On 2023-07-21 16:25, Paul Eggert wrote:
Even mbcel strains GCC's capabilities to optimize. When I look at the
x86-64 code it generates I see clear opportunities for tighter code. If
I have time I'll fire off a suggestion or two to the GCC maintainers.
With luck these suggestions would also help mbu?iterf?.
I sent one suggestion here:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110823
This sped up the microbenchmark for mbcel on ASCII by 55% on a newer
x86-64 platform, and by 74% on an older one. I assume mbu?iterf? would
also benefit.
Although I suspect the resulting machine code could be improved further,
there are diminishing returns....
