Robin Dapp <[email protected]>於 2025年11月4日 週二，15:36寫道：
> > On 11/3/25 6:06 PM, KuanLin Chen wrote: > >> I'll rename it in the next version. > >> I'm curious why use_zero_stride_load should be 'false'. It seems to be > >> the trigger of 'define_insn_and_split > >> "*pred_strided_broadcast<mode>"'. > >> I would appreciate it if you could let me know the reason. > > To date we haven't seen a single micro-architecture where the the zero > > strided load idiom performs better than a scalar load followed by a > > vector broadcast. > > At this point we might even want to reconsider the default for "generic"? > However all the SiFive cores seem to want/prefer it. Kito is this still > correct? Sifive core has that optimization for part of the cores like x280, but not for p470/p670, and seems like Tenstorrent Ascalon also doing that optimization as well? (they set that on both LLVM and GCC). > > > -- > Regards > Robin > >
