Ioannis Assiouras

Hi @msimberg > Do you think there's a chance this could be fixed to be a constant time operation in future releases (and if so, should I open a separate...

Hi @msimberg a fix for this has been added with this change: https://github.com/ROCm/clr/commit/8bdda9007104ddd90f8a77955821d2c765979ad3 And it has made it into the 6.1 release. Could you please confirm that it resolves the...

Hi @pxl-th can you please attach a reproducer for the issue. Can you reproduce the hang in C++ as well?

Hi @pxl-th I think the slowness in async malloc/free is caused by a bug triggered by small (

Hi @pxl-th the fix for < 8-byte allocations has been merged in develop https://github.com/ROCm/clr/commit/2ede1c9adb76e04627fea8126f9cea9c03740574 and it should appear in future release. Regarding the 16-byte allocations timing test: There is an...

I am using rocm5.7.1 on ubuntu 2204.03 and gfx1030, the only difference is that my card is a RX 6900 XT but I cannot reproduce the issue. I will try...

Thanks @JackAKirk I think this is a slightly older version of the driver that seems to correspond to 5.6 so you can try to upgrade that but first can you...

@JackAKirk please see https://rocm.docs.amd.com/en/latest/release/gpu_os_support.html The PCIe atomic requirement for ROCm is listed under "CPU support" and it applies to all supported amd cards. device assert (similar to printf() and device-side...

Hi @JackAKirk yes I think that the relevant part is this: AtomicOpsCap: Routing- 32bit- 64bit- 128bitCAS- can you also provide the output of lspci -t

Hi @JackAKirk on this new machine can you now check the atomics for 00:01.0 `sudo lspci -s 00:01.0 -vv | grep AtomicOpsCap` I expect it will show something like AtomicOpsCap:...