Closed ambrad closed 6 years ago
Has a GPU bug, but can chase that down later. [edit: never mind; hadn't updated repo on GPU platform]
Closing so I don't clutter the board.
Tests suggest that although this branch is slower on GPU, it tends to be faster on HSW/KNL. So I'll reopen for consideration. I think there's a memory uninit bug or something like that in here b/c tests occasionally fail, so that has to be cleaned up if we decide to go with this branch.
Closing but keeping the branch.
On HSW, this may be the best combination. Need to evaluate on other platforms.