GCC 8.0 Moves On To Only Regression/Documentation Fixes

[phoronix]

Phoronix: GCC 8.0 Moves On To Only Regression/Documentation Fixes

The GCC 8 compiler is on to its last stage of development...

http://www.phoronix.com/scan.php?page=news_item&px=GCC-8.0-Doc-Reg-Fixes-Only

[rene]

gcc became really slow, I build $sysroot for my mips64 R10000 Sgi Octane the other week: https://t2sde.org/hardware/workstation/Sgi/Octane/
and going from gcc-4.1 to gcc-7.2 is approximately twice as slow, like 20 seconds instead of 10 seconds for a regular open source application .c file, ... :-/ compiling perl takes 2 hours?
Build [9] at 01/15/2018 from 11:58:30 to 14:07:39 UTC
:-/ thanks god I cross compile most on a datacenter Epyc but still, ...
I remember gcc 2.95.3 on a 200 MHz IDT Winchip2 or AMD Athlon, that must indeed feel blazing fast instead.
I only upgraded the Octane because some packages like glib and friends started to fail without __thread TLS, ... but the price in compile time is just :-/

[Brane215]

gcc became really slow, I build $sysroot for my mips64 R10000 Sgi Octane the other week: https://t2sde.org/hardware/workstation/Sgi/Octane/
and going from gcc-4.1 to gcc-7.2 is approximately twice as slow, like 20 seconds instead of 10 seconds for a regular open source application .c file, ... :-/ compiling perl takes 2 hours?

Why don't you crosscompile packages for such ancient machine ?

[geearf]

gcc became really slow, I build $sysroot for my mips64 R10000 Sgi Octane the other week: https://t2sde.org/hardware/workstation/Sgi/Octane/
and going from gcc-4.1 to gcc-7.2 is approximately twice as slow, like 20 seconds instead of 10 seconds for a regular open source application .c file, ... :-/ compiling perl takes 2 hours?
Build [9] at 01/15/2018 from 11:58:30 to 14:07:39 UTC
:-/ thanks god I cross compile most on a datacenter Epyc but still, ...
I remember gcc 2.95.3 on a 200 MHz IDT Winchip2 or AMD Athlon, that must indeed feel blazing fast instead.
I only upgraded the Octane because some packages like glib and friends started to fail without __thread TLS, ... but the price in compile time is just :-/

How about the binaries though, are they faster or smaller?

[rene]

How about the binaries though, are they faster or smaller?

maybe sox/play decodes .flac files 1% faster, ...

[rene]

Why don't you crosscompile packages for such ancient machine ?

of course I do cross compile most of the $sysroot. but although our #t2sde is quite good at that, unfortunately not all the packages are cross compile friendly. https://t2sde.org
PS: I already noticed on my Epyc datacenter server that building latest t2/trunk:HEAD takes like twice as long as a build with downdated gcc/binutils, ... quite a difference when you want to test something and you have to wait 4 hours instead of 2, ..!

[geearf]

maybe sox/play decodes .flac files 1% faster, ...

That does not seem worth it :/

[rene]

That does not seem worth it :/

some quick test while at it. for simple hello world.c GCC compile-time became ~20% slower just between 4.9 and 7.2: https://rene.rebe.de/2018-01-18/gcc-...er-and-slower/
(mips64, Sgi Octane, latest glibc & binutils, N32 user-land, linux kernel 4.12 https://t2sde.org/packages/linux-ip30)

[geearf]

some quick test while at it. for simple hello world.c GCC compile-time became ~20% slower just between 4.9 and 7.2: https://rene.rebe.de/2018-01-18/gcc-...er-and-slower/
(mips64, Sgi Octane, latest glibc & binutils, N32 user-land, linux kernel 4.12 https://t2sde.org/packages/linux-ip30)

How about the binary size?
Do you have LTO on 7.2? It slows down compile time quite a bit.

[rene]

How about the binary size?
Do you have LTO on 7.2? It slows down compile time quite a bit.

No LTO, that would be even slower, … https://www.youtube.com/watch?v=AU_RV8uoTIo&t=25s