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 "shader/prog_parameter.h"
  35 #include "shader/prog_print.h"
  36 #include "tnl/t_vp_build.h"
  37
  38 #include "pipe/p_context.h"
  39 #include "pipe/p_defines.h"
  40 #include "pipe/p_winsys.h"
  41 #include "pipe/tgsi/mesa/mesa_to_tgsi.h"
  42 #include "pipe/tgsi/exec/tgsi_core.h"
  43
  44 #include "st_context.h"
  45 #include "st_cache.h"
  46 #include "st_atom.h"
  47 #include "st_program.h"
  48
  49
  50 #define TGSI_DEBUG 1
  51
  52
  53 /**
  54  * Translate a Mesa vertex shader into a TGSI shader.
  55  * \param outputMapping  to map vertex program output registers to TGSI
  56  *                       output slots
  57  * \param tokensOut  destination for TGSI tokens
  58  * \return  pointer to cached pipe_shader object.
  59  */
  60 const struct cso_vertex_shader *
  61 st_translate_vertex_program(struct st_context *st,
  62                             struct st_vertex_program *stvp,
  63                             const GLuint outputMapping[],
  64                             struct tgsi_token *tokensOut,
  65                             GLuint maxTokens)
  66 {
  67    GLuint defaultOutputMapping[VERT_RESULT_MAX];
  68    struct pipe_shader_state vs;
  69    const struct cso_vertex_shader *cso;
  70    GLuint attr, i;
  71
  72    memset(&vs, 0, sizeof(vs));
  73
  74    /*
  75     * Determine number of inputs, the mappings between VERT_ATTRIB_x
  76     * and TGSI generic input indexes, plus input attrib semantic info.
  77     */
  78    for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
  79       if (stvp->Base.Base.InputsRead & (1 << attr)) {
  80          const GLuint slot = vs.num_inputs;
  81
  82          vs.num_inputs++;
  83
  84          stvp->input_to_index[attr] = slot;
  85          stvp->index_to_input[slot] = attr;
  86
  87          switch (attr) {
  88          case VERT_ATTRIB_POS:
  89             vs.input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
  90             vs.input_semantic_index[slot] = 0;
  91             break;
  92          case VERT_ATTRIB_WEIGHT:
  93             /* fall-through */
  94          case VERT_ATTRIB_NORMAL:
  95             /* just label as a generic */
  96             vs.input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
  97             vs.input_semantic_index[slot] = 0;
  98             break;
  99          case VERT_ATTRIB_COLOR0:
 100             vs.input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 101             vs.input_semantic_index[slot] = 0;
 102             break;
 103          case VERT_ATTRIB_COLOR1:
 104             vs.input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 105             vs.input_semantic_index[slot] = 1;
 106             break;
 107          case VERT_ATTRIB_FOG:
 108             vs.input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
 109             vs.input_semantic_index[slot] = 0;
 110             break;
 111          case VERT_ATTRIB_TEX0:
 112          case VERT_ATTRIB_TEX1:
 113          case VERT_ATTRIB_TEX2:
 114          case VERT_ATTRIB_TEX3:
 115          case VERT_ATTRIB_TEX4:
 116          case VERT_ATTRIB_TEX5:
 117          case VERT_ATTRIB_TEX6:
 118          case VERT_ATTRIB_TEX7:
 119             vs.input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 120             vs.input_semantic_index[slot] = attr - VERT_ATTRIB_TEX0;
 121             break;
 122          case VERT_ATTRIB_GENERIC0:
 123          case VERT_ATTRIB_GENERIC1:
 124          case VERT_ATTRIB_GENERIC2:
 125          case VERT_ATTRIB_GENERIC3:
 126          case VERT_ATTRIB_GENERIC4:
 127          case VERT_ATTRIB_GENERIC5:
 128          case VERT_ATTRIB_GENERIC6:
 129          case VERT_ATTRIB_GENERIC7:
 130             assert(attr < VERT_ATTRIB_MAX);
 131             vs.input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 132             vs.input_semantic_index[slot] = attr - VERT_ATTRIB_GENERIC0;
 133             break;
 134          default:
 135             assert(0);
 136          }
 137       }
 138    }
 139
 140    /* initialize output semantics to defaults */
 141    for (i = 0; i < PIPE_MAX_SHADER_OUTPUTS; i++) {
 142       vs.output_semantic_name[i] = TGSI_SEMANTIC_GENERIC;
 143       vs.output_semantic_index[i] = 0;
 144    }
 145
 146    /*
 147     * Determine number of outputs, the (default) output register
 148     * mapping and the semantic information for each output.
 149     */
 150    for (attr = 0; attr < VERT_RESULT_MAX; attr++) {
 151       if (stvp->Base.Base.OutputsWritten & (1 << attr)) {
 152          GLuint slot;
 153
 154          if (outputMapping) {
 155             slot = outputMapping[attr];
 156             assert(slot != ~0);
 157          }
 158          else {
 159             slot = vs.num_outputs;
 160             vs.num_outputs++;
 161             defaultOutputMapping[attr] = slot;
 162          }
 163
 164          /*
 165          printf("Output %u -> slot %u\n", attr, slot);
 166          */
 167
 168          switch (attr) {
 169          case VERT_RESULT_HPOS:
 170             vs.output_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
 171             vs.output_semantic_index[slot] = 0;
 172             break;
 173          case VERT_RESULT_COL0:
 174             vs.output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 175             vs.output_semantic_index[slot] = 0;
 176             break;
 177          case VERT_RESULT_COL1:
 178             vs.output_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 179             vs.output_semantic_index[slot] = 1;
 180             break;
 181          case VERT_RESULT_BFC0:
 182             vs.output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
 183             vs.output_semantic_index[slot] = 0;
 184             break;
 185          case VERT_RESULT_BFC1:
 186             vs.output_semantic_name[slot] = TGSI_SEMANTIC_BCOLOR;
 187             vs.output_semantic_index[slot] = 1;
 188             break;
 189          case VERT_RESULT_FOGC:
 190             vs.output_semantic_name[slot] = TGSI_SEMANTIC_FOG;
 191             vs.output_semantic_index[slot] = 0;
 192             break;
 193          case VERT_RESULT_PSIZ:
 194             vs.output_semantic_name[slot] = TGSI_SEMANTIC_PSIZE;
 195             vs.output_semantic_index[slot] = 0;
 196             break;
 197          case VERT_RESULT_EDGE:
 198             assert(0);
 199             break;
 200          case VERT_RESULT_TEX0:
 201          case VERT_RESULT_TEX1:
 202          case VERT_RESULT_TEX2:
 203          case VERT_RESULT_TEX3:
 204          case VERT_RESULT_TEX4:
 205          case VERT_RESULT_TEX5:
 206          case VERT_RESULT_TEX6:
 207          case VERT_RESULT_TEX7:
 208             vs.output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 209             vs.output_semantic_index[slot] = attr - VERT_RESULT_TEX0;
 210             break;
 211          case VERT_RESULT_VAR0:
 212             /* fall-through */
 213          default:
 214             assert(attr - VERT_RESULT_VAR0 < MAX_VARYING);
 215             vs.output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 216             vs.output_semantic_index[slot] = attr - VERT_RESULT_VAR0;
 217          }
 218       }
 219    }
 220
 221
 222    if (outputMapping) {
 223       /* find max output slot referenced to compute vs.num_outputs */
 224       GLuint maxSlot = 0;
 225       for (attr = 0; attr < VERT_RESULT_MAX; attr++) {
 226          if (outputMapping[attr] != ~0 && outputMapping[attr] > maxSlot)
 227             maxSlot = outputMapping[attr];
 228       }
 229       vs.num_outputs = maxSlot + 1;
 230    }
 231    else {
 232       outputMapping = defaultOutputMapping;
 233    }
 234
 235    /* XXX: fix static allocation of tokens:
 236     */
 237    tgsi_mesa_compile_vp_program( &stvp->Base,
 238                                  /* inputs */
 239                                  vs.num_inputs,
 240                                  stvp->input_to_index,
 241                                  vs.input_semantic_name,
 242                                  vs.input_semantic_index,
 243                                  /* outputs */
 244                                  vs.num_outputs,
 245                                  outputMapping,
 246                                  vs.output_semantic_name,
 247                                  vs.output_semantic_index,
 248                                  /* tokenized result */
 249                                  tokensOut, maxTokens);
 250
 251    vs.tokens = tokensOut;
 252    cso = st_cached_vs_state(st, &vs);
 253    stvp->vs = cso;
 254
 255    if (TGSI_DEBUG)
 256       tgsi_dump( tokensOut, 0 );
 257
 258 #if defined(USE_X86_ASM) || defined(SLANG_X86)
 259    if (stvp->sse2_program.csr == stvp->sse2_program.store)
 260       tgsi_emit_sse2( tokensOut, &stvp->sse2_program );
 261
 262    if (!cso->state.executable)
 263       ((struct cso_vertex_shader*)cso)->state.executable = (void *) x86_get_func( &stvp->sse2_program );
 264 #endif
 265
 266    return cso;
 267 }
 268
 269
 270
 271 /**
 272  * Translate a Mesa fragment shader into a TGSI shader.
 273  * \param inputMapping  to map fragment program input registers to TGSI
 274  *                      input slots
 275  * \param tokensOut  destination for TGSI tokens
 276  * \return  pointer to cached pipe_shader object.
 277  */
 278 const struct cso_fragment_shader *
 279 st_translate_fragment_program(struct st_context *st,
 280                               struct st_fragment_program *stfp,
 281                               const GLuint inputMapping[],
 282                               struct tgsi_token *tokensOut,
 283                               GLuint maxTokens)
 284 {
 285    GLuint outputMapping[FRAG_RESULT_MAX];
 286    GLuint defaultInputMapping[FRAG_ATTRIB_MAX];
 287    struct pipe_shader_state fs;
 288    const struct cso_fragment_shader *cso;
 289    GLuint interpMode[16];  /* XXX size? */
 290    GLuint attr;
 291    GLbitfield inputsRead = stfp->Base.Base.InputsRead;
 292
 293    /* Check if all fragment programs need the fragment position (in order
 294     * to do perspective-corrected interpolation).
 295     */
 296    /* XXX temporary! */
 297    if (st->pipe->get_param(st->pipe, PIPE_PARAM_FS_NEEDS_POS))
 298       inputsRead |= FRAG_BIT_WPOS;
 299
 300    memset(&fs, 0, sizeof(fs));
 301
 302    /*
 303     * Convert Mesa program inputs to TGSI input register semantics.
 304     */
 305    for (attr = 0; attr < FRAG_ATTRIB_MAX; attr++) {
 306       if (inputsRead & (1 << attr)) {
 307          const GLuint slot = fs.num_inputs;
 308
 309          fs.num_inputs++;
 310
 311          defaultInputMapping[attr] = slot;
 312
 313          switch (attr) {
 314          case FRAG_ATTRIB_WPOS:
 315             fs.input_semantic_name[slot] = TGSI_SEMANTIC_POSITION;
 316             fs.input_semantic_index[slot] = 0;
 317             interpMode[slot] = TGSI_INTERPOLATE_CONSTANT;
 318             break;
 319          case FRAG_ATTRIB_COL0:
 320             fs.input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 321             fs.input_semantic_index[slot] = 0;
 322             interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
 323             break;
 324          case FRAG_ATTRIB_COL1:
 325             fs.input_semantic_name[slot] = TGSI_SEMANTIC_COLOR;
 326             fs.input_semantic_index[slot] = 1;
 327             interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
 328             break;
 329          case FRAG_ATTRIB_FOGC:
 330             fs.input_semantic_name[slot] = TGSI_SEMANTIC_FOG;
 331             fs.input_semantic_index[slot] = 0;
 332             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 333             break;
 334          case FRAG_ATTRIB_TEX0:
 335          case FRAG_ATTRIB_TEX1:
 336          case FRAG_ATTRIB_TEX2:
 337          case FRAG_ATTRIB_TEX3:
 338          case FRAG_ATTRIB_TEX4:
 339          case FRAG_ATTRIB_TEX5:
 340          case FRAG_ATTRIB_TEX6:
 341          case FRAG_ATTRIB_TEX7:
 342             fs.input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 343             fs.input_semantic_index[slot] = attr - FRAG_ATTRIB_TEX0;
 344             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 345             break;
 346          case FRAG_ATTRIB_VAR0:
 347             /* fall-through */
 348          default:
 349             fs.input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
 350             fs.input_semantic_index[slot] = attr - FRAG_ATTRIB_VAR0;
 351             interpMode[slot] = TGSI_INTERPOLATE_PERSPECTIVE;
 352          }
 353       }
 354    }
 355
 356    /*
 357     * Semantics for outputs
 358     */
 359    for (attr = 0; attr < FRAG_RESULT_MAX; attr++) {
 360       if (stfp->Base.Base.OutputsWritten & (1 << attr)) {
 361          switch (attr) {
 362          case FRAG_RESULT_DEPR:
 363             fs.output_semantic_name[fs.num_outputs] = TGSI_SEMANTIC_POSITION;
 364             outputMapping[attr] = fs.num_outputs;
 365             break;
 366          case FRAG_RESULT_COLR:
 367             fs.output_semantic_name[fs.num_outputs] = TGSI_SEMANTIC_COLOR;
 368             outputMapping[attr] = fs.num_outputs;
 369             break;
 370          default:
 371             assert(0);
 372          }
 373          fs.num_outputs++;
 374       }
 375    }
 376
 377    if (!inputMapping)
 378       inputMapping = defaultInputMapping;
 379
 380    /* XXX: fix static allocation of tokens:
 381     */
 382    tgsi_mesa_compile_fp_program( &stfp->Base,
 383                                  /* inputs */
 384                                  fs.num_inputs,
 385                                  inputMapping,
 386                                  fs.input_semantic_name,
 387                                  fs.input_semantic_index,
 388                                  interpMode,
 389                                  /* outputs */
 390                                  outputMapping,
 391                                  /* tokenized result */
 392                                  tokensOut, maxTokens);
 393
 394
 395    fs.tokens = tokensOut;
 396
 397    cso = st_cached_fs_state(st, &fs);
 398    stfp->fs = cso;
 399
 400    if (TGSI_DEBUG)
 401       tgsi_dump( tokensOut, 0/*TGSI_DUMP_VERBOSE*/ );
 402
 403 #if defined(USE_X86_ASM) || defined(SLANG_X86)
 404    if (stfp->sse2_program.csr == stfp->sse2_program.store)
 405       tgsi_emit_sse2_fs( tokensOut, &stfp->sse2_program );
 406
 407    if (!cso->state.executable)
 408       ((struct cso_fragment_shader*)cso)->state.executable = (void *) x86_get_func( &stfp->sse2_program );
 409 #endif
 410
 411    return cso;
 412 }
 413