[NOT FOR LANDING] experimental NVSHMEM integration #146593
Draft
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/146593
Note: Links to docs will display an error until the docs builds have been completed. ❌ 7 New Failures, 2 Unrelated FailuresAs of commit bcd4a36 with merge base 07b9fe0 ( NEW FAILURES - The following jobs have failed:
FLAKY - The following jobs failed but were likely due to flakiness present on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
cc H-Huang awgu kwen2501 wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
cc H-Huang awgu kwen2501 wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
seth-howell
reviewed
Feb 14, 2025
| set_target_properties(nvshmem_extension PROPERTIES CUDA_SEPARABLE_COMPILATION ON) | ||
| target_compile_options(nvshmem_extension PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:-rdc=true>) | ||
| target_link_libraries(nvshmem_extension PRIVATE | ||
| ${NVSHMEM_LIB_DIR}/libnvshmem.a |
There was a problem hiding this comment.
Typically we dynamically link libnvshmem_host.so and statically link libnvshmem_device.a.
You also don't need to link the extension to nvshmem_bootstrap_uid.so. It will by dynamically opened by NVSHMEM.
Alternatively, if you are only using host APIs, you can forego linking to libnvshmem_device.a and dynamically load libnvshmem_host.so. Which would mean you wouldn't actually have any build-time NVSHMEM dependencies from your module.
Collaborator
Author
There was a problem hiding this comment.
Alternatively, if you are only using host APIs, you can forego linking to libnvshmem_device.a and dynamically load libnvshmem_host.so
@seth-howell I tried this but it didn't work. I might've done something wrong, but I got a dynamic linker error complaining about some missing symboI (I forgot the name). I tried building with the host compiler, including only the host header and only calling nvshmem_init but they didn't help.
There was a problem hiding this comment.
Sorry, I probably should have provided a little more context. If you are dynamically loading the library you will want to use the nvshmemx_hostlib_init_attr and nvshmemx_hostlib_finalize APIs instead. https://docs.nvidia.com/nvshmem/api/gen/api/setup.html#nvshmemx-hostlib-init-attr
| target_compile_definitions(torch_cuda PRIVATE USE_NCCL) | ||
| endif() | ||
|
|
||
| # Use env var for these for now for prototyping purposes |
There was a problem hiding this comment.
FWIW, Outside of the prototyping phase, NVSHMEM does support cmake's find_package command.
yifuwang
pushed a commit
to yifuwang/pytorch
that referenced
this pull request
Feb 25, 2025
ghstack-source-id: dc491e4 Pull Request resolved: pytorch#146593
cc H-Huang awgu kwen2501 wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o [ghstack-poisoned]
kwen2501
added a commit
that referenced
this pull request
Apr 29, 2025
Adding NVSHMEM as a backend for `SymmetricMemory`, implementation of which is in `NVSHMEMSymmetricMemory.cu`. Moving some helper functions in `CUDASymmetricMemory.cu` to `CUDASymmetricMemoryUtils.cpp`, so that they can be shared by `NVSHMEMSymmetricMemory`. These functions are mostly side-band exchange helpers (`store_all_gather`, `IpcChannel`, etc). Adding `TORCH_SYMMEM` to control which implementation to use for CUDA tensors, currently support: `CUDA` (in-house impl), `NVSHMEM`. The NVSHMEM feature is gated by build-time flag: `USE_NVSHMEM=1`. And `NVSHMEM_HOME` setting is required (TODO). Ported most code from #146593. cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k [ghstack-poisoned]
kwen2501
added a commit
that referenced
this pull request
Apr 29, 2025
Adding NVSHMEM as a backend for `SymmetricMemory`, implementation of which is in `NVSHMEMSymmetricMemory.cu`. Moving some helper functions in `CUDASymmetricMemory.cu` to `CUDASymmetricMemoryUtils.cpp`, so that they can be shared by `NVSHMEMSymmetricMemory`. These functions are mostly side-band exchange helpers (`store_all_gather`, `IpcChannel`, etc). Adding `TORCH_SYMMEM` to control which implementation to use for CUDA tensors, currently support: `CUDA` (in-house impl), `NVSHMEM`. The NVSHMEM feature is gated by build-time flag: `USE_NVSHMEM=1`. And `NVSHMEM_HOME` setting is required (TODO). Ported most code from #146593. cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k [ghstack-poisoned]
kwen2501
added a commit
that referenced
this pull request
May 1, 2025
Adding NVSHMEM as a backend for `SymmetricMemory`, implementation of which is in `NVSHMEMSymmetricMemory.cu`. Moving some helper functions in `CUDASymmetricMemory.cu` to `CUDASymmetricMemoryUtils.cpp`, so that they can be shared by `NVSHMEMSymmetricMemory`. These functions are mostly side-band exchange helpers (`store_all_gather`, `IpcChannel`, etc). Adding `TORCH_SYMMEM` to control which implementation to use for CUDA tensors, currently support: `CUDA` (in-house impl), `NVSHMEM`. The NVSHMEM feature is gated by build-time flag: `USE_NVSHMEM=1`. And `NVSHMEM_HOME` setting is required (TODO). Ported most code from #146593. cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k [ghstack-poisoned]
kwen2501
added a commit
that referenced
this pull request
May 1, 2025
Adding NVSHMEM as a backend for `SymmetricMemory`, implementation of which is in `NVSHMEMSymmetricMemory.cu`. Moving some helper functions in `CUDASymmetricMemory.cu` to `CUDASymmetricMemoryUtils.cpp`, so that they can be shared by `NVSHMEMSymmetricMemory`. These functions are mostly side-band exchange helpers (`store_all_gather`, `IpcChannel`, etc). Adding `TORCH_SYMMEM` to control which implementation to use for CUDA tensors, currently support: `CUDA` (in-house impl), `NVSHMEM`. The NVSHMEM feature is gated by build-time flag: `USE_NVSHMEM=1`. And `NVSHMEM_HOME` setting is required (TODO). Ported most code from #146593. cc H-Huang awgu wanchaol fegin fduwjj wz337 wconstab d4l3k [ghstack-poisoned]
pytorchmergebot
pushed a commit
that referenced
this pull request
May 1, 2025
Adding NVSHMEM as a backend for `SymmetricMemory`, implementation of which is in `NVSHMEMSymmetricMemory.cu`. Moving some helper functions in `CUDASymmetricMemory.cu` to `CUDASymmetricMemoryUtils.cpp`, so that they can be shared by `NVSHMEMSymmetricMemory`. These functions are mostly side-band exchange helpers (`store_all_gather`, `IpcChannel`, etc). Adding `TORCH_SYMMEM` to control which implementation to use for CUDA tensors, currently support: `CUDA` (in-house impl), `NVSHMEM`. The NVSHMEM feature is gated by build-time flag: `USE_NVSHMEM=1`. And `NVSHMEM_HOME` setting is required (TODO). Ported most code from #146593. Pull Request resolved: #151261 Approved by: https://github.com/fegin, https://github.com/fduwjj
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Stack from ghstack (oldest at bottom):
cc @H-Huang @awgu @kwen2501 @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @c-p-i-o