luau/tools/stack-usage-reporter.py
Andy Friesen a5c6a38b10
Sync to upstream/release/559 (#804)
* Fix autocompletion of if-then-else expressions
* Fix a potential crash surrounding improper use of `%*` in a string
format specifier
* All Python scripts now invoke Python via `python3` rather than
`python`.
* Improved error handling for string interpolation with too many
arguments.

Co-authored-by: Arseny Kapoulkine <arseny.kapoulkine@gmail.com>
Co-authored-by: Vyacheslav Egorov <vegorov@roblox.com>
2023-01-13 14:10:01 -08:00

173 lines
6 KiB
Python

#!/usr/bin/python3
# This file is part of the Luau programming language and is licensed under MIT License; see LICENSE.txt for details
# The purpose of this script is to analyze disassembly generated by objdump or
# dumpbin to print (or to compare) the stack usage of functions/methods.
# This is a quickly written script, so it is quite possible it may not handle
# all code properly.
#
# The script expects the user to create a text assembly dump to be passed to
# the script.
#
# objdump Example
# objdump --demangle --disassemble objfile.o > objfile.s
#
# dumpbin Example
# dumpbin /disasm objfile.obj > objfile.s
#
# If the script is passed a single file, then all stack size information that
# is found it printed. If two files are passed, then the script compares the
# stack usage of the two files (useful for A/B comparisons).
# Currently more than two input files are not supported. (But adding support shouldn't
# be very difficult.)
#
# Note: The script only handles x64 disassembly. Supporting x86 is likely
# trivial, but ARM support could be difficult.
# Thus far the script has been tested with MSVC on Win64 and clang on OSX.
import argparse
import re
blank_re = re.compile('\s*')
class LineReader:
def __init__(self, lines):
self.lines = list(reversed(lines))
def get_line(self):
return self.lines.pop(-1)
def peek_line(self):
return self.lines[-1]
def consume_blank_lines(self):
while blank_re.fullmatch(self.peek_line()):
self.get_line()
def is_empty(self):
return len(self.lines) == 0
def parse_objdump_assembly(in_file):
results = {}
text_section_re = re.compile('Disassembly of section __TEXT,__text:\s*')
symbol_re = re.compile('[^<]*<(.*)>:\s*')
stack_alloc = re.compile('.*subq\s*\$(\d*), %rsp\s*')
lr = LineReader(in_file.readlines())
def find_stack_alloc_size():
while True:
if lr.is_empty():
return None
if blank_re.fullmatch(lr.peek_line()):
return None
line = lr.get_line()
mo = stack_alloc.fullmatch(line)
if mo:
lr.consume_blank_lines()
return int(mo.group(1))
# Find beginning of disassembly
while not text_section_re.fullmatch(lr.get_line()):
pass
# Scan for symbols
while not lr.is_empty():
lr.consume_blank_lines()
if lr.is_empty():
break
line = lr.get_line()
mo = symbol_re.fullmatch(line)
# Found a symbol
if mo:
symbol = mo.group(1)
stack_size = find_stack_alloc_size()
if stack_size != None:
results[symbol] = stack_size
return results
def parse_dumpbin_assembly(in_file):
results = {}
file_type_re = re.compile('File Type: COFF OBJECT\s*')
symbol_re = re.compile('[^(]*\((.*)\):\s*')
summary_re = re.compile('\s*Summary\s*')
stack_alloc = re.compile('.*sub\s*rsp,([A-Z0-9]*)h\s*')
lr = LineReader(in_file.readlines())
def find_stack_alloc_size():
while True:
if lr.is_empty():
return None
if blank_re.fullmatch(lr.peek_line()):
return None
line = lr.get_line()
mo = stack_alloc.fullmatch(line)
if mo:
lr.consume_blank_lines()
return int(mo.group(1), 16) # return value in decimal
# Find beginning of disassembly
while not file_type_re.fullmatch(lr.get_line()):
pass
# Scan for symbols
while not lr.is_empty():
lr.consume_blank_lines()
if lr.is_empty():
break
line = lr.get_line()
if summary_re.fullmatch(line):
break
mo = symbol_re.fullmatch(line)
# Found a symbol
if mo:
symbol = mo.group(1)
stack_size = find_stack_alloc_size()
if stack_size != None:
results[symbol] = stack_size
return results
def main():
parser = argparse.ArgumentParser(description='Tool used for reporting or comparing the stack usage of functions/methods')
parser.add_argument('--format', choices=['dumpbin', 'objdump'], required=True, help='Specifies the program used to generate the input files')
parser.add_argument('--input', action='append', required=True, help='Input assembly file. This option may be specified multiple times.')
parser.add_argument('--md-output', action='store_true', help='Show table output in markdown format')
parser.add_argument('--only-diffs', action='store_true', help='Only show stack info when it differs between the input files')
args = parser.parse_args()
parsers = {'dumpbin': parse_dumpbin_assembly, 'objdump' : parse_objdump_assembly}
parse_func = parsers[args.format]
input_results = []
for input_name in args.input:
with open(input_name) as in_file:
results = parse_func(in_file)
input_results.append(results)
if len(input_results) == 1:
# Print out the results sorted by size
size_sorted = sorted([(size, symbol) for symbol, size in results.items()], reverse=True)
print(input_name)
for size, symbol in size_sorted:
print(f'{size:10}\t{symbol}')
print()
elif len(input_results) == 2:
common_symbols = set(input_results[0].keys()).intersection(set(input_results[1].keys()))
print(f'Found {len(common_symbols)} common symbols')
stack_sizes = sorted([(input_results[0][sym], input_results[1][sym], sym) for sym in common_symbols], reverse=True)
if args.md_output:
print('Before | After | Symbol')
print('-- | -- | --')
for size0, size1, symbol in stack_sizes:
if args.only_diffs and size0 == size1:
continue
if args.md_output:
print(f'{size0} | {size1} | {symbol}')
else:
print(f'{size0:10}\t{size1:10}\t{symbol}')
else:
print("TODO support more than 2 inputs")
if __name__ == '__main__':
main()