2 # Copyright (c) 2018 Linaro Limited
4 # This library is free software; you can redistribute it and/or
5 # modify it under the terms of the GNU Lesser General Public
6 # License as published by the Free Software Foundation; either
7 # version 2 of the License, or (at your option) any later version.
9 # This library is distributed in the hope that it will be useful,
10 # but WITHOUT ANY WARRANTY; without even the implied warranty of
11 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 # Lesser General Public License for more details.
14 # You should have received a copy of the GNU Lesser General Public
15 # License along with this library; if not, see <http://www.gnu.org/licenses/>.
19 # Generate a decoding tree from a specification file.
20 # See the syntax and semantics in docs/devel/decodetree.rst.
35 translate_prefix
= 'trans'
36 translate_scope
= 'static '
41 decode_function
= 'decode'
43 re_ident
= '[a-zA-Z][a-zA-Z0-9_]*'
46 def error_with_file(file, lineno
, *args
):
47 """Print an error message from file:line and args and exit."""
52 r
= '{0}:{1}: error:'.format(file, lineno
)
54 r
= '{0}: error:'.format(file)
61 if output_file
and output_fd
:
63 os
.remove(output_file
)
66 def error(lineno
, *args
):
67 error_with_file(input_file
, lineno
, args
)
75 if sys
.version_info
>= (3, 4):
76 re_fullmatch
= re
.fullmatch
78 def re_fullmatch(pat
, str):
79 return re
.match('^' + pat
+ '$', str)
83 output('/* This file is autogenerated by scripts/decodetree.py. */\n\n')
87 """Return a string with C spaces"""
91 def str_fields(fields
):
92 """Return a string uniquely identifing FIELDS"""
94 for n
in sorted(fields
.keys()):
99 def str_match_bits(bits
, mask
):
100 """Return a string pretty-printing BITS/MASK"""
103 i
= 1 << (insnwidth
- 1)
121 """Return true iff X is equal to a power of 2."""
122 return (x
& (x
- 1)) == 0
126 """Return the number of times 2 factors into X."""
128 while ((x
>> r
) & 1) == 0:
133 def is_contiguous(bits
):
135 if is_pow2((bits
>> shift
) + 1):
141 def eq_fields_for_args(flds_a
, flds_b
):
142 if len(flds_a
) != len(flds_b
):
144 for k
, a
in flds_a
.items():
150 def eq_fields_for_fmts(flds_a
, flds_b
):
151 if len(flds_a
) != len(flds_b
):
153 for k
, a
in flds_a
.items():
157 if a
.__class
__ != b
.__class
__ or a
!= b
:
163 """Class representing a simple instruction field"""
164 def __init__(self
, sign
, pos
, len):
168 self
.mask
= ((1 << len) - 1) << pos
175 return str(self
.pos
) + ':' + s
+ str(self
.len)
177 def str_extract(self
):
182 return '{0}(insn, {1}, {2})'.format(extr
, self
.pos
, self
.len)
184 def __eq__(self
, other
):
185 return self
.sign
== other
.sign
and self
.sign
== other
.sign
187 def __ne__(self
, other
):
188 return not self
.__eq
__(other
)
193 """Class representing a compound instruction field"""
194 def __init__(self
, subs
, mask
):
196 self
.sign
= subs
[0].sign
200 return str(self
.subs
)
202 def str_extract(self
):
205 for f
in reversed(self
.subs
):
207 ret
= f
.str_extract()
209 ret
= 'deposit32({0}, {1}, {2}, {3})' \
210 .format(ret
, pos
, 32 - pos
, f
.str_extract())
214 def __ne__(self
, other
):
215 if len(self
.subs
) != len(other
.subs
):
217 for a
, b
in zip(self
.subs
, other
.subs
):
218 if a
.__class
__ != b
.__class
__ or a
!= b
:
222 def __eq__(self
, other
):
223 return not self
.__ne
__(other
)
228 """Class representing an argument field with constant value"""
229 def __init__(self
, value
):
232 self
.sign
= value
< 0
235 return str(self
.value
)
237 def str_extract(self
):
238 return str(self
.value
)
240 def __cmp__(self
, other
):
241 return self
.value
- other
.value
246 """Class representing a field passed through an expander"""
247 def __init__(self
, func
, base
):
248 self
.mask
= base
.mask
249 self
.sign
= base
.sign
254 return self
.func
+ '(' + str(self
.base
) + ')'
256 def str_extract(self
):
257 return self
.func
+ '(' + self
.base
.str_extract() + ')'
259 def __eq__(self
, other
):
260 return self
.func
== other
.func
and self
.base
== other
.base
262 def __ne__(self
, other
):
263 return not self
.__eq
__(other
)
268 """Class representing the extracted fields of a format"""
269 def __init__(self
, nm
, flds
, extern
):
272 self
.fields
= sorted(flds
)
275 return self
.name
+ ' ' + str(self
.fields
)
277 def struct_name(self
):
278 return 'arg_' + self
.name
280 def output_def(self
):
282 output('typedef struct {\n')
283 for n
in self
.fields
:
284 output(' int ', n
, ';\n')
285 output('} ', self
.struct_name(), ';\n\n')
290 """Common code between instruction formats and instruction patterns"""
291 def __init__(self
, name
, lineno
, base
, fixb
, fixm
, udfm
, fldm
, flds
):
293 self
.file = input_file
296 self
.fixedbits
= fixb
297 self
.fixedmask
= fixm
298 self
.undefmask
= udfm
299 self
.fieldmask
= fldm
305 r
= r
+ ' ' + self
.base
.name
307 r
= r
+ ' ' + str(self
.fields
)
308 r
= r
+ ' ' + str_match_bits(self
.fixedbits
, self
.fixedmask
)
312 return str_indent(i
) + self
.__str
__()
316 class Format(General
):
317 """Class representing an instruction format"""
319 def extract_name(self
):
320 return 'extract_' + self
.name
322 def output_extract(self
):
323 output('static void ', self
.extract_name(), '(',
324 self
.base
.struct_name(), ' *a, ', insntype
, ' insn)\n{\n')
325 for n
, f
in self
.fields
.items():
326 output(' a->', n
, ' = ', f
.str_extract(), ';\n')
331 class Pattern(General
):
332 """Class representing an instruction pattern"""
334 def output_decl(self
):
335 global translate_scope
336 global translate_prefix
337 output('typedef ', self
.base
.base
.struct_name(),
338 ' arg_', self
.name
, ';\n')
339 output(translate_scope
, 'bool ', translate_prefix
, '_', self
.name
,
340 '(DisasContext *ctx, arg_', self
.name
, ' *a);\n')
342 def output_code(self
, i
, extracted
, outerbits
, outermask
):
343 global translate_prefix
345 arg
= self
.base
.base
.name
346 output(ind
, '/* ', self
.file, ':', str(self
.lineno
), ' */\n')
348 output(ind
, self
.base
.extract_name(), '(&u.f_', arg
, ', insn);\n')
349 for n
, f
in self
.fields
.items():
350 output(ind
, 'u.f_', arg
, '.', n
, ' = ', f
.str_extract(), ';\n')
351 output(ind
, 'return ', translate_prefix
, '_', self
.name
,
352 '(ctx, &u.f_', arg
, ');\n')
356 def parse_field(lineno
, name
, toks
):
357 """Parse one instruction field from TOKS at LINENO"""
362 # A "simple" field will have only one entry;
363 # a "multifield" will have several.
368 if re_fullmatch('!function=' + re_ident
, t
):
370 error(lineno
, 'duplicate function')
375 if re_fullmatch('[0-9]+:s[0-9]+', t
):
376 # Signed field extract
377 subtoks
= t
.split(':s')
379 elif re_fullmatch('[0-9]+:[0-9]+', t
):
380 # Unsigned field extract
381 subtoks
= t
.split(':')
384 error(lineno
, 'invalid field token "{0}"'.format(t
))
387 if po
+ le
> insnwidth
:
388 error(lineno
, 'field {0} too large'.format(t
))
389 f
= Field(sign
, po
, le
)
393 if width
> insnwidth
:
394 error(lineno
, 'field too large')
401 error(lineno
, 'field components overlap')
403 f
= MultiField(subs
, mask
)
405 f
= FunctionField(func
, f
)
408 error(lineno
, 'duplicate field', name
)
413 def parse_arguments(lineno
, name
, toks
):
414 """Parse one argument set from TOKS at LINENO"""
421 if re_fullmatch('!extern', t
):
424 if not re_fullmatch(re_ident
, t
):
425 error(lineno
, 'invalid argument set token "{0}"'.format(t
))
427 error(lineno
, 'duplicate argument "{0}"'.format(t
))
430 if name
in arguments
:
431 error(lineno
, 'duplicate argument set', name
)
432 arguments
[name
] = Arguments(name
, flds
, extern
)
433 # end parse_arguments
436 def lookup_field(lineno
, name
):
440 error(lineno
, 'undefined field', name
)
443 def add_field(lineno
, flds
, new_name
, f
):
445 error(lineno
, 'duplicate field', new_name
)
450 def add_field_byname(lineno
, flds
, new_name
, old_name
):
451 return add_field(lineno
, flds
, new_name
, lookup_field(lineno
, old_name
))
454 def infer_argument_set(flds
):
456 global decode_function
458 for arg
in arguments
.values():
459 if eq_fields_for_args(flds
, arg
.fields
):
462 name
= decode_function
+ str(len(arguments
))
463 arg
= Arguments(name
, flds
.keys(), False)
464 arguments
[name
] = arg
468 def infer_format(arg
, fieldmask
, flds
):
471 global decode_function
475 for n
, c
in flds
.items():
481 # Look for an existing format with the same argument set and fields
482 for fmt
in formats
.values():
483 if arg
and fmt
.base
!= arg
:
485 if fieldmask
!= fmt
.fieldmask
:
487 if not eq_fields_for_fmts(flds
, fmt
.fields
):
489 return (fmt
, const_flds
)
491 name
= decode_function
+ '_Fmt_' + str(len(formats
))
493 arg
= infer_argument_set(flds
)
495 fmt
= Format(name
, 0, arg
, 0, 0, 0, fieldmask
, var_flds
)
498 return (fmt
, const_flds
)
502 def parse_generic(lineno
, is_format
, name
, toks
):
503 """Parse one instruction format from TOKS at LINENO"""
520 # '&Foo' gives a format an explcit argument set.
524 error(lineno
, 'multiple argument sets')
528 error(lineno
, 'undefined argument set', t
)
531 # '@Foo' gives a pattern an explicit format.
535 error(lineno
, 'multiple formats')
539 error(lineno
, 'undefined format', t
)
542 # '%Foo' imports a field.
545 flds
= add_field_byname(lineno
, flds
, tt
, tt
)
548 # 'Foo=%Bar' imports a field with a different name.
549 if re_fullmatch(re_ident
+ '=%' + re_ident
, t
):
550 (fname
, iname
) = t
.split('=%')
551 flds
= add_field_byname(lineno
, flds
, fname
, iname
)
554 # 'Foo=number' sets an argument field to a constant value
555 if re_fullmatch(re_ident
+ '=[0-9]+', t
):
556 (fname
, value
) = t
.split('=')
558 flds
= add_field(lineno
, flds
, fname
, ConstField(value
))
561 # Pattern of 0s, 1s, dots and dashes indicate required zeros,
562 # required ones, or dont-cares.
563 if re_fullmatch('[01.-]+', t
):
565 fms
= t
.replace('0', '1')
566 fms
= fms
.replace('.', '0')
567 fms
= fms
.replace('-', '0')
568 fbs
= t
.replace('.', '0')
569 fbs
= fbs
.replace('-', '0')
570 ubm
= t
.replace('1', '0')
571 ubm
= ubm
.replace('.', '0')
572 ubm
= ubm
.replace('-', '1')
576 fixedbits
= (fixedbits
<< shift
) | fbs
577 fixedmask
= (fixedmask
<< shift
) | fms
578 undefmask
= (undefmask
<< shift
) | ubm
579 # Otherwise, fieldname:fieldwidth
580 elif re_fullmatch(re_ident
+ ':s?[0-9]+', t
):
581 (fname
, flen
) = t
.split(':')
586 shift
= int(flen
, 10)
587 f
= Field(sign
, insnwidth
- width
- shift
, shift
)
588 flds
= add_field(lineno
, flds
, fname
, f
)
593 error(lineno
, 'invalid token "{0}"'.format(t
))
596 # We should have filled in all of the bits of the instruction.
597 if not (is_format
and width
== 0) and width
!= insnwidth
:
598 error(lineno
, 'definition has {0} bits'.format(width
))
600 # Do not check for fields overlaping fields; one valid usage
601 # is to be able to duplicate fields via import.
603 for f
in flds
.values():
606 # Fix up what we've parsed to match either a format or a pattern.
608 # Formats cannot reference formats.
610 error(lineno
, 'format referencing format')
611 # If an argument set is given, then there should be no fields
612 # without a place to store it.
614 for f
in flds
.keys():
615 if f
not in arg
.fields
:
616 error(lineno
, 'field {0} not in argument set {1}'
617 .format(f
, arg
.name
))
619 arg
= infer_argument_set(flds
)
621 error(lineno
, 'duplicate format name', name
)
622 fmt
= Format(name
, lineno
, arg
, fixedbits
, fixedmask
,
623 undefmask
, fieldmask
, flds
)
626 # Patterns can reference a format ...
628 # ... but not an argument simultaneously
630 error(lineno
, 'pattern specifies both format and argument set')
631 if fixedmask
& fmt
.fixedmask
:
632 error(lineno
, 'pattern fixed bits overlap format fixed bits')
633 fieldmask |
= fmt
.fieldmask
634 fixedbits |
= fmt
.fixedbits
635 fixedmask |
= fmt
.fixedmask
636 undefmask |
= fmt
.undefmask
638 (fmt
, flds
) = infer_format(arg
, fieldmask
, flds
)
640 for f
in flds
.keys():
641 if f
not in arg
.fields
:
642 error(lineno
, 'field {0} not in argument set {1}'
643 .format(f
, arg
.name
))
644 if f
in fmt
.fields
.keys():
645 error(lineno
, 'field {0} set by format and pattern'.format(f
))
647 if f
not in flds
.keys() and f
not in fmt
.fields
.keys():
648 error(lineno
, 'field {0} not initialized'.format(f
))
649 pat
= Pattern(name
, lineno
, fmt
, fixedbits
, fixedmask
,
650 undefmask
, fieldmask
, flds
)
653 # Validate the masks that we have assembled.
654 if fieldmask
& fixedmask
:
655 error(lineno
, 'fieldmask overlaps fixedmask (0x{0:08x} & 0x{1:08x})'
656 .format(fieldmask
, fixedmask
))
657 if fieldmask
& undefmask
:
658 error(lineno
, 'fieldmask overlaps undefmask (0x{0:08x} & 0x{1:08x})'
659 .format(fieldmask
, undefmask
))
660 if fixedmask
& undefmask
:
661 error(lineno
, 'fixedmask overlaps undefmask (0x{0:08x} & 0x{1:08x})'
662 .format(fixedmask
, undefmask
))
664 allbits
= fieldmask | fixedmask | undefmask
665 if allbits
!= insnmask
:
666 error(lineno
, 'bits left unspecified (0x{0:08x})'
667 .format(allbits ^ insnmask
))
672 """Parse all of the patterns within a file"""
674 # Read all of the lines of the file. Concatenate lines
675 # ending in backslash; discard empty lines and comments.
688 # Next line after continuation
702 error(lineno
, 'short line')
707 # Determine the type of object needing to be parsed.
709 parse_field(lineno
, name
[1:], toks
)
711 parse_arguments(lineno
, name
[1:], toks
)
713 parse_generic(lineno
, True, name
[1:], toks
)
715 parse_generic(lineno
, False, name
, toks
)
721 """Class representing a node in a decode tree"""
723 def __init__(self
, fm
, tm
):
731 r
= '{0}{1:08x}'.format(ind
, self
.fixedmask
)
733 r
+= ' ' + self
.format
.name
735 for (b
, s
) in self
.subs
:
736 r
+= '{0} {1:08x}:\n'.format(ind
, b
)
737 r
+= s
.str1(i
+ 4) + '\n'
744 def output_code(self
, i
, extracted
, outerbits
, outermask
):
747 # If we identified all nodes below have the same format,
748 # extract the fields now.
749 if not extracted
and self
.base
:
750 output(ind
, self
.base
.extract_name(),
751 '(&u.f_', self
.base
.base
.name
, ', insn);\n')
754 # Attempt to aid the compiler in producing compact switch statements.
755 # If the bits in the mask are contiguous, extract them.
756 sh
= is_contiguous(self
.thismask
)
758 # Propagate SH down into the local functions.
759 def str_switch(b
, sh
=sh
):
760 return '(insn >> {0}) & 0x{1:x}'.format(sh
, b
>> sh
)
762 def str_case(b
, sh
=sh
):
763 return '0x{0:x}'.format(b
>> sh
)
766 return 'insn & 0x{0:08x}'.format(b
)
769 return '0x{0:08x}'.format(b
)
771 output(ind
, 'switch (', str_switch(self
.thismask
), ') {\n')
772 for b
, s
in sorted(self
.subs
):
773 assert (self
.thismask
& ~s
.fixedmask
) == 0
774 innermask
= outermask | self
.thismask
775 innerbits
= outerbits | b
776 output(ind
, 'case ', str_case(b
), ':\n')
778 str_match_bits(innerbits
, innermask
), ' */\n')
779 s
.output_code(i
+ 4, extracted
, innerbits
, innermask
)
781 output(ind
, 'return false;\n')
785 def build_tree(pats
, outerbits
, outermask
):
786 # Find the intersection of all remaining fixedmask.
787 innermask
= ~outermask
& insnmask
789 innermask
&= i
.fixedmask
794 pnames
.append(p
.name
+ ':' + p
.file + ':' + str(p
.lineno
))
795 error_with_file(pats
[0].file, pats
[0].lineno
,
796 'overlapping patterns:', pnames
)
798 fullmask
= outermask | innermask
800 # Sort each element of pats into the bin selected by the mask.
803 fb
= i
.fixedbits
& innermask
809 # We must recurse if any bin has more than one element or if
810 # the single element in the bin has not been fully matched.
811 t
= Tree(fullmask
, innermask
)
813 for b
, l
in bins
.items():
815 if len(l
) > 1 or s
.fixedmask
& ~fullmask
!= 0:
816 s
= build_tree(l
, b | outerbits
, fullmask
)
817 t
.subs
.append((b
, s
))
823 def prop_format(tree
):
824 """Propagate Format objects into the decode tree"""
826 # Depth first search.
827 for (b
, s
) in tree
.subs
:
828 if isinstance(s
, Tree
):
831 # If all entries in SUBS have the same format, then
832 # propagate that into the tree.
834 for (b
, s
) in tree
.subs
:
849 global translate_scope
850 global translate_prefix
857 global decode_function
859 decode_scope
= 'static '
861 long_opts
= ['decode=', 'translate=', 'output=', 'insnwidth=']
863 (opts
, args
) = getopt
.getopt(sys
.argv
[1:], 'o:w:', long_opts
)
864 except getopt
.GetoptError
as err
:
867 if o
in ('-o', '--output'):
869 elif o
== '--decode':
872 elif o
== '--translate':
875 elif o
in ('-w', '--insnwidth'):
878 insntype
= 'uint16_t'
880 elif insnwidth
!= 32:
881 error(0, 'cannot handle insns of width', insnwidth
)
883 assert False, 'unhandled option'
886 error(0, 'missing input file')
887 for filename
in args
:
888 input_file
= filename
889 f
= open(filename
, 'r')
893 t
= build_tree(patterns
, 0, 0)
897 output_fd
= open(output_file
, 'w')
899 output_fd
= sys
.stdout
902 for n
in sorted(arguments
.keys()):
906 # A single translate function can be invoked for different patterns.
907 # Make sure that the argument sets are the same, and declare the
908 # function only once.
911 if i
.name
in out_pats
:
913 if i
.base
.base
!= p
.base
.base
:
914 error(0, i
.name
, ' has conflicting argument sets')
920 for n
in sorted(formats
.keys()):
924 output(decode_scope
, 'bool ', decode_function
,
925 '(DisasContext *ctx, ', insntype
, ' insn)\n{\n')
928 output(i4
, 'union {\n')
929 for n
in sorted(arguments
.keys()):
931 output(i4
, i4
, f
.struct_name(), ' f_', f
.name
, ';\n')
932 output(i4
, '} u;\n\n')
934 t
.output_code(4, False, 0, 0)
943 if __name__
== '__main__':