Closed makslevental closed 4 days ago
Can we avoid adding these binary elf files used in the tests? E.g. generating them instead?
Obv I can but
Ok that I realize that doesn't make complete sense but basically what I'm saying is since all this stuff is so brittle, when something breaks, we have basically no way of knowing where exactly a break happens. Now usually you could narrow the tests by reproducing the environment but the issue for this PR was I have no idea what version of peano is on the runner (that's why one of the commits was just to print time stamps for that directory, which only revealed it's from December). So the only way to debug that was this extremely narrow unit test. Going forward we should obviously pin to a hash of peano and that'll be possible soon (when they're satisfied with https://github.com/Xilinx/llvm-aie/pull/109). At that point I'll be able to delete these.
Will move stuff around to remove binaries from commit.
This PR moves us to using Xilinx/aie-rt; in particular the
release/main_aig
branch which is part of the Linux Foundation upstreaming agreement/deal.Note, this one and https://github.com/nod-ai/iree-amd-aie/pull/501 are logically independent even though they touch some of the same files/lines. Whichever lands first, I'll just rebase the other.
cc @kumardeepakamd