On Fri, 19 Mar 2010, John Baldwin wrote:
On Friday 19 March 2010 12:13:00 pm Ivan Voras wrote:
SSE in the userland you mean? Regardless, I don't think there is now
reason for compiling everything as for i386. E.g. why not add at least
-mtune=generic or even also -march=i686 to default gcc options?
http://gcc.gnu.org/onlinedocs/gcc/i386-and-x86_002d64-Options.html
I actually suggested using -mtune=i686 several years ago for i386. Using
-mtune=generic probably would be a good thing to use when CPUTYPE is not
specified for i386 and amd64 now.
Er, isn't -mtune=generic the default in gcc-4.2? I once fought with
gcc-4.2 to produce kernels with similar size and compile time to ones
produced by gcc-3 (impossible for the runtime since gcc got another
40% slower and kernel source size and complications expanded another
40% even when the object size didn't bloat much). It took -i386
and a couple of other flags like -fno-inline-functions-called-once
(the latter is also needed to unbreak debugging with ddb and profiling).
The flags had very little difference on runtime efficiency, partly because
I ran the kernels mainly on Athlons which run old i386-optimized code
amazingly well but gain on benefits from i686 optimizations. Optimizing
Athlon kernels for Athlons also makes little difference but I usually do
it.
Bruce
_______________________________________________
svn-src-head@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/svn-src-head
To unsubscribe, send any mail to "svn-src-head-unsubscr...@freebsd.org"
