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_TXQ_LZ
||
64 opcode
== TGSI_OPCODE_LODQ
;
68 * Is the opcode a "true" texture instruction which samples from a
72 is_texture_inst(unsigned opcode
)
74 return (!is_mem_query_inst(opcode
) &&
75 tgsi_get_opcode_info(opcode
)->is_tex
);
80 * Is the opcode an instruction which computes a derivative explicitly or
84 computes_derivative(unsigned opcode
)
86 if (tgsi_get_opcode_info(opcode
)->is_tex
) {
87 return opcode
!= TGSI_OPCODE_TG4
&&
88 opcode
!= TGSI_OPCODE_TXD
&&
89 opcode
!= TGSI_OPCODE_TXF
&&
90 opcode
!= TGSI_OPCODE_TXL
&&
91 opcode
!= TGSI_OPCODE_TXL2
&&
92 opcode
!= TGSI_OPCODE_TXQ
&&
93 opcode
!= TGSI_OPCODE_TXQ_LZ
&&
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
,
111 bool is_interp_instruction
,
114 int ind
= src
->Register
.Index
;
116 /* Mark which inputs are effectively used */
117 if (src
->Register
.File
== TGSI_FILE_INPUT
) {
118 if (src
->Register
.Indirect
) {
119 for (ind
= 0; ind
< info
->num_inputs
; ++ind
) {
120 info
->input_usage_mask
[ind
] |= usage_mask
;
124 assert(ind
< PIPE_MAX_SHADER_INPUTS
);
125 info
->input_usage_mask
[ind
] |= usage_mask
;
128 if (info
->processor
== PIPE_SHADER_FRAGMENT
) {
129 unsigned name
, index
, input
;
131 if (src
->Register
.Indirect
&& src
->Indirect
.ArrayID
)
132 input
= info
->input_array_first
[src
->Indirect
.ArrayID
];
134 input
= src
->Register
.Index
;
136 name
= info
->input_semantic_name
[input
];
137 index
= info
->input_semantic_index
[input
];
139 if (name
== TGSI_SEMANTIC_POSITION
&&
140 (src
->Register
.SwizzleX
== TGSI_SWIZZLE_Z
||
141 src
->Register
.SwizzleY
== TGSI_SWIZZLE_Z
||
142 src
->Register
.SwizzleZ
== TGSI_SWIZZLE_Z
||
143 src
->Register
.SwizzleW
== TGSI_SWIZZLE_Z
))
144 info
->reads_z
= TRUE
;
146 if (name
== TGSI_SEMANTIC_COLOR
) {
148 (1 << src
->Register
.SwizzleX
) |
149 (1 << src
->Register
.SwizzleY
) |
150 (1 << src
->Register
.SwizzleZ
) |
151 (1 << src
->Register
.SwizzleW
);
153 info
->colors_read
|= mask
<< (index
* 4);
156 /* Process only interpolated varyings. Don't include POSITION.
157 * Don't include integer varyings, because they are not
158 * interpolated. Don't process inputs interpolated by INTERP
159 * opcodes. Those are tracked separately.
161 if ((!is_interp_instruction
|| src_index
!= 0) &&
162 (name
== TGSI_SEMANTIC_GENERIC
||
163 name
== TGSI_SEMANTIC_TEXCOORD
||
164 name
== TGSI_SEMANTIC_COLOR
||
165 name
== TGSI_SEMANTIC_BCOLOR
||
166 name
== TGSI_SEMANTIC_FOG
||
167 name
== TGSI_SEMANTIC_CLIPDIST
)) {
168 switch (info
->input_interpolate
[input
]) {
169 case TGSI_INTERPOLATE_COLOR
:
170 case TGSI_INTERPOLATE_PERSPECTIVE
:
171 switch (info
->input_interpolate_loc
[input
]) {
172 case TGSI_INTERPOLATE_LOC_CENTER
:
173 info
->uses_persp_center
= TRUE
;
175 case TGSI_INTERPOLATE_LOC_CENTROID
:
176 info
->uses_persp_centroid
= TRUE
;
178 case TGSI_INTERPOLATE_LOC_SAMPLE
:
179 info
->uses_persp_sample
= TRUE
;
183 case TGSI_INTERPOLATE_LINEAR
:
184 switch (info
->input_interpolate_loc
[input
]) {
185 case TGSI_INTERPOLATE_LOC_CENTER
:
186 info
->uses_linear_center
= TRUE
;
188 case TGSI_INTERPOLATE_LOC_CENTROID
:
189 info
->uses_linear_centroid
= TRUE
;
191 case TGSI_INTERPOLATE_LOC_SAMPLE
:
192 info
->uses_linear_sample
= TRUE
;
196 /* TGSI_INTERPOLATE_CONSTANT doesn't do any interpolation. */
202 if (info
->processor
== PIPE_SHADER_TESS_CTRL
&&
203 src
->Register
.File
== TGSI_FILE_OUTPUT
) {
206 if (src
->Register
.Indirect
&& src
->Indirect
.ArrayID
)
207 input
= info
->output_array_first
[src
->Indirect
.ArrayID
];
209 input
= src
->Register
.Index
;
211 switch (info
->output_semantic_name
[input
]) {
212 case TGSI_SEMANTIC_PATCH
:
213 info
->reads_perpatch_outputs
= true;
215 case TGSI_SEMANTIC_TESSINNER
:
216 case TGSI_SEMANTIC_TESSOUTER
:
217 info
->reads_tessfactor_outputs
= true;
220 info
->reads_pervertex_outputs
= true;
224 /* check for indirect register reads */
225 if (src
->Register
.Indirect
) {
226 info
->indirect_files
|= (1 << src
->Register
.File
);
227 info
->indirect_files_read
|= (1 << src
->Register
.File
);
229 /* record indirect constant buffer indexing */
230 if (src
->Register
.File
== TGSI_FILE_CONSTANT
) {
231 if (src
->Register
.Dimension
) {
232 if (src
->Dimension
.Indirect
)
233 info
->const_buffers_indirect
= info
->const_buffers_declared
;
235 info
->const_buffers_indirect
|= 1u << src
->Dimension
.Index
;
237 info
->const_buffers_indirect
|= 1;
242 if (src
->Register
.Dimension
&& src
->Dimension
.Indirect
)
243 info
->dim_indirect_files
|= 1u << src
->Register
.File
;
245 /* Texture samplers */
246 if (src
->Register
.File
== TGSI_FILE_SAMPLER
) {
247 const unsigned index
= src
->Register
.Index
;
249 assert(fullinst
->Instruction
.Texture
);
250 assert(index
< ARRAY_SIZE(info
->is_msaa_sampler
));
251 assert(index
< PIPE_MAX_SAMPLERS
);
253 if (is_texture_inst(fullinst
->Instruction
.Opcode
)) {
254 const unsigned target
= fullinst
->Texture
.Texture
;
255 assert(target
< TGSI_TEXTURE_UNKNOWN
);
256 /* for texture instructions, check that the texture instruction
257 * target matches the previous sampler view declaration (if there
260 if (info
->sampler_targets
[index
] == TGSI_TEXTURE_UNKNOWN
) {
261 /* probably no sampler view declaration */
262 info
->sampler_targets
[index
] = target
;
264 /* Make sure the texture instruction's sampler/target info
265 * agrees with the sampler view declaration.
267 assert(info
->sampler_targets
[index
] == target
);
270 if (target
== TGSI_TEXTURE_2D_MSAA
||
271 target
== TGSI_TEXTURE_2D_ARRAY_MSAA
) {
272 info
->is_msaa_sampler
[src
->Register
.Index
] = TRUE
;
277 if (is_memory_file(src
->Register
.File
) &&
278 !is_mem_query_inst(fullinst
->Instruction
.Opcode
)) {
281 if (tgsi_get_opcode_info(fullinst
->Instruction
.Opcode
)->is_store
) {
282 info
->writes_memory
= TRUE
;
284 if (src
->Register
.File
== TGSI_FILE_IMAGE
) {
285 if (src
->Register
.Indirect
)
286 info
->images_writemask
= info
->images_declared
;
288 info
->images_writemask
|= 1 << src
->Register
.Index
;
289 } else if (src
->Register
.File
== TGSI_FILE_BUFFER
) {
290 if (src
->Register
.Indirect
)
291 info
->shader_buffers_atomic
= info
->shader_buffers_declared
;
293 info
->shader_buffers_atomic
|= 1 << src
->Register
.Index
;
296 if (src
->Register
.File
== TGSI_FILE_BUFFER
) {
297 if (src
->Register
.Indirect
)
298 info
->shader_buffers_load
= info
->shader_buffers_declared
;
300 info
->shader_buffers_load
|= 1 << src
->Register
.Index
;
308 scan_instruction(struct tgsi_shader_info
*info
,
309 const struct tgsi_full_instruction
*fullinst
,
310 unsigned *current_depth
)
313 bool is_mem_inst
= false;
314 bool is_interp_instruction
= false;
316 assert(fullinst
->Instruction
.Opcode
< TGSI_OPCODE_LAST
);
317 info
->opcode_count
[fullinst
->Instruction
.Opcode
]++;
319 switch (fullinst
->Instruction
.Opcode
) {
321 case TGSI_OPCODE_UIF
:
322 case TGSI_OPCODE_BGNLOOP
:
324 info
->max_depth
= MAX2(info
->max_depth
, *current_depth
);
326 case TGSI_OPCODE_ENDIF
:
327 case TGSI_OPCODE_ENDLOOP
:
334 if (fullinst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_CENTROID
||
335 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_OFFSET
||
336 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_SAMPLE
) {
337 const struct tgsi_full_src_register
*src0
= &fullinst
->Src
[0];
340 is_interp_instruction
= true;
342 if (src0
->Register
.Indirect
&& src0
->Indirect
.ArrayID
)
343 input
= info
->input_array_first
[src0
->Indirect
.ArrayID
];
345 input
= src0
->Register
.Index
;
347 /* For the INTERP opcodes, the interpolation is always
348 * PERSPECTIVE unless LINEAR is specified.
350 switch (info
->input_interpolate
[input
]) {
351 case TGSI_INTERPOLATE_COLOR
:
352 case TGSI_INTERPOLATE_CONSTANT
:
353 case TGSI_INTERPOLATE_PERSPECTIVE
:
354 switch (fullinst
->Instruction
.Opcode
) {
355 case TGSI_OPCODE_INTERP_CENTROID
:
356 info
->uses_persp_opcode_interp_centroid
= TRUE
;
358 case TGSI_OPCODE_INTERP_OFFSET
:
359 info
->uses_persp_opcode_interp_offset
= TRUE
;
361 case TGSI_OPCODE_INTERP_SAMPLE
:
362 info
->uses_persp_opcode_interp_sample
= TRUE
;
367 case TGSI_INTERPOLATE_LINEAR
:
368 switch (fullinst
->Instruction
.Opcode
) {
369 case TGSI_OPCODE_INTERP_CENTROID
:
370 info
->uses_linear_opcode_interp_centroid
= TRUE
;
372 case TGSI_OPCODE_INTERP_OFFSET
:
373 info
->uses_linear_opcode_interp_offset
= TRUE
;
375 case TGSI_OPCODE_INTERP_SAMPLE
:
376 info
->uses_linear_opcode_interp_sample
= TRUE
;
383 if (fullinst
->Instruction
.Opcode
>= TGSI_OPCODE_F2D
&&
384 fullinst
->Instruction
.Opcode
<= TGSI_OPCODE_DSSG
)
385 info
->uses_doubles
= TRUE
;
387 for (i
= 0; i
< fullinst
->Instruction
.NumSrcRegs
; i
++) {
388 scan_src_operand(info
, fullinst
, &fullinst
->Src
[i
], i
,
389 tgsi_util_get_inst_usage_mask(fullinst
, i
),
390 is_interp_instruction
, &is_mem_inst
);
393 if (fullinst
->Instruction
.Texture
) {
394 for (i
= 0; i
< fullinst
->Texture
.NumOffsets
; i
++) {
395 struct tgsi_full_src_register src
= {{0}};
397 src
.Register
.File
= fullinst
->TexOffsets
[i
].File
;
398 src
.Register
.Index
= fullinst
->TexOffsets
[i
].Index
;
399 src
.Register
.SwizzleX
= fullinst
->TexOffsets
[i
].SwizzleX
;
400 src
.Register
.SwizzleY
= fullinst
->TexOffsets
[i
].SwizzleY
;
401 src
.Register
.SwizzleZ
= fullinst
->TexOffsets
[i
].SwizzleZ
;
403 /* The usage mask is suboptimal but should be safe. */
404 scan_src_operand(info
, fullinst
, &src
, 0, TGSI_WRITEMASK_XYZ
,
405 false, &is_mem_inst
);
409 /* check for indirect register writes */
410 for (i
= 0; i
< fullinst
->Instruction
.NumDstRegs
; i
++) {
411 const struct tgsi_full_dst_register
*dst
= &fullinst
->Dst
[i
];
412 if (dst
->Register
.Indirect
) {
413 info
->indirect_files
|= (1 << dst
->Register
.File
);
414 info
->indirect_files_written
|= (1 << dst
->Register
.File
);
417 if (dst
->Register
.Dimension
&& dst
->Dimension
.Indirect
)
418 info
->dim_indirect_files
|= 1u << dst
->Register
.File
;
420 if (is_memory_file(dst
->Register
.File
)) {
421 assert(fullinst
->Instruction
.Opcode
== TGSI_OPCODE_STORE
);
424 info
->writes_memory
= TRUE
;
426 if (dst
->Register
.File
== TGSI_FILE_IMAGE
) {
427 if (dst
->Register
.Indirect
)
428 info
->images_writemask
= info
->images_declared
;
430 info
->images_writemask
|= 1 << dst
->Register
.Index
;
431 } else if (dst
->Register
.File
== TGSI_FILE_BUFFER
) {
432 if (dst
->Register
.Indirect
)
433 info
->shader_buffers_store
= info
->shader_buffers_declared
;
435 info
->shader_buffers_store
|= 1 << dst
->Register
.Index
;
441 info
->num_memory_instructions
++;
443 if (computes_derivative(fullinst
->Instruction
.Opcode
))
444 info
->uses_derivatives
= true;
446 info
->num_instructions
++;
451 scan_declaration(struct tgsi_shader_info
*info
,
452 const struct tgsi_full_declaration
*fulldecl
)
454 const uint file
= fulldecl
->Declaration
.File
;
455 const unsigned procType
= info
->processor
;
458 if (fulldecl
->Declaration
.Array
) {
459 unsigned array_id
= fulldecl
->Array
.ArrayID
;
462 case TGSI_FILE_INPUT
:
463 assert(array_id
< ARRAY_SIZE(info
->input_array_first
));
464 info
->input_array_first
[array_id
] = fulldecl
->Range
.First
;
465 info
->input_array_last
[array_id
] = fulldecl
->Range
.Last
;
467 case TGSI_FILE_OUTPUT
:
468 assert(array_id
< ARRAY_SIZE(info
->output_array_first
));
469 info
->output_array_first
[array_id
] = fulldecl
->Range
.First
;
470 info
->output_array_last
[array_id
] = fulldecl
->Range
.Last
;
473 info
->array_max
[file
] = MAX2(info
->array_max
[file
], array_id
);
476 for (reg
= fulldecl
->Range
.First
; reg
<= fulldecl
->Range
.Last
; reg
++) {
477 unsigned semName
= fulldecl
->Semantic
.Name
;
478 unsigned semIndex
= fulldecl
->Semantic
.Index
+
479 (reg
- fulldecl
->Range
.First
);
481 unsigned index
, target
, type
;
483 /* only first 32 regs will appear in this bitfield */
484 info
->file_mask
[file
] |= (1 << reg
);
485 info
->file_count
[file
]++;
486 info
->file_max
[file
] = MAX2(info
->file_max
[file
], (int)reg
);
489 case TGSI_FILE_CONSTANT
:
492 if (fulldecl
->Declaration
.Dimension
)
493 buffer
= fulldecl
->Dim
.Index2D
;
495 info
->const_file_max
[buffer
] =
496 MAX2(info
->const_file_max
[buffer
], (int)reg
);
497 info
->const_buffers_declared
|= 1u << buffer
;
500 case TGSI_FILE_IMAGE
:
501 info
->images_declared
|= 1u << reg
;
502 if (fulldecl
->Image
.Resource
== TGSI_TEXTURE_BUFFER
)
503 info
->images_buffers
|= 1 << reg
;
506 case TGSI_FILE_BUFFER
:
507 info
->shader_buffers_declared
|= 1u << reg
;
510 case TGSI_FILE_INPUT
:
511 info
->input_semantic_name
[reg
] = (ubyte
) semName
;
512 info
->input_semantic_index
[reg
] = (ubyte
) semIndex
;
513 info
->input_interpolate
[reg
] = (ubyte
)fulldecl
->Interp
.Interpolate
;
514 info
->input_interpolate_loc
[reg
] = (ubyte
)fulldecl
->Interp
.Location
;
515 info
->input_cylindrical_wrap
[reg
] = (ubyte
)fulldecl
->Interp
.CylindricalWrap
;
517 /* Vertex shaders can have inputs with holes between them. */
518 info
->num_inputs
= MAX2(info
->num_inputs
, reg
+ 1);
520 if (semName
== TGSI_SEMANTIC_PRIMID
)
521 info
->uses_primid
= TRUE
;
522 else if (procType
== PIPE_SHADER_FRAGMENT
) {
523 if (semName
== TGSI_SEMANTIC_POSITION
)
524 info
->reads_position
= TRUE
;
525 else if (semName
== TGSI_SEMANTIC_FACE
)
526 info
->uses_frontface
= TRUE
;
530 case TGSI_FILE_SYSTEM_VALUE
:
531 index
= fulldecl
->Range
.First
;
533 info
->system_value_semantic_name
[index
] = semName
;
534 info
->num_system_values
= MAX2(info
->num_system_values
, index
+ 1);
537 case TGSI_SEMANTIC_INSTANCEID
:
538 info
->uses_instanceid
= TRUE
;
540 case TGSI_SEMANTIC_VERTEXID
:
541 info
->uses_vertexid
= TRUE
;
543 case TGSI_SEMANTIC_VERTEXID_NOBASE
:
544 info
->uses_vertexid_nobase
= TRUE
;
546 case TGSI_SEMANTIC_BASEVERTEX
:
547 info
->uses_basevertex
= TRUE
;
549 case TGSI_SEMANTIC_PRIMID
:
550 info
->uses_primid
= TRUE
;
552 case TGSI_SEMANTIC_INVOCATIONID
:
553 info
->uses_invocationid
= TRUE
;
555 case TGSI_SEMANTIC_POSITION
:
556 info
->reads_position
= TRUE
;
558 case TGSI_SEMANTIC_FACE
:
559 info
->uses_frontface
= TRUE
;
561 case TGSI_SEMANTIC_SAMPLEMASK
:
562 info
->reads_samplemask
= TRUE
;
564 case TGSI_SEMANTIC_TESSINNER
:
565 case TGSI_SEMANTIC_TESSOUTER
:
566 info
->reads_tess_factors
= true;
571 case TGSI_FILE_OUTPUT
:
572 info
->output_semantic_name
[reg
] = (ubyte
) semName
;
573 info
->output_semantic_index
[reg
] = (ubyte
) semIndex
;
574 info
->output_usagemask
[reg
] |= fulldecl
->Declaration
.UsageMask
;
575 info
->num_outputs
= MAX2(info
->num_outputs
, reg
+ 1);
577 if (fulldecl
->Declaration
.UsageMask
& TGSI_WRITEMASK_X
) {
578 info
->output_streams
[reg
] |= (ubyte
)fulldecl
->Semantic
.StreamX
;
579 info
->num_stream_output_components
[fulldecl
->Semantic
.StreamX
]++;
581 if (fulldecl
->Declaration
.UsageMask
& TGSI_WRITEMASK_Y
) {
582 info
->output_streams
[reg
] |= (ubyte
)fulldecl
->Semantic
.StreamY
<< 2;
583 info
->num_stream_output_components
[fulldecl
->Semantic
.StreamY
]++;
585 if (fulldecl
->Declaration
.UsageMask
& TGSI_WRITEMASK_Z
) {
586 info
->output_streams
[reg
] |= (ubyte
)fulldecl
->Semantic
.StreamZ
<< 4;
587 info
->num_stream_output_components
[fulldecl
->Semantic
.StreamZ
]++;
589 if (fulldecl
->Declaration
.UsageMask
& TGSI_WRITEMASK_W
) {
590 info
->output_streams
[reg
] |= (ubyte
)fulldecl
->Semantic
.StreamW
<< 6;
591 info
->num_stream_output_components
[fulldecl
->Semantic
.StreamW
]++;
595 case TGSI_SEMANTIC_PRIMID
:
596 info
->writes_primid
= true;
598 case TGSI_SEMANTIC_VIEWPORT_INDEX
:
599 info
->writes_viewport_index
= true;
601 case TGSI_SEMANTIC_LAYER
:
602 info
->writes_layer
= true;
604 case TGSI_SEMANTIC_PSIZE
:
605 info
->writes_psize
= true;
607 case TGSI_SEMANTIC_CLIPVERTEX
:
608 info
->writes_clipvertex
= true;
610 case TGSI_SEMANTIC_COLOR
:
611 info
->colors_written
|= 1 << semIndex
;
613 case TGSI_SEMANTIC_STENCIL
:
614 info
->writes_stencil
= true;
616 case TGSI_SEMANTIC_SAMPLEMASK
:
617 info
->writes_samplemask
= true;
619 case TGSI_SEMANTIC_EDGEFLAG
:
620 info
->writes_edgeflag
= true;
622 case TGSI_SEMANTIC_POSITION
:
623 if (procType
== PIPE_SHADER_FRAGMENT
)
624 info
->writes_z
= true;
626 info
->writes_position
= true;
631 case TGSI_FILE_SAMPLER
:
632 STATIC_ASSERT(sizeof(info
->samplers_declared
) * 8 >= PIPE_MAX_SAMPLERS
);
633 info
->samplers_declared
|= 1u << reg
;
636 case TGSI_FILE_SAMPLER_VIEW
:
637 target
= fulldecl
->SamplerView
.Resource
;
638 type
= fulldecl
->SamplerView
.ReturnTypeX
;
640 assert(target
< TGSI_TEXTURE_UNKNOWN
);
641 if (info
->sampler_targets
[reg
] == TGSI_TEXTURE_UNKNOWN
) {
642 /* Save sampler target for this sampler index */
643 info
->sampler_targets
[reg
] = target
;
644 info
->sampler_type
[reg
] = type
;
646 /* if previously declared, make sure targets agree */
647 assert(info
->sampler_targets
[reg
] == target
);
648 assert(info
->sampler_type
[reg
] == type
);
657 scan_immediate(struct tgsi_shader_info
*info
)
659 uint reg
= info
->immediate_count
++;
660 uint file
= TGSI_FILE_IMMEDIATE
;
662 info
->file_mask
[file
] |= (1 << reg
);
663 info
->file_count
[file
]++;
664 info
->file_max
[file
] = MAX2(info
->file_max
[file
], (int)reg
);
669 scan_property(struct tgsi_shader_info
*info
,
670 const struct tgsi_full_property
*fullprop
)
672 unsigned name
= fullprop
->Property
.PropertyName
;
673 unsigned value
= fullprop
->u
[0].Data
;
675 assert(name
< ARRAY_SIZE(info
->properties
));
676 info
->properties
[name
] = value
;
679 case TGSI_PROPERTY_NUM_CLIPDIST_ENABLED
:
680 info
->num_written_clipdistance
= value
;
681 info
->clipdist_writemask
|= (1 << value
) - 1;
683 case TGSI_PROPERTY_NUM_CULLDIST_ENABLED
:
684 info
->num_written_culldistance
= value
;
685 info
->culldist_writemask
|= (1 << value
) - 1;
692 * Scan the given TGSI shader to collect information such as number of
693 * registers used, special instructions used, etc.
694 * \return info the result of the scan
697 tgsi_scan_shader(const struct tgsi_token
*tokens
,
698 struct tgsi_shader_info
*info
)
701 struct tgsi_parse_context parse
;
702 unsigned current_depth
= 0;
704 memset(info
, 0, sizeof(*info
));
705 for (i
= 0; i
< TGSI_FILE_COUNT
; i
++)
706 info
->file_max
[i
] = -1;
707 for (i
= 0; i
< ARRAY_SIZE(info
->const_file_max
); i
++)
708 info
->const_file_max
[i
] = -1;
709 info
->properties
[TGSI_PROPERTY_GS_INVOCATIONS
] = 1;
710 for (i
= 0; i
< ARRAY_SIZE(info
->sampler_targets
); i
++)
711 info
->sampler_targets
[i
] = TGSI_TEXTURE_UNKNOWN
;
714 ** Setup to begin parsing input shader
716 if (tgsi_parse_init( &parse
, tokens
) != TGSI_PARSE_OK
) {
717 debug_printf("tgsi_parse_init() failed in tgsi_scan_shader()!\n");
720 procType
= parse
.FullHeader
.Processor
.Processor
;
721 assert(procType
== PIPE_SHADER_FRAGMENT
||
722 procType
== PIPE_SHADER_VERTEX
||
723 procType
== PIPE_SHADER_GEOMETRY
||
724 procType
== PIPE_SHADER_TESS_CTRL
||
725 procType
== PIPE_SHADER_TESS_EVAL
||
726 procType
== PIPE_SHADER_COMPUTE
);
727 info
->processor
= procType
;
730 ** Loop over incoming program tokens/instructions
732 while (!tgsi_parse_end_of_tokens(&parse
)) {
735 tgsi_parse_token( &parse
);
737 switch( parse
.FullToken
.Token
.Type
) {
738 case TGSI_TOKEN_TYPE_INSTRUCTION
:
739 scan_instruction(info
, &parse
.FullToken
.FullInstruction
,
742 case TGSI_TOKEN_TYPE_DECLARATION
:
743 scan_declaration(info
, &parse
.FullToken
.FullDeclaration
);
745 case TGSI_TOKEN_TYPE_IMMEDIATE
:
746 scan_immediate(info
);
748 case TGSI_TOKEN_TYPE_PROPERTY
:
749 scan_property(info
, &parse
.FullToken
.FullProperty
);
752 assert(!"Unexpected TGSI token type");
756 info
->uses_kill
= (info
->opcode_count
[TGSI_OPCODE_KILL_IF
] ||
757 info
->opcode_count
[TGSI_OPCODE_KILL
]);
759 /* The dimensions of the IN decleration in geometry shader have
760 * to be deduced from the type of the input primitive.
762 if (procType
== PIPE_SHADER_GEOMETRY
) {
763 unsigned input_primitive
=
764 info
->properties
[TGSI_PROPERTY_GS_INPUT_PRIM
];
765 int num_verts
= u_vertices_per_prim(input_primitive
);
767 info
->file_count
[TGSI_FILE_INPUT
] = num_verts
;
768 info
->file_max
[TGSI_FILE_INPUT
] =
769 MAX2(info
->file_max
[TGSI_FILE_INPUT
], num_verts
- 1);
770 for (j
= 0; j
< num_verts
; ++j
) {
771 info
->file_mask
[TGSI_FILE_INPUT
] |= (1 << j
);
775 tgsi_parse_free(&parse
);
779 * Collect information about the arrays of a given register file.
781 * @param tokens TGSI shader
782 * @param file the register file to scan through
783 * @param max_array_id number of entries in @p arrays; should be equal to the
784 * highest array id, i.e. tgsi_shader_info::array_max[file].
785 * @param arrays info for array of each ID will be written to arrays[ID - 1].
788 tgsi_scan_arrays(const struct tgsi_token
*tokens
,
790 unsigned max_array_id
,
791 struct tgsi_array_info
*arrays
)
793 struct tgsi_parse_context parse
;
795 if (tgsi_parse_init(&parse
, tokens
) != TGSI_PARSE_OK
) {
796 debug_printf("tgsi_parse_init() failed in tgsi_scan_arrays()!\n");
800 memset(arrays
, 0, sizeof(arrays
[0]) * max_array_id
);
802 while (!tgsi_parse_end_of_tokens(&parse
)) {
803 struct tgsi_full_instruction
*inst
;
805 tgsi_parse_token(&parse
);
807 if (parse
.FullToken
.Token
.Type
== TGSI_TOKEN_TYPE_DECLARATION
) {
808 struct tgsi_full_declaration
*decl
= &parse
.FullToken
.FullDeclaration
;
810 if (decl
->Declaration
.Array
&& decl
->Declaration
.File
== file
&&
811 decl
->Array
.ArrayID
> 0 && decl
->Array
.ArrayID
<= max_array_id
) {
812 struct tgsi_array_info
*array
= &arrays
[decl
->Array
.ArrayID
- 1];
813 assert(!array
->declared
);
814 array
->declared
= true;
815 array
->range
= decl
->Range
;
819 if (parse
.FullToken
.Token
.Type
!= TGSI_TOKEN_TYPE_INSTRUCTION
)
822 inst
= &parse
.FullToken
.FullInstruction
;
823 for (unsigned i
= 0; i
< inst
->Instruction
.NumDstRegs
; i
++) {
824 const struct tgsi_full_dst_register
*dst
= &inst
->Dst
[i
];
825 if (dst
->Register
.File
!= file
)
828 if (dst
->Register
.Indirect
) {
829 if (dst
->Indirect
.ArrayID
> 0 &&
830 dst
->Indirect
.ArrayID
<= max_array_id
) {
831 arrays
[dst
->Indirect
.ArrayID
- 1].writemask
|= dst
->Register
.WriteMask
;
833 /* Indirect writes without an ArrayID can write anywhere. */
834 for (unsigned j
= 0; j
< max_array_id
; ++j
)
835 arrays
[j
].writemask
|= dst
->Register
.WriteMask
;
838 /* Check whether the write falls into any of the arrays anyway. */
839 for (unsigned j
= 0; j
< max_array_id
; ++j
) {
840 struct tgsi_array_info
*array
= &arrays
[j
];
841 if (array
->declared
&&
842 dst
->Register
.Index
>= array
->range
.First
&&
843 dst
->Register
.Index
<= array
->range
.Last
)
844 array
->writemask
|= dst
->Register
.WriteMask
;
850 tgsi_parse_free(&parse
);
857 * Check if the given shader is a "passthrough" shader consisting of only
858 * MOV instructions of the form: MOV OUT[n], IN[n]
862 tgsi_is_passthrough_shader(const struct tgsi_token
*tokens
)
864 struct tgsi_parse_context parse
;
867 ** Setup to begin parsing input shader
869 if (tgsi_parse_init(&parse
, tokens
) != TGSI_PARSE_OK
) {
870 debug_printf("tgsi_parse_init() failed in tgsi_is_passthrough_shader()!\n");
875 ** Loop over incoming program tokens/instructions
877 while (!tgsi_parse_end_of_tokens(&parse
)) {
879 tgsi_parse_token(&parse
);
881 switch (parse
.FullToken
.Token
.Type
) {
882 case TGSI_TOKEN_TYPE_INSTRUCTION
:
884 struct tgsi_full_instruction
*fullinst
=
885 &parse
.FullToken
.FullInstruction
;
886 const struct tgsi_full_src_register
*src
=
888 const struct tgsi_full_dst_register
*dst
=
891 /* Do a whole bunch of checks for a simple move */
892 if (fullinst
->Instruction
.Opcode
!= TGSI_OPCODE_MOV
||
893 (src
->Register
.File
!= TGSI_FILE_INPUT
&&
894 src
->Register
.File
!= TGSI_FILE_SYSTEM_VALUE
) ||
895 dst
->Register
.File
!= TGSI_FILE_OUTPUT
||
896 src
->Register
.Index
!= dst
->Register
.Index
||
898 src
->Register
.Negate
||
899 src
->Register
.Absolute
||
901 src
->Register
.SwizzleX
!= TGSI_SWIZZLE_X
||
902 src
->Register
.SwizzleY
!= TGSI_SWIZZLE_Y
||
903 src
->Register
.SwizzleZ
!= TGSI_SWIZZLE_Z
||
904 src
->Register
.SwizzleW
!= TGSI_SWIZZLE_W
||
906 dst
->Register
.WriteMask
!= TGSI_WRITEMASK_XYZW
)
908 tgsi_parse_free(&parse
);
914 case TGSI_TOKEN_TYPE_DECLARATION
:
916 case TGSI_TOKEN_TYPE_IMMEDIATE
:
918 case TGSI_TOKEN_TYPE_PROPERTY
:
925 tgsi_parse_free(&parse
);
927 /* if we get here, it's a pass-through shader */