[Bug target/115458] [15 regression] [RISC-V] ICE in lra_split_hard_reg_for, at lra-assigns.cc:1868 unable to find a register to spill since r15-518-g99b1daae18c095

[law at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115458

--- Comment #11 from Jeffrey A. Law <law at gcc dot gnu.org> ---
Hmm.  Are we somehow mis-computing the size of the mask operand thinking it
takes v0..v7 in the LMUL8 scenario?  If so that would mean this instruction
consumes the entire vector register file.  In isolation that should be
manageable, but it may become problematical if we have to do any spilling.

I guess the first question is why did we spill.  Does the .reload dump file
give any good clues?  Do we perhaps re-use the mask and end up with crossing
lifetimes for the mask?