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
  60    memset(info, 0, sizeof(*info));
  61    for (i = 0; i < TGSI_FILE_COUNT; i++)
  62       info->file_max[i] = -1;
  63    for (i = 0; i < Elements(info->const_file_max); i++)
  64       info->const_file_max[i] = -1;
  65    info->properties[TGSI_PROPERTY_GS_INVOCATIONS] = 1;
  66
  67    /**
  68     ** Setup to begin parsing input shader
  69     **/
  70    if (tgsi_parse_init( &parse, tokens ) != TGSI_PARSE_OK) {
  71       debug_printf("tgsi_parse_init() failed in tgsi_scan_shader()!\n");
  72       return;
  73    }
  74    procType = parse.FullHeader.Processor.Processor;
  75    assert(procType == TGSI_PROCESSOR_FRAGMENT ||
  76           procType == TGSI_PROCESSOR_VERTEX ||
  77           procType == TGSI_PROCESSOR_GEOMETRY ||
  78           procType == TGSI_PROCESSOR_TESS_CTRL ||
  79           procType == TGSI_PROCESSOR_TESS_EVAL ||
  80           procType == TGSI_PROCESSOR_COMPUTE);
  81    info->processor = procType;
  82
  83
  84    /**
  85     ** Loop over incoming program tokens/instructions
  86     */
  87    while( !tgsi_parse_end_of_tokens( &parse ) ) {
  88
  89       info->num_tokens++;
  90
  91       tgsi_parse_token( &parse );
  92
  93       switch( parse.FullToken.Token.Type ) {
  94       case TGSI_TOKEN_TYPE_INSTRUCTION:
  95          {
  96             const struct tgsi_full_instruction *fullinst
  97                = &parse.FullToken.FullInstruction;
  98             uint i;
  99
 100             assert(fullinst->Instruction.Opcode < TGSI_OPCODE_LAST);
 101             info->opcode_count[fullinst->Instruction.Opcode]++;
 102
 103             if (fullinst->Instruction.Opcode >= TGSI_OPCODE_F2D &&
 104                 fullinst->Instruction.Opcode <= TGSI_OPCODE_DSSG)
 105                info->uses_doubles = true;
 106
 107             for (i = 0; i < fullinst->Instruction.NumSrcRegs; i++) {
 108                const struct tgsi_full_src_register *src =
 109                   &fullinst->Src[i];
 110                int ind = src->Register.Index;
 111
 112                /* Mark which inputs are effectively used */
 113                if (src->Register.File == TGSI_FILE_INPUT) {
 114                   unsigned usage_mask;
 115                   usage_mask = tgsi_util_get_inst_usage_mask(fullinst, i);
 116                   if (src->Register.Indirect) {
 117                      for (ind = 0; ind < info->num_inputs; ++ind) {
 118                         info->input_usage_mask[ind] |= usage_mask;
 119                      }
 120                   } else {
 121                      assert(ind >= 0);
 122                      assert(ind < PIPE_MAX_SHADER_INPUTS);
 123                      info->input_usage_mask[ind] |= usage_mask;
 124                   }
 125
 126                   if (procType == TGSI_PROCESSOR_FRAGMENT &&
 127                       info->reads_position &&
 128                       src->Register.Index == 0 &&
 129                       (src->Register.SwizzleX == TGSI_SWIZZLE_Z ||
 130                        src->Register.SwizzleY == TGSI_SWIZZLE_Z ||
 131                        src->Register.SwizzleZ == TGSI_SWIZZLE_Z ||
 132                        src->Register.SwizzleW == TGSI_SWIZZLE_Z)) {
 133                      info->reads_z = TRUE;
 134                   }
 135                }
 136
 137                /* check for indirect register reads */
 138                if (src->Register.Indirect) {
 139                   info->indirect_files |= (1 << src->Register.File);
 140                   info->indirect_files_read |= (1 << src->Register.File);
 141                }
 142
 143                /* MSAA samplers */
 144                if (src->Register.File == TGSI_FILE_SAMPLER) {
 145                   assert(fullinst->Instruction.Texture);
 146                   assert(src->Register.Index < Elements(info->is_msaa_sampler));
 147
 148                   if (fullinst->Instruction.Texture &&
 149                       (fullinst->Texture.Texture == TGSI_TEXTURE_2D_MSAA ||
 150                        fullinst->Texture.Texture == TGSI_TEXTURE_2D_ARRAY_MSAA)) {
 151                      info->is_msaa_sampler[src->Register.Index] = TRUE;
 152                   }
 153                }
 154             }
 155
 156             /* check for indirect register writes */
 157             for (i = 0; i < fullinst->Instruction.NumDstRegs; i++) {
 158                const struct tgsi_full_dst_register *dst = &fullinst->Dst[i];
 159                if (dst->Register.Indirect) {
 160                   info->indirect_files |= (1 << dst->Register.File);
 161                   info->indirect_files_written |= (1 << dst->Register.File);
 162                }
 163             }
 164
 165             info->num_instructions++;
 166          }
 167          break;
 168
 169       case TGSI_TOKEN_TYPE_DECLARATION:
 170          {
 171             const struct tgsi_full_declaration *fulldecl
 172                = &parse.FullToken.FullDeclaration;
 173             const uint file = fulldecl->Declaration.File;
 174             uint reg;
 175
 176             if (fulldecl->Declaration.Array) {
 177                unsigned array_id = fulldecl->Array.ArrayID;
 178
 179                switch (file) {
 180                case TGSI_FILE_INPUT:
 181                   assert(array_id < ARRAY_SIZE(info->input_array_first));
 182                   info->input_array_first[array_id] = fulldecl->Range.First;
 183                   info->input_array_last[array_id] = fulldecl->Range.Last;
 184                   break;
 185                case TGSI_FILE_OUTPUT:
 186                   assert(array_id < ARRAY_SIZE(info->output_array_first));
 187                   info->output_array_first[array_id] = fulldecl->Range.First;
 188                   info->output_array_last[array_id] = fulldecl->Range.Last;
 189                   break;
 190                }
 191                info->array_max[file] = MAX2(info->array_max[file], array_id);
 192             }
 193
 194             for (reg = fulldecl->Range.First;
 195                  reg <= fulldecl->Range.Last;
 196                  reg++) {
 197                unsigned semName = fulldecl->Semantic.Name;
 198                unsigned semIndex =
 199                   fulldecl->Semantic.Index + (reg - fulldecl->Range.First);
 200
 201                /* only first 32 regs will appear in this bitfield */
 202                info->file_mask[file] |= (1 << reg);
 203                info->file_count[file]++;
 204                info->file_max[file] = MAX2(info->file_max[file], (int)reg);
 205
 206                if (file == TGSI_FILE_CONSTANT) {
 207                   int buffer = 0;
 208
 209                   if (fulldecl->Declaration.Dimension)
 210                      buffer = fulldecl->Dim.Index2D;
 211
 212                   info->const_file_max[buffer] =
 213                         MAX2(info->const_file_max[buffer], (int)reg);
 214                }
 215                else if (file == TGSI_FILE_INPUT) {
 216                   info->input_semantic_name[reg] = (ubyte) semName;
 217                   info->input_semantic_index[reg] = (ubyte) semIndex;
 218                   info->input_interpolate[reg] = (ubyte)fulldecl->Interp.Interpolate;
 219                   info->input_interpolate_loc[reg] = (ubyte)fulldecl->Interp.Location;
 220                   info->input_cylindrical_wrap[reg] = (ubyte)fulldecl->Interp.CylindricalWrap;
 221                   info->num_inputs++;
 222
 223                   if (fulldecl->Interp.Location == TGSI_INTERPOLATE_LOC_CENTROID)
 224                      info->uses_centroid = TRUE;
 225
 226                   if (semName == TGSI_SEMANTIC_PRIMID)
 227                      info->uses_primid = TRUE;
 228                   else if (procType == TGSI_PROCESSOR_FRAGMENT) {
 229                      if (semName == TGSI_SEMANTIC_POSITION)
 230                         info->reads_position = TRUE;
 231                      else if (semName == TGSI_SEMANTIC_FACE)
 232                         info->uses_frontface = TRUE;
 233                   }
 234                }
 235                else if (file == TGSI_FILE_SYSTEM_VALUE) {
 236                   unsigned index = fulldecl->Range.First;
 237
 238                   info->system_value_semantic_name[index] = semName;
 239                   info->num_system_values = MAX2(info->num_system_values,
 240                                                  index + 1);
 241
 242                   if (semName == TGSI_SEMANTIC_INSTANCEID) {
 243                      info->uses_instanceid = TRUE;
 244                   }
 245                   else if (semName == TGSI_SEMANTIC_VERTEXID) {
 246                      info->uses_vertexid = TRUE;
 247                   }
 248                   else if (semName == TGSI_SEMANTIC_VERTEXID_NOBASE) {
 249                      info->uses_vertexid_nobase = TRUE;
 250                   }
 251                   else if (semName == TGSI_SEMANTIC_BASEVERTEX) {
 252                      info->uses_basevertex = TRUE;
 253                   }
 254                   else if (semName == TGSI_SEMANTIC_PRIMID) {
 255                      info->uses_primid = TRUE;
 256                   } else if (semName == TGSI_SEMANTIC_INVOCATIONID) {
 257                      info->uses_invocationid = TRUE;
 258                   }
 259                }
 260                else if (file == TGSI_FILE_OUTPUT) {
 261                   info->output_semantic_name[reg] = (ubyte) semName;
 262                   info->output_semantic_index[reg] = (ubyte) semIndex;
 263                   info->num_outputs++;
 264
 265                   if (procType == TGSI_PROCESSOR_VERTEX ||
 266                       procType == TGSI_PROCESSOR_GEOMETRY ||
 267                       procType == TGSI_PROCESSOR_TESS_CTRL ||
 268                       procType == TGSI_PROCESSOR_TESS_EVAL) {
 269                      if (semName == TGSI_SEMANTIC_CLIPDIST) {
 270                         info->num_written_clipdistance +=
 271                            util_bitcount(fulldecl->Declaration.UsageMask);
 272                         info->clipdist_writemask |=
 273                            fulldecl->Declaration.UsageMask << (semIndex*4);
 274                      }
 275                      else if (semName == TGSI_SEMANTIC_CULLDIST) {
 276                         info->num_written_culldistance +=
 277                            util_bitcount(fulldecl->Declaration.UsageMask);
 278                         info->culldist_writemask |=
 279                            fulldecl->Declaration.UsageMask << (semIndex*4);
 280                      }
 281                      else if (semName == TGSI_SEMANTIC_VIEWPORT_INDEX) {
 282                         info->writes_viewport_index = TRUE;
 283                      }
 284                      else if (semName == TGSI_SEMANTIC_LAYER) {
 285                         info->writes_layer = TRUE;
 286                      }
 287                      else if (semName == TGSI_SEMANTIC_PSIZE) {
 288                         info->writes_psize = TRUE;
 289                      }
 290                      else if (semName == TGSI_SEMANTIC_CLIPVERTEX) {
 291                         info->writes_clipvertex = TRUE;
 292                      }
 293                   }
 294
 295                   if (procType == TGSI_PROCESSOR_FRAGMENT) {
 296                      if (semName == TGSI_SEMANTIC_POSITION) {
 297                         info->writes_z = TRUE;
 298                      }
 299                      else if (semName == TGSI_SEMANTIC_STENCIL) {
 300                         info->writes_stencil = TRUE;
 301                      }
 302                   }
 303
 304                   if (procType == TGSI_PROCESSOR_VERTEX) {
 305                      if (semName == TGSI_SEMANTIC_EDGEFLAG) {
 306                         info->writes_edgeflag = TRUE;
 307                      }
 308                   }
 309                }
 310             }
 311          }
 312          break;
 313
 314       case TGSI_TOKEN_TYPE_IMMEDIATE:
 315          {
 316             uint reg = info->immediate_count++;
 317             uint file = TGSI_FILE_IMMEDIATE;
 318
 319             info->file_mask[file] |= (1 << reg);
 320             info->file_count[file]++;
 321             info->file_max[file] = MAX2(info->file_max[file], (int)reg);
 322          }
 323          break;
 324
 325       case TGSI_TOKEN_TYPE_PROPERTY:
 326          {
 327             const struct tgsi_full_property *fullprop
 328                = &parse.FullToken.FullProperty;
 329             unsigned name = fullprop->Property.PropertyName;
 330
 331             assert(name < Elements(info->properties));
 332             info->properties[name] = fullprop->u[0].Data;
 333          }
 334          break;
 335
 336       default:
 337          assert( 0 );
 338       }
 339    }
 340
 341    info->uses_kill = (info->opcode_count[TGSI_OPCODE_KILL_IF] ||
 342                       info->opcode_count[TGSI_OPCODE_KILL]);
 343
 344    /* The dimensions of the IN decleration in geometry shader have
 345     * to be deduced from the type of the input primitive.
 346     */
 347    if (procType == TGSI_PROCESSOR_GEOMETRY) {
 348       unsigned input_primitive =
 349             info->properties[TGSI_PROPERTY_GS_INPUT_PRIM];
 350       int num_verts = u_vertices_per_prim(input_primitive);
 351       int j;
 352       info->file_count[TGSI_FILE_INPUT] = num_verts;
 353       info->file_max[TGSI_FILE_INPUT] =
 354             MAX2(info->file_max[TGSI_FILE_INPUT], num_verts - 1);
 355       for (j = 0; j < num_verts; ++j) {
 356          info->file_mask[TGSI_FILE_INPUT] |= (1 << j);
 357       }
 358    }
 359
 360    tgsi_parse_free (&parse);
 361 }
 362
 363
 364
 365 /**
 366  * Check if the given shader is a "passthrough" shader consisting of only
 367  * MOV instructions of the form:  MOV OUT[n], IN[n]
 368  *
 369  */
 370 boolean
 371 tgsi_is_passthrough_shader(const struct tgsi_token *tokens)
 372 {
 373    struct tgsi_parse_context parse;
 374
 375    /**
 376     ** Setup to begin parsing input shader
 377     **/
 378    if (tgsi_parse_init(&parse, tokens) != TGSI_PARSE_OK) {
 379       debug_printf("tgsi_parse_init() failed in tgsi_is_passthrough_shader()!\n");
 380       return FALSE;
 381    }
 382
 383    /**
 384     ** Loop over incoming program tokens/instructions
 385     */
 386    while (!tgsi_parse_end_of_tokens(&parse)) {
 387
 388       tgsi_parse_token(&parse);
 389
 390       switch (parse.FullToken.Token.Type) {
 391       case TGSI_TOKEN_TYPE_INSTRUCTION:
 392          {
 393             struct tgsi_full_instruction *fullinst =
 394                &parse.FullToken.FullInstruction;
 395             const struct tgsi_full_src_register *src =
 396                &fullinst->Src[0];
 397             const struct tgsi_full_dst_register *dst =
 398                &fullinst->Dst[0];
 399
 400             /* Do a whole bunch of checks for a simple move */
 401             if (fullinst->Instruction.Opcode != TGSI_OPCODE_MOV ||
 402                 (src->Register.File != TGSI_FILE_INPUT &&
 403                  src->Register.File != TGSI_FILE_SYSTEM_VALUE) ||
 404                 dst->Register.File != TGSI_FILE_OUTPUT ||
 405                 src->Register.Index != dst->Register.Index ||
 406
 407                 src->Register.Negate ||
 408                 src->Register.Absolute ||
 409
 410                 src->Register.SwizzleX != TGSI_SWIZZLE_X ||
 411                 src->Register.SwizzleY != TGSI_SWIZZLE_Y ||
 412                 src->Register.SwizzleZ != TGSI_SWIZZLE_Z ||
 413                 src->Register.SwizzleW != TGSI_SWIZZLE_W ||
 414
 415                 dst->Register.WriteMask != TGSI_WRITEMASK_XYZW)
 416             {
 417                tgsi_parse_free(&parse);
 418                return FALSE;
 419             }
 420          }
 421          break;
 422
 423       case TGSI_TOKEN_TYPE_DECLARATION:
 424          /* fall-through */
 425       case TGSI_TOKEN_TYPE_IMMEDIATE:
 426          /* fall-through */
 427       case TGSI_TOKEN_TYPE_PROPERTY:
 428          /* fall-through */
 429       default:
 430          ; /* no-op */
 431       }
 432    }
 433
 434    tgsi_parse_free(&parse);
 435
 436    /* if we get here, it's a pass-through shader */
 437    return TRUE;
 438 }