GCC 5.1 Release Candidate Now Available

[phoronix]

Phoronix: GCC 5.1 Release Candidate Now Available

A short time after branching GCC 5 and initiating GCC 6.0 development, the first GCC 5.1 Release Candidate has surfaced in marking the big GCC 5...

http://www.phoronix.com/scan.php?page=news_item&px=GCC-5.1-Release-Candidate

[hubicka]

See David's blog on GCC's JIT http://developerblog.redhat.com/2015...using-gcc-5-2/
and also I got around to write something about the LTO and inter-procedural optimization changes in GCC 5

http://hubicka.blogspot.ca/2015/04/GCC5-IPA-LTO-news.html

[mlau]

Oh great! Time to let gcc 5 loose on my MIPS and SH4 systems.

[curaga]

@hubicka

Did you get around to testing webkit?

[hubicka]

@hubicka

Did you get around to testing webkit?

Just to the degree of building it, chasing out all compiler crashes, and verifying it starts & is smaller in binary size that non-LTO and clang binary. Did not figure out how to benchmark anything about it yet.

What I plan to look into next is getting Chromium builds down from 10GB of memory use now when GCC 5 is (mostly) out.

[curaga]

In webkitFLTK, I added a render bench app for this. It quits as soon as the page is fully rendered.

example:
time ./bench/webkitbench path/to/some/large/SVG

[Krejzi]

@hubicka

Did you get around to testing webkit?

I have built webkitgtk-2.8.0 with gcc-5-20150405 snapshot and the renderer process crashed a lot when I started epiphany. Rebuilding it with clang-3.6.0 fixed the issue and there were no issues when it was built with gcc-4.9.2 either. I can't say about build speeds though, but I did notice certain speedups in some packages' compilation.

[hubicka]

I have built webkitgtk-2.8.0 with gcc-5-20150405 snapshot and the renderer process crashed a lot when I started epiphany. Rebuilding it with clang-3.6.0 fixed the issue and there were no issues when it was built with gcc-4.9.2 either. I can't say about build speeds though, but I did notice certain speedups in some packages' compilation.

Well, as usual, this needs to be analyzed to work out if it is bug in webkit or GCC (which in case of such large beast may not be completely trivial to do). I will check if I get some interesting crashes from my Chromium builds.

[Krejzi]

Well, as usual, this needs to be analyzed to work out if it is bug in webkit or GCC (which in case of such large beast may not be completely trivial to do). I will check if I get some interesting crashes from my Chromium builds.

I couldn't even build the latest chromium stable release due to a bug in chromium code about which Google devs are aware but don't want to do anything at the moment. That's the second package I had to compile with clang (out of ~1000).

Some GStreamer-1.4.5 plugin crashed Firefox when accessing youtube (some reference to gstreamer memory allocator), so that's the third set of packages I needed to compile with clang. I suspect either ffmpeg/libav plugin which is sensitive to GCC changes (or at least it was in the past).

Everything else was able to be built and work with GCC-5 when the porting was done for some packages (which the awesome Fedora/Debian devs already did for 90% of the packages I built). So far, good work! I'll re-check Gstreamer packages as soon as 5.1 stable is released, as the bug might have already been fixed.

[hubicka]

I couldn't even build the latest chromium stable release due to a bug in chromium code about which Google devs are aware but don't want to do anything at the moment. That's the second package I had to compile with clang (out of ~1000).

Yep, Chromium at my side needs a work around for incorrect narrowing conversions. I played with fixing it at Chromium side bug got lost in code generators, so ended up hacking GCC frontend to ignore it.