Major enhancements to the IB signal forwarding mechanisms
(`host-no-atomic` mode), primarily adding support for GDRCopy and MLX5
Direct Verbs, and refactoring the signal forwarding path for IB
HostNoAtomic mode. The changes fix memory consistency issues and reduce
signaling latency.
- GDRCopy and MLX5 Direct Verbs MR integration
- Signal forwarding path redesign
- Semaphore and connection API updates
- Environment (`MSCCLPP_FORCE_DISABLE_GDR`) and documentation updates
This pull request updates the deployment pipeline to allow custom CMake
arguments to be passed to the pip install process on remote VMs.
---------
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
## Summary
- Fix `run-remote.sh` to correctly execute multi-command scripts (e.g.,
multiple `mpirun` calls)
- The old approach piped decoded script through `base64 -d | bash`,
which feeds the script via bash's **stdin**. When `mpirun` (or its child
processes) runs, it can consume the remaining stdin, causing bash to
never see subsequent commands — only the first command would execute.
- The fix decodes the script to a **temp file** and runs `bash -euxo
pipefail "$TMP"` instead, so bash reads commands from the file and
`mpirun` consuming stdin has no effect.
- Applied to both the docker path (pssh + docker exec) and the
non-docker path (pssh only).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
## Summary
- Replace the two-step `signal()` implementation (`incOutbound()` +
`atomicStore()`) with a single fire-and-forget PTX
`red.release.sys.global.add.u64` instruction
- This eliminates one local atomic fetch-add and replaces a remote store
with a remote atomic add that has no return value — more efficient on
both NVIDIA (PTX `red`) and AMD (compiler optimizes `(void)fetch_add` to
fire-and-forget `flat_atomic_add_x2`)
- Add a C++ perf test (`PERF_TEST`) in `mp_unit` for signal+wait
ping-pong latency
### Performance (H100, 2 ranks, signal+wait round-trip)
```
SemaphorePerfTest.SignalPingPong:
Store-based (old): 2.595 us/iter
Red-based (new): 2.345 us/iter
Speedup: 1.11x
```
## Test plan
- [x] Builds successfully (`make mp_unit_tests`)
- [x] `mpirun -np 2 ./build/bin/mp_unit_tests --filter
"SemaphorePerfTest"` — 1.11x speedup
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
- Removes the GTest dependency, replacing it with a minimal custom
framework (`test/framework.*`) that covers only what the tests actually
use — a unified `TEST()` macro with SFINAE-based fixture auto-detection,
`EXPECT_*`/`ASSERT_*` assertions, environments, and setup/teardown.
- `--exclude-perf-tests` flag and substring-based negative filtering
- `MSCCLPP_ENABLE_COVERAGE` CMake option with gcov/lcov; CI uploads to
Codecov
- Merges standalone `test/perf/` into main test targets
- Refactors Azure pipelines to reduce redundancies & make more readable
---------
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: Changho Hwang <changhohwang@microsoft.com>
## Summary
This PR addresses a multicast resource leak, fixes `cuMemMap` offset
handling for multicast handles, renames NVLS allreduce algorithm classes
for clarity, and adds a new unit test for `SwitchChannel`.
### Bug Fixes
#### 1. Fix multicast allocation handle leak in `createMulticast()`
(`gpu_ipc_mem.cc`)
`GpuIpcMemHandle::createMulticast()` called
`cuMulticastCreate(&allocHandle, ...)` but never released the local
`allocHandle` after exporting it to shareable handles (POSIX FD /
Fabric). This caused a reference count leak — the multicast object was
never freed even after all mappings and imported handles were released.
Per the [CUDA Driver API docs for
`cuMemRelease`](https://docs.nvidia.com/cuda/cuda-driver-api/group__CUDA__VA.html):
> *"The memory allocation will be freed when all outstanding mappings to
the memory are unmapped and when all outstanding references to the
handle (including its shareable counterparts) are also released."*
The fix adds `cuMemRelease(allocHandle)` after export, matching the
existing pattern used for regular allocations in
`GpuIpcMemHandle::create()`.
**Impact:** Without this fix, repeated creation/destruction of NVLS
connections causes OOM after ~120 iterations when allocating 1GB
multicast buffers on H100.
#### 2. Fix `cuMemMap` offset for multicast handles (`gpu_ipc_mem.cc`)
`cuMemMap` requires `offset=0` for multicast handles. Previously, the
code attempted to map at a non-zero offset within the multicast object,
leading to errors when binding multiple buffers to the same
`NvlsConnection`. The fix maps the entire range `[0, mcOffset +
bufferSize)` and returns the pointer offset by `mcOffset`. This only
consumes extra virtual address space; no additional physical memory is
used.
### Refactoring
#### 3. Rename NVLS allreduce algorithm classes
Renamed for clarity:
- `AllreduceNvls` → `AllreduceNvlsZeroCopy`
- `AllreduceNvlsWithCopy` → `AllreduceNvlsWarpPipeline`
- `AllreduceNvlsWithCopy2` → `AllreduceNvlsBlockPipeline`
Updated all references in builder, selector, docs, and examples.
#### 4. Move `nvlsConnections` setup to `initialize()`
Moved `nvlsConnections_` from `AlgorithmCtx` (which no longer has this
member) to individual algorithm class members, initialized in their
`initialize()` methods.
### Tests
#### 5. Add `TwoChannelsSameConnection` test
New unit test that creates two `SwitchChannel` instances from the same
`NvlsConnection`, performs reduce operations on both, and verifies
correctness. This exercises the multi-bind path that triggered the
`cuMemMap` offset fix.
### Files Changed
- `src/core/gpu_ipc_mem.cc` — multicast handle leak fix + cuMemMap
offset fix
- `src/ext/collectives/allreduce/allreduce_nvls_zero_copy.cu` (renamed)
- `src/ext/collectives/allreduce/allreduce_nvls_warp_pipeline.cu`
(renamed)
- `src/ext/collectives/allreduce/allreduce_nvls_block_pipeline.cu`
(renamed)
- `src/ext/collectives/allreduce/allreduce_nvls_packet.cu` —
nvlsConnections fix
- `src/ext/collectives/include/allreduce/*.hpp` — renamed headers
- `src/ext/collectives/algorithm_collection_builder.cc` — updated
references
- `src/ext/nccl/algorithm_selector.cc` — updated algorithm names
- `test/mp_unit/switch_channel_tests.cu` — new test
- `docs/guide/mscclpp-torch-integration.md` — updated names
- `examples/torch-integration/customized_comm_with_default_algo.py` —
updated names
This pull request updates the way the `nlohmann/json` library is fetched
and upgrades it to a newer version in both the main build and test
configuration files.
Addressed installation issue in some env
This PR refactors the algorithm selection logic in MSCCL++ and
introduces support for symmetric memory configuration through
environment variables.
1. Algorithm Selection Refactoring
Use separate class for algo selection. Could introduce more complex
logic for algo selection based on message size, arch, if cuda graph is
enabled and memory allocation method
2. Symmetric Memory Support
Introduced symmetricMemory parameter in algorithm context key
generation. Remove disableChannelCache env as is ambiguous
3. Add new args for build_default_algorithms
Add flag_buffer, and flag_buffer_size args to build default algorithm.
Then we could use unified flag buffer for different algorithms, avoid
application hanging when switch algo for different message size.
---------
Co-authored-by: chhwang <8018170+chhwang@users.noreply.github.com>
Co-authored-by: Qinghua Zhou <qinghuazhou@microsoft.com>
Co-authored-by: Caio Rocha <caiorocha@microsoft.com>
* Added configurable InfiniBand (IB) signaling mode.
`EndpointConfig::Ib::Mode` enum selects the mode (`Default`, `Host`,
`HostNoAtomic`). `Default` is equivalent to `Host` unless specified
different by envrionment `MSCCLPP_IBV_MODE`. `Host` corresponds to the
previous implementation using RDMA atomics for signaling, while
`HostNoAtomic` uses write-with-immediate instead.
* Regarding updates in Python bindings and API.
Introduce handle cache for AMD platform.
Avoid reaching handle limitation if we open too much IPC handles
For nvidia, we don't need this feature since nvidia will count the
handle reference internally and reuse the same handle if already be
opened
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Binyang2014 <9415966+Binyang2014@users.noreply.github.com>
Co-authored-by: Changho Hwang <changhohwang@microsoft.com>
* Added `port` and `gidIndex` field in the IB endpoint config (and
`deviceIndex` field for future usages)
* Added `MSCCLPP_IBV_SO` env variable to specify a custom libibverbs.so
* Added `--ib_gid_index` CLI option to `mp_unit_tests`
* Other minor fixes
The key purpose is handling all mscclpp objects' memory internally by
hiding shared pointers from user APIs.
* `Connection` class is now a wrapper of `BaseConnection` class that is
equivalent to the previous `Connection` class
* `connect()` methods now return `Connection` instead of
`std::shared_ptr<Connection>`
* Removed `connectOnSetup()` method
* Always use `ibv_reg_dmabuf_mr` when DMABUF is supported
* Do not check `nvidia-peermem` when unnecessary
* More rigorous check on IB port availability
* Fixed ibverbs wrappers
* Fixed `IbPeerToPeerTest.SimpleAtomicAdd` test
* Add a compile flag `MSCCLPP_USE_IB` that explicitly specifies IB
on/off
* Fix `nvidia-peermem` check; no need for DMABUF supported systems
* Fix `mp_unit_tests` to skip all IB tests when built with
`-DMSCCLPP_USE_IB=OFF`
Use unix socket to share fd to other processes. Used for nvls handle sharing
Update nccl interface to support worldSize=1
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Running NVLS test, `test/nvls_test.cu` in CUDA 12.9 leads to illegal
memory access at
571fee16fb/test/nvls_test.cu (L151)
.
This PR addresses this error by moving cudaMemset after memory mapping.
Integrate MSCCL++ with torch
Introduce `NCCL audit shim library`, use can use following commands to
launch torch library. Also avoid break build pipeline in the CPU machine
```bash
export LD_AUDIT=$MSCCLPP_INSTALL_DIR/libmscclpp_audit_nccl.so
export LD_LIBRARY_PATH=$MSCCLPP_INSTALL_DIR:$LD_LIBRARY_PATH
torchrun --nnodes=1 --nproc_per_node=8 your_script.py
```
The PR contains following changes:
Python side:
- Channel based DSL implementation: decouple channel with chunk.
- Users create channel explicitly, only need local_rank, remote_rank and
channel_type
- Adjust executor json file, add remote_buffer fields, different op can
use different channel and remote buffers combination.
- Reimplement operation fusion, data dependency check mechanism
- Add new op such as semaphore, pipeline
- Clean code and enhance document
C++ side:
- Support new execution file json format
- Support semaphore and pipeline operation
- code clean, support non-zero copy scenario
---------
Co-authored-by: Caio Rocha <caiorocha@microsoft.com>
Co-authored-by: Changho Hwang <changhohwang@microsoft.com>
Change to use smart pointer for IB structure. Registered memory will own
ibMr, ibCtx will not held the reference
- Use smart pointer for IbQp and IbMr
- Update memoryChannel API, keep localRegisteredMemory
- Close fd when registedMemory released
---------
Co-authored-by: Changho Hwang <changhohwang@microsoft.com>
More intuitive interfaces for creating semaphores and channels. Also
allows channel construction using third-party bootstrappers directly
without overriding MSCCL++ Bootstrap.
* Add a FIFO test code that reproduced a correctness issue
* Fix the correctness issue by using pinned memory instead of cudaMemcpy
---------
Co-authored-by: Binyang Li <binyli@microsoft.com>
* Revert `MSCCLPP_FIFO_USE_TAIL_REPLICA=1` back to the default.
* Optimize `FifoDeviceHandle`.
* Do not use `cudaHostAllocWriteCombined` that increases latency.
* Pin host memory for `Host2DeviceSemaphore::outboundSemaphore_`.
* Fix proxy NUMA binding issues.
* Prevent graph capture inside proxy threads.
* Now `CudaIpcConnection` skips stream sync when unnecessary.
* Now any type of connection needs to hold a shared pointer to the
context for memory safety.
* Now a context should be always managed by a shared pointer for memory
safety.
* Minor docs & interface improvements.
* Minor fix in `mscclpp-test` correctness test.
Previous `gpuCalloc*()` creates a new stream for each allocation, which
messes the timeline up in profiler traces. Now `GpuStreamPool` allows
reusing the temporal streams.
* In cases when the same `tag` is used for receiving data from the same
remote rank, #514 changed the behavior of `Communicator::connect` and
`Communicator::recvMemory` to receive data in the order of
`std::shared_future::get()` is called, instead of the original behvaior
that receive data in the order of the method calls. Since the original
behavior is more intuitive, we get that back. Now when `get()` is called
on a future, the async function will first call `wait()` on the latest
previously returned future. In a recursive manner, this will call
`wait()` on all previous futures that are not yet ready.
* Removed all deprecated API calls and replaced into the new ones.
