Hi!
> > BTW, I bet your dynamic relocation tables are a bit larger too.
>
> Somewhat, but does it matter? They are not kept in memory anyways.
>
> It's all about how much data a ld.so relocation has to touch. But
> preloading will help out here, even though that isn't in wide spread
> use just yet.
>
> And I was talking about user stack usage, not the kernel kind
> :-)
>
> Andi, do something very simple like run -m32 vs -m64 microbenchmarks,
> I bet -m32 beats -m64 in all the lmbench lat_proc tests. On sparc64
> it's (on a 2-way SMP system):
>
> -m32 fork+exit: 360.8328 microseconds
> -m32 fork+execve: 1342.2213 microseconds
> -m32 fork+/bin/sh: 5497.0149 microseconds
>
> -m64 fork+exit: 553.9076 microseconds
> -m64 fork+execve: 1904.6315 microseconds
> -m64 fork+/bin/sh: 6268.6932 microseconds
>
> NOTE: make sure you change /bin/sh to be 32-bit/64-bit as
> appropriate in the tests above.
>
> So what is this on x86_64? :-) I think lat_proc is great becuase it
> shows pure libc overhead in continually relocating the exit()
> etc. symbols in the child for fork+exit, for example.
>
> The reason I'm making such a stink about this is that I don't want
> people believing that "the code generation improvements due to the
> extra x86_64 registers available nullifies the bloat cost from
> going to 64-bit"
Actually, it tends to nullify the bloat cost and then make it few
percent faster... For most of spec2000 modulo two or three cache-bound
tests that are 50% slower :-(.
Pavel
-- Worst form of spam? Adding advertisment signatures ala sourceforge.net. What goes next? Inserting advertisment *into* email? - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sat Dec 07 2002 - 22:00:24 EST