* Add torch.take and Tensor.put_
These are similar to numpy.take and numpy.put. The take function allows
you to linearly index into a tensor without viewing it as a 1D tensor
first. The output has the same shape as the indices. The put function
copies value into a tensor also using linear indices.
This includes some changes to the dispatch code for torch.xxx functions:
- Since Variable.addmm is an instance-method, the self argument has to
come first. The dispatch code swaps the first two arguments if
necessary to suppor the deprecated signatures where 'alpha' or 'beta'
comes before the 'self' tensor.
- Delete IMPLEMENT_STATELESS_REVERSED. These functions require output
arguments to be passed in using the keyword 'out'. They were meant to
handle torch.gt(out, a, b), but we haven't allowed that for a while.
ATen has it's own default CPU RNG. Use this as the default in PyTorch so
that random functions called through ATen have the same behavior as
random functions called through TensorMethods
Along the way I added converters for Variable and TracingInput. Variable should
probably be moved to a more widely known spot.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Previously, our AST was a DAG, where shared Nodes indicated a computation
should be reused. This commit rewrites the IR into a new functional
representation which represents sharing explicitly using variable
bindings.
We offer a few justifications for this new style:
1. The new representation is not all that different from the
old one; it is about as easy to construct, and the lack of an
explicit graph doesn't negatively impact our ability to interpret
the graph, since we've chosen, as a matter of design, to NOT have
the IR participate in the actual execution of a graph.
2. The new let-binding representation has an implicit ordering,
which we can use to conveniently keep track of the original order
the trace showed up as. This automatically gives us a topsort,
and gives us an easier to read textual representation of our
IR:
%14 = Embedding %11, %0, -1, None, 2, False, False
%15 = Dropout %14, 0.2, True, False
%16 = Index %12, 0
%17 = Index %12, 1
%18 = Index %13, 0
%19 = Index %13, 1
%20 = Index %15, 0
%21 = Linear %20, %1, %3
%22 = Linear %16, %2, %4
3. It moves us closer to a Futhark style language
(http://futhark-lang.org/publications/pldi17.pdf).
Major aspects of the diff
- Node is replaced with Expr and Arg, a pair of mutually recursive
structures which represent our new language. In BNF, the language
looks like this:
a ::= c | %i
e ::= %i, ... = e
| PyOp e, ...
| Ret %i, ...
Technically, Ret is not actually a return (no control flow is involved),
it just tuples up a series of tensors (identified by variables).
One important invariant is that locals are always tensors; they
are never constants (this is asymmetric with Args.)
- Arguments support Python constants. This is an important piece because
many operators take extra Python literals like integers and tuples in
order to specify extra parameters about how an operator operates. Adding
this was essential to getting word_language_model to work.
- As both Expr and Arg have multiple variants, there is new infrastructure
for doing case on the variants using ExprVisitor and ArgVisitor. The
strategy here is adapted from WebAssembly's visitors, although we have
generalized to permit arbitrary argument forwarding, which is necessary
to support tail-recursive visitor calls. TCO is important because our
interpreter may recurse arbitrarily deep into a stack of nested lets.
If users wish, they can also manually case on the type tag.
- Tracing is now turned on and off using _tracer_enter/_tracer_exit in
torch._C. _tracer_enter accepts a list of variables which are to be
treated as arguments; _tracer_exit accepts the list of traced variables
which should be returned when you reexecute the trace, and returns
the trace expression which can be reexecuted. GlobalTracingState
is a global variable which tracks whether or not we are tracing or not.
- You use run_forward to execute a trace on some set of parameters.
- When under tracing, variables keep track, via trace_local, what the
name of their variables in the IR are.
Here is a simple runner which leaks memory but can be used to JIT models:
import torch.autograd.function as F
import torch._C
def jit(model):
import types
real_forward = model.forward
def forward(self, *args):
def flatten(x):
return tuple(F._iter_variables(x))
if not hasattr(self, "saved_trace"):
torch._C._tracer_enter(tuple(self.parameters()) + flatten(args))
out = real_forward(*args)
self.saved_trace = torch._C._tracer_exit(flatten(out))
self.saved_outs = out
return out
else:
flat_out = Variable._execution_engine.run_forward(self.saved_trace, tuple(self.parameters()) + flatten(args))
return F._unflatten(flat_out, self.saved_outs)
Major problems:
- Sanity checking is spotty at best, especially when users pass in variables.
- The interpreter leaks tensor memory from the store. When we add back def-use
we should be able to deallocate tensors as soon as we know they are no longer
necessary.
- The interpreter needs to reach feature parity with the old execution engine.
From there, we need to see if backwards can be subsumed as well.
- I still have no confidence in having memory managed everything correctly.
This requires a close look.
- Rather than return an *open* expression as a trace, we should return a
*lambda* instead, which knows about how many formal parameters it
requires.
- The IR is not introspectable from Python at the moment, but this is simply a
matter of implementing all the binding code.
- The tracer is NOT reentrant (you can't trace while you're inside a trace.)
Furthermore, no sanity checking is done if you try to incorrectly reuse
things from one trace in another.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Setting torch.utils.backcompat.broadcast.warning.enabled=True
will cause Python warnings in the case where broadcast occurs
but previously 1-d view style pointwise ops occured.
Otherwise, on many machines, the size of the OpenMP thread pool will
change between MKL and our OpenMP enabled functions. The constant thread
creation and destruction results in worse performance and leaks memory
on GCC 5.4
The core autograd Variable, Function, and Engine no longer depend on the
Python API. This let's us implement functions in C++. In the future, we
can also multithread engine and release the GIL for most of the
non-Python backwards.
