third-party/benchmark/tools/strip_asm.py

   1 #!/usr/bin/env python
   2
   3 """
   4 strip_asm.py - Cleanup ASM output for the specified file
   5 """
   6
   7 from argparse import ArgumentParser
   8 import sys
   9 import os
  10 import re
  11
  12
  13 def find_used_labels(asm):
  14     found = set()
  15     label_re = re.compile("\s*j[a-z]+\s+\.L([a-zA-Z0-9][a-zA-Z0-9_]*)")
  16     for l in asm.splitlines():
  17         m = label_re.match(l)
  18         if m:
  19             found.add(".L%s" % m.group(1))
  20     return found
  21
  22
  23 def normalize_labels(asm):
  24     decls = set()
  25     label_decl = re.compile("^[.]{0,1}L([a-zA-Z0-9][a-zA-Z0-9_]*)(?=:)")
  26     for l in asm.splitlines():
  27         m = label_decl.match(l)
  28         if m:
  29             decls.add(m.group(0))
  30     if len(decls) == 0:
  31         return asm
  32     needs_dot = next(iter(decls))[0] != "."
  33     if not needs_dot:
  34         return asm
  35     for ld in decls:
  36         asm = re.sub("(^|\s+)" + ld + "(?=:|\s)", "\\1." + ld, asm)
  37     return asm
  38
  39
  40 def transform_labels(asm):
  41     asm = normalize_labels(asm)
  42     used_decls = find_used_labels(asm)
  43     new_asm = ""
  44     label_decl = re.compile("^\.L([a-zA-Z0-9][a-zA-Z0-9_]*)(?=:)")
  45     for l in asm.splitlines():
  46         m = label_decl.match(l)
  47         if not m or m.group(0) in used_decls:
  48             new_asm += l
  49             new_asm += "\n"
  50     return new_asm
  51
  52
  53 def is_identifier(tk):
  54     if len(tk) == 0:
  55         return False
  56     first = tk[0]
  57     if not first.isalpha() and first != "_":
  58         return False
  59     for i in range(1, len(tk)):
  60         c = tk[i]
  61         if not c.isalnum() and c != "_":
  62             return False
  63     return True
  64
  65
  66 def process_identifiers(l):
  67     """
  68     process_identifiers - process all identifiers and modify them to have
  69     consistent names across all platforms; specifically across ELF and MachO.
  70     For example, MachO inserts an additional understore at the beginning of
  71     names. This function removes that.
  72     """
  73     parts = re.split(r"([a-zA-Z0-9_]+)", l)
  74     new_line = ""
  75     for tk in parts:
  76         if is_identifier(tk):
  77             if tk.startswith("__Z"):
  78                 tk = tk[1:]
  79             elif (
  80                 tk.startswith("_") and len(tk) > 1 and tk[1].isalpha() and tk[1] != "Z"
  81             ):
  82                 tk = tk[1:]
  83         new_line += tk
  84     return new_line
  85
  86
  87 def process_asm(asm):
  88     """
  89     Strip the ASM of unwanted directives and lines
  90     """
  91     new_contents = ""
  92     asm = transform_labels(asm)
  93
  94     # TODO: Add more things we want to remove
  95     discard_regexes = [
  96         re.compile("\s+\..*$"),  # directive
  97         re.compile("\s*#(NO_APP|APP)$"),  # inline ASM
  98         re.compile("\s*#.*$"),  # comment line
  99         re.compile("\s*\.globa?l\s*([.a-zA-Z_][a-zA-Z0-9$_.]*)"),  # global directive
 100         re.compile(
 101             "\s*\.(string|asciz|ascii|[1248]?byte|short|word|long|quad|value|zero)"
 102         ),
 103     ]
 104     keep_regexes = []
 105     fn_label_def = re.compile("^[a-zA-Z_][a-zA-Z0-9_.]*:")
 106     for l in asm.splitlines():
 107         # Remove Mach-O attribute
 108         l = l.replace("@GOTPCREL", "")
 109         add_line = True
 110         for reg in discard_regexes:
 111             if reg.match(l) is not None:
 112                 add_line = False
 113                 break
 114         for reg in keep_regexes:
 115             if reg.match(l) is not None:
 116                 add_line = True
 117                 break
 118         if add_line:
 119             if fn_label_def.match(l) and len(new_contents) != 0:
 120                 new_contents += "\n"
 121             l = process_identifiers(l)
 122             new_contents += l
 123             new_contents += "\n"
 124     return new_contents
 125
 126
 127 def main():
 128     parser = ArgumentParser(description="generate a stripped assembly file")
 129     parser.add_argument(
 130         "input", metavar="input", type=str, nargs=1, help="An input assembly file"
 131     )
 132     parser.add_argument(
 133         "out", metavar="output", type=str, nargs=1, help="The output file"
 134     )
 135     args, unknown_args = parser.parse_known_args()
 136     input = args.input[0]
 137     output = args.out[0]
 138     if not os.path.isfile(input):
 139         print(("ERROR: input file '%s' does not exist") % input)
 140         sys.exit(1)
 141     contents = None
 142     with open(input, "r") as f:
 143         contents = f.read()
 144     new_contents = process_asm(contents)
 145     with open(output, "w") as f:
 146         f.write(new_contents)
 147
 148
 149 if __name__ == "__main__":
 150     main()
 151
 152 # vim: tabstop=4 expandtab shiftwidth=4 softtabstop=4
 153 # kate: tab-width: 4; replace-tabs on; indent-width 4; tab-indents: off;
 154 # kate: indent-mode python; remove-trailing-spaces modified;