On Mon, Aug 18, 2014 at 3:17 PM, Patrick Alken <patrick.al...@colorado.edu> wrote:
> On 08/18/2014 03:11 PM, Hong Xu wrote: > >> Hi everyone, >> >> I have founded a very interesting problem: if I compile the gsl library >> with the CFLAGS of '-O2 -march=x86-64' or '-march=core2' (or one of >> sandybridge ivybridge haswell), all tests can pass. However, if I >> compile with CFLAGS='-O2 -march=[any of core2 sandybridge ivybridge >> haswell]', the tests won't pass. The file specfunc/testsuite.log is >> pasted in this link: http://pastebin.com/hcNhHz3b >> >> This would limit the usage of the advantage of newer instructions, and >> potentially there may be a bug. Would anyone look into this? >> >> Thanks! >> Hong >> >> >> >> Could this be a gcc problem? A quick google search shows some gcc > versions have bugs with these architectures > > I've tested it under gcc 4.8.2 on Ubuntu 14.04 and 4.9.1 on Arch Linux. Is 'core2' also buggy? I saw kernel compilation supports core2 but not further more. And, how can I show whether it's a problem of gcc or a problem in gsl? Hong