX2:
 138          case VERT_ATTRIB_TEX3:
 139          case VERT_ATTRIB_TEX4:
 140          case VERT_ATTRIB_TEX5:
 141          case VERT_ATTRIB_TEX6:
 142          case VERT_ATTRIB_TEX7:
 143             vs_input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 144             vs_input_semantic_index[slot] = num_generic++;
 145             break;
 146          case VERT_ATTRIB_GENERIC0:
 147          case VERT_ATTRIB_GENERIC1:
 148          case VERT_ATTRIB_GENERIC2:
 149          case VERT_ATTRIB_GENERIC3:
 150          case VERT_ATTRIB_GENERIC4:
 151          case VERT_ATTRIB_GENERIC5:
 152          case VERT_ATTRIB_GENERIC6:
 153          case VERT_ATTRIB_GENERIC7:
 154             assert(attr < VERT_ATTRIB_MAX);
 155             vs_input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 156             vs_input_semantic_index[slot] = num_generic++;
 157             break;
 158          default:
 159             assert(0);
 160          }
 161       }
 162    }
 163
 164    /* initialize output semantics to defaults */
 165    for (i = 0; i < PIPE_MAX_SHADER_OUTPUTS; i++) {
 166       vs_output_semantic_name[i] = TGSI_SEMANTIC_GENERIC;
 167       vs_output_semantic_index[i] = 0;
 168    }
 169
 170    num_generic = 0;
 171    /*
 172     * Determine number of outputs, the (default) output register
 173     * mapping and the semantic information for each output.
 174     */
 175    for (attr = 0; attr < VERT_RESULT_MAX; attr++) {
 176       if (stvp->Base.Base.OutputsWritten & (1 << attr)) {
 177          GLuint slot;
 178
 179          if (outputMapping) {
 180             slot = outputMapping[attr];
 181             assert(slot != ~0);
 182          }
 183          else {
 184             slot = vs_num_outputs;
 185             vs_num_outputs++;
 186             defaultOutputMapping[attr] = slot;
 187          }
 188
 189          /*
 190          printf("Output %u -> slot %u\n", attr, slot);
 191          */
 192
 193          switch (attr) {
 194          case VERT_RESULT_HPOS:
 195             assert(slot == 0);
 196             vs_output_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
 197             vs_output_semantic_index[slot] = 0;
 198             break;
 199          case VERT_RESULT_COL0:
 200             vs_output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 201             vs_output_semantic_index[slot] = 0;
 202             break;
 203          case VERT_RESULT_COL1:
 204             vs_output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 205             vs_output_semantic_index[slot] = 1;
 206             break;
 207          case VERT_RESULT_BFC0:
 208             vs_output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
 209             vs_output_semantic_index[slot] = 0;
 210             break;
 211          case VERT_RESULT_BFC1:
 212             vs_output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
 213             vs_output_semantic_index[slot] = 1;
 214             break;
 215          case VERT_RESULT_FOGC:
 216             vs_output_semantic_name[slot] = TGSI_SEMANTIC_FOG;
 217             vs_output_semantic_index[slot] = 0;
 218             break;
 219          case VERT_RESULT_PSIZ:
 220             vs_output_semantic_name[slot] = TGSI_SEMANTIC_PSIZE;
 221             vs_output_semantic_index[slot] = 0;
 222             break;
 223          case VERT_RESULT_EDGE:
 224             assert(0);
 225             break;
 226          case VERT_RESULT_TEX0:
 227          case VERT_RESULT_TEX1:
 228          case VERT_RESULT_TEX2:
 229          case VERT_RESULT_TEX3:
 230          case VERT_RESULT_TEX4:
 231          case VERT_RESULT_TEX5:
 232          case VERT_RESULT_TEX6:
 233          case VERT_RESULT_TEX7:
 234             vs_output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 235             vs_output_semantic_index[slot] = num_generic++;
 236             break;
 237          case VERT_RESULT_VAR0:
 238             /* fall-through */
 239          default:
 240             assert(attr - VERT_RESULT_VAR0 < MAX_VARYING);
 241             vs_output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 242             vs_output_semantic_index[slot] = num_generic++;
 243          }
 244       }
 245    }
 246
 247    assert(vs_output_semantic_name[0] == TGSI_SEMANTIC_POSITION);
 248
 249
 250    if (outputMapping) {
 251       /* find max output slot referenced to compute vs_num_outputs */
 252       GLuint maxSlot = 0;
 253       for (attr = 0; attr < VERT_RESULT_MAX; attr++) {
 254          if (outputMapping[attr] != ~0 && outputMapping[attr] > maxSlot)
 255             maxSlot = outputMapping[attr];
 256       }
 257       vs_num_outputs = maxSlot + 1;
 258    }
 259    else {
 260       outputMapping = defaultOutputMapping;
 261    }
 262
 263    /* XXX: fix static allocation of tokens:
 264     */
 265    num_tokens = tgsi_translate_mesa_program( TGSI_PROCESSOR_VERTEX,
 266                                 &stvp->Base.Base,
 267                                 /* inputs */
 268                                 vs_num_inputs,
 269                                 stvp->input_to_index,
 270                                 vs_input_semantic_name,
 271                                 vs_input_semantic_index,
 272                                 NULL,
 273                                 /* outputs */
 274                                 vs_num_outputs,
 275                                 outputMapping,
 276                                 vs_output_semantic_name,
 277                                 vs_output_semantic_index,
 278                                 /* tokenized result */
 279                                 tokens, ST_MAX_SHADER_TOKENS);
 280
 281    vs.tokens = (struct tgsi_token *)
 282       mem_dup(tokens, num_tokens * sizeof(tokens[0]));
 283
 284    stvp->num_inputs = vs_num_inputs;
 285    stvp->state = vs; /* struct copy */
 286    stvp->driver_shader = pipe->create_vs_state(pipe, &vs);
 287
 288    if (0)
 289       _mesa_print_program(&stvp->Base.Base);
 290
 291    if (TGSI_DEBUG)
 292       tgsi_dump( vs.tokens, 0 );
 293 }
 294
 295
 296
 297 /**
 298  * Translate a Mesa fragment shader into a TGSI shader.
 299  * \param inputMapping  to map fragment program input registers to TGSI
 300  *                      input slots
 301  * \param tokensOut  destination for TGSI tokens
 302  * \return  pointer to cached pipe_shader object.
 303  */
 304 const struct cso_fragment_shader *
 305 st_translate_fragment_program(struct st_context *st,
 306                               struct st_fragment_program *stfp,
 307                               const GLuint inputMapping[])
 308 {
 309    struct pipe_context *pipe = st->pipe;
 310    struct tgsi_token tokens[ST_MAX_SHADER_TOKENS];
 311    GLuint outputMapping[FRAG_RESULT_MAX];
 312    GLuint defaultInputMapping[FRAG_ATTRIB_MAX];
 313    struct pipe_shader_state fs;
 314    const struct cso_fragment_shader *cso;
 315    GLuint interpMode[16];  /* XXX size? */
 316    GLuint attr;
 317    const GLbitfield inputsRead = stfp->Base.Base.InputsRead;
 318    GLuint vslot = 0;
 319    GLuint num_generic = 0;
 320    GLuint num_tokens;
 321
 322    ubyte fs_input_semantic_name[PIPE_MAX_SHADER_INPUTS];
 323    ubyte fs_input_semantic_index[PIPE_MAX_SHADER_INPUTS];
 324    uint fs_num_inputs = 0;
 325
 326    ubyte fs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
 327    ubyte fs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
 328    uint fs_num_outputs = 0;
 329
 330    memset(&fs, 0, sizeof(fs));
 331
 332    /* which vertex output goes to the first fragment input: */
 333    if (inputsRead & FRAG_BIT_WPOS)
 334       vslot = 0;
 335    else
 336       vslot = 1;
 337
 338    /*
 339     * Convert Mesa program inputs to TGSI input register semantics.
 340     */
 341    for (attr = 0; attr < FRAG_ATTRIB_MAX; attr++) {
 342       if (inputsRead & (1 << attr)) {
 343          const GLuint slot = fs_num_inputs;
 344
 345          defaultInputMapping[attr] = slot;
 346
 347          stfp->input_map[slot] = vslot++;
 348
 349          fs_num_inputs++;
 350
 351          switch (attr) {
 352          case FRAG_ATTRIB_WPOS:
 353             fs_input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
 354             fs_input_semantic_index[slot] = 0;
 355             interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
 356             break;
 357          case FRAG_ATTRIB_COL0:
 358             fs_input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 359             fs_input_semantic_index[slot] = 0;
 360             interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
 361             break;
 362          case FRAG_ATTRIB_COL1:
 363             fs_input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 364             fs_input_semantic_index[slot] = 1;
 365             interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
 366             break;
 367          case FRAG_ATTRIB_FOGC:
 368             fs_input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
 369             fs_input_semantic_index[slot] = 0;
 370             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 371             break;
 372          case FRAG_ATTRIB_TEX0:
 373          case FRAG_ATTRIB_TEX1:
 374          case FRAG_ATTRIB_TEX2:
 375          case FRAG_ATTRIB_TEX3:
 376          case FRAG_ATTRIB_TEX4:
 377          case FRAG_ATTRIB_TEX5:
 378          case FRAG_ATTRIB_TEX6:
 379          case FRAG_ATTRIB_TEX7:
 380             fs_input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 381             fs_input_semantic_index[slot] = num_generic++;
 382             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 383             break;
 384          case FRAG_ATTRIB_VAR0:
 385             /* fall-through */
 386          default:
 387             fs_input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 388             fs_input_semantic_index[slot] = num_generic++;
 389             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 390          }
 391       }
 392    }
 393
 394    /*
 395     * Semantics and mapping for outputs
 396     */
 397    {
 398       uint numColors = 0;
 399       GLbitfield outputsWritten = stfp->Base.Base.OutputsWritten;
 400
 401       /* if z is written, emit that first */
 402       if (outputsWritten & (1 << FRAG_RESULT_DEPR)) {
 403          fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION;
 404          fs_output_semantic_index[fs_num_outputs] = 0;
 405          outputMapping[FRAG_RESULT_DEPR] = fs_num_outputs;
 406          fs_num_outputs++;
 407          outputsWritten &= ~(1 << FRAG_RESULT_DEPR);
 408       }
 409
 410       /* handle remaning outputs (color) */
 411       for (attr = 0; attr < FRAG_RESULT_MAX; attr++) {
 412          if (outputsWritten & (1 << attr)) {
 413             switch (attr) {
 414             case FRAG_RESULT_DEPR:
 415                /* handled above */
 416                assert(0);
 417                break;
 418             case FRAG_RESULT_COLR:
 419                fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_COLOR;
 420                fs_output_semantic_index[fs_num_outputs] = numColors;
 421                outputMapping[attr] = fs_num_outputs;
 422                numColors++;
 423                break;
 424             default:
 425                assert(0);
 426             }
 427             fs_num_outputs++;
 428          }
 429       }
 430    }
 431
 432    if (!inputMapping)
 433       inputMapping = defaultInputMapping;
 434
 435    /* XXX: fix static allocation of tokens:
 436     */
 437    num_tokens = tgsi_translate_mesa_program( TGSI_PROCESSOR_FRAGMENT,
 438                                 &stfp->Base.Base,
 439                                 /* inputs */
 440                                 fs_num_inputs,
 441                                 inputMapping,
 442                                 fs_input_semantic_name,
 443                                 fs_input_semantic_index,
 444                                 interpMode,
 445                                 /* outputs */
 446                                 fs_num_outputs,
 447                                 outputMapping,
 448                                 fs_output_semantic_name,
 449                                 fs_output_semantic_index,
 450                                 /* tokenized result */
 451                                 tokens, ST_MAX_SHADER_TOKENS);
 452
 453    fs.tokens = (struct tgsi_token *)
 454       mem_dup(tokens, num_tokens * sizeof(tokens[0]));
 455
 456    stfp->state = fs; /* struct copy */
 457    stfp->driver_shader = pipe->create_fs_state(pipe, &fs);
 458
 459    if (0)
 460       _mesa_print_program(&stfp->Base.Base);
 461
 462    if (TGSI_DEBUG)
 463       tgsi_dump( fs.tokens, 0/*TGSI_DUMP_VERBOSE*/ );
 464
 465    return cso;
 466 }
 467