[PATCH] D31827: Add a flag that enables more aggressive load PRE

[Daniel Berlin via Phabricator via llvm-commits]

dberlin added a comment.

Any performance numbers?

Note that this probably ends up equivalent to this algorithm: http://www1.cse.wustl.edu/~cytron/cs531/Resources/Papers/valnum.pdf

In which case, we should do it in rank order to minimize the number of iterations it will take to float the loads upwards as far as they go.


https://reviews.llvm.org/D31827