Jeff Breidenbach wrote:
Upstream says:
GCC 4.9 strikes again.
Strangely,while it built in the past with 4.8, I got an ICE when I tried
the current version of libwebp with the current gcc-4.8. Switching to
4.8 was the first thing I tried to workaround this issue. Switching to
4.7 was the second thing I tried (4.7 didn't suffer from the ICE but
suffered from another failure in a differnt file). Using -O1 was the
third thing I tried, using -O0 was the final thing I tried.
CFLAGS='-O2 -frename-registers' should do [the trick].
I have tested -frename-registers successfully on manual compiles of the
individual files that failed before, Testing on the package as a whole
is in progress.
If it succeeds should I go ahead and NMU? if so what if any delay would
you like me to use on the NMU?
I'm generally comfortable with NMU high urgency situations
like this one.
It's moderate urgency. The porters are in the process of pushing arm64
into the official archive and would like to see it in a healthy state
sooner rather than later. This bug is a blocker for reaching said
healthy state but it's not the only one and setting up the buildd and
actually doing the building will also take time so a few days delay to
get your opinion on this bug and/or see if you have any other fixes in
the pipieline that could go in the same upload probablly won't delay
anything else.
I prefer to keep the optimizations if possible,
because they are substantial.
* NEON assembly additions:
- ~25% faster lossy decode / encode (-m 4)
- ~10% faster lossless decode
- ~5-10% faster lossless encode (-m 3/4)
Obviously if we can get it building with optimisation enabled so much
the better.
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
