[Pkg-opencl-devel] beignet: weird test results

[Andreas Beckmann]

On 2014-11-02 13:43, Rebecca N. Palmer wrote:
> While the official test suite has many failures on 0.8 and few to none
> on 0.9.3, my own testing (see attached) suggests that those bugs (erf(c)
> diverges in tails, tgamma is actually lgamma, pow/pown give wrong sign)
> actually exist in *both* versions; the main difference between the two
> on my hardware appears to be that sin/cos/tan are much faster but less
> accurate in 0.9.3.

Hmm, this should be a (upstream) bug report ...

Andreas

PS: no need to Cc: me on mails going to the list