patches for increased performance of matmul, dotprod, transpose

[Tim Prince]

Tim Prince wrote:

>  gcc might be considered deficient in that it does not automatically 
> perform scalar reductions (to move store instructions out of a loop, 

A C compiler requires appropriate restrict qualifiers to permit the 
desired automatic scalar reductions.  gcc doesn't take any advantage of 
restrict here, so my comment about "possible FIXME" is still somewhat 
valid.    In principle, the original code could get more advantage from 
restrict, particularly if vectorization kicked in.