Hi Peng,
implementing custom kernels aside, the main alternative probably is looping over the segments. If you have sorted segments and not too many (say 10-100), it should not be unreasonably inefficient.
Best regards
Thomas
Hi Peng,
implementing custom kernels aside, the main alternative probably is looping over the segments. If you have sorted segments and not too many (say 10-100), it should not be unreasonably inefficient.
Best regards
Thomas