1 # Copyright (c) 2014 ARM Limited
4 # The license below extends only to copyright in the software and shall
5 # not be construed as granting a license to any other intellectual
6 # property including but not limited to intellectual property relating
7 # to a hardware implementation of the functionality of the software
8 # licensed hereunder. You may use the software subject to the license
9 # terms below provided that you ensure that this notice is replicated
10 # unmodified and in its entirety in all distributions of the software,
11 # modified or unmodified, in source code or in binary form.
13 # Copyright (c) 2003-2005 The Regents of The University of Michigan
14 # Copyright (c) 2013 Advanced Micro Devices, Inc.
15 # All rights reserved.
17 # Redistribution and use in source and binary forms, with or without
18 # modification, are permitted provided that the following conditions are
19 # met: redistributions of source code must retain the above copyright
20 # notice, this list of conditions and the following disclaimer;
21 # redistributions in binary form must reproduce the above copyright
22 # notice, this list of conditions and the following disclaimer in the
23 # documentation and/or other materials provided with the distribution;
24 # neither the name of the copyright holders nor the names of its
25 # contributors may be used to endorse or promote products derived from
26 # this software without specific prior written permission.
28 # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
29 # "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
30 # LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
31 # A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
32 # OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
33 # SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
34 # LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
35 # DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
36 # THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
37 # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
38 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
40 # Authors: Steve Reinhardt
42 from __future__
import with_statement
47 import inspect
, traceback
51 from m5
.util
.grammar
import Grammar
59 # Indent every line in string 's' by two spaces
60 # (except preprocessor directives).
61 # Used to make nested code blocks look pretty.
64 return re
.sub(r
'(?m)^(?!#)', ' ', s
)
67 # Munge a somewhat arbitrarily formatted piece of Python code
68 # (e.g. from a format 'let' block) into something whose indentation
69 # will get by the Python parser.
71 # The two keys here are that Python will give a syntax error if
72 # there's any whitespace at the beginning of the first line, and that
73 # all lines at the same lexical nesting level must have identical
74 # indentation. Unfortunately the way code literals work, an entire
75 # let block tends to have some initial indentation. Rather than
76 # trying to figure out what that is and strip it off, we prepend 'if
77 # 1:' to make the let code the nested block inside the if (and have
78 # the parser automatically deal with the indentation for us).
80 # We don't want to do this if (1) the code block is empty or (2) the
81 # first line of the block doesn't have any whitespace at the front.
83 def fixPythonIndentation(s
):
84 # get rid of blank lines first
85 s
= re
.sub(r
'(?m)^\s*\n', '', s
);
86 if (s
!= '' and re
.match(r
'[ \t]', s
[0])):
90 class ISAParserError(Exception):
91 """Error handler for parser errors"""
92 def __init__(self
, first
, second
=None):
97 if hasattr(first
, 'lexer'):
98 first
= first
.lexer
.lineno
102 def display(self
, filename_stack
, print_traceback
=debug
):
103 # Output formatted to work under Emacs compile-mode. Optional
104 # 'print_traceback' arg, if set to True, prints a Python stack
105 # backtrace too (can be handy when trying to debug the parser
109 for (filename
, line
) in filename_stack
[:-1]:
110 print "%sIn file included from %s:" % (spaces
, filename
)
113 # Print a Python stack backtrace if requested.
114 if print_traceback
or not self
.lineno
:
115 traceback
.print_exc()
117 line_str
= "%s:" % (filename_stack
[-1][0], )
119 line_str
+= "%d:" % (self
.lineno
, )
121 return "%s%s %s" % (spaces
, line_str
, self
.string
)
123 def exit(self
, filename_stack
, print_traceback
=debug
):
126 sys
.exit(self
.display(filename_stack
, print_traceback
))
129 raise ISAParserError(*args
)
134 # Template objects are format strings that allow substitution from
135 # the attribute spaces of other objects (e.g. InstObjParams instances).
137 labelRE
= re
.compile(r
'(?<!%)%\(([^\)]+)\)[sd]')
139 class Template(object):
140 def __init__(self
, parser
, t
):
147 # Protect non-Python-dict substitutions (e.g. if there's a printf
148 # in the templated C++ code)
149 template
= self
.parser
.protectNonSubstPercents(self
.template
)
150 # CPU-model-specific substitutions are handled later (in GenCode).
151 template
= self
.parser
.protectCpuSymbols(template
)
153 # Build a dict ('myDict') to use for the template substitution.
154 # Start with the template namespace. Make a copy since we're
155 # going to modify it.
156 myDict
= self
.parser
.templateMap
.copy()
158 if isinstance(d
, InstObjParams
):
159 # If we're dealing with an InstObjParams object, we need
160 # to be a little more sophisticated. The instruction-wide
161 # parameters are already formed, but the parameters which
162 # are only function wide still need to be generated.
165 myDict
.update(d
.__dict
__)
166 # The "operands" and "snippets" attributes of the InstObjParams
167 # objects are for internal use and not substitution.
168 del myDict
['operands']
169 del myDict
['snippets']
171 snippetLabels
= [l
for l
in labelRE
.findall(template
)
172 if d
.snippets
.has_key(l
)]
174 snippets
= dict([(s
, self
.parser
.mungeSnippet(d
.snippets
[s
]))
175 for s
in snippetLabels
])
177 myDict
.update(snippets
)
179 compositeCode
= ' '.join(map(str, snippets
.values()))
181 # Add in template itself in case it references any
182 # operands explicitly (like Mem)
183 compositeCode
+= ' ' + template
185 operands
= SubOperandList(self
.parser
, compositeCode
, d
.operands
)
187 myDict
['op_decl'] = operands
.concatAttrStrings('op_decl')
188 if operands
.readPC
or operands
.setPC
:
189 myDict
['op_decl'] += 'TheISA::PCState __parserAutoPCState;\n'
191 # In case there are predicated register reads and write, declare
192 # the variables for register indicies. It is being assumed that
193 # all the operands in the OperandList are also in the
194 # SubOperandList and in the same order. Otherwise, it is
195 # expected that predication would not be used for the operands.
196 if operands
.predRead
:
197 myDict
['op_decl'] += 'uint8_t _sourceIndex = 0;\n'
198 if operands
.predWrite
:
199 myDict
['op_decl'] += 'uint8_t M5_VAR_USED _destIndex = 0;\n'
201 is_src
= lambda op
: op
.is_src
202 is_dest
= lambda op
: op
.is_dest
204 myDict
['op_src_decl'] = \
205 operands
.concatSomeAttrStrings(is_src
, 'op_src_decl')
206 myDict
['op_dest_decl'] = \
207 operands
.concatSomeAttrStrings(is_dest
, 'op_dest_decl')
209 myDict
['op_src_decl'] += \
210 'TheISA::PCState __parserAutoPCState;\n'
212 myDict
['op_dest_decl'] += \
213 'TheISA::PCState __parserAutoPCState;\n'
215 myDict
['op_rd'] = operands
.concatAttrStrings('op_rd')
217 myDict
['op_rd'] = '__parserAutoPCState = xc->pcState();\n' + \
220 # Compose the op_wb string. If we're going to write back the
221 # PC state because we changed some of its elements, we'll need to
222 # do that as early as possible. That allows later uncoordinated
223 # modifications to the PC to layer appropriately.
224 reordered
= list(operands
.items
)
227 pcWbStr
= 'xc->pcState(__parserAutoPCState);\n'
228 for op_desc
in reordered
:
229 if op_desc
.isPCPart() and op_desc
.is_dest
:
230 op_wb_str
= op_desc
.op_wb
+ pcWbStr
+ op_wb_str
233 op_wb_str
= op_desc
.op_wb
+ op_wb_str
234 myDict
['op_wb'] = op_wb_str
236 elif isinstance(d
, dict):
237 # if the argument is a dictionary, we just use it.
239 elif hasattr(d
, '__dict__'):
240 # if the argument is an object, we use its attribute map.
241 myDict
.update(d
.__dict
__)
243 raise TypeError, "Template.subst() arg must be or have dictionary"
244 return template
% myDict
246 # Convert to string. This handles the case when a template with a
247 # CPU-specific term gets interpolated into another template or into
250 return self
.parser
.expandCpuSymbolsToString(self
.template
)
255 # A format object encapsulates an instruction format. It must provide
256 # a defineInst() method that generates the code for an instruction
259 class Format(object):
260 def __init__(self
, id, params
, code
):
263 label
= 'def format ' + id
264 self
.user_code
= compile(fixPythonIndentation(code
), label
, 'exec')
265 param_list
= string
.join(params
, ", ")
266 f
= '''def defInst(_code, _context, %s):
267 my_locals = vars().copy()
268 exec _code in _context, my_locals
269 return my_locals\n''' % param_list
270 c
= compile(f
, label
+ ' wrapper', 'exec')
274 def defineInst(self
, parser
, name
, args
, lineno
):
275 parser
.updateExportContext()
276 context
= parser
.exportContext
.copy()
278 Name
= name
[0].upper()
281 context
.update({ 'name' : name
, 'Name' : Name
})
283 vars = self
.func(self
.user_code
, context
, *args
[0], **args
[1])
284 except Exception, exc
:
287 error(lineno
, 'error defining "%s": %s.' % (name
, exc
))
288 for k
in vars.keys():
289 if k
not in ('header_output', 'decoder_output',
290 'exec_output', 'decode_block'):
292 return GenCode(parser
, **vars)
294 # Special null format to catch an implicit-format instruction
295 # definition outside of any format block.
296 class NoFormat(object):
298 self
.defaultInst
= ''
300 def defineInst(self
, parser
, name
, args
, lineno
):
302 'instruction definition "%s" with no active format!' % name
)
307 # The GenCode class encapsulates generated code destined for various
308 # output files. The header_output and decoder_output attributes are
309 # strings containing code destined for decoder.hh and decoder.cc
310 # respectively. The decode_block attribute contains code to be
311 # incorporated in the decode function itself (that will also end up in
312 # decoder.cc). The exec_output attribute is a dictionary with a key
313 # for each CPU model name; the value associated with a particular key
314 # is the string of code for that CPU model's exec.cc file. The
315 # has_decode_default attribute is used in the decode block to allow
316 # explicit default clauses to override default default clauses.
318 class GenCode(object):
319 # Constructor. At this point we substitute out all CPU-specific
320 # symbols. For the exec output, these go into the per-model
321 # dictionary. For all other output types they get collapsed into
323 def __init__(self
, parser
,
324 header_output
= '', decoder_output
= '', exec_output
= '',
325 decode_block
= '', has_decode_default
= False):
327 self
.header_output
= parser
.expandCpuSymbolsToString(header_output
)
328 self
.decoder_output
= parser
.expandCpuSymbolsToString(decoder_output
)
329 self
.exec_output
= exec_output
330 self
.decode_block
= decode_block
331 self
.has_decode_default
= has_decode_default
333 # Write these code chunks out to the filesystem. They will be properly
334 # interwoven by the write_top_level_files().
336 if self
.header_output
:
337 self
.parser
.get_file('header').write(self
.header_output
)
338 if self
.decoder_output
:
339 self
.parser
.get_file('decoder').write(self
.decoder_output
)
341 self
.parser
.get_file('exec').write(self
.exec_output
)
342 if self
.decode_block
:
343 self
.parser
.get_file('decode_block').write(self
.decode_block
)
345 # Override '+' operator: generate a new GenCode object that
346 # concatenates all the individual strings in the operands.
347 def __add__(self
, other
):
348 return GenCode(self
.parser
,
349 self
.header_output
+ other
.header_output
,
350 self
.decoder_output
+ other
.decoder_output
,
351 self
.exec_output
+ other
.exec_output
,
352 self
.decode_block
+ other
.decode_block
,
353 self
.has_decode_default
or other
.has_decode_default
)
355 # Prepend a string (typically a comment) to all the strings.
356 def prepend_all(self
, pre
):
357 self
.header_output
= pre
+ self
.header_output
358 self
.decoder_output
= pre
+ self
.decoder_output
359 self
.decode_block
= pre
+ self
.decode_block
360 self
.exec_output
= pre
+ self
.exec_output
362 # Wrap the decode block in a pair of strings (e.g., 'case foo:'
363 # and 'break;'). Used to build the big nested switch statement.
364 def wrap_decode_block(self
, pre
, post
= ''):
365 self
.decode_block
= pre
+ indent(self
.decode_block
) + post
367 #####################################################################
369 # Bitfield Operator Support
371 #####################################################################
373 bitOp1ArgRE
= re
.compile(r
'<\s*(\w+)\s*:\s*>')
375 bitOpWordRE
= re
.compile(r
'(?<![\w\.])([\w\.]+)<\s*(\w+)\s*:\s*(\w+)\s*>')
376 bitOpExprRE
= re
.compile(r
'\)<\s*(\w+)\s*:\s*(\w+)\s*>')
378 def substBitOps(code
):
379 # first convert single-bit selectors to two-index form
380 # i.e., <n> --> <n:n>
381 code
= bitOp1ArgRE
.sub(r
'<\1:\1>', code
)
382 # simple case: selector applied to ID (name)
383 # i.e., foo<a:b> --> bits(foo, a, b)
384 code
= bitOpWordRE
.sub(r
'bits(\1, \2, \3)', code
)
385 # if selector is applied to expression (ending in ')'),
386 # we need to search backward for matching '('
387 match
= bitOpExprRE
.search(code
)
389 exprEnd
= match
.start()
393 if code
[here
] == '(':
395 elif code
[here
] == ')':
399 sys
.exit("Didn't find '('!")
401 newExpr
= r
'bits(%s, %s, %s)' % (code
[exprStart
:exprEnd
+1],
402 match
.group(1), match
.group(2))
403 code
= code
[:exprStart
] + newExpr
+ code
[match
.end():]
404 match
= bitOpExprRE
.search(code
)
408 #####################################################################
412 # The remaining code is the support for automatically extracting
413 # instruction characteristics from pseudocode.
415 #####################################################################
417 # Force the argument to be a list. Useful for flags, where a caller
418 # can specify a singleton flag or a list of flags. Also usful for
419 # converting tuples to lists so they can be modified.
421 if isinstance(arg
, list):
423 elif isinstance(arg
, tuple):
430 class Operand(object):
431 '''Base class for operand descriptors. An instance of this class
432 (or actually a class derived from this one) represents a specific
433 operand for a code block (e.g, "Rc.sq" as a dest). Intermediate
434 derived classes encapsulates the traits of a particular operand
435 type (e.g., "32-bit integer register").'''
437 def buildReadCode(self
, func
= None):
438 subst_dict
= {"name": self
.base_name
,
440 "reg_idx": self
.reg_spec
,
442 if hasattr(self
, 'src_reg_idx'):
443 subst_dict
['op_idx'] = self
.src_reg_idx
444 code
= self
.read_code
% subst_dict
445 return '%s = %s;\n' % (self
.base_name
, code
)
447 def buildWriteCode(self
, func
= None):
448 subst_dict
= {"name": self
.base_name
,
450 "reg_idx": self
.reg_spec
,
452 "final_val": self
.base_name
}
453 if hasattr(self
, 'dest_reg_idx'):
454 subst_dict
['op_idx'] = self
.dest_reg_idx
455 code
= self
.write_code
% subst_dict
460 if (traceData) { traceData->setData(final_val); }
461 }''' % (self
.dflt_ctype
, self
.base_name
, code
)
463 def __init__(self
, parser
, full_name
, ext
, is_src
, is_dest
):
464 self
.full_name
= full_name
467 self
.is_dest
= is_dest
468 # The 'effective extension' (eff_ext) is either the actual
469 # extension, if one was explicitly provided, or the default.
472 elif hasattr(self
, 'dflt_ext'):
473 self
.eff_ext
= self
.dflt_ext
475 if hasattr(self
, 'eff_ext'):
476 self
.ctype
= parser
.operandTypeMap
[self
.eff_ext
]
478 # Finalize additional fields (primarily code fields). This step
479 # is done separately since some of these fields may depend on the
480 # register index enumeration that hasn't been performed yet at the
481 # time of __init__(). The register index enumeration is affected
482 # by predicated register reads/writes. Hence, we forward the flags
483 # that indicate whether or not predication is in use.
484 def finalize(self
, predRead
, predWrite
):
485 self
.flags
= self
.getFlags()
486 self
.constructor
= self
.makeConstructor(predRead
, predWrite
)
487 self
.op_decl
= self
.makeDecl()
490 self
.op_rd
= self
.makeRead(predRead
)
491 self
.op_src_decl
= self
.makeDecl()
494 self
.op_src_decl
= ''
497 self
.op_wb
= self
.makeWrite(predWrite
)
498 self
.op_dest_decl
= self
.makeDecl()
501 self
.op_dest_decl
= ''
509 def isFloatReg(self
):
518 def isControlReg(self
):
525 return self
.isPCState() and self
.reg_spec
527 def hasReadPred(self
):
528 return self
.read_predicate
!= None
530 def hasWritePred(self
):
531 return self
.write_predicate
!= None
534 # note the empty slice '[:]' gives us a copy of self.flags[0]
535 # instead of a reference to it
536 my_flags
= self
.flags
[0][:]
538 my_flags
+= self
.flags
[1]
540 my_flags
+= self
.flags
[2]
544 # Note that initializations in the declarations are solely
545 # to avoid 'uninitialized variable' errors from the compiler.
546 return self
.ctype
+ ' ' + self
.base_name
+ ' = 0;\n';
548 class IntRegOperand(Operand
):
555 def makeConstructor(self
, predRead
, predWrite
):
560 c_src
= '\n\t_srcRegIdx[_numSrcRegs++] = %s;' % (self
.reg_spec
)
561 if self
.hasReadPred():
562 c_src
= '\n\tif (%s) {%s\n\t}' % \
563 (self
.read_predicate
, c_src
)
566 c_dest
= '\n\t_destRegIdx[_numDestRegs++] = %s;' % \
568 c_dest
+= '\n\t_numIntDestRegs++;'
569 if self
.hasWritePred():
570 c_dest
= '\n\tif (%s) {%s\n\t}' % \
571 (self
.write_predicate
, c_dest
)
573 return c_src
+ c_dest
575 def makeRead(self
, predRead
):
576 if (self
.ctype
== 'float' or self
.ctype
== 'double'):
577 error('Attempt to read integer register as FP')
578 if self
.read_code
!= None:
579 return self
.buildReadCode('readIntRegOperand')
583 int_reg_val
= 'xc->readIntRegOperand(this, _sourceIndex++)'
584 if self
.hasReadPred():
585 int_reg_val
= '(%s) ? %s : 0' % \
586 (self
.read_predicate
, int_reg_val
)
588 int_reg_val
= 'xc->readIntRegOperand(this, %d)' % self
.src_reg_idx
590 return '%s = %s;\n' % (self
.base_name
, int_reg_val
)
592 def makeWrite(self
, predWrite
):
593 if (self
.ctype
== 'float' or self
.ctype
== 'double'):
594 error('Attempt to write integer register as FP')
595 if self
.write_code
!= None:
596 return self
.buildWriteCode('setIntRegOperand')
600 if self
.hasWritePred():
601 wp
= self
.write_predicate
603 wcond
= 'if (%s)' % (wp
)
604 windex
= '_destIndex++'
607 windex
= '%d' % self
.dest_reg_idx
613 xc->setIntRegOperand(this, %s, final_val);\n
614 if (traceData) { traceData->setData(final_val); }
615 }''' % (wcond
, self
.ctype
, self
.base_name
, windex
)
619 class FloatRegOperand(Operand
):
623 def isFloatReg(self
):
626 def makeConstructor(self
, predRead
, predWrite
):
631 c_src
= '\n\t_srcRegIdx[_numSrcRegs++] = %s + FP_Reg_Base;' % \
636 '\n\t_destRegIdx[_numDestRegs++] = %s + FP_Reg_Base;' % \
638 c_dest
+= '\n\t_numFPDestRegs++;'
640 return c_src
+ c_dest
642 def makeRead(self
, predRead
):
644 if (self
.ctype
== 'float' or self
.ctype
== 'double'):
645 func
= 'readFloatRegOperand'
647 func
= 'readFloatRegOperandBits'
648 if self
.read_code
!= None:
649 return self
.buildReadCode(func
)
652 rindex
= '_sourceIndex++'
654 rindex
= '%d' % self
.src_reg_idx
656 return '%s = xc->%s(this, %s);\n' % \
657 (self
.base_name
, func
, rindex
)
659 def makeWrite(self
, predWrite
):
660 if (self
.ctype
== 'float' or self
.ctype
== 'double'):
661 func
= 'setFloatRegOperand'
663 func
= 'setFloatRegOperandBits'
664 if self
.write_code
!= None:
665 return self
.buildWriteCode(func
)
670 wp
= '%d' % self
.dest_reg_idx
671 wp
= 'xc->%s(this, %s, final_val);' % (func
, wp
)
677 if (traceData) { traceData->setData(final_val); }
678 }''' % (self
.ctype
, self
.base_name
, wp
)
681 class CCRegOperand(Operand
):
688 def makeConstructor(self
, predRead
, predWrite
):
693 c_src
= '\n\t_srcRegIdx[_numSrcRegs++] = %s + CC_Reg_Base;' % \
695 if self
.hasReadPred():
696 c_src
= '\n\tif (%s) {%s\n\t}' % \
697 (self
.read_predicate
, c_src
)
701 '\n\t_destRegIdx[_numDestRegs++] = %s + CC_Reg_Base;' % \
703 c_dest
+= '\n\t_numCCDestRegs++;'
704 if self
.hasWritePred():
705 c_dest
= '\n\tif (%s) {%s\n\t}' % \
706 (self
.write_predicate
, c_dest
)
708 return c_src
+ c_dest
710 def makeRead(self
, predRead
):
711 if (self
.ctype
== 'float' or self
.ctype
== 'double'):
712 error('Attempt to read condition-code register as FP')
713 if self
.read_code
!= None:
714 return self
.buildReadCode('readCCRegOperand')
718 int_reg_val
= 'xc->readCCRegOperand(this, _sourceIndex++)'
719 if self
.hasReadPred():
720 int_reg_val
= '(%s) ? %s : 0' % \
721 (self
.read_predicate
, int_reg_val
)
723 int_reg_val
= 'xc->readCCRegOperand(this, %d)' % self
.src_reg_idx
725 return '%s = %s;\n' % (self
.base_name
, int_reg_val
)
727 def makeWrite(self
, predWrite
):
728 if (self
.ctype
== 'float' or self
.ctype
== 'double'):
729 error('Attempt to write condition-code register as FP')
730 if self
.write_code
!= None:
731 return self
.buildWriteCode('setCCRegOperand')
735 if self
.hasWritePred():
736 wp
= self
.write_predicate
738 wcond
= 'if (%s)' % (wp
)
739 windex
= '_destIndex++'
742 windex
= '%d' % self
.dest_reg_idx
748 xc->setCCRegOperand(this, %s, final_val);\n
749 if (traceData) { traceData->setData(final_val); }
750 }''' % (wcond
, self
.ctype
, self
.base_name
, windex
)
754 class ControlRegOperand(Operand
):
758 def isControlReg(self
):
761 def makeConstructor(self
, predRead
, predWrite
):
767 '\n\t_srcRegIdx[_numSrcRegs++] = %s + Misc_Reg_Base;' % \
772 '\n\t_destRegIdx[_numDestRegs++] = %s + Misc_Reg_Base;' % \
775 return c_src
+ c_dest
777 def makeRead(self
, predRead
):
779 if (self
.ctype
== 'float' or self
.ctype
== 'double'):
780 error('Attempt to read control register as FP')
781 if self
.read_code
!= None:
782 return self
.buildReadCode('readMiscRegOperand')
785 rindex
= '_sourceIndex++'
787 rindex
= '%d' % self
.src_reg_idx
789 return '%s = xc->readMiscRegOperand(this, %s);\n' % \
790 (self
.base_name
, rindex
)
792 def makeWrite(self
, predWrite
):
793 if (self
.ctype
== 'float' or self
.ctype
== 'double'):
794 error('Attempt to write control register as FP')
795 if self
.write_code
!= None:
796 return self
.buildWriteCode('setMiscRegOperand')
799 windex
= '_destIndex++'
801 windex
= '%d' % self
.dest_reg_idx
803 wb
= 'xc->setMiscRegOperand(this, %s, %s);\n' % \
804 (windex
, self
.base_name
)
805 wb
+= 'if (traceData) { traceData->setData(%s); }' % \
810 class MemOperand(Operand
):
814 def makeConstructor(self
, predRead
, predWrite
):
818 # Note that initializations in the declarations are solely
819 # to avoid 'uninitialized variable' errors from the compiler.
820 # Declare memory data variable.
821 return '%s %s = 0;\n' % (self
.ctype
, self
.base_name
)
823 def makeRead(self
, predRead
):
824 if self
.read_code
!= None:
825 return self
.buildReadCode()
828 def makeWrite(self
, predWrite
):
829 if self
.write_code
!= None:
830 return self
.buildWriteCode()
833 class PCStateOperand(Operand
):
834 def makeConstructor(self
, predRead
, predWrite
):
837 def makeRead(self
, predRead
):
839 # A component of the PC state.
840 return '%s = __parserAutoPCState.%s();\n' % \
841 (self
.base_name
, self
.reg_spec
)
843 # The whole PC state itself.
844 return '%s = xc->pcState();\n' % self
.base_name
846 def makeWrite(self
, predWrite
):
848 # A component of the PC state.
849 return '__parserAutoPCState.%s(%s);\n' % \
850 (self
.reg_spec
, self
.base_name
)
852 # The whole PC state itself.
853 return 'xc->pcState(%s);\n' % self
.base_name
856 ctype
= 'TheISA::PCState'
859 # Note that initializations in the declarations are solely
860 # to avoid 'uninitialized variable' errors from the compiler.
861 return '%s %s = 0;\n' % (ctype
, self
.base_name
)
866 class OperandList(object):
867 '''Find all the operands in the given code block. Returns an operand
868 descriptor list (instance of class OperandList).'''
869 def __init__(self
, parser
, code
):
872 # delete strings and comments so we don't match on operands inside
873 for regEx
in (stringRE
, commentRE
):
874 code
= regEx
.sub('', code
)
875 # search for operands
878 match
= parser
.operandsRE
.search(code
, next_pos
)
880 # no more matches: we're done
883 # regexp groups are operand full name, base, and extension
884 (op_full
, op_base
, op_ext
) = op
885 # if the token following the operand is an assignment, this is
886 # a destination (LHS), else it's a source (RHS)
887 is_dest
= (assignRE
.match(code
, match
.end()) != None)
889 # see if we've already seen this one
890 op_desc
= self
.find_base(op_base
)
892 if op_desc
.ext
!= op_ext
:
893 error('Inconsistent extensions for operand %s' % \
895 op_desc
.is_src
= op_desc
.is_src
or is_src
896 op_desc
.is_dest
= op_desc
.is_dest
or is_dest
898 # new operand: create new descriptor
899 op_desc
= parser
.operandNameMap
[op_base
](parser
,
900 op_full
, op_ext
, is_src
, is_dest
)
902 # start next search after end of current match
903 next_pos
= match
.end()
905 # enumerate source & dest register operands... used in building
909 self
.numFPDestRegs
= 0
910 self
.numIntDestRegs
= 0
911 self
.numCCDestRegs
= 0
912 self
.numMiscDestRegs
= 0
913 self
.memOperand
= None
915 # Flags to keep track if one or more operands are to be read/written
917 self
.predRead
= False
918 self
.predWrite
= False
920 for op_desc
in self
.items
:
923 op_desc
.src_reg_idx
= self
.numSrcRegs
926 op_desc
.dest_reg_idx
= self
.numDestRegs
927 self
.numDestRegs
+= 1
928 if op_desc
.isFloatReg():
929 self
.numFPDestRegs
+= 1
930 elif op_desc
.isIntReg():
931 self
.numIntDestRegs
+= 1
932 elif op_desc
.isCCReg():
933 self
.numCCDestRegs
+= 1
934 elif op_desc
.isControlReg():
935 self
.numMiscDestRegs
+= 1
936 elif op_desc
.isMem():
938 error("Code block has more than one memory operand.")
939 self
.memOperand
= op_desc
941 # Check if this operand has read/write predication. If true, then
942 # the microop will dynamically index source/dest registers.
943 self
.predRead
= self
.predRead
or op_desc
.hasReadPred()
944 self
.predWrite
= self
.predWrite
or op_desc
.hasWritePred()
946 if parser
.maxInstSrcRegs
< self
.numSrcRegs
:
947 parser
.maxInstSrcRegs
= self
.numSrcRegs
948 if parser
.maxInstDestRegs
< self
.numDestRegs
:
949 parser
.maxInstDestRegs
= self
.numDestRegs
950 if parser
.maxMiscDestRegs
< self
.numMiscDestRegs
:
951 parser
.maxMiscDestRegs
= self
.numMiscDestRegs
953 # now make a final pass to finalize op_desc fields that may depend
954 # on the register enumeration
955 for op_desc
in self
.items
:
956 op_desc
.finalize(self
.predRead
, self
.predWrite
)
959 return len(self
.items
)
961 def __getitem__(self
, index
):
962 return self
.items
[index
]
964 def append(self
, op_desc
):
965 self
.items
.append(op_desc
)
966 self
.bases
[op_desc
.base_name
] = op_desc
968 def find_base(self
, base_name
):
969 # like self.bases[base_name], but returns None if not found
970 # (rather than raising exception)
971 return self
.bases
.get(base_name
)
973 # internal helper function for concat[Some]Attr{Strings|Lists}
974 def __internalConcatAttrs(self
, attr_name
, filter, result
):
975 for op_desc
in self
.items
:
977 result
+= getattr(op_desc
, attr_name
)
980 # return a single string that is the concatenation of the (string)
981 # values of the specified attribute for all operands
982 def concatAttrStrings(self
, attr_name
):
983 return self
.__internalConcatAttrs
(attr_name
, lambda x
: 1, '')
985 # like concatAttrStrings, but only include the values for the operands
986 # for which the provided filter function returns true
987 def concatSomeAttrStrings(self
, filter, attr_name
):
988 return self
.__internalConcatAttrs
(attr_name
, filter, '')
990 # return a single list that is the concatenation of the (list)
991 # values of the specified attribute for all operands
992 def concatAttrLists(self
, attr_name
):
993 return self
.__internalConcatAttrs
(attr_name
, lambda x
: 1, [])
995 # like concatAttrLists, but only include the values for the operands
996 # for which the provided filter function returns true
997 def concatSomeAttrLists(self
, filter, attr_name
):
998 return self
.__internalConcatAttrs
(attr_name
, filter, [])
1001 self
.items
.sort(lambda a
, b
: a
.sort_pri
- b
.sort_pri
)
1003 class SubOperandList(OperandList
):
1004 '''Find all the operands in the given code block. Returns an operand
1005 descriptor list (instance of class OperandList).'''
1006 def __init__(self
, parser
, code
, master_list
):
1009 # delete strings and comments so we don't match on operands inside
1010 for regEx
in (stringRE
, commentRE
):
1011 code
= regEx
.sub('', code
)
1012 # search for operands
1015 match
= parser
.operandsRE
.search(code
, next_pos
)
1017 # no more matches: we're done
1020 # regexp groups are operand full name, base, and extension
1021 (op_full
, op_base
, op_ext
) = op
1022 # find this op in the master list
1023 op_desc
= master_list
.find_base(op_base
)
1025 error('Found operand %s which is not in the master list!' \
1026 ' This is an internal error' % op_base
)
1028 # See if we've already found this operand
1029 op_desc
= self
.find_base(op_base
)
1031 # if not, add a reference to it to this sub list
1032 self
.append(master_list
.bases
[op_base
])
1034 # start next search after end of current match
1035 next_pos
= match
.end()
1037 self
.memOperand
= None
1038 # Whether the whole PC needs to be read so parts of it can be accessed
1040 # Whether the whole PC needs to be written after parts of it were
1043 # Whether this instruction manipulates the whole PC or parts of it.
1044 # Mixing the two is a bad idea and flagged as an error.
1047 # Flags to keep track if one or more operands are to be read/written
1049 self
.predRead
= False
1050 self
.predWrite
= False
1052 for op_desc
in self
.items
:
1053 if op_desc
.isPCPart():
1058 if op_desc
.isPCState():
1059 if self
.pcPart
is not None:
1060 if self
.pcPart
and not op_desc
.isPCPart() or \
1061 not self
.pcPart
and op_desc
.isPCPart():
1062 error("Mixed whole and partial PC state operands.")
1063 self
.pcPart
= op_desc
.isPCPart()
1067 error("Code block has more than one memory operand.")
1068 self
.memOperand
= op_desc
1070 # Check if this operand has read/write predication. If true, then
1071 # the microop will dynamically index source/dest registers.
1072 self
.predRead
= self
.predRead
or op_desc
.hasReadPred()
1073 self
.predWrite
= self
.predWrite
or op_desc
.hasWritePred()
1075 # Regular expression object to match C++ strings
1076 stringRE
= re
.compile(r
'"([^"\\]|\\.)*"')
1078 # Regular expression object to match C++ comments
1079 # (used in findOperands())
1080 commentRE
= re
.compile(r
'(^)?[^\S\n]*/(?:\*(.*?)\*/[^\S\n]*|/[^\n]*)($)?',
1081 re
.DOTALL | re
.MULTILINE
)
1083 # Regular expression object to match assignment statements
1084 # (used in findOperands())
1085 assignRE
= re
.compile(r
'\s*=(?!=)', re
.MULTILINE
)
1087 def makeFlagConstructor(flag_list
):
1088 if len(flag_list
) == 0:
1090 # filter out repeated flags
1093 while i
< len(flag_list
):
1094 if flag_list
[i
] == flag_list
[i
-1]:
1100 code
= pre
+ string
.join(flag_list
, post
+ pre
) + post
1103 # Assume all instruction flags are of the form 'IsFoo'
1104 instFlagRE
= re
.compile(r
'Is.*')
1106 # OpClass constants end in 'Op' except No_OpClass
1107 opClassRE
= re
.compile(r
'.*Op|No_OpClass')
1109 class InstObjParams(object):
1110 def __init__(self
, parser
, mnem
, class_name
, base_class
= '',
1111 snippets
= {}, opt_args
= []):
1112 self
.mnemonic
= mnem
1113 self
.class_name
= class_name
1114 self
.base_class
= base_class
1115 if not isinstance(snippets
, dict):
1116 snippets
= {'code' : snippets
}
1117 compositeCode
= ' '.join(map(str, snippets
.values()))
1118 self
.snippets
= snippets
1120 self
.operands
= OperandList(parser
, compositeCode
)
1122 # The header of the constructor declares the variables to be used
1123 # in the body of the constructor.
1125 header
+= '\n\t_numSrcRegs = 0;'
1126 header
+= '\n\t_numDestRegs = 0;'
1127 header
+= '\n\t_numFPDestRegs = 0;'
1128 header
+= '\n\t_numIntDestRegs = 0;'
1129 header
+= '\n\t_numCCDestRegs = 0;'
1131 self
.constructor
= header
+ \
1132 self
.operands
.concatAttrStrings('constructor')
1134 self
.flags
= self
.operands
.concatAttrLists('flags')
1136 self
.op_class
= None
1138 # Optional arguments are assumed to be either StaticInst flags
1139 # or an OpClass value. To avoid having to import a complete
1140 # list of these values to match against, we do it ad-hoc
1143 if instFlagRE
.match(oa
):
1144 self
.flags
.append(oa
)
1145 elif opClassRE
.match(oa
):
1148 error('InstObjParams: optional arg "%s" not recognized '
1149 'as StaticInst::Flag or OpClass.' % oa
)
1151 # Make a basic guess on the operand class if not set.
1152 # These are good enough for most cases.
1153 if not self
.op_class
:
1154 if 'IsStore' in self
.flags
:
1155 self
.op_class
= 'MemWriteOp'
1156 elif 'IsLoad' in self
.flags
or 'IsPrefetch' in self
.flags
:
1157 self
.op_class
= 'MemReadOp'
1158 elif 'IsFloating' in self
.flags
:
1159 self
.op_class
= 'FloatAddOp'
1161 self
.op_class
= 'IntAluOp'
1163 # add flag initialization to contructor here to include
1164 # any flags added via opt_args
1165 self
.constructor
+= makeFlagConstructor(self
.flags
)
1167 # if 'IsFloating' is set, add call to the FP enable check
1168 # function (which should be provided by isa_desc via a declare)
1169 if 'IsFloating' in self
.flags
:
1170 self
.fp_enable_check
= 'fault = checkFpEnableFault(xc);'
1172 self
.fp_enable_check
= ''
1175 # Stack: a simple stack object. Used for both formats (formatStack)
1176 # and default cases (defaultStack). Simply wraps a list to give more
1177 # stack-like syntax and enable initialization with an argument list
1178 # (as opposed to an argument that's a list).
1181 def __init__(self
, *items
):
1182 list.__init
__(self
, items
)
1184 def push(self
, item
):
1190 #######################
1193 # parses ISA DSL and emits C++ headers and source
1196 class ISAParser(Grammar
):
1197 class CpuModel(object):
1198 def __init__(self
, name
, filename
, includes
, strings
):
1200 self
.filename
= filename
1201 self
.includes
= includes
1202 self
.strings
= strings
1204 def __init__(self
, output_dir
):
1205 super(ISAParser
, self
).__init
__()
1206 self
.output_dir
= output_dir
1208 self
.filename
= None # for output file watermarking/scaremongering
1211 ISAParser
.CpuModel('ExecContext',
1212 'generic_cpu_exec.cc',
1213 '#include "cpu/exec_context.hh"',
1214 { "CPU_exec_context" : "ExecContext" }),
1217 # variable to hold templates
1218 self
.templateMap
= {}
1220 # This dictionary maps format name strings to Format objects.
1223 # Track open files and, if applicable, how many chunks it has been
1224 # split into so far.
1228 # isa_name / namespace identifier from namespace declaration.
1229 # before the namespace declaration, None.
1230 self
.isa_name
= None
1231 self
.namespace
= None
1234 self
.formatStack
= Stack(NoFormat())
1236 # The default case stack.
1237 self
.defaultStack
= Stack(None)
1239 # Stack that tracks current file and line number. Each
1240 # element is a tuple (filename, lineno) that records the
1241 # *current* filename and the line number in the *previous*
1242 # file where it was included.
1243 self
.fileNameStack
= Stack()
1245 symbols
= ('makeList', 're', 'string')
1246 self
.exportContext
= dict([(s
, eval(s
)) for s
in symbols
])
1248 self
.maxInstSrcRegs
= 0
1249 self
.maxInstDestRegs
= 0
1250 self
.maxMiscDestRegs
= 0
1252 def __getitem__(self
, i
): # Allow object (self) to be
1253 return getattr(self
, i
) # passed to %-substitutions
1255 # Change the file suffix of a base filename:
1256 # (e.g.) decoder.cc -> decoder-g.cc.inc for 'global' outputs
1257 def suffixize(self
, s
, sec
):
1258 extn
= re
.compile('(\.[^\.]+)$') # isolate extension
1260 return extn
.sub(r
'-ns\1.inc', s
) # insert some text on either side
1262 return extn
.sub(r
'-g\1.inc', s
)
1264 # Get the file object for emitting code into the specified section
1265 # (header, decoder, exec, decode_block).
1266 def get_file(self
, section
):
1267 if section
== 'decode_block':
1268 filename
= 'decode-method.cc.inc'
1270 if section
== 'header':
1273 file = '%s.cc' % section
1274 filename
= self
.suffixize(file, section
)
1276 return self
.files
[filename
]
1277 except KeyError: pass
1279 f
= self
.open(filename
)
1280 self
.files
[filename
] = f
1282 # The splittable files are the ones with many independent
1283 # per-instruction functions - the decoder's instruction constructors
1284 # and the instruction execution (execute()) methods. These both have
1285 # the suffix -ns.cc.inc, meaning they are within the namespace part
1286 # of the ISA, contain object-emitting C++ source, and are included
1287 # into other top-level files. These are the files that need special
1288 # #define's to allow parts of them to be compiled separately. Rather
1289 # than splitting the emissions into separate files, the monolithic
1290 # output of the ISA parser is maintained, but the value (or lack
1291 # thereof) of the __SPLIT definition during C preprocessing will
1292 # select the different chunks. If no 'split' directives are used,
1293 # the cpp emissions have no effect.
1294 if re
.search('-ns.cc.inc$', filename
):
1295 print >>f
, '#if !defined(__SPLIT) || (__SPLIT == 1)'
1297 # ensure requisite #include's
1298 elif filename
in ['decoder-g.cc.inc', 'exec-g.cc.inc']:
1299 print >>f
, '#include "decoder.hh"'
1300 elif filename
== 'decoder-g.hh.inc':
1301 print >>f
, '#include "base/bitfield.hh"'
1305 # Weave together the parts of the different output sections by
1306 # #include'ing them into some very short top-level .cc/.hh files.
1307 # These small files make it much clearer how this tool works, since
1308 # you directly see the chunks emitted as files that are #include'd.
1309 def write_top_level_files(self
):
1310 dep
= self
.open('inc.d', bare
=True)
1312 # decoder header - everything depends on this
1314 with self
.open(file) as f
:
1317 fn
= 'decoder-g.hh.inc'
1318 assert(fn
in self
.files
)
1319 f
.write('#include "%s"\n' % fn
)
1322 fn
= 'decoder-ns.hh.inc'
1323 assert(fn
in self
.files
)
1324 f
.write('namespace %s {\n#include "%s"\n}\n'
1325 % (self
.namespace
, fn
))
1328 print >>dep
, file+':', ' '.join(inc
)
1330 # decoder method - cannot be split
1332 with self
.open(file) as f
:
1335 fn
= 'decoder-g.cc.inc'
1336 assert(fn
in self
.files
)
1337 f
.write('#include "%s"\n' % fn
)
1340 fn
= 'decode-method.cc.inc'
1341 # is guaranteed to have been written for parse to complete
1342 f
.write('#include "%s"\n' % fn
)
1345 inc
.append("decoder.hh")
1346 print >>dep
, file+':', ' '.join(inc
)
1348 extn
= re
.compile('(\.[^\.]+)$')
1350 # instruction constructors
1351 splits
= self
.splits
[self
.get_file('decoder')]
1352 file_
= 'inst-constrs.cc'
1353 for i
in range(1, splits
+1):
1355 file = extn
.sub(r
'-%d\1' % i
, file_
)
1358 with self
.open(file) as f
:
1361 fn
= 'decoder-g.cc.inc'
1362 assert(fn
in self
.files
)
1363 f
.write('#include "%s"\n' % fn
)
1366 fn
= 'decoder-ns.cc.inc'
1367 assert(fn
in self
.files
)
1368 print >>f
, 'namespace %s {' % self
.namespace
1370 print >>f
, '#define __SPLIT %u' % i
1371 print >>f
, '#include "%s"' % fn
1375 inc
.append("decoder.hh")
1376 print >>dep
, file+':', ' '.join(inc
)
1378 # instruction execution per-CPU model
1379 splits
= self
.splits
[self
.get_file('exec')]
1380 for cpu
in self
.cpuModels
:
1381 for i
in range(1, splits
+1):
1383 file = extn
.sub(r
'_%d\1' % i
, cpu
.filename
)
1386 with self
.open(file) as f
:
1389 fn
= 'exec-g.cc.inc'
1390 assert(fn
in self
.files
)
1391 f
.write('#include "%s"\n' % fn
)
1394 f
.write(cpu
.includes
+"\n")
1396 fn
= 'exec-ns.cc.inc'
1397 assert(fn
in self
.files
)
1398 print >>f
, 'namespace %s {' % self
.namespace
1399 print >>f
, '#define CPU_EXEC_CONTEXT %s' \
1400 % cpu
.strings
['CPU_exec_context']
1402 print >>f
, '#define __SPLIT %u' % i
1403 print >>f
, '#include "%s"' % fn
1407 inc
.append("decoder.hh")
1408 print >>dep
, file+':', ' '.join(inc
)
1411 self
.update('max_inst_regs.hh',
1412 '''namespace %(namespace)s {
1413 const int MaxInstSrcRegs = %(maxInstSrcRegs)d;
1414 const int MaxInstDestRegs = %(maxInstDestRegs)d;
1415 const int MaxMiscDestRegs = %(maxMiscDestRegs)d;\n}\n''' % self
)
1416 print >>dep
, 'max_inst_regs.hh:'
1421 scaremonger_template
='''// DO NOT EDIT
1422 // This file was automatically generated from an ISA description:
1427 #####################################################################
1431 # The PLY lexer module takes two things as input:
1432 # - A list of token names (the string list 'tokens')
1433 # - A regular expression describing a match for each token. The
1434 # regexp for token FOO can be provided in two ways:
1435 # - as a string variable named t_FOO
1436 # - as the doc string for a function named t_FOO. In this case,
1437 # the function is also executed, allowing an action to be
1438 # associated with each token match.
1440 #####################################################################
1442 # Reserved words. These are listed separately as they are matched
1443 # using the same regexp as generic IDs, but distinguished in the
1444 # t_ID() function. The PLY documentation suggests this approach.
1446 'BITFIELD', 'DECODE', 'DECODER', 'DEFAULT', 'DEF', 'EXEC', 'FORMAT',
1447 'HEADER', 'LET', 'NAMESPACE', 'OPERAND_TYPES', 'OPERANDS',
1448 'OUTPUT', 'SIGNED', 'SPLIT', 'TEMPLATE'
1451 # List of tokens. The lex module requires this.
1452 tokens
= reserved
+ (
1465 # ( ) [ ] { } < > , ; . : :: *
1467 'LBRACKET', 'RBRACKET',
1469 'LESS', 'GREATER', 'EQUALS',
1470 'COMMA', 'SEMI', 'DOT', 'COLON', 'DBLCOLON',
1473 # C preprocessor directives
1476 # The following are matched but never returned. commented out to
1477 # suppress PLY warning
1485 # Regular expressions for token matching
1502 # Identifiers and reserved words
1505 reserved_map
[r
.lower()] = r
1509 t
.type = self
.reserved_map
.get(t
.value
, 'ID')
1513 def t_INTLIT(self
, t
):
1514 r
'-?(0x[\da-fA-F]+)|\d+'
1516 t
.value
= int(t
.value
,0)
1518 error(t
, 'Integer value "%s" too large' % t
.value
)
1522 # String literal. Note that these use only single quotes, and
1523 # can span multiple lines.
1524 def t_STRLIT(self
, t
):
1527 t
.value
= t
.value
[1:-1]
1528 t
.lexer
.lineno
+= t
.value
.count('\n')
1532 # "Code literal"... like a string literal, but delimiters are
1533 # '{{' and '}}' so they get formatted nicely under emacs c-mode
1534 def t_CODELIT(self
, t
):
1535 r
"(?m)\{\{([^\}]|}(?!\}))+\}\}"
1537 t
.value
= t
.value
[2:-2]
1538 t
.lexer
.lineno
+= t
.value
.count('\n')
1541 def t_CPPDIRECTIVE(self
, t
):
1543 t
.lexer
.lineno
+= t
.value
.count('\n')
1546 def t_NEWFILE(self
, t
):
1547 r
'^\#\#newfile\s+"[^"]*"'
1548 self
.fileNameStack
.push((t
.value
[11:-1], t
.lexer
.lineno
))
1551 def t_ENDFILE(self
, t
):
1553 (old_filename
, t
.lexer
.lineno
) = self
.fileNameStack
.pop()
1556 # The functions t_NEWLINE, t_ignore, and t_error are
1557 # special for the lex module.
1561 def t_NEWLINE(self
, t
):
1563 t
.lexer
.lineno
+= t
.value
.count('\n')
1566 def t_comment(self
, t
):
1569 # Completely ignored characters
1570 t_ignore
= ' \t\x0c'
1573 def t_error(self
, t
):
1574 error(t
, "illegal character '%s'" % t
.value
[0])
1577 #####################################################################
1581 # Every function whose name starts with 'p_' defines a grammar
1582 # rule. The rule is encoded in the function's doc string, while
1583 # the function body provides the action taken when the rule is
1584 # matched. The argument to each function is a list of the values
1585 # of the rule's symbols: t[0] for the LHS, and t[1..n] for the
1586 # symbols on the RHS. For tokens, the value is copied from the
1587 # t.value attribute provided by the lexer. For non-terminals, the
1588 # value is assigned by the producing rule; i.e., the job of the
1589 # grammar rule function is to set the value for the non-terminal
1590 # on the LHS (by assigning to t[0]).
1591 #####################################################################
1593 # The LHS of the first grammar rule is used as the start symbol
1594 # (in this case, 'specification'). Note that this rule enforces
1595 # that there will be exactly one namespace declaration, with 0 or
1596 # more global defs/decls before and after it. The defs & decls
1597 # before the namespace decl will be outside the namespace; those
1598 # after will be inside. The decoder function is always inside the
1600 def p_specification(self
, t
):
1601 'specification : opt_defs_and_outputs top_level_decode_block'
1603 for f
in self
.splits
.iterkeys():
1604 f
.write('\n#endif\n')
1606 for f
in self
.files
.itervalues(): # close ALL the files;
1607 f
.close() # not doing so can cause compilation to fail
1609 self
.write_top_level_files()
1613 # 'opt_defs_and_outputs' is a possibly empty sequence of def and/or
1614 # output statements. Its productions do the hard work of eventually
1615 # instantiating a GenCode, which are generally emitted (written to disk)
1616 # as soon as possible, except for the decode_block, which has to be
1617 # accumulated into one large function of nested switch/case blocks.
1618 def p_opt_defs_and_outputs_0(self
, t
):
1619 'opt_defs_and_outputs : empty'
1621 def p_opt_defs_and_outputs_1(self
, t
):
1622 'opt_defs_and_outputs : defs_and_outputs'
1624 def p_defs_and_outputs_0(self
, t
):
1625 'defs_and_outputs : def_or_output'
1627 def p_defs_and_outputs_1(self
, t
):
1628 'defs_and_outputs : defs_and_outputs def_or_output'
1630 # The list of possible definition/output statements.
1631 # They are all processed as they are seen.
1632 def p_def_or_output(self
, t
):
1633 '''def_or_output : name_decl
1636 | def_bitfield_struct
1644 # Utility function used by both invocations of splitting - explicit
1645 # 'split' keyword and split() function inside "let {{ }};" blocks.
1646 def split(self
, sec
, write
=False):
1647 assert(sec
!= 'header' and "header cannot be split")
1649 f
= self
.get_file(sec
)
1651 s
= '\n#endif\n#if __SPLIT == %u\n' % self
.splits
[f
]
1657 # split output file to reduce compilation time
1658 def p_split(self
, t
):
1659 'split : SPLIT output_type SEMI'
1660 assert(self
.isa_name
and "'split' not allowed before namespace decl")
1662 self
.split(t
[2], True)
1664 def p_output_type(self
, t
):
1665 '''output_type : DECODER
1670 # ISA name declaration looks like "namespace <foo>;"
1671 def p_name_decl(self
, t
):
1672 'name_decl : NAMESPACE ID SEMI'
1673 assert(self
.isa_name
== None and "Only 1 namespace decl permitted")
1674 self
.isa_name
= t
[2]
1675 self
.namespace
= t
[2] + 'Inst'
1677 # Output blocks 'output <foo> {{...}}' (C++ code blocks) are copied
1678 # directly to the appropriate output section.
1680 # Massage output block by substituting in template definitions and
1681 # bit operators. We handle '%'s embedded in the string that don't
1682 # indicate template substitutions (or CPU-specific symbols, which
1683 # get handled in GenCode) by doubling them first so that the
1684 # format operation will reduce them back to single '%'s.
1685 def process_output(self
, s
):
1686 s
= self
.protectNonSubstPercents(s
)
1687 # protects cpu-specific symbols too
1688 s
= self
.protectCpuSymbols(s
)
1689 return substBitOps(s
% self
.templateMap
)
1691 def p_output(self
, t
):
1692 'output : OUTPUT output_type CODELIT SEMI'
1693 kwargs
= { t
[2]+'_output' : self
.process_output(t
[3]) }
1694 GenCode(self
, **kwargs
).emit()
1696 # global let blocks 'let {{...}}' (Python code blocks) are
1697 # executed directly when seen. Note that these execute in a
1698 # special variable context 'exportContext' to prevent the code
1699 # from polluting this script's namespace.
1700 def p_global_let(self
, t
):
1701 'global_let : LET CODELIT SEMI'
1703 return self
.split(sec
)
1704 self
.updateExportContext()
1705 self
.exportContext
["header_output"] = ''
1706 self
.exportContext
["decoder_output"] = ''
1707 self
.exportContext
["exec_output"] = ''
1708 self
.exportContext
["decode_block"] = ''
1709 self
.exportContext
["split"] = _split
1713 globals()[sec + '_output'] += func(sec)
1718 # This tricky setup (immediately above) allows us to just write
1719 # (e.g.) "split('exec')" in the Python code and the split #ifdef's
1720 # will automatically be added to the exec_output variable. The inner
1721 # Python execution environment doesn't know about the split points,
1722 # so we carefully inject and wrap a closure that can retrieve the
1723 # next split's #define from the parser and add it to the current
1724 # emission-in-progress.
1726 exec split_setup
+fixPythonIndentation(t
[2]) in self
.exportContext
1727 except Exception, exc
:
1730 error(t
, 'error: %s in global let block "%s".' % (exc
, t
[2]))
1732 header_output
=self
.exportContext
["header_output"],
1733 decoder_output
=self
.exportContext
["decoder_output"],
1734 exec_output
=self
.exportContext
["exec_output"],
1735 decode_block
=self
.exportContext
["decode_block"]).emit()
1737 # Define the mapping from operand type extensions to C++ types and
1738 # bit widths (stored in operandTypeMap).
1739 def p_def_operand_types(self
, t
):
1740 'def_operand_types : DEF OPERAND_TYPES CODELIT SEMI'
1742 self
.operandTypeMap
= eval('{' + t
[3] + '}')
1743 except Exception, exc
:
1747 'error: %s in def operand_types block "%s".' % (exc
, t
[3]))
1749 # Define the mapping from operand names to operand classes and
1750 # other traits. Stored in operandNameMap.
1751 def p_def_operands(self
, t
):
1752 'def_operands : DEF OPERANDS CODELIT SEMI'
1753 if not hasattr(self
, 'operandTypeMap'):
1754 error(t
, 'error: operand types must be defined before operands')
1756 user_dict
= eval('{' + t
[3] + '}', self
.exportContext
)
1757 except Exception, exc
:
1760 error(t
, 'error: %s in def operands block "%s".' % (exc
, t
[3]))
1761 self
.buildOperandNameMap(user_dict
, t
.lexer
.lineno
)
1763 # A bitfield definition looks like:
1764 # 'def [signed] bitfield <ID> [<first>:<last>]'
1765 # This generates a preprocessor macro in the output file.
1766 def p_def_bitfield_0(self
, t
):
1767 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT COLON INTLIT GREATER SEMI'
1768 expr
= 'bits(machInst, %2d, %2d)' % (t
[6], t
[8])
1769 if (t
[2] == 'signed'):
1770 expr
= 'sext<%d>(%s)' % (t
[6] - t
[8] + 1, expr
)
1771 hash_define
= '#undef %s\n#define %s\t%s\n' % (t
[4], t
[4], expr
)
1772 GenCode(self
, header_output
=hash_define
).emit()
1774 # alternate form for single bit: 'def [signed] bitfield <ID> [<bit>]'
1775 def p_def_bitfield_1(self
, t
):
1776 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT GREATER SEMI'
1777 expr
= 'bits(machInst, %2d, %2d)' % (t
[6], t
[6])
1778 if (t
[2] == 'signed'):
1779 expr
= 'sext<%d>(%s)' % (1, expr
)
1780 hash_define
= '#undef %s\n#define %s\t%s\n' % (t
[4], t
[4], expr
)
1781 GenCode(self
, header_output
=hash_define
).emit()
1783 # alternate form for structure member: 'def bitfield <ID> <ID>'
1784 def p_def_bitfield_struct(self
, t
):
1785 'def_bitfield_struct : DEF opt_signed BITFIELD ID id_with_dot SEMI'
1787 error(t
, 'error: structure bitfields are always unsigned.')
1788 expr
= 'machInst.%s' % t
[5]
1789 hash_define
= '#undef %s\n#define %s\t%s\n' % (t
[4], t
[4], expr
)
1790 GenCode(self
, header_output
=hash_define
).emit()
1792 def p_id_with_dot_0(self
, t
):
1796 def p_id_with_dot_1(self
, t
):
1797 'id_with_dot : ID DOT id_with_dot'
1798 t
[0] = t
[1] + t
[2] + t
[3]
1800 def p_opt_signed_0(self
, t
):
1801 'opt_signed : SIGNED'
1804 def p_opt_signed_1(self
, t
):
1805 'opt_signed : empty'
1808 def p_def_template(self
, t
):
1809 'def_template : DEF TEMPLATE ID CODELIT SEMI'
1810 if t
[3] in self
.templateMap
:
1811 print "warning: template %s already defined" % t
[3]
1812 self
.templateMap
[t
[3]] = Template(self
, t
[4])
1814 # An instruction format definition looks like
1815 # "def format <fmt>(<params>) {{...}};"
1816 def p_def_format(self
, t
):
1817 'def_format : DEF FORMAT ID LPAREN param_list RPAREN CODELIT SEMI'
1818 (id, params
, code
) = (t
[3], t
[5], t
[7])
1819 self
.defFormat(id, params
, code
, t
.lexer
.lineno
)
1821 # The formal parameter list for an instruction format is a
1822 # possibly empty list of comma-separated parameters. Positional
1823 # (standard, non-keyword) parameters must come first, followed by
1824 # keyword parameters, followed by a '*foo' parameter that gets
1825 # excess positional arguments (as in Python). Each of these three
1826 # parameter categories is optional.
1828 # Note that we do not support the '**foo' parameter for collecting
1829 # otherwise undefined keyword args. Otherwise the parameter list
1830 # is (I believe) identical to what is supported in Python.
1832 # The param list generates a tuple, where the first element is a
1833 # list of the positional params and the second element is a dict
1834 # containing the keyword params.
1835 def p_param_list_0(self
, t
):
1836 'param_list : positional_param_list COMMA nonpositional_param_list'
1839 def p_param_list_1(self
, t
):
1840 '''param_list : positional_param_list
1841 | nonpositional_param_list'''
1844 def p_positional_param_list_0(self
, t
):
1845 'positional_param_list : empty'
1848 def p_positional_param_list_1(self
, t
):
1849 'positional_param_list : ID'
1852 def p_positional_param_list_2(self
, t
):
1853 'positional_param_list : positional_param_list COMMA ID'
1854 t
[0] = t
[1] + [t
[3]]
1856 def p_nonpositional_param_list_0(self
, t
):
1857 'nonpositional_param_list : keyword_param_list COMMA excess_args_param'
1860 def p_nonpositional_param_list_1(self
, t
):
1861 '''nonpositional_param_list : keyword_param_list
1862 | excess_args_param'''
1865 def p_keyword_param_list_0(self
, t
):
1866 'keyword_param_list : keyword_param'
1869 def p_keyword_param_list_1(self
, t
):
1870 'keyword_param_list : keyword_param_list COMMA keyword_param'
1871 t
[0] = t
[1] + [t
[3]]
1873 def p_keyword_param(self
, t
):
1874 'keyword_param : ID EQUALS expr'
1875 t
[0] = t
[1] + ' = ' + t
[3].__repr
__()
1877 def p_excess_args_param(self
, t
):
1878 'excess_args_param : ASTERISK ID'
1879 # Just concatenate them: '*ID'. Wrap in list to be consistent
1880 # with positional_param_list and keyword_param_list.
1881 t
[0] = [t
[1] + t
[2]]
1883 # End of format definition-related rules.
1887 # A decode block looks like:
1888 # decode <field1> [, <field2>]* [default <inst>] { ... }
1890 def p_top_level_decode_block(self
, t
):
1891 'top_level_decode_block : decode_block'
1893 codeObj
.wrap_decode_block('''
1895 %(isa_name)s::Decoder::decodeInst(%(isa_name)s::ExtMachInst machInst)
1897 using namespace %(namespace)s;
1902 def p_decode_block(self
, t
):
1903 'decode_block : DECODE ID opt_default LBRACE decode_stmt_list RBRACE'
1904 default_defaults
= self
.defaultStack
.pop()
1906 # use the "default defaults" only if there was no explicit
1907 # default statement in decode_stmt_list
1908 if not codeObj
.has_decode_default
:
1909 codeObj
+= default_defaults
1910 codeObj
.wrap_decode_block('switch (%s) {\n' % t
[2], '}\n')
1913 # The opt_default statement serves only to push the "default
1914 # defaults" onto defaultStack. This value will be used by nested
1915 # decode blocks, and used and popped off when the current
1916 # decode_block is processed (in p_decode_block() above).
1917 def p_opt_default_0(self
, t
):
1918 'opt_default : empty'
1919 # no default specified: reuse the one currently at the top of
1921 self
.defaultStack
.push(self
.defaultStack
.top())
1922 # no meaningful value returned
1925 def p_opt_default_1(self
, t
):
1926 'opt_default : DEFAULT inst'
1927 # push the new default
1929 codeObj
.wrap_decode_block('\ndefault:\n', 'break;\n')
1930 self
.defaultStack
.push(codeObj
)
1931 # no meaningful value returned
1934 def p_decode_stmt_list_0(self
, t
):
1935 'decode_stmt_list : decode_stmt'
1938 def p_decode_stmt_list_1(self
, t
):
1939 'decode_stmt_list : decode_stmt decode_stmt_list'
1940 if (t
[1].has_decode_default
and t
[2].has_decode_default
):
1941 error(t
, 'Two default cases in decode block')
1945 # Decode statement rules
1947 # There are four types of statements allowed in a decode block:
1948 # 1. Format blocks 'format <foo> { ... }'
1949 # 2. Nested decode blocks
1950 # 3. Instruction definitions.
1951 # 4. C preprocessor directives.
1954 # Preprocessor directives found in a decode statement list are
1955 # passed through to the output, replicated to all of the output
1956 # code streams. This works well for ifdefs, so we can ifdef out
1957 # both the declarations and the decode cases generated by an
1958 # instruction definition. Handling them as part of the grammar
1959 # makes it easy to keep them in the right place with respect to
1960 # the code generated by the other statements.
1961 def p_decode_stmt_cpp(self
, t
):
1962 'decode_stmt : CPPDIRECTIVE'
1963 t
[0] = GenCode(self
, t
[1], t
[1], t
[1], t
[1])
1965 # A format block 'format <foo> { ... }' sets the default
1966 # instruction format used to handle instruction definitions inside
1967 # the block. This format can be overridden by using an explicit
1968 # format on the instruction definition or with a nested format
1970 def p_decode_stmt_format(self
, t
):
1971 'decode_stmt : FORMAT push_format_id LBRACE decode_stmt_list RBRACE'
1972 # The format will be pushed on the stack when 'push_format_id'
1973 # is processed (see below). Once the parser has recognized
1974 # the full production (though the right brace), we're done
1975 # with the format, so now we can pop it.
1976 self
.formatStack
.pop()
1979 # This rule exists so we can set the current format (& push the
1980 # stack) when we recognize the format name part of the format
1982 def p_push_format_id(self
, t
):
1983 'push_format_id : ID'
1985 self
.formatStack
.push(self
.formatMap
[t
[1]])
1986 t
[0] = ('', '// format %s' % t
[1])
1988 error(t
, 'instruction format "%s" not defined.' % t
[1])
1990 # Nested decode block: if the value of the current field matches
1991 # the specified constant(s), do a nested decode on some other field.
1992 def p_decode_stmt_decode(self
, t
):
1993 'decode_stmt : case_list COLON decode_block'
1996 # just wrap the decoding code from the block as a case in the
1997 # outer switch statement.
1998 codeObj
.wrap_decode_block('\n%s\n' % ''.join(case_list
))
1999 codeObj
.has_decode_default
= (case_list
== ['default:'])
2002 # Instruction definition (finally!).
2003 def p_decode_stmt_inst(self
, t
):
2004 'decode_stmt : case_list COLON inst SEMI'
2007 codeObj
.wrap_decode_block('\n%s' % ''.join(case_list
), 'break;\n')
2008 codeObj
.has_decode_default
= (case_list
== ['default:'])
2011 # The constant list for a decode case label must be non-empty, and must
2012 # either be the keyword 'default', or made up of one or more
2013 # comma-separated integer literals or strings which evaluate to
2014 # constants when compiled as C++.
2015 def p_case_list_0(self
, t
):
2016 'case_list : DEFAULT'
2019 def prep_int_lit_case_label(self
, lit
):
2021 return 'case ULL(%#x): ' % lit
2023 return 'case %#x: ' % lit
2025 def prep_str_lit_case_label(self
, lit
):
2026 return 'case %s: ' % lit
2028 def p_case_list_1(self
, t
):
2029 'case_list : INTLIT'
2030 t
[0] = [self
.prep_int_lit_case_label(t
[1])]
2032 def p_case_list_2(self
, t
):
2033 'case_list : STRLIT'
2034 t
[0] = [self
.prep_str_lit_case_label(t
[1])]
2036 def p_case_list_3(self
, t
):
2037 'case_list : case_list COMMA INTLIT'
2039 t
[0].append(self
.prep_int_lit_case_label(t
[3]))
2041 def p_case_list_4(self
, t
):
2042 'case_list : case_list COMMA STRLIT'
2044 t
[0].append(self
.prep_str_lit_case_label(t
[3]))
2046 # Define an instruction using the current instruction format
2047 # (specified by an enclosing format block).
2048 # "<mnemonic>(<args>)"
2049 def p_inst_0(self
, t
):
2050 'inst : ID LPAREN arg_list RPAREN'
2051 # Pass the ID and arg list to the current format class to deal with.
2052 currentFormat
= self
.formatStack
.top()
2053 codeObj
= currentFormat
.defineInst(self
, t
[1], t
[3], t
.lexer
.lineno
)
2054 args
= ','.join(map(str, t
[3]))
2055 args
= re
.sub('(?m)^', '//', args
)
2056 args
= re
.sub('^//', '', args
)
2057 comment
= '\n// %s::%s(%s)\n' % (currentFormat
.id, t
[1], args
)
2058 codeObj
.prepend_all(comment
)
2061 # Define an instruction using an explicitly specified format:
2062 # "<fmt>::<mnemonic>(<args>)"
2063 def p_inst_1(self
, t
):
2064 'inst : ID DBLCOLON ID LPAREN arg_list RPAREN'
2066 format
= self
.formatMap
[t
[1]]
2068 error(t
, 'instruction format "%s" not defined.' % t
[1])
2070 codeObj
= format
.defineInst(self
, t
[3], t
[5], t
.lexer
.lineno
)
2071 comment
= '\n// %s::%s(%s)\n' % (t
[1], t
[3], t
[5])
2072 codeObj
.prepend_all(comment
)
2075 # The arg list generates a tuple, where the first element is a
2076 # list of the positional args and the second element is a dict
2077 # containing the keyword args.
2078 def p_arg_list_0(self
, t
):
2079 'arg_list : positional_arg_list COMMA keyword_arg_list'
2080 t
[0] = ( t
[1], t
[3] )
2082 def p_arg_list_1(self
, t
):
2083 'arg_list : positional_arg_list'
2086 def p_arg_list_2(self
, t
):
2087 'arg_list : keyword_arg_list'
2090 def p_positional_arg_list_0(self
, t
):
2091 'positional_arg_list : empty'
2094 def p_positional_arg_list_1(self
, t
):
2095 'positional_arg_list : expr'
2098 def p_positional_arg_list_2(self
, t
):
2099 'positional_arg_list : positional_arg_list COMMA expr'
2100 t
[0] = t
[1] + [t
[3]]
2102 def p_keyword_arg_list_0(self
, t
):
2103 'keyword_arg_list : keyword_arg'
2106 def p_keyword_arg_list_1(self
, t
):
2107 'keyword_arg_list : keyword_arg_list COMMA keyword_arg'
2111 def p_keyword_arg(self
, t
):
2112 'keyword_arg : ID EQUALS expr'
2113 t
[0] = { t
[1] : t
[3] }
2116 # Basic expressions. These constitute the argument values of
2117 # "function calls" (i.e. instruction definitions in the decode
2118 # block) and default values for formal parameters of format
2121 # Right now, these are either strings, integers, or (recursively)
2122 # lists of exprs (using Python square-bracket list syntax). Note
2123 # that bare identifiers are trated as string constants here (since
2124 # there isn't really a variable namespace to refer to).
2126 def p_expr_0(self
, t
):
2133 def p_expr_1(self
, t
):
2134 '''expr : LBRACKET list_expr RBRACKET'''
2137 def p_list_expr_0(self
, t
):
2141 def p_list_expr_1(self
, t
):
2142 'list_expr : list_expr COMMA expr'
2143 t
[0] = t
[1] + [t
[3]]
2145 def p_list_expr_2(self
, t
):
2150 # Empty production... use in other rules for readability.
2152 def p_empty(self
, t
):
2156 # Parse error handler. Note that the argument here is the
2157 # offending *token*, not a grammar symbol (hence the need to use
2159 def p_error(self
, t
):
2161 error(t
, "syntax error at '%s'" % t
.value
)
2163 error("unknown syntax error")
2165 # END OF GRAMMAR RULES
2167 def updateExportContext(self
):
2169 # create a continuation that allows us to grab the current parser
2170 def wrapInstObjParams(*args
):
2171 return InstObjParams(self
, *args
)
2172 self
.exportContext
['InstObjParams'] = wrapInstObjParams
2173 self
.exportContext
.update(self
.templateMap
)
2175 def defFormat(self
, id, params
, code
, lineno
):
2176 '''Define a new format'''
2178 # make sure we haven't already defined this one
2179 if id in self
.formatMap
:
2180 error(lineno
, 'format %s redefined.' % id)
2182 # create new object and store in global map
2183 self
.formatMap
[id] = Format(id, params
, code
)
2185 def expandCpuSymbolsToDict(self
, template
):
2186 '''Expand template with CPU-specific references into a
2187 dictionary with an entry for each CPU model name. The entry
2188 key is the model name and the corresponding value is the
2189 template with the CPU-specific refs substituted for that
2192 # Protect '%'s that don't go with CPU-specific terms
2193 t
= re
.sub(r
'%(?!\(CPU_)', '%%', template
)
2195 for cpu
in self
.cpuModels
:
2196 result
[cpu
.name
] = t
% cpu
.strings
2199 def expandCpuSymbolsToString(self
, template
):
2200 '''*If* the template has CPU-specific references, return a
2201 single string containing a copy of the template for each CPU
2202 model with the corresponding values substituted in. If the
2203 template has no CPU-specific references, it is returned
2206 if template
.find('%(CPU_') != -1:
2207 return reduce(lambda x
,y
: x
+y
,
2208 self
.expandCpuSymbolsToDict(template
).values())
2212 def protectCpuSymbols(self
, template
):
2213 '''Protect CPU-specific references by doubling the
2214 corresponding '%'s (in preparation for substituting a different
2215 set of references into the template).'''
2217 return re
.sub(r
'%(?=\(CPU_)', '%%', template
)
2219 def protectNonSubstPercents(self
, s
):
2220 '''Protect any non-dict-substitution '%'s in a format string
2221 (i.e. those not followed by '(')'''
2223 return re
.sub(r
'%(?!\()', '%%', s
)
2225 def buildOperandNameMap(self
, user_dict
, lineno
):
2227 for op_name
, val
in user_dict
.iteritems():
2229 # Check if extra attributes have been specified.
2231 error(lineno
, 'error: too many attributes for operand "%s"' %
2234 # Pad val with None in case optional args are missing
2235 val
+= (None, None, None, None)
2236 base_cls_name
, dflt_ext
, reg_spec
, flags
, sort_pri
, \
2237 read_code
, write_code
, read_predicate
, write_predicate
= val
[:9]
2239 # Canonical flag structure is a triple of lists, where each list
2240 # indicates the set of flags implied by this operand always, when
2241 # used as a source, and when used as a dest, respectively.
2242 # For simplicity this can be initialized using a variety of fairly
2243 # obvious shortcuts; we convert these to canonical form here.
2245 # no flags specified (e.g., 'None')
2246 flags
= ( [], [], [] )
2247 elif isinstance(flags
, str):
2248 # a single flag: assumed to be unconditional
2249 flags
= ( [ flags
], [], [] )
2250 elif isinstance(flags
, list):
2251 # a list of flags: also assumed to be unconditional
2252 flags
= ( flags
, [], [] )
2253 elif isinstance(flags
, tuple):
2254 # it's a tuple: it should be a triple,
2255 # but each item could be a single string or a list
2256 (uncond_flags
, src_flags
, dest_flags
) = flags
2257 flags
= (makeList(uncond_flags
),
2258 makeList(src_flags
), makeList(dest_flags
))
2260 # Accumulate attributes of new operand class in tmp_dict
2262 attrList
= ['reg_spec', 'flags', 'sort_pri',
2263 'read_code', 'write_code',
2264 'read_predicate', 'write_predicate']
2266 dflt_ctype
= self
.operandTypeMap
[dflt_ext
]
2267 attrList
.extend(['dflt_ctype', 'dflt_ext'])
2268 for attr
in attrList
:
2269 tmp_dict
[attr
] = eval(attr
)
2270 tmp_dict
['base_name'] = op_name
2272 # New class name will be e.g. "IntReg_Ra"
2273 cls_name
= base_cls_name
+ '_' + op_name
2274 # Evaluate string arg to get class object. Note that the
2275 # actual base class for "IntReg" is "IntRegOperand", i.e. we
2276 # have to append "Operand".
2278 base_cls
= eval(base_cls_name
+ 'Operand')
2281 'error: unknown operand base class "%s"' % base_cls_name
)
2282 # The following statement creates a new class called
2283 # <cls_name> as a subclass of <base_cls> with the attributes
2284 # in tmp_dict, just as if we evaluated a class declaration.
2285 operand_name
[op_name
] = type(cls_name
, (base_cls
,), tmp_dict
)
2287 self
.operandNameMap
= operand_name
2289 # Define operand variables.
2290 operands
= user_dict
.keys()
2291 extensions
= self
.operandTypeMap
.keys()
2293 operandsREString
= r
'''
2294 (?<!\w) # neg. lookbehind assertion: prevent partial matches
2295 ((%s)(?:_(%s))?) # match: operand with optional '_' then suffix
2296 (?!\w) # neg. lookahead assertion: prevent partial matches
2297 ''' % (string
.join(operands
, '|'), string
.join(extensions
, '|'))
2299 self
.operandsRE
= re
.compile(operandsREString
, re
.MULTILINE|re
.VERBOSE
)
2301 # Same as operandsREString, but extension is mandatory, and only two
2302 # groups are returned (base and ext, not full name as above).
2303 # Used for subtituting '_' for '.' to make C++ identifiers.
2304 operandsWithExtREString
= r
'(?<!\w)(%s)_(%s)(?!\w)' \
2305 % (string
.join(operands
, '|'), string
.join(extensions
, '|'))
2307 self
.operandsWithExtRE
= \
2308 re
.compile(operandsWithExtREString
, re
.MULTILINE
)
2310 def substMungedOpNames(self
, code
):
2311 '''Munge operand names in code string to make legal C++
2312 variable names. This means getting rid of the type extension
2313 if any. Will match base_name attribute of Operand object.)'''
2314 return self
.operandsWithExtRE
.sub(r
'\1', code
)
2316 def mungeSnippet(self
, s
):
2317 '''Fix up code snippets for final substitution in templates.'''
2318 if isinstance(s
, str):
2319 return self
.substMungedOpNames(substBitOps(s
))
2323 def open(self
, name
, bare
=False):
2324 '''Open the output file for writing and include scary warning.'''
2325 filename
= os
.path
.join(self
.output_dir
, name
)
2326 f
= open(filename
, 'w')
2329 f
.write(ISAParser
.scaremonger_template
% self
)
2332 def update(self
, file, contents
):
2333 '''Update the output file only. Scons should handle the case when
2334 the new contents are unchanged using its built-in hash feature.'''
2339 # This regular expression matches '##include' directives
2340 includeRE
= re
.compile(r
'^\s*##include\s+"(?P<filename>[^"]*)".*$',
2343 def replace_include(self
, matchobj
, dirname
):
2344 """Function to replace a matched '##include' directive with the
2345 contents of the specified file (with nested ##includes
2346 replaced recursively). 'matchobj' is an re match object
2347 (from a match of includeRE) and 'dirname' is the directory
2348 relative to which the file path should be resolved."""
2350 fname
= matchobj
.group('filename')
2351 full_fname
= os
.path
.normpath(os
.path
.join(dirname
, fname
))
2352 contents
= '##newfile "%s"\n%s\n##endfile\n' % \
2353 (full_fname
, self
.read_and_flatten(full_fname
))
2356 def read_and_flatten(self
, filename
):
2357 """Read a file and recursively flatten nested '##include' files."""
2359 current_dir
= os
.path
.dirname(filename
)
2361 contents
= open(filename
).read()
2363 error('Error including file "%s"' % filename
)
2365 self
.fileNameStack
.push((filename
, 0))
2367 # Find any includes and include them
2368 def replace(matchobj
):
2369 return self
.replace_include(matchobj
, current_dir
)
2370 contents
= self
.includeRE
.sub(replace
, contents
)
2372 self
.fileNameStack
.pop()
2375 AlreadyGenerated
= {}
2377 def _parse_isa_desc(self
, isa_desc_file
):
2378 '''Read in and parse the ISA description.'''
2380 # The build system can end up running the ISA parser twice: once to
2381 # finalize the build dependencies, and then to actually generate
2382 # the files it expects (in src/arch/$ARCH/generated). This code
2383 # doesn't do anything different either time, however; the SCons
2384 # invocations just expect different things. Since this code runs
2385 # within SCons, we can just remember that we've already run and
2386 # not perform a completely unnecessary run, since the ISA parser's
2387 # effect is idempotent.
2388 if isa_desc_file
in ISAParser
.AlreadyGenerated
:
2391 # grab the last three path components of isa_desc_file
2392 self
.filename
= '/'.join(isa_desc_file
.split('/')[-3:])
2394 # Read file and (recursively) all included files into a string.
2395 # PLY requires that the input be in a single string so we have to
2397 isa_desc
= self
.read_and_flatten(isa_desc_file
)
2399 # Initialize filename stack with outer file.
2400 self
.fileNameStack
.push((isa_desc_file
, 0))
2403 self
.parse_string(isa_desc
)
2405 ISAParser
.AlreadyGenerated
[isa_desc_file
] = None
2407 def parse_isa_desc(self
, *args
, **kwargs
):
2409 self
._parse
_isa
_desc
(*args
, **kwargs
)
2410 except ISAParserError
, e
:
2411 e
.exit(self
.fileNameStack
)
2413 # Called as script: get args from command line.
2414 # Args are: <isa desc file> <output dir>
2415 if __name__
== '__main__':
2416 ISAParser(sys
.argv
[2]).parse_isa_desc(sys
.argv
[1])