OT: Re: A very slow program

On 02/10/2024 13:44, Bonita Montero wrote:
> Ive developed a Sieve of Erastosthenes in C++20 with an unbeaten
> performance. With this code I get all primes <= 2 ^ 32 in 140ms
> on my AMD 7950X 16-core Zen4-system.
> It calculates first only the primes up to the square root of the
> maximum. The remaining bits are partitioned according to the num-
> ber of CPU-cores. Within each thread the bitmap is partitioned in
> parts that fit into the L2-cache (queried via CPUID).
> All operations are aligned on a cacheline-boundary to avoid false
> sharing and locking of shared words.

Unbeaten?

This isn't the same program you posted before over on comp.lang.c++, but 
it builds and runs fine on my machine.

for 4e9 primes my machine take nearly 10 seconds for 4294967295 primes. 
The one I posted before over there takes 1.5 seconds.

This version is if anything slightly slower than your other one.

You might like to have a look.

(And Keith, yes, I realise there is no point whatsoever in doing this... 
but it's been fun. First time in years I've written anything I'm not 
getting paid for!)

Andy