https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121051
--- Comment #4 from Robin Dapp <rdapp at gcc dot gnu.org> --- I managed to somehow get the patch series to work on trunk but there's bit-rot already. We really shouldn't postpone that any further than next stage 1. The code improves (to around 100 insns vs 140 before) but is still nowhere near to where I'd like it to be and we two-three vector regs per slide. Going to experiment some more.
