https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70586
--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> --- There is currently no cost model. One easily could add a estimate_num_insns based one (with a --param) to bb_no_side_effects_p. Note that I originally invented ifcombine to replace LOGICAL_OP_NON_SHORT_CIRCUIT, that is, make sure what fold would do early we also do late and IIRC fold only checks for side-effects when transforming TRUTH_ANDIF to TRUTH_AND.
