Re: discussion reset! thumb2/thumbee code as on armv7-a
On Thursday 15 July 2010 21:11:50 Paul Brook wrote:
> Choice of ISA is independent of the ABI. Generally speaking ARM code is
> still faster. This is another instance of where you want multiple builds
> of the same package within the same port, so users can choose between
> speed (-marm -O2) and size (-mthumb -Os) optimized builds.
That's still not entirely true. In a benchmark for Eigen3 library -with NEON
enabled- the binary (bench_gemm) compiled with -mthumb gave ~0.1GFLOPS more
(0.93, vs 0.83GFLOPS) than the one without -both on hardfp, btw. The reason is
simple, there simply was more room in the L1/L2 caches for data and the code
was equally fast to process them.
So, thumb is actually not a bad thing for speed either.