In looking at:
Array a is a local array(3,3) of doubles, and ajj is a local array(6,6) of doubles.
It appears to me that the lea of the base address could be combined with the offset to omit the add of the offset following the lea (occurs twice).
Also, the mask in k1 is getting redundantly rebuilt to the same value.
IVF V16.0 update 1
Jim Dempsey
Link Copied
Here is even a worse case:
ajj is local array(6,6) of doubles, aj1 is local array of (6,6,n) of doubles (n unknown to routine)
The following could simply move a block of 36 doubles from a section of aj1 to ajj
Jim Dempsey
