1 /**************************************************************************
3 * Copyright 2008 VMware, Inc.
5 * Copyright 2008 VMware, Inc. All rights Reserved.
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:
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
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.
27 **************************************************************************/
30 * TGSI program scan utility.
31 * Used to determine which registers and instructions are used by a shader.
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_info.h"
42 #include "tgsi/tgsi_parse.h"
43 #include "tgsi/tgsi_util.h"
44 #include "tgsi/tgsi_scan.h"
48 is_memory_file(unsigned file
)
50 return file
== TGSI_FILE_SAMPLER
||
51 file
== TGSI_FILE_SAMPLER_VIEW
||
52 file
== TGSI_FILE_IMAGE
||
53 file
== TGSI_FILE_BUFFER
;
58 is_mem_query_inst(unsigned opcode
)
60 return opcode
== TGSI_OPCODE_RESQ
||
61 opcode
== TGSI_OPCODE_TXQ
||
62 opcode
== TGSI_OPCODE_TXQS
||
63 opcode
== TGSI_OPCODE_LODQ
;
67 * Is the opcode a "true" texture instruction which samples from a
71 is_texture_inst(unsigned opcode
)
73 return (!is_mem_query_inst(opcode
) &&
74 tgsi_get_opcode_info(opcode
)->is_tex
);
79 * Is the opcode an instruction which computes a derivative explicitly or
83 computes_derivative(unsigned opcode
)
85 if (tgsi_get_opcode_info(opcode
)->is_tex
) {
86 return opcode
!= TGSI_OPCODE_TG4
&&
87 opcode
!= TGSI_OPCODE_TXD
&&
88 opcode
!= TGSI_OPCODE_TXF
&&
89 opcode
!= TGSI_OPCODE_TXF_LZ
&&
90 opcode
!= TGSI_OPCODE_TEX_LZ
&&
91 opcode
!= TGSI_OPCODE_TXL
&&
92 opcode
!= TGSI_OPCODE_TXL2
&&
93 opcode
!= TGSI_OPCODE_TXQ
&&
94 opcode
!= TGSI_OPCODE_TXQS
;
97 return opcode
== TGSI_OPCODE_DDX
|| opcode
== TGSI_OPCODE_DDX_FINE
||
98 opcode
== TGSI_OPCODE_DDY
|| opcode
== TGSI_OPCODE_DDY_FINE
||
99 opcode
== TGSI_OPCODE_SAMPLE
||
100 opcode
== TGSI_OPCODE_SAMPLE_B
||
101 opcode
== TGSI_OPCODE_SAMPLE_C
;
106 scan_src_operand(struct tgsi_shader_info
*info
,
107 const struct tgsi_full_instruction
*fullinst
,
108 const struct tgsi_full_src_register
*src
,
110 unsigned usage_mask_after_swizzle
,
111 bool is_interp_instruction
,
114 int ind
= src
->Register
.Index
;
116 if (info
->processor
== PIPE_SHADER_COMPUTE
&&
117 src
->Register
.File
== TGSI_FILE_SYSTEM_VALUE
) {
120 name
= info
->system_value_semantic_name
[src
->Register
.Index
];
123 case TGSI_SEMANTIC_THREAD_ID
:
124 case TGSI_SEMANTIC_BLOCK_ID
:
125 mask
= usage_mask_after_swizzle
& TGSI_WRITEMASK_XYZ
;
127 unsigned i
= u_bit_scan(&mask
);
129 if (name
== TGSI_SEMANTIC_THREAD_ID
)
130 info
->uses_thread_id
[i
] = true;
132 info
->uses_block_id
[i
] = true;
135 case TGSI_SEMANTIC_BLOCK_SIZE
:
136 /* The block size is translated to IMM with a fixed block size. */
137 if (info
->properties
[TGSI_PROPERTY_CS_FIXED_BLOCK_WIDTH
] == 0)
138 info
->uses_block_size
= true;
140 case TGSI_SEMANTIC_GRID_SIZE
:
141 info
->uses_grid_size
= true;
146 /* Mark which inputs are effectively used */
147 if (src
->Register
.File
== TGSI_FILE_INPUT
) {
148 if (src
->Register
.Indirect
) {
149 for (ind
= 0; ind
< info
->num_inputs
; ++ind
) {
150 info
->input_usage_mask
[ind
] |= usage_mask_after_swizzle
;
154 assert(ind
< PIPE_MAX_SHADER_INPUTS
);
155 info
->input_usage_mask
[ind
] |= usage_mask_after_swizzle
;
158 if (info
->processor
== PIPE_SHADER_FRAGMENT
) {
159 unsigned name
, index
, input
;
161 if (src
->Register
.Indirect
&& src
->Indirect
.ArrayID
)
162 input
= info
->input_array_first
[src
->Indirect
.ArrayID
];
164 input
= src
->Register
.Index
;
166 name
= info
->input_semantic_name
[input
];
167 index
= info
->input_semantic_index
[input
];
169 if (name
== TGSI_SEMANTIC_POSITION
&&
170 usage_mask_after_swizzle
& TGSI_WRITEMASK_Z
)
171 info
->reads_z
= true;
173 if (name
== TGSI_SEMANTIC_COLOR
)
174 info
->colors_read
|= usage_mask_after_swizzle
<< (index
* 4);
176 /* Process only interpolated varyings. Don't include POSITION.
177 * Don't include integer varyings, because they are not
178 * interpolated. Don't process inputs interpolated by INTERP
179 * opcodes. Those are tracked separately.
181 if ((!is_interp_instruction
|| src_index
!= 0) &&
182 (name
== TGSI_SEMANTIC_GENERIC
||
183 name
== TGSI_SEMANTIC_TEXCOORD
||
184 name
== TGSI_SEMANTIC_COLOR
||
185 name
== TGSI_SEMANTIC_BCOLOR
||
186 name
== TGSI_SEMANTIC_FOG
||
187 name
== TGSI_SEMANTIC_CLIPDIST
)) {
188 switch (info
->input_interpolate
[input
]) {
189 case TGSI_INTERPOLATE_COLOR
:
190 case TGSI_INTERPOLATE_PERSPECTIVE
:
191 switch (info
->input_interpolate_loc
[input
]) {
192 case TGSI_INTERPOLATE_LOC_CENTER
:
193 info
->uses_persp_center
= TRUE
;
195 case TGSI_INTERPOLATE_LOC_CENTROID
:
196 info
->uses_persp_centroid
= TRUE
;
198 case TGSI_INTERPOLATE_LOC_SAMPLE
:
199 info
->uses_persp_sample
= TRUE
;
203 case TGSI_INTERPOLATE_LINEAR
:
204 switch (info
->input_interpolate_loc
[input
]) {
205 case TGSI_INTERPOLATE_LOC_CENTER
:
206 info
->uses_linear_center
= TRUE
;
208 case TGSI_INTERPOLATE_LOC_CENTROID
:
209 info
->uses_linear_centroid
= TRUE
;
211 case TGSI_INTERPOLATE_LOC_SAMPLE
:
212 info
->uses_linear_sample
= TRUE
;
216 /* TGSI_INTERPOLATE_CONSTANT doesn't do any interpolation. */
222 if (info
->processor
== PIPE_SHADER_TESS_CTRL
&&
223 src
->Register
.File
== TGSI_FILE_OUTPUT
) {
226 if (src
->Register
.Indirect
&& src
->Indirect
.ArrayID
)
227 input
= info
->output_array_first
[src
->Indirect
.ArrayID
];
229 input
= src
->Register
.Index
;
231 switch (info
->output_semantic_name
[input
]) {
232 case TGSI_SEMANTIC_PATCH
:
233 info
->reads_perpatch_outputs
= true;
235 case TGSI_SEMANTIC_TESSINNER
:
236 case TGSI_SEMANTIC_TESSOUTER
:
237 info
->reads_tessfactor_outputs
= true;
240 info
->reads_pervertex_outputs
= true;
244 /* check for indirect register reads */
245 if (src
->Register
.Indirect
) {
246 info
->indirect_files
|= (1 << src
->Register
.File
);
247 info
->indirect_files_read
|= (1 << src
->Register
.File
);
249 /* record indirect constant buffer indexing */
250 if (src
->Register
.File
== TGSI_FILE_CONSTANT
) {
251 if (src
->Register
.Dimension
) {
252 if (src
->Dimension
.Indirect
)
253 info
->const_buffers_indirect
= info
->const_buffers_declared
;
255 info
->const_buffers_indirect
|= 1u << src
->Dimension
.Index
;
257 info
->const_buffers_indirect
|= 1;
262 if (src
->Register
.Dimension
&& src
->Dimension
.Indirect
)
263 info
->dim_indirect_files
|= 1u << src
->Register
.File
;
265 /* Texture samplers */
266 if (src
->Register
.File
== TGSI_FILE_SAMPLER
) {
267 const unsigned index
= src
->Register
.Index
;
269 assert(fullinst
->Instruction
.Texture
);
270 assert(index
< ARRAY_SIZE(info
->is_msaa_sampler
));
271 assert(index
< PIPE_MAX_SAMPLERS
);
273 if (is_texture_inst(fullinst
->Instruction
.Opcode
)) {
274 const unsigned target
= fullinst
->Texture
.Texture
;
275 assert(target
< TGSI_TEXTURE_UNKNOWN
);
276 /* for texture instructions, check that the texture instruction
277 * target matches the previous sampler view declaration (if there
280 if (info
->sampler_targets
[index
] == TGSI_TEXTURE_UNKNOWN
) {
281 /* probably no sampler view declaration */
282 info
->sampler_targets
[index
] = target
;
284 /* Make sure the texture instruction's sampler/target info
285 * agrees with the sampler view declaration.
287 assert(info
->sampler_targets
[index
] == target
);
290 if (target
== TGSI_TEXTURE_2D_MSAA
||
291 target
== TGSI_TEXTURE_2D_ARRAY_MSAA
) {
292 info
->is_msaa_sampler
[src
->Register
.Index
] = TRUE
;
297 if (is_memory_file(src
->Register
.File
) &&
298 !is_mem_query_inst(fullinst
->Instruction
.Opcode
)) {
301 if (tgsi_get_opcode_info(fullinst
->Instruction
.Opcode
)->is_store
) {
302 info
->writes_memory
= TRUE
;
304 if (src
->Register
.File
== TGSI_FILE_IMAGE
) {
305 if (src
->Register
.Indirect
)
306 info
->images_atomic
= info
->images_declared
;
308 info
->images_atomic
|= 1 << src
->Register
.Index
;
309 } else if (src
->Register
.File
== TGSI_FILE_BUFFER
) {
310 if (src
->Register
.Indirect
)
311 info
->shader_buffers_atomic
= info
->shader_buffers_declared
;
313 info
->shader_buffers_atomic
|= 1 << src
->Register
.Index
;
316 if (src
->Register
.File
== TGSI_FILE_IMAGE
) {
317 if (src
->Register
.Indirect
)
318 info
->images_load
= info
->images_declared
;
320 info
->images_load
|= 1 << src
->Register
.Index
;
321 } else if (src
->Register
.File
== TGSI_FILE_BUFFER
) {
322 if (src
->Register
.Indirect
)
323 info
->shader_buffers_load
= info
->shader_buffers_declared
;
325 info
->shader_buffers_load
|= 1 << src
->Register
.Index
;
333 scan_instruction(struct tgsi_shader_info
*info
,
334 const struct tgsi_full_instruction
*fullinst
,
335 unsigned *current_depth
)
338 bool is_mem_inst
= false;
339 bool is_interp_instruction
= false;
340 unsigned sampler_src
;
342 assert(fullinst
->Instruction
.Opcode
< TGSI_OPCODE_LAST
);
343 info
->opcode_count
[fullinst
->Instruction
.Opcode
]++;
345 switch (fullinst
->Instruction
.Opcode
) {
347 case TGSI_OPCODE_UIF
:
348 case TGSI_OPCODE_BGNLOOP
:
350 info
->max_depth
= MAX2(info
->max_depth
, *current_depth
);
352 case TGSI_OPCODE_ENDIF
:
353 case TGSI_OPCODE_ENDLOOP
:
356 case TGSI_OPCODE_TEX
:
357 case TGSI_OPCODE_TEX_LZ
:
358 case TGSI_OPCODE_TXB
:
359 case TGSI_OPCODE_TXD
:
360 case TGSI_OPCODE_TXL
:
361 case TGSI_OPCODE_TXP
:
362 case TGSI_OPCODE_TXQ
:
363 case TGSI_OPCODE_TXQS
:
364 case TGSI_OPCODE_TXF
:
365 case TGSI_OPCODE_TXF_LZ
:
366 case TGSI_OPCODE_TEX2
:
367 case TGSI_OPCODE_TXB2
:
368 case TGSI_OPCODE_TXL2
:
369 case TGSI_OPCODE_TG4
:
370 case TGSI_OPCODE_LODQ
:
371 sampler_src
= fullinst
->Instruction
.NumSrcRegs
- 1;
372 if (fullinst
->Src
[sampler_src
].Register
.File
!= TGSI_FILE_SAMPLER
)
373 info
->uses_bindless_samplers
= true;
375 case TGSI_OPCODE_RESQ
:
376 case TGSI_OPCODE_LOAD
:
377 case TGSI_OPCODE_ATOMUADD
:
378 case TGSI_OPCODE_ATOMXCHG
:
379 case TGSI_OPCODE_ATOMCAS
:
380 case TGSI_OPCODE_ATOMAND
:
381 case TGSI_OPCODE_ATOMOR
:
382 case TGSI_OPCODE_ATOMXOR
:
383 case TGSI_OPCODE_ATOMUMIN
:
384 case TGSI_OPCODE_ATOMUMAX
:
385 case TGSI_OPCODE_ATOMIMIN
:
386 case TGSI_OPCODE_ATOMIMAX
:
387 if (tgsi_is_bindless_image_file(fullinst
->Src
[0].Register
.File
))
388 info
->uses_bindless_images
= true;
390 case TGSI_OPCODE_STORE
:
391 if (tgsi_is_bindless_image_file(fullinst
->Dst
[0].Register
.File
))
392 info
->uses_bindless_images
= true;
398 if (fullinst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_CENTROID
||
399 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_OFFSET
||
400 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_SAMPLE
) {
401 const struct tgsi_full_src_register
*src0
= &fullinst
->Src
[0];
404 is_interp_instruction
= true;
406 if (src0
->Register
.Indirect
&& src0
->Indirect
.ArrayID
)
407 input
= info
->input_array_first
[src0
->Indirect
.ArrayID
];
409 input
= src0
->Register
.Index
;
411 /* For the INTERP opcodes, the interpolation is always
412 * PERSPECTIVE unless LINEAR is specified.
414 switch (info
->input_interpolate
[input
]) {
415 case TGSI_INTERPOLATE_COLOR
:
416 case TGSI_INTERPOLATE_CONSTANT
:
417 case TGSI_INTERPOLATE_PERSPECTIVE
:
418 switch (fullinst
->Instruction
.Opcode
) {
419 case TGSI_OPCODE_INTERP_CENTROID
:
420 info
->uses_persp_opcode_interp_centroid
= TRUE
;
422 case TGSI_OPCODE_INTERP_OFFSET
:
423 info
->uses_persp_opcode_interp_offset
= TRUE
;
425 case TGSI_OPCODE_INTERP_SAMPLE
:
426 info
->uses_persp_opcode_interp_sample
= TRUE
;
431 case TGSI_INTERPOLATE_LINEAR
:
432 switch (fullinst
->Instruction
.Opcode
) {
433 case TGSI_OPCODE_INTERP_CENTROID
:
434 info
->uses_linear_opcode_interp_centroid
= TRUE
;
436 case TGSI_OPCODE_INTERP_OFFSET
:
437 info
->uses_linear_opcode_interp_offset
= TRUE
;
439 case TGSI_OPCODE_INTERP_SAMPLE
:
440 info
->uses_linear_opcode_interp_sample
= TRUE
;
447 if ((fullinst
->Instruction
.Opcode
>= TGSI_OPCODE_F2D
&&
448 fullinst
->Instruction
.Opcode
<= TGSI_OPCODE_DSSG
) ||
449 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_DFMA
||
450 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_DDIV
||
451 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_D2U64
||
452 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_D2I64
||
453 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_U642D
||
454 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_I642D
)
455 info
->uses_doubles
= TRUE
;
457 for (i
= 0; i
< fullinst
->Instruction
.NumSrcRegs
; i
++) {
458 scan_src_operand(info
, fullinst
, &fullinst
->Src
[i
], i
,
459 tgsi_util_get_inst_usage_mask(fullinst
, i
),
460 is_interp_instruction
, &is_mem_inst
);
462 if (fullinst
->Src
[i
].Register
.Indirect
) {
463 struct tgsi_full_src_register src
= {{0}};
465 src
.Register
.File
= fullinst
->Src
[i
].Indirect
.File
;
466 src
.Register
.Index
= fullinst
->Src
[i
].Indirect
.Index
;
468 scan_src_operand(info
, fullinst
, &src
, -1,
469 1 << fullinst
->Src
[i
].Indirect
.Swizzle
,
473 if (fullinst
->Src
[i
].Register
.Dimension
&&
474 fullinst
->Src
[i
].Dimension
.Indirect
) {
475 struct tgsi_full_src_register src
= {{0}};
477 src
.Register
.File
= fullinst
->Src
[i
].DimIndirect
.File
;
478 src
.Register
.Index
= fullinst
->Src
[i
].DimIndirect
.Index
;
480 scan_src_operand(info
, fullinst
, &src
, -1,
481 1 << fullinst
->Src
[i
].DimIndirect
.Swizzle
,
486 if (fullinst
->Instruction
.Texture
) {
487 for (i
= 0; i
< fullinst
->Texture
.NumOffsets
; i
++) {
488 struct tgsi_full_src_register src
= {{0}};
490 src
.Register
.File
= fullinst
->TexOffsets
[i
].File
;
491 src
.Register
.Index
= fullinst
->TexOffsets
[i
].Index
;
493 /* The usage mask is suboptimal but should be safe. */
494 scan_src_operand(info
, fullinst
, &src
, -1,
495 (1 << fullinst
->TexOffsets
[i
].SwizzleX
) |
496 (1 << fullinst
->TexOffsets
[i
].SwizzleY
) |
497 (1 << fullinst
->TexOffsets
[i
].SwizzleZ
),
498 false, &is_mem_inst
);
502 /* check for indirect register writes */
503 for (i
= 0; i
< fullinst
->Instruction
.NumDstRegs
; i
++) {
504 const struct tgsi_full_dst_register
*dst
= &fullinst
->Dst
[i
];
506 if (dst
->Register
.Indirect
) {
507 struct tgsi_full_src_register src
= {{0}};
509 src
.Register
.File
= dst
->Indirect
.File
;
510 src
.Register
.Index
= dst
->Indirect
.Index
;
512 scan_src_operand(info
, fullinst
, &src
, -1,
513 1 << dst
->Indirect
.Swizzle
, false, NULL
);
515 info
->indirect_files
|= (1 << dst
->Register
.File
);
516 info
->indirect_files_written
|= (1 << dst
->Register
.File
);
519 if (dst
->Register
.Dimension
&& dst
->Dimension
.Indirect
) {
520 struct tgsi_full_src_register src
= {{0}};
522 src
.Register
.File
= dst
->DimIndirect
.File
;
523 src
.Register
.Index
= dst
->DimIndirect
.Index
;
525 scan_src_operand(info
, fullinst
, &src
, -1,
526 1 << dst
->DimIndirect
.Swizzle
, false, NULL
);
528 info
->dim_indirect_files
|= 1u << dst
->Register
.File
;
531 if (is_memory_file(dst
->Register
.File
)) {
532 assert(fullinst
->Instruction
.Opcode
== TGSI_OPCODE_STORE
);
535 info
->writes_memory
= TRUE
;
537 if (dst
->Register
.File
== TGSI_FILE_IMAGE
) {
538 if (dst
->Register
.Indirect
)
539 info
->images_store
= info
->images_declared
;
541 info
->images_store
|= 1 << dst
->Register
.Index
;
542 } else if (dst
->Register
.File
== TGSI_FILE_BUFFER
) {
543 if (dst
->Register
.Indirect
)
544 info
->shader_buffers_store
= info
->shader_buffers_declared
;
546 info
->shader_buffers_store
|= 1 << dst
->Register
.Index
;
552 info
->num_memory_instructions
++;
554 if (computes_derivative(fullinst
->Instruction
.Opcode
))
555 info
->uses_derivatives
= true;
557 info
->num_instructions
++;
562 scan_declaration(struct tgsi_shader_info
*info
,
563 const struct tgsi_full_declaration
*fulldecl
)
565 const uint file
= fulldecl
->Declaration
.File
;
566 const unsigned procType
= info
->processor
;
569 if (fulldecl
->Declaration
.Array
) {
570 unsigned array_id
= fulldecl
->Array
.ArrayID
;
573 case TGSI_FILE_INPUT
:
574 assert(array_id
< ARRAY_SIZE(info
->input_array_first
));
575 info
->input_array_first
[array_id
] = fulldecl
->Range
.First
;
576 info
->input_array_last
[array_id
] = fulldecl
->Range
.Last
;
578 case TGSI_FILE_OUTPUT
:
579 assert(array_id
< ARRAY_SIZE(info
->output_array_first
));
580 info
->output_array_first
[array_id
] = fulldecl
->Range
.First
;
581 info
->output_array_last
[array_id
] = fulldecl
->Range
.Last
;
584 info
->array_max
[file
] = MAX2(info
->array_max
[file
], array_id
);
587 for (reg
= fulldecl
->Range
.First
; reg
<= fulldecl
->Range
.Last
; reg
++) {
588 unsigned semName
= fulldecl
->Semantic
.Name
;
589 unsigned semIndex
= fulldecl
->Semantic
.Index
+
590 (reg
- fulldecl
->Range
.First
);
592 unsigned index
, target
, type
;
594 /* only first 32 regs will appear in this bitfield */
595 info
->file_mask
[file
] |= (1 << reg
);
596 info
->file_count
[file
]++;
597 info
->file_max
[file
] = MAX2(info
->file_max
[file
], (int)reg
);
600 case TGSI_FILE_CONSTANT
:
603 if (fulldecl
->Declaration
.Dimension
)
604 buffer
= fulldecl
->Dim
.Index2D
;
606 info
->const_file_max
[buffer
] =
607 MAX2(info
->const_file_max
[buffer
], (int)reg
);
608 info
->const_buffers_declared
|= 1u << buffer
;
611 case TGSI_FILE_IMAGE
:
612 info
->images_declared
|= 1u << reg
;
613 if (fulldecl
->Image
.Resource
== TGSI_TEXTURE_BUFFER
)
614 info
->images_buffers
|= 1 << reg
;
617 case TGSI_FILE_BUFFER
:
618 info
->shader_buffers_declared
|= 1u << reg
;
621 case TGSI_FILE_INPUT
:
622 info
->input_semantic_name
[reg
] = (ubyte
) semName
;
623 info
->input_semantic_index
[reg
] = (ubyte
) semIndex
;
624 info
->input_interpolate
[reg
] = (ubyte
)fulldecl
->Interp
.Interpolate
;
625 info
->input_interpolate_loc
[reg
] = (ubyte
)fulldecl
->Interp
.Location
;
626 info
->input_cylindrical_wrap
[reg
] = (ubyte
)fulldecl
->Interp
.CylindricalWrap
;
628 /* Vertex shaders can have inputs with holes between them. */
629 info
->num_inputs
= MAX2(info
->num_inputs
, reg
+ 1);
632 case TGSI_SEMANTIC_PRIMID
:
633 info
->uses_primid
= true;
635 case TGSI_SEMANTIC_POSITION
:
636 info
->reads_position
= true;
638 case TGSI_SEMANTIC_FACE
:
639 info
->uses_frontface
= true;
644 case TGSI_FILE_SYSTEM_VALUE
:
645 index
= fulldecl
->Range
.First
;
647 info
->system_value_semantic_name
[index
] = semName
;
648 info
->num_system_values
= MAX2(info
->num_system_values
, index
+ 1);
651 case TGSI_SEMANTIC_INSTANCEID
:
652 info
->uses_instanceid
= TRUE
;
654 case TGSI_SEMANTIC_VERTEXID
:
655 info
->uses_vertexid
= TRUE
;
657 case TGSI_SEMANTIC_VERTEXID_NOBASE
:
658 info
->uses_vertexid_nobase
= TRUE
;
660 case TGSI_SEMANTIC_BASEVERTEX
:
661 info
->uses_basevertex
= TRUE
;
663 case TGSI_SEMANTIC_PRIMID
:
664 info
->uses_primid
= TRUE
;
666 case TGSI_SEMANTIC_INVOCATIONID
:
667 info
->uses_invocationid
= TRUE
;
669 case TGSI_SEMANTIC_POSITION
:
670 info
->reads_position
= TRUE
;
672 case TGSI_SEMANTIC_FACE
:
673 info
->uses_frontface
= TRUE
;
675 case TGSI_SEMANTIC_SAMPLEMASK
:
676 info
->reads_samplemask
= TRUE
;
678 case TGSI_SEMANTIC_TESSINNER
:
679 case TGSI_SEMANTIC_TESSOUTER
:
680 info
->reads_tess_factors
= true;
685 case TGSI_FILE_OUTPUT
:
686 info
->output_semantic_name
[reg
] = (ubyte
) semName
;
687 info
->output_semantic_index
[reg
] = (ubyte
) semIndex
;
688 info
->output_usagemask
[reg
] |= fulldecl
->Declaration
.UsageMask
;
689 info
->num_outputs
= MAX2(info
->num_outputs
, reg
+ 1);
691 if (fulldecl
->Declaration
.UsageMask
& TGSI_WRITEMASK_X
) {
692 info
->output_streams
[reg
] |= (ubyte
)fulldecl
->Semantic
.StreamX
;
693 info
->num_stream_output_components
[fulldecl
->Semantic
.StreamX
]++;
695 if (fulldecl
->Declaration
.UsageMask
& TGSI_WRITEMASK_Y
) {
696 info
->output_streams
[reg
] |= (ubyte
)fulldecl
->Semantic
.StreamY
<< 2;
697 info
->num_stream_output_components
[fulldecl
->Semantic
.StreamY
]++;
699 if (fulldecl
->Declaration
.UsageMask
& TGSI_WRITEMASK_Z
) {
700 info
->output_streams
[reg
] |= (ubyte
)fulldecl
->Semantic
.StreamZ
<< 4;
701 info
->num_stream_output_components
[fulldecl
->Semantic
.StreamZ
]++;
703 if (fulldecl
->Declaration
.UsageMask
& TGSI_WRITEMASK_W
) {
704 info
->output_streams
[reg
] |= (ubyte
)fulldecl
->Semantic
.StreamW
<< 6;
705 info
->num_stream_output_components
[fulldecl
->Semantic
.StreamW
]++;
709 case TGSI_SEMANTIC_PRIMID
:
710 info
->writes_primid
= true;
712 case TGSI_SEMANTIC_VIEWPORT_INDEX
:
713 info
->writes_viewport_index
= true;
715 case TGSI_SEMANTIC_LAYER
:
716 info
->writes_layer
= true;
718 case TGSI_SEMANTIC_PSIZE
:
719 info
->writes_psize
= true;
721 case TGSI_SEMANTIC_CLIPVERTEX
:
722 info
->writes_clipvertex
= true;
724 case TGSI_SEMANTIC_COLOR
:
725 info
->colors_written
|= 1 << semIndex
;
727 case TGSI_SEMANTIC_STENCIL
:
728 info
->writes_stencil
= true;
730 case TGSI_SEMANTIC_SAMPLEMASK
:
731 info
->writes_samplemask
= true;
733 case TGSI_SEMANTIC_EDGEFLAG
:
734 info
->writes_edgeflag
= true;
736 case TGSI_SEMANTIC_POSITION
:
737 if (procType
== PIPE_SHADER_FRAGMENT
)
738 info
->writes_z
= true;
740 info
->writes_position
= true;
745 case TGSI_FILE_SAMPLER
:
746 STATIC_ASSERT(sizeof(info
->samplers_declared
) * 8 >= PIPE_MAX_SAMPLERS
);
747 info
->samplers_declared
|= 1u << reg
;
750 case TGSI_FILE_SAMPLER_VIEW
:
751 target
= fulldecl
->SamplerView
.Resource
;
752 type
= fulldecl
->SamplerView
.ReturnTypeX
;
754 assert(target
< TGSI_TEXTURE_UNKNOWN
);
755 if (info
->sampler_targets
[reg
] == TGSI_TEXTURE_UNKNOWN
) {
756 /* Save sampler target for this sampler index */
757 info
->sampler_targets
[reg
] = target
;
758 info
->sampler_type
[reg
] = type
;
760 /* if previously declared, make sure targets agree */
761 assert(info
->sampler_targets
[reg
] == target
);
762 assert(info
->sampler_type
[reg
] == type
);
771 scan_immediate(struct tgsi_shader_info
*info
)
773 uint reg
= info
->immediate_count
++;
774 uint file
= TGSI_FILE_IMMEDIATE
;
776 info
->file_mask
[file
] |= (1 << reg
);
777 info
->file_count
[file
]++;
778 info
->file_max
[file
] = MAX2(info
->file_max
[file
], (int)reg
);
783 scan_property(struct tgsi_shader_info
*info
,
784 const struct tgsi_full_property
*fullprop
)
786 unsigned name
= fullprop
->Property
.PropertyName
;
787 unsigned value
= fullprop
->u
[0].Data
;
789 assert(name
< ARRAY_SIZE(info
->properties
));
790 info
->properties
[name
] = value
;
793 case TGSI_PROPERTY_NUM_CLIPDIST_ENABLED
:
794 info
->num_written_clipdistance
= value
;
795 info
->clipdist_writemask
|= (1 << value
) - 1;
797 case TGSI_PROPERTY_NUM_CULLDIST_ENABLED
:
798 info
->num_written_culldistance
= value
;
799 info
->culldist_writemask
|= (1 << value
) - 1;
806 * Scan the given TGSI shader to collect information such as number of
807 * registers used, special instructions used, etc.
808 * \return info the result of the scan
811 tgsi_scan_shader(const struct tgsi_token
*tokens
,
812 struct tgsi_shader_info
*info
)
815 struct tgsi_parse_context parse
;
816 unsigned current_depth
= 0;
818 memset(info
, 0, sizeof(*info
));
819 for (i
= 0; i
< TGSI_FILE_COUNT
; i
++)
820 info
->file_max
[i
] = -1;
821 for (i
= 0; i
< ARRAY_SIZE(info
->const_file_max
); i
++)
822 info
->const_file_max
[i
] = -1;
823 info
->properties
[TGSI_PROPERTY_GS_INVOCATIONS
] = 1;
824 for (i
= 0; i
< ARRAY_SIZE(info
->sampler_targets
); i
++)
825 info
->sampler_targets
[i
] = TGSI_TEXTURE_UNKNOWN
;
828 ** Setup to begin parsing input shader
830 if (tgsi_parse_init( &parse
, tokens
) != TGSI_PARSE_OK
) {
831 debug_printf("tgsi_parse_init() failed in tgsi_scan_shader()!\n");
834 procType
= parse
.FullHeader
.Processor
.Processor
;
835 assert(procType
== PIPE_SHADER_FRAGMENT
||
836 procType
== PIPE_SHADER_VERTEX
||
837 procType
== PIPE_SHADER_GEOMETRY
||
838 procType
== PIPE_SHADER_TESS_CTRL
||
839 procType
== PIPE_SHADER_TESS_EVAL
||
840 procType
== PIPE_SHADER_COMPUTE
);
841 info
->processor
= procType
;
844 ** Loop over incoming program tokens/instructions
846 while (!tgsi_parse_end_of_tokens(&parse
)) {
849 tgsi_parse_token( &parse
);
851 switch( parse
.FullToken
.Token
.Type
) {
852 case TGSI_TOKEN_TYPE_INSTRUCTION
:
853 scan_instruction(info
, &parse
.FullToken
.FullInstruction
,
856 case TGSI_TOKEN_TYPE_DECLARATION
:
857 scan_declaration(info
, &parse
.FullToken
.FullDeclaration
);
859 case TGSI_TOKEN_TYPE_IMMEDIATE
:
860 scan_immediate(info
);
862 case TGSI_TOKEN_TYPE_PROPERTY
:
863 scan_property(info
, &parse
.FullToken
.FullProperty
);
866 assert(!"Unexpected TGSI token type");
870 info
->uses_kill
= (info
->opcode_count
[TGSI_OPCODE_KILL_IF
] ||
871 info
->opcode_count
[TGSI_OPCODE_KILL
]);
873 /* The dimensions of the IN decleration in geometry shader have
874 * to be deduced from the type of the input primitive.
876 if (procType
== PIPE_SHADER_GEOMETRY
) {
877 unsigned input_primitive
=
878 info
->properties
[TGSI_PROPERTY_GS_INPUT_PRIM
];
879 int num_verts
= u_vertices_per_prim(input_primitive
);
881 info
->file_count
[TGSI_FILE_INPUT
] = num_verts
;
882 info
->file_max
[TGSI_FILE_INPUT
] =
883 MAX2(info
->file_max
[TGSI_FILE_INPUT
], num_verts
- 1);
884 for (j
= 0; j
< num_verts
; ++j
) {
885 info
->file_mask
[TGSI_FILE_INPUT
] |= (1 << j
);
889 tgsi_parse_free(&parse
);
893 * Collect information about the arrays of a given register file.
895 * @param tokens TGSI shader
896 * @param file the register file to scan through
897 * @param max_array_id number of entries in @p arrays; should be equal to the
898 * highest array id, i.e. tgsi_shader_info::array_max[file].
899 * @param arrays info for array of each ID will be written to arrays[ID - 1].
902 tgsi_scan_arrays(const struct tgsi_token
*tokens
,
904 unsigned max_array_id
,
905 struct tgsi_array_info
*arrays
)
907 struct tgsi_parse_context parse
;
909 if (tgsi_parse_init(&parse
, tokens
) != TGSI_PARSE_OK
) {
910 debug_printf("tgsi_parse_init() failed in tgsi_scan_arrays()!\n");
914 memset(arrays
, 0, sizeof(arrays
[0]) * max_array_id
);
916 while (!tgsi_parse_end_of_tokens(&parse
)) {
917 struct tgsi_full_instruction
*inst
;
919 tgsi_parse_token(&parse
);
921 if (parse
.FullToken
.Token
.Type
== TGSI_TOKEN_TYPE_DECLARATION
) {
922 struct tgsi_full_declaration
*decl
= &parse
.FullToken
.FullDeclaration
;
924 if (decl
->Declaration
.Array
&& decl
->Declaration
.File
== file
&&
925 decl
->Array
.ArrayID
> 0 && decl
->Array
.ArrayID
<= max_array_id
) {
926 struct tgsi_array_info
*array
= &arrays
[decl
->Array
.ArrayID
- 1];
927 assert(!array
->declared
);
928 array
->declared
= true;
929 array
->range
= decl
->Range
;
933 if (parse
.FullToken
.Token
.Type
!= TGSI_TOKEN_TYPE_INSTRUCTION
)
936 inst
= &parse
.FullToken
.FullInstruction
;
937 for (unsigned i
= 0; i
< inst
->Instruction
.NumDstRegs
; i
++) {
938 const struct tgsi_full_dst_register
*dst
= &inst
->Dst
[i
];
939 if (dst
->Register
.File
!= file
)
942 if (dst
->Register
.Indirect
) {
943 if (dst
->Indirect
.ArrayID
> 0 &&
944 dst
->Indirect
.ArrayID
<= max_array_id
) {
945 arrays
[dst
->Indirect
.ArrayID
- 1].writemask
|= dst
->Register
.WriteMask
;
947 /* Indirect writes without an ArrayID can write anywhere. */
948 for (unsigned j
= 0; j
< max_array_id
; ++j
)
949 arrays
[j
].writemask
|= dst
->Register
.WriteMask
;
952 /* Check whether the write falls into any of the arrays anyway. */
953 for (unsigned j
= 0; j
< max_array_id
; ++j
) {
954 struct tgsi_array_info
*array
= &arrays
[j
];
955 if (array
->declared
&&
956 dst
->Register
.Index
>= array
->range
.First
&&
957 dst
->Register
.Index
<= array
->range
.Last
)
958 array
->writemask
|= dst
->Register
.WriteMask
;
964 tgsi_parse_free(&parse
);
970 check_no_subroutines(const struct tgsi_full_instruction
*inst
)
972 switch (inst
->Instruction
.Opcode
) {
973 case TGSI_OPCODE_BGNSUB
:
974 case TGSI_OPCODE_ENDSUB
:
975 case TGSI_OPCODE_CAL
:
976 unreachable("subroutines unhandled");
981 get_inst_tessfactor_writemask(const struct tgsi_shader_info
*info
,
982 const struct tgsi_full_instruction
*inst
)
984 unsigned writemask
= 0;
986 for (unsigned i
= 0; i
< inst
->Instruction
.NumDstRegs
; i
++) {
987 const struct tgsi_full_dst_register
*dst
= &inst
->Dst
[i
];
989 if (dst
->Register
.File
== TGSI_FILE_OUTPUT
&&
990 !dst
->Register
.Indirect
) {
991 unsigned name
= info
->output_semantic_name
[dst
->Register
.Index
];
993 if (name
== TGSI_SEMANTIC_TESSINNER
)
994 writemask
|= dst
->Register
.WriteMask
;
995 else if (name
== TGSI_SEMANTIC_TESSOUTER
)
996 writemask
|= dst
->Register
.WriteMask
<< 4;
1003 get_block_tessfactor_writemask(const struct tgsi_shader_info
*info
,
1004 struct tgsi_parse_context
*parse
,
1005 unsigned end_opcode
)
1007 struct tgsi_full_instruction
*inst
;
1008 unsigned writemask
= 0;
1011 tgsi_parse_token(parse
);
1012 assert(parse
->FullToken
.Token
.Type
== TGSI_TOKEN_TYPE_INSTRUCTION
);
1013 inst
= &parse
->FullToken
.FullInstruction
;
1014 check_no_subroutines(inst
);
1016 /* Recursively process nested blocks. */
1017 switch (inst
->Instruction
.Opcode
) {
1018 case TGSI_OPCODE_IF
:
1019 case TGSI_OPCODE_UIF
:
1021 get_block_tessfactor_writemask(info
, parse
, TGSI_OPCODE_ENDIF
);
1024 case TGSI_OPCODE_BGNLOOP
:
1026 get_block_tessfactor_writemask(info
, parse
, TGSI_OPCODE_ENDLOOP
);
1029 case TGSI_OPCODE_BARRIER
:
1030 unreachable("nested BARRIER is illegal");
1034 writemask
|= get_inst_tessfactor_writemask(info
, inst
);
1035 } while (inst
->Instruction
.Opcode
!= end_opcode
);
1041 get_if_block_tessfactor_writemask(const struct tgsi_shader_info
*info
,
1042 struct tgsi_parse_context
*parse
,
1043 unsigned *upper_block_tf_writemask
,
1044 unsigned *cond_block_tf_writemask
)
1046 struct tgsi_full_instruction
*inst
;
1047 unsigned then_tessfactor_writemask
= 0;
1048 unsigned else_tessfactor_writemask
= 0;
1049 bool is_then
= true;
1052 tgsi_parse_token(parse
);
1053 assert(parse
->FullToken
.Token
.Type
== TGSI_TOKEN_TYPE_INSTRUCTION
);
1054 inst
= &parse
->FullToken
.FullInstruction
;
1055 check_no_subroutines(inst
);
1057 switch (inst
->Instruction
.Opcode
) {
1058 case TGSI_OPCODE_ELSE
:
1062 /* Recursively process nested blocks. */
1063 case TGSI_OPCODE_IF
:
1064 case TGSI_OPCODE_UIF
:
1065 get_if_block_tessfactor_writemask(info
, parse
,
1066 is_then
? &then_tessfactor_writemask
:
1067 &else_tessfactor_writemask
,
1068 cond_block_tf_writemask
);
1071 case TGSI_OPCODE_BGNLOOP
:
1072 *cond_block_tf_writemask
|=
1073 get_block_tessfactor_writemask(info
, parse
, TGSI_OPCODE_ENDLOOP
);
1076 case TGSI_OPCODE_BARRIER
:
1077 unreachable("nested BARRIER is illegal");
1081 /* Process an instruction in the current block. */
1082 unsigned writemask
= get_inst_tessfactor_writemask(info
, inst
);
1086 then_tessfactor_writemask
|= writemask
;
1088 else_tessfactor_writemask
|= writemask
;
1090 } while (inst
->Instruction
.Opcode
!= TGSI_OPCODE_ENDIF
);
1092 if (then_tessfactor_writemask
|| else_tessfactor_writemask
) {
1093 /* If both statements write the same tess factor channels,
1094 * we can say that the upper block writes them too. */
1095 *upper_block_tf_writemask
|= then_tessfactor_writemask
&
1096 else_tessfactor_writemask
;
1097 *cond_block_tf_writemask
|= then_tessfactor_writemask
|
1098 else_tessfactor_writemask
;
1103 tgsi_scan_tess_ctrl(const struct tgsi_token
*tokens
,
1104 const struct tgsi_shader_info
*info
,
1105 struct tgsi_tessctrl_info
*out
)
1107 memset(out
, 0, sizeof(*out
));
1109 if (info
->processor
!= PIPE_SHADER_TESS_CTRL
)
1112 struct tgsi_parse_context parse
;
1113 if (tgsi_parse_init(&parse
, tokens
) != TGSI_PARSE_OK
) {
1114 debug_printf("tgsi_parse_init() failed in tgsi_scan_arrays()!\n");
1118 /* The pass works as follows:
1119 * If all codepaths write tess factors, we can say that all invocations
1120 * define tess factors.
1122 * Each tess factor channel is tracked separately.
1124 unsigned main_block_tf_writemask
= 0; /* if main block writes tess factors */
1125 unsigned cond_block_tf_writemask
= 0; /* if cond block writes tess factors */
1127 /* Initial value = true. Here the pass will accumulate results from multiple
1128 * segments surrounded by barriers. If tess factors aren't written at all,
1129 * it's a shader bug and we don't care if this will be true.
1131 out
->tessfactors_are_def_in_all_invocs
= true;
1133 while (!tgsi_parse_end_of_tokens(&parse
)) {
1134 tgsi_parse_token(&parse
);
1136 if (parse
.FullToken
.Token
.Type
!= TGSI_TOKEN_TYPE_INSTRUCTION
)
1139 struct tgsi_full_instruction
*inst
= &parse
.FullToken
.FullInstruction
;
1140 check_no_subroutines(inst
);
1142 /* Process nested blocks. */
1143 switch (inst
->Instruction
.Opcode
) {
1144 case TGSI_OPCODE_IF
:
1145 case TGSI_OPCODE_UIF
:
1146 get_if_block_tessfactor_writemask(info
, &parse
,
1147 &main_block_tf_writemask
,
1148 &cond_block_tf_writemask
);
1151 case TGSI_OPCODE_BGNLOOP
:
1152 cond_block_tf_writemask
|=
1153 get_block_tessfactor_writemask(info
, &parse
, TGSI_OPCODE_ENDIF
);
1156 case TGSI_OPCODE_BARRIER
:
1157 /* The following case must be prevented:
1158 * gl_TessLevelInner = ...;
1160 * if (gl_InvocationID == 1)
1161 * gl_TessLevelInner = ...;
1163 * If you consider disjoint code segments separated by barriers, each
1164 * such segment that writes tess factor channels should write the same
1165 * channels in all codepaths within that segment.
1167 if (main_block_tf_writemask
|| cond_block_tf_writemask
) {
1168 /* Accumulate the result: */
1169 out
->tessfactors_are_def_in_all_invocs
&=
1170 !(cond_block_tf_writemask
& ~main_block_tf_writemask
);
1172 /* Analyze the next code segment from scratch. */
1173 main_block_tf_writemask
= 0;
1174 cond_block_tf_writemask
= 0;
1179 main_block_tf_writemask
|= get_inst_tessfactor_writemask(info
, inst
);
1182 /* Accumulate the result for the last code segment separated by a barrier. */
1183 if (main_block_tf_writemask
|| cond_block_tf_writemask
) {
1184 out
->tessfactors_are_def_in_all_invocs
&=
1185 !(cond_block_tf_writemask
& ~main_block_tf_writemask
);
1188 tgsi_parse_free(&parse
);