Evaluation on SPEC2006

[PeimingLiu]

Hi,
I tried to evaluate Softbound on SPEC2006, but it showed a significant overhead (for bzip, 5x slower than native run). Is there any extra optimizations that I have to do to reduce the overhead?

Thanks

[santoshn]

It is possible that the checks are not getting inlined. Can you check if the checks and softboundcets wrappers are getting inlined?

[PeimingLiu]

Thanks, I inlined them, and the runtime overhead for bzip2 is now about 1.7x, is it a fair result?

[santoshn]

1.7X with spatial and temporal safety seems to be in right ballpark. That said, I have not updated this code base in 7 years. You may be interested in Tina Jung's work on eliminating bounds check in the SoftBound context; http://www.prog.uni-saarland.de/projects/pico/