This is the mail archive of the
mailing list for the GCC project.
Re: [PATCH] Disable CSE skip-blocks
> > You mean operand selection? Sure, but that's completely driven by the
> > equivalence tables.
> No, I mean that, find_best_addr (which would go away if people cared to
> have a look at fwprop), fold_rtx (which does a lot even after fwprop),
> the 1000 lines of cse_insn, and so on.
Right, that's what I thought you meant and what I mean by my comment: if
you have the full equivalence table at the start of each basic block, ALL
of that code can function with complete effectiveness without either of
those switches being turned on.
(And I don't think that find_best_addr could or should go away since it's
HIGHLY target-specific and I don't think we want to put that level of knowlege
into tree-level optimizers, but that's a separate issue.)