ponderous 'make world' times post GCC 3.3...
brandt at fokus.fraunhofer.de
Wed Aug 13 08:23:59 PDT 2003
T>On Mon, Aug 11, 2003 at 11:51:00AM -0400, Garance A Drosihn wrote:
T>> At 1:40 AM -0700 8/9/03, Kris Kennaway wrote:
T>> >On Sat, Aug 09, 2003, Garance A Drosihn wrote:
T>> >> So, apparently the problem is something a bit more subtle than
T>> >> just gcc 3.3 being slower to compile than gcc 3.2. Apparently
T>> >> the August 9th system is a lot slower at *running* than the
T>> >> early system. Do we have some other benchmarks we could run?
T>> >This suggests that something might have been pessimized with
T>> >the gcc 3.3 code generation on sparc.
The problem seems to be in the kernel, not the compiler. I have rebuilt a
gcc 3.2.2. The actual kernel (from yesterday) is slightly faster when
built with 3.3.1 as opposed to 3.2.2 (in the order of 3-4%), but VERY
slow. A kernel from June 1st built with 3.2.2 has its 'normal' speed. I'm
trying now I binary search to find the victim.
One difference between the kernels is, that a /usr/bin/time -l make
something reports twice as much involuntary context switches with the new
kernel. I don't know, however, exactly what this means.
brandt at fokus.fraunhofer.de, harti at freebsd.org
More information about the freebsd-sparc64