Closed kmp5VT closed 10 months ago
@kshitij-05 have a look at the remaining failures, all seem to be ZCP (no more CP failures), indeed perhaps sitting too close to the edge?
@evaleev @kshitij-05 I believe I was able to fix the issue. Looks like it was a near rank problem. I just turned up the rank slightly for these problems (I increased the rank by 2)
An issue found in calling
update_w_KRP
without MKL see https://github.com/ValeevGroup/BTAS/issues/167. To fix the problem I no longer call the flatten function. Now I just push the transpose work on thecontract
function. I also rewrote the Flatten code to work properly by taking an order-N tensor and flattening it to an order-3 tensor with dimensions(modes_before_n, n, modes_after_n)
and then I "permute" the tensor into the flattened order during the copy stage.