spatel added a comment. In https://reviews.llvm.org/D49606#1170448, @thakis wrote:
> Isn't implementing this in plain old C the nicest approach anyhow, even once > funnel shift exists? No, the primary reason for creating the intrinsic is that we can’t guarantee that we’ll recognize the pattern as ‘shld’ or ‘rotate’ in this C/IR form. That’s because the pattern can get split across basic blocks, so the backend can’t easily recognize it. Also, IIUC this patch only deals with the specific case of 64-bit values, so we still don’t have the functionality for other widths? I don’t object if this helps that one case, but we have made the investment in the IR intrinsics, so using them should be the general direction/goal. https://reviews.llvm.org/D49606 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
