scripts/bloat-o-meter: compile .NUMBER regex

Every often used regex is better be compiled in Python.

Speedup is about ~9.8% (whee!)

    $ perf stat -r 16 taskset -c 15 ./scripts/bloat-o-meter ../vmlinux-000 ../obj/vmlinux >/dev/null
    7.091202853 seconds time elapsed                         ( +-  0.15% )

    +re.compile
    6.397564973 seconds time elapsed                         ( +-  0.34% )

Link: http://lkml.kernel.org/r/20161119004417.GB1200@avx2
Signed-off-by: Alexey Dobriyan <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

Author: Alexey Dobriyan

Diff for: scripts/bloat-o-meter

@@ -16,6 +16,8 @@ if len(sys.argv) != 3:
     sys.stderr.write("usage: %s file1 file2\n" % sys.argv[0])
     sys.exit(-1)
 
+re_NUMBER = re.compile(r'\.[0-9]+')
+
 def getsizes(file):
     sym = {}
     with os.popen("nm --size-sort " + file) as f:
@@ -28,7 +30,7 @@ def getsizes(file):
                 if name.startswith("compat_SyS_"): continue
                 if name == "linux_banner": continue
                 # statics and some other optimizations adds random .NUMBER
-                name = re.sub(r'\.[0-9]+', '', name)
+                name = re_NUMBER.sub('', name)
                 sym[name] = sym.get(name, 0) + int(size, 16)
     return sym