Inverse permutation needs to be applied when converting unpack to DMA, not permutation. PR adds test with permutation P != P^{-1}, without which this bug is hidden. I've left a comment with that test explaining why this fix is required. Also, --cse is removed from the test lit command, to reduce noise.
Inverse permutation needs to be applied when converting unpack to DMA, not permutation. PR adds test with permutation P != P^{-1}, without which this bug is hidden. I've left a comment with that test explaining why this fix is required. Also,
--cse
is removed from the test lit command, to reduce noise.