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    /* free old shader state, if any */
 264    if (stvp->state.tokens) {
 265       FREE((void *) stvp->state.tokens);
 266       stvp->state.tokens = NULL;
 267    }
 268    if (stvp->driver_shader) {
 269       pipe->delete_vs_state(pipe, stvp->driver_shader);
 270       stvp->driver_shader = NULL;
 271    }
 272
 273    /* XXX: fix static allocation of tokens:
 274     */
 275    num_tokens = tgsi_translate_mesa_program( TGSI_PROCESSOR_VERTEX,
 276                                 &stvp->Base.Base,
 277                                 /* inputs */
 278                                 vs_num_inputs,
 279                                 stvp->input_to_index,
 280                                 vs_input_semantic_name,
 281                                 vs_input_semantic_index,
 282                                 NULL,
 283                                 /* outputs */
 284                                 vs_num_outputs,
 285                                 outputMapping,
 286                                 vs_output_semantic_name,
 287                                 vs_output_semantic_index,
 288                                 /* tokenized result */
 289                                 tokens, ST_MAX_SHADER_TOKENS);
 290
 291    vs.tokens = (struct tgsi_token *)
 292       mem_dup(tokens, num_tokens * sizeof(tokens[0]));
 293
 294    stvp->num_inputs = vs_num_inputs;
 295    stvp->state = vs; /* struct copy */
 296    stvp->driver_shader = pipe->create_vs_state(pipe, &vs);
 297
 298    if (0)
 299       _mesa_print_program(&stvp->Base.Base);
 300
 301    if (TGSI_DEBUG)
 302       tgsi_dump( vs.tokens, 0 );
 303 }
 304
 305
 306
 307 /**
 308  * Translate a Mesa fragment shader into a TGSI shader.
 309  * \param inputMapping  to map fragment program input registers to TGSI
 310  *                      input slots
 311  * \param tokensOut  destination for TGSI tokens
 312  * \return  pointer to cached pipe_shader object.
 313  */
 314 const struct cso_fragment_shader *
 315 st_translate_fragment_program(struct st_context *st,
 316                               struct st_fragment_program *stfp,
 317                               const GLuint inputMapping[])
 318 {
 319    struct pipe_context *pipe = st->pipe;
 320    struct tgsi_token tokens[ST_MAX_SHADER_TOKENS];
 321    GLuint outputMapping[FRAG_RESULT_MAX];
 322    GLuint defaultInputMapping[FRAG_ATTRIB_MAX];
 323    struct pipe_shader_state fs;
 324    const struct cso_fragment_shader *cso;
 325    GLuint interpMode[16];  /* XXX size? */
 326    GLuint attr;
 327    const GLbitfield inputsRead = stfp->Base.Base.InputsRead;
 328    GLuint vslot = 0;
 329    GLuint num_generic = 0;
 330    GLuint num_tokens;
 331
 332    ubyte fs_input_semantic_name[PIPE_MAX_SHADER_INPUTS];
 333    ubyte fs_input_semantic_index[PIPE_MAX_SHADER_INPUTS];
 334    uint fs_num_inputs = 0;
 335
 336    ubyte fs_output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
 337    ubyte fs_output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
 338    uint fs_num_outputs = 0;
 339
 340    memset(&fs, 0, sizeof(fs));
 341
 342    /* which vertex output goes to the first fragment input: */
 343    if (inputsRead & FRAG_BIT_WPOS)
 344       vslot = 0;
 345    else
 346       vslot = 1;
 347
 348    /*
 349     * Convert Mesa program inputs to TGSI input register semantics.
 350     */
 351    for (attr = 0; attr < FRAG_ATTRIB_MAX; attr++) {
 352       if (inputsRead & (1 << attr)) {
 353          const GLuint slot = fs_num_inputs;
 354
 355          defaultInputMapping[attr] = slot;
 356
 357          stfp->input_map[slot] = vslot++;
 358
 359          fs_num_inputs++;
 360
 361          switch (attr) {
 362          case FRAG_ATTRIB_WPOS:
 363             fs_input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
 364             fs_input_semantic_index[slot] = 0;
 365             interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
 366             break;
 367          case FRAG_ATTRIB_COL0:
 368             fs_input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 369             fs_input_semantic_index[slot] = 0;
 370             interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
 371             break;
 372          case FRAG_ATTRIB_COL1:
 373             fs_input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 374             fs_input_semantic_index[slot] = 1;
 375             interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
 376             break;
 377          case FRAG_ATTRIB_FOGC:
 378             fs_input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
 379             fs_input_semantic_index[slot] = 0;
 380             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 381             break;
 382          case FRAG_ATTRIB_TEX0:
 383          case FRAG_ATTRIB_TEX1:
 384          case FRAG_ATTRIB_TEX2:
 385          case FRAG_ATTRIB_TEX3:
 386          case FRAG_ATTRIB_TEX4:
 387          case FRAG_ATTRIB_TEX5:
 388          case FRAG_ATTRIB_TEX6:
 389          case FRAG_ATTRIB_TEX7:
 390             fs_input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 391             fs_input_semantic_index[slot] = num_generic++;
 392             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 393             break;
 394          case FRAG_ATTRIB_VAR0:
 395             /* fall-through */
 396          default:
 397             fs_input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 398             fs_input_semantic_index[slot] = num_generic++;
 399             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 400          }
 401       }
 402    }
 403
 404    /*
 405     * Semantics and mapping for outputs
 406     */
 407    {
 408       uint numColors = 0;
 409       GLbitfield outputsWritten = stfp->Base.Base.OutputsWritten;
 410
 411       /* if z is written, emit that first */
 412       if (outputsWritten & (1 << FRAG_RESULT_DEPR)) {
 413          fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_POSITION;
 414          fs_output_semantic_index[fs_num_outputs] = 0;
 415          outputMapping[FRAG_RESULT_DEPR] = fs_num_outputs;
 416          fs_num_outputs++;
 417          outputsWritten &= ~(1 << FRAG_RESULT_DEPR);
 418       }
 419
 420       /* handle remaning outputs (color) */
 421       for (attr = 0; attr < FRAG_RESULT_MAX; attr++) {
 422          if (outputsWritten & (1 << attr)) {
 423             switch (attr) {
 424             case FRAG_RESULT_DEPR:
 425                /* handled above */
 426                assert(0);
 427                break;
 428             case FRAG_RESULT_COLR:
 429                fs_output_semantic_name[fs_num_outputs] = TGSI_SEMANTIC_COLOR;
 430                fs_output_semantic_index[fs_num_outputs] = numColors;
 431                outputMapping[attr] = fs_num_outputs;
 432                numColors++;
 433                break;
 434             default:
 435                assert(0);
 436             }
 437             fs_num_outputs++;
 438          }
 439       }
 440    }
 441
 442    if (!inputMapping)
 443       inputMapping = defaultInputMapping;
 444
 445    /* XXX: fix static allocation of tokens:
 446     */
 447    num_tokens = tgsi_translate_mesa_program( TGSI_PROCESSOR_FRAGMENT,
 448                                 &stfp->Base.Base,
 449                                 /* inputs */
 450                                 fs_num_inputs,
 451                                 inputMapping,
 452                                 fs_input_semantic_name,
 453                                 fs_input_semantic_index,
 454                                 interpMode,
 455                                 /* outputs */
 456                                 fs_num_outputs,
 457                                 outputMapping,
 458                                 fs_output_semantic_name,
 459                                 fs_output_semantic_index,
 460                                 /* tokenized result */
 461                                 tokens, ST_MAX_SHADER_TOKENS);
 462
 463    fs.tokens = (struct tgsi_token *)
 464       mem_dup(tokens, num_tokens * sizeof(tokens[0]));
 465
 466    stfp->state = fs; /* struct copy */
 467    stfp->driver_shader = pipe->create_fs_state(pipe, &fs);
 468
 469    if (0)
 470       _mesa_print_program(&stfp->Base.Base);
 471
 472    if (TGSI_DEBUG)
 473       tgsi_dump( fs.tokens, 0/*TGSI_DUMP_VERBOSE*/ );
 474
 475    return cso;
 476 }
 477