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