Dear LLVM Developers,
The following are the performance comparisons on some different versions of llvm I used to test my project (running openmp with 2 threads on Power8):
GCC/8.2.0 - host compiler: 17 Seconds
LLVM/12.0.0 - built with gcc/8.2.0: 27 Seconds
LLVM/11.0.0 - Downloaded Pre-built Binaries (PowerPC64LE Red Hat Enterprise Linux): 27 Seconds
LLVM/8.0.1 - Downloaded Pre-built Binaries ( Red Hat Enterprise Linux 7.4 ppc64le ): 17 Seconds
It is surprising to find that the older version of llvm/8.0.1 reaches its performance that is close to gcc/8.2.0, while the newer are almost twice as slower than gcc/8.2.0.
Any idea about these low performances of newer versions?