Actually converting the C/C++ code back into pyx would be difficult, and wouldn't likely yield very pythonic (let alone the original) source. However, you could probably extract a lot from the comments, e.g.
from collections import defaultdict
import re
import sys
all = defaultdict(dict)
for line in sys.stdin:
m = re.match(r'/\* "(.*\.pyx)":(\d+)', line)
if m:
file = m.group(1)
lineno = int(m.group(2)) - 2
elif '*/' in line:
lineno = None
elif lineno is not None:
all[file][lineno] = line[2:].rstrip()
lineno += 1
for file, lines in all.items():
print '=' * 20, file, '=' * 20
for lineno, line in sorted(lines.items()):
print lineno, ':', line