sample_embedding_folder2/0746579.txt

Ticket Name: TDA2: High performance array transposition for EVE core

Query Text:
Part Number: TDA2 Hi, I have to transpose array for EVE next way: data1[y * width + y] to data2[y + x * height]. I try reales thes throw EDMA, but it is slow way. DMA doesn't work with TILER. What is the most efficient way to realise this procedure?

Responses:
Hi Danil, Using EDMA to do the transpose will not be efficient. You can use VCOP to do the transpose ( look for p_scatter store instruction in VCOP). If you are doing just transpose then DMA will become bottleneck so it would be advisable if you can combine the transpose with some more processing to utilize EVE optimally. Regards, Anshu