New user self-registration is currently disabled. Please email eigen-core-team @ lists.tuxfamily.org if you need an account.
Bug 1360 - Test fail on ppc64el
Test fail on ppc64el
Status: RESOLVED FIXED
Product: Eigen
Classification: Unclassified
Component: Core - vectorization
3.3 (current stable)
PPC - AltiVec Linux
: Normal Wrong Result
Assigned To: Konstantinos Margaritis
:
Depends on:
Blocks:
  Show dependency treegraph
 
Reported: 2016-12-09 17:17 UTC by gladk
Modified: 2016-12-23 18:44 UTC (History)
5 users (show)



Attachments

Description gladk 2016-12-09 17:17:28 UTC
Dear Eigen deveopers.

we are in Debian are having bugreport #836844 [1] which detects the matrix multiplication problem on ppc64 platform. 

I hope that the solution for this bug is already found, see the following answer [2]. But we need of course the confirmation from upstream. We have a corresponding hardware to test it.

[1] https://bugs.debian.org/836844
[2] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=836844#68

Thank you

Anton
Comment 1 Christoph Hertzberg 2016-12-09 17:52:24 UTC
I'm moving this to vectorization. (So hopefully Markos can have a look at it).
I guess pnegate(const Packet4f& a) should also use negative zero then?
Comment 2 Konstantinos Margaritis 2016-12-12 13:28:27 UTC
Going to test this with proposed negative zero on my ppc64le system and will update the ticket, sorry for the delay in replying, but it is an extremely busy period for me.
Comment 3 Gael Guennebaud 2016-12-18 22:56:08 UTC
https://bitbucket.org/eigen/eigen/commits/d7f257380d20/
(tested on gcc farm)
Comment 4 gladk 2016-12-23 18:44:03 UTC
Dear all,

sorry, I missed the discussion. Thanks for fixing it, we will integrate into Debian this fix as fast as possible.

Note You need to log in before you can comment on or make changes to this bug.