Heterogeneous Memory Management v20 Published
It's looking less and less likely like Heterogeneous Memory Management (HMM) will be mainlined for the Linux 4.12 kernel. This is the long-in-development effort by Jerome Glisse that would benefit CUDA, OpenCL, and more by allow device memory to be transparently used by any device process and for mirroring process address space on a device.
Jerome has wanted HMM in Linux 4.12 (after previously pushing for it in 4.11) but it has yet to be pulled into the -next tree and 4.11 is potentially being released this weekend. Version 20 of the HMM patches were published overnight.
HMM v20 limits this feature to x86_64 Linux, contains various fixes, and the code is now re-based on top of the recent memory hot-plugging rework.
HMM is looking more and more ready, but given the timing, unless it's added last minute will likely be punted off to Linux 4.13. Those wanting to check out this 3,400 lines of new kernel code can find them via this patch series. At least when it arrives, it should be very useful for GPGPU applications.
Jerome has wanted HMM in Linux 4.12 (after previously pushing for it in 4.11) but it has yet to be pulled into the -next tree and 4.11 is potentially being released this weekend. Version 20 of the HMM patches were published overnight.
HMM v20 limits this feature to x86_64 Linux, contains various fixes, and the code is now re-based on top of the recent memory hot-plugging rework.
HMM is looking more and more ready, but given the timing, unless it's added last minute will likely be punted off to Linux 4.13. Those wanting to check out this 3,400 lines of new kernel code can find them via this patch series. At least when it arrives, it should be very useful for GPGPU applications.
5 Comments