Shweta Khatri

This issue has been addressed by https://github.com/ROCm/rocminfo/pull/55 and will be available in ROCM 6.1 release.

@AngryLoki - Sorry, my bad! The patches you have suggested make sense for Musl libc. I am not sure about this change - [@@ -122,12 +125,6 @@ class os_thread {](https://github.com/AngryLoki/gentoo/blob/5eaf4513b4af1c0ae9ba8f30874d7168010c057d/dev-libs/rocr-runtime/files/rocr-runtime-5.7.1-musl.patch#L43)...

Sorry for the delay. @AngryLoki . This patch and https://github.com/ROCm/ROCR-Runtime/pull/143 have been integrated and will be part for ROCm 6.2 release. Thank you.

Hi @joanbm , Could you please run your test with the follow patch? This fixes the pad_size address - ``` --- a/opensrc/hsa-runtime/core/runtime/amd_blit_sdma.cpp +++ b/opensrc/hsa-runtime/core/runtime/amd_blit_sdma.cpp @@ -471,0 +472 @@ hsa_status_t BlitSdma:...

Thanks for confirming @joanbm. This fix will be available with ROCM 6.1 release.

Thank you for reporting this hot thread issue with the second command queue handling in ROCm and your detailed analysis, preda. We've assigned this issue for immediate triage.

@preda -Like you have pointed out here- https://github.com/preda/clr/commit/03858366ea2dc96d504a16f7002e1781560e2b8a, its true that kernel driver can only support a maximum of 4096 interrupt signals per device. The failure to allocate any further...

@zhoumin2 - No..this seems to be a valid change. Thanks. I have merged this patch into our internal AMD branch.