[Bug 1449119] Re: [Ubuntu 15.10] gcc: Optimize swap removal in presence of floating-point vector convert/splat sequence

[Steve Langasek]

It looks like this bug will be fixed as a matter of course when gcc 5.1
is included in the distro and becomes the default in Ubuntu 15.10.
Assigning to Matthias for confirmation.

For 14.04, this appears to be addressed by the SRU of gcc-4.8 currently
in trusty-proposed, which will bring us up to revision 222448.

** Package changed: ubuntu => gcc-defaults (Ubuntu)

** Changed in: gcc-defaults (Ubuntu)
       Status: New => In Progress

** Changed in: gcc-defaults (Ubuntu)
     Assignee: (unassigned) => Matthias Klose (doko)

-- 
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to gcc-defaults in Ubuntu.
https://bugs.launchpad.net/bugs/1449119

Title:
  [Ubuntu 15.10] gcc: Optimize swap removal in presence of floating-
  point vector convert/splat sequence

Status in gcc-defaults package in Ubuntu:
  In Progress

Bug description:
  On Little Endian systems, GCC performs optimizations to remove
  unnecessary element reversals of vectors that are sometimes required
  to adjust for behavior of the lxvd2x, etc., instructions.  In one not
  uncommon case, such optimizations are defeated when we perform a
  vector convert from double-precision to single-precision followed by a
  vector splat word.  I have pushed patches upstream to fix this,
  implemented using the following revisions:

  GCC trunk (6.0):  222351
  GCC 5.1:  222394
  GCC 4.9:  222420
  GCC 4.8:  222427

  I would like to see the backported patch applied to the gcc shipped
  with the Ubuntu 15.10 release.  We should probably have a conversation
  about the next service pack for the 14.04 LTS release as well.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/gcc-defaults/+bug/1449119/+subscriptions