History

Sean Silva 39d50ccf0d Add end-to-end testing framework for TorchScript. The E2E tests can be run with ``` npcpy frontends/pytorch/e2e_testing/torchscript/main.py ``` This commit adds a couple items supporting that end, including new sugar for annotations (no more raw use of ClassAnnotator!). Recommended review order: 1. `frontends/pytorch/e2e_testing/torchscript/main.py` for the harness + `basic.py` in that directory for examples of tests. 2. Annotation sugar in `frontends/pytorch/python/torch_mlir/torchscript/annotations.py` and unittest in `frontends/pytorch/test/ivalue_import/annotations/sugar.py` 3. Global test registry / sugar in `frontends/pytorch/python/torch_mlir/torchscript/e2e_test/registry.py` 4. `frontends/pytorch/python/torch_mlir/torchscript/e2e_test/framework.py` for the meat of the testing framework (start at `run_tests`), and looking at the backend configs in `frontends/pytorch/python/torch_mlir/torchscript/e2e_test/configs` for examples of backends. This is likely the bulk of review time. 5. Unit tests of the framework logic in `frontends/pytorch/test/torchscript_e2e_test` There's TODO's scattered throughout, but this seems functional enough to start pulling stuff into and kicking the tires. A few missing pieces: 1. Marking test expected pass/fail per backend. 2. Figuring out how best to fit this into dev workflows. 3. IREE TestConfig. Also, forgive this Python newbie... Any advice on Python code structure / library design would be much appreciated.		2021-04-20 12:00:35 -07:00
..
csrc	[cleanup] Put the root class type for exportPath first.	2021-04-01 18:40:03 -07:00
docs	Add design sketch for aten fallback.	2020-11-24 18:13:35 -08:00
e2e_testing/torchscript	Add end-to-end testing framework for TorchScript.	2021-04-20 12:00:35 -07:00
examples	Add npcomp-verify-backend-contract pass.	2021-04-20 12:00:35 -07:00
python	Add end-to-end testing framework for TorchScript.	2021-04-20 12:00:35 -07:00
test	Add end-to-end testing framework for TorchScript.	2021-04-20 12:00:35 -07:00
utils	[cleanup] Put the root class type for exportPath first.	2021-04-01 18:40:03 -07:00
CMakeLists.txt	Delete old PyTorch 1.3 type dispatch oriented code paths.	2020-11-12 22:27:05 -08:00
LICENSE	Add pytorch interface to ATen Dialect (#30 )	2020-08-21 11:22:47 -07:00
README.md	Update README.	2021-03-30 11:33:33 -07:00

README.md

NPComp - PyTorch frontend integration

This directory contains optional components for interfacing PyTorch to NPComp. Integration is targeted at multiple levels:

Via program capture with a ATen pseudo-device.
Via IR-level integration with PyTorch (via tracing or scripting interfaces).
Interfaces to facilitate checking against reference implementations and verification.

In all situations, the target dialects are maintained in the outer project, along with their lowerings to common intermediate dialects and backends. This directory should be purely about interfacing with the PyTorch/LibTorch components for extracting and executing programs.

The code in this directory is intended to integrate tightly with pytorch, and follows the code style for pytorch. See the overall documentation for frontends for further details about code layout and integration philosophy. In particular, this directory exists to provide a working frontend to an MLIR based pytorch compilation flow and is not intended to be contributed to the LLVM monorepo. If the project is successful, it makes more sense to either break it out as an independent project that depends on LLVM/MLIR/npcomp or contribute it upstream to PyTorch. However, as it will be quite some time before the components are in a state to support such a dependency, it is being carried in-tree in the interim.

Program capture with a ATen dispatch capture.

Integration with a pseudo-device is typified by code like the following:

import torch
import torch_mlir

lhs = torch.rand(2, 3)
rhs = torch.rand(3, 4)

mb = torch_mlir.ModuleBuilder()
with mb.capture_function("mm", [lhs, rhs]) as f:
  result = torch.mm(lhs, rhs)
  f.returns([result])

mb.module.operation.print()

All operations that happen under the mb.capture_function context manager are intercepted via PyTorch's dispatcher, and an IR graph is constructed into the module held by the ModuleBuilder.

This technique has several advantages and disadvantages. For training use cases, this technique generates a backward path automatically using the same method that pytorch natively uses. The resulting graph also tends to be simpler, since it will not reflect conditionals in the original python code. Lastly, it is natural if MLIR is being used as a frontend target for an actual device of some sort. In this case, the MLIR could go through a device-specific lowering path and the resulting code run on a device. The implementation of this technique is largely modeled after pytorch/xla.