AMD AOCC 1.2 Code Compiler Offers Some Performance Benefits For EPYC

[phoronix]

Phoronix: AMD AOCC 1.2 Code Compiler Offers Some Performance Benefits For EPYC

Last month AMD released the AOCC 1.2 compiler for Zen systems. This updated version of their branched LLVM/Clang compiler with extra patches/optimizations for Zen CPUs was re-based to the LLVM/Clang 6.0 code-base while also adding in experimental FLANG support for Fortran compilation and various other unlisted changes to their "znver1" patch-set. Here's a look at how the performance compares with AOCC 1.2 to LLVM Clang 6.0 and GCC 7/8 C/C++ compilers.

http://www.phoronix.com/vr.php?view=26316

[davidbepo]

yeah, lets hope all this goes upstream, also it would be good on gcc not only llvm

[AndyChow]

I get why Intel would work on their own compiler. Why AMD would, with so little resources, I don't understand. It's like their storemi thing. No upside, a whole lot of downside. Just support and work with GCC and Visual Studio, and give them good documentation. We don't need a "meeToo" here.

[GruenSein]

I get why Intel would work on their own compiler. Why AMD would, with so little resources, I don't understand. It's like their storemi thing. No upside, a whole lot of downside. Just support and work with GCC and Visual Studio, and give them good documentation. We don't need a "meeToo" here.

My guess is that this is more of a testing ground and a platform for early feature roll out since getting stuff upstream in large open source projects can be hard (Remember DAL/DC?). Maybe, AMD wants to align compiler feature rollout with hardware releases or something. Vanilla LLVM won't care about that. Otherwise, I'd agree that I don't really get the point of another compiler.
Anyway, it's not a completely new compiler written from the ground up. It basically is LLVM/Clang 6.0 with some new features. If they prove themselves worthwhile, they will most likely go into LLVM at some point.

[wizard69]

I get why Intel would work on their own compiler. Why AMD would, with so little resources, I don't understand.

It isnt so much their own compiler as it is a fork or the LLVN suite. It is a place to do development and testing where features eventually get portedcback into the LLVM base. Frankly it is the way of open source.

It's like their storemi thing. No upside, a whole lot of downside.

What is the downside to improving compiler support under LLVM?

Just support and work with GCC and Visual Studio, and give them good documentation. We don't need a "meeToo" here.

now i get it you are an LLVM hater Generally manufactures avoid Visual Studio simple because the compiler duite was not compliant in the past and frakly still has short comings. With LLVM they engaged with a vast array of compiler developers.

[AndyChow]

My guess is that this is more of a testing ground and a platform for early feature roll out since getting stuff upstream in large open source projects can be hard (Remember DAL/DC?). Maybe, AMD wants to align compiler feature rollout with hardware releases or something. Vanilla LLVM won't care about that. Otherwise, I'd agree that I don't really get the point of another compiler.
Anyway, it's not a completely new compiler written from the ground up. It basically is LLVM/Clang 6.0 with some new features. If they prove themselves worthwhile, they will most likely go into LLVM at some point.

Yes. And I hope it does get into LLVM at some point. But I doubt these new features will ever become tangible. Even Intel, with their billions in resources, have only produced sub-par results. Creating a good compiler, or even slight improvements on a mature compiler, is next to impossible.