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
||
54 file
== TGSI_FILE_HW_ATOMIC
;
59 is_mem_query_inst(enum tgsi_opcode opcode
)
61 return opcode
== TGSI_OPCODE_RESQ
||
62 opcode
== TGSI_OPCODE_TXQ
||
63 opcode
== TGSI_OPCODE_TXQS
||
64 opcode
== TGSI_OPCODE_LODQ
;
68 * Is the opcode a "true" texture instruction which samples from a
72 is_texture_inst(enum tgsi_opcode 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(enum tgsi_opcode 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_TXF_LZ
&&
91 opcode
!= TGSI_OPCODE_TEX_LZ
&&
92 opcode
!= TGSI_OPCODE_TXL
&&
93 opcode
!= TGSI_OPCODE_TXL2
&&
94 opcode
!= TGSI_OPCODE_TXQ
&&
95 opcode
!= TGSI_OPCODE_TXQS
;
98 return opcode
== TGSI_OPCODE_DDX
|| opcode
== TGSI_OPCODE_DDX_FINE
||
99 opcode
== TGSI_OPCODE_DDY
|| opcode
== TGSI_OPCODE_DDY_FINE
||
100 opcode
== TGSI_OPCODE_SAMPLE
||
101 opcode
== TGSI_OPCODE_SAMPLE_B
||
102 opcode
== TGSI_OPCODE_SAMPLE_C
;
107 scan_src_operand(struct tgsi_shader_info
*info
,
108 const struct tgsi_full_instruction
*fullinst
,
109 const struct tgsi_full_src_register
*src
,
111 unsigned usage_mask_after_swizzle
,
112 bool is_interp_instruction
,
115 int ind
= src
->Register
.Index
;
117 if (info
->processor
== PIPE_SHADER_COMPUTE
&&
118 src
->Register
.File
== TGSI_FILE_SYSTEM_VALUE
) {
121 name
= info
->system_value_semantic_name
[src
->Register
.Index
];
124 case TGSI_SEMANTIC_THREAD_ID
:
125 case TGSI_SEMANTIC_BLOCK_ID
:
126 mask
= usage_mask_after_swizzle
& TGSI_WRITEMASK_XYZ
;
128 unsigned i
= u_bit_scan(&mask
);
130 if (name
== TGSI_SEMANTIC_THREAD_ID
)
131 info
->uses_thread_id
[i
] = true;
133 info
->uses_block_id
[i
] = true;
136 case TGSI_SEMANTIC_BLOCK_SIZE
:
137 /* The block size is translated to IMM with a fixed block size. */
138 if (info
->properties
[TGSI_PROPERTY_CS_FIXED_BLOCK_WIDTH
] == 0)
139 info
->uses_block_size
= true;
141 case TGSI_SEMANTIC_GRID_SIZE
:
142 info
->uses_grid_size
= true;
147 /* Mark which inputs are effectively used */
148 if (src
->Register
.File
== TGSI_FILE_INPUT
) {
149 if (src
->Register
.Indirect
) {
150 for (ind
= 0; ind
< info
->num_inputs
; ++ind
) {
151 info
->input_usage_mask
[ind
] |= usage_mask_after_swizzle
;
155 assert(ind
< PIPE_MAX_SHADER_INPUTS
);
156 info
->input_usage_mask
[ind
] |= usage_mask_after_swizzle
;
159 if (info
->processor
== PIPE_SHADER_FRAGMENT
) {
160 unsigned name
, index
, input
;
162 if (src
->Register
.Indirect
&& src
->Indirect
.ArrayID
)
163 input
= info
->input_array_first
[src
->Indirect
.ArrayID
];
165 input
= src
->Register
.Index
;
167 name
= info
->input_semantic_name
[input
];
168 index
= info
->input_semantic_index
[input
];
170 if (name
== TGSI_SEMANTIC_POSITION
&&
171 usage_mask_after_swizzle
& TGSI_WRITEMASK_Z
)
172 info
->reads_z
= true;
174 if (name
== TGSI_SEMANTIC_COLOR
)
175 info
->colors_read
|= usage_mask_after_swizzle
<< (index
* 4);
177 /* Process only interpolated varyings. Don't include POSITION.
178 * Don't include integer varyings, because they are not
179 * interpolated. Don't process inputs interpolated by INTERP
180 * opcodes. Those are tracked separately.
182 if ((!is_interp_instruction
|| src_index
!= 0) &&
183 (name
== TGSI_SEMANTIC_GENERIC
||
184 name
== TGSI_SEMANTIC_TEXCOORD
||
185 name
== TGSI_SEMANTIC_COLOR
||
186 name
== TGSI_SEMANTIC_BCOLOR
||
187 name
== TGSI_SEMANTIC_FOG
||
188 name
== TGSI_SEMANTIC_CLIPDIST
)) {
189 switch (info
->input_interpolate
[input
]) {
190 case TGSI_INTERPOLATE_COLOR
:
191 case TGSI_INTERPOLATE_PERSPECTIVE
:
192 switch (info
->input_interpolate_loc
[input
]) {
193 case TGSI_INTERPOLATE_LOC_CENTER
:
194 info
->uses_persp_center
= TRUE
;
196 case TGSI_INTERPOLATE_LOC_CENTROID
:
197 info
->uses_persp_centroid
= TRUE
;
199 case TGSI_INTERPOLATE_LOC_SAMPLE
:
200 info
->uses_persp_sample
= TRUE
;
204 case TGSI_INTERPOLATE_LINEAR
:
205 switch (info
->input_interpolate_loc
[input
]) {
206 case TGSI_INTERPOLATE_LOC_CENTER
:
207 info
->uses_linear_center
= TRUE
;
209 case TGSI_INTERPOLATE_LOC_CENTROID
:
210 info
->uses_linear_centroid
= TRUE
;
212 case TGSI_INTERPOLATE_LOC_SAMPLE
:
213 info
->uses_linear_sample
= TRUE
;
217 /* TGSI_INTERPOLATE_CONSTANT doesn't do any interpolation. */
223 if (info
->processor
== PIPE_SHADER_TESS_CTRL
&&
224 src
->Register
.File
== TGSI_FILE_OUTPUT
) {
227 if (src
->Register
.Indirect
&& src
->Indirect
.ArrayID
)
228 input
= info
->output_array_first
[src
->Indirect
.ArrayID
];
230 input
= src
->Register
.Index
;
232 switch (info
->output_semantic_name
[input
]) {
233 case TGSI_SEMANTIC_PATCH
:
234 info
->reads_perpatch_outputs
= true;
236 case TGSI_SEMANTIC_TESSINNER
:
237 case TGSI_SEMANTIC_TESSOUTER
:
238 info
->reads_tessfactor_outputs
= true;
241 info
->reads_pervertex_outputs
= true;
245 /* check for indirect register reads */
246 if (src
->Register
.Indirect
) {
247 info
->indirect_files
|= (1 << src
->Register
.File
);
248 info
->indirect_files_read
|= (1 << src
->Register
.File
);
250 /* record indirect constant buffer indexing */
251 if (src
->Register
.File
== TGSI_FILE_CONSTANT
) {
252 if (src
->Register
.Dimension
) {
253 if (src
->Dimension
.Indirect
)
254 info
->const_buffers_indirect
= info
->const_buffers_declared
;
256 info
->const_buffers_indirect
|= 1u << src
->Dimension
.Index
;
258 info
->const_buffers_indirect
|= 1;
263 if (src
->Register
.Dimension
&& src
->Dimension
.Indirect
)
264 info
->dim_indirect_files
|= 1u << src
->Register
.File
;
266 /* Texture samplers */
267 if (src
->Register
.File
== TGSI_FILE_SAMPLER
) {
268 const unsigned index
= src
->Register
.Index
;
270 assert(fullinst
->Instruction
.Texture
);
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
);
292 if (is_memory_file(src
->Register
.File
) &&
293 !is_mem_query_inst(fullinst
->Instruction
.Opcode
)) {
296 if (tgsi_get_opcode_info(fullinst
->Instruction
.Opcode
)->is_store
) {
297 info
->writes_memory
= TRUE
;
299 if (src
->Register
.File
== TGSI_FILE_IMAGE
) {
300 if (src
->Register
.Indirect
)
301 info
->images_atomic
= info
->images_declared
;
303 info
->images_atomic
|= 1 << src
->Register
.Index
;
304 } else if (src
->Register
.File
== TGSI_FILE_BUFFER
) {
305 if (src
->Register
.Indirect
)
306 info
->shader_buffers_atomic
= info
->shader_buffers_declared
;
308 info
->shader_buffers_atomic
|= 1 << src
->Register
.Index
;
311 if (src
->Register
.File
== TGSI_FILE_IMAGE
) {
312 if (src
->Register
.Indirect
)
313 info
->images_load
= info
->images_declared
;
315 info
->images_load
|= 1 << src
->Register
.Index
;
316 } else if (src
->Register
.File
== TGSI_FILE_BUFFER
) {
317 if (src
->Register
.Indirect
)
318 info
->shader_buffers_load
= info
->shader_buffers_declared
;
320 info
->shader_buffers_load
|= 1 << src
->Register
.Index
;
328 scan_instruction(struct tgsi_shader_info
*info
,
329 const struct tgsi_full_instruction
*fullinst
,
330 unsigned *current_depth
)
333 bool is_mem_inst
= false;
334 bool is_interp_instruction
= false;
335 unsigned sampler_src
;
337 assert(fullinst
->Instruction
.Opcode
< TGSI_OPCODE_LAST
);
338 info
->opcode_count
[fullinst
->Instruction
.Opcode
]++;
340 switch (fullinst
->Instruction
.Opcode
) {
342 case TGSI_OPCODE_UIF
:
343 case TGSI_OPCODE_BGNLOOP
:
345 info
->max_depth
= MAX2(info
->max_depth
, *current_depth
);
347 case TGSI_OPCODE_ENDIF
:
348 case TGSI_OPCODE_ENDLOOP
:
351 case TGSI_OPCODE_TEX
:
352 case TGSI_OPCODE_TEX_LZ
:
353 case TGSI_OPCODE_TXB
:
354 case TGSI_OPCODE_TXD
:
355 case TGSI_OPCODE_TXL
:
356 case TGSI_OPCODE_TXP
:
357 case TGSI_OPCODE_TXQ
:
358 case TGSI_OPCODE_TXQS
:
359 case TGSI_OPCODE_TXF
:
360 case TGSI_OPCODE_TXF_LZ
:
361 case TGSI_OPCODE_TEX2
:
362 case TGSI_OPCODE_TXB2
:
363 case TGSI_OPCODE_TXL2
:
364 case TGSI_OPCODE_TG4
:
365 case TGSI_OPCODE_LODQ
:
366 sampler_src
= fullinst
->Instruction
.NumSrcRegs
- 1;
367 if (fullinst
->Src
[sampler_src
].Register
.File
!= TGSI_FILE_SAMPLER
)
368 info
->uses_bindless_samplers
= true;
370 case TGSI_OPCODE_RESQ
:
371 if (tgsi_is_bindless_image_file(fullinst
->Src
[0].Register
.File
))
372 info
->uses_bindless_images
= true;
374 case TGSI_OPCODE_LOAD
:
375 if (tgsi_is_bindless_image_file(fullinst
->Src
[0].Register
.File
)) {
376 info
->uses_bindless_images
= true;
378 if (fullinst
->Memory
.Texture
== TGSI_TEXTURE_BUFFER
)
379 info
->uses_bindless_buffer_load
= true;
381 info
->uses_bindless_image_load
= true;
384 case TGSI_OPCODE_ATOMUADD
:
385 case TGSI_OPCODE_ATOMXCHG
:
386 case TGSI_OPCODE_ATOMCAS
:
387 case TGSI_OPCODE_ATOMAND
:
388 case TGSI_OPCODE_ATOMOR
:
389 case TGSI_OPCODE_ATOMXOR
:
390 case TGSI_OPCODE_ATOMUMIN
:
391 case TGSI_OPCODE_ATOMUMAX
:
392 case TGSI_OPCODE_ATOMIMIN
:
393 case TGSI_OPCODE_ATOMIMAX
:
394 case TGSI_OPCODE_ATOMFADD
:
395 if (tgsi_is_bindless_image_file(fullinst
->Src
[0].Register
.File
)) {
396 info
->uses_bindless_images
= true;
398 if (fullinst
->Memory
.Texture
== TGSI_TEXTURE_BUFFER
)
399 info
->uses_bindless_buffer_atomic
= true;
401 info
->uses_bindless_image_atomic
= true;
404 case TGSI_OPCODE_STORE
:
405 if (tgsi_is_bindless_image_file(fullinst
->Dst
[0].Register
.File
)) {
406 info
->uses_bindless_images
= true;
408 if (fullinst
->Memory
.Texture
== TGSI_TEXTURE_BUFFER
)
409 info
->uses_bindless_buffer_store
= true;
411 info
->uses_bindless_image_store
= true;
418 if (fullinst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_CENTROID
||
419 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_OFFSET
||
420 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_SAMPLE
) {
421 const struct tgsi_full_src_register
*src0
= &fullinst
->Src
[0];
424 is_interp_instruction
= true;
426 if (src0
->Register
.Indirect
&& src0
->Indirect
.ArrayID
)
427 input
= info
->input_array_first
[src0
->Indirect
.ArrayID
];
429 input
= src0
->Register
.Index
;
431 /* For the INTERP opcodes, the interpolation is always
432 * PERSPECTIVE unless LINEAR is specified.
434 switch (info
->input_interpolate
[input
]) {
435 case TGSI_INTERPOLATE_COLOR
:
436 case TGSI_INTERPOLATE_CONSTANT
:
437 case TGSI_INTERPOLATE_PERSPECTIVE
:
438 switch (fullinst
->Instruction
.Opcode
) {
439 case TGSI_OPCODE_INTERP_CENTROID
:
440 info
->uses_persp_opcode_interp_centroid
= TRUE
;
442 case TGSI_OPCODE_INTERP_OFFSET
:
443 info
->uses_persp_opcode_interp_offset
= TRUE
;
445 case TGSI_OPCODE_INTERP_SAMPLE
:
446 info
->uses_persp_opcode_interp_sample
= TRUE
;
451 case TGSI_INTERPOLATE_LINEAR
:
452 switch (fullinst
->Instruction
.Opcode
) {
453 case TGSI_OPCODE_INTERP_CENTROID
:
454 info
->uses_linear_opcode_interp_centroid
= TRUE
;
456 case TGSI_OPCODE_INTERP_OFFSET
:
457 info
->uses_linear_opcode_interp_offset
= TRUE
;
459 case TGSI_OPCODE_INTERP_SAMPLE
:
460 info
->uses_linear_opcode_interp_sample
= TRUE
;
467 if ((fullinst
->Instruction
.Opcode
>= TGSI_OPCODE_F2D
&&
468 fullinst
->Instruction
.Opcode
<= TGSI_OPCODE_DSSG
) ||
469 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_DFMA
||
470 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_DDIV
||
471 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_D2U64
||
472 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_D2I64
||
473 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_U642D
||
474 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_I642D
)
475 info
->uses_doubles
= TRUE
;
477 for (i
= 0; i
< fullinst
->Instruction
.NumSrcRegs
; i
++) {
478 scan_src_operand(info
, fullinst
, &fullinst
->Src
[i
], i
,
479 tgsi_util_get_inst_usage_mask(fullinst
, i
),
480 is_interp_instruction
, &is_mem_inst
);
482 if (fullinst
->Src
[i
].Register
.Indirect
) {
483 struct tgsi_full_src_register src
= {{0}};
485 src
.Register
.File
= fullinst
->Src
[i
].Indirect
.File
;
486 src
.Register
.Index
= fullinst
->Src
[i
].Indirect
.Index
;
488 scan_src_operand(info
, fullinst
, &src
, -1,
489 1 << fullinst
->Src
[i
].Indirect
.Swizzle
,
493 if (fullinst
->Src
[i
].Register
.Dimension
&&
494 fullinst
->Src
[i
].Dimension
.Indirect
) {
495 struct tgsi_full_src_register src
= {{0}};
497 src
.Register
.File
= fullinst
->Src
[i
].DimIndirect
.File
;
498 src
.Register
.Index
= fullinst
->Src
[i
].DimIndirect
.Index
;
500 scan_src_operand(info
, fullinst
, &src
, -1,
501 1 << fullinst
->Src
[i
].DimIndirect
.Swizzle
,
506 if (fullinst
->Instruction
.Texture
) {
507 for (i
= 0; i
< fullinst
->Texture
.NumOffsets
; i
++) {
508 struct tgsi_full_src_register src
= {{0}};
510 src
.Register
.File
= fullinst
->TexOffsets
[i
].File
;
511 src
.Register
.Index
= fullinst
->TexOffsets
[i
].Index
;
513 /* The usage mask is suboptimal but should be safe. */
514 scan_src_operand(info
, fullinst
, &src
, -1,
515 (1 << fullinst
->TexOffsets
[i
].SwizzleX
) |
516 (1 << fullinst
->TexOffsets
[i
].SwizzleY
) |
517 (1 << fullinst
->TexOffsets
[i
].SwizzleZ
),
518 false, &is_mem_inst
);
522 /* check for indirect register writes */
523 for (i
= 0; i
< fullinst
->Instruction
.NumDstRegs
; i
++) {
524 const struct tgsi_full_dst_register
*dst
= &fullinst
->Dst
[i
];
526 if (dst
->Register
.Indirect
) {
527 struct tgsi_full_src_register src
= {{0}};
529 src
.Register
.File
= dst
->Indirect
.File
;
530 src
.Register
.Index
= dst
->Indirect
.Index
;
532 scan_src_operand(info
, fullinst
, &src
, -1,
533 1 << dst
->Indirect
.Swizzle
, false, NULL
);
535 info
->indirect_files
|= (1 << dst
->Register
.File
);
536 info
->indirect_files_written
|= (1 << dst
->Register
.File
);
539 if (dst
->Register
.Dimension
&& dst
->Dimension
.Indirect
) {
540 struct tgsi_full_src_register src
= {{0}};
542 src
.Register
.File
= dst
->DimIndirect
.File
;
543 src
.Register
.Index
= dst
->DimIndirect
.Index
;
545 scan_src_operand(info
, fullinst
, &src
, -1,
546 1 << dst
->DimIndirect
.Swizzle
, false, NULL
);
548 info
->dim_indirect_files
|= 1u << dst
->Register
.File
;
551 if (is_memory_file(dst
->Register
.File
)) {
552 assert(fullinst
->Instruction
.Opcode
== TGSI_OPCODE_STORE
);
555 info
->writes_memory
= TRUE
;
557 if (dst
->Register
.File
== TGSI_FILE_IMAGE
) {
558 if (dst
->Register
.Indirect
)
559 info
->images_store
= info
->images_declared
;
561 info
->images_store
|= 1 << dst
->Register
.Index
;
562 } else if (dst
->Register
.File
== TGSI_FILE_BUFFER
) {
563 if (dst
->Register
.Indirect
)
564 info
->shader_buffers_store
= info
->shader_buffers_declared
;
566 info
->shader_buffers_store
|= 1 << dst
->Register
.Index
;
572 info
->num_memory_instructions
++;
574 if (computes_derivative(fullinst
->Instruction
.Opcode
))
575 info
->uses_derivatives
= true;
577 info
->num_instructions
++;
582 scan_declaration(struct tgsi_shader_info
*info
,
583 const struct tgsi_full_declaration
*fulldecl
)
585 const uint file
= fulldecl
->Declaration
.File
;
586 const unsigned procType
= info
->processor
;
589 if (fulldecl
->Declaration
.Array
) {
590 unsigned array_id
= fulldecl
->Array
.ArrayID
;
593 case TGSI_FILE_INPUT
:
594 assert(array_id
< ARRAY_SIZE(info
->input_array_first
));
595 info
->input_array_first
[array_id
] = fulldecl
->Range
.First
;
596 info
->input_array_last
[array_id
] = fulldecl
->Range
.Last
;
598 case TGSI_FILE_OUTPUT
:
599 assert(array_id
< ARRAY_SIZE(info
->output_array_first
));
600 info
->output_array_first
[array_id
] = fulldecl
->Range
.First
;
601 info
->output_array_last
[array_id
] = fulldecl
->Range
.Last
;
604 info
->array_max
[file
] = MAX2(info
->array_max
[file
], array_id
);
607 for (reg
= fulldecl
->Range
.First
; reg
<= fulldecl
->Range
.Last
; reg
++) {
608 unsigned semName
= fulldecl
->Semantic
.Name
;
609 unsigned semIndex
= fulldecl
->Semantic
.Index
+
610 (reg
- fulldecl
->Range
.First
);
612 unsigned index
, target
, type
;
615 * only first 32 regs will appear in this bitfield, if larger
616 * bits will wrap around.
618 info
->file_mask
[file
] |= (1u << (reg
& 31));
619 info
->file_count
[file
]++;
620 info
->file_max
[file
] = MAX2(info
->file_max
[file
], (int)reg
);
623 case TGSI_FILE_CONSTANT
:
626 if (fulldecl
->Declaration
.Dimension
)
627 buffer
= fulldecl
->Dim
.Index2D
;
629 info
->const_file_max
[buffer
] =
630 MAX2(info
->const_file_max
[buffer
], (int)reg
);
631 info
->const_buffers_declared
|= 1u << buffer
;
634 case TGSI_FILE_IMAGE
:
635 info
->images_declared
|= 1u << reg
;
636 if (fulldecl
->Image
.Resource
== TGSI_TEXTURE_BUFFER
)
637 info
->images_buffers
|= 1 << reg
;
640 case TGSI_FILE_BUFFER
:
641 info
->shader_buffers_declared
|= 1u << reg
;
644 case TGSI_FILE_INPUT
:
645 info
->input_semantic_name
[reg
] = (ubyte
) semName
;
646 info
->input_semantic_index
[reg
] = (ubyte
) semIndex
;
647 info
->input_interpolate
[reg
] = (ubyte
)fulldecl
->Interp
.Interpolate
;
648 info
->input_interpolate_loc
[reg
] = (ubyte
)fulldecl
->Interp
.Location
;
649 info
->input_cylindrical_wrap
[reg
] = (ubyte
)fulldecl
->Interp
.CylindricalWrap
;
651 /* Vertex shaders can have inputs with holes between them. */
652 info
->num_inputs
= MAX2(info
->num_inputs
, reg
+ 1);
655 case TGSI_SEMANTIC_PRIMID
:
656 info
->uses_primid
= true;
658 case TGSI_SEMANTIC_POSITION
:
659 info
->reads_position
= true;
661 case TGSI_SEMANTIC_FACE
:
662 info
->uses_frontface
= true;
667 case TGSI_FILE_SYSTEM_VALUE
:
668 index
= fulldecl
->Range
.First
;
670 info
->system_value_semantic_name
[index
] = semName
;
671 info
->num_system_values
= MAX2(info
->num_system_values
, index
+ 1);
674 case TGSI_SEMANTIC_INSTANCEID
:
675 info
->uses_instanceid
= TRUE
;
677 case TGSI_SEMANTIC_VERTEXID
:
678 info
->uses_vertexid
= TRUE
;
680 case TGSI_SEMANTIC_VERTEXID_NOBASE
:
681 info
->uses_vertexid_nobase
= TRUE
;
683 case TGSI_SEMANTIC_BASEVERTEX
:
684 info
->uses_basevertex
= TRUE
;
686 case TGSI_SEMANTIC_PRIMID
:
687 info
->uses_primid
= TRUE
;
689 case TGSI_SEMANTIC_INVOCATIONID
:
690 info
->uses_invocationid
= TRUE
;
692 case TGSI_SEMANTIC_POSITION
:
693 info
->reads_position
= TRUE
;
695 case TGSI_SEMANTIC_FACE
:
696 info
->uses_frontface
= TRUE
;
698 case TGSI_SEMANTIC_SAMPLEMASK
:
699 info
->reads_samplemask
= TRUE
;
701 case TGSI_SEMANTIC_TESSINNER
:
702 case TGSI_SEMANTIC_TESSOUTER
:
703 info
->reads_tess_factors
= true;
708 case TGSI_FILE_OUTPUT
:
709 info
->output_semantic_name
[reg
] = (ubyte
) semName
;
710 info
->output_semantic_index
[reg
] = (ubyte
) semIndex
;
711 info
->output_usagemask
[reg
] |= fulldecl
->Declaration
.UsageMask
;
712 info
->num_outputs
= MAX2(info
->num_outputs
, reg
+ 1);
714 if (fulldecl
->Declaration
.UsageMask
& TGSI_WRITEMASK_X
) {
715 info
->output_streams
[reg
] |= (ubyte
)fulldecl
->Semantic
.StreamX
;
716 info
->num_stream_output_components
[fulldecl
->Semantic
.StreamX
]++;
718 if (fulldecl
->Declaration
.UsageMask
& TGSI_WRITEMASK_Y
) {
719 info
->output_streams
[reg
] |= (ubyte
)fulldecl
->Semantic
.StreamY
<< 2;
720 info
->num_stream_output_components
[fulldecl
->Semantic
.StreamY
]++;
722 if (fulldecl
->Declaration
.UsageMask
& TGSI_WRITEMASK_Z
) {
723 info
->output_streams
[reg
] |= (ubyte
)fulldecl
->Semantic
.StreamZ
<< 4;
724 info
->num_stream_output_components
[fulldecl
->Semantic
.StreamZ
]++;
726 if (fulldecl
->Declaration
.UsageMask
& TGSI_WRITEMASK_W
) {
727 info
->output_streams
[reg
] |= (ubyte
)fulldecl
->Semantic
.StreamW
<< 6;
728 info
->num_stream_output_components
[fulldecl
->Semantic
.StreamW
]++;
732 case TGSI_SEMANTIC_PRIMID
:
733 info
->writes_primid
= true;
735 case TGSI_SEMANTIC_VIEWPORT_INDEX
:
736 info
->writes_viewport_index
= true;
738 case TGSI_SEMANTIC_LAYER
:
739 info
->writes_layer
= true;
741 case TGSI_SEMANTIC_PSIZE
:
742 info
->writes_psize
= true;
744 case TGSI_SEMANTIC_CLIPVERTEX
:
745 info
->writes_clipvertex
= true;
747 case TGSI_SEMANTIC_COLOR
:
748 info
->colors_written
|= 1 << semIndex
;
750 case TGSI_SEMANTIC_STENCIL
:
751 info
->writes_stencil
= true;
753 case TGSI_SEMANTIC_SAMPLEMASK
:
754 info
->writes_samplemask
= true;
756 case TGSI_SEMANTIC_EDGEFLAG
:
757 info
->writes_edgeflag
= true;
759 case TGSI_SEMANTIC_POSITION
:
760 if (procType
== PIPE_SHADER_FRAGMENT
)
761 info
->writes_z
= true;
763 info
->writes_position
= true;
768 case TGSI_FILE_SAMPLER
:
769 STATIC_ASSERT(sizeof(info
->samplers_declared
) * 8 >= PIPE_MAX_SAMPLERS
);
770 info
->samplers_declared
|= 1u << reg
;
773 case TGSI_FILE_SAMPLER_VIEW
:
774 target
= fulldecl
->SamplerView
.Resource
;
775 type
= fulldecl
->SamplerView
.ReturnTypeX
;
777 assert(target
< TGSI_TEXTURE_UNKNOWN
);
778 if (info
->sampler_targets
[reg
] == TGSI_TEXTURE_UNKNOWN
) {
779 /* Save sampler target for this sampler index */
780 info
->sampler_targets
[reg
] = target
;
781 info
->sampler_type
[reg
] = type
;
783 /* if previously declared, make sure targets agree */
784 assert(info
->sampler_targets
[reg
] == target
);
785 assert(info
->sampler_type
[reg
] == type
);
794 scan_immediate(struct tgsi_shader_info
*info
)
796 uint reg
= info
->immediate_count
++;
797 uint file
= TGSI_FILE_IMMEDIATE
;
799 info
->file_mask
[file
] |= (1 << reg
);
800 info
->file_count
[file
]++;
801 info
->file_max
[file
] = MAX2(info
->file_max
[file
], (int)reg
);
806 scan_property(struct tgsi_shader_info
*info
,
807 const struct tgsi_full_property
*fullprop
)
809 unsigned name
= fullprop
->Property
.PropertyName
;
810 unsigned value
= fullprop
->u
[0].Data
;
812 assert(name
< ARRAY_SIZE(info
->properties
));
813 info
->properties
[name
] = value
;
816 case TGSI_PROPERTY_NUM_CLIPDIST_ENABLED
:
817 info
->num_written_clipdistance
= value
;
818 info
->clipdist_writemask
|= (1 << value
) - 1;
820 case TGSI_PROPERTY_NUM_CULLDIST_ENABLED
:
821 info
->num_written_culldistance
= value
;
822 info
->culldist_writemask
|= (1 << value
) - 1;
829 * Scan the given TGSI shader to collect information such as number of
830 * registers used, special instructions used, etc.
831 * \return info the result of the scan
834 tgsi_scan_shader(const struct tgsi_token
*tokens
,
835 struct tgsi_shader_info
*info
)
838 struct tgsi_parse_context parse
;
839 unsigned current_depth
= 0;
841 memset(info
, 0, sizeof(*info
));
842 for (i
= 0; i
< TGSI_FILE_COUNT
; i
++)
843 info
->file_max
[i
] = -1;
844 for (i
= 0; i
< ARRAY_SIZE(info
->const_file_max
); i
++)
845 info
->const_file_max
[i
] = -1;
846 info
->properties
[TGSI_PROPERTY_GS_INVOCATIONS
] = 1;
847 for (i
= 0; i
< ARRAY_SIZE(info
->sampler_targets
); i
++)
848 info
->sampler_targets
[i
] = TGSI_TEXTURE_UNKNOWN
;
851 ** Setup to begin parsing input shader
853 if (tgsi_parse_init( &parse
, tokens
) != TGSI_PARSE_OK
) {
854 debug_printf("tgsi_parse_init() failed in tgsi_scan_shader()!\n");
857 procType
= parse
.FullHeader
.Processor
.Processor
;
858 assert(procType
== PIPE_SHADER_FRAGMENT
||
859 procType
== PIPE_SHADER_VERTEX
||
860 procType
== PIPE_SHADER_GEOMETRY
||
861 procType
== PIPE_SHADER_TESS_CTRL
||
862 procType
== PIPE_SHADER_TESS_EVAL
||
863 procType
== PIPE_SHADER_COMPUTE
);
864 info
->processor
= procType
;
865 info
->num_tokens
= tgsi_num_tokens(parse
.Tokens
);
868 ** Loop over incoming program tokens/instructions
870 while (!tgsi_parse_end_of_tokens(&parse
)) {
871 tgsi_parse_token( &parse
);
873 switch( parse
.FullToken
.Token
.Type
) {
874 case TGSI_TOKEN_TYPE_INSTRUCTION
:
875 scan_instruction(info
, &parse
.FullToken
.FullInstruction
,
878 case TGSI_TOKEN_TYPE_DECLARATION
:
879 scan_declaration(info
, &parse
.FullToken
.FullDeclaration
);
881 case TGSI_TOKEN_TYPE_IMMEDIATE
:
882 scan_immediate(info
);
884 case TGSI_TOKEN_TYPE_PROPERTY
:
885 scan_property(info
, &parse
.FullToken
.FullProperty
);
888 assert(!"Unexpected TGSI token type");
892 info
->uses_kill
= (info
->opcode_count
[TGSI_OPCODE_KILL_IF
] ||
893 info
->opcode_count
[TGSI_OPCODE_KILL
]);
895 /* The dimensions of the IN decleration in geometry shader have
896 * to be deduced from the type of the input primitive.
898 if (procType
== PIPE_SHADER_GEOMETRY
) {
899 unsigned input_primitive
=
900 info
->properties
[TGSI_PROPERTY_GS_INPUT_PRIM
];
901 int num_verts
= u_vertices_per_prim(input_primitive
);
903 info
->file_count
[TGSI_FILE_INPUT
] = num_verts
;
904 info
->file_max
[TGSI_FILE_INPUT
] =
905 MAX2(info
->file_max
[TGSI_FILE_INPUT
], num_verts
- 1);
906 for (j
= 0; j
< num_verts
; ++j
) {
907 info
->file_mask
[TGSI_FILE_INPUT
] |= (1 << j
);
911 tgsi_parse_free(&parse
);
915 * Collect information about the arrays of a given register file.
917 * @param tokens TGSI shader
918 * @param file the register file to scan through
919 * @param max_array_id number of entries in @p arrays; should be equal to the
920 * highest array id, i.e. tgsi_shader_info::array_max[file].
921 * @param arrays info for array of each ID will be written to arrays[ID - 1].
924 tgsi_scan_arrays(const struct tgsi_token
*tokens
,
926 unsigned max_array_id
,
927 struct tgsi_array_info
*arrays
)
929 struct tgsi_parse_context parse
;
931 if (tgsi_parse_init(&parse
, tokens
) != TGSI_PARSE_OK
) {
932 debug_printf("tgsi_parse_init() failed in tgsi_scan_arrays()!\n");
936 memset(arrays
, 0, sizeof(arrays
[0]) * max_array_id
);
938 while (!tgsi_parse_end_of_tokens(&parse
)) {
939 struct tgsi_full_instruction
*inst
;
941 tgsi_parse_token(&parse
);
943 if (parse
.FullToken
.Token
.Type
== TGSI_TOKEN_TYPE_DECLARATION
) {
944 struct tgsi_full_declaration
*decl
= &parse
.FullToken
.FullDeclaration
;
946 if (decl
->Declaration
.Array
&& decl
->Declaration
.File
== file
&&
947 decl
->Array
.ArrayID
> 0 && decl
->Array
.ArrayID
<= max_array_id
) {
948 struct tgsi_array_info
*array
= &arrays
[decl
->Array
.ArrayID
- 1];
949 assert(!array
->declared
);
950 array
->declared
= true;
951 array
->range
= decl
->Range
;
955 if (parse
.FullToken
.Token
.Type
!= TGSI_TOKEN_TYPE_INSTRUCTION
)
958 inst
= &parse
.FullToken
.FullInstruction
;
959 for (unsigned i
= 0; i
< inst
->Instruction
.NumDstRegs
; i
++) {
960 const struct tgsi_full_dst_register
*dst
= &inst
->Dst
[i
];
961 if (dst
->Register
.File
!= file
)
964 if (dst
->Register
.Indirect
) {
965 if (dst
->Indirect
.ArrayID
> 0 &&
966 dst
->Indirect
.ArrayID
<= max_array_id
) {
967 arrays
[dst
->Indirect
.ArrayID
- 1].writemask
|= dst
->Register
.WriteMask
;
969 /* Indirect writes without an ArrayID can write anywhere. */
970 for (unsigned j
= 0; j
< max_array_id
; ++j
)
971 arrays
[j
].writemask
|= dst
->Register
.WriteMask
;
974 /* Check whether the write falls into any of the arrays anyway. */
975 for (unsigned j
= 0; j
< max_array_id
; ++j
) {
976 struct tgsi_array_info
*array
= &arrays
[j
];
977 if (array
->declared
&&
978 dst
->Register
.Index
>= array
->range
.First
&&
979 dst
->Register
.Index
<= array
->range
.Last
)
980 array
->writemask
|= dst
->Register
.WriteMask
;
986 tgsi_parse_free(&parse
);
992 check_no_subroutines(const struct tgsi_full_instruction
*inst
)
994 switch (inst
->Instruction
.Opcode
) {
995 case TGSI_OPCODE_BGNSUB
:
996 case TGSI_OPCODE_ENDSUB
:
997 case TGSI_OPCODE_CAL
:
998 unreachable("subroutines unhandled");
1003 get_inst_tessfactor_writemask(const struct tgsi_shader_info
*info
,
1004 const struct tgsi_full_instruction
*inst
)
1006 unsigned writemask
= 0;
1008 for (unsigned i
= 0; i
< inst
->Instruction
.NumDstRegs
; i
++) {
1009 const struct tgsi_full_dst_register
*dst
= &inst
->Dst
[i
];
1011 if (dst
->Register
.File
== TGSI_FILE_OUTPUT
&&
1012 !dst
->Register
.Indirect
) {
1013 unsigned name
= info
->output_semantic_name
[dst
->Register
.Index
];
1015 if (name
== TGSI_SEMANTIC_TESSINNER
)
1016 writemask
|= dst
->Register
.WriteMask
;
1017 else if (name
== TGSI_SEMANTIC_TESSOUTER
)
1018 writemask
|= dst
->Register
.WriteMask
<< 4;
1025 get_block_tessfactor_writemask(const struct tgsi_shader_info
*info
,
1026 struct tgsi_parse_context
*parse
,
1027 unsigned end_opcode
)
1029 struct tgsi_full_instruction
*inst
;
1030 unsigned writemask
= 0;
1032 tgsi_parse_token(parse
);
1033 assert(parse
->FullToken
.Token
.Type
== TGSI_TOKEN_TYPE_INSTRUCTION
);
1034 inst
= &parse
->FullToken
.FullInstruction
;
1035 check_no_subroutines(inst
);
1037 while (inst
->Instruction
.Opcode
!= end_opcode
) {
1039 /* Recursively process nested blocks. */
1040 switch (inst
->Instruction
.Opcode
) {
1041 case TGSI_OPCODE_IF
:
1042 case TGSI_OPCODE_UIF
:
1044 get_block_tessfactor_writemask(info
, parse
, TGSI_OPCODE_ENDIF
);
1047 case TGSI_OPCODE_BGNLOOP
:
1049 get_block_tessfactor_writemask(info
, parse
, TGSI_OPCODE_ENDLOOP
);
1052 case TGSI_OPCODE_BARRIER
:
1053 unreachable("nested BARRIER is illegal");
1057 writemask
|= get_inst_tessfactor_writemask(info
, inst
);
1060 tgsi_parse_token(parse
);
1061 assert(parse
->FullToken
.Token
.Type
== TGSI_TOKEN_TYPE_INSTRUCTION
);
1062 inst
= &parse
->FullToken
.FullInstruction
;
1063 check_no_subroutines(inst
);
1070 get_if_block_tessfactor_writemask(const struct tgsi_shader_info
*info
,
1071 struct tgsi_parse_context
*parse
,
1072 unsigned *upper_block_tf_writemask
,
1073 unsigned *cond_block_tf_writemask
)
1075 struct tgsi_full_instruction
*inst
;
1076 unsigned then_tessfactor_writemask
= 0;
1077 unsigned else_tessfactor_writemask
= 0;
1079 bool is_then
= true;
1081 tgsi_parse_token(parse
);
1082 assert(parse
->FullToken
.Token
.Type
== TGSI_TOKEN_TYPE_INSTRUCTION
);
1083 inst
= &parse
->FullToken
.FullInstruction
;
1084 check_no_subroutines(inst
);
1086 while (inst
->Instruction
.Opcode
!= TGSI_OPCODE_ENDIF
) {
1088 switch (inst
->Instruction
.Opcode
) {
1089 case TGSI_OPCODE_ELSE
:
1093 /* Recursively process nested blocks. */
1094 case TGSI_OPCODE_IF
:
1095 case TGSI_OPCODE_UIF
:
1096 get_if_block_tessfactor_writemask(info
, parse
,
1097 is_then
? &then_tessfactor_writemask
:
1098 &else_tessfactor_writemask
,
1099 cond_block_tf_writemask
);
1102 case TGSI_OPCODE_BGNLOOP
:
1103 *cond_block_tf_writemask
|=
1104 get_block_tessfactor_writemask(info
, parse
, TGSI_OPCODE_ENDLOOP
);
1107 case TGSI_OPCODE_BARRIER
:
1108 unreachable("nested BARRIER is illegal");
1111 /* Process an instruction in the current block. */
1112 writemask
= get_inst_tessfactor_writemask(info
, inst
);
1116 then_tessfactor_writemask
|= writemask
;
1118 else_tessfactor_writemask
|= writemask
;
1122 tgsi_parse_token(parse
);
1123 assert(parse
->FullToken
.Token
.Type
== TGSI_TOKEN_TYPE_INSTRUCTION
);
1124 inst
= &parse
->FullToken
.FullInstruction
;
1125 check_no_subroutines(inst
);
1128 if (then_tessfactor_writemask
|| else_tessfactor_writemask
) {
1129 /* If both statements write the same tess factor channels,
1130 * we can say that the upper block writes them too. */
1131 *upper_block_tf_writemask
|= then_tessfactor_writemask
&
1132 else_tessfactor_writemask
;
1133 *cond_block_tf_writemask
|= then_tessfactor_writemask
|
1134 else_tessfactor_writemask
;
1139 tgsi_scan_tess_ctrl(const struct tgsi_token
*tokens
,
1140 const struct tgsi_shader_info
*info
,
1141 struct tgsi_tessctrl_info
*out
)
1143 memset(out
, 0, sizeof(*out
));
1145 if (info
->processor
!= PIPE_SHADER_TESS_CTRL
)
1148 struct tgsi_parse_context parse
;
1149 if (tgsi_parse_init(&parse
, tokens
) != TGSI_PARSE_OK
) {
1150 debug_printf("tgsi_parse_init() failed in tgsi_scan_arrays()!\n");
1154 /* The pass works as follows:
1155 * If all codepaths write tess factors, we can say that all invocations
1156 * define tess factors.
1158 * Each tess factor channel is tracked separately.
1160 unsigned main_block_tf_writemask
= 0; /* if main block writes tess factors */
1161 unsigned cond_block_tf_writemask
= 0; /* if cond block writes tess factors */
1163 /* Initial value = true. Here the pass will accumulate results from multiple
1164 * segments surrounded by barriers. If tess factors aren't written at all,
1165 * it's a shader bug and we don't care if this will be true.
1167 out
->tessfactors_are_def_in_all_invocs
= true;
1169 while (!tgsi_parse_end_of_tokens(&parse
)) {
1170 tgsi_parse_token(&parse
);
1172 if (parse
.FullToken
.Token
.Type
!= TGSI_TOKEN_TYPE_INSTRUCTION
)
1175 struct tgsi_full_instruction
*inst
= &parse
.FullToken
.FullInstruction
;
1176 check_no_subroutines(inst
);
1178 /* Process nested blocks. */
1179 switch (inst
->Instruction
.Opcode
) {
1180 case TGSI_OPCODE_IF
:
1181 case TGSI_OPCODE_UIF
:
1182 get_if_block_tessfactor_writemask(info
, &parse
,
1183 &main_block_tf_writemask
,
1184 &cond_block_tf_writemask
);
1187 case TGSI_OPCODE_BGNLOOP
:
1188 cond_block_tf_writemask
|=
1189 get_block_tessfactor_writemask(info
, &parse
, TGSI_OPCODE_ENDLOOP
);
1192 case TGSI_OPCODE_BARRIER
:
1193 /* The following case must be prevented:
1194 * gl_TessLevelInner = ...;
1196 * if (gl_InvocationID == 1)
1197 * gl_TessLevelInner = ...;
1199 * If you consider disjoint code segments separated by barriers, each
1200 * such segment that writes tess factor channels should write the same
1201 * channels in all codepaths within that segment.
1203 if (main_block_tf_writemask
|| cond_block_tf_writemask
) {
1204 /* Accumulate the result: */
1205 out
->tessfactors_are_def_in_all_invocs
&=
1206 !(cond_block_tf_writemask
& ~main_block_tf_writemask
);
1208 /* Analyze the next code segment from scratch. */
1209 main_block_tf_writemask
= 0;
1210 cond_block_tf_writemask
= 0;
1215 main_block_tf_writemask
|= get_inst_tessfactor_writemask(info
, inst
);
1218 /* Accumulate the result for the last code segment separated by a barrier. */
1219 if (main_block_tf_writemask
|| cond_block_tf_writemask
) {
1220 out
->tessfactors_are_def_in_all_invocs
&=
1221 !(cond_block_tf_writemask
& ~main_block_tf_writemask
);
1224 tgsi_parse_free(&parse
);