User Profile

[bridgman]

You should see some changes in OpenCL in the upcoming 18.10 release - waiting for some last bugs to be addressed but still on track for this month AF...

[bridgman]

We are continuing to extend DKMS support for existing distro releases, including some broader OS coverage....

[bridgman]

As far as I know IBM's z/OS is pretty much a direct evolution of the mainframe OS code we were running in the 60's - still supporting all the same application...

[bridgman]

In fairness, OpenGL was like this maybe 20 years ago......

[bridgman]

Might be as soon as the next week or so, but will check on Monday....

[bridgman]

I initially read this as "it's more stable than Bridgman" and thought "yeah, true..."....

[bridgman]

It's not just "CUDA-to-something-different" though (like CUDA-to-OpenCL), it's "CUDA to portable C++ that runs on both environments"....

[bridgman]

Actually no - the CUDA to HIP/HCC part went pretty smoothly. What took the time (about 7 months between Tensorflow 1.3 and 1.7) was MIOpen, an open source...

[bridgman]

Yep. I ended up jumping from 68000 to the National 16032, which was a very slick design with a good page-based MMU, but it took a lot longer than expected...

[bridgman]

I am referring to the 18.10 packaged driver release - the one coming after 17.50:

https://support.amd.com/en-us/kb-articles/Pages/Radeon...

[bridgman]

I don't understand. You can take HIP output and run it through either NVidia or AMD compilers - the whole idea is single framework based on C++17....

[bridgman]

Ahh, that makes sense - thanks.

The 18.10 release (some time this month) should include an OpenCL implementation which does not require PCIE...

[bridgman]

Yeah, I should have said "most early 68xxx minicomputer systems" (typically running Unix).

For embedded systems MMUs were pretty...

[bridgman]

Are you talking about a regression in 4.16 or a problem which has always been there ?...

[bridgman]

When you say "interesting part of lspci" is that without the boot parms ? Guessing it is because you are using radeon, just checking.
...

[bridgman]

I don't think I understand what you are asking - Vega chips are "ROCm supporting hardware".

Are you just asking about running a...

[bridgman]

Yep... the 68010 was the first part that was able to recover from a page fault, so with 68000 you could use an MMU but not use fault-based mapping. That...

[bridgman]

It's easy to forget that mainframe CPUs have been using speculative execution for decades. Pretty sure the IBM 360/91 used it back in 1964 - not sure...

[bridgman]

Ahh, you think they interpreted that as "all hardware supported by amdgpu" not "all hardware supported by DC" ?

I wondered...

[bridgman]

I don't think Michael said anything about amdgpu and GCN 1.0 - not aware of any change in defaults....