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    info->processor = procType;
  76
  77
  78    /**
  79     ** Loop over incoming program tokens/instructions
  80     */
  81    while( !tgsi_parse_end_of_tokens( &parse ) ) {
  82
  83       info->num_tokens++;
  84
  85       tgsi_parse_token( &parse );
  86
  87       switch( parse.FullToken.Token.Type ) {
  88       case TGSI_TOKEN_TYPE_INSTRUCTION:
  89          {
  90             const struct tgsi_full_instruction *fullinst
  91                = &parse.FullToken.FullInstruction;
  92             uint i;
  93
  94             assert(fullinst->Instruction.Opcode < TGSI_OPCODE_LAST);
  95             info->opcode_count[fullinst->Instruction.Opcode]++;
  96
  97             for (i = 0; i < fullinst->Instruction.NumSrcRegs; i++) {
  98                const struct tgsi_full_src_register *src =
  99                   &fullinst->Src[i];
 100                int ind = src->Register.Index;
 101
 102                /* Mark which inputs are effectively used */
 103                if (src->Register.File == TGSI_FILE_INPUT) {
 104                   unsigned usage_mask;
 105                   usage_mask = tgsi_util_get_inst_usage_mask(fullinst, i);
 106                   if (src->Register.Indirect) {
 107                      for (ind = 0; ind < info->num_inputs; ++ind) {
 108                         info->input_usage_mask[ind] |= usage_mask;
 109                      }
 110                   } else {
 111                      assert(ind >= 0);
 112                      assert(ind < PIPE_MAX_SHADER_INPUTS);
 113                      info->input_usage_mask[ind] |= usage_mask;
 114                   }
 115
 116                   if (procType == TGSI_PROCESSOR_FRAGMENT &&
 117                       src->Register.File == TGSI_FILE_INPUT &&
 118                       info->reads_position &&
 119                       src->Register.Index == 0 &&
 120                       (src->Register.SwizzleX == TGSI_SWIZZLE_Z ||
 121                        src->Register.SwizzleY == TGSI_SWIZZLE_Z ||
 122                        src->Register.SwizzleZ == TGSI_SWIZZLE_Z ||
 123                        src->Register.SwizzleW == TGSI_SWIZZLE_Z)) {
 124                      info->reads_z = TRUE;
 125                   }
 126                }
 127
 128                /* check for indirect register reads */
 129                if (src->Register.Indirect) {
 130                   info->indirect_files |= (1 << src->Register.File);
 131                }
 132             }
 133
 134             /* check for indirect register writes */
 135             for (i = 0; i < fullinst->Instruction.NumDstRegs; i++) {
 136                const struct tgsi_full_dst_register *dst = &fullinst->Dst[i];
 137                if (dst->Register.Indirect) {
 138                   info->indirect_files |= (1 << dst->Register.File);
 139                }
 140             }
 141
 142             info->num_instructions++;
 143          }
 144          break;
 145
 146       case TGSI_TOKEN_TYPE_DECLARATION:
 147          {
 148             const struct tgsi_full_declaration *fulldecl
 149                = &parse.FullToken.FullDeclaration;
 150             const uint file = fulldecl->Declaration.File;
 151             uint reg;
 152             for (reg = fulldecl->Range.First;
 153                  reg <= fulldecl->Range.Last;
 154                  reg++) {
 155
 156                /* only first 32 regs will appear in this bitfield */
 157                info->file_mask[file] |= (1 << reg);
 158                info->file_count[file]++;
 159                info->file_max[file] = MAX2(info->file_max[file], (int)reg);
 160
 161                if (file == TGSI_FILE_INPUT) {
 162                   info->input_semantic_name[reg] = (ubyte)fulldecl->Semantic.Name;
 163                   info->input_semantic_index[reg] = (ubyte)fulldecl->Semantic.Index;
 164                   info->input_interpolate[reg] = (ubyte)fulldecl->Interp.Interpolate;
 165                   info->input_centroid[reg] = (ubyte)fulldecl->Interp.Centroid;
 166                   info->input_cylindrical_wrap[reg] = (ubyte)fulldecl->Interp.CylindricalWrap;
 167                   info->num_inputs++;
 168
 169                   if (procType == TGSI_PROCESSOR_FRAGMENT &&
 170                       fulldecl->Semantic.Name == TGSI_SEMANTIC_POSITION)
 171                         info->reads_position = TRUE;
 172                }
 173                else if (file == TGSI_FILE_SYSTEM_VALUE) {
 174                   unsigned index = fulldecl->Range.First;
 175                   unsigned semName = fulldecl->Semantic.Name;
 176
 177                   info->system_value_semantic_name[index] = semName;
 178                   info->num_system_values = MAX2(info->num_system_values,
 179                                                  index + 1);
 180
 181                   /*
 182                   info->system_value_semantic_name[info->num_system_values++] =
 183                      fulldecl->Semantic.Name;
 184                   */
 185
 186                   if (fulldecl->Semantic.Name == TGSI_SEMANTIC_INSTANCEID) {
 187                      info->uses_instanceid = TRUE;
 188                   }
 189                   else if (fulldecl->Semantic.Name == TGSI_SEMANTIC_VERTEXID) {
 190                      info->uses_vertexid = TRUE;
 191                   } else if (fulldecl->Semantic.Name == TGSI_SEMANTIC_PRIMID) {
 192                      info->uses_primid = TRUE;
 193                   }
 194                }
 195                else if (file == TGSI_FILE_OUTPUT) {
 196                   info->output_semantic_name[reg] = (ubyte)fulldecl->Semantic.Name;
 197                   info->output_semantic_index[reg] = (ubyte)fulldecl->Semantic.Index;
 198                   info->num_outputs++;
 199
 200                   if ((procType == TGSI_PROCESSOR_VERTEX || procType == TGSI_PROCESSOR_GEOMETRY) &&
 201                       fulldecl->Semantic.Name == TGSI_SEMANTIC_CLIPDIST) {
 202                      info->num_written_clipdistance += util_bitcount(fulldecl->Declaration.UsageMask);
 203                   }
 204                   if ((procType == TGSI_PROCESSOR_VERTEX || procType == TGSI_PROCESSOR_GEOMETRY) &&
 205                       fulldecl->Semantic.Name == TGSI_SEMANTIC_CULLDIST) {
 206                      info->num_written_culldistance += util_bitcount(fulldecl->Declaration.UsageMask);
 207                   }
 208                   /* extra info for special outputs */
 209                   if (procType == TGSI_PROCESSOR_FRAGMENT &&
 210                       fulldecl->Semantic.Name == TGSI_SEMANTIC_POSITION)
 211                         info->writes_z = TRUE;
 212                   if (procType == TGSI_PROCESSOR_FRAGMENT &&
 213                       fulldecl->Semantic.Name == TGSI_SEMANTIC_STENCIL)
 214                         info->writes_stencil = TRUE;
 215                   if (procType == TGSI_PROCESSOR_VERTEX &&
 216                       fulldecl->Semantic.Name == TGSI_SEMANTIC_EDGEFLAG) {
 217                      info->writes_edgeflag = TRUE;
 218                   }
 219
 220                   if (procType == TGSI_PROCESSOR_GEOMETRY &&
 221                       fulldecl->Semantic.Name ==
 222                       TGSI_SEMANTIC_VIEWPORT_INDEX) {
 223                      info->writes_viewport_index = TRUE;
 224                   }
 225                   if (procType == TGSI_PROCESSOR_GEOMETRY &&
 226                       fulldecl->Semantic.Name ==
 227                       TGSI_SEMANTIC_LAYER) {
 228                      info->writes_layer = TRUE;
 229                   }
 230                }
 231
 232              }
 233          }
 234          break;
 235
 236       case TGSI_TOKEN_TYPE_IMMEDIATE:
 237          {
 238             uint reg = info->immediate_count++;
 239             uint file = TGSI_FILE_IMMEDIATE;
 240
 241             info->file_mask[file] |= (1 << reg);
 242             info->file_count[file]++;
 243             info->file_max[file] = MAX2(info->file_max[file], (int)reg);
 244          }
 245          break;
 246
 247       case TGSI_TOKEN_TYPE_PROPERTY:
 248          {
 249             const struct tgsi_full_property *fullprop
 250                = &parse.FullToken.FullProperty;
 251
 252             info->properties[info->num_properties].name =
 253                fullprop->Property.PropertyName;
 254             memcpy(info->properties[info->num_properties].data,
 255                    fullprop->u, 8 * sizeof(unsigned));;
 256
 257             ++info->num_properties;
 258          }
 259          break;
 260
 261       default:
 262          assert( 0 );
 263       }
 264    }
 265
 266    info->uses_kill = (info->opcode_count[TGSI_OPCODE_KILL_IF] ||
 267                       info->opcode_count[TGSI_OPCODE_KILL]);
 268
 269    /* extract simple properties */
 270    for (i = 0; i < info->num_properties; ++i) {
 271       switch (info->properties[i].name) {
 272       case TGSI_PROPERTY_FS_COORD_ORIGIN:
 273          info->origin_lower_left = info->properties[i].data[0];
 274          break;
 275       case TGSI_PROPERTY_FS_COORD_PIXEL_CENTER:
 276          info->pixel_center_integer = info->properties[i].data[0];
 277          break;
 278       case TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS:
 279          info->color0_writes_all_cbufs = info->properties[i].data[0];
 280          break;
 281       case TGSI_PROPERTY_GS_INPUT_PRIM:
 282          /* The dimensions of the IN decleration in geometry shader have
 283           * to be deduced from the type of the input primitive.
 284           */
 285          if (procType == TGSI_PROCESSOR_GEOMETRY) {
 286             unsigned input_primitive = info->properties[i].data[0];
 287             int num_verts = u_vertices_per_prim(input_primitive);
 288             int j;
 289             info->file_count[TGSI_FILE_INPUT] = num_verts;
 290             info->file_max[TGSI_FILE_INPUT] =
 291                MAX2(info->file_max[TGSI_FILE_INPUT], num_verts - 1);
 292             for (j = 0; j < num_verts; ++j) {
 293                info->file_mask[TGSI_FILE_INPUT] |= (1 << j);
 294             }
 295          }
 296          break;
 297       default:
 298          ;
 299       }
 300    }
 301
 302    tgsi_parse_free (&parse);
 303 }
 304
 305
 306
 307 /**
 308  * Check if the given shader is a "passthrough" shader consisting of only
 309  * MOV instructions of the form:  MOV OUT[n], IN[n]
 310  *
 311  */
 312 boolean
 313 tgsi_is_passthrough_shader(const struct tgsi_token *tokens)
 314 {
 315    struct tgsi_parse_context parse;
 316
 317    /**
 318     ** Setup to begin parsing input shader
 319     **/
 320    if (tgsi_parse_init(&parse, tokens) != TGSI_PARSE_OK) {
 321       debug_printf("tgsi_parse_init() failed in tgsi_is_passthrough_shader()!\n");
 322       return FALSE;
 323    }
 324
 325    /**
 326     ** Loop over incoming program tokens/instructions
 327     */
 328    while (!tgsi_parse_end_of_tokens(&parse)) {
 329
 330       tgsi_parse_token(&parse);
 331
 332       switch (parse.FullToken.Token.Type) {
 333       case TGSI_TOKEN_TYPE_INSTRUCTION:
 334          {
 335             struct tgsi_full_instruction *fullinst =
 336                &parse.FullToken.FullInstruction;
 337             const struct tgsi_full_src_register *src =
 338                &fullinst->Src[0];
 339             const struct tgsi_full_dst_register *dst =
 340                &fullinst->Dst[0];
 341
 342             /* Do a whole bunch of checks for a simple move */
 343             if (fullinst->Instruction.Opcode != TGSI_OPCODE_MOV ||
 344                 (src->Register.File != TGSI_FILE_INPUT &&
 345                  src->Register.File != TGSI_FILE_SYSTEM_VALUE) ||
 346                 dst->Register.File != TGSI_FILE_OUTPUT ||
 347                 src->Register.Index != dst->Register.Index ||
 348
 349                 src->Register.Negate ||
 350                 src->Register.Absolute ||
 351
 352                 src->Register.SwizzleX != TGSI_SWIZZLE_X ||
 353                 src->Register.SwizzleY != TGSI_SWIZZLE_Y ||
 354                 src->Register.SwizzleZ != TGSI_SWIZZLE_Z ||
 355                 src->Register.SwizzleW != TGSI_SWIZZLE_W ||
 356
 357                 dst->Register.WriteMask != TGSI_WRITEMASK_XYZW)
 358             {
 359                tgsi_parse_free(&parse);
 360                return FALSE;
 361             }
 362          }
 363          break;
 364
 365       case TGSI_TOKEN_TYPE_DECLARATION:
 366          /* fall-through */
 367       case TGSI_TOKEN_TYPE_IMMEDIATE:
 368          /* fall-through */
 369       case TGSI_TOKEN_TYPE_PROPERTY:
 370          /* fall-through */
 371       default:
 372          ; /* no-op */
 373       }
 374    }
 375
 376    tgsi_parse_free(&parse);
 377
 378    /* if we get here, it's a pass-through shader */
 379    return TRUE;
 380 }