src/gallium/auxiliary/tgsi/tgsi_scan.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2008 VMware, Inc.
   4  * All Rights Reserved.
   5  * Copyright 2008 VMware, Inc.  All rights Reserved.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a
   8  * copy of this software and associated documentation files (the
   9  * "Software"), to deal in the Software without restriction, including
  10  * without limitation the rights to use, copy, modify, merge, publish,
  11  * distribute, sub license, and/or sell copies of the Software, and to
  12  * permit persons to whom the Software is furnished to do so, subject to
  13  * the following conditions:
  14  *
  15  * The above copyright notice and this permission notice (including the
  16  * next paragraph) shall be included in all copies or substantial portions
  17  * of the Software.
  18  *
  19  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  20  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  21  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  22  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  23  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  24  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  25  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  26  *
  27  **************************************************************************/
  28
  29 /**
  30  * TGSI program scan utility.
  31  * Used to determine which registers and instructions are used by a shader.
  32  *
  33  * Authors:  Brian Paul
  34  */
  35
  36
  37 #include "util/u_debug.h"
  38 #include "util/u_math.h"
  39 #include "util/u_memory.h"
  40 #include "util/u_prim.h"
  41 #include "tgsi/tgsi_parse.h"
  42 #include "tgsi/tgsi_util.h"
  43 #include "tgsi/tgsi_scan.h"
  44
  45
  46
  47
  48 /**
  49  * Scan the given TGSI shader to collect information such as number of
  50  * registers used, special instructions used, etc.
  51  * \return info  the result of the scan
  52  */
  53 void
  54 tgsi_scan_shader(const struct tgsi_token *tokens,
  55                  struct tgsi_shader_info *info)
  56 {
  57    uint procType, i;
  58    struct tgsi_parse_context parse;
  59    unsigned current_depth = 0;
  60
  61    memset(info, 0, sizeof(*info));
  62    for (i = 0; i < TGSI_FILE_COUNT; i++)
  63       info->file_max[i] = -1;
  64    for (i = 0; i < Elements(info->const_file_max); i++)
  65       info->const_file_max[i] = -1;
  66    info->properties[TGSI_PROPERTY_GS_INVOCATIONS] = 1;
  67
  68    /**
  69     ** Setup to begin parsing input shader
  70     **/
  71    if (tgsi_parse_init( &parse, tokens ) != TGSI_PARSE_OK) {
  72       debug_printf("tgsi_parse_init() failed in tgsi_scan_shader()!\n");
  73       return;
  74    }
  75    procType = parse.FullHeader.Processor.Processor;
  76    assert(procType == TGSI_PROCESSOR_FRAGMENT ||
  77           procType == TGSI_PROCESSOR_VERTEX ||
  78           procType == TGSI_PROCESSOR_GEOMETRY ||
  79           procType == TGSI_PROCESSOR_TESS_CTRL ||
  80           procType == TGSI_PROCESSOR_TESS_EVAL ||
  81           procType == TGSI_PROCESSOR_COMPUTE);
  82    info->processor = procType;
  83
  84
  85    /**
  86     ** Loop over incoming program tokens/instructions
  87     */
  88    while( !tgsi_parse_end_of_tokens( &parse ) ) {
  89
  90       info->num_tokens++;
  91
  92       tgsi_parse_token( &parse );
  93
  94       switch( parse.FullToken.Token.Type ) {
  95       case TGSI_TOKEN_TYPE_INSTRUCTION:
  96          {
  97             const struct tgsi_full_instruction *fullinst
  98                = &parse.FullToken.FullInstruction;
  99             uint i;
 100
 101             assert(fullinst->Instruction.Opcode < TGSI_OPCODE_LAST);
 102             info->opcode_count[fullinst->Instruction.Opcode]++;
 103
 104             switch (fullinst->Instruction.Opcode) {
 105             case TGSI_OPCODE_IF:
 106             case TGSI_OPCODE_UIF:
 107             case TGSI_OPCODE_BGNLOOP:
 108                current_depth++;
 109                info->max_depth = MAX2(info->max_depth, current_depth);
 110                break;
 111             case TGSI_OPCODE_ENDIF:
 112             case TGSI_OPCODE_ENDLOOP:
 113                current_depth--;
 114                break;
 115             default:
 116                break;
 117             }
 118
 119             if (fullinst->Instruction.Opcode == TGSI_OPCODE_INTERP_CENTROID ||
 120                 fullinst->Instruction.Opcode == TGSI_OPCODE_INTERP_OFFSET ||
 121                 fullinst->Instruction.Opcode == TGSI_OPCODE_INTERP_SAMPLE) {
 122                const struct tgsi_full_src_register *src0 = &fullinst->Src[0];
 123                unsigned input;
 124
 125                if (src0->Register.Indirect && src0->Indirect.ArrayID)
 126                   input = info->input_array_first[src0->Indirect.ArrayID];
 127                else
 128                   input = src0->Register.Index;
 129
 130                /* For the INTERP opcodes, the interpolation is always
 131                 * PERSPECTIVE unless LINEAR is specified.
 132                 */
 133                switch (info->input_interpolate[input]) {
 134                case TGSI_INTERPOLATE_COLOR:
 135                case TGSI_INTERPOLATE_CONSTANT:
 136                case TGSI_INTERPOLATE_PERSPECTIVE:
 137                   switch (fullinst->Instruction.Opcode) {
 138                   case TGSI_OPCODE_INTERP_CENTROID:
 139                      info->uses_persp_opcode_interp_centroid = true;
 140                      break;
 141                   case TGSI_OPCODE_INTERP_OFFSET:
 142                      info->uses_persp_opcode_interp_offset = true;
 143                      break;
 144                   case TGSI_OPCODE_INTERP_SAMPLE:
 145                      info->uses_persp_opcode_interp_sample = true;
 146                      break;
 147                   }
 148                   break;
 149
 150                case TGSI_INTERPOLATE_LINEAR:
 151                   switch (fullinst->Instruction.Opcode) {
 152                   case TGSI_OPCODE_INTERP_CENTROID:
 153                      info->uses_linear_opcode_interp_centroid = true;
 154                      break;
 155                   case TGSI_OPCODE_INTERP_OFFSET:
 156                      info->uses_linear_opcode_interp_offset = true;
 157                      break;
 158                   case TGSI_OPCODE_INTERP_SAMPLE:
 159                      info->uses_linear_opcode_interp_sample = true;
 160                      break;
 161                   }
 162                   break;
 163                }
 164             }
 165
 166             if (fullinst->Instruction.Opcode >= TGSI_OPCODE_F2D &&
 167                 fullinst->Instruction.Opcode <= TGSI_OPCODE_DSSG)
 168                info->uses_doubles = true;
 169
 170             for (i = 0; i < fullinst->Instruction.NumSrcRegs; i++) {
 171                const struct tgsi_full_src_register *src =
 172                   &fullinst->Src[i];
 173                int ind = src->Register.Index;
 174
 175                /* Mark which inputs are effectively used */
 176                if (src->Register.File == TGSI_FILE_INPUT) {
 177                   unsigned usage_mask;
 178                   usage_mask = tgsi_util_get_inst_usage_mask(fullinst, i);
 179                   if (src->Register.Indirect) {
 180                      for (ind = 0; ind < info->num_inputs; ++ind) {
 181                         info->input_usage_mask[ind] |= usage_mask;
 182                      }
 183                   } else {
 184                      assert(ind >= 0);
 185                      assert(ind < PIPE_MAX_SHADER_INPUTS);
 186                      info->input_usage_mask[ind] |= usage_mask;
 187                   }
 188
 189                   if (procType == TGSI_PROCESSOR_FRAGMENT &&
 190                       info->reads_position &&
 191                       src->Register.Index == 0 &&
 192                       (src->Register.SwizzleX == TGSI_SWIZZLE_Z ||
 193                        src->Register.SwizzleY == TGSI_SWIZZLE_Z ||
 194                        src->Register.SwizzleZ == TGSI_SWIZZLE_Z ||
 195                        src->Register.SwizzleW == TGSI_SWIZZLE_Z)) {
 196                      info->reads_z = TRUE;
 197                   }
 198                }
 199
 200                /* check for indirect register reads */
 201                if (src->Register.Indirect) {
 202                   info->indirect_files |= (1 << src->Register.File);
 203                   info->indirect_files_read |= (1 << src->Register.File);
 204                }
 205
 206                /* MSAA samplers */
 207                if (src->Register.File == TGSI_FILE_SAMPLER) {
 208                   assert(fullinst->Instruction.Texture);
 209                   assert(src->Register.Index < Elements(info->is_msaa_sampler));
 210
 211                   if (fullinst->Instruction.Texture &&
 212                       (fullinst->Texture.Texture == TGSI_TEXTURE_2D_MSAA ||
 213                        fullinst->Texture.Texture == TGSI_TEXTURE_2D_ARRAY_MSAA)) {
 214                      info->is_msaa_sampler[src->Register.Index] = TRUE;
 215                   }
 216                }
 217             }
 218
 219             /* check for indirect register writes */
 220             for (i = 0; i < fullinst->Instruction.NumDstRegs; i++) {
 221                const struct tgsi_full_dst_register *dst = &fullinst->Dst[i];
 222                if (dst->Register.Indirect) {
 223                   info->indirect_files |= (1 << dst->Register.File);
 224                   info->indirect_files_written |= (1 << dst->Register.File);
 225                }
 226             }
 227
 228             info->num_instructions++;
 229          }
 230          break;
 231
 232       case TGSI_TOKEN_TYPE_DECLARATION:
 233          {
 234             const struct tgsi_full_declaration *fulldecl
 235                = &parse.FullToken.FullDeclaration;
 236             const uint file = fulldecl->Declaration.File;
 237             uint reg;
 238
 239             if (fulldecl->Declaration.Array) {
 240                unsigned array_id = fulldecl->Array.ArrayID;
 241
 242                switch (file) {
 243                case TGSI_FILE_INPUT:
 244                   assert(array_id < ARRAY_SIZE(info->input_array_first));
 245                   info->input_array_first[array_id] = fulldecl->Range.First;
 246                   info->input_array_last[array_id] = fulldecl->Range.Last;
 247                   break;
 248                case TGSI_FILE_OUTPUT:
 249                   assert(array_id < ARRAY_SIZE(info->output_array_first));
 250                   info->output_array_first[array_id] = fulldecl->Range.First;
 251                   info->output_array_last[array_id] = fulldecl->Range.Last;
 252                   break;
 253                }
 254                info->array_max[file] = MAX2(info->array_max[file], array_id);
 255             }
 256
 257             for (reg = fulldecl->Range.First;
 258                  reg <= fulldecl->Range.Last;
 259                  reg++) {
 260                unsigned semName = fulldecl->Semantic.Name;
 261                unsigned semIndex =
 262                   fulldecl->Semantic.Index + (reg - fulldecl->Range.First);
 263
 264                /* only first 32 regs will appear in this bitfield */
 265                info->file_mask[file] |= (1 << reg);
 266                info->file_count[file]++;
 267                info->file_max[file] = MAX2(info->file_max[file], (int)reg);
 268
 269                if (file == TGSI_FILE_CONSTANT) {
 270                   int buffer = 0;
 271
 272                   if (fulldecl->Declaration.Dimension)
 273                      buffer = fulldecl->Dim.Index2D;
 274
 275                   info->const_file_max[buffer] =
 276                         MAX2(info->const_file_max[buffer], (int)reg);
 277                }
 278                else if (file == TGSI_FILE_INPUT) {
 279                   info->input_semantic_name[reg] = (ubyte) semName;
 280                   info->input_semantic_index[reg] = (ubyte) semIndex;
 281                   info->input_interpolate[reg] = (ubyte)fulldecl->Interp.Interpolate;
 282                   info->input_interpolate_loc[reg] = (ubyte)fulldecl->Interp.Location;
 283                   info->input_cylindrical_wrap[reg] = (ubyte)fulldecl->Interp.CylindricalWrap;
 284                   info->num_inputs++;
 285
 286                   /* Only interpolated varyings. Don't include POSITION.
 287                    * Don't include integer varyings, because they are not
 288                    * interpolated.
 289                    */
 290                   if (semName == TGSI_SEMANTIC_GENERIC ||
 291                       semName == TGSI_SEMANTIC_TEXCOORD ||
 292                       semName == TGSI_SEMANTIC_COLOR ||
 293                       semName == TGSI_SEMANTIC_BCOLOR ||
 294                       semName == TGSI_SEMANTIC_FOG ||
 295                       semName == TGSI_SEMANTIC_CLIPDIST ||
 296                       semName == TGSI_SEMANTIC_CULLDIST) {
 297                      switch (fulldecl->Interp.Interpolate) {
 298                      case TGSI_INTERPOLATE_COLOR:
 299                      case TGSI_INTERPOLATE_PERSPECTIVE:
 300                         switch (fulldecl->Interp.Location) {
 301                         case TGSI_INTERPOLATE_LOC_CENTER:
 302                            info->uses_persp_center = true;
 303                            break;
 304                         case TGSI_INTERPOLATE_LOC_CENTROID:
 305                            info->uses_persp_centroid = true;
 306                            break;
 307                         case TGSI_INTERPOLATE_LOC_SAMPLE:
 308                            info->uses_persp_sample = true;
 309                            break;
 310                         }
 311                         break;
 312                      case TGSI_INTERPOLATE_LINEAR:
 313                         switch (fulldecl->Interp.Location) {
 314                         case TGSI_INTERPOLATE_LOC_CENTER:
 315                            info->uses_linear_center = true;
 316                            break;
 317                         case TGSI_INTERPOLATE_LOC_CENTROID:
 318                            info->uses_linear_centroid = true;
 319                            break;
 320                         case TGSI_INTERPOLATE_LOC_SAMPLE:
 321                            info->uses_linear_sample = true;
 322                            break;
 323                         }
 324                         break;
 325                      /* TGSI_INTERPOLATE_CONSTANT doesn't do any interpolation. */
 326                      }
 327                   }
 328
 329                   if (semName == TGSI_SEMANTIC_PRIMID)
 330                      info->uses_primid = TRUE;
 331                   else if (procType == TGSI_PROCESSOR_FRAGMENT) {
 332                      if (semName == TGSI_SEMANTIC_POSITION)
 333                         info->reads_position = TRUE;
 334                      else if (semName == TGSI_SEMANTIC_FACE)
 335                         info->uses_frontface = TRUE;
 336                   }
 337                }
 338                else if (file == TGSI_FILE_SYSTEM_VALUE) {
 339                   unsigned index = fulldecl->Range.First;
 340
 341                   info->system_value_semantic_name[index] = semName;
 342                   info->num_system_values = MAX2(info->num_system_values,
 343                                                  index + 1);
 344
 345                   if (semName == TGSI_SEMANTIC_INSTANCEID) {
 346                      info->uses_instanceid = TRUE;
 347                   }
 348                   else if (semName == TGSI_SEMANTIC_VERTEXID) {
 349                      info->uses_vertexid = TRUE;
 350                   }
 351                   else if (semName == TGSI_SEMANTIC_VERTEXID_NOBASE) {
 352                      info->uses_vertexid_nobase = TRUE;
 353                   }
 354                   else if (semName == TGSI_SEMANTIC_BASEVERTEX) {
 355                      info->uses_basevertex = TRUE;
 356                   }
 357                   else if (semName == TGSI_SEMANTIC_PRIMID) {
 358                      info->uses_primid = TRUE;
 359                   } else if (semName == TGSI_SEMANTIC_INVOCATIONID) {
 360                      info->uses_invocationid = TRUE;
 361                   }
 362                }
 363                else if (file == TGSI_FILE_OUTPUT) {
 364                   info->output_semantic_name[reg] = (ubyte) semName;
 365                   info->output_semantic_index[reg] = (ubyte) semIndex;
 366                   info->num_outputs++;
 367
 368                   if (procType == TGSI_PROCESSOR_VERTEX ||
 369                       procType == TGSI_PROCESSOR_GEOMETRY ||
 370                       procType == TGSI_PROCESSOR_TESS_CTRL ||
 371                       procType == TGSI_PROCESSOR_TESS_EVAL) {
 372                      if (semName == TGSI_SEMANTIC_CLIPDIST) {
 373                         info->num_written_clipdistance +=
 374                            util_bitcount(fulldecl->Declaration.UsageMask);
 375                         info->clipdist_writemask |=
 376                            fulldecl->Declaration.UsageMask << (semIndex*4);
 377                      }
 378                      else if (semName == TGSI_SEMANTIC_CULLDIST) {
 379                         info->num_written_culldistance +=
 380                            util_bitcount(fulldecl->Declaration.UsageMask);
 381                         info->culldist_writemask |=
 382                            fulldecl->Declaration.UsageMask << (semIndex*4);
 383                      }
 384                      else if (semName == TGSI_SEMANTIC_VIEWPORT_INDEX) {
 385                         info->writes_viewport_index = TRUE;
 386                      }
 387                      else if (semName == TGSI_SEMANTIC_LAYER) {
 388                         info->writes_layer = TRUE;
 389                      }
 390                      else if (semName == TGSI_SEMANTIC_PSIZE) {
 391                         info->writes_psize = TRUE;
 392                      }
 393                      else if (semName == TGSI_SEMANTIC_CLIPVERTEX) {
 394                         info->writes_clipvertex = TRUE;
 395                      }
 396                   }
 397
 398                   if (procType == TGSI_PROCESSOR_FRAGMENT) {
 399                      if (semName == TGSI_SEMANTIC_POSITION) {
 400                         info->writes_z = TRUE;
 401                      }
 402                      else if (semName == TGSI_SEMANTIC_STENCIL) {
 403                         info->writes_stencil = TRUE;
 404                      }
 405                   }
 406
 407                   if (procType == TGSI_PROCESSOR_VERTEX) {
 408                      if (semName == TGSI_SEMANTIC_EDGEFLAG) {
 409                         info->writes_edgeflag = TRUE;
 410                      }
 411                   }
 412                }
 413             }
 414          }
 415          break;
 416
 417       case TGSI_TOKEN_TYPE_IMMEDIATE:
 418          {
 419             uint reg = info->immediate_count++;
 420             uint file = TGSI_FILE_IMMEDIATE;
 421
 422             info->file_mask[file] |= (1 << reg);
 423             info->file_count[file]++;
 424             info->file_max[file] = MAX2(info->file_max[file], (int)reg);
 425          }
 426          break;
 427
 428       case TGSI_TOKEN_TYPE_PROPERTY:
 429          {
 430             const struct tgsi_full_property *fullprop
 431                = &parse.FullToken.FullProperty;
 432             unsigned name = fullprop->Property.PropertyName;
 433
 434             assert(name < Elements(info->properties));
 435             info->properties[name] = fullprop->u[0].Data;
 436          }
 437          break;
 438
 439       default:
 440          assert( 0 );
 441       }
 442    }
 443
 444    info->uses_kill = (info->opcode_count[TGSI_OPCODE_KILL_IF] ||
 445                       info->opcode_count[TGSI_OPCODE_KILL]);
 446
 447    /* The dimensions of the IN decleration in geometry shader have
 448     * to be deduced from the type of the input primitive.
 449     */
 450    if (procType == TGSI_PROCESSOR_GEOMETRY) {
 451       unsigned input_primitive =
 452             info->properties[TGSI_PROPERTY_GS_INPUT_PRIM];
 453       int num_verts = u_vertices_per_prim(input_primitive);
 454       int j;
 455       info->file_count[TGSI_FILE_INPUT] = num_verts;
 456       info->file_max[TGSI_FILE_INPUT] =
 457             MAX2(info->file_max[TGSI_FILE_INPUT], num_verts - 1);
 458       for (j = 0; j < num_verts; ++j) {
 459          info->file_mask[TGSI_FILE_INPUT] |= (1 << j);
 460       }
 461    }
 462
 463    tgsi_parse_free (&parse);
 464 }
 465
 466
 467
 468 /**
 469  * Check if the given shader is a "passthrough" shader consisting of only
 470  * MOV instructions of the form:  MOV OUT[n], IN[n]
 471  *
 472  */
 473 boolean
 474 tgsi_is_passthrough_shader(const struct tgsi_token *tokens)
 475 {
 476    struct tgsi_parse_context parse;
 477
 478    /**
 479     ** Setup to begin parsing input shader
 480     **/
 481    if (tgsi_parse_init(&parse, tokens) != TGSI_PARSE_OK) {
 482       debug_printf("tgsi_parse_init() failed in tgsi_is_passthrough_shader()!\n");
 483       return FALSE;
 484    }
 485
 486    /**
 487     ** Loop over incoming program tokens/instructions
 488     */
 489    while (!tgsi_parse_end_of_tokens(&parse)) {
 490
 491       tgsi_parse_token(&parse);
 492
 493       switch (parse.FullToken.Token.Type) {
 494       case TGSI_TOKEN_TYPE_INSTRUCTION:
 495          {
 496             struct tgsi_full_instruction *fullinst =
 497                &parse.FullToken.FullInstruction;
 498             const struct tgsi_full_src_register *src =
 499                &fullinst->Src[0];
 500             const struct tgsi_full_dst_register *dst =
 501                &fullinst->Dst[0];
 502
 503             /* Do a whole bunch of checks for a simple move */
 504             if (fullinst->Instruction.Opcode != TGSI_OPCODE_MOV ||
 505                 (src->Register.File != TGSI_FILE_INPUT &&
 506                  src->Register.File != TGSI_FILE_SYSTEM_VALUE) ||
 507                 dst->Register.File != TGSI_FILE_OUTPUT ||
 508                 src->Register.Index != dst->Register.Index ||
 509
 510                 src->Register.Negate ||
 511                 src->Register.Absolute ||
 512
 513                 src->Register.SwizzleX != TGSI_SWIZZLE_X ||
 514                 src->Register.SwizzleY != TGSI_SWIZZLE_Y ||
 515                 src->Register.SwizzleZ != TGSI_SWIZZLE_Z ||
 516                 src->Register.SwizzleW != TGSI_SWIZZLE_W ||
 517
 518                 dst->Register.WriteMask != TGSI_WRITEMASK_XYZW)
 519             {
 520                tgsi_parse_free(&parse);
 521                return FALSE;
 522             }
 523          }
 524          break;
 525
 526       case TGSI_TOKEN_TYPE_DECLARATION:
 527          /* fall-through */
 528       case TGSI_TOKEN_TYPE_IMMEDIATE:
 529          /* fall-through */
 530       case TGSI_TOKEN_TYPE_PROPERTY:
 531          /* fall-through */
 532       default:
 533          ; /* no-op */
 534       }
 535    }
 536
 537    tgsi_parse_free(&parse);
 538
 539    /* if we get here, it's a pass-through shader */
 540    return TRUE;
 541 }