This removes volatile from Variable. The functionality is mostly
replaced by a global (thread-local) flag, which is controlled by
torch.set_grad_enabled() and the context manager torch.no_grad().
In C++, the flag is exposed through GradMode::is_enabled() and GradMode::set_enabled()
Fixes#3627
* Trace ATen non-primitive functions as themselves, not their implementations.
Previously, if I invoked an ATen non-primitive function foo, which in turn
called subfoo, I would always see 'subfoo' in the trace (e.g., tracing
'inlines' all of these operations.) Such inlining is bad for ONNX
(and can be bad for optimization) as it prevents high-level
optimizations from taking advantage of the structure. It might
be right to inline, but give the optimizer a chance to work before
inlining happens!
The implementation here is surprisingly simple, because it uses
the "DCE trick". Essentially, it doesn't matter if the constituent
calls perform tracing, because you can always trace it again, and
override the trace nodes associated with the returned variables.
The original trace becomes dead and can be DCE'd.
While implementing this, I also refactored how 'isTracing' and
'trace_outputs' works:
- isTracing was previously a single function with overloads for
both Tensor and Variable arguments. Unfortunately, such overloads
are not safe, because of how C++ implicit conversions work. You
would think that C++ should never confuse an overload for
Variable with ArrayRef<Tensor>, but this is exactly what can
happen: Tensor is convertible to both Variable and ArrayRef<Tensor>,
thus it's ambiguous and C++ doesn't like it. The last time I ran
into this problem, I applied initializer lists to everything and
called it a day. A more robust fix is to separate out the
Variable and Tensor overloads, which I have done in this patch.
- trace_outputs was fed as an initializer list, which doesn't work
when you have heterogenous inputs. So instead we first feed
everything through 'flatten', which has overloads for each of the
argument patterns in ATen, which then goes on to the recordTrace
(which takes an ArrayRef). This is *no less efficient*, because
we were allocating a vector anyway (to do the conversion from
vector of Tensor to vector of Variable).
This fixes mean that 'index' can properly be traced... although the
JIT still does not support it. A failing test case has been added to
this effect.
Some knock-on effects:
- The fuser now knows about chunk as well as split. They're pretty
similar so there is no problem.
- There is a new 'canonicalize' pass in the JIT which renumbers a graph
so that all structurally equivalent graphs render the same.
- We run DCE before the fuser tests, to make sure dead nodes don't
block fusion.
- There are new ONNX exports for the newly introduced higher level ATen
operations. This includes type_as (no-op case only), chunk, select.
Zach didn't like the extra use of 'native' in the new codegen, so
we've introduced a new concept, 'abstract'. An abstract function
is one that is implemented in derived types (e.g., CPUDoubleType),
where as a concrete one is implemented in the base type (Type).
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
SavedVariable.unpack() may throw std::runtime_error which may lead to
program termination with SIGABRT without the exception beeing handled
in Python
Fixes#3860
* Add interpreter support for Handles/PythonOp/CppOp
This treats Handles as a first-class type in the interpreter
since this turned out to be conceptually simpler than treating
them as a separate concept, which requires a second channel for
register allocating and moving data from one op to the next.
Notes:
* The refcounting nature of tensors is factored into its own base type
so that it can be shared with other refcounted types such as handle.
* Some methods redundant with TensorBase have been deleted from Tensor
* The interpreter uses raw refcounted handles. In addition to being
able to treat Tensors and Handles as the same base object, it removes
a lot of redundant refcounting as objects moved from tensors to input/
output lists.
* aten_dispatch has been updated to work directly on the raw refcounted
lists to avoid refcounting and duplicate lists.
* Removing jit_closure.cpp, The interpreter can now handle all pathways.
* Functions like `unsafeToTensorShare` describe how
ownership transfers in the interpreter. The `Steal` variants
take rvalue references as arguments, and invalidate those
arguments to prevent potential problems.
* Make TensorTemporary is not a subtype relationship because it is too easy to
do something horribly unsafe:
```
void foo(at::Tensor bar) {
// bar destructor call release on a temporary!
}
foo(TensorTemporary(retainable)); // structure slicing!
```
Previously, an in-place operation that saves its output (such as
relu/threshold) would create a reference cycle when applied to the a
view. There were two cycles created:
1) The cycle base.grad_fn.fn.input_.base
base.grad_fn is a CopySlices
base.grad_fn.fn is ThresholdBackward
base.grad_fn.fn.input_ is a SavedVariable with base pointing to base
2) The cycle base.grad_fn.fn.input_.grad_fn.next_functions[0]
base.grad_fn.fn.input_.grad_fn is AsStridedBackward
and next_functions[0] points to base.grad_fn
Generally, we avoid cycles because the AD graph is mostly immutable. Two
notable exceptions are:
a) Variable.grad_fn can change to point to a new grad_fn
b) SavedVariables in a function can be set after the function is created
The first case is not a problem if grad_fns do not hold strong references
to Variables. Removing "base" from SavedVariable removes the strong ref.
For the second case, we need to avoid saving the grad_fn of outputs. We
were incorrectly saving the grad_fns of outputs when they were the
result of in-place ops on views.
This commit adds a Value type similar to the one @ezyang suggested a while
ago for handling multi-return nodes.
Previously if we had a graph like:
a = op1(b)
c, d = op2(a)
Then its in-memory format would look like:
%0 = op1(b)
%1 = op2(%0)
%2 = select(%1, 0)
%2 = select(%1, 1)
Select nodes were used only to handle the multi-output case. In the
single-output case ops referred directly to their uses.
This required special handling for the single- and multi- output cases,
and was confusing when used with ONNX which distinguishes values (the
inputs/outputs of a node) from the nodes themselves (e.g. a Conv).
This commit adds the Node/Value distinction to the IR. In the example
above, `a`, `b`, `c`, and `d` are now Value objects, while `op1` and
`op2` are now Node objects. Inputs/Outputs to the graph are values.
* Nodes now always have multiple outputs, accessible through their `output()`
method.
* Methods exist for adding/removing outputs from a node.
* Nodes own their output Values, destroying a node destroys its outputs and it
is only valid to destroy a node when no uses of its outputs remain.
* Unlike select, Values do not appear in the nodes list.
* The method `node()` on `Value` retrieves its defining node. Calling it
is always valid. For inputs, its kind is "Param". Like "Return" there is a single Param
node representing all inputs.
* For single-output Nodes, the method `output()` retrieves the single
output Value, asserting that the node is in-fact single output.
* Functions are the same, but some functions like `type()` have moved to
Value.
* `replaceAllUsesWith` is now sanely defined for both Values and Nodes.
In the case of Nodes, it replaces all outputs of the node with the outputs
of the replacement node.
* stage is defined both on Node/Value. This is because Inputs require a stage.
* Apart from changing data types from Node->Value most passes remain the same.
Things that previously assumed single-output nodes now have to call output()
to get the node.
* This removes the uses = [...] field in the outputs because it was
getting confusing even before this commit when uses would refer to nodes,
but we print the names of Values. The lint pass validates the use list,
so printing it out seems less necessary.
Allow in-place operations on views
Adds VariableViewImpl, a subclass of VariableImpl which has a pointer to
the base Variable on which it is a view. In-place operations on views
change the grad_fn of the base.
Note that in-place operations only work on views that are the first output of the function that created them. All C++/ATen implemented functions have this behavior, but it's possible to write Python-implemented autograd functions that do not. In-place operations on these view will raise an exception.
Fixes#3313
Replace None grad_inputs with zero tensors in some cases
In Python-implemented autograd functions, we sometimes return None as
the grad_input if the output is marked "non-differentiable". This
replaces those None values with zero-filled Variables if the
corresponding input has requires_grad=True.
C++ implemented autograd functions expect the input (grad_outputs) to
be defined if they're executed. They always return non-null grad_inputs
if should_compute_output(i) is true. This could lead to segfaults if a
subsequent Python-implemented function returned None.
See #3412, #3241
I've also made the version counter and the "live" reference count
atomics.
Note that it's not safe to set the version counter (operator=) from
multiple threads, because shared_ptr assignment isn't thread safe.
Currently, the only call sites to these functions are on newly created
variables before they can be accessed from other threads.
See #3111
Variable is now a subclass of at::Tensor backed by a VariableImpl* pImpl. The implementation of the ATen functions is defined in the auto-generated VariableType.h/cpp file.
Currently, only functions which fall through to the base type, such as sizes() and isCuda() are implemented. Differentiable ops like add() and mul() will be added in a subsequent PR.
