Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
5564 | serge | 1 | /************************************************************************** |
2 | * |
||
3 | * Copyright 2008 VMware, Inc. |
||
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 VMWARE 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_TESSCTRL || |
||
78 | procType == TGSI_PROCESSOR_TESSEVAL || |
||
79 | procType == TGSI_PROCESSOR_COMPUTE); |
||
80 | info->processor = procType; |
||
81 | |||
82 | |||
83 | /** |
||
84 | ** Loop over incoming program tokens/instructions |
||
85 | */ |
||
86 | while( !tgsi_parse_end_of_tokens( &parse ) ) { |
||
87 | |||
88 | info->num_tokens++; |
||
89 | |||
90 | tgsi_parse_token( &parse ); |
||
91 | |||
92 | switch( parse.FullToken.Token.Type ) { |
||
93 | case TGSI_TOKEN_TYPE_INSTRUCTION: |
||
94 | { |
||
95 | const struct tgsi_full_instruction *fullinst |
||
96 | = &parse.FullToken.FullInstruction; |
||
97 | uint i; |
||
98 | |||
99 | assert(fullinst->Instruction.Opcode < TGSI_OPCODE_LAST); |
||
100 | info->opcode_count[fullinst->Instruction.Opcode]++; |
||
101 | |||
102 | for (i = 0; i < fullinst->Instruction.NumSrcRegs; i++) { |
||
103 | const struct tgsi_full_src_register *src = |
||
104 | &fullinst->Src[i]; |
||
105 | int ind = src->Register.Index; |
||
106 | |||
107 | /* Mark which inputs are effectively used */ |
||
108 | if (src->Register.File == TGSI_FILE_INPUT) { |
||
109 | unsigned usage_mask; |
||
110 | usage_mask = tgsi_util_get_inst_usage_mask(fullinst, i); |
||
111 | if (src->Register.Indirect) { |
||
112 | for (ind = 0; ind < info->num_inputs; ++ind) { |
||
113 | info->input_usage_mask[ind] |= usage_mask; |
||
114 | } |
||
115 | } else { |
||
116 | assert(ind >= 0); |
||
117 | assert(ind < PIPE_MAX_SHADER_INPUTS); |
||
118 | info->input_usage_mask[ind] |= usage_mask; |
||
119 | } |
||
120 | |||
121 | if (procType == TGSI_PROCESSOR_FRAGMENT && |
||
122 | info->reads_position && |
||
123 | src->Register.Index == 0 && |
||
124 | (src->Register.SwizzleX == TGSI_SWIZZLE_Z || |
||
125 | src->Register.SwizzleY == TGSI_SWIZZLE_Z || |
||
126 | src->Register.SwizzleZ == TGSI_SWIZZLE_Z || |
||
127 | src->Register.SwizzleW == TGSI_SWIZZLE_Z)) { |
||
128 | info->reads_z = TRUE; |
||
129 | } |
||
130 | } |
||
131 | |||
132 | /* check for indirect register reads */ |
||
133 | if (src->Register.Indirect) { |
||
134 | info->indirect_files |= (1 << src->Register.File); |
||
135 | info->indirect_files_read |= (1 << src->Register.File); |
||
136 | } |
||
137 | |||
138 | /* MSAA samplers */ |
||
139 | if (src->Register.File == TGSI_FILE_SAMPLER) { |
||
140 | assert(fullinst->Instruction.Texture); |
||
141 | assert(src->Register.Index < Elements(info->is_msaa_sampler)); |
||
142 | |||
143 | if (fullinst->Instruction.Texture && |
||
144 | (fullinst->Texture.Texture == TGSI_TEXTURE_2D_MSAA || |
||
145 | fullinst->Texture.Texture == TGSI_TEXTURE_2D_ARRAY_MSAA)) { |
||
146 | info->is_msaa_sampler[src->Register.Index] = TRUE; |
||
147 | } |
||
148 | } |
||
149 | } |
||
150 | |||
151 | /* check for indirect register writes */ |
||
152 | for (i = 0; i < fullinst->Instruction.NumDstRegs; i++) { |
||
153 | const struct tgsi_full_dst_register *dst = &fullinst->Dst[i]; |
||
154 | if (dst->Register.Indirect) { |
||
155 | info->indirect_files |= (1 << dst->Register.File); |
||
156 | info->indirect_files_written |= (1 << dst->Register.File); |
||
157 | } |
||
158 | } |
||
159 | |||
160 | info->num_instructions++; |
||
161 | } |
||
162 | break; |
||
163 | |||
164 | case TGSI_TOKEN_TYPE_DECLARATION: |
||
165 | { |
||
166 | const struct tgsi_full_declaration *fulldecl |
||
167 | = &parse.FullToken.FullDeclaration; |
||
168 | const uint file = fulldecl->Declaration.File; |
||
169 | uint reg; |
||
170 | if (fulldecl->Declaration.Array) |
||
171 | info->array_max[file] = MAX2(info->array_max[file], fulldecl->Array.ArrayID); |
||
172 | for (reg = fulldecl->Range.First; |
||
173 | reg <= fulldecl->Range.Last; |
||
174 | reg++) { |
||
175 | unsigned semName = fulldecl->Semantic.Name; |
||
176 | unsigned semIndex = fulldecl->Semantic.Index; |
||
177 | |||
178 | /* only first 32 regs will appear in this bitfield */ |
||
179 | info->file_mask[file] |= (1 << reg); |
||
180 | info->file_count[file]++; |
||
181 | info->file_max[file] = MAX2(info->file_max[file], (int)reg); |
||
182 | |||
183 | if (file == TGSI_FILE_CONSTANT) { |
||
184 | int buffer = 0; |
||
185 | |||
186 | if (fulldecl->Declaration.Dimension) |
||
187 | buffer = fulldecl->Dim.Index2D; |
||
188 | |||
189 | info->const_file_max[buffer] = |
||
190 | MAX2(info->const_file_max[buffer], (int)reg); |
||
191 | } |
||
192 | else if (file == TGSI_FILE_INPUT) { |
||
193 | info->input_semantic_name[reg] = (ubyte) semName; |
||
194 | info->input_semantic_index[reg] = (ubyte) semIndex; |
||
195 | info->input_interpolate[reg] = (ubyte)fulldecl->Interp.Interpolate; |
||
196 | info->input_interpolate_loc[reg] = (ubyte)fulldecl->Interp.Location; |
||
197 | info->input_cylindrical_wrap[reg] = (ubyte)fulldecl->Interp.CylindricalWrap; |
||
198 | info->num_inputs++; |
||
199 | |||
200 | if (fulldecl->Interp.Location == TGSI_INTERPOLATE_LOC_CENTROID) |
||
201 | info->uses_centroid = TRUE; |
||
202 | |||
203 | if (semName == TGSI_SEMANTIC_PRIMID) |
||
204 | info->uses_primid = TRUE; |
||
205 | else if (procType == TGSI_PROCESSOR_FRAGMENT) { |
||
206 | if (semName == TGSI_SEMANTIC_POSITION) |
||
207 | info->reads_position = TRUE; |
||
208 | else if (semName == TGSI_SEMANTIC_FACE) |
||
209 | info->uses_frontface = TRUE; |
||
210 | } |
||
211 | } |
||
212 | else if (file == TGSI_FILE_SYSTEM_VALUE) { |
||
213 | unsigned index = fulldecl->Range.First; |
||
214 | |||
215 | info->system_value_semantic_name[index] = semName; |
||
216 | info->num_system_values = MAX2(info->num_system_values, |
||
217 | index + 1); |
||
218 | |||
219 | if (semName == TGSI_SEMANTIC_INSTANCEID) { |
||
220 | info->uses_instanceid = TRUE; |
||
221 | } |
||
222 | else if (semName == TGSI_SEMANTIC_VERTEXID) { |
||
223 | info->uses_vertexid = TRUE; |
||
224 | } |
||
225 | else if (semName == TGSI_SEMANTIC_VERTEXID_NOBASE) { |
||
226 | info->uses_vertexid_nobase = TRUE; |
||
227 | } |
||
228 | else if (semName == TGSI_SEMANTIC_BASEVERTEX) { |
||
229 | info->uses_basevertex = TRUE; |
||
230 | } |
||
231 | else if (semName == TGSI_SEMANTIC_PRIMID) { |
||
232 | info->uses_primid = TRUE; |
||
233 | } |
||
234 | } |
||
235 | else if (file == TGSI_FILE_OUTPUT) { |
||
236 | info->output_semantic_name[reg] = (ubyte) semName; |
||
237 | info->output_semantic_index[reg] = (ubyte) semIndex; |
||
238 | info->num_outputs++; |
||
239 | |||
240 | if (procType == TGSI_PROCESSOR_VERTEX || |
||
241 | procType == TGSI_PROCESSOR_GEOMETRY || |
||
242 | procType == TGSI_PROCESSOR_TESSCTRL || |
||
243 | procType == TGSI_PROCESSOR_TESSEVAL) { |
||
244 | if (semName == TGSI_SEMANTIC_CLIPDIST) { |
||
245 | info->num_written_clipdistance += |
||
246 | util_bitcount(fulldecl->Declaration.UsageMask); |
||
247 | info->clipdist_writemask |= |
||
248 | fulldecl->Declaration.UsageMask << (semIndex*4); |
||
249 | } |
||
250 | else if (semName == TGSI_SEMANTIC_CULLDIST) { |
||
251 | info->num_written_culldistance += |
||
252 | util_bitcount(fulldecl->Declaration.UsageMask); |
||
253 | info->culldist_writemask |= |
||
254 | fulldecl->Declaration.UsageMask << (semIndex*4); |
||
255 | } |
||
256 | else if (semName == TGSI_SEMANTIC_VIEWPORT_INDEX) { |
||
257 | info->writes_viewport_index = TRUE; |
||
258 | } |
||
259 | else if (semName == TGSI_SEMANTIC_LAYER) { |
||
260 | info->writes_layer = TRUE; |
||
261 | } |
||
262 | else if (semName == TGSI_SEMANTIC_PSIZE) { |
||
263 | info->writes_psize = TRUE; |
||
264 | } |
||
265 | else if (semName == TGSI_SEMANTIC_CLIPVERTEX) { |
||
266 | info->writes_clipvertex = TRUE; |
||
267 | } |
||
268 | } |
||
269 | |||
270 | if (procType == TGSI_PROCESSOR_FRAGMENT) { |
||
271 | if (semName == TGSI_SEMANTIC_POSITION) { |
||
272 | info->writes_z = TRUE; |
||
273 | } |
||
274 | else if (semName == TGSI_SEMANTIC_STENCIL) { |
||
275 | info->writes_stencil = TRUE; |
||
276 | } |
||
277 | } |
||
278 | |||
279 | if (procType == TGSI_PROCESSOR_VERTEX) { |
||
280 | if (semName == TGSI_SEMANTIC_EDGEFLAG) { |
||
281 | info->writes_edgeflag = TRUE; |
||
282 | } |
||
283 | } |
||
284 | } |
||
285 | } |
||
286 | } |
||
287 | break; |
||
288 | |||
289 | case TGSI_TOKEN_TYPE_IMMEDIATE: |
||
290 | { |
||
291 | uint reg = info->immediate_count++; |
||
292 | uint file = TGSI_FILE_IMMEDIATE; |
||
293 | |||
294 | info->file_mask[file] |= (1 << reg); |
||
295 | info->file_count[file]++; |
||
296 | info->file_max[file] = MAX2(info->file_max[file], (int)reg); |
||
297 | } |
||
298 | break; |
||
299 | |||
300 | case TGSI_TOKEN_TYPE_PROPERTY: |
||
301 | { |
||
302 | const struct tgsi_full_property *fullprop |
||
303 | = &parse.FullToken.FullProperty; |
||
304 | unsigned name = fullprop->Property.PropertyName; |
||
305 | |||
306 | assert(name < Elements(info->properties)); |
||
307 | info->properties[name] = fullprop->u[0].Data; |
||
308 | } |
||
309 | break; |
||
310 | |||
311 | default: |
||
312 | assert( 0 ); |
||
313 | } |
||
314 | } |
||
315 | |||
316 | info->uses_kill = (info->opcode_count[TGSI_OPCODE_KILL_IF] || |
||
317 | info->opcode_count[TGSI_OPCODE_KILL]); |
||
318 | |||
319 | /* The dimensions of the IN decleration in geometry shader have |
||
320 | * to be deduced from the type of the input primitive. |
||
321 | */ |
||
322 | if (procType == TGSI_PROCESSOR_GEOMETRY) { |
||
323 | unsigned input_primitive = |
||
324 | info->properties[TGSI_PROPERTY_GS_INPUT_PRIM]; |
||
325 | int num_verts = u_vertices_per_prim(input_primitive); |
||
326 | int j; |
||
327 | info->file_count[TGSI_FILE_INPUT] = num_verts; |
||
328 | info->file_max[TGSI_FILE_INPUT] = |
||
329 | MAX2(info->file_max[TGSI_FILE_INPUT], num_verts - 1); |
||
330 | for (j = 0; j < num_verts; ++j) { |
||
331 | info->file_mask[TGSI_FILE_INPUT] |= (1 << j); |
||
332 | } |
||
333 | } |
||
334 | |||
335 | tgsi_parse_free (&parse); |
||
336 | } |
||
337 | |||
338 | |||
339 | |||
340 | /** |
||
341 | * Check if the given shader is a "passthrough" shader consisting of only |
||
342 | * MOV instructions of the form: MOV OUT[n], IN[n] |
||
343 | * |
||
344 | */ |
||
345 | boolean |
||
346 | tgsi_is_passthrough_shader(const struct tgsi_token *tokens) |
||
347 | { |
||
348 | struct tgsi_parse_context parse; |
||
349 | |||
350 | /** |
||
351 | ** Setup to begin parsing input shader |
||
352 | **/ |
||
353 | if (tgsi_parse_init(&parse, tokens) != TGSI_PARSE_OK) { |
||
354 | debug_printf("tgsi_parse_init() failed in tgsi_is_passthrough_shader()!\n"); |
||
355 | return FALSE; |
||
356 | } |
||
357 | |||
358 | /** |
||
359 | ** Loop over incoming program tokens/instructions |
||
360 | */ |
||
361 | while (!tgsi_parse_end_of_tokens(&parse)) { |
||
362 | |||
363 | tgsi_parse_token(&parse); |
||
364 | |||
365 | switch (parse.FullToken.Token.Type) { |
||
366 | case TGSI_TOKEN_TYPE_INSTRUCTION: |
||
367 | { |
||
368 | struct tgsi_full_instruction *fullinst = |
||
369 | &parse.FullToken.FullInstruction; |
||
370 | const struct tgsi_full_src_register *src = |
||
371 | &fullinst->Src[0]; |
||
372 | const struct tgsi_full_dst_register *dst = |
||
373 | &fullinst->Dst[0]; |
||
374 | |||
375 | /* Do a whole bunch of checks for a simple move */ |
||
376 | if (fullinst->Instruction.Opcode != TGSI_OPCODE_MOV || |
||
377 | (src->Register.File != TGSI_FILE_INPUT && |
||
378 | src->Register.File != TGSI_FILE_SYSTEM_VALUE) || |
||
379 | dst->Register.File != TGSI_FILE_OUTPUT || |
||
380 | src->Register.Index != dst->Register.Index || |
||
381 | |||
382 | src->Register.Negate || |
||
383 | src->Register.Absolute || |
||
384 | |||
385 | src->Register.SwizzleX != TGSI_SWIZZLE_X || |
||
386 | src->Register.SwizzleY != TGSI_SWIZZLE_Y || |
||
387 | src->Register.SwizzleZ != TGSI_SWIZZLE_Z || |
||
388 | src->Register.SwizzleW != TGSI_SWIZZLE_W || |
||
389 | |||
390 | dst->Register.WriteMask != TGSI_WRITEMASK_XYZW) |
||
391 | { |
||
392 | tgsi_parse_free(&parse); |
||
393 | return FALSE; |
||
394 | } |
||
395 | } |
||
396 | break; |
||
397 | |||
398 | case TGSI_TOKEN_TYPE_DECLARATION: |
||
399 | /* fall-through */ |
||
400 | case TGSI_TOKEN_TYPE_IMMEDIATE: |
||
401 | /* fall-through */ |
||
402 | case TGSI_TOKEN_TYPE_PROPERTY: |
||
403 | /* fall-through */ |
||
404 | default: |
||
405 | ; /* no-op */ |
||
406 | } |
||
407 | } |
||
408 | |||
409 | tgsi_parse_free(&parse); |
||
410 | |||
411 | /* if we get here, it's a pass-through shader */ |
||
412 | return TRUE; |
||
413 | }><>>>><>><>><>><>=>><>><>>>><>><>>>>>>> |