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
  64    /**
  65     ** Setup to begin parsing input shader
  66     **/
  67    if (tgsi_parse_init( &parse, tokens ) != TGSI_PARSE_OK) {
  68       debug_printf("tgsi_parse_init() failed in tgsi_scan_shader()!\n");
  69       return;
  70    }
  71    procType = parse.FullHeader.Processor.Processor;
  72    assert(procType == TGSI_PROCESSOR_FRAGMENT ||
  73           procType == TGSI_PROCESSOR_VERTEX ||
  74           procType == TGSI_PROCESSOR_GEOMETRY ||
  75           procType == TGSI_PROCESSOR_COMPUTE);
  76    info->processor = procType;
  77
  78
  79    /**
  80     ** Loop over incoming program tokens/instructions
  81     */
  82    while( !tgsi_parse_end_of_tokens( &parse ) ) {
  83
  84       info->num_tokens++;
  85
  86       tgsi_parse_token( &parse );
  87
  88       switch( parse.FullToken.Token.Type ) {
  89       case TGSI_TOKEN_TYPE_INSTRUCTION:
  90          {
  91             const struct tgsi_full_instruction *fullinst
  92                = &parse.FullToken.FullInstruction;
  93             uint i;
  94
  95             assert(fullinst->Instruction.Opcode < TGSI_OPCODE_LAST);
  96             info->opcode_count[fullinst->Instruction.Opcode]++;
  97
  98             for (i = 0; i < fullinst->Instruction.NumSrcRegs; i++) {
  99                const struct tgsi_full_src_register *src =
 100                   &fullinst->Src[i];
 101                int ind = src->Register.Index;
 102
 103                /* Mark which inputs are effectively used */
 104                if (src->Register.File == TGSI_FILE_INPUT) {
 105                   unsigned usage_mask;
 106                   usage_mask = tgsi_util_get_inst_usage_mask(fullinst, i);
 107                   if (src->Register.Indirect) {
 108                      for (ind = 0; ind < info->num_inputs; ++ind) {
 109                         info->input_usage_mask[ind] |= usage_mask;
 110                      }
 111                   } else {
 112                      assert(ind >= 0);
 113                      assert(ind < PIPE_MAX_SHADER_INPUTS);
 114                      info->input_usage_mask[ind] |= usage_mask;
 115                   }
 116
 117                   if (procType == TGSI_PROCESSOR_FRAGMENT &&
 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                /* MSAA samplers */
 134                if (src->Register.File == TGSI_FILE_SAMPLER) {
 135                   assert(fullinst->Instruction.Texture);
 136                   assert(src->Register.Index < Elements(info->is_msaa_sampler));
 137
 138                   if (fullinst->Instruction.Texture &&
 139                       (fullinst->Texture.Texture == TGSI_TEXTURE_2D_MSAA ||
 140                        fullinst->Texture.Texture == TGSI_TEXTURE_2D_ARRAY_MSAA)) {
 141                      info->is_msaa_sampler[src->Register.Index] = TRUE;
 142                   }
 143                }
 144             }
 145
 146             /* check for indirect register writes */
 147             for (i = 0; i < fullinst->Instruction.NumDstRegs; i++) {
 148                const struct tgsi_full_dst_register *dst = &fullinst->Dst[i];
 149                if (dst->Register.Indirect) {
 150                   info->indirect_files |= (1 << dst->Register.File);
 151                }
 152             }
 153
 154             info->num_instructions++;
 155          }
 156          break;
 157
 158       case TGSI_TOKEN_TYPE_DECLARATION:
 159          {
 160             const struct tgsi_full_declaration *fulldecl
 161                = &parse.FullToken.FullDeclaration;
 162             const uint file = fulldecl->Declaration.File;
 163             uint reg;
 164             for (reg = fulldecl->Range.First;
 165                  reg <= fulldecl->Range.Last;
 166                  reg++) {
 167                unsigned semName = fulldecl->Semantic.Name;
 168                unsigned semIndex = fulldecl->Semantic.Index;
 169
 170                /* only first 32 regs will appear in this bitfield */
 171                info->file_mask[file] |= (1 << reg);
 172                info->file_count[file]++;
 173                info->file_max[file] = MAX2(info->file_max[file], (int)reg);
 174
 175                if (file == TGSI_FILE_INPUT) {
 176                   info->input_semantic_name[reg] = (ubyte) semName;
 177                   info->input_semantic_index[reg] = (ubyte) semIndex;
 178                   info->input_interpolate[reg] = (ubyte)fulldecl->Interp.Interpolate;
 179                   info->input_centroid[reg] = (ubyte)fulldecl->Interp.Centroid;
 180                   info->input_cylindrical_wrap[reg] = (ubyte)fulldecl->Interp.CylindricalWrap;
 181                   info->num_inputs++;
 182
 183                   if (procType == TGSI_PROCESSOR_FRAGMENT) {
 184                      if (semName == TGSI_SEMANTIC_POSITION)
 185                         info->reads_position = TRUE;
 186                      else if (semName == TGSI_SEMANTIC_PRIMID)
 187                         info->uses_primid = TRUE;
 188                      else if (semName == TGSI_SEMANTIC_FACE)
 189                         info->uses_frontface = TRUE;
 190                   }
 191                }
 192                else if (file == TGSI_FILE_SYSTEM_VALUE) {
 193                   unsigned index = fulldecl->Range.First;
 194
 195                   info->system_value_semantic_name[index] = semName;
 196                   info->num_system_values = MAX2(info->num_system_values,
 197                                                  index + 1);
 198
 199                   if (semName == TGSI_SEMANTIC_INSTANCEID) {
 200                      info->uses_instanceid = TRUE;
 201                   }
 202                   else if (semName == TGSI_SEMANTIC_VERTEXID) {
 203                      info->uses_vertexid = TRUE;
 204                   }
 205                   else if (semName == TGSI_SEMANTIC_PRIMID) {
 206                      info->uses_primid = TRUE;
 207                   }
 208                }
 209                else if (file == TGSI_FILE_OUTPUT) {
 210                   info->output_semantic_name[reg] = (ubyte) semName;
 211                   info->output_semantic_index[reg] = (ubyte) semIndex;
 212                   info->num_outputs++;
 213
 214                   if (procType == TGSI_PROCESSOR_VERTEX ||
 215                       procType == TGSI_PROCESSOR_GEOMETRY) {
 216                      if (semName == TGSI_SEMANTIC_CLIPDIST) {
 217                         info->num_written_clipdistance +=
 218                            util_bitcount(fulldecl->Declaration.UsageMask);
 219                      }
 220                      else if (semName == TGSI_SEMANTIC_CULLDIST) {
 221                         info->num_written_culldistance +=
 222                            util_bitcount(fulldecl->Declaration.UsageMask);
 223                      }
 224                   }
 225
 226                   if (procType == TGSI_PROCESSOR_FRAGMENT) {
 227                      if (semName == TGSI_SEMANTIC_POSITION) {
 228                         info->writes_z = TRUE;
 229                      }
 230                      else if (semName == TGSI_SEMANTIC_STENCIL) {
 231                         info->writes_stencil = TRUE;
 232                      }
 233                   }
 234
 235                   if (procType == TGSI_PROCESSOR_VERTEX) {
 236                      if (semName == TGSI_SEMANTIC_EDGEFLAG) {
 237                         info->writes_edgeflag = TRUE;
 238                      }
 239                   }
 240
 241                   if (procType == TGSI_PROCESSOR_GEOMETRY) {
 242                      if (semName == TGSI_SEMANTIC_VIEWPORT_INDEX) {
 243                         info->writes_viewport_index = TRUE;
 244                      }
 245                      else if (semName == TGSI_SEMANTIC_LAYER) {
 246                         info->writes_layer = TRUE;
 247                      }
 248                   }
 249                }
 250             }
 251          }
 252          break;
 253
 254       case TGSI_TOKEN_TYPE_IMMEDIATE:
 255          {
 256             uint reg = info->immediate_count++;
 257             uint file = TGSI_FILE_IMMEDIATE;
 258
 259             info->file_mask[file] |= (1 << reg);
 260             info->file_count[file]++;
 261             info->file_max[file] = MAX2(info->file_max[file], (int)reg);
 262          }
 263          break;
 264
 265       case TGSI_TOKEN_TYPE_PROPERTY:
 266          {
 267             const struct tgsi_full_property *fullprop
 268                = &parse.FullToken.FullProperty;
 269
 270             info->properties[info->num_properties].name =
 271                fullprop->Property.PropertyName;
 272             memcpy(info->properties[info->num_properties].data,
 273                    fullprop->u, 8 * sizeof(unsigned));;
 274
 275             ++info->num_properties;
 276          }
 277          break;
 278
 279       default:
 280          assert( 0 );
 281       }
 282    }
 283
 284    info->uses_kill = (info->opcode_count[TGSI_OPCODE_KILL_IF] ||
 285                       info->opcode_count[TGSI_OPCODE_KILL]);
 286
 287    /* extract simple properties */
 288    for (i = 0; i < info->num_properties; ++i) {
 289       switch (info->properties[i].name) {
 290       case TGSI_PROPERTY_FS_COORD_ORIGIN:
 291          info->origin_lower_left = info->properties[i].data[0];
 292          break;
 293       case TGSI_PROPERTY_FS_COORD_PIXEL_CENTER:
 294          info->pixel_center_integer = info->properties[i].data[0];
 295          break;
 296       case TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS:
 297          info->color0_writes_all_cbufs = info->properties[i].data[0];
 298          break;
 299       case TGSI_PROPERTY_GS_INPUT_PRIM:
 300          /* The dimensions of the IN decleration in geometry shader have
 301           * to be deduced from the type of the input primitive.
 302           */
 303          if (procType == TGSI_PROCESSOR_GEOMETRY) {
 304             unsigned input_primitive = info->properties[i].data[0];
 305             int num_verts = u_vertices_per_prim(input_primitive);
 306             int j;
 307             info->file_count[TGSI_FILE_INPUT] = num_verts;
 308             info->file_max[TGSI_FILE_INPUT] =
 309                MAX2(info->file_max[TGSI_FILE_INPUT], num_verts - 1);
 310             for (j = 0; j < num_verts; ++j) {
 311                info->file_mask[TGSI_FILE_INPUT] |= (1 << j);
 312             }
 313          }
 314          break;
 315       default:
 316          ;
 317       }
 318    }
 319
 320    tgsi_parse_free (&parse);
 321 }
 322
 323
 324
 325 /**
 326  * Check if the given shader is a "passthrough" shader consisting of only
 327  * MOV instructions of the form:  MOV OUT[n], IN[n]
 328  *
 329  */
 330 boolean
 331 tgsi_is_passthrough_shader(const struct tgsi_token *tokens)
 332 {
 333    struct tgsi_parse_context parse;
 334
 335    /**
 336     ** Setup to begin parsing input shader
 337     **/
 338    if (tgsi_parse_init(&parse, tokens) != TGSI_PARSE_OK) {
 339       debug_printf("tgsi_parse_init() failed in tgsi_is_passthrough_shader()!\n");
 340       return FALSE;
 341    }
 342
 343    /**
 344     ** Loop over incoming program tokens/instructions
 345     */
 346    while (!tgsi_parse_end_of_tokens(&parse)) {
 347
 348       tgsi_parse_token(&parse);
 349
 350       switch (parse.FullToken.Token.Type) {
 351       case TGSI_TOKEN_TYPE_INSTRUCTION:
 352          {
 353             struct tgsi_full_instruction *fullinst =
 354                &parse.FullToken.FullInstruction;
 355             const struct tgsi_full_src_register *src =
 356                &fullinst->Src[0];
 357             const struct tgsi_full_dst_register *dst =
 358                &fullinst->Dst[0];
 359
 360             /* Do a whole bunch of checks for a simple move */
 361             if (fullinst->Instruction.Opcode != TGSI_OPCODE_MOV ||
 362                 (src->Register.File != TGSI_FILE_INPUT &&
 363                  src->Register.File != TGSI_FILE_SYSTEM_VALUE) ||
 364                 dst->Register.File != TGSI_FILE_OUTPUT ||
 365                 src->Register.Index != dst->Register.Index ||
 366
 367                 src->Register.Negate ||
 368                 src->Register.Absolute ||
 369
 370                 src->Register.SwizzleX != TGSI_SWIZZLE_X ||
 371                 src->Register.SwizzleY != TGSI_SWIZZLE_Y ||
 372                 src->Register.SwizzleZ != TGSI_SWIZZLE_Z ||
 373                 src->Register.SwizzleW != TGSI_SWIZZLE_W ||
 374
 375                 dst->Register.WriteMask != TGSI_WRITEMASK_XYZW)
 376             {
 377                tgsi_parse_free(&parse);
 378                return FALSE;
 379             }
 380          }
 381          break;
 382
 383       case TGSI_TOKEN_TYPE_DECLARATION:
 384          /* fall-through */
 385       case TGSI_TOKEN_TYPE_IMMEDIATE:
 386          /* fall-through */
 387       case TGSI_TOKEN_TYPE_PROPERTY:
 388          /* fall-through */
 389       default:
 390          ; /* no-op */
 391       }
 392    }
 393
 394    tgsi_parse_free(&parse);
 395
 396    /* if we get here, it's a pass-through shader */
 397    return TRUE;
 398 }