[gold patch] Add inlined fast path for LEB128 routines

[Cary Coutant <ccoutant at google dot com>]

This patch adds an inlined fast path for the common case of a
single-byte LEB128 number. This provides significant speed-ups for
DWARF processing (mostly for the as-yet-unsubmitted changes to
generate the .gdb-index section). In my testing,
read_unsigned_LEB_128() accounted for more than 5% of the link time
when generating a gdb index; with this change, the LEB128 routines are
negligible, resulting in a 5% improvement. This should also improve
line number lookup and ODR detection.

I've also added new unit tests for the LEB128 readers.

Tested on x86_64. OK?

-cary


2012-01-24  Cary Coutant  <ccoutant@google.com>

	* int_encoding.cc (read_unsigned_LEB_128): Replaced with inline
	definition and ...
	(read_unsigned_LEB_128_x): ... this new function.
	(read_signed_LEB_128): Replaced with inline definition and ...
	(read_signed_LEB_128_x): ... this new function.
	* int_encoding.h  (read_unsigned_LEB_128_x): New function.
	(read_unsigned_LEB_128): Add inline definition.
	(read_signed_LEB_128_x): New function.
	(read_signed_LEB_128): Add inline definition.
	* testsuite/Makefile.am (leb128_unittest): New unit test.
	* testsuite/Makefile.in: Regenerate.
	* testsuite/leb128_unittest.cc: New unit test.

Attachment: gold-fast-leb128-patch.txt
Description: Text document