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