Name Date Size #Lines LOC

..--

README.mdH A D25-Apr-2025782 159

gen_data_pad_mm.pyH A D25-Apr-20254.6 KiB150117

gen_pad_mm_a100.shH A D25-Apr-2025111 62

gen_pad_mm_h100.shH A D25-Apr-2025111 62

generate_heuristic_pad_mm.shH A D25-Apr-2025959 3115

get_padmm_dataset.shH A D25-Apr-2025199 96

test_pad_mm.pyH A D25-Apr-20255.5 KiB138120

train_decision_pad_mm.pyH A D25-Apr-2025752 2817

train_pad_mm.pyH A D25-Apr-2025725 2817

train_regression_pad_mm.pyH A D25-Apr-2025717 2817

README.md

1If you just want to re-generate existing heuristics with already collected data for pad_mm for A100, run the following scripts:
2
3`bash get_padmm_dataset.sh # Downloads A100`
4`bash gen_pad_mm_a100.sh # Generates A100 heuristic`
5
6If you want to collect new data, or generate a heuristic for another GPU, use the `generate_heuristic_pad_mm.sh` script:
7First, go into the generate_heuristic_mm.sh and modify the variables according to the comments. Then, run the script to perform benchmarks and collect training data:
8
9`bash generate_heuristic_pad_mm.sh collect`
10
11This will collect training data on random inputs. Depending on how many GPUs you are using, this might take a day.
12Afterwards, run the script in order to learn the heuristic:
13
14`bash generate_heuristic_pad_mm.sh generate`
15