On Sat, Feb 11, 2012 at 11:01 PM, Richard Henderson <r...@redhat.com> wrote:
>>>> * compare-elim.c (find_comparisons_in_bb): Eliminate only compares >>>> having mode compatible with the mode of previous compare. Substitute >>>> compare mode of previous compare with the mode, compatible >>>> with eliminated and previous compare. >>> >>> This patch is ok for 4.8. >> >> Unfortunately, we need to update all uses of flag register with a new, >> compatible mode, as well, similar to how compatible mode is handled in >> CSE2 pass with cse_condition_code_reg in cse.c > > We do? What subsequent pass really cares? Yes, please see an example in [1] how CSE2 pass handles this. I don't know if this is necessary, but when we merge arithmetic insn with compare, we _do_ update the flags users, in the same way as CSE2 pass. > What goes wrong leaving things as they are? x86 tolerates wrong modes in flags users, so as far as x86 is concerned, this is allowed. But I don't know if wrong, although compatible modes, should be generated from a generic pass. [1] http://gcc.gnu.org/ml/gcc-patches/2012-02/msg00547.htm Uros.