Open zyn810039594 opened 7 months ago
Hi,
They way that this is implemented in litex (for instance), is that there is an AXI to tilelink bridge which then goes toward the L2 cache.
So all the axi trafic has to go through l2.
The main issue is that converting Axi to Tilelink is tricky / bottlneck, as AXI ordering need to be preserved for similar ID, while tilelink is fully out of order. So it all depend how much efficiency / bandwidth / bridge implemention / axi ordering requirements you have
Hi! Is there anyway to read a block of sram that I have previously read but has been overwritten by another host(like a dma)? I use the Axi4 interface so there's not any cache coherency solution... Maybe there's someway to flush the cache?