Comment by wallscratch

Can anyone comment on how efficient the Warp code is compared to manually written / fine-tuned CUDA?