Preliminary registered-buffer collective support via Inductor#138029
Closed
yifuwang wants to merge 14 commits intogh/yifuwang/148/basefrom
Closed
Preliminary registered-buffer collective support via Inductor#138029yifuwang wants to merge 14 commits intogh/yifuwang/148/basefrom
yifuwang wants to merge 14 commits intogh/yifuwang/148/basefrom
Conversation
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/138029
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit f2f8e1f with merge base 23d590e ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
…tor" ``` NOTE [lowering-time collective optimization] In collective communication libraries such as NCCL, every rank maintains communication buffers that are remotely accessible by some peers. Depending on the underlying transport, remote accessibility may be established via mechanisms such as ib_reg_mr, CUDA P2P, or CUDA multicast. Typically, these buffers are private to the communication library by default, and communication ops copy user data in and out of these buffers. To prevent these copies, an optimization commonly known as "user buffer registration" can be employed. This allows direct establishment of remote accessibility on user buffers, eliminating the need for copying. However, this optimization introduces stringent usage requirements, which are typically hard to satisfy without being intrusive to the user code: - Establishing remote accessibility is expensive and often done ahead of time. In such implementations, all ranks must agree on the set of allocations used for every collective op. Failing to meet this requirement can lead to runtime errors or even silent correctness issues. - Even if the collective communication library supports gracefully falling back to "unregistered" implementations, the fallback mechanism would nullify the optimization. - Some communication mechanisms impose stricter requirements than others. For example, CUDA's multicast + multi-mem instructions require all ranks to agree not only on the allocations used for every collective but also on the offsets within these allocations. To support all different mechanisms with optimal results, we aim to satisfy the strictest requirement for this family of optimizations - we ensures that every collective op invocation is guaranteed to operate on the same allocation, at the same offset, in every iteration. For eligible collective ops, we identify communication buffers at lowering time and optionally choose to lower the op to a different kernel (ommunication libraries like NCCL handle both registered and non-registered buffers transparently within the same op, though some may require different ops for different cases). Later, the codegen will perform "persistent allocation" to satisfy the aforementioned constraints, and optionally, perform buffer planning to optimize overall memory usage. ``` ### Changes - Created `comm_lowering.py` for the lowerings of `_c10d_functional` ops. This is to prevent cluttering `lowering.py` as we add more lowering-time collective optimizations. This PR moved the lowerings for `all_reduce` and `all_reduce_` to the file. - Added `comm_buffer_type: Dict[str, str]` to `GraphLowering` to track whether a buffer is a comm buffer and the type of the comm buffer. - Added codegen allocation support for comm buffers of type "symm_mem". - Added support for auto-lowering `_c10d_functional.all_reduce_` to `symm_mem.one_shot_all_reduce`. - Added an Inductor config for collective optimizations in general (`config._collective`). ### Limitation Currently, each persistently allocated comm buffer is dedicated to a single callsite. This is not viable in terms of memory usage. However, this is a neccesary intermediate state before we tackle memory planning for comm buffers. cc XilunWu H-Huang awgu kwen2501 wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Chillee
reviewed
Oct 16, 2024
…tor" ``` NOTE [lowering-time collective optimization] In collective communication libraries such as NCCL, every rank maintains communication buffers that are remotely accessible by some peers. Depending on the underlying transport, remote accessibility may be established via mechanisms such as ib_reg_mr, CUDA P2P, or CUDA multicast. Typically, these buffers are private to the communication library by default, and communication ops copy user data in and out of these buffers. To prevent these copies, an optimization commonly known as "user buffer registration" can be employed. This allows direct establishment of remote accessibility on user buffers, eliminating the need for copying. However, this optimization introduces stringent usage requirements, which are typically hard to satisfy without being intrusive to the user code: - Establishing remote accessibility is expensive and often done ahead of time. In such implementations, all ranks must agree on the set of allocations used for every collective op. Failing to meet this requirement can lead to runtime errors or even silent correctness issues. - Even if the collective communication library supports gracefully falling back to "unregistered" implementations, the fallback mechanism would nullify the optimization. - Some communication mechanisms impose stricter requirements than others. For example, CUDA's multicast + multi-mem instructions require all ranks to agree not only on the allocations used for every collective but also on the offsets within these allocations. To support all different mechanisms with optimal results, we aim to satisfy the strictest requirement for this family of optimizations - we ensures that every collective op invocation is guaranteed to operate on the same allocation, at the same offset, in every iteration. For eligible collective ops, we identify communication buffers at lowering time and optionally choose to lower the op to a different kernel (ommunication libraries like NCCL handle both registered and non-registered buffers transparently within the same op, though some may require different ops for different cases). Later, the codegen will perform "persistent allocation" to satisfy the aforementioned constraints, and optionally, perform buffer planning to optimize overall memory usage. ``` ### Changes - Created `comm_lowering.py` for the lowerings of `_c10d_functional` ops. This is to prevent cluttering `lowering.py` as we add more lowering-time collective optimizations. This PR moved the lowerings for `all_reduce` and `all_reduce_` to the file. - Added `comm_buffer_type: Dict[str, str]` to `GraphLowering` to track whether a buffer is a comm buffer and the type of the comm buffer. - Added codegen allocation support for comm buffers of type "symm_mem". - Added support for auto-lowering `_c10d_functional.all_reduce_` to `symm_mem.one_shot_all_reduce`. - Added an Inductor config for collective optimizations in general (`config._collective`). ### Limitation Currently, each persistently allocated comm buffer is dedicated to a single callsite. This is not viable in terms of memory usage. However, this is a neccesary intermediate state before we tackle memory planning for comm buffers. cc XilunWu H-Huang awgu kwen2501 wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
…tor" ``` NOTE [lowering-time collective optimization] In collective communication libraries such as NCCL, every rank maintains communication buffers that are remotely accessible by some peers. Depending on the underlying transport, remote accessibility may be established via mechanisms such as ib_reg_mr, CUDA P2P, or CUDA multicast. Typically, these buffers are private to the communication library by default, and communication ops copy user data in and out of these buffers. To prevent these copies, an optimization commonly known as "user buffer registration" can be employed. This allows direct establishment of remote accessibility on user buffers, eliminating the need for copying. However, this optimization introduces stringent usage requirements, which are typically hard to satisfy without being intrusive to the user code: - Establishing remote accessibility is expensive and often done ahead of time. In such implementations, all ranks must agree on the set of allocations used for every collective op. Failing to meet this requirement can lead to runtime errors or even silent correctness issues. - Even if the collective communication library supports gracefully falling back to "unregistered" implementations, the fallback mechanism would nullify the optimization. - Some communication mechanisms impose stricter requirements than others. For example, CUDA's multicast + multi-mem instructions require all ranks to agree not only on the allocations used for every collective but also on the offsets within these allocations. To support all different mechanisms with optimal results, we aim to satisfy the strictest requirement for this family of optimizations - we ensures that every collective op invocation is guaranteed to operate on the same allocation, at the same offset, in every iteration. For eligible collective ops, we identify communication buffers at lowering time and optionally choose to lower the op to a different kernel (ommunication libraries like NCCL handle both registered and non-registered buffers transparently within the same op, though some may require different ops for different cases). Later, the codegen will perform "persistent allocation" to satisfy the aforementioned constraints, and optionally, perform buffer planning to optimize overall memory usage. ``` ### Changes - Created `comm_lowering.py` for the lowerings of `_c10d_functional` ops. This is to prevent cluttering `lowering.py` as we add more lowering-time collective optimizations. This PR moved the lowerings for `all_reduce` and `all_reduce_` to the file. - Added `comm_buffer_type: Dict[str, str]` to `GraphLowering` to track whether a buffer is a comm buffer and the type of the comm buffer. - Added codegen allocation support for comm buffers of type "symm_mem". - Added support for auto-lowering `_c10d_functional.all_reduce_` to `symm_mem.one_shot_all_reduce`. - Added an Inductor config for collective optimizations in general (`config._collective`). ### Limitation Currently, each persistently allocated comm buffer is dedicated to a single callsite. This is not viable in terms of memory usage. However, this is a neccesary intermediate state before we tackle memory planning for comm buffers. cc XilunWu H-Huang awgu kwen2501 wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Closed
…tor" ``` NOTE [lowering-time collective optimization] In collective communication libraries such as NCCL, every rank maintains communication buffers that are remotely accessible by some peers. Depending on the underlying transport, remote accessibility may be established via mechanisms such as ib_reg_mr, CUDA P2P, or CUDA multicast. Typically, these buffers are private to the communication library by default, and communication ops copy user data in and out of these buffers. To prevent these copies, an optimization commonly known as "user buffer registration" can be employed. This allows direct establishment of remote accessibility on user buffers, eliminating the need for copying. However, this optimization introduces stringent usage requirements, which are typically hard to satisfy without being intrusive to the user code: - Establishing remote accessibility is expensive and often done ahead of time. In such implementations, all ranks must agree on the set of allocations used for every collective op. Failing to meet this requirement can lead to runtime errors or even silent correctness issues. - Even if the collective communication library supports gracefully falling back to "unregistered" implementations, the fallback mechanism would nullify the optimization. - Some communication mechanisms impose stricter requirements than others. For example, CUDA's multicast + multi-mem instructions require all ranks to agree not only on the allocations used for every collective but also on the offsets within these allocations. To support all different mechanisms with optimal results, we aim to satisfy the strictest requirement for this family of optimizations - we ensures that every collective op invocation is guaranteed to operate on the same allocation, at the same offset, in every iteration. For eligible collective ops, we identify communication buffers at lowering time and optionally choose to lower the op to a different kernel (ommunication libraries like NCCL handle both registered and non-registered buffers transparently within the same op, though some may require different ops for different cases). Later, the codegen will perform "persistent allocation" to satisfy the aforementioned constraints, and optionally, perform buffer planning to optimize overall memory usage. ``` ### Changes - Created `comm_lowering.py` for the lowerings of `_c10d_functional` ops. This is to prevent cluttering `lowering.py` as we add more lowering-time collective optimizations. This PR moved the lowerings for `all_reduce` and `all_reduce_` to the file. - Added `comm_buffer_type: Dict[str, str]` to `GraphLowering` to track whether a buffer is a comm buffer and the type of the comm buffer. - Added codegen allocation support for comm buffers of type "symm_mem". - Added support for auto-lowering `_c10d_functional.all_reduce_` to `symm_mem.one_shot_all_reduce`. - Added an Inductor config for collective optimizations in general (`config._collective`). ### Limitation Currently, each persistently allocated comm buffer is dedicated to a single callsite. This is not viable in terms of memory usage. However, this is a neccesary intermediate state before we tackle memory planning for comm buffers. cc XilunWu H-Huang awgu kwen2501 wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
…tor" ``` NOTE [lowering-time collective optimization] In collective communication libraries such as NCCL, every rank maintains communication buffers that are remotely accessible by some peers. Depending on the underlying transport, remote accessibility may be established via mechanisms such as ib_reg_mr, CUDA P2P, or CUDA multicast. Typically, these buffers are private to the communication library by default, and communication ops copy user data in and out of these buffers. To prevent these copies, an optimization commonly known as "user buffer registration" can be employed. This allows direct establishment of remote accessibility on user buffers, eliminating the need for copying. However, this optimization introduces stringent usage requirements, which are typically hard to satisfy without being intrusive to the user code: - Establishing remote accessibility is expensive and often done ahead of time. In such implementations, all ranks must agree on the set of allocations used for every collective op. Failing to meet this requirement can lead to runtime errors or even silent correctness issues. - Even if the collective communication library supports gracefully falling back to "unregistered" implementations, the fallback mechanism would nullify the optimization. - Some communication mechanisms impose stricter requirements than others. For example, CUDA's multicast + multi-mem instructions require all ranks to agree not only on the allocations used for every collective but also on the offsets within these allocations. To support all different mechanisms with optimal results, we aim to satisfy the strictest requirement for this family of optimizations - we ensures that every collective op invocation is guaranteed to operate on the same allocation, at the same offset, in every iteration. For eligible collective ops, we identify communication buffers at lowering time and optionally choose to lower the op to a different kernel (ommunication libraries like NCCL handle both registered and non-registered buffers transparently within the same op, though some may require different ops for different cases). Later, the codegen will perform "persistent allocation" to satisfy the aforementioned constraints, and optionally, perform buffer planning to optimize overall memory usage. ``` ### Changes - Created `comm_lowering.py` for the lowerings of `_c10d_functional` ops. This is to prevent cluttering `lowering.py` as we add more lowering-time collective optimizations. This PR moved the lowerings for `all_reduce` and `all_reduce_` to the file. - Added `comm_buffer_type: Dict[str, str]` to `GraphLowering` to track whether a buffer is a comm buffer and the type of the comm buffer. - Added codegen allocation support for comm buffers of type "symm_mem". - Added support for auto-lowering `_c10d_functional.all_reduce_` to `symm_mem.one_shot_all_reduce`. - Added an Inductor config for collective optimizations in general (`config._collective`). ### Limitation Currently, each persistently allocated comm buffer is dedicated to a single callsite. This is not viable in terms of memory usage. However, this is a neccesary intermediate state before we tackle memory planning for comm buffers. cc XilunWu H-Huang awgu kwen2501 wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
…tor" ``` NOTE [lowering-time collective optimization] In collective communication libraries such as NCCL, every rank maintains communication buffers that are remotely accessible by some peers. Depending on the underlying transport, remote accessibility may be established via mechanisms such as ib_reg_mr, CUDA P2P, or CUDA multicast. Typically, these buffers are private to the communication library by default, and communication ops copy user data in and out of these buffers. To prevent these copies, an optimization commonly known as "user buffer registration" can be employed. This allows direct establishment of remote accessibility on user buffers, eliminating the need for copying. However, this optimization introduces stringent usage requirements, which are typically hard to satisfy without being intrusive to the user code: - Establishing remote accessibility is expensive and often done ahead of time. In such implementations, all ranks must agree on the set of allocations used for every collective op. Failing to meet this requirement can lead to runtime errors or even silent correctness issues. - Even if the collective communication library supports gracefully falling back to "unregistered" implementations, the fallback mechanism would nullify the optimization. - Some communication mechanisms impose stricter requirements than others. For example, CUDA's multicast + multi-mem instructions require all ranks to agree not only on the allocations used for every collective but also on the offsets within these allocations. To support all different mechanisms with optimal results, we aim to satisfy the strictest requirement for this family of optimizations - we ensures that every collective op invocation is guaranteed to operate on the same allocation, at the same offset, in every iteration. For eligible collective ops, we identify communication buffers at lowering time and optionally choose to lower the op to a different kernel (ommunication libraries like NCCL handle both registered and non-registered buffers transparently within the same op, though some may require different ops for different cases). Later, the codegen will perform "persistent allocation" to satisfy the aforementioned constraints, and optionally, perform buffer planning to optimize overall memory usage. ``` ### Changes - Created `comm_lowering.py` for the lowerings of `_c10d_functional` ops. This is to prevent cluttering `lowering.py` as we add more lowering-time collective optimizations. This PR moved the lowerings for `all_reduce` and `all_reduce_` to the file. - Added `comm_buffer_type: Dict[str, str]` to `GraphLowering` to track whether a buffer is a comm buffer and the type of the comm buffer. - Added codegen allocation support for comm buffers of type "symm_mem". - Added support for auto-lowering `_c10d_functional.all_reduce_` to `symm_mem.one_shot_all_reduce`. - Added an Inductor config for collective optimizations in general (`config._collective`). ### Limitation Currently, each persistently allocated comm buffer is dedicated to a single callsite. This is not viable in terms of memory usage. However, this is a neccesary intermediate state before we tackle memory planning for comm buffers. cc XilunWu H-Huang awgu kwen2501 wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
…tor" ``` NOTE [lowering-time collective optimization] In collective communication libraries such as NCCL, every rank maintains communication buffers that are remotely accessible by some peers. Depending on the underlying transport, remote accessibility may be established via mechanisms such as ib_reg_mr, CUDA P2P, or CUDA multicast. Typically, these buffers are private to the communication library by default, and communication ops copy user data in and out of these buffers. To prevent these copies, an optimization commonly known as "user buffer registration" can be employed. This allows direct establishment of remote accessibility on user buffers, eliminating the need for copying. However, this optimization introduces stringent usage requirements, which are typically hard to satisfy without being intrusive to the user code: - Establishing remote accessibility is expensive and often done ahead of time. In such implementations, all ranks must agree on the set of allocations used for every collective op. Failing to meet this requirement can lead to runtime errors or even silent correctness issues. - Even if the collective communication library supports gracefully falling back to "unregistered" implementations, the fallback mechanism would nullify the optimization. - Some communication mechanisms impose stricter requirements than others. For example, CUDA's multicast + multi-mem instructions require all ranks to agree not only on the allocations used for every collective but also on the offsets within these allocations. To support all different mechanisms with optimal results, we aim to satisfy the strictest requirement for this family of optimizations - we ensures that every collective op invocation is guaranteed to operate on the same allocation, at the same offset, in every iteration. For eligible collective ops, we identify communication buffers at lowering time and optionally choose to lower the op to a different kernel (ommunication libraries like NCCL handle both registered and non-registered buffers transparently within the same op, though some may require different ops for different cases). Later, the codegen will perform "persistent allocation" to satisfy the aforementioned constraints, and optionally, perform buffer planning to optimize overall memory usage. ``` ### Changes - Created `comm_lowering.py` for the lowerings of `_c10d_functional` ops. This is to prevent cluttering `lowering.py` as we add more lowering-time collective optimizations. This PR moved the lowerings for `all_reduce` and `all_reduce_` to the file. - Added `comm_buffer_type: Dict[str, str]` to `GraphLowering` to track whether a buffer is a comm buffer and the type of the comm buffer. - Added codegen allocation support for comm buffers of type "symm_mem". - Added support for auto-lowering `_c10d_functional.all_reduce_` to `symm_mem.one_shot_all_reduce`. - Added an Inductor config for collective optimizations in general (`config._collective`). ### Limitation Currently, each persistently allocated comm buffer is dedicated to a single callsite. This is not viable in terms of memory usage. However, this is a neccesary intermediate state before we tackle memory planning for comm buffers. cc XilunWu H-Huang awgu kwen2501 wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Chillee
requested changes
Oct 18, 2024
| try: | ||
| # Only add empty_strided_p2p() if distributed and SymmetricMemory | ||
| # is available | ||
| from torch._C._distributed_c10d import _SymmetricMemory # noqa: F401 |
Collaborator
There was a problem hiding this comment.
I feel like it'd be good for us to refactor this code in some manner so that we only import the dependencies we need.
| ) | ||
|
|
||
|
|
||
| _bufs_to_skip_wait: OrderedSet[Tuple[int, str]] = OrderedSet() |
Chillee
approved these changes
Oct 18, 2024
Collaborator
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Collaborator
…tor" ``` NOTE [lowering-time collective optimization] In collective communication libraries such as NCCL, every rank maintains communication buffers that are remotely accessible by some peers. Depending on the underlying transport, remote accessibility may be established via mechanisms such as ib_reg_mr, CUDA P2P, or CUDA multicast. Typically, these buffers are private to the communication library by default, and communication ops copy user data in and out of these buffers. To prevent these copies, an optimization commonly known as "user buffer registration" can be employed. This allows direct establishment of remote accessibility on user buffers, eliminating the need for copying. However, this optimization introduces stringent usage requirements, which are typically hard to satisfy without being intrusive to the user code: - Establishing remote accessibility is expensive and often done ahead of time. In such implementations, all ranks must agree on the set of allocations used for every collective op. Failing to meet this requirement can lead to runtime errors or even silent correctness issues. - Even if the collective communication library supports gracefully falling back to "unregistered" implementations, the fallback mechanism would nullify the optimization. - Some communication mechanisms impose stricter requirements than others. For example, CUDA's multicast + multi-mem instructions require all ranks to agree not only on the allocations used for every collective but also on the offsets within these allocations. To support all different mechanisms with optimal results, we aim to satisfy the strictest requirement for this family of optimizations - we ensures that every collective op invocation is guaranteed to operate on the same allocation, at the same offset, in every iteration. For eligible collective ops, we identify communication buffers at lowering time and optionally choose to lower the op to a different kernel (ommunication libraries like NCCL handle both registered and non-registered buffers transparently within the same op, though some may require different ops for different cases). Later, the codegen will perform "persistent allocation" to satisfy the aforementioned constraints, and optionally, perform buffer planning to optimize overall memory usage. ``` ### Changes - Created `comm_lowering.py` for the lowerings of `_c10d_functional` ops. This is to prevent cluttering `lowering.py` as we add more lowering-time collective optimizations. This PR moved the lowerings for `all_reduce` and `all_reduce_` to the file. - Added `comm_buffer_type: Dict[str, str]` to `GraphLowering` to track whether a buffer is a comm buffer and the type of the comm buffer. - Added codegen allocation support for comm buffers of type "symm_mem". - Added support for auto-lowering `_c10d_functional.all_reduce_` to `symm_mem.one_shot_all_reduce`. - Added an Inductor config for collective optimizations in general (`config._collective`). ### Limitation Currently, each persistently allocated comm buffer is dedicated to a single callsite. This is not viable in terms of memory usage. However, this is a neccesary intermediate state before we tackle memory planning for comm buffers. cc H-Huang awgu kwen2501 wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang aakhundov XilunWu [ghstack-poisoned]
Collaborator
Author
|
@pytorchbot merge |
Collaborator
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Collaborator
|
The merge job was canceled or timed out. This most often happen if two merge requests were issued for the same PR, or if merge job was waiting for more than 6 hours for tests to finish. In later case, please do not hesitate to reissue the merge command |
…tor" ``` NOTE [lowering-time collective optimization] In collective communication libraries such as NCCL, every rank maintains communication buffers that are remotely accessible by some peers. Depending on the underlying transport, remote accessibility may be established via mechanisms such as ib_reg_mr, CUDA P2P, or CUDA multicast. Typically, these buffers are private to the communication library by default, and communication ops copy user data in and out of these buffers. To prevent these copies, an optimization commonly known as "user buffer registration" can be employed. This allows direct establishment of remote accessibility on user buffers, eliminating the need for copying. However, this optimization introduces stringent usage requirements, which are typically hard to satisfy without being intrusive to the user code: - Establishing remote accessibility is expensive and often done ahead of time. In such implementations, all ranks must agree on the set of allocations used for every collective op. Failing to meet this requirement can lead to runtime errors or even silent correctness issues. - Even if the collective communication library supports gracefully falling back to "unregistered" implementations, the fallback mechanism would nullify the optimization. - Some communication mechanisms impose stricter requirements than others. For example, CUDA's multicast + multi-mem instructions require all ranks to agree not only on the allocations used for every collective but also on the offsets within these allocations. To support all different mechanisms with optimal results, we aim to satisfy the strictest requirement for this family of optimizations - we ensures that every collective op invocation is guaranteed to operate on the same allocation, at the same offset, in every iteration. For eligible collective ops, we identify communication buffers at lowering time and optionally choose to lower the op to a different kernel (ommunication libraries like NCCL handle both registered and non-registered buffers transparently within the same op, though some may require different ops for different cases). Later, the codegen will perform "persistent allocation" to satisfy the aforementioned constraints, and optionally, perform buffer planning to optimize overall memory usage. ``` ### Changes - Created `comm_lowering.py` for the lowerings of `_c10d_functional` ops. This is to prevent cluttering `lowering.py` as we add more lowering-time collective optimizations. This PR moved the lowerings for `all_reduce` and `all_reduce_` to the file. - Added `comm_buffer_type: Dict[str, str]` to `GraphLowering` to track whether a buffer is a comm buffer and the type of the comm buffer. - Added codegen allocation support for comm buffers of type "symm_mem". - Added support for auto-lowering `_c10d_functional.all_reduce_` to `symm_mem.one_shot_all_reduce`. - Added an Inductor config for collective optimizations in general (`config._collective`). ### Limitation Currently, each persistently allocated comm buffer is dedicated to a single callsite. This is not viable in terms of memory usage. However, this is a neccesary intermediate state before we tackle memory planning for comm buffers. cc H-Huang awgu kwen2501 wanchaol fegin fduwjj wz337 wconstab d4l3k c-p-i-o voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang aakhundov XilunWu [ghstack-poisoned]
Collaborator
Author
|
@pytorchbot rebase |
Collaborator
|
@pytorchbot started a rebase job onto refs/remotes/origin/viable/strict. Check the current status here |
Collaborator
|
Successfully rebased |
Collaborator
Author
|
@pytorchbot merge |
Collaborator
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Contributor
|
@yifuwang I think this change is failing on ROCm, could you help take a look at the failure? distributed/test_symmetric_memory.py::LoweringTest::test_lowering_one_shot_all_reduce GH job link HUD commit link |
pytorchmergebot
pushed a commit
that referenced
this pull request
Nov 1, 2024
#139414) I'm not sure this is expected to run if it requires buffer-registration support CC @yifuwang @huydhn @syed-ahmed #138029 Pull Request resolved: #139414 Approved by: https://github.com/huydhn, https://github.com/yifuwang
rahulsingh-intel
pushed a commit
to rahulsingh-intel/pytorch
that referenced
this pull request
Nov 5, 2024
…h#138029) ``` NOTE [lowering-time collective optimization] In collective communication libraries such as NCCL, every rank maintains communication buffers that are remotely accessible by some peers. Depending on the underlying transport, remote accessibility may be established via mechanisms such as ib_reg_mr, CUDA P2P, or CUDA multicast. Typically, these buffers are private to the communication library by default, and communication ops copy user data in and out of these buffers. To prevent these copies, an optimization commonly known as "user buffer registration" can be employed. This allows direct establishment of remote accessibility on user buffers, eliminating the need for copying. However, this optimization introduces stringent usage requirements, which are typically hard to satisfy without being intrusive to the user code: - Establishing remote accessibility is expensive and often done ahead of time. In such implementations, all ranks must agree on the set of allocations used for every collective op. Failing to meet this requirement can lead to runtime errors or even silent correctness issues. - Even if the collective communication library supports gracefully falling back to "unregistered" implementations, the fallback mechanism would nullify the optimization. - Some communication mechanisms impose stricter requirements than others. For example, CUDA's multicast + multi-mem instructions require all ranks to agree not only on the allocations used for every collective but also on the offsets within these allocations. To support all different mechanisms with optimal results, we aim to satisfy the strictest requirement for this family of optimizations - we ensures that every collective op invocation is guaranteed to operate on the same allocation, at the same offset, in every iteration. For eligible collective ops, we identify communication buffers at lowering time and optionally choose to lower the op to a different kernel (ommunication libraries like NCCL handle both registered and non-registered buffers transparently within the same op, though some may require different ops for different cases). Later, the codegen will perform "persistent allocation" to satisfy the aforementioned constraints, and optionally, perform buffer planning to optimize overall memory usage. ``` ### Changes - Created `comm_lowering.py` for the lowerings of `_c10d_functional` ops. This is to prevent cluttering `lowering.py` as we add more lowering-time collective optimizations. This PR moved the lowerings for `all_reduce` and `all_reduce_` to the file. - Added `comm_buffer_type: Dict[str, str]` to `GraphLowering` to track whether a buffer is a comm buffer and the type of the comm buffer. - Added codegen allocation support for comm buffers of type "symm_mem". - Added support for auto-lowering `_c10d_functional.all_reduce_` to `symm_mem.one_shot_all_reduce`. - Added an Inductor config for collective optimizations in general (`config._collective`). ### Limitation Currently, each persistently allocated comm buffer is dedicated to a single callsite. This is not viable in terms of memory usage. However, this is a neccesary intermediate state before we tackle memory planning for comm buffers. Pull Request resolved: pytorch#138029 Approved by: https://github.com/Chillee ghstack dependencies: pytorch#138028
rahulsingh-intel
pushed a commit
to rahulsingh-intel/pytorch
that referenced
this pull request
Nov 5, 2024
pytorch#139414) I'm not sure this is expected to run if it requires buffer-registration support CC @yifuwang @huydhn @syed-ahmed pytorch#138029 Pull Request resolved: pytorch#139414 Approved by: https://github.com/huydhn, https://github.com/yifuwang
This was referenced Jan 13, 2026
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
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):
Changes
comm_lowering.pyfor the lowerings of_c10d_functionalops. This is to prevent clutteringlowering.pyas we add more lowering-time collective optimizations. This PR moved the lowerings forall_reduceandall_reduce_to the file.comm_buffer_type: Dict[str, str]toGraphLoweringto track whether a buffer is a comm buffer and the type of the comm buffer._c10d_functional.all_reduce_tosymm_mem.one_shot_all_reduce.config._collective).Limitation
Currently, each persistently allocated comm buffer is dedicated to a single callsite. This is not viable in terms of memory usage. However, this is a neccesary intermediate state before we tackle memory planning for comm buffers.
cc @H-Huang @awgu @kwen2501 @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @c-p-i-o @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @yf225 @chenyang78 @kadeng @muchulee8 @ColinPeppler @amjames @desertfire @chauhang @aakhundov @XilunWu