gallivm: Use lp_build_div instead of lp_build_mul + lp_build_rcp.

Single divide, so let lp_build_div decide how to implement this.

This will save a multiplication in architectures which don't have
a RCP intrinsic.
1 file changed