https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122767
--- Comment #7 from Alexandre Oliva <aoliva at gcc dot gnu.org> --- Created attachment 62870 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=62870&action=edit alternative without space overhead and likely speedup Having realized that it was just the compare function that needed improvement, I figured I'd extend the HARD_REG_SET API with something suitable for qsort compares. Avoiding the byte accesses and shuffling from the hashing function probably makes it faster, without any space overhead. Yay! Now, gotta wait for stage1, because this doesn't look like stage3 material :-(
