cvs commit: src/test/cpuperf Makefile cpu_add.S cpu_call.S cpu_cmpadd.S cpu_cmpexg.S cpu_lcmpexg.S cputest.c src/test/sysperf blib.c blib.h
Matthew Dillon
dillon at apollo.backplane.com
Tue Feb 10 10:21:22 PST 2004
:Some ssh sessions, ica client, that's about it. Within X.
:
:> The timing
:> tests are very, very sensitive. The second addl
:> taking 77nS is the locked add instruction? Obviously not
:> the 3 x addl $1,%eax, not at 77ns :-). I've gotten negative
:> numbers from the first addl test, but I've never gotten
:> negative numbers from the call/ret test.
:
:ls /tmp/cpu*, that order I ran them.
:
:And yea, I've done it about 6-7 times and 5/7th are negative.
:
:--
:Jeroen Ruigrok van der Werven <asmodai(at)wxs.nl> / asmodai / kita no mono
If you see any mistakes I'm all ears. I aligned the assembly code but
maybe the main loop in C is broken and needs to be aligned as well.
Try using -O or gcc3. We could be seeing cache line alignment effects.
-Matt
Matthew Dillon
<dillon at xxxxxxxxxxxxx>
More information about the Commits
mailing list