* add reduce arg to PoissonNLLLoss
* fixed comments except reference function
* fixed unit test
* small indentation fix
* fixing last comments by richard
* lint check
* another linting issue
* Comprehensive rewrite of Torch CuDNN bindings / a bit of ATen infra
The executive summary is that this moves the torch/csrc/cudnn
library into ATen, adding a number of new cudnn_ methods to ATen
for batchnorm, convolution, affine grid generator and grid sampler.
ATen infra changes:
- TensorGeometry was moved to ATen
- TensorGeometry was modified to make its interface resemble that of
Tensor; in particular, sizes is no longer a field, it's a method.
- AT_CUDA_ENABLED macro is set via ATen/Config.h header which is
generated at cmake configure time.
Fixes https://github.com/zdevito/ATen/issues/168
- Change AT_CUDA_ENABLED macro to be a function macro, so that we
error if it is not defined
- Introduce a new TensorArg class, which is a Tensor plus a little
metadata. This helps us give good error messages when checking
dimensions/shapes of tensors.
Fixes https://github.com/zdevito/ATen/issues/169
- Also introduce a TensorGeometryArg class, for when you don't
need the actual tensor data (which is most of the time.)
- Add ATen/Check.h, which contains a number of utility functions
for testing shapes, types and devices of input tensors. This
will be particulary useful for native methods, which don't get
code generated input testing code. These functions take a
'CheckedFrom' argument, at the moment just a string, which
specifies some extra information about what function was
doing the actual checking; this greatly improves error messages.
- Many check functions take initializer lists, which let you
test that all tensors have some property. This API is
peculiar, in that we IGNORE undefined tensors in this case.
This is handled by filterDefined.
- Add AT_CUDNN_ENABLED macro
- CuDNN linking from ATen was improved; for example, we now actually
add the CuDNN headers to our include path.
- Add some missing override specifiers to some methods
- We now actually build tests with CUDA functionality accessible
(previously, AT_CUDA_ENABLED was not defined, meaning that
the headers were missing all CUDA-only functionality.)
- Native functions now support giving explicit names to return
outputs in yaml. This makes it possible to hook into the NN
autogenerated derivatives codepath using native functions.
CuDNN rewrite changes:
- torch/csrc/cudnn now uses ATen (rather than passing around
THVoidTensor) and lives in ATen. This lets us remove tensorPointer
shenanigans. The functions are exposed to ATen as native functions
described in aten/src/ATen/cudnn/cuDNN.yaml
- ATen now builds and links against CuDNN when enabled. The cmake
package script was taken from Caffe2.
- Some header reorganization was done to help reduce dependencies
on headers (this reorg is no longer used but I've kept it)
- Rename CHECK to CUDNN_CHECK
- Rip out old shape/type testing code in favor of modern ATen/Check.h
interface using TensorArg. In many cases, increase the robustness of
the checking code.
- Change the inputs of the public facing functions, so that they can
be bound by ATen
- Delete THCState*; this is retrieved from the global ATen context
- Delete cudnnHandle_t, this is retrieved from the global Handles.h
- Delete cudnnDataType_t, this is retrieved from the Tensor type
- Delete Convolution class, instead its constituent arguments are
passed individually
- Change functions to return tensors, rather than take an appropriately
sized output tensor as an input.
- Redo how transposed convolution / backward convolution is implemented
(knock on effect of returning tensors). Previously it was assumed
that you would always pass an appropriately sized output tensor, but
we don't want to do this anymore. For backwards, we instead give
the desired output tensor (input, really) size, because that is
readily available. For *transposed* convolution, however, we take
output_padding, and otherwise do the shape calculation.
- Redo how legacy group convolution is implemented (knock on effect from
porting cudnn to ATen.) Previously, group convolution was implemented
by manually constructing sizes and strides and then outputting
appropriate, with macros switching between individual groups and
all-at-once based on CuDNN version. Now, the code looks exactly what
you'd expect: there's a top-level wrapping function that supports
group convolution no matter the version of CuDNN, and a low-level
wrapper which supports only what CuDNN supports. The top-level
function conditions on CuDNN version, and invokes the low-level
interface 1 or n times.
- There is now a debugging printer for tensor descriptors.
- Convolution struct is replaced with ConvolutionArgs, which is not
part of the public API but is used internally to conveniently
pass around all of the arguments needed for Convolution.
- Add some constexprs for well-known dimensions, reduce amount of
magic numbers in code.
- Put 'deterministic' in to ConvParams. Fixes#3659
- Lots more comments.
- Some pessimizations, in the name of code clarity:
- The descriptors are initialized on every invocation of convolution
forward/backward. Previously, the descriptors were cached, so that
you didn't have to initialize them again on backwards. This is
difficult to support in the ATen interface so I didn't support it.
- Legacy group convolution initializes its workspace for *every* group
it performs. I did not feel motivated to fix this because the
legacy codepath is already quite slow.
- Affine grid generator and grid sampler automatically call contiguous
on their arguments as necessary.
- Batchnorm input checking is greatly beefed up, it now checks for
the following input characteristics:
- Definedness
- GPU location
- Type
- Contiguity
- Size
PyTorch binding code changes
- batchnorm now uses consistent var/data naming
- batchnorm and convolution make use of new ATen bindings
- Affine grid generator and grid sampler make use of ATen CuDNN
bindings via derivatives.yaml. This means I had to restructure
the code a little, since the THNN bindings still go through
a legacy Python class.
- I fixed some warnings:
- s/friend class/friend struct/ on InterpreterStateImpl
- Removed pessimizing move 'detached' in torch/csrc/autograd/variable.cpp
- Removed unused pack_list on Scalar
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
GCC 4.8 buildfix
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Add TensorGeometry to ATen.h
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
CUDNN_CHECK
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Update TODO comment
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Delete return in cudnn_grid_sampler
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
s/cudnnSetStreamToCurrent/setCuDNNStreamToCurrent/g
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Don't allocate a new vector when filtering defined.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Remove Check overloads, convert to pass references.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Some more microbenchmarking.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
For example, this splits threshold into threshold(), which is now
never in-place, and threshold_() which is always in-place.
This simplifies the in-place vs. non-in-place logic in
gen_variable_type.py, which was bug-prone.
This operator is a warmup I was doing before tackling convolution, as it
has many properties that make it a "first" for implementing things. In
particular, it is the first operator whose backwards have multiple
returns; this means its double backwards is the first backwards for a
function with multiple differentiable outputs. This exercises new code
for output_mask and set_flags.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
* Prevent numerical issues with poisson_nll_loss when log_input=False
Evaluation of the logarithm of the input variable in poisson negative log likelihood leads to NaN loss if variable being evaluated is zero. Small epsilon is added to prevent this. See equivalent Keras epsilon here: https://github.com/fchollet/keras/blob/master/keras/losses.py#L68
* PEP8 fix
* Add epsilon support to PoissonNLLLoss in nn.modules.loss
* API changes
* Implement reduce for THNN ClassNLLCriterion
* Implement reduce keyword for THCUNN ClassNLLCriterion
* Implement reduce for THNN SpatialClassNLLCriterion
* Implement reduce for THCUNN SpatialClassNLLCriterion
* Make legacy NLLLoss work
* Docs for NLLLoss reduce
* reduce keyword for double backwards NLLLoss
* reduce=False tests
* Addressed comments
* Fix trailing whitespace
* Fix test failures in legacy nn
* Rebase: add reduce keyword to aten declarations of NLLLoss
* Add reference functions for all NLLLoss and NLLLoss2d test cases
* Replaced slow get/set fns. Don't use int64_t in kernels.
* Use TH_INDEX_BASE in NLLLoss for consistency
* Fix legacy ClassNLLCriterion tests
- Cleaned up THNN and THCUNN code and kernels
- Improved THCUNN kernel performance 5x, making it match cuDNN performance
- Added support for computing softmax over arbitrary dims
NOTE: The default dim for 3D inputs is now 1 (used to be 0)
- Both functions now accept inputs with arbitrarily many dimensions
- Autograd functions no longer save the input (it's unnecessary)
- Added cuDNN bindings for softmax, but they are unused as THCUNN
matches or even exceeds cuDNN performance
* Fix docs for nn.Embedding and F.embedding.
- add description of 'sparse' argument (#3104)
- fix F.embedding example (resulted in RuntimeError)
* Make EmbeddingBag a New Style Function.
* Add a functional interface for EmbeddingBag
* Fix failing tests: add max_norm and norm_type to context,
and fix typo in backend call.
* Docfix: remove torch.manual_seed from example code.
* Add a note about using sparse keyword in Embedding function.
* Add reduce keyword to MSECriterion API
* Move gradOutput usage from py to backend
* Implement reduce keyword for THNN MSECriterion
* Implement reduce keyword for THCUNN MSECriterion
* Implement reduce keyword for MSE double backwards
* Tests for MSECriterion with reduce keyword
* Documentation for reduce for MSELoss
* Make legacy nn work with reduce keyword by ignoring it
* Apply linter suggestions
* Address comments (small changes)
* Revert "Tests for MSECriterion with reduce keyword"
This reverts commit 1c0be0defa49d336d023d7d9795db4037c92b6fe.
* Undo changes to legacy nn tests
* Reuse module test for MSELoss by creating a wrapper class for MSELoss
* Address comments: refactor MSECriterion.cu to be nicer
* Fix lint & build errors
* Add examples in functional.py
Added examples for F.cross_entropy, F.binary_cross_entropy and F.binary_cross_entropy_with_logits.
* Add ` for PyTorch docs
Added ` for PyTorch docs.
* Add examples in loss.py
Added examples for nn.BCELoss and nn.BCEWithLogitLoss.
* added tests + removed explicit expand of weight in bce with logits
* add auto broadcasting of weight to BCELoss
* remove the need for _BCELoss
* formatting of warning
* remove TODO
* move across assert from _functions/thnn/loss.py
* flake8 fixes
* add dropout2d and dropout3d to functional
added some loss functions to functional
added tests
using dropout from backend
added docs
fixes
* edited loss modules to call functional
This takes advantage of the broadcasting behavior of torch.matmul to
support inputs with more than two dimensions. The extra dimensions are
treated like part of the batch dimension, much like nn.Bottle in Lua
Torch.
There are a few related small performance changes:
* Addmm computes the gradient in column-major for inputs in
column-major format
* Variable.mm calls Addmm in-place with the desired output buffer
* Add SELU activation function
* Remove unnecessary case
* Add Function for SELU + tests and fix RReLU inplace
* Fix extra line in doc
* Fix tests
Remove in-place tests for RReLU. For some reason they fail on legacy nn, but passes on nn
* SELU in new-style Function
It also supports double backprop, verifyed with gradgradcheck
* Fix flake8
Here's the command I used to invoke autopep8 (in parallel!):
git ls-files | grep '\.py$' | xargs -n1 -P`nproc` autopep8 -i
Several rules are ignored in setup.cfg. The goal is to let autopep8
handle everything which it can handle safely, and to disable any rules
which are tricky or controversial to address. We may want to come back
and re-enable some of these rules later, but I'm trying to make this
patch as safe as possible.
Also configures flake8 to match pep8's behavior.
Also configures TravisCI to check the whole project for lint.
* Always compile .numpy() for all types
* Add torch.nn.functional docs and hidden headers
* Use sphinx to generate torchvision docs
* Remove unused import in ffi utils
