* Added RCCL env params to control setting the SO_REUSEADDR and SO_LINGER socket options. This can allow control over the number of file descriptors created during bootstrapping.
* Casted the linger value to `int` sooner to avoid a scope of unknown typed-ness.
* Added CHANGELOG entry for this feature.
* ext-src: add MSCCLPP memory registration APIs
* update mem-reg patch with mscclpp helper routine to check if buffer is registered
* RCCL integration of MSCCL++ user-buffer registration APIs
* only include mscclpp_nccl header if ENABLE_MSCCLPP is defined
* ext-src: update mscclpp mem-reg patch
* add helper routine to patch
* check handle before MSCCL++ deregister
* fix typo to replace send buff with recv buff
* in case of no mscclpp registration, dduring deRegister call, ont fall back to rccl deRegister which will return an error
* Apply suggestions from code review
Whitespace suggestions and reducing diffs to avoid future merge conflicts
Co-authored-by: corey-derochie-amd <161367113+corey-derochie-amd@users.noreply.github.com>
* rename helper functions and change their return type
* set RCCL user-buffer registration to occur if attempting MSCCL++ registration with a buffer in managed memory
---------
Co-authored-by: isaki001 <Ioannis.Sakiotis@amd.com>
Co-authored-by: isaki001 <36317038+isaki001@users.noreply.github.com>
Co-authored-by: corey-derochie-amd <161367113+corey-derochie-amd@users.noreply.github.com>
Network user buffer support for collectives
* Leverage user buffer registration to achieve zero-copy
inter-node communications for Ring, NVLS and Collnet
Add RAS subsystem
* Create a RAS thread keeping track of all NCCL communicators.
* Add a ncclras tool contacting the RAS thread and getting a
report.
Add fp8 support
* Add support for e5m2 and e4m3 8-bit floating point operations.
* Use Tree/PAT algorithms when possible for better numerical
stability.
Add NIC fusion
* Add a NET API to ask the network plugin to fuse a set of
interfaces together.
* Fuse multiple NICs under the same PCI switch as a single,
larger NIC.
Socket connection failure retry
* Retry in case of socket connection failure (unreachable host)
* Avoid "Software caused connection abort" errors on retries
QP connection failure retry
* Retry in case of IB QP connection failure during ibv_modify_qp.
NET API improvements
* Allow plugins to force a flush in case data and completion
ordering is not guaranteed.
* Indicate when completion is not needed (e.g. for the LL128
protocol), allowing plugins to skip generating a completion.
* Allow for full offload of allgather operations when using one
GPU per node.
NCCL_ALGO/NCCL_PROTO strict enforcement
* Extend NCCL_ALGO/NCCL_PROTO syntax to be able to specify
ALGO/PROTO filters for each collective operation.
* Strictly enforce the ALGO/PROTO filters, no longer fall back
on the ring algorithm when the filtering leaves no option and
error out instead.
Enable CUMEM host allocations
* Use cumem functions for host memory allocation by default.
Improved profiler plugin API
* Avoid dependencies with NCCL includes.
* Add information on whether the buffer is registered or not
Adjust PAT tuning
* Improve transition between PAT and ring at scale.
Fix hangs when running with different CPU architectures
* Detect when we use a mix of GPU architectures
* Ensure Algo/Proto decisions are made based on that unified
state.
Fix FD leak in UDS
* Fix a leak when mapping buffers intra-node with cumem IPCs.
Fix crash when mixing buffer registration and graph buffer registration.
* Separate local and graph registration to avoid crashes when we free
buffers.
Fix user buffer registration with dmabuf
* Make ncclSend/ncclRecv communication with buffer registration functional
on network plugins relying on dmabuf for buffer registration.
Fix crash in IB code caused by uninitialized fields.
Fix non-blocking ncclSend/ncclRecv
* Fix case where ncclSend/ncclRecv would return ncclSuccess in non-blocking
mode even though the operation was not enqueued onto the stream.
* Issue #1495
Various compiler tweaks and fixes
* PR #758
Fix typo in ncclTopoPrintGraph
* Issue #1468
It seems like here wants to check xgmi_node instead. If checks node for "nvlink", it will verify the link_info everytime.
If checks node for "xgmi", when get yes answer, it won't need check vsmi topo interface.
* Switched calls to `cudaMemcpyAsync` to be `cudaMemcpy` in `ncclTransportP2pSetup` to avoid race condition with `cudaIpcOpenMemHandle` inside p2p `connect`. See `ncclP2pImportShareableBuffer`.
* Moved synchronize outside of the loop, as it isn't necessary to sync between every iteration of the loop.
* Add Topologies for 16-GPU gfx942 SuperNode
- Add GigaIO topologies to tools/topo_expl for dev and testing
- Add GigaIO Columba 16 GPU romeModel and adjust topology
matching algorithm in rome_models for 16 GPU system
- Fix bug which failed to match Rome Model when using subsets
of system resources (i.e. ROCR_VISIBLE_DEVICES is set)
- Fixes for topo_expl
* Fix bug w/ 1H16P
* Changing C-strings to be const.
* Changed variable-length arrays to std::vector to avoid warnings. VLA is a compiler extension.
* Changed `#define` inside functions into `constexpr int` to preserve scoping and avoid macro redefinition warnings.
* Disabled warnings for modifying `CMAKE_CXX_FLAGS` caused by `check_symbol_exists`, which temporarily modifies the flag to do a compile check.
* Fixed VLA in rccl UT.
* Make sure the target device is used for MSCCL
* Enable single process mode by default to use MSCCL in MT
* Create a per-rank state when GPUs share a thread
Add scalable init API
* Add new ncclCommInitRankScalable to allow for passing multiple
unique IDs to the init function.
* Spreads the load onto multiple bootstrap roots, allowing for
constant bootstrap time.
* Requires multiple ranks to create a unique ID, and the CPU-side
ID exchange code to call allgather[v] instead of broadcast.
Accelerate init bootstrap operations
* Reduce the number of calls to allgather.
* Allow roots to reply early to ranks when information is already
available.
* Add an option to use ncclNet instead of sockets to perform
bootstrap allgather operations.
Add PAT algorithms for Allgather and ReduceScatter
* Parallel Aggregated Trees, variation of Bruck algorithm.
* Logarithmic number of network steps for small sizes at scale.
* Only supports one rank per node at the moment.
Add support for registered buffers for intra-node communication.
* Allow registered user buffers to be accessed directly intra-node
* Avoids extra copies in algorithms which permit it, saving
memory bandwidth and helping with compute overlap.
Add profiler plugin API
* New plugin API for profiling
* Supports various levels of profiling, with a hierarchy.
Asynchronous graph allocation
* Make calls to cudaMalloc and cudaMemcpy during graph allocation
asynchronous.
* Significantly speeds up graph capture.
Use fatal IB asynchronous events to stop network operation
* Avoids many other error messages
* Only fatal errors are affected; potentially transient errors
(e.g. port down) do not cause an immediate stop.
Set P2P level to PXB on AMD CPUs when using more than 2 GPUs per node
* P2P would cause a significant performance degradation when using
many GPUs, and therefore many interleaved data flows.
* Disable P2P through the CPU when we have 3+ GPUs per node; keep it
enabled when we only have 2 GPUs.
Improve the init logs to report the real NCCL function.
* Make the log report ncclCommInitRank or ncclCommSplit, rather than
the generic ncclCommInitRankFunc.
Add a parameter to set the location of the user configuration file.
* Add NCCL_CONF_FILE environment variable to set where the user's
configuration file resides.
Increase default IB timeout
* Increase IB timeout value from 18 to 20.
* Should help avoid fatal errors on large RoCE systems.
Add new check for nvidia peermem
* On linux kernels 6.6+, /sys/kernel/mm/memory_peers is no longer
present; check for /sys/module/nvidia_peermem/version instead.
Fix old performance regression when mixing small and large operations.
* Improves distribution of work on channels.
Fix crash when NUMA IDs are equal to -1.
* Can happen when a NIC is a virtual NIC, or when linux doesn't
know which NUMA node a device is attached to
* Issue NVIDIA/nccl-tests#233
Fix tree graph search when NCCL_CROSS_NIC is set to 1.
* Would force NCCL to use the balanced_tree pattern, thereby
disabling LL128 on platforms with 1 GPU+1 NIC per PCI switch.
* Would also try to use alternate rings even though it was not
needed.
Compiler tweaks and fixes
* PR #1177
* PR #1228
Fix stack smash
* PR #1325
Fixes for multi-node NVLink + IB operation
Coverity fixes and comments.
* Moved call to `mscclpp_ncclGetUniqueId` into `ncclCommInitRankFunc` to avoid setting up transport early in environments where MSCCL++ isn't valid.
* Checking `mscclEnabled` for the process and the topology to gate MSCCL++.
* Allowed `mscclForceEnable` to enable MSCCL++.
BertanDogancay
corey-derochie-amd
isaki001
Nusrat Islam
Dingming Wu
Luna
Sylvain Jeaugey
qiwei_ji
Xeonacid
WoosungMyung
Mustafa Abduljabbar
Ziyue Yang
Hujingbo
Benjamin Kitor
gilbertlee-amd
Bertan Dogancay
darren-amd
Abhishek Kulkarni
Avinash
Arm Patinyasakdikul
Wenkai Du
akolliasAMD
Nilesh M Negi
Mustafa Abduljabbar