> nothing beats multithreaded C and vectorized assembly code

I think Terra would give you a run for your money:

http://theory.stanford.edu/~aiken/publications/papers/pldi13...

From the paper:

> Our Terra-based auto- tuner for BLAS routines performs within 20% of ATLAS, and our DSL for stencil computations runs 2.3x faster than hand-written C.

See also:

http://terralang.org/