search

stanford-ppl / spatial-lang

Spatial: "Specify Parameterized Accelerators Through Inordinately Abstract Language"
MIT License
99 stars 12 forks source link

Accel Inside Loop #259

Closed mattfel1 closed 6 years ago

mattfel1 commented 6 years ago

Should this kind of thing be supported? Right now, at least in vcs, something gets messed up in the generated accel. On the board, I don't think it will work today either because we aren't controlling the reset signal from sw and we reprogram the fpga only at the top of main. 

for (i <- 0 until 10) {
  Accel{
    ...
  }
  // Some CPU stuff
}

I don't think it would be unusual for someone with a relatively complicated app to want to do some CPU stuff between iterations of the accel. It just seems like we need to do a little busywork to get it working

dkoeplin commented 6 years ago

Yeah ideally we should support this kind of thing. That'll ultimately make it easier to figure out where Accel should go when we work on work distribution/allocation.

raghup17 commented 6 years ago

Agreed that we need this. I think fundamentally we don't have a distinction between a board-level reset (which wipes out the configuration) and a memory-mapped "design reset" that restores state elements to a clean state before Accel.

mattfel1 commented 6 years ago

Done. Codegen just didn't enter the RangeForeach node