src/gallium/auxiliary/tgsi/tgsi_scan.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
   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 TUNGSTEN GRAPHICS 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                }
 134
 135                /* MSAA samplers */
 136                if (src->Register.File == TGSI_FILE_SAMPLER) {
 137                   assert(fullinst->Instruction.Texture);
 138                   assert(src->Register.Index < Elements(info->is_msaa_sampler));
 139
 140                   if (fullinst->Instruction.Texture &&
 141                       (fullinst->Texture.Texture == TGSI_TEXTURE_2D_MSAA ||
 142                        fullinst->Texture.Texture == TGSI_TEXTURE_2D_ARRAY_MSAA)) {
 143                      info->is_msaa_sampler[src->Register.Index] = TRUE;
 144                   }
 145                }
 146             }
 147
 148             /* check for indirect register writes */
 149             for (i = 0; i < fullinst->Instruction.NumDstRegs; i++) {
 150                const struct tgsi_full_dst_register *dst = &fullinst->Dst[i];
 151                if (dst->Register.Indirect) {
 152                   info->indirect_files |= (1 << dst->Register.File);
 153                }
 154             }
 155
 156             info->num_instructions++;
 157          }
 158          break;
 159
 160       case TGSI_TOKEN_TYPE_DECLARATION:
 161          {
 162             const struct tgsi_full_declaration *fulldecl
 163                = &parse.FullToken.FullDeclaration;
 164             const uint file = fulldecl->Declaration.File;
 165             uint reg;
 166             for (reg = fulldecl->Range.First;
 167                  reg <= fulldecl->Range.Last;
 168                  reg++) {
 169                unsigned semName = fulldecl->Semantic.Name;
 170                unsigned semIndex = fulldecl->Semantic.Index;
 171
 172                /* only first 32 regs will appear in this bitfield */
 173                info->file_mask[file] |= (1 << reg);
 174                info->file_count[file]++;
 175                info->file_max[file] = MAX2(info->file_max[file], (int)reg);
 176
 177                if (file == TGSI_FILE_CONSTANT) {
 178                   int buffer = 0;
 179
 180                   if (fulldecl->Declaration.Dimension)
 181                      buffer = fulldecl->Dim.Index2D;
 182
 183                   info->const_file_max[buffer] =
 184                         MAX2(info->const_file_max[buffer], (int)reg);
 185                }
 186                else if (file == TGSI_FILE_INPUT) {
 187                   info->input_semantic_name[reg] = (ubyte) semName;
 188                   info->input_semantic_index[reg] = (ubyte) semIndex;
 189                   info->input_interpolate[reg] = (ubyte)fulldecl->Interp.Interpolate;
 190                   info->input_centroid[reg] = (ubyte)fulldecl->Interp.Centroid;
 191                   info->input_cylindrical_wrap[reg] = (ubyte)fulldecl->Interp.CylindricalWrap;
 192                   info->num_inputs++;
 193
 194                   if (procType == TGSI_PROCESSOR_FRAGMENT) {
 195                      if (semName == TGSI_SEMANTIC_POSITION)
 196                         info->reads_position = TRUE;
 197                      else if (semName == TGSI_SEMANTIC_PRIMID)
 198                         info->uses_primid = TRUE;
 199                      else if (semName == TGSI_SEMANTIC_FACE)
 200                         info->uses_frontface = TRUE;
 201                   }
 202                }
 203                else if (file == TGSI_FILE_SYSTEM_VALUE) {
 204                   unsigned index = fulldecl->Range.First;
 205
 206                   info->system_value_semantic_name[index] = semName;
 207                   info->num_system_values = MAX2(info->num_system_values,
 208                                                  index + 1);
 209
 210                   if (semName == TGSI_SEMANTIC_INSTANCEID) {
 211                      info->uses_instanceid = TRUE;
 212                   }
 213                   else if (semName == TGSI_SEMANTIC_VERTEXID) {
 214                      info->uses_vertexid = TRUE;
 215                   }
 216                   else if (semName == TGSI_SEMANTIC_PRIMID) {
 217                      info->uses_primid = TRUE;
 218                   }
 219                }
 220                else if (file == TGSI_FILE_OUTPUT) {
 221                   info->output_semantic_name[reg] = (ubyte) semName;
 222                   info->output_semantic_index[reg] = (ubyte) semIndex;
 223                   info->num_outputs++;
 224
 225                   if (procType == TGSI_PROCESSOR_VERTEX ||
 226                       procType == TGSI_PROCESSOR_GEOMETRY) {
 227                      if (semName == TGSI_SEMANTIC_CLIPDIST) {
 228                         info->num_written_clipdistance +=
 229                            util_bitcount(fulldecl->Declaration.UsageMask);
 230                      }
 231                      else if (semName == TGSI_SEMANTIC_CULLDIST) {
 232                         info->num_written_culldistance +=
 233                            util_bitcount(fulldecl->Declaration.UsageMask);
 234                      }
 235                   }
 236
 237                   if (procType == TGSI_PROCESSOR_FRAGMENT) {
 238                      if (semName == TGSI_SEMANTIC_POSITION) {
 239                         info->writes_z = TRUE;
 240                      }
 241                      else if (semName == TGSI_SEMANTIC_STENCIL) {
 242                         info->writes_stencil = TRUE;
 243                      }
 244                   }
 245
 246                   if (procType == TGSI_PROCESSOR_VERTEX) {
 247                      if (semName == TGSI_SEMANTIC_EDGEFLAG) {
 248                         info->writes_edgeflag = TRUE;
 249                      }
 250                   }
 251
 252                   if (procType == TGSI_PROCESSOR_GEOMETRY) {
 253                      if (semName == TGSI_SEMANTIC_VIEWPORT_INDEX) {
 254                         info->writes_viewport_index = TRUE;
 255                      }
 256                      else if (semName == TGSI_SEMANTIC_LAYER) {
 257                         info->writes_layer = TRUE;
 258                      }
 259                   }
 260                }
 261             }
 262          }
 263          break;
 264
 265       case TGSI_TOKEN_TYPE_IMMEDIATE:
 266          {
 267             uint reg = info->immediate_count++;
 268             uint file = TGSI_FILE_IMMEDIATE;
 269
 270             info->file_mask[file] |= (1 << reg);
 271             info->file_count[file]++;
 272             info->file_max[file] = MAX2(info->file_max[file], (int)reg);
 273          }
 274          break;
 275
 276       case TGSI_TOKEN_TYPE_PROPERTY:
 277          {
 278             const struct tgsi_full_property *fullprop
 279                = &parse.FullToken.FullProperty;
 280
 281             info->properties[info->num_properties].name =
 282                fullprop->Property.PropertyName;
 283             memcpy(info->properties[info->num_properties].data,
 284                    fullprop->u, 8 * sizeof(unsigned));;
 285
 286             ++info->num_properties;
 287          }
 288          break;
 289
 290       default:
 291          assert( 0 );
 292       }
 293    }
 294
 295    info->uses_kill = (info->opcode_count[TGSI_OPCODE_KILL_IF] ||
 296                       info->opcode_count[TGSI_OPCODE_KILL]);
 297
 298    /* extract simple properties */
 299    for (i = 0; i < info->num_properties; ++i) {
 300       switch (info->properties[i].name) {
 301       case TGSI_PROPERTY_FS_COORD_ORIGIN:
 302          info->origin_lower_left = info->properties[i].data[0];
 303          break;
 304       case TGSI_PROPERTY_FS_COORD_PIXEL_CENTER:
 305          info->pixel_center_integer = info->properties[i].data[0];
 306          break;
 307       case TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS:
 308          info->color0_writes_all_cbufs = info->properties[i].data[0];
 309          break;
 310       case TGSI_PROPERTY_GS_INPUT_PRIM:
 311          /* The dimensions of the IN decleration in geometry shader have
 312           * to be deduced from the type of the input primitive.
 313           */
 314          if (procType == TGSI_PROCESSOR_GEOMETRY) {
 315             unsigned input_primitive = info->properties[i].data[0];
 316             int num_verts = u_vertices_per_prim(input_primitive);
 317             int j;
 318             info->file_count[TGSI_FILE_INPUT] = num_verts;
 319             info->file_max[TGSI_FILE_INPUT] =
 320                MAX2(info->file_max[TGSI_FILE_INPUT], num_verts - 1);
 321             for (j = 0; j < num_verts; ++j) {
 322                info->file_mask[TGSI_FILE_INPUT] |= (1 << j);
 323             }
 324          }
 325          break;
 326       default:
 327          ;
 328       }
 329    }
 330
 331    tgsi_parse_free (&parse);
 332 }
 333
 334
 335
 336 /**
 337  * Check if the given shader is a "passthrough" shader consisting of only
 338  * MOV instructions of the form:  MOV OUT[n], IN[n]
 339  *
 340  */
 341 boolean
 342 tgsi_is_passthrough_shader(const struct tgsi_token *tokens)
 343 {
 344    struct tgsi_parse_context parse;
 345
 346    /**
 347     ** Setup to begin parsing input shader
 348     **/
 349    if (tgsi_parse_init(&parse, tokens) != TGSI_PARSE_OK) {
 350       debug_printf("tgsi_parse_init() failed in tgsi_is_passthrough_shader()!\n");
 351       return FALSE;
 352    }
 353
 354    /**
 355     ** Loop over incoming program tokens/instructions
 356     */
 357    while (!tgsi_parse_end_of_tokens(&parse)) {
 358
 359       tgsi_parse_token(&parse);
 360
 361       switch (parse.FullToken.Token.Type) {
 362       case TGSI_TOKEN_TYPE_INSTRUCTION:
 363          {
 364             struct tgsi_full_instruction *fullinst =
 365                &parse.FullToken.FullInstruction;
 366             const struct tgsi_full_src_register *src =
 367                &fullinst->Src[0];
 368             const struct tgsi_full_dst_register *dst =
 369                &fullinst->Dst[0];
 370
 371             /* Do a whole bunch of checks for a simple move */
 372             if (fullinst->Instruction.Opcode != TGSI_OPCODE_MOV ||
 373                 (src->Register.File != TGSI_FILE_INPUT &&
 374                  src->Register.File != TGSI_FILE_SYSTEM_VALUE) ||
 375                 dst->Register.File != TGSI_FILE_OUTPUT ||
 376                 src->Register.Index != dst->Register.Index ||
 377
 378                 src->Register.Negate ||
 379                 src->Register.Absolute ||
 380
 381                 src->Register.SwizzleX != TGSI_SWIZZLE_X ||
 382                 src->Register.SwizzleY != TGSI_SWIZZLE_Y ||
 383                 src->Register.SwizzleZ != TGSI_SWIZZLE_Z ||
 384                 src->Register.SwizzleW != TGSI_SWIZZLE_W ||
 385
 386                 dst->Register.WriteMask != TGSI_WRITEMASK_XYZW)
 387             {
 388                tgsi_parse_free(&parse);
 389                return FALSE;
 390             }
 391          }
 392          break;
 393
 394       case TGSI_TOKEN_TYPE_DECLARATION:
 395          /* fall-through */
 396       case TGSI_TOKEN_TYPE_IMMEDIATE:
 397          /* fall-through */
 398       case TGSI_TOKEN_TYPE_PROPERTY:
 399          /* fall-through */
 400       default:
 401          ; /* no-op */
 402       }
 403    }
 404
 405    tgsi_parse_free(&parse);
 406
 407    /* if we get here, it's a pass-through shader */
 408    return TRUE;
 409 }