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 #include "shader/programopt.h"
  38
  39 #include "pipe/p_context.h"
  40 #include "pipe/p_defines.h"
  41 #include "pipe/p_shader_tokens.h"
  42 #include "draw/draw_context.h"
  43 #include "tgsi/tgsi_dump.h"
  44
  45 #include "st_debug.h"
  46 #include "st_context.h"
  47 #include "st_atom.h"
  48 #include "st_program.h"
  49 #include "st_mesa_to_tgsi.h"
  50 #include "cso_cache/cso_context.h"
  51
  52
  53 /**
  54  * Translate a Mesa vertex shader into a TGSI shader.
  55  * \param outputMapping  to map vertex program output registers (VERT_RESULT_x)
  56  *       to TGSI output slots
  57  * \param tokensOut  destination for TGSI tokens
  58  * \return  pointer to cached pipe_shader object.
  59  */
  60 void
  61 st_translate_vertex_program(struct st_context *st,
  62                             struct st_vertex_program *stvp,
  63                             const GLuint outputMapping[],
  64                             const ubyte *outputSemanticName,
  65                             const ubyte *outputSemanticIndex)
  66 {
  67    struct pipe_context *pipe = st->pipe;
  68    GLuint defaultOutputMapping[VERT_RESULT_MAX];
  69    GLuint attr, i;
  70    GLuint num_generic = 0;
  71
  72    ubyte vs_input_semantic_name[PIPE_MAX_SHADER_INPUTS];
  73    ubyte vs_input_semantic_index[PIPE_MAX_SHADER_INPUTS];
  74    uint vs_num_inputs = 0;
  75
  76    ubyte vs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
  77    ubyte vs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
  78    uint vs_num_outputs = 0;
  79
  80    GLbitfield input_flags[MAX_PROGRAM_INPUTS];
  81    GLbitfield output_flags[MAX_PROGRAM_OUTPUTS];
  82
  83 //   memset(&vs, 0, sizeof(vs));
  84    memset(input_flags, 0, sizeof(input_flags));
  85    memset(output_flags, 0, sizeof(output_flags));
  86
  87    if (stvp->Base.IsPositionInvariant)
  88       _mesa_insert_mvp_code(st->ctx, &stvp->Base);
  89
  90    /*
  91     * Determine number of inputs, the mappings between VERT_ATTRIB_x
  92     * and TGSI generic input indexes, plus input attrib semantic info.
  93     */
  94    for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
  95       if (stvp->Base.Base.InputsRead & (1 << attr)) {
  96          const GLuint slot = vs_num_inputs;
  97
  98          vs_num_inputs++;
  99
 100          stvp->input_to_index[attr] = slot;
 101          stvp->index_to_input[slot] = attr;
 102
 103          switch (attr) {
 104          case VERT_ATTRIB_POS:
 105             vs_input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
 106             vs_input_semantic_index[slot] = 0;
 107             break;
 108          case VERT_ATTRIB_WEIGHT:
 109             /* fall-through */
 110          case VERT_ATTRIB_NORMAL:
 111             /* just label as a generic */
 112             vs_input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 113             vs_input_semantic_index[slot] = 0;
 114             break;
 115          case VERT_ATTRIB_COLOR0:
 116             vs_input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 117             vs_input_semantic_index[slot] = 0;
 118             break;
 119          case VERT_ATTRIB_COLOR1:
 120             vs_input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 121             vs_input_semantic_index[slot] = 1;
 122             break;
 123          case VERT_ATTRIB_FOG:
 124             vs_input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
 125             vs_input_semantic_index[slot] = 0;
 126             break;
 127          case VERT_ATTRIB_POINT_SIZE:
 128             vs_input_semantic_name[slot] = TGSI_SEMANTIC_PSIZE;
 129             vs_input_semantic_index[slot] = 0;
 130             break;
 131          case VERT_ATTRIB_TEX0:
 132          case VERT_ATTRIB_TEX1:
 133          case VERT_ATTRIB_TEX2:
 134          case VERT_ATTRIB_TEX3:
 135          case VERT_ATTRIB_TEX4:
 136          case VERT_ATTRIB_TEX5:
 137          case VERT_ATTRIB_TEX6:
 138          case VERT_ATTRIB_TEX7:
 139             assert(slot < Elements(vs_input_semantic_name));
 140             vs_input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 141             vs_input_semantic_index[slot] = num_generic++;
 142             break;
 143          case VERT_ATTRIB_GENERIC0:
 144          case VERT_ATTRIB_GENERIC1:
 145          case VERT_ATTRIB_GENERIC2:
 146          case VERT_ATTRIB_GENERIC3:
 147          case VERT_ATTRIB_GENERIC4:
 148          case VERT_ATTRIB_GENERIC5:
 149          case VERT_ATTRIB_GENERIC6:
 150          case VERT_ATTRIB_GENERIC7:
 151          case VERT_ATTRIB_GENERIC8:
 152          case VERT_ATTRIB_GENERIC9:
 153          case VERT_ATTRIB_GENERIC10:
 154          case VERT_ATTRIB_GENERIC11:
 155          case VERT_ATTRIB_GENERIC12:
 156          case VERT_ATTRIB_GENERIC13:
 157          case VERT_ATTRIB_GENERIC14:
 158          case VERT_ATTRIB_GENERIC15:
 159             assert(attr < VERT_ATTRIB_MAX);
 160             assert(slot < Elements(vs_input_semantic_name));
 161             vs_input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 162             vs_input_semantic_index[slot] = num_generic++;
 163             break;
 164          default:
 165             assert(0);
 166          }
 167
 168          input_flags[slot] = stvp->Base.Base.InputFlags[attr];
 169       }
 170    }
 171
 172 #if 0
 173    if (outputMapping && outputSemanticName) {
 174       printf("VERT_RESULT  written  out_slot  semantic_name  semantic_index\n");
 175       for (attr = 0; attr < VERT_RESULT_MAX; attr++) {
 176          printf("    %-2d          %c       %3d          %2d              %2d\n",
 177                 attr,
 178                 ((stvp->Base.Base.OutputsWritten & (1 << attr)) ? 'Y' : ' '),
 179                 outputMapping[attr],
 180                 outputSemanticName[attr],
 181                 outputSemanticIndex[attr]);
 182       }
 183    }
 184 #endif
 185
 186    /* initialize output semantics to defaults */
 187    for (i = 0; i < PIPE_MAX_SHADER_OUTPUTS; i++) {
 188       assert(i < Elements(vs_output_semantic_name));
 189       vs_output_semantic_name[i] = TGSI_SEMANTIC_GENERIC;
 190       vs_output_semantic_index[i] = 0;
 191       output_flags[i] = 0x0;
 192    }
 193
 194    num_generic = 0;
 195    /*
 196     * Determine number of outputs, the (default) output register
 197     * mapping and the semantic information for each output.
 198     */
 199    for (attr = 0; attr < VERT_RESULT_MAX; attr++) {
 200       if (stvp->Base.Base.OutputsWritten & (1 << attr)) {
 201          GLuint slot;
 202
 203          /* XXX
 204           * Pass in the fragment program's input's semantic info.
 205           * Use the generic semantic indexes from there, instead of
 206           * guessing below.
 207           */
 208
 209          if (outputMapping) {
 210             slot = outputMapping[attr];
 211             assert(slot != ~0);
 212          }
 213          else {
 214             slot = vs_num_outputs;
 215             vs_num_outputs++;
 216             defaultOutputMapping[attr] = slot;
 217          }
 218
 219          switch (attr) {
 220          case VERT_RESULT_HPOS:
 221             assert(slot == 0);
 222             vs_output_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
 223             vs_output_semantic_index[slot] = 0;
 224             break;
 225          case VERT_RESULT_COL0:
 226             vs_output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 227             vs_output_semantic_index[slot] = 0;
 228             break;
 229          case VERT_RESULT_COL1:
 230             vs_output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 231             vs_output_semantic_index[slot] = 1;
 232             break;
 233          case VERT_RESULT_BFC0:
 234             vs_output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
 235             vs_output_semantic_index[slot] = 0;
 236             break;
 237          case VERT_RESULT_BFC1:
 238             vs_output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
 239             vs_output_semantic_index[slot] = 1;
 240             break;
 241          case VERT_RESULT_FOGC:
 242             vs_output_semantic_name[slot] = TGSI_SEMANTIC_FOG;
 243             vs_output_semantic_index[slot] = 0;
 244             break;
 245          case VERT_RESULT_PSIZ:
 246             vs_output_semantic_name[slot] = TGSI_SEMANTIC_PSIZE;
 247             vs_output_semantic_index[slot] = 0;
 248             break;
 249          case VERT_RESULT_EDGE:
 250             assert(0);
 251             break;
 252          case VERT_RESULT_TEX0:
 253          case VERT_RESULT_TEX1:
 254          case VERT_RESULT_TEX2:
 255          case VERT_RESULT_TEX3:
 256          case VERT_RESULT_TEX4:
 257          case VERT_RESULT_TEX5:
 258          case VERT_RESULT_TEX6:
 259          case VERT_RESULT_TEX7:
 260             /* fall-through */
 261          case VERT_RESULT_VAR0:
 262             /* fall-through */
 263          default:
 264             assert(slot < Elements(vs_output_semantic_name));
 265             if (outputSemanticName) {
 266                /* use provided semantic into */
 267                assert(outputSemanticName[attr] != TGSI_SEMANTIC_COUNT);
 268                vs_output_semantic_name[slot] = outputSemanticName[attr];
 269                vs_output_semantic_index[slot] = outputSemanticIndex[attr];
 270             }
 271             else {
 272                /* use default semantic info */
 273                vs_output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 274                vs_output_semantic_index[slot] = num_generic++;
 275             }
 276          }
 277
 278          assert(slot < Elements(output_flags));
 279          output_flags[slot] = stvp->Base.Base.OutputFlags[attr];
 280       }
 281    }
 282
 283    if (outputMapping) {
 284       /* find max output slot referenced to compute vs_num_outputs */
 285       GLuint maxSlot = 0;
 286       for (attr = 0; attr < VERT_RESULT_MAX; attr++) {
 287          if (outputMapping[attr] != ~0 && outputMapping[attr] > maxSlot)
 288             maxSlot = outputMapping[attr];
 289       }
 290       vs_num_outputs = maxSlot + 1;
 291    }
 292    else {
 293       outputMapping = defaultOutputMapping;
 294    }
 295
 296 #if 0 /* debug */
 297    {
 298       GLuint i;
 299       printf("outputMapping? %d\n", outputMapping ? 1 : 0);
 300       if (outputMapping) {
 301          printf("attr -> slot\n");
 302          for (i = 0; i < 16;  i++) {
 303             printf(" %2d       %3d\n", i, outputMapping[i]);
 304          }
 305       }
 306       printf("slot    sem_name  sem_index\n");
 307       for (i = 0; i < vs_num_outputs; i++) {
 308          printf(" %2d         %d         %d\n",
 309                 i,
 310                 vs_output_semantic_name[i],
 311                 vs_output_semantic_index[i]);
 312       }
 313    }
 314 #endif
 315
 316    /* free old shader state, if any */
 317    if (stvp->state.tokens) {
 318       st_free_tokens(stvp->state.tokens);
 319       stvp->state.tokens = NULL;
 320    }
 321    if (stvp->driver_shader) {
 322       cso_delete_vertex_shader(st->cso_context, stvp->driver_shader);
 323       stvp->driver_shader = NULL;
 324    }
 325
 326    stvp->state.tokens =
 327       st_translate_mesa_program(st->ctx,
 328                                 TGSI_PROCESSOR_VERTEX,
 329                                 &stvp->Base.Base,
 330                                 /* inputs */
 331                                 vs_num_inputs,
 332                                 stvp->input_to_index,
 333                                 vs_input_semantic_name,
 334                                 vs_input_semantic_index,
 335                                 NULL,
 336                                 input_flags,
 337                                 /* outputs */
 338                                 vs_num_outputs,
 339                                 outputMapping,
 340                                 vs_output_semantic_name,
 341                                 vs_output_semantic_index,
 342                                 output_flags );
 343
 344    stvp->num_inputs = vs_num_inputs;
 345    stvp->driver_shader = pipe->create_vs_state(pipe, &stvp->state);
 346
 347    if ((ST_DEBUG & DEBUG_TGSI) && (ST_DEBUG & DEBUG_MESA)) {
 348       _mesa_print_program(&stvp->Base.Base);
 349       debug_printf("\n");
 350    }
 351
 352    if (ST_DEBUG & DEBUG_TGSI) {
 353       tgsi_dump( stvp->state.tokens, 0 );
 354       debug_printf("\n");
 355    }
 356 }
 357
 358
 359
 360 /**
 361  * Translate a Mesa fragment shader into a TGSI shader.
 362  * \param inputMapping  to map fragment program input registers to TGSI
 363  *                      input slots
 364  * \return  pointer to cached pipe_shader object.
 365  */
 366 void
 367 st_translate_fragment_program(struct st_context *st,
 368                               struct st_fragment_program *stfp,
 369                               const GLuint inputMapping[])
 370 {
 371    struct pipe_context *pipe = st->pipe;
 372    GLuint outputMapping[FRAG_RESULT_MAX];
 373    GLuint defaultInputMapping[FRAG_ATTRIB_MAX];
 374    GLuint interpMode[16];  /* XXX size? */
 375    GLuint attr;
 376    const GLbitfield inputsRead = stfp->Base.Base.InputsRead;
 377    GLuint vslot = 0;
 378    GLuint num_generic = 0;
 379
 380    uint fs_num_inputs = 0;
 381
 382    ubyte fs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
 383    ubyte fs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
 384    uint fs_num_outputs = 0;
 385
 386    GLbitfield input_flags[MAX_PROGRAM_INPUTS];
 387    GLbitfield output_flags[MAX_PROGRAM_OUTPUTS];
 388
 389 //   memset(&fs, 0, sizeof(fs));
 390    memset(input_flags, 0, sizeof(input_flags));
 391    memset(output_flags, 0, sizeof(output_flags));
 392
 393    /* which vertex output goes to the first fragment input: */
 394    if (inputsRead & FRAG_BIT_WPOS)
 395       vslot = 0;
 396    else
 397       vslot = 1;
 398
 399    /*
 400     * Convert Mesa program inputs to TGSI input register semantics.
 401     */
 402    for (attr = 0; attr < FRAG_ATTRIB_MAX; attr++) {
 403       if (inputsRead & (1 << attr)) {
 404          const GLuint slot = fs_num_inputs;
 405
 406          defaultInputMapping[attr] = slot;
 407
 408          stfp->input_map[slot] = vslot++;
 409
 410          fs_num_inputs++;
 411
 412          switch (attr) {
 413          case FRAG_ATTRIB_WPOS:
 414             stfp->input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
 415             stfp->input_semantic_index[slot] = 0;
 416             interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
 417             break;
 418          case FRAG_ATTRIB_COL0:
 419             stfp->input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 420             stfp->input_semantic_index[slot] = 0;
 421             interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
 422             break;
 423          case FRAG_ATTRIB_COL1:
 424             stfp->input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 425             stfp->input_semantic_index[slot] = 1;
 426             interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
 427             break;
 428          case FRAG_ATTRIB_FOGC:
 429             stfp->input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
 430             stfp->input_semantic_index[slot] = 0;
 431             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 432             break;
 433          case FRAG_ATTRIB_FACE:
 434             stfp->input_semantic_name[slot] = TGSI_SEMANTIC_FACE;
 435             stfp->input_semantic_index[slot] = num_generic++;
 436             interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
 437             break;
 438          case FRAG_ATTRIB_PNTC:
 439             stfp->input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 440             stfp->input_semantic_index[slot] = num_generic++;
 441             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 442             break;
 443          case FRAG_ATTRIB_TEX0:
 444          case FRAG_ATTRIB_TEX1:
 445          case FRAG_ATTRIB_TEX2:
 446          case FRAG_ATTRIB_TEX3:
 447          case FRAG_ATTRIB_TEX4:
 448          case FRAG_ATTRIB_TEX5:
 449          case FRAG_ATTRIB_TEX6:
 450          case FRAG_ATTRIB_TEX7:
 451             stfp->input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 452             stfp->input_semantic_index[slot] = num_generic++;
 453             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 454             break;
 455          case FRAG_ATTRIB_VAR0:
 456             /* fall-through */
 457          default:
 458             stfp->input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 459             stfp->input_semantic_index[slot] = num_generic++;
 460             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 461          }
 462
 463          input_flags[slot] = stfp->Base.Base.InputFlags[attr];
 464       }
 465    }
 466
 467    /*
 468     * Semantics and mapping for outputs
 469     */
 470    {
 471       uint numColors = 0;
 472       GLbitfield outputsWritten = stfp->Base.Base.OutputsWritten;
 473
 474       /* if z is written, emit that first */
 475       if (outputsWritten & (1 << FRAG_RESULT_DEPTH)) {
 476          fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION;
 477          fs_output_semantic_index[fs_num_outputs] = 0;
 478          outputMapping[FRAG_RESULT_DEPTH] = fs_num_outputs;
 479          fs_num_outputs++;
 480          outputsWritten &= ~(1 << FRAG_RESULT_DEPTH);
 481       }
 482
 483       /* handle remaning outputs (color) */
 484       for (attr = 0; attr < FRAG_RESULT_MAX; attr++) {
 485          if (outputsWritten & (1 << attr)) {
 486             switch (attr) {
 487             case FRAG_RESULT_DEPTH:
 488                /* handled above */
 489                assert(0);
 490                break;
 491             default:
 492                assert(attr == FRAG_RESULT_COLOR ||
 493                       (FRAG_RESULT_DATA0 <= attr && attr < FRAG_RESULT_MAX));
 494                fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_COLOR;
 495                fs_output_semantic_index[fs_num_outputs] = numColors;
 496                outputMapping[attr] = fs_num_outputs;
 497                numColors++;
 498                break;
 499             }
 500
 501             output_flags[fs_num_outputs] = stfp->Base.Base.OutputFlags[attr];
 502
 503             fs_num_outputs++;
 504          }
 505       }
 506    }
 507
 508    if (!inputMapping)
 509       inputMapping = defaultInputMapping;
 510
 511    stfp->state.tokens =
 512       st_translate_mesa_program(st->ctx,
 513                                 TGSI_PROCESSOR_FRAGMENT,
 514                                 &stfp->Base.Base,
 515                                 /* inputs */
 516                                 fs_num_inputs,
 517                                 inputMapping,
 518                                 stfp->input_semantic_name,
 519                                 stfp->input_semantic_index,
 520                                 interpMode,
 521                                 input_flags,
 522                                 /* outputs */
 523                                 fs_num_outputs,
 524                                 outputMapping,
 525                                 fs_output_semantic_name,
 526                                 fs_output_semantic_index,
 527                                 output_flags );
 528
 529    stfp->driver_shader = pipe->create_fs_state(pipe, &stfp->state);
 530
 531    if ((ST_DEBUG & DEBUG_TGSI) && (ST_DEBUG & DEBUG_MESA)) {
 532       _mesa_print_program(&stfp->Base.Base);
 533       debug_printf("\n");
 534    }
 535
 536    if (ST_DEBUG & DEBUG_TGSI) {
 537       tgsi_dump( stfp->state.tokens, 0/*TGSI_DUMP_VERBOSE*/ );
 538       debug_printf("\n");
 539    }
 540 }
 541
 542
 543 /**
 544  * Debug- print current shader text
 545  */
 546 void
 547 st_print_shaders(GLcontext *ctx)
 548 {
 549    struct gl_shader_program *shProg = ctx->Shader.CurrentProgram;
 550    if (shProg) {
 551       GLuint i;
 552       for (i = 0; i < shProg->NumShaders; i++) {
 553          printf("GLSL shader %u of %u:\n", i, shProg->NumShaders);
 554          printf("%s\n", shProg->Shaders[i]->Source);
 555       }
 556    }
 557 }