torch-mlir/README.md

76 lines
2.5 KiB
Markdown
Raw Normal View History

2020-04-27 06:50:23 +08:00
# npcomp - An aspirational MLIR based numpy compiler
2020-04-27 11:14:20 +08:00
This is a research prototype of MLIR dialects for representing
numpy programs, and a set of reference tracing/compiler tools.
2020-04-30 07:11:18 +08:00
The primary purpose at this point is to establish a solid modeling
of restricted Python programs and Numpy based computations in MLIR.
While this project will provide some reference implementations to prove
the design, the intention is to align this with the broader set of
tools that exist at this level of abstraction.
2020-04-27 06:50:23 +08:00
2020-04-30 01:55:35 +08:00
## Design Notes
As I work through things, I've been jotting down some design notes:
* [Type Extraction - April 15, 2020](https://gist.github.com/stellaraccident/ec1ab0f633cfca0a05866fd77705b4e4)
* [Ufunc modeling Part 1 - April 29, 2020](https://gist.github.com/stellaraccident/4fcd2a24a66b6588f92b22b2b8ab974f)
2020-05-06 04:49:05 +08:00
* [Array funcs and op granularity - May 5, 2020](https://gist.github.com/stellaraccident/2c11652cfdee1457921bc7c98807b462)
2020-04-30 01:55:35 +08:00
2020-04-27 07:32:10 +08:00
## Quick start
2020-04-27 06:50:23 +08:00
2020-04-27 07:32:10 +08:00
```
2020-04-30 07:33:48 +08:00
LLVM_VERSION=10
export CC=clang-$LLVM_VERSION
export CXX=clang++-$LLVM_VERSION
export LDFLAGS=-fuse-ld=$(which ld.lld-$LLVM_VERSION)
2020-04-27 07:32:10 +08:00
export LLVM_SRC_DIR=/path/to/llvm-project
2020-04-30 07:33:48 +08:00
# Check out last known good commit.
(cd $LLVM_SRC_DIR && git checkout 0c4aab27b3da05dd1b0c0c39472525325fda5e23)
2020-04-27 07:32:10 +08:00
./tools/install_mlir.sh
./tools/cmake_configure.sh
cd build
ninja
2020-04-30 08:05:45 +08:00
ninja check-npcomp-opt
# Note: currently, python tests run separately
2020-04-27 07:32:10 +08:00
./python/run_tests.py
2020-04-27 06:50:23 +08:00
```
2020-05-03 10:54:49 +08:00
### Things to look at:
* `python/npcomp/tracing/mlir_trace_test.py` : Simple test case of tracing a function to an MLIR module.
2020-04-27 06:50:23 +08:00
### Installing pybind11
The native extension relies on pybind11. In a perfect world, this could just
be installed with your system package manager. However, at least on
2020-04-27 11:14:20 +08:00
some tested versions of Ubuntu, the system package installed with broken cmake
files.
2020-04-27 06:50:23 +08:00
2020-04-27 11:14:20 +08:00
If this happens, you must install pybind11 from source.
2020-04-27 06:50:23 +08:00
### Building the python native library
```shell
# From the build directory
ninja NPCOMPNativePyExt
2020-04-27 11:14:20 +08:00
ninja check-npcomp
python3 ./python/run_tests.py
# Setup PYTHONPATH for interactive use.
2020-04-27 06:50:23 +08:00
export PYTHONPATH=$(pwd)/tools/npcomp/python
```
Notes:
* Python sources are symlinked to the output directory at configure time.
Adding sources will require a reconfigure. Editing should not.
* It is a very common issue to have both python 2.7 (aka. "python") and python
3.x (aka. "python3") on a system at a time (and we can only hope that one
day this ends). Since the native library at development time binds to a
specific version, if you try to run with a different python, you will get
an error about the "native" module not being found.
2020-04-27 11:14:20 +08:00