Summary:
My team has been hitting a mysterious crash for a few months on a windows binary that uses Caffe2 inside a worker thread.
When this thread gets destroyed, there is an error at this line in context_gpu.h where the state of this operation gives CUDNN_STATUS_INTERNAL_ERROR instead of CUDNN_STATUS_SUCCESS.
When enabling cudnn debug logs (via the env variables nvidia specifies), I can see that the context is destroyed twice, even though this code only destroys it once, so something mysterious is causing a double free.
This seems very very similar to the issue/fix described here for pytorch:
https://github.com/pytorch/pytorch/issues/17658https://github.com/apache/tvm/pull/8267
And pytorch handles this in the same way, by just not calling cudnnDestroy
This seems to have become an issue with cuda11, but I tested cuda12 as well and found that the issue persists so this needs to be somehow fixed.
Test Plan:
CI
I checked that the specific windows binary I am using is able to create and drestroy caffe2-invoking threads without causing the application to crash.
buck run arvr/mode/win/cuda11/opt //arvr/projects/nimble/prod/tools/MonoHandTrackingVis
Differential Revision: D43538017
Pull Request resolved: https://github.com/pytorch/pytorch/pull/95382
Approved by: https://github.com/malfet
Summary:
Follow-up to https://github.com/pytorch/pytorch/issues/18584. This PR covers the remaining places where event or stream query might result in not ready errors.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/61554
Reviewed By: mrshenli
Differential Revision: D29763973
Pulled By: ezyang
fbshipit-source-id: 41d988d1826b2309cc6b01a81144094b353abdf9
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/55121
This is done by allow -1 as a stream ID, meaning "don't change
the stream", in SwitchToDevice
Fixes#54830
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Test Plan: Imported from OSS
Reviewed By: malfet
Differential Revision: D27527544
Pulled By: ezyang
fbshipit-source-id: c54983d6fc79a8fa1c65a71559a57425e40ba717
Summary:
Throwing from destructor leads to undefined behaviour (most often to segault)
So it's better to leak memory then segault
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34756
Test Plan: Run `test_pytorch_onnx_caffe2`
Differential Revision: D20504228
Pulled By: malfet
fbshipit-source-id: 7a05776fea9036f602e95b8182f8493cb5886dab
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/17623
Despite it's generic sounding name, caffe2::DeviceGuard actually
only worked on CUDA devices. Rename it to something that more
clearly spells out its applicability.
I'm not sure if it's the right call, but in this patch I added
'using CUDAGuard = c10::cuda::CUDAGuard', as this seems to be more
in-line with how the Caffe2 codebase is currently written. More
idiomatic c10 namespace style would be to say cuda::CUDAGuard.
Willing to change this if people shout.
This is a respin of D13156470 (#14284)
Reviewed By: dzhulgakov
Differential Revision: D14285504
fbshipit-source-id: 93b8ab938b064572b3b010c307e1261fde0fff3d
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/16399
Catching cudaError_t return values in a few places, because it's nodiscard in rocm. Unless we add -Wno-unused-result, it'll end up with a compilation error.
Also in c10/cuda/test, check whether a host has GPU or not. We were silently throwing out the error before (so not really testing the cuda api).
Reviewed By: bddppq
Differential Revision: D13828281
fbshipit-source-id: 587d1cc31c20b836ce9594e3c18f067d322b2934
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/16721
The very key line is we have to set the stream to the default
stream before calling the allocator. This is very interesting.
It shouldn't be necessary, but seemingly is!
Reviewed By: dzhulgakov
Differential Revision: D13943193
fbshipit-source-id: c21014917d9fe504fab0ad8abbc025787f559287
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/15692
It was leading to ocassional crashes with dynamically linked CUDA because runtime was already destroyed.
Also, unique_ptr<T[]> is more suitable than deque<T> for the purpose.
Reviewed By: Yangqing
Differential Revision: D13571988
fbshipit-source-id: 37eb26dfbe361c49160367b53f87bd037c6c0e46
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/15316
This starts cleaning up the files in c10 according to the module structure we decided on.
Move to c10/util:
- Half.h, Half-inl.h, Half.cpp, bitcasts.h
Move to c10/core:
- Device.h, Device.cpp
- DeviceType.h, DeviceType.cpp
i-am-not-moving-c2-to-c10
Reviewed By: dzhulgakov
Differential Revision: D13498493
fbshipit-source-id: dfcf1c490474a12ab950c72ca686b8ad86428f63
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/15174
Previously, Caffe2 maintained a separate per-thread per-device
current logical CUDA stream ID. In this PR, we switch Caffe2 over
to using c10::Stream to manage the current stream, and also
manage the allocation of cudaStream_t objects.
This results in a slight behavior change: previously, Caffe2
would have been willing to allocate an arbitrary number of
CUDA streams, depending on how high the logical stream IDs
went. The c10::Stream pool has a fixed number of streams, once
you exceed it, it wraps around.
Reviewed By: dzhulgakov
Differential Revision: D13451550
fbshipit-source-id: da6cf33ee026932a2d873835f6e090f7b8a7d8dc
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/15126
I want to make people stop manufacturing StreamId from thin air,
and a first step is to make people use the default stream.
Reviewed By: dzhulgakov
Differential Revision: D13432922
fbshipit-source-id: 9f0d8d70646c50d979bde5ba3c3addeebac48a3d
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/14282
This also is a substantive change, as 'DeviceIndex' and 'StreamId' are
narrower types than 'int'.
Reviewed By: Yangqing, smessmer
Differential Revision: D13156471
fbshipit-source-id: 08aa0f70c4142415b6bd4d17c57da0641c1d0e9a
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/14268
Removes the need for Context in Tensor by doing simple dispatch for CopyBytes. It'd eventually be subsumed by Roy Li's changes of proper copy_ op, but before that is done, let's get a clear logic of how copies are implemented and clean up some craft in CopyFrom implementation.
Note, that with these changes, one can probably can get rid of Context::CopyFromCPU/CopyToCPU, but it's a matter for follow up diffs.
This diff doesn't change the API of Tensor yet, but relies on the fact that passing `Context` to CopyFrom makes copy async if the device is CUDA and doesn't have any effect otherwise (that's how Context methods are implemented).
This doesn't change semantics of copy async implementation - as before it blindly calls cudaMemcpyAsync which probably means that it can be misused if invoked separately outside of operator body. I'll leave it for the follow up copy_ unification.
For Extend() we always do async copy - it makes sense as it's an in-place device-device operation and only any further op would be observable.
Note: there are now three ways of invoking copy in C2 code - templated CopyBytes, virtual CopyFromCPU/etc, and double-dispatch free method here. Hopefully we can get rid of the second one.
Also, please advise whether it's c10-worthy :)
Reviewed By: ezyang
Differential Revision: D13117987
fbshipit-source-id: a6772d6dcf3effaf06717da3a656fc9873b310b5
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/13080
This is the first step to untangle this logic:
- moves stream id to thread local mechanically
- relies on the fact that the value of thread local is valid in conjunction with CUDAContext only until the next SwitchToDevice is called - we should move to proper RAII in the following diffs
Follow up diffs are going to move more stuff outside of CUDAContext (by making gpu_id thread local too) and simplify the CopyFrom.
The only expected change in behavior is that before CopyFrom would do copy on stream logical id 0 if the context was created on the fly and now it'd do so on the current stream. Since it'd block explicitly, I don't think it matters much.
Also, observers were semi-broken by waiting on the potentially wrong stream. It can be fixed later - I renamed the method to avoid abuse.
Reviewed By: ezyang
Differential Revision: D10525134
fbshipit-source-id: 5d495a21490bebe060a76389f1b47bdf12cbc59e
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/12304
- make ExtractDeviceOption to be a free function.
- Add a Strorage(at::Device) constructor in order to preserve the device_id.
Reviewed By: dzhulgakov
Differential Revision: D10069839
fbshipit-source-id: a5f3994a39bdf1b7503b39bb42c228e438b52bfa
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/12029
In order to remove New() function in StaticContext(to remove StaticContext) and converge to the Allocator design, we'll first change the return type of New to at::DataPtr.
Reviewed By: ezyang
Differential Revision: D9889990
fbshipit-source-id: 3257c763530b987025f428741bdd2e089d11bad4
Summary:
Original commit changeset: f5614a5d2607
D9986213 is causing Multifeed Aggregator a [huge performance different](https://our.intern.facebook.com/intern/ads/analyze_canary/412951953278781781/) and is blocking aggregator push since last Friday night: https://fburl.com/feedtools/b6izvwjz
We need to land this revert ASAP to unblock aggregator push.
Reviewed By: orionr
Differential Revision: D10123245
fbshipit-source-id: d83da8e00a1250f5d09811a0a587c127e377aab2
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/11688
As a first step to remove static context(merge with allocator), we'll create a
global registries for context constructors, and remove CreateContext function from tensor.
Reviewed By: ezyang, dzhulgakov
Differential Revision: D9779821
fbshipit-source-id: 8b239ea50af7a0556fde2382f58f79194f0e3dc1
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/11656
The mis-capitalization really sticks up my craw. I know why (we
already have a static function named GetDeviceType), but let's
name it differently.
```
codemod -d . --extensions cc,cpp,cu,cuh,h,py,hpp,TARGETS GetDevicetype device_type
```
Reviewed By: jerryzh168
Differential Revision: D9813544
fbshipit-source-id: fe462f4bc40b03e74921f8cf5ebd9cfc52e7e636
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/11587
To help debug the issue in T33295362, we add some checks in the function.
Possible crashing site in `GetTensorInfo`
1. tc is nullptr, which is checked.
2. tc->capacity_nbytes() hits nullptr, this is unlikely because storage is not a pointer and compute of capacity_nbytes doesn't involve pointers. It's numel * itermsize().
3. tc->ExtractDeviceOption hits nullpt. One possibility raw_data() is nullptr because tc->ExtractDeviceOption will use that. This is checked.
4. Tensor itself which is not a reference. This is also checked.
Reviewed By: salexspb
Differential Revision: D9793484
fbshipit-source-id: 3fc72746fc310a23ae45553bbe0d269a4b9edb72
Summary:
This is an experimental build on top of what orionr and mingzhe09088 built.
Essentially, the idea is that we will need separate *_API versions for different shared libraries. If this theory is right, I'll try to clean up the design a bit and document it properly.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/11266
Reviewed By: orionr
Differential Revision: D9682942
Pulled By: Yangqing
fbshipit-source-id: c79653199e67a1500c9174f39f8b0357324763f3
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/11254
Previously we use DeviceType in caffe2.proto directly, but it's an `enum` and have implicit conversion to int, which does not have type safety, e.g. we have to explicitly check for a device type is valid in event.h:
```
template <int d>
struct EventCreateFunctionRegisterer {
explicit EventCreateFunctionRegisterer(EventCreateFunction f) {
static_assert(d < MaxDeviceTypes, "");
Event::event_creator_[d] = f;
}
};
```
at::DeviceType is an `enum class`, and it does not have implicit conversion to int, and provides better type safety guarantees. In this diff we have done the following refactor(taking CPU as an example):
1. caffe2::DeviceType → caffe2::DeviceTypeProto
2. caffe2::CPU → caffe2::PROTO_CPU
3. caffe2::DeviceType = at::DeviceType
4. caffe2::CPU = at::DeviceType::CPU
codemod -d caffe2/caffe2 --extensions h,cc,cpp 'device_type\(\), ' 'device_type(), PROTO_'
+ some manual changes
In short, after this diff, in c++, caffe2::CPU refers to the at::DeviceType::CPU and the old proto caffe2::CPU will be caffe2::PROTO_CPU.
In python side, we have a temporary workaround that alias `caffe2_pb2.CPU = caffe2_pb2.PROOT_CPU` to make the change easier to review and this will be removed later.
Reviewed By: ezyang
Differential Revision: D9545704
fbshipit-source-id: 461a28a4ca74e616d3ee183a607078a717fd38a7
Summary:
This completely removes BUILD_CAFFE2 from CMake. There is still a little bit of "full build" stuff in setup.py that enables USE_CUDNN and BUILD_PYTHON, but otherwise everything should be enabled for PyTorch as well as Caffe2. This gets us a lot closer to full unification.
cc mingzhe09088, pjh5, ezyang, smessmer, Yangqing
Pull Request resolved: https://github.com/pytorch/pytorch/pull/8338
Reviewed By: mingzhe09088
Differential Revision: D9600513
Pulled By: orionr
fbshipit-source-id: 9f6ca49df35b920d3439dcec56e7b26ad4768b7d
Summary:
Properly annotated all apis for cpu front. Checked with cmake using
cmake -DUSE_ATEN=ON -DUSE_CUDA=OFF -DBUILD_ATEN=ON
and resulting libcaffe2.so has about 11k symbols.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/10504
Reviewed By: ezyang
Differential Revision: D9316491
Pulled By: Yangqing
fbshipit-source-id: 215659abf350af7032e9a4b0f28a856babab2454
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/9939
Pull Request resolved: https://github.com/facebookresearch/weakly-supervised-action-detection/pull/13
Pull Request resolved: https://github.com/pytorch/translate/pull/166
Pull Request resolved: https://github.com/pytorch/pytorch/pull/9125
Closes https://github.com/pytorch/pytorch/pull/9125
Use inheritance for polymorphism, and remove template parameter
This is to change the templating in call sites, the core implementations will change later
Before Caffe2 Tensor class was compile-time fixed to bind to a particular device/context. With this change, we're making it a runtime property (stored inside the tensor), but preserve the same semantics. For example, one has to specify device type in order to create a Tensor - there are no uninitialized tensors. More specifically the changes are:
1. We added an extra argument *DeviceType* to most of the constructors of the tensor, e.g. (Tensor(DeviceType type)),
2. Semantics of constructor Tensor(const Tensor<SrcContext>& src, ContextForCopy* context); is changed, in this constructor, the second context is passed in to enable us to call the templated Copy function, it could be in a different context as source and target previously, now we'll enforce that the context should have same device type as src, if it is provided.
3. To preserve 'get-or-construct' semantics of Blob, we added specialized getter Blob::GetMutableTensor that verifies both that Blob contains a Tensor and that it's of a correct type
4. Specifically, Tensor type is not default-constructible any more (as we don't have unknown device tensors) and thus some of the code handling STL containers needs to change
Note: Some changes are postponed just to keep this diff a bit smaller. Please see `TODO`s.
Reviewed By: ezyang, houseroad
Differential Revision: D9024330
fbshipit-source-id: e0b8295d2dc6ebe2963383ded5af799ad17164ba
Summary:
Pull Request resolved: https://github.com/facebookresearch/weakly-supervised-action-detection/pull/13
Pull Request resolved: https://github.com/pytorch/translate/pull/166
Pull Request resolved: https://github.com/pytorch/pytorch/pull/9125
Closes https://github.com/pytorch/pytorch/pull/9125
Use inheritance for polymorphism, and remove template parameter
This is to change the templating in call sites, the core implementations will change later
Before Caffe2 Tensor class was compile-time fixed to bind to a particular device/context. With this change, we're making it a runtime property (stored inside the tensor), but preserve the same semantics. For example, one has to specify device type in order to create a Tensor - there are no uninitialized tensors. More specifically the changes are:
1. We added an extra argument *DeviceType* to most of the constructors of the tensor, e.g. (Tensor(DeviceType type)),
2. Semantics of constructor Tensor(const Tensor<SrcContext>& src, ContextForCopy* context); is changed, in this constructor, the second context is passed in to enable us to call the templated Copy function, it could be in a different context as source and target previously, now we'll enforce that the context should have same device type as src, if it is provided.
3. To preserve 'get-or-construct' semantics of Blob, we added specialized getter Blob::GetMutableTensor that verifies both that Blob contains a Tensor and that it's of a correct type
4. Specifically, Tensor type is not default-constructible any more (as we don't have unknown device tensors) and thus some of the code handling STL containers needs to change
Note: Some changes are postponed just to keep this diff a bit smaller. Please see `TODO`s.
Reviewed By: xw285cornell
Differential Revision: D8121878
fbshipit-source-id: 4a5e9a677ba4ac82095df959851a054c81eccf81
* Fix some signed/unsigned mismatches
* Skip unused result warning
* Explict fallthrough for murmur hash
* Enable aligned new support to eliminate warning
* Switch to int instead of unsigned in some cases
* [GanH][Easy]: Add assertion to adaptive weighting layer
0 weight causes numeric instability and exploding ne
* [Easy] Add cast op before computing norm in diagnose options
As LpNorm only takes floats we add a manual casting here.
* Introduce a new caching device allocator
`cudaMalloc` and `cudaFree` calls are slow, and become slower the
more GPUs there are. Essentially, they grab a host-wide (not device-wide) lock
because GPU memory is transparently shared across all GPUs. Normally, this
isn't much of a concern since workloads allocate memory upfront, and reuse it
during later computation.
However, under some computation models (specifically, memory conserving
approaches like checkpoint-and-recompute, see
https://medium.com/@yaroslavvb/fitting-larger-networks-into-memory-583e3c758ff9)
this assumption is no longer true. In these situations, `cudaMalloc` and
`cudaFree` are common and frequent. Furthermore, in data parallel contexts,
these calls happen at nearly the same time from all GPUs worsening lock
contention.
A common solution to this problem is to add a custom allocator. In fact,
nVIDIA provides one out of the box: CUB, which Caffe2 already supports.
Unfortunately, the CUB allocator suffers from very high fragmentation. This is
primarily because it is a "buddy" allocator which neither splits nor merges
free cached blocks. Study
https://github.com/NVlabs/cub/blob/1.8.0/cub/util_allocator.cuh#L357 if you
want to convince yourself.
This diff adapts a caching allocator from the Torch codebase
https://github.com/torch/cutorch/blob/master/lib/THC/THCCachingAllocator.cpp
which does splitting and merging and ends up working really well, at least for
workloads like the checkpoint-and-recompute computation models noted above.
I simplified the implementation a little bit, made it a bit more C++-like. I
also removed a bunch of stream synchronization primitives for this diff. I
plan to add them back in subsequent diffs.
* Report reader progress in fblearner workflows
Integrate with fblearner progress reporting API and add support to report training progress from reader nodes.
If reader is constructed with batch limits, report based on finished batch vs total batch. The finished batch may be more than total batch because we evaludate if we should stop processing everytime we dequeue a split.
If no limit for the reader, report based on finished splits (Hive files) vs total splits. This is fairly accurate.
* [GanH][Diagnose]: fix plotting
1. ganh diagnose needs to set plot options
2. modifier's blob name is used for metric field can need to be fixed before
generating net
* Automatic update of fbcode/onnx to 985af3f5a0f7e7d29bc0ee6b13047e7ead9c90c8
* Make CompositeReader stops as soon as one reader finishes
Previously, CompositeReader calls all readers before stopping. It results in flaky test since the last batch may be read by different threads; resulting in dropped data.
* [dper] make sure loss is not nan
as desc.
* [rosetta2] [mobile-vision] Option to export NHWC order for RoIWarp/RoIAlign
Thanks for finding this @stzpz and @wangyanghan. Looks like NHWC is more
optimized. For OCR though it doesn't yet help since NHWC uses more mem b/w but
will soon become important.
* Intra-op parallel FC operator
Intra-op parallel FC operator
* [C2 Proto] extra info in device option
passing extra information in device option
design doc: https://fb.quip.com/yAiuAXkRXZGx
* Unregister MKL fallbacks for NCHW conversions
* Tracing for more executors
Modified Tracer to work with other executors and add more tracing
* Remove ShiftActivationDevices()
* Check for blob entry iff it is present
When processing the placeholders ops, ignore if the blob is not present in the blob_to_device.
* Internalize use of eigen tensor
Move use of eigen tensor out of the header file so we don't get template partial specialization errors when building other libraries.
* feature importance for transformed features.
* - Fix unused parameter warnings
The changes in this diff comments out unused parameters.
This will allow us to enable -Wunused-parameter as error.
#accept2ship
* add opencv dependencies to caffe2
The video input op requires additional opencv packages. This is to add them to
cmake so that it can build
* Add clip_by_value option in gradient clipping
Add clip_by_value option in gradient clipping
when the value is bigger than max or smaller than min, do the clip
* std::round compat
