Re: OT: Re: Sieve of Erastosthenes optimized to the max

Am 20.08.2024 um 18:55 schrieb Vir Campestris:

That's really odd. I too have an AMD - although my AMD Ryzen 5 3400G is chosen for low power, not speed. Surely it can't be because I use Linux, not Windows?

My code spends its time almost only in userspace. The compiler is clang
17.0.3, the clang-version currently shipped as an alternative to MSVC
with Visual Studio.

 I'm leaning towards thinking you may have a point on the mod30 code. There are more operations on the innermost loop with mod30, although the loop goes around fewer times. It also means that the store is forced to be byte wide - I find that my original odd-only code is significantly faster - about 30% - when the store is 64 bit wide rather than byte.

Make the algorithm cache-aware, i.e. calculate chunks which fit into
the L2-cache.

Speed of writing the primes out? I don't care. The only reason I put it in there was to allow me to check the output. If I did care I'd be putting a more efficient enumeration function too.

My output code isn't even parallelized and writes only a gigabyte per
second to the SSD.