Steve Bergman wrote:
I have a few questions. How much work has been done to benchmark the
advantages/disadvantages of using -Os for various FC packages on
X86_64? I know that on x86 -Os is used fort he kernel. On X86_64, it
looks to me like -Os is not used. In fact, it seems that I read
somewhere that one should not use -Os on X86_64, or shouldn't use it on
the kernel, or shouldn't use it in FC4, or something like that and I
can't seem to relocate that reference. Anyone know anything about that?
Also, how useful would it be if did some real world benchmarking on my
Athlon64 2800+ system running FC4/X86_64? Would I be wasting my time,
duplicating someone else's previous efforts? What packages might
benefit? Is it possible that compilintg the whole distro -Os might be a
win? (Intuitively, it seems that everything being a little smaller could
have a cummulative effect on the speed of the whole system, saving disk
reads, main memory, cache memory, etc.)
I'm no expert at this stuff, so I figured asking on the list before I do
much work would be a good idea. Searches of the archives don't seem to
turn up much except that "-Os can sometimes help".
So far, I've just played around a little bit with bzip2. -Os seems to
be a pretty good win for bzip2 compression, giving over about a 12%
Thank you for any input.
None of the stuff I know of hasn't looked at x86_64.
You might look at the Code-Size Benchmark Environment (CSiBE):
However, they do not seem to be running the tests for x86_64 at this time.
During the fall of 2004 a couple of students at North Carolina State
University did some experiments compiling mozilla and the related
libraries with -Os on x86 machines to see the effect. They found that
there was some reduction in size and startup time.