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