Fix: #107315
This PR enables dynamo to trace through the `pytree` API by inlining its functions. In
order to do so, a few details of `pytree` had to be changed.
In summary, this PR:
- Introduces `TreeSpecVariable` for representing `TreeSpec` instances
- Specializes `<type>.__bases__` call, returning a `TupleVariable`
- Enables the call to `id` builtin function for every variable that implements
`as_python_constant` method
- Specializes `ConstantVariable.call_method` for its (un)flatten functions
- Implements `UserDefinedObjectVariable.as_python_constant`
- Modifies `pytree` by:
- Make `SUPPORTED_NODES` a map of ids (instead of types) to `NodeDef`
- Removed `functools.wraps` function, since it can't be inlined
Pull Request resolved: https://github.com/pytorch/pytorch/pull/108533
Approved by: https://github.com/ezyang, https://github.com/voznesenskym
ghstack dependencies: #109201
This combines a bunch of python global state guards into a single C++ guard and switches to checking them 100% of the time. It also adds a few new guards for things that change inductor's behavior. Even though we are checking more things, I expect this to be much faster.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/108624
Approved by: https://github.com/anijain2305
**This PR is a 99% copy paste of Sam Gross** (@colesbury) work at https://github.com/pytorch/pytorch/pull/100642. Copied from there
--------
The NN_MODULE guard now subsumes guards on Module attributes. The check_fn will fail if the module attributes are changed (such as Module.training), parameters, submodules, and buffers are added or removed, and if fields are changed on the type itself.
This gives up specificity in the guard check -- if any field is changed the check_fn fails -- for faster overall checks.
-----
Pull Request resolved: https://github.com/pytorch/pytorch/pull/108528
Approved by: https://github.com/ezyang
**Motivation:**
When input FakeTensor to torch.compile has SymInt sizes (e.g. make_fx(opt_f, tracing_mode="symbolic"):
1. We cannot create a FakeTensor from that input in dynamo due to the SymInts.
2. We cannot check input tensors in guard check function and will abort due to tensor check calls sizes/strides.
For 1, we specialize the FakeTensor's SymInts using their hints. This is mostly safe since inputs mostly have concrete shapes and not computed from some DynamicOutputShape ops. We'll throw a data dependent error if the symint is unbacked.
For 2, we replace size/stride calls with the sym_* variants in TENSOR_CHECK guards' check function.
**Test Plan:**
See added tests.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/107662
Approved by: https://github.com/ezyang
The story here is relatively simple - when we go to wrap a tensor, we (1) ensure that it is a real, not fake tensor (2) check if we have seen it before. (3) If we have seen it, we create a positive alias guard and return the associated variable. If not, we proceed.
By short circuiting here, we avoid lifting it to a graph input, and guarantee that the only names passed to tensors are unique. This allows us to guard on the unique relationships (pyboject addresses, aka IDs, cannot match) to give us guards for negative aliases.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/104921
Approved by: https://github.com/jansel, https://github.com/ezyang
Tweaks the TENSOR_MATCH guard logic to avoid saving sizes / strides for the case of dynamic shapes. Instead, the dim() is stored, which is enough for both dense tensors and NTs.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/97896
Approved by: https://github.com/ezyang
Apply clang-tidy check modernize-use-emplace. This is slightly more efficient by using an inplace constructor and is the recommended style in parts of the codebase covered by clang-tidy. This just manually applies the check to rest of the codebase. Pinging @ezyang as this is related to my other PRs he reviewed like #89000
Pull Request resolved: https://github.com/pytorch/pytorch/pull/91077
Approved by: https://github.com/ezyang
Inductor internally models any `size=1` dimension as having `stride=0` to simplify indexing formulas (sympy will remove these terms from the expression).
This caused a bug in our generate stride assert in detectron2_maskrcnn_r_50_fpn, where we asserted the wrong stride of a size==1 dimension.
This fixes that bug, and moves size/stride assert logic to C++ which should be a small perf gain.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/87028
Approved by: https://github.com/anijain2305
