Make Functorch interpreters serializable most of the time, so that we can save the guards on functorch states.
## Test Cases:
0. torch.compile() without functorch layers present. Guard should fail with any layer being pushed.
1. torch.compile() nested in vmap.
2. torch.compile() nested in grad.
3. torch.compile() nested in jvp + vmap
4. torch.compile() nested functionalize
5. torch.compile() nested in vmap + grad
Differential Revision: [D74008787](https://our.internmc.facebook.com/intern/diff/D74008787/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/152616
Approved by: https://github.com/zou3519
ghstack dependencies: #152615
I tried `beginAllocateToPool` instead of `_cuda_beginAllocateCurrentStreamToPool` and the error in #151199 does not happen any more.
However, this approach is unsafe for multithreading. When multiple run_eager happens concurrently, we expect memory allocation to different mem_pool. Since beginAllocateToPool does not check stream, these memory allocation may happen on the same mem_pool.
So, I use `_cuda_beginAllocateCurrentThreadToPool` to direct all memory allocation on the same thread to a given mem_pool. In particular, `_cuda_beginAllocateCurrentThreadToPool` records the launching thread id, and during runtime checks if the current thread id matches the launching thread id.
Fixes#151199
Pull Request resolved: https://github.com/pytorch/pytorch/pull/152472
Approved by: https://github.com/eellison, https://github.com/ngimel
I tried `beginAllocateToPool` instead of `_cuda_beginAllocateCurrentStreamToPool` and the error in #151199 does not happen any more.
However, this approach is unsafe for multithreading. When multiple run_eager happens concurrently, we expect memory allocation to different mem_pool. Since beginAllocateToPool does not check stream, these memory allocation may happen on the same mem_pool.
So, I use `_cuda_beginAllocateCurrentThreadToPool` to direct all memory allocation on the same thread to a given mem_pool. In particular, `_cuda_beginAllocateCurrentThreadToPool` records the launching thread id, and during runtime checks if the current thread id matches the launching thread id.
Fixes#151199
Pull Request resolved: https://github.com/pytorch/pytorch/pull/152472
Approved by: https://github.com/eellison
Summary:
# Context:
When memory leak happens, it usually trigger the OOM in the later iterations. The snapshot of full iteration will be huge and hard to interpret.
On CUDA side, they provide OOM observer which generates snapshot when OOM happens with latest 1,500,000 entries for debugging.
In this diff, we want to implement the feature on MTIA side
Test Plan:
Run this test with last diff in the stack.
```
buck run @//mode/opt kineto/libkineto/fb/mtia/integration_tests:mtia_memory_auto_trace_test
```
As shown, the memory_snapshot is generated when oom happens
Log: P1794792326
Snapshot: https://fburl.com/pytorch_memory_visualizer/lx73y6s3 {F1977402355}
Differential Revision: D71993315
Pull Request resolved: https://github.com/pytorch/pytorch/pull/152160
Approved by: https://github.com/sraikund16
MemPool is a separate pool of memory handled by the caching allocator. This PR adds the option let the caching allocator try to use this pool as a last resort instead of OOMing by associating a use_on_oom bool with each MemPool.
Usage:
Users can optionally specify a ``use_on_oom`` bool (which is False by default) during MemPool creation. If true, then the CUDACachingAllocator will be able to use memory in this pool as a last resort instead of OOMing.
```
pool = torch.cuda.MemPool(allocator, use_on_oom=True)
with torch.cuda.use_mem_pool(pool):
a = torch.randn(40 * 1024 * 1024, dtype=torch.uint8, device="cuda")
del a
# at the memory limit, this will succeed by using pool's memory in order to avoid the oom
b = torch.randn(40 * 1024 * 1024, dtype=torch.uint8, device="cuda")
```
Testing:
```
python test/test_cuda.py -k test_mempool_limited_memory_with_allocator
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/151487
Approved by: https://github.com/eqy, https://github.com/syed-ahmed, https://github.com/ngimel
This is a proof-of-concept of how we could serialize a guard and deserialize it back from the bytes.
The main behavioral change introduced in this diff is on CheckFunctionManager:
```
check_fn_manager = CheckFunctionManager(code, output_graph, guards_serialization_mode="save")
guards_state: bytes = check_fn_manager.guards_state
```
Once `guards_serialization_mode` is set to `save`, CheckFunctionManager will return an addtional `bytes` object called `guards_state` which should contain all the information needed for deserializing guards later.
When we load back guards state, we will set `guards_serialization_mode` is set to `load`:
```
output_graph_state = pickle.loads(guards_state)
check_fn_manager = CheckFunctionManager(code, output_graph_state, guards_serialization_mode="load")
```
# TENSOR_MATCH
Since we have many types of guards to support, we will break the work into small diffs instead of a single diff to support every guards.
We kick off the work from TENSOR_MATCH from this diff.
# Testing
For each type of guard we will test it like the following:
1. Use guard_filter_fn to select 1 type of guard each time.
2. Call InstructionTranslator directly on an example function to get OutputGraph and CheckFunctionManager (reference guard manager)
3. Serialize->deserialize the output graph state and re-build the guards with a new CheckFunctionManager (loaded guard manager)
4. Throw a set of example inputs to both reference and loaded guard manager to see if their behavior match.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/151318
Approved by: https://github.com/jansel, https://github.com/anijain2305
# Motivation
We propose adding support for the Python with statement on `torch.accelerator.device_index` to enable device switching functionality. This enhancement would simplify writing device-agnostic code and provide benefits across all accelerators. Its device-specific counterparts include [`torch.cuda.device`](00199acdb8/torch/cuda/__init__.py (L482)) and [`torch.cuda._DeviceGuard`](00199acdb8/torch/cuda/__init__.py (L469)).
**Design Philosophy**
It accepts either an `Int` or `None` as input. When `None` is passed, no device switch is performed. Supporting `None` is important for compatibility, as it's possible to encounter `None` values from `torch.device.index`.
Therefore, with this PR, we can do like this
```python
src = 0
dst = 1
# Set src to current device
torch.accelerator.set_device_index(src)
with torch.accelerator.device_index(dst):
# Inside with statement, we set dst to current device
assert torch.accelerator.get_device_index() == dst
# Here the current device should be src
assert torch.accelerator.get_device_index() == src
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148864
Approved by: https://github.com/albanD
This is a proof-of-concept of how we could serialize a guard and deserialize it back from the bytes.
The main behavioral change introduced in this diff is on CheckFunctionManager:
```
check_fn_manager = CheckFunctionManager(code, output_graph, guards_serialization_mode="save")
guards_state: bytes = check_fn_manager.guards_state
```
Once `guards_serialization_mode` is set to `save`, CheckFunctionManager will return an addtional `bytes` object called `guards_state` which should contain all the information needed for deserializing guards later.
When we load back guards state, we will set `guards_serialization_mode` is set to `load`:
```
output_graph_state = pickle.loads(guards_state)
check_fn_manager = CheckFunctionManager(code, output_graph_state, guards_serialization_mode="load")
```
# TENSOR_MATCH
Since we have many types of guards to support, we will break the work into small diffs instead of a single diff to support every guards.
We kick off the work from TENSOR_MATCH from this diff.
# Testing
For each type of guard we will test it like the following:
1. Use guard_filter_fn to select 1 type of guard each time.
2. Call InstructionTranslator directly on an example function to get OutputGraph and CheckFunctionManager (reference guard manager)
3. Serialize->deserialize the output graph state and re-build the guards with a new CheckFunctionManager (loaded guard manager)
4. Throw a set of example inputs to both reference and loaded guard manager to see if their behavior match.
Differential Revision: [D72987485](https://our.internmc.facebook.com/intern/diff/D72987485/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/151318
Approved by: https://github.com/jansel, https://github.com/anijain2305
Summary:
I can confirm that `torch.jit.Error.mro()` contains `Exception` in the inheritance hierarchy.
This avoids a bunch of `pyre-ignore`s in D73352417.
Test Plan: Sandcastle
Differential Revision: D73464544
Pull Request resolved: https://github.com/pytorch/pytorch/pull/151947
Approved by: https://github.com/Skylion007
This adds a non-blocking mode to queue_pop. This allows for workers to poll if work is ready without blocking the main loop. This is useful for the case where you want to have a GPU have maximum utilization when something only periodically is sent on the queue.
We also expose a `torch.distributed.QueueEmptyError` so users can catch the error and handle it accordingly.
Test plan:
```
pytest test/distributed/test_store.py -k queue -v -s -x
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/151485
Approved by: https://github.com/fduwjj, https://github.com/tianfengfrank
Summary: add one option to allow skipping all reduce unused parameters, this could help improve training throughput significantly when the number of unused parameters is large in the model.
Test Plan: unit tests, CI
Differential Revision: D72282069
Pull Request resolved: https://github.com/pytorch/pytorch/pull/151503
Approved by: https://github.com/mrshenli
# Motivation
Adapt `torch.accelerator.device_count` for multi-process usage. For example, `torch.cuda.device_count` avoids poisoning fork, then `torch.accelerator.device_count` should meet the same requirement.
Now that `torch.get_device_module(device).device_count` supports this, `torch.accelerator.device_count` should align with this behavior as well.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/149924
Approved by: https://github.com/albanD
ghstack dependencies: #147507
This adds lazy initialization support to ProcessGroupGloo via `TORCH_GLOO_LAZY_INIT` or via `create_device(..., lazy_init=True)`
This is still a draft PR as there's one race condition when doing coalesced operations that needs to be fixed upstream in Gloo first. Depends on https://github.com/facebookincubator/gloo/pull/427 landing first
This also updates the gloo submodule to include the required changes.
Test plan:
added lazy init test variants
```
pytest -v test/distributed/test_c10d_gloo.py -k Lazy
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/150801
Approved by: https://github.com/fduwjj
This PR creates two utils for generating a schema for hops from example inputs and use base hop as an exmaple.
1. HopArgumentInfoGen creates an argument or an output schema with mutation information.
2. CFuncitonSchemaGen piece together the argument info of inputs and outputs and produces torch._C.FunctionSchema.
is_write attribute of argument info can be computed. Note that the is_write annotation only works when the inputs are flattened (e.g. cannot support mutation inside tuple). We need special handling the case where we have tuple inputs like cond.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/149688
Approved by: https://github.com/zou3519
Relanding #148590 due to merge conflict.
This PR has multiple changes to `ProcessGroupNCCL` (which unfortunately are related):
1. When async_op=False, we directly launch the collective on "current" stream, instead of a trampoline stream and join back.
- Resolves#147729
- Resolves#146881
- Also saves two event syncs (which have overhead in case of HIP) and one pybind when we call `work.wait()` in distributed_c10d.py on behalf of user.
2. Entirely remove `record_stream` and use CPU-side stashing for managing tensor lifetime against recycling.
- Resolves#147168
3. Remove tensor life management when async_op=False; only use it when async_op=True.
4. To guard against user not calling `work.wait()`, we ask watchdog to unstash tensors after detecting completion of collectives, to prevent us from holding reference to tensors forever. This is a safety net, rather than a service guarantee, see discussion [here](https://github.com/pytorch/pytorch/issues/147168#issuecomment-2660142460).
5. Profile in async_op=False mode would look different -- collective kernels would show up in the same line and compute kernels.
Joint work with @cenzhaometa who wants to remove the event sync overhead.
Squashed contents:
* [ptd][nccl] use current-stream as nccl-stream under async=False mode (#147820)
PTD current workflow:
- PTD creates its own dedicated `ncclStream` for comm operation
- it will first add a dependency on current-stream (typically the compute stream) to ensure tensors are ready before invoking collective
such stream synchronization become expensive in Inference world (cpu overhead: 70us vs GPU kernel time: 160us).
This diff:
- async=False [default], will use current-stream as nccl-stream and avoid the stream-sync overhead
- async=True, will retain existing logic: create new nccl-stream, let it wait on current-stream to ensure tensors are ready
- pass down async from c10d down to NCCL-PG
this helps shave off 50% CPU overhead **(70us -> 35us)**, which reduce total CPU/GPU from **230us to 195us by 15%**
* [PGNCCL] Make avoid-record-stream default
* [c10d] Add asyncOp argument to Ops
* Change python side wait
* Pass asyncOp at ProcessGroup level
* Watchdog unstashing tensors as a safety net
* Stash tensors for reduce_scatter_v and all_gather_v
Pull Request approved: https://github.com/pytorch/pytorch/pull/149753
* [c10d] Move unstashing from watchdog to main thread
Pull Request approved: https://github.com/pytorch/pytorch/pull/150079
* [PGNCCL][BE] Merge mutex into TensorShelf for encapsulation
Pull Request approved: https://github.com/pytorch/pytorch/pull/150130
Pull Request resolved: https://github.com/pytorch/pytorch/pull/150398
Approved by: https://github.com/atalman
Summary:
Oftentimes, users complain that a bunch of extra events are prepended to their desired GPU snapshot. This is because they usually attach an OOM logger without knowing and when they go to collect the actual snapshot, it adds all the OOM logger contents. Since OOM and regular snapshot use the same backend, we currently don't have the infra in place to split these snapshots.
As a solution we add a flag to the snapshot frontend to clear out the history when starting the auto-trace record memory history.
A more thorough solution would be to have a user pass in a handle and to have snapshots per handle to seperate the events. However, this would likely be complicated and more work than it is worth as we would have to change the callbacks in the caching allocator and pass these objects between python and cpp.
Test Plan:
See diff below
Differential Revision: D71159720
Pull Request resolved: https://github.com/pytorch/pytorch/pull/149352
Approved by: https://github.com/eqy, https://github.com/aaronenyeshi
This is a new version of https://github.com/pytorch/pytorch/pull/148561 fixing the ROCM test failure
Putting this up for a first pass review, though I will likely make a bunch of changes before landing to add more features, etc.
This diff implements a first version of a static CUDA kernel launcher in `torch._C`. The goal here is to take a cubin file and some metadata from a CompiledKernel from `triton`, and launch the cubin file directly.
Background doc: https://docs.google.com/document/d/1rjRcHl6MfauHG30nCoQX-9UKvKyIs4WWMy_GsGyqb9g/edit?tab=t.0#heading=h.ut5lf39lzq66
Normally, using triton's CompiledKernel.make_launcher(), we would pay the cost of codegenning C++ and running it at compile time. With this new approach, we can use one statically compiled library to launch the kernel.
The tradeoff here is that this new kernel launcher will not be able to use codegen to deal with different lengths/types of arguments. So we use templating to handle up to 10 arguments for now. We also allocate 8 bytes on the stack per argument no matter the argument type, which can take more memory than codegenning. On the other hand, we improve compile time on cold and warm start by not having to call the C++ compiler at all.
This diff does not add the launcher to torch, but introduces a basic test suite.
A list of TODOs that are not yet complete:
- Handle `nvTmaDesc` and `cuTensorMap`, which triton handles
- Embed the grid logic instead of passing in gridX,Y,Z
- Handle launch_enter and exit hooks? (Not sure if inductor has these)
- Benchmarking to see if there's runtime performance loss
- Probably lots of features of the triton C++ generated code that I haven't handled yet.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/149238
Approved by: https://github.com/oulgen
Putting this up for a first pass review, though I will likely make a bunch of changes before landing to add more features, etc.
This diff implements a first version of a static CUDA kernel launcher in `torch._C`. The goal here is to take a cubin file and some metadata from a CompiledKernel from `triton`, and launch the cubin file directly.
Background doc: https://docs.google.com/document/d/1rjRcHl6MfauHG30nCoQX-9UKvKyIs4WWMy_GsGyqb9g/edit?tab=t.0#heading=h.ut5lf39lzq66
Normally, using triton's CompiledKernel.make_launcher(), we would pay the cost of codegenning C++ and running it at compile time. With this new approach, we can use one statically compiled library to launch the kernel.
The tradeoff here is that this new kernel launcher will not be able to use codegen to deal with different lengths/types of arguments. So we use templating to handle up to 10 arguments for now. We also allocate 8 bytes on the stack per argument no matter the argument type, which can take more memory than codegenning. On the other hand, we improve compile time on cold and warm start by not having to call the C++ compiler at all.
This diff does not add the launcher to torch, but introduces a basic test suite.
A list of TODOs that are not yet complete, will do in separate diff:
- Handle `nvTmaDesc` and `cuTensorMap`, which triton handles
- Embed the grid logic instead of passing in gridX,Y,Z. With https://github.com/pytorch/pytorch/pull/147583, we should be able to handle all of the grid logic directly in _StaticCudaLauncher.launch_kernel, and get rid of the python evaluation.
- Handle launch_enter and exit hooks? (Not sure if inductor has these)
- Benchmarking to see if there's runtime performance loss
- Hooking it up with a config to inductor
- Testing harness to test against torch generated triton kernels
Differential Revision: [D69926783](https://our.internmc.facebook.com/intern/diff/D69926783/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148561
Approved by: https://github.com/aorenste, https://github.com/syed-ahmed
Summary:
**Codegen**
- Skip some codegen parts for torchbind (such as arg decleration) because they are loaded in proxy executor, so we do not need to declare torchbind args in cpp code
- Added a helper method to get the schema of CallTorchBind HOP. The returned schema is only the schema of `obj.method()`.
**Serialization**
Add support for torchbind object in serialization
- For CallTorchBind HOP, we need to handle it specially because of it's schema. The output serialized args is in the format of `(obj, method, *args, **kwargs)`.
- it.TorchBindObject inputs are serialized to `as_custom_obj` Argument.
**Packaging**
Add torchbind objects file and `custom_objs_config.json` file to generated files output of `aot_compile`.
The json file is stored in the `data/aotinductor/<model_name>` folder in pt2 archive.
The torchbind objects are stored in data/constants/ folder in pt2 archive.
The format of torchbind objects are `f"{CUSTOM_OBJ_FILENAME_PREFIX}{custom_obj_idx}"`. e.g. `custom_obj_0`.
CustomClassHolder objects implement their own pickle methods.
Note that this `custom_objs_config.json` file is different from the `model_constants_config.json` file produced in package_sigmoid(). The keys in `custom_objs_config` directly correspond to the arg name in extern nodes json.
The key in `model_constants_config.json` produced by `package_sigmoid` is the attribute name in the user mode code.
This is required for both internal and OSS torchbind support.
For OSS torchbind support, we also need to package torchbind_constants into the .pt2 output.
**Work Left**
We still need to add torchbind support in ProxyExecutor for inductor.aoti_load_package to work. See other diffs in the stack.
Test Plan:
```
buck run fbcode//mode/dev-nosan //caffe2/test/inductor:torchbind -- -r schema
buck run fbcode//mode/dev-nosan //caffe2/test/inductor:torchbind -- -r aot_compile
```
Differential Revision: D69490718
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148506
Approved by: https://github.com/angelayi
This PR has multiple changes to `ProcessGroupNCCL` (which unfortunately are related):
1. When async_op=False, we directly launch the collective on "current" stream, instead of a trampoline stream and join back.
- Resolves#147729
- Resolves#146881
- Also saves two event syncs (which have overhead in case of HIP) and one pybind when we call `work.wait()` in distributed_c10d.py on behalf of user.
2. Entirely remove `record_stream` and use CPU-side stashing for managing tensor lifetime against recycling.
- Resolves#147168
3. Remove tensor life management when async_op=False; only use it when async_op=True.
4. To guard against user not calling `work.wait()`, we ask watchdog to unstash tensors after detecting completion of collectives, to prevent us from holding reference to tensors forever. This is a safety net, rather than a service guarantee, see discussion [here](https://github.com/pytorch/pytorch/issues/147168#issuecomment-2660142460).
5. Profile in async_op=False mode would look different -- collective kernels would show up in the same line and compute kernels.
Joint work with @cenzhaometa who wants to remove the event sync overhead.
Cc: @ngimel @awgu @Aidyn-A @skyw @wconstab @leonardo0lyj
Differential Revision: [D70937982](https://our.internmc.facebook.com/intern/diff/D70937982)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148590
Approved by: https://github.com/eqy, https://github.com/Aidyn-A, https://github.com/fduwjj
This PR has multiple changes to `ProcessGroupNCCL` (which unfortunately are related):
1. When async_op=False, we directly launch the collective on "current" stream, instead of a trampoline stream and join back.
- Resolves#147729
- Resolves#146881
- Also saves two event syncs (which have overhead in case of HIP) and one pybind when we call `work.wait()` in distributed_c10d.py on behalf of user.
2. Entirely remove `record_stream` and use CPU-side stashing for managing tensor lifetime against recycling.
- Resolves#147168
3. Remove tensor life management when async_op=False; only use it when async_op=True.
4. To guard against user not calling `work.wait()`, we ask watchdog to unstash tensors after detecting completion of collectives, to prevent us from holding reference to tensors forever. This is a safety net, rather than a service guarantee, see discussion [here](https://github.com/pytorch/pytorch/issues/147168#issuecomment-2660142460).
5. Profile in async_op=False mode would look different -- collective kernels would show up in the same line and compute kernels.
Joint work with @cenzhaometa who wants to remove the event sync overhead.
Cc: @ngimel @awgu @Aidyn-A @skyw @wconstab @leonardo0lyj
Differential Revision: [D70835197](https://our.internmc.facebook.com/intern/diff/D70835197)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148590
Approved by: https://github.com/eqy, https://github.com/Aidyn-A, https://github.com/fduwjj
This PR has multiple changes to `ProcessGroupNCCL` (which unfortunately are related):
1. When async_op=False, we directly launch the collective on "current" stream, instead of a trampoline stream and join back.
- Resolves#147729
- Resolves#146881
- Also saves two event syncs (which have overhead in case of HIP) and one pybind when we call `work.wait()` in distributed_c10d.py on behalf of user.
2. Entirely remove `record_stream` and use CPU-side stashing for managing tensor lifetime against recycling.
- Resolves#147168
3. Remove tensor life management when async_op=False; only use it when async_op=True.
4. To guard against user not calling `work.wait()`, we ask watchdog to unstash tensors after detecting completion of collectives, to prevent us from holding reference to tensors forever. This is a safety net, rather than a service guarantee, see discussion [here](https://github.com/pytorch/pytorch/issues/147168#issuecomment-2660142460).
5. Profile in async_op=False mode would look different -- collective kernels would show up in the same line and compute kernels.
Joint work with @cenzhaometa who wants to remove the event sync overhead.
Cc: @ngimel @awgu @Aidyn-A @skyw @wconstab @leonardo0lyj
Differential Revision: [D70835197](https://our.internmc.facebook.com/intern/diff/D70835197)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148590
Approved by: https://github.com/eqy, https://github.com/Aidyn-A, https://github.com/fduwjj
This adds `abort` and `shutdown` to `Backend` and `ProcessGroup` objects. This simplifies the logic in `distributed_c10d.py` by having a default noop implementation for all PGs.
This will be useful for torchft and upcoming versions of NCCL which will handle abort correctly. Currently `torchft` would have to call internal methods `_abort` on the PGNCCL object directly but with this change we can now just call `.abort()` and have it work for any PG implementation.
Test plan:
```
pytest distributed/test_backends.py distributed/test_c10d_common.py distributed/test_c10d_pypg.py
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148798
Approved by: https://github.com/kwen2501
This PR adds support for non-functional collectives under `FakeTensorMode` and `fake_pg`. It helps eliminate the patching of collectives for memory and runtime estimation.
It also modifies the `ModTracker` to enable the post-backward hook call for modules whose inputs don't require gradients but parameters do.
For the memory tracking, we now enable tracking DTensor dispatcher for custom dispatch functions like `entropy_loss`.
Dispatcher is only enabled for the memory tracking part and disabled as soon as it is done.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/147566
Approved by: https://github.com/weifengpy
This is an initial attempt to provide some statistics for the pinned host memory allocations flowing through CachingHostAllocator. Many times in the past we have had inexplicable slowdowns that would be much easier to diagnose if we had some host memory characteristics.
This change tries very hard not to disrupt the initial design of the allocator, and it uses existing locking mechanism, whenever possible, to gather statistics "for free". Only deviation from that is on the "slow path" where we incur CUDA calls anyway, so taking a short lock is not going to hurt the performance much, especially in the steady state where most allocations will come from cache.
As mentioned before, this is the first PR, to introduce the concept and to see if it fits the right paradigm. We can always add more later.
Metrics that would require more involved changes to the code base and locks, like requested memory, have been punted for now. I also tried to reuse the Stat structure used in CUDA caching allocator, in order to maintain symmetry.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/147660
Approved by: https://github.com/ngimel
Currently, recorded profiler events for aten ops do not store overload names. It would be useful to know which overloads are actually called to analyse performance.
For example, consider the following dispatch trace which occurs if there is a fallthrough kernel registered for aten::add:
```
[call] op=[aten::add.Tensor], key=[AutogradCPU]
[redispatch] op=[aten::add.Tensor], key=[Undefined]
[call] op=[aten::empty.memory_format], key=[BackendSelect]
[redispatch] op=[aten::empty.memory_format], key=[CPU]
[call] op=[aten::add.out], key=[CPU]
```
In this case, aten::add.out is a child of aten::add.Tensor, however the current profiler trace provides no way to differentiate aten op calls.
See the added unit test for a more detailed example.
Fixes #ISSUE_NUMBER
Pull Request resolved: https://github.com/pytorch/pytorch/pull/143114
Approved by: https://github.com/sraikund16
1. My company is using privateuseone to connect new hardware device and requires the use of `batch_isend_irecv` function. However, `batch_isend_irecv` is currently only open to CUDA, so I add `supports_coalescing` property in `c10d::Backend` to determine whether backend supports coalescing.
2. If `pg._has_hooks` return True, We don't need to determine if the current device is CUDA. So privateuseone can also support `pg._wait_for_pending_works`
Pull Request resolved: https://github.com/pytorch/pytorch/pull/135338
Approved by: https://github.com/kwen2501, https://github.com/albanD
This is an initial attempt to provide some statistics for the pinned host memory allocations flowing through CachingHostAllocator. Many times in the past we have had inexplicable slowdowns that would be much easier to diagnose if we had some host memory characteristics.
This change tries very hard not to disrupt the initial design of the allocator, and it uses existing locking mechanism, whenever possible, to gather statistics "for free". Only deviation from that is on the "slow path" where we incur CUDA calls anyway, so taking a short lock is not going to hurt the performance much, especially in the steady state where most allocations will come from cache.
As mentioned before, this is the first PR, to introduce the concept and to see if it fits the right paradigm. We can always add more later.
Metrics that would require more involved changes to the code base and locks, like requested memory, have been punted for now. I also tried to reuse the Stat structure used in CUDA caching allocator, in order to maintain symmetry.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/147660
Approved by: https://github.com/ngimel
Resubmission of #144974 which was reverted for unrelated reasons.
Newer matmul kernels, e.g. those targeting Hopper GPUs, sometime use a "persistent" schedule which consists in launching as many CUDA blocks as there are SMs on the GPU, with each such block then working on multiple output tiles in a row. This allows to eliminate the overhead of starting and finishing each tile, effectively doing cross-tile pipelining. In previous generations these latencies could be hidden by having multiple CUDA blocks per SM but, with blocks becoming larger, only one can run at a time per SM and thus this needs to be taken care of in software.
Persistent kernels become an issue when other kernels are running concurrently. The classical example is a NCCL communication kernel running in the background. In such cases the matmul expects to be able to use all the SMs but is prevented from doing so because some of the are busy. This can lead to its blocks being scheduled as two separate waves on the available SMs. This "wave quantization" can double the latency of the matmul kernels.
While we wait for smarter solutions, such as automatic load balancing among the blocks, an easy way to unblock ourselves is to tell the matmuls to only use a subset of the GPU's SMs. For this, I am introducing a global `sm_carveout` flag which can be used to specify how many SMs should be left available for other kernels.
For now I only change the cuBLAS kernels and the scaled-mm CUTLASS kernel. More kernels can be opted-in later.
I tested this change manually, by using the Kineto profiler to look up the grid size of a scaled-mm kernel with different values of `sm_carveout`, and making sure it changed. Suggestions are welcome for a more automated test.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/147966
Approved by: https://github.com/danthe3rd
Newer matmul kernels, e.g. those targeting Hopper GPUs, sometime use a "persistent" schedule which consists in launching as many CUDA blocks as there are SMs on the GPU, with each such block then working on multiple output tiles in a row. This allows to eliminate the overhead of starting and finishing each tile, effectively doing cross-tile pipelining. In previous generations these latencies could be hidden by having multiple CUDA blocks per SM but, with blocks becoming larger, only one can run at a time per SM and thus this needs to be taken care of in software.
Persistent kernels become an issue when other kernels are running concurrently. The classical example is a NCCL communication kernel running in the background. In such cases the matmul expects to be able to use all the SMs but is prevented from doing so because some of the are busy. This can lead to its blocks being scheduled as two separate waves on the available SMs. This "wave quantization" can double the latency of the matmul kernels.
While we wait for smarter solutions, such as automatic load balancing among the blocks, an easy way to unblock ourselves is to tell the matmuls to only use a subset of the GPU's SMs. For this, I am introducing a global `sm_carveout` flag which can be used to specify how many SMs should be left available for other kernels.
For now I only change the cuBLAS kernels and the scaled-mm CUTLASS kernel. More kernels can be opted-in later.
I tested this change manually, by using the Kineto profiler to look up the grid size of a scaled-mm kernel with different values of `sm_carveout`, and making sure it changed. Suggestions are welcome for a more automated test.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/144974
Approved by: https://github.com/eqy, https://github.com/albanD
This PR and the previous:
- Moves parts of `eval_frame.c` to C++.
- Reduces code duplication in `dynamo__custom_eval_frame` and makes the control flow more clear.
- Enables `convert_frame` to signal to `eval_frame.cpp` in a general manner how to evaluate this frame, recursive frames, and future frames with the same code object (default/compile, skip, run-only). e.g. this will allow us to change skipping/cache limit hit eval_frame behavior directly from convert_frame without requiring changes to C/C++.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/146355
Approved by: https://github.com/jansel
ghstack dependencies: #145603
# Motivation
Add context variable `torch.bachend.mkldnn.allow_tf32` to control tf32 computation in convolution kernels at XPU side. The tf32 data type is beneficial to improve the performance of deep learning workloads during training/inference. Current PR uses the [oneDNN API fpmath_mode](https://oneapi-src.github.io/oneDNN/dev_guide_attributes_fpmath_mode.html#the-floating-point-math-mode-attribute) to trigger the tf32 acceleration in convolution kernels.
# Valiadation
* ut to test context variable
`python test/xpu/test_conv.py -k test_mkldnn_allow_tf32_get_set`
* Runtime exemplification
```
onednn_verbose,primitive,exec,gpu:0,convolution,jit:ir,forward_training,src_f32::blocked:abcd::f0 wei_f32::blocked:abcd::f0 bia_f32::blocked:a::f0 dst_f32::blocked:abcd::f0,attr-scratchpad:user attr-fpmath:tf32,alg:convolution_direct,mb20_ic16oc33_ih50oh24kh3sh2dh0ph0_iw100ow49kw3sw2dw0pw0,0.649902
onednn_verbose,primitive,exec,gpu:0,convolution,jit:ir,forward_training,src_f32::blocked:abcd::f0 wei_f32::blocked:abcd::f0 bia_f32::blocked:a::f0 dst_f32::blocked:abcd::f0,attr-scratchpad:user attr-fpmath:tf32,alg:convolution_direct,mb20_ic33oc33_ih24oh24kh3sh1dh0ph1_iw49ow49kw3sw1dw0pw1,0.151855
onednn_verbose,primitive,exec,gpu:0,convolution,jit:ir,backward_data,src_f32::blocked:abcd::f0 wei_f32::blocked:abcd::f0 bia_undef::undef::: dst_f32::blocked:abcd::f0,attr-scratchpad:user attr-fpmath:tf32,alg:convolution_direct,mb20_ic33oc33_ih24oh24kh3sh1dh0ph1_iw49ow49kw3sw1dw0pw1,0.167969
onednn_verbose,primitive,exec,gpu:0,convolution,jit:ir,backward_weights,src_f32::blocked:abcd::f0 wei_f32::blocked:abcd::f0 bia_f32::blocked:a::f0 dst_f32::blocked:abcd::f0,attr-scratchpad:user attr-fpmath:tf32,alg:convolution_direct,mb20_ic33oc33_ih24oh24kh3sh1dh0ph1_iw49ow49kw3sw1dw0pw1,0.26709
onednn_verbose,primitive,exec,gpu:0,convolution,jit:ir,backward_weights,src_f32::blocked:abcd::f0 wei_f32::blocked:abcd::f0 bia_f32::blocked:a::f0 dst_f32::blocked:abcd::f0,attr-scratchpad:user attr-fpmath:tf32,alg:convolution_direct,mb20_ic16oc33_ih50oh24kh3sh2dh0ph0_iw100ow49kw3sw2dw0pw0,0.219971
```
According to the field `fpmath:tf32` in verbose, we could see that, current context setting utils could successfully trigger tf32 computation in conv forward/backward_data/backward_weights kernels.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/137570
Approved by: https://github.com/guangyey, https://github.com/EikanWang, https://github.com/atalman, https://github.com/malfet
Co-authored-by: Yu, Guangye <guangye.yu@intel.com>
Summary: Public summary (shared with Github): This diff implements a C++-Python binding to enable `reset_peak_memory_stats`.
Test Plan: The test is implemented in the following diff.
Reviewed By: yuhc
Differential Revision: D68988673
Pull Request resolved: https://github.com/pytorch/pytorch/pull/146710
Approved by: https://github.com/nautsimon
See the comment [here](https://github.com/pytorch/pytorch/issues/132014#issuecomment-2379547400) (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 @rec) - this PR updates `_unsafe_set_version_counter` to accept a list of tensors, for overhead-sensitive users (e.g. distributed) who need to hide VC bumps from autograd on a large list of tensors without wanting to suffer the overhead of going from python->C++ separately for every tensor in the list.
I left the binding in pybind, and used a `std::vector`. if we **really** need to optimize overhead even further, we could write a manual cpython binding.
I use this updated API in the next PR to fix FSDP2, so that it properly hides the VC of all `all_gather_buffer` tensors in its call to `split_with_sizes_copy.out(all_gather_buffers)`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/137921
Approved by: https://github.com/awgu, https://github.com/albanD
This PR implements a small UI improvement over #133603.
It prepares a NCCL memory allocator in torch cpp and then pybind's it out, so that user can directly use it.
UI:
```
pool = torch.cuda.MemPool(backend.mem_allocator)
with torch.cuda.use_mem_pool(pool):
tensor = torch.arange(1024 * 1024 * 2, device=device)
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/145675
Approved by: https://github.com/syed-ahmed, https://github.com/wconstab
This PR implements a small UI improvement over #133603.
It prepares a NCCL memory allocator in torch cpp and then pybind's it out, so that user can directly use it.
UI:
```
pool = torch.cuda.MemPool(backend.mem_allocator)
with torch.cuda.use_mem_pool(pool):
tensor = torch.arange(1024 * 1024 * 2, device=device)
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/145675
Approved by: https://github.com/syed-ahmed, https://github.com/wconstab
This PR implements a small UI improvement over #133603.
It prepares a NCCL memory allocator in torch cpp and then pybind's it out, so that user can directly use it.
UI:
```
pool = torch.cuda.MemPool(backend.mem_allocator)
with torch.cuda.use_mem_pool(pool):
tensor = torch.arange(1024 * 1024 * 2, device=device)
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/145675
Approved by: https://github.com/syed-ahmed, https://github.com/wconstab
Fixes#144976
Using appoach ① `IO[bytes]`, but could also try with a protocol.
## Notes:
- moved `torch.serialization.FILE_LIKE` to `torch.types.FileLike`
- Use `FileLike` annotation where it makes sense
- made sure those functions also support `os.PathLike`
- Replaced `isinstance(x, io.BytesIO)` with `isinstance(x, (io.IOBase, IO))` where appropriate.
- Replaced `BinaryIO` with `IO[bytes]` (the two ABCs are almost identical, the only difference is that `BinaryIO` allows `bytearray` input to `write`, whereas `IO[bytes]` only `bytes`)
- needed to make `torch.serialization._opener` generic to avoid LSP violations.
- skipped `torch/onnx/verification` for now (functions use `BytesIO.getvalue` which is not part of the `IO[bytes]` ABC, but it kind of seems that this is redundant, as e.g. `onnx.load` supports `str | PathLike[str] | IO[bytes]` directly...
Pull Request resolved: https://github.com/pytorch/pytorch/pull/144994
Approved by: https://github.com/ezyang, https://github.com/Skylion007
Summary:
This PR is basically a replacement of
https://github.com/pytorch/pytorch/pull/140087, which caused some perf
drop due to frequent TCPStore check in watchdog thread. The fix is to move the
tcpstore check in monitoring thread
If unhealthy, the user should be able to get the type of errors, e.g.,
timeout,nccl error or remote error.
This API is applied to PG level, compared to the
work.get_future_result() API which is applied to Work Level.
Error detection at PG level is much more convenient for users to handle
the PG failure as a whole, e.g, restarting the PG.
Error handling at the work level is still useful for users to attach
work specific context and debug the RC of the specific failing
work/collective
Note it is critical for all ranks in the PG to be notified about an
error as soon as it occurs, so we introduce an errorType of
REMOTE_ERROR, which is 'broadcasted' from a src rank (which detects a
local error) to all other ranks in the PG, the broadcast is done through
TCPStore currently
Tags:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/144498
Approved by: https://github.com/kwen2501
Fix: #141974
This PR makes `ViewMeta` sequence, present in functional tensors,
serializable with pickle. In order to accomplish that, it makes
`ViewMeta` an abstract class with overridable `forward` and `reverse`
functions. In this context, each operation that once instanciated
`ViewMeta`, should now create a new specialized class that inherits from
`ViewMeta. Therefore, this PR also uses codegen for creating these
specializations.
In summary, these are the changes this PR introduces:
- `ViewMeta` is turned into an abstract class (see
_FunctionalStorageImpl.cpp_). `forward` and `reverse` are pure virtual
functions that need to be implemented. `to_out_index` should be
implemented by operations that might return more than 1 output.
- New `ViewMeta` specializations for `resize_` and `_unsafe_view` are
created (see _FunctionalizeFallbackKernel.h_).
- New templates _ViewMetaClasses.{cpp,h}_ are created. They hold the
declaration and definition of the `ViewMeta` specializations, which
are automatically generated in the ATen codegen (see _gen.py_).
- New `_functionalization` Python sub-module is created (see
_Module.cpp_). It serves as namespace for the `ViewMeta`
specializations and `InverseReturnMode` enum.
- New template _ViewMetaClassesPythonBinding.cpp_ is created. It holds
the automatically generated Python bindings for the `ViewMeta`
specialization, which are generated in the torch codegen (see
_generate_code.py_).
Note that this PR makes use of codegen at 2 different moments:
- ATen codegen (_gen.py_): generates the `ViewMeta` specialized classes.
- Torch codegen (_generate_code.py_): generated the Python bindings for
them.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/143712
Approved by: https://github.com/bdhirsh
I.e. when `MTL_CAPTURE_ENABLED` environment variable is set to 1, one should be able to invoke wrap the code with `torch.mps.profiler.capture_metal` to generate gputrace for shaders invoked inside the context manager.
For example, code below:
```python
import torch
import os
def foo(x):
return x[:,::2].sin() + x[:, 1::2].cos()
if __name__ == "__main__":
os.environ["MTL_CAPTURE_ENABLED"] = "1"
x = torch.rand(32, 1024, device="mps")
with torch.mps.profiler.metal_capture("compiled_shader"):
torch.compile(foo)(x)
```
should capture the execution of a `torch.compile` generated shader
<img width="734" alt="image" src="https://github.com/user-attachments/assets/718ff64e-103b-4b11-b66c-c89cfc770b5d" />
Pull Request resolved: https://github.com/pytorch/pytorch/pull/144561
Approved by: https://github.com/manuelcandales
ghstack dependencies: #144559, #144560
Replace https://github.com/pytorch/pytorch/pull/138947 for re-import.
Replaces https://github.com/ROCm/pytorch/pull/1592
This PR contains the initial implementation of SDPA with composable_kernel backend. The CK path can be forced by simply calling torch.backends.cuda.preferred_rocm_fa_library("ck"). Similarly, you can force the incumbent aotriton implementation by passing in "aotriton" or "default". As you'd expect, not setting this option will result in aotriton to be used as the backend. In the case of CK, if pytorch deems flash attention usable, then it will use the CK path in all the same places aotriton would have been used. This PR makes no changes to the heuristics which select which attention scheme to use (i.e. flash attention vs memory efficient attention vs math etc etc). It only gets called when flash attention is both enabled (via USE_FLASH_ATTENTION) and is selected at runtime by the existing heuristics.
Files located in pytorch/aten/src/ATen/native/transformers/hip/flash_attn/ck/mha* have been pulled from https://github.com/Dao-AILab/flash-attention courtesy of @tridao's hard work who is the co-author
NOTE: In order to use this backend, the user MUST set USE_CK_FLASH_ATTENTION=1 in their environment when they build PyTorch.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/143695
Approved by: https://github.com/malfet
Co-authored-by: Andy Lugo <Andy.LugoReyes@amd.com>
Co-authored-by: Jithun Nair <jithun.nair@amd.com>
Description:
1. Quantize Linear Layer Weights to 4-bits:
Quantize the weights of the Linear layer to 4 bits, using symmetric quantization.
Pack two 4-bit weights into one uint8 container.
Choose a quantization scheme (channel-wise or group-wise), with the group size being a multiple of 32.
2. Prepare Quantized Weights, Scales, and Optional Bias:
After quantizing, obtain the quantized_weights, scales, and groupsize.
If the original Linear layer has a bias, prepare it as well.
3. Pack the Weights Efficiently:
Use torch.ops.aten._dyn_quant_pack_4bit_weight to optimally pack the weights, scales, and optional bias.
```python
packed_weights = torch.ops.aten._dyn_quant_pack_4bit_weight(weight, scales_and_zeros, bias, groupsize, in_features, out_features)
```
Input parameters should include:
in_features and out_features (the same as the Linear layer’s corresponding parameters).
4. Perform Dynamic Quantized Matrix Multiplication:
Use torch.ops.aten._dyn_quant_matmul_4bit to perform matrix multiplication with quantized weights.
```python
output = torch.ops.aten._dyn_quant_matmul_4bit(input, packed_weights, groupsize, in_features, out_features)
```
Inputs required include:
The input tensor, packed_weights , groupsize, and the in_features and out_features.
API Usage: https://github.com/pytorch/pytorch/issues/143289
Model Perf :
7B Transformer model:
Prefill : 340 t/s
Decode : 40 t/s
2B Transformer model
Prefill : 747 t/s
Decode : 80 t/s
Tests:
python test/test_linalg.py -k test__dyn_quant_pack_4bit_weight
Ran 1 test in 0.016s
OK
python test/test_linalg.py -k test__dyn_quant_matmul_4bit
Ran 8 tests in 0.077s
OK
python test/test_linalg.py -k test_compile_dyn_quant_matmul_4bit
Ran 8 tests in 11.454s
Change-Id: Ia1672bad5e6ec94e64d8bb1971395d60f4b3a452
Fixes #ISSUE_NUMBER
Pull Request resolved: https://github.com/pytorch/pytorch/pull/134124
Approved by: https://github.com/digantdesai, https://github.com/malfet
Fixes#130559
* Intro
This PR adds support for `@contextmanager` in Dynamo. We chose to limit the
scope of this work to only `@contextmanager` and plan to handle generators fully
in #141055 (still in draft).
* Motivation
Dynamo lacks support for generator functions. When it encounters one, it traces
it as if it were a regular function. This is problematic because it can lead to
incorrect behavior. To illustrate, consider the test case below:
```python
import torch
import contextlib
@contextlib.contextmanager
def set_default_dtype(dtype):
old_dtype = torch.get_default_dtype()
try:
torch.set_default_dtype(dtype)
yield
finally:
torch.set_default_dtype(old_dtype)
@torch.compile(backend="eager", fullgraph=True)
def fn():
with set_default_dtype(torch.float64):
x = torch.tensor([3.0, 3.0 + 5.0j])
return x
```
Before this work, Dynamo would not stop at the `yield`, and the graph produced
would contain both calls to `set_default_dtype` executed one after the other.
This is incorrect because the context manager should execute code before and
after the `yield`.
* List of changes
`YIELD_VALUE` now raises an exception (`YieldValueOp`) to signal that control
flow must be suspended and returned to the caller. Additionally, `RETURN_VALUE`
behaves differently in a generator function. Unlike regular functions, where
`RETURN_VALUE` indicates the final result, in generators it signifies that the
generator is exhausted and implicitly raises `StopIteration`.
A new `VariableTracker` named `FunctionDecoratedByContextlibContextManagerVariable`
was introduced to handle `@contextmanager`. This variable tracker acts not just
as a wrapper for the original function but also maintains an internal `tx`
(InstructionTranslator) object to suspend and return control flow to the parent
tracer when a `yield` is encountered.
* Corner cases
Returning a context manager from a compiled function is not supported. This
would require PyTorch to synchronize the generator state between Dynamo and the
interpreter. Any attempt to return it will result in an `IncorrectUsage`
exception.
Graph breaks require special handling as well. In the event of a graph break,
the frame associated with the context manager is skipped, and the context
manager runs in eager mode.
* This PR is breaking my code
There is a configuration flag (`enable_trace_contextlib`) that can be set to
`False` to disable tracing context managers. If this still causes crashes,
please revert this PR.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/136033
Approved by: https://github.com/zou3519
Description:
1. Quantize Linear Layer Weights to 4-bits:
Quantize the weights of the Linear layer to 4 bits, using symmetric quantization.
Pack two 4-bit weights into one uint8 container.
Choose a quantization scheme (channel-wise or group-wise), with the group size being a multiple of 32.
2. Prepare Quantized Weights, Scales, and Optional Bias:
After quantizing, obtain the quantized_weights, scales, and groupsize.
If the original Linear layer has a bias, prepare it as well.
3. Pack the Weights Efficiently:
Use torch.ops.aten._dyn_quant_pack_4bit_weight to optimally pack the weights, scales, and optional bias.
```python
packed_weights = torch.ops.aten._dyn_quant_pack_4bit_weight(weight, scales_and_zeros, bias, groupsize, in_features, out_features)
```
Input parameters should include:
in_features and out_features (the same as the Linear layer’s corresponding parameters).
4. Perform Dynamic Quantized Matrix Multiplication:
Use torch.ops.aten._dyn_quant_matmul_4bit to perform matrix multiplication with quantized weights.
```python
output = torch.ops.aten._dyn_quant_matmul_4bit(input, packed_weights, groupsize, in_features, out_features)
```
Inputs required include:
The input tensor, packed_weights , groupsize, and the in_features and out_features.
API Usage: https://github.com/pytorch/pytorch/issues/143289
Model Perf :
7B Transformer model:
Prefill : 340 t/s
Decode : 40 t/s
2B Transformer model
Prefill : 747 t/s
Decode : 80 t/s
Tests:
python test/test_linalg.py -k test__dyn_quant_pack_4bit_weight
Ran 1 test in 0.016s
OK
python test/test_linalg.py -k test__dyn_quant_matmul_4bit
Ran 8 tests in 0.077s
OK
python test/test_linalg.py -k test_compile_dyn_quant_matmul_4bit
Ran 8 tests in 11.454s
Change-Id: Ia1672bad5e6ec94e64d8bb1971395d60f4b3a452
Fixes #ISSUE_NUMBER
Pull Request resolved: https://github.com/pytorch/pytorch/pull/134124
Approved by: https://github.com/digantdesai, https://github.com/malfet
Description:
1. Quantize Linear Layer Weights to 4-bits:
Quantize the weights of the Linear layer to 4 bits, using symmetric quantization.
Pack two 4-bit weights into one uint8 container.
Choose a quantization scheme (channel-wise or group-wise), with the group size being a multiple of 32.
2. Prepare Quantized Weights, Scales, and Optional Bias:
After quantizing, obtain the quantized_weights, scales, and groupsize.
If the original Linear layer has a bias, prepare it as well.
3. Pack the Weights Efficiently:
Use torch.ops.aten._dyn_quant_pack_4bit_weight to optimally pack the weights, scales, and optional bias.
```python
packed_weights = torch.ops.aten._dyn_quant_pack_4bit_weight(weight, scales_and_zeros, bias, groupsize, in_features, out_features)
```
Input parameters should include:
in_features and out_features (the same as the Linear layer’s corresponding parameters).
4. Perform Dynamic Quantized Matrix Multiplication:
Use torch.ops.aten._dyn_quant_matmul_4bit to perform matrix multiplication with quantized weights.
```python
output = torch.ops.aten._dyn_quant_matmul_4bit(input, packed_weights, groupsize, in_features, out_features)
```
Inputs required include:
The input tensor, packed_weights , groupsize, and the in_features and out_features.
API Usage: https://github.com/pytorch/pytorch/issues/143289
Model Perf :
7B Transformer model:
Prefill : 340 t/s
Decode : 40 t/s
2B Transformer model
Prefill : 747 t/s
Decode : 80 t/s
Tests:
python test/test_linalg.py -k test__dyn_quant_pack_4bit_weight
Ran 1 test in 0.016s
OK
python test/test_linalg.py -k test__dyn_quant_matmul_4bit
Ran 8 tests in 0.077s
OK
python test/test_linalg.py -k test_compile_dyn_quant_matmul_4bit
Ran 8 tests in 11.454s
Change-Id: Ia1672bad5e6ec94e64d8bb1971395d60f4b3a452
Fixes #ISSUE_NUMBER
Pull Request resolved: https://github.com/pytorch/pytorch/pull/134124
Approved by: https://github.com/digantdesai, https://github.com/malfet
Implements https://github.com/pytorch/pytorch/issues/93753 - move frame local guard accessors to C++.
Before, we used dict accessors on a Python dict representing the frame's fastlocals that we manually build. We move this accessor to C++ and additionally use the fastlocal index whenever possible.
Some implementation notes:
- `FrameLocalsMapping` is now initialized as a C++ vector of `PyObject`s. We do not just use the frame's localsplus/fastlocals buffer because we also unbox cells.
- `FrameLocalsMapping` can still be converted into a Python dict representing the frame's fastlocals, but it is done lazily.
- We update `LeafGuard`, `GuardAccessor`, and `GuardManager`'s `check_nopybind` methods to accept `FrameLocalsMapping`. By default, we convert the `FrameLocalsMapping` to a Python dict and run the original `check_nopybind` on it, but in some cases, conversion is not needed.
- We add a new guard accessor `FrameLocalsGuardAccessor`, which is similar to `DictGetItemGuardAccessor` but has special handling for `FrameLocalsMapping`. We create a separate class to emphasize different use cases, but we could probably combine these two (can do in a follow up)
dynamo_guard_eval.py microbenchmark update:
- 713.2us -> 630.0us (3.10)
- 598.8us -> 530.7us (3.12)
Other followups:
- Add `FrameLocalsMapping` version for `check_verbose_nopybind` in order to match behavior between `check_nopybind` and `check_verbose_nopybind`. This can prevent difficult debugging situations where guards fail (`check_nopybind` returns false) but no guard error message is generated (`check_verbose_nopybind` succeeds).
- Rewrite the `SHAPE_ENV` guard into C++ - it is a fairly common guard that results in `FrameLocalsMapping` needing to convert to a dict
Pull Request resolved: https://github.com/pytorch/pytorch/pull/140063
Approved by: https://github.com/jansel
ghstack dependencies: #142117, #142430
Replaces https://github.com/ROCm/pytorch/pull/1592
This PR contains the initial implementation of SDPA with composable_kernel backend. The CK path can be forced by simply calling `torch.backends.cuda.preferred_rocm_fa_library("ck")`. Similarly, you can force the incumbent aotriton implementation by passing in "aotriton" or "default". As you'd expect, not setting this option will result in aotriton to be used as the backend. In the case of CK, if pytorch deems flash attention usable, then it will use the CK path in all the same places aotriton would have been used. This PR makes no changes to the heuristics which select which attention scheme to use (i.e. flash attention vs memory efficient attention vs math etc etc). It only gets called when flash attention is both enabled (via `USE_FLASH_ATTENTION`) and is selected at runtime by the existing heuristics.
Files located in pytorch/aten/src/ATen/native/transformers/hip/flash_attn/ck/mha* have been pulled from https://github.com/Dao-AILab/flash-attention courtesy of @tridao's hard work who is the co-author
NOTE: In order to use this backend, the user MUST set USE_CK_FLASH_ATTENTION=1 in their environment when they build PyTorch.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/138947
Approved by: https://github.com/pruthvistony, https://github.com/xw285cornell, https://github.com/leitian
Co-authored-by: Xiaodong Wang <xw285@cornell.edu>
TunableOp's rotating buffer feature cannot be properly tested because the environment variable that controls this feature is sticky. A Python API is introduced to modify this value.
Additional items in this PR:
* UT for rotating buffer API
* Clean up UTs that were setting the rotating buffer via the environment variable
* Align behavior of environment variable and Python API when a negative value (< 0) is set.
* Update documentation.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/143172
Approved by: https://github.com/jeffdaily
### Motivation:
As design illustrated in Intel distributed support RFC https://github.com/pytorch/pytorch/issues/141741, two sections are needed to enable intel distributed backend (`XCCL`) support in PyTorch.
1. Intel GPU distributed Backend integration in PyTorch `torch-xpu-ops`.
2. **Intel distributed Backend register in PyTorch distributed package**. This PR is to contribute section 2 change.
### Example:
Here is a simple example of using spawn to launch XCCL backend and perform allreduce on XPU tensors.
```
import os
import torch
import torch.distributed as dist
import torch.multiprocessing as mp
def setup(rank, world_size):
os.environ['MASTER_ADDR'] = 'localhost'
os.environ['MASTER_PORT'] = '29500'
dist.init_process_group(rank=rank, world_size=world_size)
def cleanup():
dist.destroy_process_group()
def run_allreduce(rank, world_size):
setup(rank, world_size)
device = torch.device('xpu:{}'.format(rank))
x = torch.randn([2, 2], device=device)
dist.all_reduce(x)
cleanup()
if __name__ == '__main__':
world_size = 2
mp.spawn(run_allreduce, args=(world_size,), nprocs=world_size, join=True)
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/141856
Approved by: https://github.com/kwen2501, https://github.com/gujinghui, https://github.com/albanD
This PR moves the logic for computing the overlapping relations between input tensors that
share a storage instance to C++.
In summary, this PR:
- Moves both `tensors_definitely_do_not_overlap` and part of `compute_overlapping_tensors`
to C++
- Introduces a `check_overlapping` function that re-runs `compute_overlapping_tensors`,
checking that the result is consistent with what is expected
- Introduces the `StorageOverlapChecker` class
- Keeps track of overlapping and non-overlapping tensors
- Actually checks the overlapping relation (call `check_overlapping`) when all tensors
are collected
- Introduces the `STORAGE_OVERLAPPING` relational guard
- Has a reference to a `StorageOverlapChecker`
- Stores the to-be-checked tensors in the checker, and triggers its check
- Introduces `install_storage_overlapping_guard` python function
- Creates an instance of `StorageOverlapChecker`
- Creates 2 instances of the `STORAGE_OVERLAPPING` guard (for overlapping and
non-overlapping tensors), referencing the same `StorageOverlapChecker` instance
**Why is `StorageOverlapChecker` needed?**
The way `GuardManager` is implemented, we have no control over the order in which the
check methods are called, i.e. no control over the order the tensors are collected. So, we
can't easily split them in "overlapping" and non-overlapping kinds.
Instead, we create 2 instances of `STORAGE_OVERLAPPING` guard, each of which helps
collecting the tensors for one of the kinds mentioned above. They are then used in a
single `StorageOverlapChecker` instance.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/140013
Approved by: https://github.com/bdhirsh
ghstack dependencies: #139554, #139555
Certain `cpp_wrapper`-enabled tests were OOM-ing in the CI pipeline, with error messages suggesting that sufficient memory was accessible. This ultimately resulted from an internal memory limitation that was not queryable in the API. This PR adds querying for that limit.
Additionally, the failing tests had incorrect memory availability checks, and are updated with measured memory requirements.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/140620
Approved by: https://github.com/malfet, https://github.com/eqy
ghstack dependencies: #141367
Before this PR, users need to call `empty_strided_p2p()` with a `group_name`:
```python
tensor = _SymmetricMemory.empty_strided_p2p((1024,), (1,), device=device, group_name="0")
symm_mem = _SymmetricMemory.rendezvous(tensor)
```
Users can now omit `group_name` at allocation time and specify it later at rendezvous time:
```python
tensor = _SymmetricMemory.empty_strided_p2p((1024,), (1,), device=device)
symm_mem = _SymmetricMemory.rendezvous(tensor, group_name="0")
```
Rationales for this change:
- This allows the same allocation to establish symmetric memory under different groups
- Specifying `group_name` at rendezvous time instead of allocation time is a more natural UX
Pull Request resolved: https://github.com/pytorch/pytorch/pull/139529
Approved by: https://github.com/lw
In `match_nested_cell`, Dynamo tried to identify pre-existing captured
cells by `(cell_name, id(cell_contents))`. This works in most cases, but
as the test added in this patch shows, it's not a complete solution.
This patch
1. changes `match_nested_cell` to `lookup_variable_for_captured_cell`,
and does the lookup based on id of cell objects, not their contents.
This requires plumbing a tuple of captured cell objects from
different CPython versions all the way to
`InstructionTranslator.__init__`, where we store a mapping from the
ids of these cell objects, and use it later in
`UserFunctionVariable.bind_args` to look for these unboxed cells.
2. builds off (1) -- rather than using a `VariableTracker` that
represents the content of the unboxed cells, use `ClosureVariable`,
which enables codegen in case these cells escape as closure of a
`NestedUserFunctionVariable`.
The patch adds a regression test for each of the scenarios above:
1. `test_write_to_cells_with_name_shadowing` where Dynamo mistakenly
thought the program is writing to a cell captured by root frame (which
it doesn't support atm), which resulted in
```
File "/Users/ryanguo99/Documents/work/pytorch/torch/_dynamo/symbolic_convert.py", line 3340, in STORE_DEREF
unimplemented("write to __closure__ while inlining")
File "/Users/ryanguo99/Documents/work/pytorch/torch/_dynamo/exc.py", line 313, in unimplemented
raise Unsupported(msg, case_name=case_name)
torch._dynamo.exc.Unsupported: write to __closure__ while inlining
```
2. `test_existing_func_that_creates_capturing_nested_func` where Dynamo
ended up trying to codegen a `NestedUserFunctionVariable` that
captures a cell which was also captured by the root frame, so it was
unboxed and ends up emitting `LOAD_DEREF` rather than
`LOAD_FAST/LOAD_CLOSURE` during codegen, resulting in
```
File "/Users/ryanguo99/Documents/work/pytorch/torch/_dynamo/variables/functions.py", line 105, in _create_nested_fn
func = FunctionType(code, f_globals, name, defaults, closure)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: arg 5 (closure) expected cell, found int
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/140436
Approved by: https://github.com/jansel, https://github.com/williamwen42
ghstack dependencies: #140330, #140152
This patch introduces a `DynamoFrameType` to serve as a layer between
Dynamo and different versions of Python frame object. In
`DynamoFrameType`, we only register attributes Dynamo cares about (e.g.,
`f_code`, `f_locals`, etc.
This will be helpful when it comes to adding new attributes to this
`DynamoFrameType`, or dealing with Python version changes.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/140330
Approved by: https://github.com/jansel, https://github.com/williamwen42
Summary:
If unhealthy, the user should be able to get the type of errors, e.g.,
timeout,nccl error or remote error.
This API is applied to PG level, compared to the work.get_future_result() API which is applied to Work Level.
Error detection at PG level is much more convenient for users to handle the PG failure as a whole, e.g, restarting the PG.
Error handling at the work level is still useful for users to attach work specific context and debug the RC of the specific failing work/collective
Note it is critical for all ranks in the PG to be notified about an error as soon as it occurs, so we introduce an errorType of REMOTE_ERROR, which is 'broadcasted' from a src rank (which detects a local error) to all other ranks in the PG, the broadcast is done through TCPStore currently
Tags:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/140087
Approved by: https://github.com/kwen2501
Differential Revision: D63206258
This diff introduces a mechanism to generate a json-compatible deserializer in cpp using nlohmann json (already being used by AOTI).
Why we need this? Because there will be a lot of cases where people don't want to use Python to load the graph (e.g. cpp runtime), and instead they can use this header to deserialize the JSON graph.
Every time we call update_schema.py to update the schema, the header will be auto generated and included into the source files.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/136398
Approved by: https://github.com/angelayi
This PR updates the binding for `stream_write_value32` to be consistent with `memset32` which IMO makes more sense for this type of utilities:
- Changed the API to take a uint32 tensor as argument, instead of a device pointer
- Changed the Python binding to be a static method of `_SymmetricMemory`, instead of a object method
- Use the dispatcher for device dispatching, as opposed to `SymmetricMemory` backends
Pull Request resolved: https://github.com/pytorch/pytorch/pull/139934
Approved by: https://github.com/weifengpy
ghstack dependencies: #139227
This PR updates the binding for `stream_write_value32` to be consistent with `memset32` which IMO makes more sense for this type of utilities:
- Changed the API to take a uint32 tensor as argument, instead of a device pointer
- Changed the Python binding to be a static method of `_SymmetricMemory`, instead of a object method
- Use the dispatcher for device dispatching, as opposed to `SymmetricMemory` backends
Pull Request resolved: https://github.com/pytorch/pytorch/pull/139934
Approved by: https://github.com/weifengpy
ghstack dependencies: #139227
This PR introduces the following:
### torch.ops.symm_mem._async_input_mm
`_async_input_mm(Tensor a, Tensor b, Tensor a_chunk_signals, int a_chunk_pivot) -> Tensor`
An mm impl that supports consuming asynchronous input. It guarantees the following rasterization order, and that the corresponding signal arrives before an input chunk is consumed.
```
num_chunks = a_chunks_signals.numel()
for chunk_idx in range(a_chunk_pivot, num_chunks + a_chunk_pivot):
chunk_idx = chunk_idx % num_chunks
wait_signal(a_chunk_signals, chunk_idx)
# Compute output tiles that consumes the input chunk
```
### PersistentAsyncInputScheduler
This is a forked version of PersistentScheduler that supports consuming asynchronous input. This tile scheduler introduces the following arguments:
- `tiles_per_chunk_m` – Specifies the size of an M chunk. Chunks are the granularity at which the asynchronous input becomes ready. It must be an interger multiple of the size of an M tile.
- `chunk_signals` – `chunk_signals[i] == 1` indicates that chunk i is ready. Before returning a work tile, get_current_work() waits for the signal to ensure that the corresponding chunk is ready.
- `tile_idx_pivot_m` – After applying swizzling, apply `pivot(m) => (m + tile_idx_pivot_m) % tiles_m` to `m`. In a distributed setting, this allows different ranks to process different m indices at the same time, thus avoiding communication hotspots.
Note that this scheduler currently only supports the `KernelTmaWarpSpecializedCooperative` kernel schedule. This is enforced via the template argument `KernelSchedule`.
Usage:
```
using GemmKernel = cutlass::gemm::kernel::GemmUniversal<
Shape<int, int, int, int>,
CollectiveMainloop,
CollectiveEpilogue,
cutlass::gemm::PersistentAsyncInputScheduler<KernelSchedule>>;
```
### _fused_all_gather_matmul_native
An ag-mm impl that combines `torch.ops.symm_mem._async_input_mm` and progress-aware all-gather. This is not yet enabled via the async-tp passes. We will use it as a backend to optimize the current decomposition-based async-tp impl.
## Benchmarks
### 4096x3584x8192
- cublas + nccl: 539us
- decomp-based async-tp w/o cuda graph: 694us
- decomp-based async-tp w/ cuda graph: 478us
- new cutlass kernel: 408us
<img width="478" alt="image" src="https://github.com/user-attachments/assets/39f316ab-36c5-4b41-af77-07854a385dfc">
### 2048x3584x8192
- cublas + nccl: 301us
- decomp-based async-tp w/o cuda graph: 687us
- decomp-based async-tp w/ cuda graph: 356us
- new cutlass kernel: 276us
<img width="441" alt="image" src="https://github.com/user-attachments/assets/9e23ce21-863b-43dd-a562-fb05d3a5a144">
## Next Steps
- Add tuning logic
- Use `_fused_all_gather_matmul_native` as a backend for the decomp-based async-tp impl
Differential temp Revision: [D65623152](https://our.internmc.facebook.com/intern/diff/D65623152)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/139227
Approved by: https://github.com/weifengpy, https://github.com/Chillee
This PR introduces the following:
### torch.ops.symm_mem._async_input_mm
`_async_input_mm(Tensor a, Tensor b, Tensor a_chunk_signals, int a_chunk_pivot) -> Tensor`
An mm impl that supports consuming asynchronous input. It guarantees the following rasterization order, and that the corresponding signal arrives before an input chunk is consumed.
```
num_chunks = a_chunks_signals.numel()
for chunk_idx in range(a_chunk_pivot, num_chunks + a_chunk_pivot):
chunk_idx = chunk_idx % num_chunks
wait_signal(a_chunk_signals, chunk_idx)
# Compute output tiles that consumes the input chunk
```
### PersistentAsyncInputScheduler
This is a forked version of PersistentScheduler that supports consuming asynchronous input. This tile scheduler introduces the following arguments:
- `tiles_per_chunk_m` – Specifies the size of an M chunk. Chunks are the granularity at which the asynchronous input becomes ready. It must be an interger multiple of the size of an M tile.
- `chunk_signals` – `chunk_signals[i] == 1` indicates that chunk i is ready. Before returning a work tile, get_current_work() waits for the signal to ensure that the corresponding chunk is ready.
- `tile_idx_pivot_m` – After applying swizzling, apply `pivot(m) => (m + tile_idx_pivot_m) % tiles_m` to `m`. In a distributed setting, this allows different ranks to process different m indices at the same time, thus avoiding communication hotspots.
Note that this scheduler currently only supports the `KernelTmaWarpSpecializedCooperative` kernel schedule. This is enforced via the template argument `KernelSchedule`.
Usage:
```
using GemmKernel = cutlass::gemm::kernel::GemmUniversal<
Shape<int, int, int, int>,
CollectiveMainloop,
CollectiveEpilogue,
cutlass::gemm::PersistentAsyncInputScheduler<KernelSchedule>>;
```
### _fused_all_gather_matmul_native
An ag-mm impl that combines `torch.ops.symm_mem._async_input_mm` and progress-aware all-gather. This is not yet enabled via the async-tp passes. We will use it as a backend to optimize the current decomposition-based async-tp impl.
## Benchmarks
### 4096x3584x8192
- cublas + nccl: 539us
- decomp-based async-tp w/o cuda graph: 694us
- decomp-based async-tp w/ cuda graph: 478us
- new cutlass kernel: 408us
<img width="478" alt="image" src="https://github.com/user-attachments/assets/39f316ab-36c5-4b41-af77-07854a385dfc">
### 2048x3584x8192
- cublas + nccl: 301us
- decomp-based async-tp w/o cuda graph: 687us
- decomp-based async-tp w/ cuda graph: 356us
- new cutlass kernel: 276us
<img width="441" alt="image" src="https://github.com/user-attachments/assets/9e23ce21-863b-43dd-a562-fb05d3a5a144">
## Next Steps
- Add tuning logic
- Use `_fused_all_gather_matmul_native` as a backend for the decomp-based async-tp impl
Pull Request resolved: https://github.com/pytorch/pytorch/pull/139227
Approved by: https://github.com/weifengpy, https://github.com/Chillee
Summary:
This PR adds Auto-Trace implementation for Trace ID. By default, the python side will generate a uuid in the same format as the one set in the backend by kineto. Upon running an auto-trace, the python generated trace id will overwrite the one set in kineto using the Config variable. Since we don't expect users to generate on-demand traces after an auto-trace we can simply keep overwriting the backend trace id whenever autotrace is ran. If we one day want to eventually do something like this, we simply have to add a call in kineto on the backend to generate a new ID upon start of profiling.
We also implement a custom callback in the frontend such that users can generate their own trace ids if they wish to. This works similarly as the default, only difference being that they have to manually set this callback after a profiler is generated. We use a specific call to set this rather then putting it in the frontend initializer in case users want to change the trace_id for different repeats.
Test Plan: Tested both default and custom callbacks using the verbose prints added. Trace ids on the frontend and the prints on the backend for the manifold upload matched.
Differential Revision: D65178308
Pull Request resolved: https://github.com/pytorch/pytorch/pull/139310
Approved by: https://github.com/shengfukevin
When we see a custom op:
- check that its mutation annotations are correct
- check that its aliasing constraints matches our constraints for custom
ops.
Otherwise, there may be undefined behavior.
Test Plan:
- new tests
Pull Request resolved: https://github.com/pytorch/pytorch/pull/139212
Approved by: https://github.com/angelayi
This PR aims to support the following use case:
```python
def all_reduce_eager(x):
y = x * x
req = dist.all_reduce(y, op=dist.ReduceOp.SUM, async_op=True)
assert isinstance(req, torch.distributed.Work)
return y
@torch.compile(fullgraph=True)
def all_reduce_wait_compiled(y):
torch.ops.c10d_functional.wait_tensor(y)
return y * y
x = torch.ones(1280, 1280, device="cuda") + self.rank
with allow_inflight_collective_as_graph_input_ctx():
y = all_reduce_eager(x)
z = all_reduce_wait_compiled(y)
```
where the collective is issued in eager (with `async_op=True`) but waited in compiled region.
This is important for internal use cases such as TorchRec, where we issue collectives in eager for SparseArch all_to_all but want to wait for them in compiled region at beginning of OverArch, so that the all_to_all can be overlapped with the DenseArch compute that runs in parallel.
----
**Update**: Did two items to prevent regression to existing use cases:
1. Added memory-stressed test case to test_c10d_nccl.py `test_unwaited` to cover existing user's "not calling work.wait() for non-functional collective" use case
2. Gated all new `register_work()` / `unregister_work()` calls with `c10d::allow_inflight_collective_as_graph_input()` check, which is a new context manager that requires explicit user enablement (i.e. not on by default, so should not affect existing users).
The risk of this new version of PR causing regression should be very low.
------
Test commands:
- `pytest -rA test/distributed/test_inductor_collectives.py::TestCollectivesMultiProc::test_eager_async_allreduce_inductor_wait`
- `pytest -rA test/test_fx.py::TestDCE::test_keep_collectives`
- `pytest -rA test/test_fx.py::TestDCE::test_keep_collectives_no_overload`
- `pytest -rA test/distributed/test_c10d_functional_native.py::TestWithNCCL::test_wait_tensor`
- `pytest -rA test/distributed/test_c10d_functional_native.py::TestWithNCCL::test_unwaited`
- `pytest -rA test/distributed/test_c10d_nccl.py::CommTest::test_wait_tensor`
- `pytest -rA test/distributed/test_c10d_nccl.py::CommTest::test_unwaited`
- `pytest -rA test/distributed/_tensor/test_tensor_ops.py::DistTensorOpsTest::test_equal`
- `pytest -rA test/distributed/_tensor/test_random_ops.py::DistTensorRandomOpTest::test_manual_seed`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_ddp_baseline_aot_eager_multiprocess`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_fsdp_aot_eager`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_fsdp_setattr`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_fsdp_unspecialized_forced_getattr_inline`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_fsdp_unspecialized_forced_getattr_no_inline`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_asymmetric_compilation`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_compiler_collectives_automatic_dynamic_scalar`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_compiler_collectives_automatic_dynamic_speculation_divergence`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_compiler_collectives_automatic_dynamic_tensor`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_compiler_collectives_dim_mismatch`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_compiler_collectives_graph_break_empty_graph_still_collective`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_compiler_collectives_missing_source`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_compiler_collectives_scalar_missing_source`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_compiler_collectives_type_mismatch`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_ddp_activation_checkpointing`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_ddp_baseline_aot_eager_multiprocess`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_fsdp_activation_checkpointing`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_fsdp_aot_eager`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_fsdp_inductor`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_fsdp_setattr`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_fsdp_unspecialized_forced_getattr_inline`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_fsdp_unspecialized_forced_getattr_no_inline`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_hf_bert_ddp_aot_eager`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_hf_bert_ddp_aot_eager_static_graph`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_hf_bert_ddp_inductor`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_hf_bert_ddp_inductor_static_graph`
- `pytest -rA test/distributed/test_dynamo_distributed.py::TestMultiProc::test_hf_bert_fsdp_activation_checkpointing`
- `pytest -rA test/distributed/_tensor/test_experimental_ops.py::DistOtherOpsTest::test_bernoulli`
- `pytest -rA test/distributed/_tensor/test_dtensor_compile.py::TestDTensorCompileE2E::test_tp_compile_fullgraph_is_seq_parallel_True`
- `pytest -rA test/distributed/test_inductor_collectives.py::TestCollectivesMultiProc::test_allreduce_inductor_cudagraph_trees`
- `python benchmarks/dynamo/torchbench.py --ci --accuracy --timing --explain --inductor --device cuda --inference --bfloat16 --total-partitions 2 --partition-id 1 --output inference_torchbench.csv --only moco`
------
Differential Revision: [D65023311](https://our.internmc.facebook.com/intern/diff/D65023311)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/137763
Approved by: https://github.com/yifuwang
This PR enables you to inspect PyObjects in C using `INSPECT(...)` without requiring https://docs.python.org/3/howto/gdb_helpers.html. `torch._dynamo.eval_frame.raise_sigtrap` can also be used to set gdb breakpoints while running Python code, e.g.
```python
x = x + 1
torch._dynamo.eval_frame.raise_sigtrap();
# can breakpoint on ceval.c:CALL to breakpoint the `sin` call in C.
x = torch.sin(x)
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/138030
Approved by: https://github.com/jansel
# Motivation
According to [[RFC]A device-agnostic Python runtime API design for stream-based accelerators](https://github.com/pytorch/pytorch/issues/128403), this PR intends to introduce a device-agnostic runtime API design.
I personally prefer the **Simple Version** APIs that no longer accept the device type as an input argument. It means we will leverage `getAccelerator` to fetch the current accelerator. And it is flexible to expand these APIs to handle multiple types of accelerator scenarios. The design does **NOT** break the previous design philosophies.
I also believe that namespace torch.accelerator is better. It lets users know that the APIs they are calling are running on an accelerator rather than CPU. This is important. Meanwhile, we can follow a simple API design principle:
1. Device-agnostic APIs should be placed under the torch.accelerator namespace and not accept a device_type optional parameter.
2. Device-specific APIs should be placed under device-specific submodules.
3. APIS required by both CPU and accelerators should be placed under the torch namespace and accept a device_type optional parameter.
Also, I list the pros and cons of **Simple Version** here:
Pros:
- `torch.accelerator.foo` will have the same input argument as `torch.xxx.foo`, bringing a better user experience;
- more concise, facilitate the developer to write a device-agnostic code.
Cons:
- no obvious drawbacks.
# Additional Context
I list the new APIs here:
```python
torch.accelerator.is_available() -> bool:
torch.accelerator.current_accelerator() -> torch.device:
torch.accelerator.device_count() -> int:
torch.accelerator.current_device_idx() -> int:
torch.accelerator.set_device_idx(device: Union[torch.device, str, int, None]) -> None:
torch.accelerator.current_stream(device: Union[torch.device, str, int, None]) -> torch.Stream:
torch.accelerator.set_stream(stream: torch.Stream) -> None:
torch.accelerator.synchronize(device: Union[torch.device, str, int, None]) -> None:
```
According to the discussion with Alban, we decide to change the API name `set_device` to `set_device_idx` and `current_device` to `current_device_idx` for more explicit. And will submit other PR to support device and stream context manager.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/132204
Approved by: https://github.com/EikanWang, https://github.com/abhilash1910, https://github.com/gujinghui, https://github.com/albanD
This PR refactors some ref-counting functionality out of `beginAllocateToPool` and `releasePool`. The ref-counting logic is then used in construction and destruction of `torch.cuda.MemPool`.
The `use_count` variable in the CUDACachingAllocator is essentially a refcount of how many context managers are using the pool. Since we are now lifting up the MemPool abstraction to the user, the MemPool object itself now needs to hold a an extra reference as well.
Part of https://github.com/pytorch/pytorch/issues/124807.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/133600
Approved by: https://github.com/eqy, https://github.com/ezyang
- composable_kernel as a third_party submodule
- "ck" as a `torch.backends.cuda.preferred_linalg_library()`
- reference CK gemm implementations for float, bfloat16, and half types
Pull Request resolved: https://github.com/pytorch/pytorch/pull/131004
Approved by: https://github.com/xw285cornell, https://github.com/pruthvistony
Co-authored-by: Andres Lugo <Andy.LugoReyes@amd.com>
Co-authored-by: Pruthvi Madugundu <pruthvigithub@gmail.com>
Thanks @eqy for reminding me of this RFC: https://github.com/pytorch/pytorch/issues/119797
This PR is meant to:
- provide a way to abort multiple PGs without deadlocking each other.
- provide a possibility to manually handle comm errors or timeouts (and potentially recovery of such).
One can find an example from: https://github.com/NVIDIA/nccl/issues/1013
## How is it different from `destroy_process_group`?
`destroy_process_group` is meant for normal exit, while `_abort_process_group` is meant for bailout upon hangs or failures. Similar to `ncclCommDestroy` vs `ncclCommAbort`.
## What's new in `_abort_process_group`?
It added support for "group abort" semantic. The "group abort" semantic is capable of aborting multiple NCCL comms concurrently, avoiding deadlock in otherwise serialized `ncclCommAbort` executions. Details are in the [RFC](https://github.com/pytorch/pytorch/issues/119797) targeting [the hang issue in multi-comm case](https://github.com/NVIDIA/nccl/issues/1013). `Group abort` semantic is added in NCCL 2.22.
## What's next?
Ideally, the watchdog's behavior should support "group abort" too. But this is hard to implement today due to a lack of "global view" by each PG's individual watchdog. A big semi-big refactor may be needed to "uplift" the watchdogs to a global level or consolidate them into one (i.e. one dog watching multiple PGs).
In any case, it may not be a bad idea to experiment the "group abort" feature with a manual API first and then extend to the automatic mode (watchdog).
Pull Request resolved: https://github.com/pytorch/pytorch/pull/132291
Approved by: https://github.com/eqy
**Motivation**
Enable SVE vectorization with `torch.compile`
Extends PR: #119571
* This PR enables vectorization for codegen part using SVE-256 (vec length)
* The changes can be extended to other SVE vec lengths
I've done some comparisons against existing NEON implementation with SVE vectorization enabled route for `torch.compile`
Test results are for 8 cores on ARM Neoverse_V1
<img width="359" alt="Screenshot 2024-08-28 at 16 02 07" src="https://github.com/user-attachments/assets/6961fbea-8285-4ca3-b92e-934a2db50ee2">
It's worth mentioning, for standalone `SiLU op` there's a `~1.8x` speedup with `torch.compile`
Pull Request resolved: https://github.com/pytorch/pytorch/pull/134672
Approved by: https://github.com/jgong5, https://github.com/malfet
This reverts commit 7743149b2b.
Reverts
* https://github.com/pytorch/pytorch/pull/135503
* https://github.com/pytorch/pytorch/pull/135502
* https://github.com/pytorch/pytorch/pull/135422
This passes this test. Earlier, the getitem would stay like a getitem in the Fx graph. But now the fake tensor propagations fails saying that .item is called. It seems that torch function is not getting triggered while fake tensor propagation.
```
import torch
from torch.nn.attention.flex_attention import BlockMask, _mask_mod_signature, _score_mod_signature, flex_attention
from torch._inductor.lowering import make_pointwise, register_lowering
from torch._inductor.virtualized import ops
from torch.nn.attention.flex_attention import create_block_mask
torch.set_default_device('cuda')
flex_attention = torch.compile(flex_attention, dynamic=False)
prefix_lengths = torch.arange(8)
def prefix_lm(b, h, q, kv):
return prefix_lengths[b] >= kv
mask = create_block_mask(prefix_lm, 8, None, 512, 512, _compile=True)
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/136590
Approved by: https://github.com/Chillee
We introduced the dispatchable backend for a ProcessGroup and collective in https://github.com/pytorch/pytorch/issues/86225. This PR is a follow-up cleanup to clean up the option of a ProcessGroup and ask users to either set timeout or backend later on or directly create backend after creating a PG.
Also PGNCCL is using option class from ProcessGroup but we actually should use Option from backend class. So this PR is to make the type or name to be aligned with what we are doing in cpp side. I don't change the signature for the public API, so they still use args named "pg_options"
We need to make changes to the test to make it aligned with the change.
This is try to reland D62008954 by fixing internal errors.
Differential Revision: [D62483294](https://our.internmc.facebook.com/intern/diff/D62483294/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/135653
Approved by: https://github.com/wz337, https://github.com/H-Huang
* Added a cpp loader, AOTIModelPackageLoader, which can load the .pt2, build the .so, and create a runner. The python-facing API is that users can directly call the `run` function, whereas in cpp users can directly access the `runner_` if they are more familiar with that. I couldn't figure out how to bind the `get_runner()` function to python...
* Added a new config, `aot_inductor.package_cpp_only` which will **not** package the so. This means that whenever the package is loaded, we will need to build the so. This is turned off by default so that new environments do not need to rebuild their so. The `package_cpp_only` is a feature which torchchat intends to use to provide flexibility to users.
* Added a new config, `aot_inductor.metadata` which stores user-provided metadata, serialized to the pt2 as a json file. It also stores the device used when exporting, "cuda" or "cpu", so that during load time, we can use that data to determine which AOTIModelContainerRunner to use. The metadata can be accessed through `loader.get_metadata()`. TODO is to move this metadata to the toplevel `package_aoti` function so that we can remove the metadata as a config.
* Separated out `package_aoti` as a standalone function, instead of it automatically being called in inductor. This is to prepare for the case where users will compile multiple models, and want to bundle it in one package. The specific use case is in torchchat, where we want to package the separately-exported encoder and decoder layers. An example of how to use this is in `test_multiple_methods`.
* `load_package` will load a singular model, given the model name.
* The loader doesn't support windows for now, I think I need to add some more casing to make the build commands work on windows?
Differential Revision: [D62329906](https://our.internmc.facebook.com/intern/diff/D62329906)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/135374
Approved by: https://github.com/desertfire, https://github.com/malfet
We introduced the dispatchable backend for a ProcessGroup and collective in https://github.com/pytorch/pytorch/issues/86225. This PR is a follow-up cleanup to clean up the option of a ProcessGroup and ask users to either set timeout or backend later on or directly create backend after creating a PG.
Also PGNCCL is using option class from ProcessGroup but we actually should use Option from backend class. So this PR is to make the type or name to be aligned with what we are doing in cpp side. I don't change the signature for the public API, so they still use args named "pg_options"
We need to make changes to the test to make it aligned with the change.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/132931
Approved by: https://github.com/H-Huang
PYTHONPATH=$(pwd) python benchmarks/update_hint_benchmark.py out
as of this diff, compile_time_instruction_count counts the number of instruction from within
convert_frame.compile_inner
```
update_hint_regression,compile_time_instruction_count,10522459165
```
will add result from CI once populated.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/133834
Approved by: https://github.com/aorenste
Summary:
We want to add compile IDs and frames to each Torch-Compiled Region in order to help users cross reference the section they are checking alongside data obtained from tools, such as tlparse.
This diff operates on the assumption that each graph section will enter and exit a CompileContext before it is ran to either compile the graph or look it up in the cache. Based on this assuption, we can save the value of the graph section from the exited CompileContext in eval_frame.c using a Python C API. After this, we can create a new interface in cpp shim to wrap around the record_function in order to pass in the new keyword argument for "context".
Test Plan:
Enhance test_profiler_dynamo_compiled_region to look for kwinputs as well as a name to see that the context is now labeled. Also changed test to run graph with more contexts so that we test a wider range of profiling.
Differential Revision: D60803317
Pull Request resolved: https://github.com/pytorch/pytorch/pull/132765
Approved by: https://github.com/anijain2305
Summary:
This PR adds in cuSPARSELt as a backend to PyTorch.
It is now possible to see if cuSPARSELt is available and the version if
it is with
```
torch.backends.cusparselt.is_available()
torch.backends.cusparselt.version()
```
Test Plan:
```
python test/test_sparse_semi_structured.py -k test_cusparselt_backend
```
Reviewers:
Subscribers:
Tasks:
Tags:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/128534
Approved by: https://github.com/cpuhrsch, https://github.com/eqy, https://github.com/syed-ahmed
This PR adds a C function to check if all torch function is disabled.
Recall that there are three torch function enablement states:
* All disabled
* Torch Function Subclass disabled
* All enabled
The API before this change provides two functions:
* `_is_torch_function_enabled` - returns True iff the current TF state is All enabled
* `_is_torch_function_mode_enabled` - returns True iff the state is not All disabled and the torch function mode stack is non-empty.
The crux of why a new API is needed is the following: If dynamo enters a frame with the torch function mode stack empty, `_is_torch_function_enabled` == False, it is impossible to determine if after a new mode is pushed whether we should enter the mode or not. This is because we don't know if the enablement state is All disabled or only subclass disabled. Adding this API to check if All disabled is True allows us to disambiguate this case.
In the next PR, Dynamo InstructionTranslator will have clearer flags than the underlying C API:
* A flag to indicate if subclasses are disabled (ie All disabled or Subclass Disabled is the current state)
* A flag to indicate if modes are disabled (ie if All disabled is the current state)
* A symbolic stack which can be checked if any modes are present
Pull Request resolved: https://github.com/pytorch/pytorch/pull/133136
Approved by: https://github.com/bdhirsh
ghstack dependencies: #133130, #133729, #133131, #133132, #133133, #133134
This PR adds a guard on the torch function mode stack state at the beginning of tracing. The way this is implemented is via a new leaf guard which is passed the initial stack state at construction and compares it to the stack state at the time the guard is run.
Details:
The stack state is extracted via popping all modes, appending them to a list, and pushing all modes back. This list is stored on the output graph and read during guard construction to pass to the stack mode guard. There the length and types of the modes are recorded. Next time the guard is run it compares this recorded state to the current mode stack state.
To implement this in python a helper function was added to utils.py and this is used if cpp guards are not enabled.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/133130
Approved by: https://github.com/anijain2305
Summary:
It seems we have multiple places deserializing torchbind objects. Moving the code around so that every load essentially share the same implementation.
Also added a test case "package_reader_testing" which load back the archive file in Python and eagerly validate the numerical result.
Test Plan: buck test mode/opt sigmoid/inference/test:e2e_test_cpu
Reviewed By: SherlockNoMad
Differential Revision: D61235770
Pull Request resolved: https://github.com/pytorch/pytorch/pull/133463
Approved by: https://github.com/ydwu4
Fixes#10536
Reattempt of #61467. Thank you so much to @mskoh52 for your excellent work!
As I was trying to create a more efficient LLM data collator, I realized that `pad_sequence` only supports right padding, even though left padding is a very common format for LLMs, like Llama and Mistral.
The proposed alternative implementation was to use multiple flips, which tends to be 1.5x-2x slower. Instead we can add a [`padding_side` parameter as there is for for Hugging Face tokenizers](9d6c0641c4/src/transformers/tokenization_utils_base.py (L1565)), which requires only a very small change in the C++ code.
Here are the benchmarks of the new implementation!
`float32`:
`bool`:
Code:
```python
from __future__ import annotations
import random
import time
from typing import Literal
import numpy as np
import torch
def pad_sequence_with_flips(
sequences: list[torch.Tensor],
batch_first: bool = False,
padding_value: int | float | bool = 0.0,
padding_side: Literal["left", "right"] | str = "left",
) -> torch.Tensor:
if padding_side == 'right':
padded_sequence = torch._C._nn.pad_sequence([t.flatten() for t in sequences], batch_first=batch_first, padding_value=padding_value)
elif padding_side=='left':
padded_sequence = torch._C._nn.pad_sequence([t.flatten().flip(0) for t in sequences], batch_first=batch_first, padding_value=padding_value) # pyright: ignore[reportArgumentType]
padded_sequence = padded_sequence.flip(int(batch_first))
else:
raise ValueError(f"padding_side should be either 'right' or 'left', but got {padding_side}")
return padded_sequence
sequence_lengths: list[int] = []
flip_left_pad_times: list[float] = []
flip_left_pad_times_std: list[float] = []
left_pad_times: list[float] = []
left_pad_times_std: list[float] = []
RUNS_PER_LOOP: int = 100
for i in range(1, 7):
sequence_length = i * int(1e6) // 6
sequence_lengths.append(sequence_length)
sequences = [torch.randint(0, 2, (random.randint(1, sequence_length),), dtype=torch.bool) for _ in range(64)]
inner_left_pad_times: list[float] = []
inner_right_pad_times: list[float] = []
inner_flip_left_pad_times: list[float] = []
inner_flip_right_pad_times: list[float] = []
for _ in range(RUNS_PER_LOOP):
start = time.perf_counter()
torch._C._nn.pad_sequence(sequences, batch_first=True, padding_value=False, padding_side="left")
end = time.perf_counter()
inner_left_pad_times.append(end - start)
start = time.perf_counter()
pad_sequence_with_flips(sequences, batch_first=True, padding_value=False, padding_side="left")
end = time.perf_counter()
inner_flip_left_pad_times.append(end - start)
left_pad_times.append(sum(inner_left_pad_times) / len(inner_left_pad_times))
left_pad_times_std.append(np.std(inner_left_pad_times))
flip_left_pad_times.append(sum(inner_flip_left_pad_times) / len(inner_flip_left_pad_times))
flip_left_pad_times_std.append(np.std(inner_flip_left_pad_times))
print(f"Sequence Length: {sequence_length}, Left Pad Time: {left_pad_times[-1]}, Left with Flips Pad Time: {flip_left_pad_times[-1]}")
import matplotlib.pyplot as plt
plt.plot(sequence_lengths, left_pad_times, label="new pad_sequence left")
plt.scatter(sequence_lengths, left_pad_times)
plt.errorbar(sequence_lengths, left_pad_times, yerr=left_pad_times_std, linestyle='None', marker='^')
plt.plot(sequence_lengths, flip_left_pad_times, label="old pad_sequence left (2 flips)")
plt.scatter(sequence_lengths, flip_left_pad_times)
plt.errorbar(sequence_lengths, flip_left_pad_times, yerr=flip_left_pad_times_std, linestyle='None', marker='^')
plt.xlabel("Sequence Length")
plt.ylabel("Time (s)")
plt.legend(loc="upper right")
# Sequence Length: 166666, Left Pad Time: 0.06147645162009212, Left with Flips Pad Time: 0.09842291727001794
# Sequence Length: 333333, Left Pad Time: 0.08933195920990329, Left with Flips Pad Time: 0.15597836187991562
# Sequence Length: 500000, Left Pad Time: 0.08863158334006585, Left with Flips Pad Time: 0.15224887342999863
# Sequence Length: 666666, Left Pad Time: 0.10524682551997103, Left with Flips Pad Time: 0.18177212480995877
# Sequence Length: 833333, Left Pad Time: 0.11801802741003485, Left with Flips Pad Time: 0.20821274195001024
# Sequence Length: 1000000, Left Pad Time: 0.131894061660023, Left with Flips Pad Time: 0.23223503091008751
```
Co-authored-by: mskoh52 <mskoh52@users.noreply.github.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/131884
Approved by: https://github.com/ezyang
The regression from https://github.com/pytorch/pytorch/issues/132281 pinpoints e4ace1a396 as the cause. The main delta that commit introduces is that we now manually check `is_inference()` and call `increment_version()` (a pybind call) on every mutated input tensor to the graph.
This PR attempts to reduce overhead a bit by bundling up all of those checks into a single pybind call, by:
(1) updating `torch.autograd.graph.increment_version()` to accept a `Union[Tensor, List[Tensor]]`
(2) updating its semantics to no-op if you pass in a tensor with no version counter, instead of erroring
Pull Request resolved: https://github.com/pytorch/pytorch/pull/132652
Approved by: https://github.com/albanD
We provide an API for user to add ephemeral timeout across all PGs within one rank and the timeout will reset when the first collective issued after the timeout added finishes.
Each extension only covers collectives after the issue and before the first collective finished. The diagram below shows how the timeout changes:
<img width="1174" alt="image" src="https://github.com/user-attachments/assets/354923b7-581c-40de-ae0f-1cd3da273ccc">
While this feature provides flexibility in specific scenarios, it introduces statefulness to timeout setting. Therefore, it is advisable to use this API sparingly and consider alternative approaches, such as directly setting the timeout or utilizing a barrier collective (one can set any timeout to the barrier), whenever feasible.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/130905
Approved by: https://github.com/ezyang
Implements donated buffer feature and adds unit tests. Donated buffer is a saved tensor that is not aliased with forward inputs, fw_outputs (except saved tensors), and bw_outputs. We detect donated buffers during `aot_dispatch_autograd` and store donated buffers in `ViewAndMutationMetadata`, such that it can be accssed in inductor.
Fixes#129496
Pull Request resolved: https://github.com/pytorch/pytorch/pull/130580
Approved by: https://github.com/bdhirsh
