Building Corpora#

Building a Corpus from an Individual description#

To build a corpus from an individual description, run the following command from the root directory of this repository:

PYTHONPATH="./" python3 ./llvm_ir_dataset_utils/tools/corpus_from_description.py \
  --base_dir=<path to build> \
  --corpus_dir=<path to corpus> \
  --corpus_description=<path to corpus description json>

The script will take the application description, clone the source, build the application with the appropriate flags, and then extract unoptimized IR from the build, placing it in a subdirectory of the directory passed to –corpus_dir in the ml-compiler-opt corpus format.