tilelink2: allow preemption of Fragmenter and WidthWidget

src/main/scala/uncore/tilelink2/Fragmenter.scala

@@ -191,7 +191,8 @@ class TLFragmenter(minSize: Int, maxSize: Int, alwaysMin: Boolean = false) exten
     val maxLgHints       = maxHints      .map(m => if (m == 0) lgMinSize else UInt(log2Ceil(m)))
 
     // Make the request Irrevocable
-    val in_a = Queue(in.a, 1, flow=true)
+    val repeat = Wire(Bool())
+    val in_a = Repeater(in.a, repeat)
 
     // If this is infront of a single manager, these become constants
     val find = manager.findFast(edgeIn.address(in_a.bits))
@@ -226,10 +227,8 @@ class TLFragmenter(minSize: Int, maxSize: Int, alwaysMin: Boolean = false) exten
 
     when (out.a.fire()) { gennum := new_gennum }
 
-    val delay = !aHasData && aFragnum =/= UInt(0)
-    out.a.valid := in_a.valid
-    in_a.ready := out.a.ready && !delay
-    out.a.bits := in_a.bits
+    repeat := !aHasData && aFragnum =/= UInt(0)
+    out.a <> in_a
     out.a.bits.addr_hi := in_a.bits.addr_hi | (~aFragnum << log2Ceil(minSize/beatBytes) & aOrigOH1 >> log2Ceil(beatBytes))
     out.a.bits.source := Cat(in_a.bits.source, aFragnum)
     out.a.bits.size := aFrag

src/main/scala/uncore/tilelink2/Repeater.scala

@@ -0,0 +1,37 @@
+// See LICENSE for license details.
+
+package uncore.tilelink2
+
+import Chisel._
+
+// A Repeater passes it's input to it's output, unless repeat is asserted.
+// When repeat is asserted, the Repeater copies the input and repeats it next cycle.
+class Repeater[T <: Data](gen: T) extends Module
+{
+  val io = new Bundle {
+    val repeat = Bool(INPUT)
+    val enq = Decoupled(gen).flip
+    val deq = Decoupled(gen)
+  }
+
+  val full = RegInit(Bool(false))
+  val saved = Reg(gen)
+
+  // When !full, a repeater is pass-through
+  io.deq.valid := io.enq.valid || full
+  io.enq.ready := io.deq.ready && !full
+  io.deq.bits := Mux(full, saved, io.enq.bits)
+
+  when (io.enq.fire() &&  io.repeat) { full := Bool(true); saved := io.enq.bits }
+  when (io.deq.fire() && !io.repeat) { full := Bool(false) }
+}
+
+object Repeater
+{
+  def apply[T <: Data](enq: DecoupledIO[T], repeat: Bool): DecoupledIO[T] = {
+    val repeater = Module(new Repeater(enq.bits))
+    repeater.io.repeat := repeat
+    repeater.io.enq := enq
+    repeater.io.deq
+  }
+}

src/main/scala/uncore/tilelink2/WidthWidget.scala

@@ -120,9 +120,7 @@ class TLWidthWidget(innerBeatBytes: Int) extends LazyModule
       val dataOut = if (edgeIn.staticHasData(in.bits) == Some(false)) UInt(0) else Mux1H(select, dataSlices)
       val maskOut = Mux1H(select, maskSlices)
 
-      in.ready := out.ready && last
-      out.valid := in.valid
-      out.bits := in.bits
+      out <> in
       edgeOut.data(out.bits) := dataOut
 
       out.bits match {
@@ -133,6 +131,9 @@ class TLWidthWidget(innerBeatBytes: Int) extends LazyModule
       }
 
       // addr_lo gets truncated automagically
+
+      // Repeat the input if we're not last
+      !last
     }
     
     def splice[T <: TLDataChannel](edgeIn: TLEdge, in: DecoupledIO[T], edgeOut: TLEdge, out: DecoupledIO[T]) = {
@@ -141,7 +142,8 @@ class TLWidthWidget(innerBeatBytes: Int) extends LazyModule
         out <> in
       } else if (edgeIn.manager.beatBytes > edgeOut.manager.beatBytes) {
         // split input to output
-        split(edgeIn, Queue(in, 1, flow=true), edgeOut, out)
+        val repeat = Wire(Bool())
+        repeat := split(edgeIn, Repeater(in, repeat), edgeOut, out)
       } else {
         // merge input to output
         merge(edgeIn, in, edgeOut, out)