search

stanford-ppl / spatial-lang

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

Dimension as Function of DSE Parameters #225

Closed mattfel1 closed 6 years ago

mattfel1 commented 6 years ago

Want to do something like this:

  val coltile = 32 (16 -> 16 -> 1280)

  def Convolution[T:Type:Num](output: DRAM2[T], 
                      input: DRAM2[T],
                      filter: LUT2[T],
                      colstride: scala.Int, rowstride: scala.Int,
                      mode: ConvMode): Unit = {
....
    val lineout = SRAM[T](coltile/colstride)
....
dkoeplin commented 6 years ago

SRAM should already be supported, implicit argIns for DRAM dimensions added on compile

dkoeplin commented 6 years ago

On develop now