> When we rescan the operands, we get a different set of V_MAY_DEFS, > specifically we lose the V_MAY_DEF for SFT.3_20. Why? It should be copying subvars to the new vectorizer variable too. At least, i believe i added that.