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_prim.h"
  40 #include "tgsi/tgsi_parse.h"
  41 #include "tgsi/tgsi_util.h"
  42 #include "tgsi/tgsi_scan.h"
  43
  44
  45
  46
  47 /**
  48  * Scan the given TGSI shader to collect information such as number of
  49  * registers used, special instructions used, etc.
  50  * \return info  the result of the scan
  51  */
  52 void
  53 tgsi_scan_shader(const struct tgsi_token *tokens,
  54                  struct tgsi_shader_info *info)
  55 {
  56    uint procType, i;
  57    struct tgsi_parse_context parse;
  58
  59    memset(info, 0, sizeof(*info));
  60    for (i = 0; i < TGSI_FILE_COUNT; i++)
  61       info->file_max[i] = -1;
  62
  63    /**
  64     ** Setup to begin parsing input shader
  65     **/
  66    if (tgsi_parse_init( &parse, tokens ) != TGSI_PARSE_OK) {
  67       debug_printf("tgsi_parse_init() failed in tgsi_scan_shader()!\n");
  68       return;
  69    }
  70    procType = parse.FullHeader.Processor.Processor;
  71    assert(procType == TGSI_PROCESSOR_FRAGMENT ||
  72           procType == TGSI_PROCESSOR_VERTEX ||
  73           procType == TGSI_PROCESSOR_GEOMETRY ||
  74           procType == TGSI_PROCESSOR_COMPUTE);
  75
  76
  77    /**
  78     ** Loop over incoming program tokens/instructions
  79     */
  80    while( !tgsi_parse_end_of_tokens( &parse ) ) {
  81
  82       info->num_tokens++;
  83
  84       tgsi_parse_token( &parse );
  85
  86       switch( parse.FullToken.Token.Type ) {
  87       case TGSI_TOKEN_TYPE_INSTRUCTION:
  88          {
  89             const struct tgsi_full_instruction *fullinst
  90                = &parse.FullToken.FullInstruction;
  91             uint i;
  92
  93             assert(fullinst->Instruction.Opcode < TGSI_OPCODE_LAST);
  94             info->opcode_count[fullinst->Instruction.Opcode]++;
  95
  96             for (i = 0; i < fullinst->Instruction.NumSrcRegs; i++) {
  97                const struct tgsi_full_src_register *src =
  98                   &fullinst->Src[i];
  99                int ind = src->Register.Index;
 100
 101                /* Mark which inputs are effectively used */
 102                if (src->Register.File == TGSI_FILE_INPUT) {
 103                   unsigned usage_mask;
 104                   usage_mask = tgsi_util_get_inst_usage_mask(fullinst, i);
 105                   if (src->Register.Indirect) {
 106                      for (ind = 0; ind < info->num_inputs; ++ind) {
 107                         info->input_usage_mask[ind] |= usage_mask;
 108                      }
 109                   } else {
 110                      assert(ind >= 0);
 111                      assert(ind < PIPE_MAX_SHADER_INPUTS);
 112                      info->input_usage_mask[ind] |= usage_mask;
 113                   }
 114
 115                   if (procType == TGSI_PROCESSOR_FRAGMENT &&
 116                       src->Register.File == TGSI_FILE_INPUT &&
 117                       info->reads_position &&
 118                       src->Register.Index == 0 &&
 119                       (src->Register.SwizzleX == TGSI_SWIZZLE_Z ||
 120                        src->Register.SwizzleY == TGSI_SWIZZLE_Z ||
 121                        src->Register.SwizzleZ == TGSI_SWIZZLE_Z ||
 122                        src->Register.SwizzleW == TGSI_SWIZZLE_Z)) {
 123                      info->reads_z = TRUE;
 124                   }
 125                }
 126
 127                /* check for indirect register reads */
 128                if (src->Register.Indirect) {
 129                   info->indirect_files |= (1 << src->Register.File);
 130                }
 131             }
 132
 133             /* check for indirect register writes */
 134             for (i = 0; i < fullinst->Instruction.NumDstRegs; i++) {
 135                const struct tgsi_full_dst_register *dst = &fullinst->Dst[i];
 136                if (dst->Register.Indirect) {
 137                   info->indirect_files |= (1 << dst->Register.File);
 138                }
 139             }
 140
 141             info->num_instructions++;
 142          }
 143          break;
 144
 145       case TGSI_TOKEN_TYPE_DECLARATION:
 146          {
 147             const struct tgsi_full_declaration *fulldecl
 148                = &parse.FullToken.FullDeclaration;
 149             const uint file = fulldecl->Declaration.File;
 150             uint reg;
 151             for (reg = fulldecl->Range.First;
 152                  reg <= fulldecl->Range.Last;
 153                  reg++) {
 154
 155                /* only first 32 regs will appear in this bitfield */
 156                info->file_mask[file] |= (1 << reg);
 157                info->file_count[file]++;
 158                info->file_max[file] = MAX2(info->file_max[file], (int)reg);
 159
 160                if (file == TGSI_FILE_INPUT) {
 161                   info->input_semantic_name[reg] = (ubyte)fulldecl->Semantic.Name;
 162                   info->input_semantic_index[reg] = (ubyte)fulldecl->Semantic.Index;
 163                   info->input_interpolate[reg] = (ubyte)fulldecl->Interp.Interpolate;
 164                   info->input_centroid[reg] = (ubyte)fulldecl->Interp.Centroid;
 165                   info->input_cylindrical_wrap[reg] = (ubyte)fulldecl->Interp.CylindricalWrap;
 166                   info->num_inputs++;
 167
 168                   if (procType == TGSI_PROCESSOR_FRAGMENT &&
 169                       fulldecl->Semantic.Name == TGSI_SEMANTIC_POSITION)
 170                         info->reads_position = TRUE;
 171                }
 172                else if (file == TGSI_FILE_SYSTEM_VALUE) {
 173                   unsigned index = fulldecl->Range.First;
 174                   unsigned semName = fulldecl->Semantic.Name;
 175
 176                   info->system_value_semantic_name[index] = semName;
 177                   info->num_system_values = MAX2(info->num_system_values,
 178                                                  index + 1);
 179
 180                   /*
 181                   info->system_value_semantic_name[info->num_system_values++] =
 182                      fulldecl->Semantic.Name;
 183                   */
 184
 185                   if (fulldecl->Semantic.Name == TGSI_SEMANTIC_INSTANCEID) {
 186                      info->uses_instanceid = TRUE;
 187                   }
 188                   else if (fulldecl->Semantic.Name == TGSI_SEMANTIC_VERTEXID) {
 189                      info->uses_vertexid = TRUE;
 190                   } else if (fulldecl->Semantic.Name == TGSI_SEMANTIC_PRIMID) {
 191                      info->uses_primid = TRUE;
 192                   }
 193                }
 194                else if (file == TGSI_FILE_OUTPUT) {
 195                   info->output_semantic_name[reg] = (ubyte)fulldecl->Semantic.Name;
 196                   info->output_semantic_index[reg] = (ubyte)fulldecl->Semantic.Index;
 197                   info->num_outputs++;
 198
 199                   if ((procType == TGSI_PROCESSOR_VERTEX || procType == TGSI_PROCESSOR_GEOMETRY) &&
 200                       fulldecl->Semantic.Name == TGSI_SEMANTIC_CLIPDIST) {
 201                      info->num_written_clipdistance += util_bitcount(fulldecl->Declaration.UsageMask);
 202                   }
 203                   if ((procType == TGSI_PROCESSOR_VERTEX || procType == TGSI_PROCESSOR_GEOMETRY) &&
 204                       fulldecl->Semantic.Name == TGSI_SEMANTIC_CULLDIST) {
 205                      info->num_written_culldistance += util_bitcount(fulldecl->Declaration.UsageMask);
 206                   }
 207                   /* extra info for special outputs */
 208                   if (procType == TGSI_PROCESSOR_FRAGMENT &&
 209                       fulldecl->Semantic.Name == TGSI_SEMANTIC_POSITION)
 210                         info->writes_z = TRUE;
 211                   if (procType == TGSI_PROCESSOR_FRAGMENT &&
 212                       fulldecl->Semantic.Name == TGSI_SEMANTIC_STENCIL)
 213                         info->writes_stencil = TRUE;
 214                   if (procType == TGSI_PROCESSOR_VERTEX &&
 215                       fulldecl->Semantic.Name == TGSI_SEMANTIC_EDGEFLAG) {
 216                      info->writes_edgeflag = TRUE;
 217                   }
 218
 219                   if (procType == TGSI_PROCESSOR_GEOMETRY &&
 220                       fulldecl->Semantic.Name ==
 221                       TGSI_SEMANTIC_VIEWPORT_INDEX) {
 222                      info->writes_viewport_index = TRUE;
 223                   }
 224                   if (procType == TGSI_PROCESSOR_GEOMETRY &&
 225                       fulldecl->Semantic.Name ==
 226                       TGSI_SEMANTIC_LAYER) {
 227                      info->writes_layer = TRUE;
 228                   }
 229                }
 230
 231              }
 232          }
 233          break;
 234
 235       case TGSI_TOKEN_TYPE_IMMEDIATE:
 236          {
 237             uint reg = info->immediate_count++;
 238             uint file = TGSI_FILE_IMMEDIATE;
 239
 240             info->file_mask[file] |= (1 << reg);
 241             info->file_count[file]++;
 242             info->file_max[file] = MAX2(info->file_max[file], (int)reg);
 243          }
 244          break;
 245
 246       case TGSI_TOKEN_TYPE_PROPERTY:
 247          {
 248             const struct tgsi_full_property *fullprop
 249                = &parse.FullToken.FullProperty;
 250
 251             info->properties[info->num_properties].name =
 252                fullprop->Property.PropertyName;
 253             memcpy(info->properties[info->num_properties].data,
 254                    fullprop->u, 8 * sizeof(unsigned));;
 255
 256             ++info->num_properties;
 257          }
 258          break;
 259
 260       default:
 261          assert( 0 );
 262       }
 263    }
 264
 265    info->uses_kill = (info->opcode_count[TGSI_OPCODE_KIL] ||
 266                       info->opcode_count[TGSI_OPCODE_KILP]);
 267
 268    /* extract simple properties */
 269    for (i = 0; i < info->num_properties; ++i) {
 270       switch (info->properties[i].name) {
 271       case TGSI_PROPERTY_FS_COORD_ORIGIN:
 272          info->origin_lower_left = info->properties[i].data[0];
 273          break;
 274       case TGSI_PROPERTY_FS_COORD_PIXEL_CENTER:
 275          info->pixel_center_integer = info->properties[i].data[0];
 276          break;
 277       case TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS:
 278          info->color0_writes_all_cbufs = info->properties[i].data[0];
 279          break;
 280       case TGSI_PROPERTY_GS_INPUT_PRIM:
 281          /* The dimensions of the IN decleration in geometry shader have
 282           * to be deduced from the type of the input primitive.
 283           */
 284          if (procType == TGSI_PROCESSOR_GEOMETRY) {
 285             unsigned input_primitive = info->properties[i].data[0];
 286             int num_verts = u_vertices_per_prim(input_primitive);
 287             unsigned j;
 288             info->file_count[TGSI_FILE_INPUT] = num_verts;
 289             info->file_max[TGSI_FILE_INPUT] =
 290                MAX2(info->file_max[TGSI_FILE_INPUT], num_verts - 1);
 291             for (j = 0; j < num_verts; ++j) {
 292                info->file_mask[TGSI_FILE_INPUT] |= (1 << j);
 293             }
 294          }
 295          break;
 296       default:
 297          ;
 298       }
 299    }
 300
 301    tgsi_parse_free (&parse);
 302 }
 303
 304
 305
 306 /**
 307  * Check if the given shader is a "passthrough" shader consisting of only
 308  * MOV instructions of the form:  MOV OUT[n], IN[n]
 309  *
 310  */
 311 boolean
 312 tgsi_is_passthrough_shader(const struct tgsi_token *tokens)
 313 {
 314    struct tgsi_parse_context parse;
 315
 316    /**
 317     ** Setup to begin parsing input shader
 318     **/
 319    if (tgsi_parse_init(&parse, tokens) != TGSI_PARSE_OK) {
 320       debug_printf("tgsi_parse_init() failed in tgsi_is_passthrough_shader()!\n");
 321       return FALSE;
 322    }
 323
 324    /**
 325     ** Loop over incoming program tokens/instructions
 326     */
 327    while (!tgsi_parse_end_of_tokens(&parse)) {
 328
 329       tgsi_parse_token(&parse);
 330
 331       switch (parse.FullToken.Token.Type) {
 332       case TGSI_TOKEN_TYPE_INSTRUCTION:
 333          {
 334             struct tgsi_full_instruction *fullinst =
 335                &parse.FullToken.FullInstruction;
 336             const struct tgsi_full_src_register *src =
 337                &fullinst->Src[0];
 338             const struct tgsi_full_dst_register *dst =
 339                &fullinst->Dst[0];
 340
 341             /* Do a whole bunch of checks for a simple move */
 342             if (fullinst->Instruction.Opcode != TGSI_OPCODE_MOV ||
 343                 (src->Register.File != TGSI_FILE_INPUT &&
 344                  src->Register.File != TGSI_FILE_SYSTEM_VALUE) ||
 345                 dst->Register.File != TGSI_FILE_OUTPUT ||
 346                 src->Register.Index != dst->Register.Index ||
 347
 348                 src->Register.Negate ||
 349                 src->Register.Absolute ||
 350
 351                 src->Register.SwizzleX != TGSI_SWIZZLE_X ||
 352                 src->Register.SwizzleY != TGSI_SWIZZLE_Y ||
 353                 src->Register.SwizzleZ != TGSI_SWIZZLE_Z ||
 354                 src->Register.SwizzleW != TGSI_SWIZZLE_W ||
 355
 356                 dst->Register.WriteMask != TGSI_WRITEMASK_XYZW)
 357             {
 358                tgsi_parse_free(&parse);
 359                return FALSE;
 360             }
 361          }
 362          break;
 363
 364       case TGSI_TOKEN_TYPE_DECLARATION:
 365          /* fall-through */
 366       case TGSI_TOKEN_TYPE_IMMEDIATE:
 367          /* fall-through */
 368       case TGSI_TOKEN_TYPE_PROPERTY:
 369          /* fall-through */
 370       default:
 371          ; /* no-op */
 372       }
 373    }
 374
 375    tgsi_parse_free(&parse);
 376
 377    /* if we get here, it's a pass-through shader */
 378    return TRUE;
 379 }