I know that Blitz ++ gets its performance plus the widespread use of expression patterns and template metaprograms. But at some point you wonβt be able to get more out of your code using these methods β you will have to multiply and summarize some pop-ups. At this stage, you can get the final result using highly optimized (especially for special architectures) BLAS procedures. Does the current Blitz ++ implementation affect BLAS procedures when possible?
source
share