GCC 11 Is Moving Closer But Still Challenged By Many Regressions

[phoronix]

Phoronix: GCC 11 Is Moving Closer But Still Challenged By Many Regressions

GCC 11 is slated to enter "Stage 4" development at the end of this weekend after which only regression and documentation fixes will be permitted. The first GCC 11 stable release should be out in 2~3 months, but at the moment there is an increasing number of P1 regressions that are of the highest priority...

http://www.phoronix.com/scan.php?page=news_item&px=GCC-11-Nears-Stage-4

[Peter Fodrek]

https://www.phoronix.com/scan.php?page=news_item&px=Glibc-2.32-Released

This is big boost for AMD if reaches glibc 2.33

Stemming from Glibc semantics that effectively "cripple AMD" in just checking for Intel CPUs while AMD CPUs with Glibc are not even taking advantage of Haswell era CPU features, AMD developers are now looking at properly plumbing AMD Zen platform support into this important C library for Linux users.

https://www.phoronix.com/scan.php?page=news_item&px=glibc-hwcaps-RFC

I want to stop AMD performance "crippling"

Will glibc 2.33 help me,please?

[jabl]

https://www.phoronix.com/scan.php?pa...-2.32-Released

This is big boost for AMD if reaches glibc 2.33

https://www.phoronix.com/scan.php?page=news_item&px=glibc-hwcaps-RFC

I want to stop AMD performance "crippling"

Will glibc 2.33 help me,please?

I hope so too, but that's unrelated to the upcoming GCC 11 release, which the article was about.

[Peter Fodrek]

I hope so too, but that's unrelated to the upcoming GCC 11 release, which the article was about.

but it is related to AMD performance mentioned in article

The only downside of moving to stage four now abandons hopes of seeing more complete AMD Zen 3 (znver3) support for this release. The initial code was merged last month but lacking various tuning and other improvements over the prior znver1/znver2 targets

[Brane215]

Beside support for new toys from Intel/AMD etc, what is big selling point of version 11 ?
Feature and improvements list seems rather short.

[baryluk]

Beside support for new toys from Intel/AMD etc, what is big selling point of version 11 ?
Feature and improvements list seems rather short.

Full C++17 support, support of some C++20 features, better diagnostic and warnings all around, hardware assisted address sanitizer support on 64-bit ARM, GCC JIT support on WIndows, number of new intrinsics for new instructions, mainly on x86, autovectorizer improvements, newer versions of Phobos/D runtime, better paralalelization of complex loops when using OpenMP, and many more, i.e. optimizer can now recognize redundant if conditions better against same variable or sub-expression and optimize it better, support for A64FX CPUs, including vector intrinsics and autovectorizer improvements (which should help other SVE targets too), support for x86-64 micro-architecture levels and psABI support compatible with clang, which hopefully will make it easier to ship more optimized libraries around that utilize more modern x86-64 features (like SSE3 and SSE4, AVX, AVX2, AVX512, LZCNT, POPCNT, CMPXCHG16B, etc) in distros.

As of the platform / architecture / targets support. There were no removed targets, or changes in tiers on supported targets. Same as GCC 10.

Also the compiler source code itself now uses quite a bit of C++11 features, which is a big deal. It can't be compiled with C++98 anymore (i.e. gcc 3.4 for bootstrap will not work, and gcc 4.8 or later is required, or other C++11 compiler).

[pal666]

Full C++17 support, support of some C++20 features

c++17 was already fully supported, now c++20 is almost fully supported

[usta]

c++17 fully support exists for a long time, with gcc11 c++17 will be the default standard for c++ compiles.