[Bug 1661051] [NEW] vec_xl and vec_xst intrinsics are broken in GCC 6 for ppc64 and ppc64el

[Michael Hohnbaum]

Steve,

This one looks like it is Foundations.

                     Michael


On 02/01/2017 09:29 AM, Launchpad Bug Tracker wrote:
> bugproxy (bugproxy) has assigned this bug to you for Ubuntu:
>
> ---Problem Description---
> Please apply the patch from revision 245075 of the FSF gcc-6-branch as a bug fix for upstream bug PR79268, wherever GCC 6 is available.  Without the fix, wrong code will be generated in some cases for the vec_xl and vec_xst intrinsic functions, which provide access to unaligned vector load and store instructions on POWER8 and later.  The bug also shows up as a compile-time error when using a pointer-to-vector-pixel type as an argument to vec_xl or vec_xst.  This latter failure mode has been seen in the field.  A workaround for all failure modes is to replace vec_xl by vec_vsx_ld, and replace vex_xst by vec_vsx_st, but of course this is an invasive solution, and a fix to the compiler is preferred.  Thank you!
>
> ---Steps to Reproduce---
>  #include <altivec.h>
> extern vector pixel vp, *vpp;
> ...
> vp = vec_xl (0, vpp);
>  
> Bill Schmidt (wschmidt at us.ibm.com), Peter Bergner (bergner at us.ibm.com)
>
> ** Affects: ubuntu
>      Importance: Undecided
>      Assignee: Taco Screen team (taco-screen-team)
>          Status: New
>
>
> ** Tags: architecture-ppc64le bugnameltc-151137 severity-high targetmilestone-inin1704

-- 
Michael Hohnbaum
OIL Program Manager
Power (ppc64el) Development Project Manager
Canonical, Ltd.

-- 
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to gcc-4.9 in Ubuntu.
https://bugs.launchpad.net/bugs/1661051

Title:
  vec_xl and vec_xst intrinsics are broken in GCC 6 for ppc64 and
  ppc64el

Status in gcc-4.9 package in Ubuntu:
  New

Bug description:
  ---Problem Description---
  Please apply the patch from revision 245075 of the FSF gcc-6-branch as a bug fix for upstream bug PR79268, wherever GCC 6 is available.  Without the fix, wrong code will be generated in some cases for the vec_xl and vec_xst intrinsic functions, which provide access to unaligned vector load and store instructions on POWER8 and later.  The bug also shows up as a compile-time error when using a pointer-to-vector-pixel type as an argument to vec_xl or vec_xst.  This latter failure mode has been seen in the field.  A workaround for all failure modes is to replace vec_xl by vec_vsx_ld, and replace vex_xst by vec_vsx_st, but of course this is an invasive solution, and a fix to the compiler is preferred.  Thank you!

  ---Steps to Reproduce---
   #include <altivec.h>
  extern vector pixel vp, *vpp;
  ...
  vp = vec_xl (0, vpp);
   
  Bill Schmidt (wschmidt at us.ibm.com), Peter Bergner (bergner at us.ibm.com)

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/gcc-4.9/+bug/1661051/+subscriptions