Re: Has anyone run fft2 successfully on linux?

[Simon L Peyton Jones]

Bonard

What this says is that GHC2.04 gives different output for fft2 than did some 
earlier version of GHC.  The numbers involved are rather small and I really
don't know which of the two is (closer to being) correct!

If anyone has a C version, or is better with floats than me, it would
be delightful to either declare GHC2.04 correct, or fix the bug if there is
one.

Help welcomed!

Simon

| I've made several attempts to run the nofib suite under the x86-linux ports of
| ghc (versions 0.29 and 2.04).  It has run well, except for some floating
| point code.  For example, the output of fft2:
| 
| expected stdout not matched by reality
| *** fft2.stdout Mon Jan  8 15:04:40 1996
| --- /tmp/runtest15948.1 Wed Jul  2 23:04:10 1997
| ***************
| *** 1,3 ****
| ! result1 = 2.6734264806727734e-12
| ! result2 = 2.6734264806727734e-12
| ! result3 = 4.8316280754079344e-8
| --- 1,3 ----
| ! result1 = 2.6519392517661656e-12
| ! result2 = 2.6519392517661656e-12
| ! result3 = 4.829263389183325e-8
| make: *** [runtest_fft2] Error 1
| 
| Any info would be most appreciated and a very warm thanks to the Glasgow team
| for such an excellent system!