Is there a pass in LLVM that can optimize this pattern https://godbolt.org/z/hdKsj4nc4?

The expression a*b + a*c + a*b*d may be optimized into a*(b + c + b*d) saving one multiplication.

It looks like ReassociatePass cannot optimize this due to the single-use check for adding the expression into the tree. Is this a hard limitation?

@RKSimon can you please comment?