GitLab

These are the benchmarks uesd in our ISMM'08 paper on Parallel GC

In some GHC versions these operators have the wrong precedence

This allows you to use nofib in a tree built with validate

The current way didn't work when running nofib with ghci.

Some of these tests were using 'take' in their attempts to duplicate
some work and hence run longer, but since GHC just started deforesting
'take' this stopped working.  Now we use {-# NOILINE #-} as a
hopefully more robust way to subvert optimisations.

remove slow way; it generates too much output

add x86_64 output

update output

slow output

add slow output

mode=slow

That completes the imaginary and spectral suites.  Most of the
programs in here will now run in 1-2s with mode=slow (on a 3GHz P4).
I had to modify the programs in a few places, so there will be some
slight discrepancies in the size of Main.o due to these changes.

mode=slow

back out: attempted repetition wasn't really working

mode=slow

mode=slow

fish doesn't take any args; this looks like a cut-n-pasto

mode=slow