> Sadly, that showed a performance-regression; from 5227456 cycles > (with cc0) to 5238564 (CC_REG). I'll add a note about this > 0.21% regression to pr93372 and hope to fix it soon.
IIRC in a few cases the old way of eliminating redundant compares would find one more occurrence because the dbr pass had reordered something. I also used CoreMark for the Visium transition and recouped the (tiny) losses in the end. -- Eric Botcazou
