GCC 7 Has Been Branched, GCC 8.0 Now On Master

[phoronix]

Phoronix: GCC 7 Has Been Branched, GCC 8.0 Now On Master

The GCC 7 mainline code-base hit the important milestone today of having zero P1 regressions -- issues of the highest priority -- and as such they branched the GCC7 code-base and GCC 7.1 RC1 is then being announced later this week as they prepare for this first stable release of GCC 7...

http://www.phoronix.com/scan.php?page=news_item&px=GCC-7-Branched

[carewolf]

Yay. Time to watch for fun stuff.

[cj.wijtmans]

Gentoo stable just moved to gcc5 and for unstable GCC6 has to be unmasked. https://packages.gentoo.org/packages/sys-devel/gcc

[Brane215]

Gentoo stable just moved to gcc5 and for unstable GCC6 has to be unmasked. https://packages.gentoo.org/packages/sys-devel/gcc

I have all my stuff compiled with 6.3.0+ latest isl since it came out.

Will try 7 as soon as it becomes available.

[mlau]

I have all my stuff compiled with 6.3.0+ latest isl since it came out.

Does graphite actually make a difference? Last time I tried, graphite on mesa caused an ICE,
and I didn't really notice any improvements, my gut feeling was to the contrary.
And there's apparently noone in gcc-land that has a good understanding of graphite to
fix the bugs.

[Brane215]

Does graphite actually make a difference? Last time I tried, graphite on mesa caused an ICE,
and I didn't really notice any improvements, my gut feeling was to the contrary.
And there's apparently noone in gcc-land that has a good understanding of graphite to
fix the bugs.

Don't know about the difference, didn't care to measure it. I have compiled mesa with gcc-6.3.0+isl-0.18 ( now I have isl-git) with
flto and graphite optimisations. Will post flags later.

[Brane215]

OK, I've just recompiled it. I've recompiled mesa-9999 ( from git repo ) with gcc-6.3.0+isl-9999 (=isl-0.18 ebuild that I've tweaked so that it uses isl-git repo).

CFLAGS="-march=native -pipe -O3 -pthread -lpthread -flto -ffat-lto-objects -fuse-linker-plugin -pipe -fgraphite-identity -floop-nest-optimize -pthread"
CXXFLAGS="-march=native -pipe -O3 -pthread -lpthread -flto -ffat-lto-objects -fuse-linker-plugin -pipe -fgraphite-identity -floop-nest-optimize -pthread -fpermissive"
LDFLAGS="-Wl,-O -O3 -pthread -lpthread -march=native -Wl,--sort-common -Wl,-flto -Wl,--as-needed -ffat-lto-objects -fgraphite-identity -floop-nest-optimize"

[Inopia]

Got 99 regressions but P1 aint' one.

[mlau]

OK, I've just recompiled it. I've recompiled mesa-9999 ( from git repo ) with gcc-6.3.0+isl-9999 (=isl-0.18 ebuild that I've tweaked so that it uses isl-git repo).

CFLAGS="-march=native -pipe -O3 -pthread -lpthread -flto -ffat-lto-objects -fuse-linker-plugin -pipe -fgraphite-identity -floop-nest-optimize -pthread"
CXXFLAGS="-march=native -pipe -O3 -pthread -lpthread -flto -ffat-lto-objects -fuse-linker-plugin -pipe -fgraphite-identity -floop-nest-optimize -pthread -fpermissive"
LDFLAGS="-Wl,-O -O3 -pthread -lpthread -march=native -Wl,--sort-common -Wl,-flto -Wl,--as-needed -ffat-lto-objects -fgraphite-identity -floop-nest-optimize"

Has mesa become any faster with graphite enabled? I did notice a slight performance improvement with the classic software renderer on a MIPS32 target: glxgears went from 7 to 12 fps due to LTO alone.

Also, the ICE I mentioned is in ffmpeg-9999, not mesa

[Brane215]

WRT to mesa speed, didn't check. It was noticeably faster with flto and since I need aggressive optimisations for flto I thought graphite couldn't hurt, so I added it.

You mentioned MIPS32. Which machine have you compiled this for ?