Any LLVM passes for integers type compaction?

A few colleagues of mine implemented an inter-procedural range analysis algorithm for LLVM 2.7 a few years back.
Their paper (see states a 13-14% reduction in variable size
for the SPEC CPU2006 benchmarks and 36% for the Stanford benchmarks.
Check out Google Code Archive - Long-term storage for Google Code Project Hosting. if you’re interested and feel free to email me if you have any questions.


Dear Henrique,

Thanks for reply, I think I saw this project before. We have an LLVM-based GPU toolchain, where variable size is usually one of the major performance factors.

  • D.

2013/5/7 Henrique Santos <>