11 Jul
2006
11 Jul
'06
1:27 a.m.
On Jul 10, 2006, at 4:23 AM, Donald Bruce Stewart wrote:
Ah! In this case, on the debian, the benchmark has been compiled _without_ -fexcess-precision, that's what's causing the big slow down. We had to turn it on, on the gp4, but it seems the flag wasn't propagated to the debian/sempron builds for some reason.
Looks like the ghc/mandelbrot benchmarks just needs to be rerun with -fexcess-precision in this case.
Ah! Thanks for noticing this -- I did not catch it. I've rerun, and the new value is more like 0.428 secs for 600. Thanks, -Brent