I think that might actually be the code which *makes* it a requirement... I don't think anyone has written the patch to *not* require atomics on dGPUs with MEC microcode that uses RMW instead of atomics but will check. I believe that patch only said "PCIE atomics not required" for APUs.
Announcement
Collapse
No announcement yet.
AMDKFD GPUVM Support Updated For Discrete Radeon GPUs, Adds Userptr Support
Collapse
X
-
This is the key feature before AMD GPUs can be deployed to infrastructure. Many of us have RHEL and/or CentOS in the infrastructure which makes current ROCM useless. I'm looking forward to the day when ROCM is just an user space library that uses the upstream kernel. That will be an interesting day.
- Likes 1
Comment
-
Originally posted by vito View PostThis is the key feature before AMD GPUs can be deployed to infrastructure. Many of us have RHEL and/or CentOS in the infrastructure which makes current ROCM useless. I'm looking forward to the day when ROCM is just an user space library that uses the upstream kernel. That will be an interesting day.Last edited by bridgman; 19 March 2018, 06:59 AM.Test signature
- Likes 1
Comment
-
Originally posted by bridgman View Post
It turns out that we may be able to run the rest of the ROC stack without PCIE atomics as well. My understanding (subject to confirmation by testing) is that on Vega at least the MEC microcode no longer uses atomic operations, and substitutes regular read-modify-write operations instead. What I'm not sure of yet is whether the driver still checks for atomics and fails if they are not available - I suspect it does. Stay tuned.
Comment
-
Originally posted by bridgman View Post
You won't need to wait for upstream code to flow through into RHEL/CentOS kernels - we are also making good progress on KCL & DKMS packaging so you can install kernel modules in stock RHEL/CentOS kernels - but yes it will be nice to finally get all this upstream.
Comment
Comment