4 ms·
... and yet in this article, the author beats by 10x a C compiled code with hand tuned assembly. (By using SIMD and unrolling, which the compiler did not. Grant
by makapuf 4y ago
... and yet in this article, the author beats by 10x a C compiled code with hand tuned assembly. (By using SIMD and unrolling, which the compiler did not. Granted linear compiler code is faster than hand made linear assembly)
- zokier 4y agoAuthor beats compiler because floating point constraints, with -ffast-math compiler vectorizes the code.. I don't have arm64 hardware to test the result, but its probably again pretty fast: https://gcc.godbolt.org/z/xvjY8P4cM https://gcc.godbolt.org/z/xvjY8P4cM