StanfordAHA/garnet

stalling during config seems to break things

Closed this issue · 0 comments

Disclaimer: Tests were done with old memory tile that is currently in master, not with lake memory tile.

I haven't root caused this, I don't know where the bug is coming from. In testing with the global buffer attached, I did the following set of steps:

  • stall the glb and tiles
  • load conv_3_3 bitstream into glb
  • use parallel config to configure the tiles
  • load image input
  • configure global buffer io
  • unstall cgra
  • soft reset
  • start

Later, when running the application, I get an X back out from the data_valid output of Tile_X05_Y00 which is paired with the output data. If I remove the stall entirely, the valid signal behaves as expected.

This seems to suggest that something is perhaps not configuring during stall, but the design is too unwieldy for me to debug this. @kong0329 and I were looking at this and aren't sure what's going on.