Radeon ROCm 2.9 Released With New "RALI" Library, rocTX

[phoronix]

Phoronix: Radeon ROCm 2.9 Released With New "RALI" Library, rocTX

Just one week after the release of Radeon Open Compute 2.8, AMD has now released ROCm 2.9 as the newest feature release for this open-source GPU Linux compute stack for Radeon hardware...

https://www.phoronix.com/scan.php?page=news_item&px=Radeon-ROCm-2.9-Released

[Madgemade]

ROCm 2.9 also introduces rocTX as a new C API for performance profiling

Yet another Profiler API. Hopefully this one actually works and provides features somewhat comparable to Nvidia.
Still no equivalent to "nvprof", C API is not going to be useful for a lot of people, a nice and easy tool would go a long way. Plus comprehensive documentation of course.

[smartbitcoin]

without Navi, please stop new release!

[coder]

AMD has remained quite

"quiet"

Also, this article is missing a comment thread: https://www.phoronix.com/scan.php?pa...lk-Moves-Lands

Otherwise, I'd have noted "The wine can be quite noticeable" should be "win".

[acowley]

Why was this release pushed out so quickly after 2.8? The release notes don’t remotely justify 2.8 having so short a life.

[Marc Driftmeyer]

One thing I've observed in this ROCm development growing pain in the ass reality, I've got plenty of time to wait on a custom AMD workstation build seeing as the goal posts for mature state in the Linux Kernel keeps getting kicked down the street, one kernel release after the next.

[tildearrow]

"quiet"

Also, this article is missing a comment thread: https://www.phoronix.com/scan.php?pa...lk-Moves-Lands

Otherwise, I'd have noted "The wine can be quite noticeable" should be "win".

I already reported the latter typo.

[coder]

I already reported the latter typo.

Really? Even now, I still can't reach the comment thread for that article!

[bridgman]

without Navi, please stop new release!

Can you help me understand how that would help ?

We have different teams with different skills working on various parts of the stack; some are heavily involved with new HW support while other components are not affected at all.

Having upper level component teams stop releasing their changes wouldn't help the other teams to work faster.

Why was this release pushed out so quickly after 2.8? The release notes don’t remotely justify 2.8 having so short a life.

IIRC the 2.8 release was delayed quite a bit waiting for a couple of last minute bug fixes, while 2.9 started to catch up with regular schedule.

[acowley]

bridgman Thank you for clarifying. I would have preferred 2.8 either be skipped, or at least come with notice that 2.9 was days away. As someone who packages ROCm (for NixOS), it’s a bit frustrating having to re-start a new release cycle before even finishing the last.