[llbm] miBench dataset has optnone function attribute set

🐛 Bug

I am not sure whether it is a bug, or perhaps it is due to the dataset?

I run the e-greedy algorithm https://github.com/facebookresearch/CompilerGym/blob/development/leaderboard/llvm_instcount/e_greedy/e_greedy.py on mibench. The results seems does not make sense at all: all rewards are 0 in terms of instruction count.

I don’t think it is due the implementation of e_greedy.py. Actually I wrote one version of greedy algorithm and it has the same results.

Why this happen? any problem in the benchmark? should we add some notes about this in the website?

To Reproduce

Steps to reproduce the behavior:

1. python e_greedy.py --n=1 --epsilon=0 --test_dataset=mibench-v0 --leaderboard_results=greedy_mibench.csv

greedy_mibench.csv

Environment

Please fill in this checklist:

• CompilerGym: 0.2.1
• How you installed CompilerGym (conda, pip, source): pip
• OS: Ubuntu 18.04
• Python version: 3.6.9
• Build command you used (if compiling from source):
• GCC/clang version (if compiling from source):
• Bazel version (if compiling from source):
• Versions of any other relevant libraries:

[ChrisCummins]

No, just update references from mibench-v0 to mibench-v1.

Cheers, Chris

[yqtianust]

Thanks. I will have a try. Do I need to manually remove the cache of the old version mibench?