Akshay-Venkatesh

UCT/CUDA_IPC: Use active-queues to track outstanding work

4

## What/Why ? Currently CUDA_IPC transport uses integer stream_count to track outstanding work but in preparation for multi-device support, this PR moves to active_queue usage similar to cuda_copy transport. This...

UCT/CUDA_IPC: use rcache instead of pgtable

21

## What/Why ? Redo https://github.com/openucx/ucx/pull/8447

UCT/API: Introduce migratable_mem_types

2

## What/Why ? Introduce migratable_mem_types field that is populated by MDs. These mem_types would include the list of the memory types whose pages can migrate between host and PCIe device...

Is emacs_chrome supported on chromebooks?

1

Hi, I'm trying to use this extension on a chromebook but due to the lack of an emacs-server, I couldn't get it to work. Are there any workarounds? Let me...

UCT/CUDA_IPC: Add VMM/mallocasync support over fabric handles

6

## What Allow VMM/Mallocasync created by user with FABRIC handles to take advantage of cuda-ipc transport

UCP/CONTEXT: Add host,cuda_managed to ODP memtypes if Grace CPU is detected

2

## What Simpler PR to enable ODP for host/cuda_managed memory on Grace compared to https://github.com/openucx/ucx/pull/9370 ## Out-of-scope and deferred to sysadmins: 1. Detecting if ODPv2 is enabled on the system...

UCT/CUDA: Detect fabric vmm memory

8

## What/Why? Breaking https://github.com/openucx/ucx/pull/9787 into parts. This part isolates: 1. Fabric memory detection in cuda_copy transport 2. Change UCM cuda hooks to treat mallocasync memory as CUDA memory and not...

## What/Why? Second part of https://github.com/openucx/ucx/pull/9787 and follow up to https://github.com/openucx/ucx/pull/9867. This implements IPC-support for fabric handle associated memory using the newer import/export API and caches import operations similar to...

ompi/coll/accelerator: implement reduce_local

6

reduce_local implementation

v4.1.x: opal/cuda: avoid direct access to cumem host numa memory

8

Memory allocated using cumemcreate API with location as {[CU_MEM_LOCATION_TYPE_HOST/CU_MEM_LOCATION_TYPE_HOST_NUMA/CU_MEM_LOCATION_TYPE_HOST _NUMA_CURRENT](https://docs.nvidia.com/cuda/cuda-driver-api/group__CUDA__TYPES.html#group__CUDA__TYPES_1g75cfd5b9fa5c1c6ee2be2547bfbe882e)} can be detected as host memory type by [pointer query API](https://docs.nvidia.com/cuda/cuda-driver-api/group__CUDA__UNIFIED.html#group__CUDA__UNIFIED_1g89f7ad29a657e574fdea2624b74d138e) but this doesn't allow the CPU to access such...