Linux Compiler Benchmarks Of LLVM Clang 3.5 vs. LLVM Clang 3.6-rc1

The first performance regression found was when running Rodinia and using the OpenMP CFD Solver. While LLVM/Clang doesn't yet have full OpenMP support, the result is still interesting in comparing from Clang 3.5 to Clang 3.6. The other Rodinia test cases did not regress on this compiler.

HMMer also seemed to slightly regress with the newer Clang code but there was a high deviation on the test's multiple runs under Clang 3.6-rc1 that doesn't make this a decisive call.

SciMark's micro-benchmark performance was slower on Clang 3.6-rc1.

Causing the drop in SciMark's performance seems to mostly come down to a regression in the Sparse Matrix Multiply test with the 3.6 result taking a nose dive.

John The Ripper's Blowfish test was slightly faster under Clang 3.6.