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_parse.h"
42 #include "tgsi/tgsi_util.h"
43 #include "tgsi/tgsi_scan.h"
49 * Scan the given TGSI shader to collect information such as number of
50 * registers used, special instructions used, etc.
51 * \return info the result of the scan
54 tgsi_scan_shader(const struct tgsi_token
*tokens
,
55 struct tgsi_shader_info
*info
)
58 struct tgsi_parse_context parse
;
59 unsigned current_depth
= 0;
61 memset(info
, 0, sizeof(*info
));
62 for (i
= 0; i
< TGSI_FILE_COUNT
; i
++)
63 info
->file_max
[i
] = -1;
64 for (i
= 0; i
< Elements(info
->const_file_max
); i
++)
65 info
->const_file_max
[i
] = -1;
66 info
->properties
[TGSI_PROPERTY_GS_INVOCATIONS
] = 1;
69 ** Setup to begin parsing input shader
71 if (tgsi_parse_init( &parse
, tokens
) != TGSI_PARSE_OK
) {
72 debug_printf("tgsi_parse_init() failed in tgsi_scan_shader()!\n");
75 procType
= parse
.FullHeader
.Processor
.Processor
;
76 assert(procType
== TGSI_PROCESSOR_FRAGMENT
||
77 procType
== TGSI_PROCESSOR_VERTEX
||
78 procType
== TGSI_PROCESSOR_GEOMETRY
||
79 procType
== TGSI_PROCESSOR_TESS_CTRL
||
80 procType
== TGSI_PROCESSOR_TESS_EVAL
||
81 procType
== TGSI_PROCESSOR_COMPUTE
);
82 info
->processor
= procType
;
86 ** Loop over incoming program tokens/instructions
88 while( !tgsi_parse_end_of_tokens( &parse
) ) {
92 tgsi_parse_token( &parse
);
94 switch( parse
.FullToken
.Token
.Type
) {
95 case TGSI_TOKEN_TYPE_INSTRUCTION
:
97 const struct tgsi_full_instruction
*fullinst
98 = &parse
.FullToken
.FullInstruction
;
101 assert(fullinst
->Instruction
.Opcode
< TGSI_OPCODE_LAST
);
102 info
->opcode_count
[fullinst
->Instruction
.Opcode
]++;
104 switch (fullinst
->Instruction
.Opcode
) {
106 case TGSI_OPCODE_UIF
:
107 case TGSI_OPCODE_BGNLOOP
:
109 info
->max_depth
= MAX2(info
->max_depth
, current_depth
);
111 case TGSI_OPCODE_ENDIF
:
112 case TGSI_OPCODE_ENDLOOP
:
119 if (fullinst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_CENTROID
||
120 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_OFFSET
||
121 fullinst
->Instruction
.Opcode
== TGSI_OPCODE_INTERP_SAMPLE
) {
122 const struct tgsi_full_src_register
*src0
= &fullinst
->Src
[0];
125 if (src0
->Register
.Indirect
&& src0
->Indirect
.ArrayID
)
126 input
= info
->input_array_first
[src0
->Indirect
.ArrayID
];
128 input
= src0
->Register
.Index
;
130 /* For the INTERP opcodes, the interpolation is always
131 * PERSPECTIVE unless LINEAR is specified.
133 switch (info
->input_interpolate
[input
]) {
134 case TGSI_INTERPOLATE_COLOR
:
135 case TGSI_INTERPOLATE_CONSTANT
:
136 case TGSI_INTERPOLATE_PERSPECTIVE
:
137 switch (fullinst
->Instruction
.Opcode
) {
138 case TGSI_OPCODE_INTERP_CENTROID
:
139 info
->uses_persp_opcode_interp_centroid
= true;
141 case TGSI_OPCODE_INTERP_OFFSET
:
142 info
->uses_persp_opcode_interp_offset
= true;
144 case TGSI_OPCODE_INTERP_SAMPLE
:
145 info
->uses_persp_opcode_interp_sample
= true;
150 case TGSI_INTERPOLATE_LINEAR
:
151 switch (fullinst
->Instruction
.Opcode
) {
152 case TGSI_OPCODE_INTERP_CENTROID
:
153 info
->uses_linear_opcode_interp_centroid
= true;
155 case TGSI_OPCODE_INTERP_OFFSET
:
156 info
->uses_linear_opcode_interp_offset
= true;
158 case TGSI_OPCODE_INTERP_SAMPLE
:
159 info
->uses_linear_opcode_interp_sample
= true;
166 if (fullinst
->Instruction
.Opcode
>= TGSI_OPCODE_F2D
&&
167 fullinst
->Instruction
.Opcode
<= TGSI_OPCODE_DSSG
)
168 info
->uses_doubles
= true;
170 for (i
= 0; i
< fullinst
->Instruction
.NumSrcRegs
; i
++) {
171 const struct tgsi_full_src_register
*src
=
173 int ind
= src
->Register
.Index
;
175 /* Mark which inputs are effectively used */
176 if (src
->Register
.File
== TGSI_FILE_INPUT
) {
178 usage_mask
= tgsi_util_get_inst_usage_mask(fullinst
, i
);
179 if (src
->Register
.Indirect
) {
180 for (ind
= 0; ind
< info
->num_inputs
; ++ind
) {
181 info
->input_usage_mask
[ind
] |= usage_mask
;
185 assert(ind
< PIPE_MAX_SHADER_INPUTS
);
186 info
->input_usage_mask
[ind
] |= usage_mask
;
189 if (procType
== TGSI_PROCESSOR_FRAGMENT
&&
190 !src
->Register
.Indirect
&&
191 info
->input_semantic_name
[src
->Register
.Index
] ==
192 TGSI_SEMANTIC_POSITION
&&
193 (src
->Register
.SwizzleX
== TGSI_SWIZZLE_Z
||
194 src
->Register
.SwizzleY
== TGSI_SWIZZLE_Z
||
195 src
->Register
.SwizzleZ
== TGSI_SWIZZLE_Z
||
196 src
->Register
.SwizzleW
== TGSI_SWIZZLE_Z
)) {
197 info
->reads_z
= TRUE
;
201 /* check for indirect register reads */
202 if (src
->Register
.Indirect
) {
203 info
->indirect_files
|= (1 << src
->Register
.File
);
204 info
->indirect_files_read
|= (1 << src
->Register
.File
);
208 if (src
->Register
.File
== TGSI_FILE_SAMPLER
) {
209 assert(fullinst
->Instruction
.Texture
);
210 assert(src
->Register
.Index
< Elements(info
->is_msaa_sampler
));
212 if (fullinst
->Instruction
.Texture
&&
213 (fullinst
->Texture
.Texture
== TGSI_TEXTURE_2D_MSAA
||
214 fullinst
->Texture
.Texture
== TGSI_TEXTURE_2D_ARRAY_MSAA
)) {
215 info
->is_msaa_sampler
[src
->Register
.Index
] = TRUE
;
220 /* check for indirect register writes */
221 for (i
= 0; i
< fullinst
->Instruction
.NumDstRegs
; i
++) {
222 const struct tgsi_full_dst_register
*dst
= &fullinst
->Dst
[i
];
223 if (dst
->Register
.Indirect
) {
224 info
->indirect_files
|= (1 << dst
->Register
.File
);
225 info
->indirect_files_written
|= (1 << dst
->Register
.File
);
229 info
->num_instructions
++;
233 case TGSI_TOKEN_TYPE_DECLARATION
:
235 const struct tgsi_full_declaration
*fulldecl
236 = &parse
.FullToken
.FullDeclaration
;
237 const uint file
= fulldecl
->Declaration
.File
;
240 if (fulldecl
->Declaration
.Array
) {
241 unsigned array_id
= fulldecl
->Array
.ArrayID
;
244 case TGSI_FILE_INPUT
:
245 assert(array_id
< ARRAY_SIZE(info
->input_array_first
));
246 info
->input_array_first
[array_id
] = fulldecl
->Range
.First
;
247 info
->input_array_last
[array_id
] = fulldecl
->Range
.Last
;
249 case TGSI_FILE_OUTPUT
:
250 assert(array_id
< ARRAY_SIZE(info
->output_array_first
));
251 info
->output_array_first
[array_id
] = fulldecl
->Range
.First
;
252 info
->output_array_last
[array_id
] = fulldecl
->Range
.Last
;
255 info
->array_max
[file
] = MAX2(info
->array_max
[file
], array_id
);
258 for (reg
= fulldecl
->Range
.First
;
259 reg
<= fulldecl
->Range
.Last
;
261 unsigned semName
= fulldecl
->Semantic
.Name
;
263 fulldecl
->Semantic
.Index
+ (reg
- fulldecl
->Range
.First
);
265 /* only first 32 regs will appear in this bitfield */
266 info
->file_mask
[file
] |= (1 << reg
);
267 info
->file_count
[file
]++;
268 info
->file_max
[file
] = MAX2(info
->file_max
[file
], (int)reg
);
270 if (file
== TGSI_FILE_CONSTANT
) {
273 if (fulldecl
->Declaration
.Dimension
)
274 buffer
= fulldecl
->Dim
.Index2D
;
276 info
->const_file_max
[buffer
] =
277 MAX2(info
->const_file_max
[buffer
], (int)reg
);
279 else if (file
== TGSI_FILE_INPUT
) {
280 info
->input_semantic_name
[reg
] = (ubyte
) semName
;
281 info
->input_semantic_index
[reg
] = (ubyte
) semIndex
;
282 info
->input_interpolate
[reg
] = (ubyte
)fulldecl
->Interp
.Interpolate
;
283 info
->input_interpolate_loc
[reg
] = (ubyte
)fulldecl
->Interp
.Location
;
284 info
->input_cylindrical_wrap
[reg
] = (ubyte
)fulldecl
->Interp
.CylindricalWrap
;
287 /* Only interpolated varyings. Don't include POSITION.
288 * Don't include integer varyings, because they are not
291 if (semName
== TGSI_SEMANTIC_GENERIC
||
292 semName
== TGSI_SEMANTIC_TEXCOORD
||
293 semName
== TGSI_SEMANTIC_COLOR
||
294 semName
== TGSI_SEMANTIC_BCOLOR
||
295 semName
== TGSI_SEMANTIC_FOG
||
296 semName
== TGSI_SEMANTIC_CLIPDIST
||
297 semName
== TGSI_SEMANTIC_CULLDIST
) {
298 switch (fulldecl
->Interp
.Interpolate
) {
299 case TGSI_INTERPOLATE_COLOR
:
300 case TGSI_INTERPOLATE_PERSPECTIVE
:
301 switch (fulldecl
->Interp
.Location
) {
302 case TGSI_INTERPOLATE_LOC_CENTER
:
303 info
->uses_persp_center
= true;
305 case TGSI_INTERPOLATE_LOC_CENTROID
:
306 info
->uses_persp_centroid
= true;
308 case TGSI_INTERPOLATE_LOC_SAMPLE
:
309 info
->uses_persp_sample
= true;
313 case TGSI_INTERPOLATE_LINEAR
:
314 switch (fulldecl
->Interp
.Location
) {
315 case TGSI_INTERPOLATE_LOC_CENTER
:
316 info
->uses_linear_center
= true;
318 case TGSI_INTERPOLATE_LOC_CENTROID
:
319 info
->uses_linear_centroid
= true;
321 case TGSI_INTERPOLATE_LOC_SAMPLE
:
322 info
->uses_linear_sample
= true;
326 /* TGSI_INTERPOLATE_CONSTANT doesn't do any interpolation. */
330 if (semName
== TGSI_SEMANTIC_PRIMID
)
331 info
->uses_primid
= TRUE
;
332 else if (procType
== TGSI_PROCESSOR_FRAGMENT
) {
333 if (semName
== TGSI_SEMANTIC_POSITION
)
334 info
->reads_position
= TRUE
;
335 else if (semName
== TGSI_SEMANTIC_FACE
)
336 info
->uses_frontface
= TRUE
;
339 else if (file
== TGSI_FILE_SYSTEM_VALUE
) {
340 unsigned index
= fulldecl
->Range
.First
;
342 info
->system_value_semantic_name
[index
] = semName
;
343 info
->num_system_values
= MAX2(info
->num_system_values
,
346 if (semName
== TGSI_SEMANTIC_INSTANCEID
) {
347 info
->uses_instanceid
= TRUE
;
349 else if (semName
== TGSI_SEMANTIC_VERTEXID
) {
350 info
->uses_vertexid
= TRUE
;
352 else if (semName
== TGSI_SEMANTIC_VERTEXID_NOBASE
) {
353 info
->uses_vertexid_nobase
= TRUE
;
355 else if (semName
== TGSI_SEMANTIC_BASEVERTEX
) {
356 info
->uses_basevertex
= TRUE
;
358 else if (semName
== TGSI_SEMANTIC_PRIMID
) {
359 info
->uses_primid
= TRUE
;
360 } else if (semName
== TGSI_SEMANTIC_INVOCATIONID
) {
361 info
->uses_invocationid
= TRUE
;
364 else if (file
== TGSI_FILE_OUTPUT
) {
365 info
->output_semantic_name
[reg
] = (ubyte
) semName
;
366 info
->output_semantic_index
[reg
] = (ubyte
) semIndex
;
369 if (semName
== TGSI_SEMANTIC_COLOR
)
370 info
->colors_written
|= 1 << semIndex
;
372 if (procType
== TGSI_PROCESSOR_VERTEX
||
373 procType
== TGSI_PROCESSOR_GEOMETRY
||
374 procType
== TGSI_PROCESSOR_TESS_CTRL
||
375 procType
== TGSI_PROCESSOR_TESS_EVAL
) {
376 if (semName
== TGSI_SEMANTIC_VIEWPORT_INDEX
) {
377 info
->writes_viewport_index
= TRUE
;
379 else if (semName
== TGSI_SEMANTIC_LAYER
) {
380 info
->writes_layer
= TRUE
;
382 else if (semName
== TGSI_SEMANTIC_PSIZE
) {
383 info
->writes_psize
= TRUE
;
385 else if (semName
== TGSI_SEMANTIC_CLIPVERTEX
) {
386 info
->writes_clipvertex
= TRUE
;
390 if (procType
== TGSI_PROCESSOR_FRAGMENT
) {
391 if (semName
== TGSI_SEMANTIC_POSITION
) {
392 info
->writes_z
= TRUE
;
394 else if (semName
== TGSI_SEMANTIC_STENCIL
) {
395 info
->writes_stencil
= TRUE
;
396 } else if (semName
== TGSI_SEMANTIC_SAMPLEMASK
) {
397 info
->writes_samplemask
= TRUE
;
401 if (procType
== TGSI_PROCESSOR_VERTEX
) {
402 if (semName
== TGSI_SEMANTIC_EDGEFLAG
) {
403 info
->writes_edgeflag
= TRUE
;
406 } else if (file
== TGSI_FILE_SAMPLER
) {
407 info
->samplers_declared
|= 1 << reg
;
413 case TGSI_TOKEN_TYPE_IMMEDIATE
:
415 uint reg
= info
->immediate_count
++;
416 uint file
= TGSI_FILE_IMMEDIATE
;
418 info
->file_mask
[file
] |= (1 << reg
);
419 info
->file_count
[file
]++;
420 info
->file_max
[file
] = MAX2(info
->file_max
[file
], (int)reg
);
424 case TGSI_TOKEN_TYPE_PROPERTY
:
426 const struct tgsi_full_property
*fullprop
427 = &parse
.FullToken
.FullProperty
;
428 unsigned name
= fullprop
->Property
.PropertyName
;
429 unsigned value
= fullprop
->u
[0].Data
;
431 assert(name
< Elements(info
->properties
));
432 info
->properties
[name
] = value
;
435 case TGSI_PROPERTY_NUM_CLIPDIST_ENABLED
:
436 info
->num_written_clipdistance
= value
;
437 info
->clipdist_writemask
|= (1 << value
) - 1;
439 case TGSI_PROPERTY_NUM_CULLDIST_ENABLED
:
440 info
->num_written_culldistance
= value
;
441 info
->culldist_writemask
|= (1 << value
) - 1;
452 info
->uses_kill
= (info
->opcode_count
[TGSI_OPCODE_KILL_IF
] ||
453 info
->opcode_count
[TGSI_OPCODE_KILL
]);
455 /* The dimensions of the IN decleration in geometry shader have
456 * to be deduced from the type of the input primitive.
458 if (procType
== TGSI_PROCESSOR_GEOMETRY
) {
459 unsigned input_primitive
=
460 info
->properties
[TGSI_PROPERTY_GS_INPUT_PRIM
];
461 int num_verts
= u_vertices_per_prim(input_primitive
);
463 info
->file_count
[TGSI_FILE_INPUT
] = num_verts
;
464 info
->file_max
[TGSI_FILE_INPUT
] =
465 MAX2(info
->file_max
[TGSI_FILE_INPUT
], num_verts
- 1);
466 for (j
= 0; j
< num_verts
; ++j
) {
467 info
->file_mask
[TGSI_FILE_INPUT
] |= (1 << j
);
471 tgsi_parse_free (&parse
);
477 * Check if the given shader is a "passthrough" shader consisting of only
478 * MOV instructions of the form: MOV OUT[n], IN[n]
482 tgsi_is_passthrough_shader(const struct tgsi_token
*tokens
)
484 struct tgsi_parse_context parse
;
487 ** Setup to begin parsing input shader
489 if (tgsi_parse_init(&parse
, tokens
) != TGSI_PARSE_OK
) {
490 debug_printf("tgsi_parse_init() failed in tgsi_is_passthrough_shader()!\n");
495 ** Loop over incoming program tokens/instructions
497 while (!tgsi_parse_end_of_tokens(&parse
)) {
499 tgsi_parse_token(&parse
);
501 switch (parse
.FullToken
.Token
.Type
) {
502 case TGSI_TOKEN_TYPE_INSTRUCTION
:
504 struct tgsi_full_instruction
*fullinst
=
505 &parse
.FullToken
.FullInstruction
;
506 const struct tgsi_full_src_register
*src
=
508 const struct tgsi_full_dst_register
*dst
=
511 /* Do a whole bunch of checks for a simple move */
512 if (fullinst
->Instruction
.Opcode
!= TGSI_OPCODE_MOV
||
513 (src
->Register
.File
!= TGSI_FILE_INPUT
&&
514 src
->Register
.File
!= TGSI_FILE_SYSTEM_VALUE
) ||
515 dst
->Register
.File
!= TGSI_FILE_OUTPUT
||
516 src
->Register
.Index
!= dst
->Register
.Index
||
518 src
->Register
.Negate
||
519 src
->Register
.Absolute
||
521 src
->Register
.SwizzleX
!= TGSI_SWIZZLE_X
||
522 src
->Register
.SwizzleY
!= TGSI_SWIZZLE_Y
||
523 src
->Register
.SwizzleZ
!= TGSI_SWIZZLE_Z
||
524 src
->Register
.SwizzleW
!= TGSI_SWIZZLE_W
||
526 dst
->Register
.WriteMask
!= TGSI_WRITEMASK_XYZW
)
528 tgsi_parse_free(&parse
);
534 case TGSI_TOKEN_TYPE_DECLARATION
:
536 case TGSI_TOKEN_TYPE_IMMEDIATE
:
538 case TGSI_TOKEN_TYPE_PROPERTY
:
545 tgsi_parse_free(&parse
);
547 /* if we get here, it's a pass-through shader */