On 2018-06-04 4:35 AM, Ole Streicher wrote:
could you make sure that this is not a numpy problem? There was a
similar case (however a different data type) here:
https://bugs.debian.org/898776
https://github.com/numpy/numpy/issues/11103
On hppa-linux, there is no support for float128. Long double and double
use the same 64-bit
IEEE format. Target is big endian.
There's nothing special in gcc's float support as far as I know.
Signalling nans use "mips" format.
The alignment for double is 8 bytes and strict. The linux kernel
handles unaligned floating
point loads and stores
If this is not a numpy problem, I would need a good description in what
hppa differs from the other archs (where the test succeeds), and a login
on a hppa machine so that I can test.
Helge, would you please provide an account for Ole on phantom.
Thanks,
Dave
--
John David Anglin dave.ang...@bell.net