Investigate use of streaming SIMD extensions to compute TDE

coryan / jaybeams

JayBeams: A Project to have fun Coding, and maybe measure relative delays in market feeds

Apache License 2.0

2 stars 2 forks source link

Open GFariasR opened 7 years ago

GFariasR commented 7 years ago

we could vectorize computations, like the one below, using the SSE instructions. How does one do that? Is there any advantage?

for (std::size_t i = 0; i != tmpa_.num_elements(); ++i) {
    tmpa_.data()[i] = std::conj(tmpa_.data()[i]) * (tmpb_.data()[i]);
}

coryan commented 7 years ago

This link:

has some relevant bits around slide 25. Basically there is a cool addsub instruction that helps with complex number arithmetic.

Also check: