TiraLib Python: A Tiramisu Compiler Python Frontend For Loading Tiramisu Programs and Building and Executing Schedules
TiraLib Python is a Python frontend for the Tiramisu compiler. It allows users to build schedules for Tiramisu programs and execute them. It also allows users to generate C++ code for their Tiramisu schedules and execute it.
To install TiraLib Python, you need to install the Tiramisu compiler first. Please follow the instructions here.
Then, you can install tiralib Python by cloning this repository and running the following command:
cd tiralib
poetry install
You can also install tiralib Python as a library from github by running the following command:
poetry add git+https://github.com/Tiramisu-Compiler/TiraLib
or using pip:
pip install git+https://github.com/Tiramisu-Compiler/TiraLib
If you installed TiraLib using poetry in a .venv environement then you need to activate the virtual environment created by Poetry:
poetry shell
TiraLib Python uses a configuration file to specify the paths to the Tiramisu compiler and the Tiramisu runtime. The configuration file is named config.yaml
and should be placed in the root directory of the project. The configuration file should have the following format (more details can be found in the config module):
tiramisu:
is_new_tiramisu: False
env_vars:
CXX: "${CXX}"
CC: "${CC}"
TIRAMISU_ROOT: "path/to/tiramisu"
TIRAMISU_TIRALIBCPP_PATH: "path/to/tiralibcpp/installation" # Optional
CONDA_ENV: "path/to/conda/env"
LD_LIBRARY_PATH: "$CONDA_ENV/lib:${TIRAMISU_ROOT}/3rdParty/Halide/build/src:${TIRAMISU_ROOT}/3rdParty/llvm/build/lib:${TIRAMISU_ROOT}/build:${TIRAMISU_ROOT}/3rdParty/isl/build/lib"
A config.yaml.example
file is provided in the root directory of the project. You can use it as a template for your configuration file.
Before running any TiraLib Python code, you need to load the configuration file using the following code:
from tiralib.config.config import BaseConfig
BaseConfig.init()
To load a Tiramisu program, you need to create a TiramisuProgram
object and pass the path to the Tiramisu program to its from_file
constructor, and set the load_annotations
and load_tree
parameters to True
if you want to load the annotations and the AST tree of the Tiramisu program respectively:
from tiralib.tiramisu import TiramisuProgram, Schedule, tiramisu_actions
from tiralib.config.config import BaseConfig
BaseConfig.init()
tiramisu_program = TiramisuProgram.from_file(
"./examples/function_blur_MINI_generator.cpp", load_annotations=True, load_tree=True
)
print(tiramisu_program.tree)
To build a schedule for a Tiramisu program, you need to create a Schedule
object and pass the TiramisuProgram
object to its constructor:
from tiralib.tiramisu import TiramisuProgram, Schedule, tiramisu_actions
from tiralib.config.config import BaseConfig
BaseConfig.init()
tiramisu_program = TiramisuProgram.from_file(
"./examples/function_blur_MINI_generator.cpp", load_annotations=True, load_tree=True
)
schedule = Schedule(tiramisu_program)
tiralib Python provides a set of code transformations that can be used to build schedules for Tiramisu programs. These transformations are implemented as TiramisuAction
objects.
To add a transformation to a schedule, you need to call the add_optimizations
method of the Schedule
object and pass the TiramisuAction
object to it:
from tiralib.tiramisu import TiramisuProgram, Schedule, tiramisu_actions
from tiralib.config.config import BaseConfig
BaseConfig.init()
tiramisu_program = TiramisuProgram.from_file(
"./examples/function_blur_MINI_generator.cpp", load_annotations=True, load_tree=True
)
schedule = Schedule(tiramisu_program)
schedule.add_optimizations([tiramisu_actions.Parallelization([("comp_blur", 0)])])
You can find the list of all the transformations implemented in tiralib Python here
To check the legality of a schedule, you need to call the is_legal
method of the Schedule
object:
from tiralib.tiramisu import TiramisuProgram, Schedule, tiramisu_actions
from tiralib.config.config import BaseConfig
BaseConfig.init()
tiramisu_program = TiramisuProgram.from_file(
"./examples/function_blur_MINI_generator.cpp", load_annotations=True, load_tree=True
)
schedule = Schedule(tiramisu_program)
schedule.add_optimizations([tiramisu_actions.Parallelization([("comp_blur", 0)])])
if schedule.is_legal():
print("The schedule is legal")
else:
print("The schedule is illegal")
To execute a schedule, you need to call the execute
method of the Schedule
object:
from tiralib.tiramisu import TiramisuProgram, Schedule, tiramisu_actions
from tiralib.config.config import BaseConfig
BaseConfig.init()
tiramisu_program = TiramisuProgram.from_file(
"./examples/function_blur_MINI_generator.cpp", load_annotations=True, load_tree=True
)
schedule = Schedule(tiramisu_program)
schedule.add_optimizations([tiramisu_actions.Parallelization([("comp_blur", 0)])])
execution_times = schedule.execute()
print(execution_times)
To run the tests, you need to activate the virtual environment created by Poetry:
poetry shell
Then, you can run the tests using the following command:
pytest
To run the tests and generate the coverage report, you need to activate the virtual environment created by Poetry:
poetry shell
Then, you can run the tests using the following command:
coverage run -m pytest
Finally, you can generate the coverage report using the following command:
coverage report
For HTML coverage report, you can use the following command:
coverage html --include="tiralib/**/*"
The library uses the ruff code formatter. To format the code, you need to activate the virtual environment created by Poetry:
poetry shell
Then, you can format the code using the following command:
ruff format .