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