StanfordAHA/Halide-to-Hardware

Running garnet->halide->map->test in Docker container

jwkirchenbauer opened this issue · 1 comments

Previously, using the aha Docker image, on a generic linux server, with Verilator installed, I was able to run the garnet generation, halide compilation, map/pnr, and the test bench sequentially on a subset of the applications the Halide-to-Hardware repo.

The problem I now have is that in the last few months I think lot of the feature work related to clockwork has moved forward and I am no longer able to coax the commands (garnet->halide->map->test) into running in the docker container. I suspect that the accumulated changes have fully deprecated the ability to run anything outside of the specific dev env the team uses that includes xcelium and leverages the Cadence IP.

Here's a series of commands that used to run for harris/resnet_layer_gen/others, that no longer does:

# within aha docker container

apt update && apt install verilator

...

python garnet.py --width 32 --height 16 --verilog --interconnect-only --no-pd

aha halide apps/harris

aha map apps/harris --width 32 --height 16

aha test apps/harris

Previously I had applied fixes in garnet.py and tbg.py inspired by

StanfordAHA/aha#1058
StanfordAHA/garnet#452

Those no longer seem to help make it past errors.

How, if at all, is the glb/ global buffer related to running these types of tests? StanfordAHA/aha#1277

I think this issue makes more sense in
https://github.com/StanfordAHA/aha

Do you mind reposting there?