Hacker News new | comments | show | ask | jobs | submit login
Benchmarking AMD's New AOCC Compiler for Ryzen (phoronix.com)
25 points by rbanffy 9 months ago | hide | past | web | favorite | 7 comments

That's pretty disappointing. I know it's 1.0, and maybe these tests aren't the greatest, but even with something that appears to be favourable for AMD specific optimisations that was only a 3% boost. Would love to see a broader test set, more stuff that AMD specific optimisations could help with.

To me, the only thing that is disappointing is how AMD has been repeatedly bitten by Intel, which has been crippling compilers to output inefficient code for AMD processors, and we're only seeing now AMD actively working to defend itself against that type of fraud.

I think this is more a testament to the quality of the current state of compilers. (I could be wrong.)

I suspect this is a require prerequisite for enabling higher performance on AMD tuned codes. Like say AMD's ACML library (much like Intel's MKL). Give that Naples isn't shipping yet, and that's the target for things like ACML the timing seems about right.

For lots of benchmarks, there is no compiler that can optimize away a vector unit half as wide.

It'd have been interesting to see how Ryzen performs with icc-generated code.

Please AMD: Just contribute to GCC and Clang.

This is based on Clang. No doubt they want to upstream the improvements.

Applications are open for YC Summer 2018

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact