Summary:
To illustrate the benefits of this commit, I'll use the time/iter I got from one of the JIT benchmarks on my machine.
| Run | Time |
|----------------------------------------------|-------------------------|
| No profiler | 45ms |
| With profiler | 56ms |
| Use `clock_gettime` instead of `std::chrono` | 48ms |
| Touch all pages on block allocation | 48ms (less jitter) |
| Use `const char*` instead of `std::string` | 47ms (even less jitter) |
Pull Request resolved: https://github.com/pytorch/pytorch/pull/11773
Differential Revision: D9886858
Pulled By: apaszke
fbshipit-source-id: 58f926f09e95df0b11ec687763a72b06b66991d0
Summary:
This eliminates the need for any heuristics regarding stack size limits.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/11534
Differential Revision: D9779866
Pulled By: resistor
fbshipit-source-id: 96753eead7904bbdc2869fb01f7bd42141032347
Summary:
Prior to this diff, there have been two ways of compiling the bulk of the torch codebase. There was no interaction between them - you had to pick one or the other.
1) with setup.py. This method
- used the setuptools C extension functionality
- worked on all platforms
- did not build test_jit/test_api binaries
- did not include the C++ api
- always included python functionality
- produced _C.so
2) with cpp_build. This method
- used CMake
- did not support Windows or ROCM
- was capable of building the test binaries
- included the C++ api
- did not build the python functionality
- produced libtorch.so
This diff combines the two.
1) cpp_build/CMakeLists.txt has become torch/CMakeLists.txt. This build
- is CMake-based
- works on all platforms
- builds the test binaries
- includes the C++ api
- does not include the python functionality
- produces libtorch.so
2) the setup.py build
- compiles the python functionality
- calls into the CMake build to build libtorch.so
- produces _C.so, which has a dependency on libtorch.so
In terms of code changes, this mostly means extending the cmake build to support the full variety of environments and platforms. There are also a small number of changes related to the fact that there are now two shared objects - in particular, windows requires annotating some symbols with dllimport/dllexport, and doesn't allow exposing thread_local globals directly.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/8792
Reviewed By: ezyang
Differential Revision: D8764181
Pulled By: anderspapitto
fbshipit-source-id: abec43834f739049da25f4583a0794b38eb0a94f
This changes type(tensor) to return `torch.Tensor` instead of
`torch.autograd.Variable`.
This requires a few implementation changes:
- torch.Tensor is now a regular Python class instead of a
pseudo-factory like torch.FloatTensor/torch.DoubleTensor
- torch.autograd.Variable is just a shell with a __new__ function.
Since no instanes are constructed it doesn't have any methods.
- Adds torch.get_default_dtype() since torch.Tensor.dtype returns
<attribute 'dtype' of 'torch._C._TensorBase' objects>
- Remove some uses of mega-header THP.h
- Use HANDLE_TH_ERRORS in functions that may throw
- Move NumPy includes to common header
- Delete unused allocator
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
* CUDA mode profiler fixes
* Enable multi-gpu CUDA tracing
We need to record per-device start events because event timing
comparison only works for events on the same device.
* Course-grained CPU-CUDA syncing of timelines
Record a __cuda_start event used to synchronize cuda/gpu timings.
This requires running some warm-up event records to ensure the
call to event record for the __cuda_start event doesn't take
longer than normal.
fix syncing
* fix cuda build and lint
* Add cudaEvent support to the profiler
This adds the ability to record cuda timings using cudaEventRecord
in the profiler. Since it doesn't require nvprof it is easier
to run than the nvprof path.
This also records a thread id for each event, which will make
tracing results easier to understand
* Add flow arrows from cpu to cuda event
* Fix no cuda build
* Review comments
* Move CUDA checks to one place
It is not an /expression/ we trace, but it is a /graph/: that is,
a closed expression which knows its parameters. Knowing the list
of parameters is helpful and helps remove a hack when interpreting.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Although ANF style developments traditionally stratifies syntactic
classes into atomic (Arg) and complex (Expr) expressions, where
atomic expressions could be variables, constants or lambdas, Zach has
successfully convinced me that we should do away with the variant here and
always require arguments to be variables. There are a few reasons for
this:
1) Tensor constants, not currently supported, could be modeled using a
"Constant" instruction, removing the need for them to be representable
directly inline. An inline constant is marginally more convenient
for peephole optimizations, but since we have gone full ANF, we are going
to need to be able to see across def-uses in any case, and it is not
too much worse to need to handle constants this way. By the way,
Swift Intermediate Language also made a similar choice, see
the slide on "Literal Instructions" in
http://llvm.org/devmtg/2015-10/slides/GroffLattner-SILHighLevelIR.pdf
2) Scalar constants, which are quite important for passing non-tensor
arguments to Python operators, are now stored out-of-band as NON
first-class values. This more closely matches the ToffeeIR design,
and makes it clear what parameters are "first class" (tensors only)
and which ones are not. However, we need to be able to unswizzle
the separate scalar/tensor lists into a unified list in the correct
format; this is what PyFunctionCConv is for.
Also, Locals got renamed into Tuple.
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>
* Implement BatchNorm double backwards as a python function called directly from C++.
This will be converted to C++ code once ATen is integrated with autograd.
* Some performance improvements via inplace ops and reusing calculations.
