...and use them to avoid creating obviously undef values as discussed in the post-commit thread for r347478. The diffs in vector div/rem show that we were missing real optimizations by creating bogus shift nodes. llvm-svn: 347502
2.3 KiB
2.3 KiB