Merge pull request #364 from ucb-bar/tl2-async-nodes

Tl2 async nodes

src/main/scala/uncore/tilelink2/Bundles.scala

@@ -3,7 +3,8 @@
 package uncore.tilelink2
 
 import Chisel._
-import chisel3.util.{Irrevocable, IrrevocableIO}
+import chisel3.util.{Irrevocable, IrrevocableIO, ReadyValidIO}
+import util.{AsyncQueueSource, AsyncQueueSink}
 
 abstract class GenericParameterizedBundle[T <: Object](val params: T) extends Bundle
 {
@@ -190,7 +191,7 @@ object TLBundle
   def apply(params: TLBundleParameters) = new TLBundle(params)
 }
 
-class IrrevocableSnoop[+T <: Data](gen: T) extends Bundle
+final class IrrevocableSnoop[+T <: Data](gen: T) extends Bundle
 {
   val ready = Bool()
   val valid = Bool()
@@ -232,3 +233,53 @@ object TLBundleSnoop
     out
   }
 }
+
+final class AsyncBundle[T <: Data](val depth: Int, gen: T) extends Bundle
+{
+  require (isPow2(depth))
+  val ridx = UInt(width = log2Up(depth)+1).flip
+  val widx = UInt(width = log2Up(depth)+1)
+  val mem  = Vec(depth, gen)
+  override def cloneType: this.type = new AsyncBundle(depth, gen).asInstanceOf[this.type]
+}
+
+object FromAsyncBundle
+{
+  def apply[T <: Data](x: AsyncBundle[T], sync: Int = 3): IrrevocableIO[T] = {
+    val sink = Module(new AsyncQueueSink(x.mem(0), x.depth, sync))
+    x.ridx := sink.io.ridx
+    sink.io.widx := x.widx
+    sink.io.mem  := x.mem
+    val out = Wire(Irrevocable(x.mem(0)))
+    out.valid := sink.io.deq.valid
+    out.bits := sink.io.deq.bits
+    sink.io.deq.ready := out.ready
+    out
+  }
+}
+
+object ToAsyncBundle
+{
+  def apply[T <: Data](x: ReadyValidIO[T], depth: Int = 8, sync: Int = 3): AsyncBundle[T] = {
+    val source = Module(new AsyncQueueSource(x.bits, depth, sync))
+    source.io.enq.valid := x.valid
+    source.io.enq.bits := x.bits
+    x.ready := source.io.enq.ready
+    val out = Wire(new AsyncBundle(depth, x.bits))
+    source.io.ridx := out.ridx
+    out.mem := source.io.mem
+    out.widx := source.io.widx
+    out
+  }
+}
+
+class TLAsyncBundleBase(params: TLAsyncBundleParameters) extends GenericParameterizedBundle(params)
+
+class TLAsyncBundle(params: TLAsyncBundleParameters) extends TLAsyncBundleBase(params)
+{
+  val a = new AsyncBundle(params.depth, new TLBundleA(params.base))
+  val b = new AsyncBundle(params.depth, new TLBundleB(params.base)).flip
+  val c = new AsyncBundle(params.depth, new TLBundleC(params.base))
+  val d = new AsyncBundle(params.depth, new TLBundleD(params.base)).flip
+  val e = new AsyncBundle(params.depth, new TLBundleE(params.base))
+}

src/main/scala/uncore/tilelink2/Crossing.scala

@@ -6,33 +6,63 @@ import Chisel._
 import chisel3.internal.sourceinfo.SourceInfo
 import util._
 
-class TLAsyncCrossing(depth: Int = 8, sync: Int = 3) extends LazyModule
+class TLAsyncCrossingSource(sync: Int = 3) extends LazyModule
 {
-  val node = TLIdentityNode()
+  val node = TLAsyncSourceNode()
 
   lazy val module = new LazyModuleImp(this) {
     val io = new Bundle {
       val in  = node.bundleIn
-      val in_clock  = Clock(INPUT)
-      val in_reset  = Bool(INPUT)
       val out = node.bundleOut
-      val out_clock = Clock(INPUT)
-      val out_reset = Bool(INPUT)
     }
 
-    // Transfer all TL2 bundles from/to the same domains
     ((io.in zip io.out) zip (node.edgesIn zip node.edgesOut)) foreach { case ((in, out), (edgeIn, edgeOut)) =>
-      out.a <> AsyncIrrevocableCrossing(io.in_clock, io.in_reset, in.a, io.out_clock, io.out_reset, depth, sync)
+      val bce = edgeIn.manager.anySupportAcquire && edgeIn.client.anySupportProbe
-      in.d <> AsyncIrrevocableCrossing(io.out_clock, io.out_reset, out.d, io.in_clock, io.in_reset, depth, sync)
+      val depth = edgeOut.manager.depth
 
-      if (edgeOut.manager.anySupportAcquire && edgeOut.client.anySupportProbe) {
+      out.a := ToAsyncBundle(in.a, depth, sync)
-        in.b <> AsyncIrrevocableCrossing(io.out_clock, io.out_reset, out.b, io.in_clock, io.in_reset, depth, sync)
+      in.d := FromAsyncBundle(out.d, sync)
-        out.c <> AsyncIrrevocableCrossing(io.in_clock, io.in_reset, in.c, io.out_clock, io.out_reset, depth, sync)
+
-        out.e <> AsyncIrrevocableCrossing(io.in_clock, io.in_reset, in.e, io.out_clock, io.out_reset, depth, sync)
+      if (bce) {
+        in.b := FromAsyncBundle(out.b, sync)
+        out.c := ToAsyncBundle(in.c, depth, sync)
+        out.e := ToAsyncBundle(in.e, depth, sync)
       } else {
         in.b.valid := Bool(false)
         in.c.ready := Bool(true)
         in.e.ready := Bool(true)
+        out.b.ridx := UInt(0)
+        out.c.widx := UInt(0)
+        out.e.widx := UInt(0)
+      }
+    }
+  }
+}
+
+class TLAsyncCrossingSink(depth: Int = 8, sync: Int = 3) extends LazyModule
+{
+  val node = TLAsyncSinkNode(depth)
+
+  lazy val module = new LazyModuleImp(this) {
+    val io = new Bundle {
+      val in  = node.bundleIn
+      val out = node.bundleOut
+    }
+
+    ((io.in zip io.out) zip (node.edgesIn zip node.edgesOut)) foreach { case ((in, out), (edgeIn, edgeOut)) =>
+      val bce = edgeOut.manager.anySupportAcquire && edgeOut.client.anySupportProbe
+
+      out.a := FromAsyncBundle(in.a, sync)
+      in.d := ToAsyncBundle(out.d, depth, sync)
+
+      if (bce) {
+        in.b := ToAsyncBundle(out.b, depth, sync)
+        out.c := FromAsyncBundle(in.c, sync)
+        out.e := FromAsyncBundle(in.e, sync)
+      } else {
+        in.b.widx := UInt(0)
+        in.c.ridx := UInt(0)
+        in.e.ridx := UInt(0)
         out.b.ready := Bool(true)
         out.c.valid := Bool(false)
         out.e.valid := Bool(false)
@@ -41,6 +71,44 @@ class TLAsyncCrossing(depth: Int = 8, sync: Int = 3) extends LazyModule
   }
 }
 
+class TLAsyncCrossing(depth: Int = 8, sync: Int = 3) extends LazyModule
+{
+  val nodeIn = TLInputNode()
+  val nodeOut = TLOutputNode()
+
+  val source = LazyModule(new TLAsyncCrossingSource(sync))
+  val sink = LazyModule(new TLAsyncCrossingSink(depth, sync))
+
+  val _    = (sink.node := source.node) // no monitor
+  val in   = (source.node := nodeIn)
+  val out  = (nodeOut := sink.node)
+
+  lazy val module = new LazyModuleImp(this) {
+    val io = new Bundle {
+      val in        = nodeIn.bundleIn
+      val in_clock  = Clock(INPUT)
+      val in_reset  = Bool(INPUT)
+      val out       = nodeOut.bundleOut
+      val out_clock = Clock(INPUT)
+      val out_reset = Bool(INPUT)
+    }
+
+    source.module.clock := io.in_clock
+    source.module.reset := io.in_reset
+    in.foreach { lm =>
+      lm.module.clock := io.in_clock
+      lm.module.reset := io.in_reset
+    }
+
+    sink.module.clock := io.out_clock
+    sink.module.reset := io.out_reset
+    out.foreach { lm =>
+      lm.module.clock := io.out_clock
+      lm.module.reset := io.out_reset
+    }
+  }
+}
+
 /** Synthesizeable unit tests */
 import unittest._
 
@@ -51,8 +119,8 @@ class TLRAMCrossing extends LazyModule {
   val cross = LazyModule(new TLAsyncCrossing)
 
   model.node := fuzz.node
-  cross.node := TLFragmenter(4, 256)(model.node)
+  cross.nodeIn := TLFragmenter(4, 256)(model.node)
-  val monitor = (ram.node := cross.node)
+  val monitor = (ram.node := cross.nodeOut)
 
   lazy val module = new LazyModuleImp(this) with HasUnitTestIO {
     io.finished := fuzz.module.io.finished

src/main/scala/uncore/tilelink2/Fuzzer.scala

@@ -224,8 +224,8 @@ class TLFuzzRAM extends LazyModule
   xbar2.node := TLAtomicAutomata()(model.node)
   ram2.node := TLFragmenter(16, 256)(xbar2.node)
   xbar.node := TLWidthWidget(16)(TLHintHandler()(xbar2.node))
-  cross.node := TLFragmenter(4, 256)(TLBuffer()(xbar.node))
+  cross.nodeIn := TLFragmenter(4, 256)(TLBuffer()(xbar.node))
-  val monitor = (ram.node := cross.node)
+  val monitor = (ram.node := cross.nodeOut)
   gpio.node := TLFragmenter(4, 32)(TLBuffer()(xbar.node))
 
   lazy val module = new LazyModuleImp(this) with HasUnitTestIO {

src/main/scala/uncore/tilelink2/IntNodes.scala

@@ -23,13 +23,13 @@ object IntRange
 
 case class IntSourceParameters(
   range:    IntRange,
-  nodePath: Seq[IntBaseNode] = Seq())
+  nodePath: Seq[BaseNode] = Seq())
 {
   val name = nodePath.lastOption.map(_.lazyModule.name).getOrElse("disconnected")
 }
 
 case class IntSinkParameters(
-  nodePath: Seq[IntBaseNode] = Seq())
+  nodePath: Seq[BaseNode] = Seq())
 {
   val name = nodePath.lastOption.map(_.lazyModule.name).getOrElse("disconnected")
 }
@@ -49,8 +49,8 @@ case class IntEdge(source: IntSourcePortParameters, sink: IntSinkPortParameters)
 
 object IntImp extends NodeImp[IntSourcePortParameters, IntSinkPortParameters, IntEdge, IntEdge, Vec[Bool]]
 {
-  def edgeO(po: IntSourcePortParameters, pi: IntSinkPortParameters): IntEdge = IntEdge(po, pi)
+  def edgeO(pd: IntSourcePortParameters, pu: IntSinkPortParameters): IntEdge = IntEdge(pd, pu)
-  def edgeI(po: IntSourcePortParameters, pi: IntSinkPortParameters): IntEdge = IntEdge(po, pi)
+  def edgeI(pd: IntSourcePortParameters, pu: IntSinkPortParameters): IntEdge = IntEdge(pd, pu)
   def bundleO(eo: Seq[IntEdge]): Vec[Vec[Bool]] = {
     if (eo.isEmpty) Vec(0, Vec(0, Bool())) else
     Vec(eo.size, Vec(eo.map(_.source.num).max, Bool()))
@@ -60,18 +60,17 @@ object IntImp extends NodeImp[IntSourcePortParameters, IntSinkPortParameters, In
     Vec(ei.size, Vec(ei.map(_.source.num).max, Bool())).flip
   }
 
-  def connect(bo: => Vec[Bool], eo: => IntEdge, bi: => Vec[Bool], ei: => IntEdge)(implicit sourceInfo: SourceInfo): (Option[LazyModule], () => Unit) = {
+  def connect(bo: => Vec[Bool], bi: => Vec[Bool], ei: => IntEdge)(implicit sourceInfo: SourceInfo): (Option[LazyModule], () => Unit) = {
     (None, () => {
-      require (eo == ei)
       // Cannot use bulk connect, because the widths could differ
       (bo zip bi) foreach { case (o, i) => i := o }
     })
   }
 
-  override def mixO(po: IntSourcePortParameters, node: IntBaseNode): IntSourcePortParameters =
+  override def mixO(pd: IntSourcePortParameters, node: OutwardNode[IntSourcePortParameters, IntSinkPortParameters, Vec[Bool]]): IntSourcePortParameters =
-   po.copy(sources = po.sources.map  { s => s.copy (nodePath = node +: s.nodePath) })
+   pd.copy(sources = pd.sources.map  { s => s.copy (nodePath = node +: s.nodePath) })
-  override def mixI(pi: IntSinkPortParameters,  node: IntBaseNode): IntSinkPortParameters =
+  override def mixI(pu: IntSinkPortParameters, node: InwardNode[IntSourcePortParameters, IntSinkPortParameters, Vec[Bool]]): IntSinkPortParameters =
-   pi.copy(sinks   = pi.sinks.map    { s => s.copy (nodePath = node +: s.nodePath) })
+   pu.copy(sinks   = pu.sinks.map    { s => s.copy (nodePath = node +: s.nodePath) })
 }
 
 case class IntIdentityNode() extends IdentityNode(IntImp)

src/main/scala/uncore/tilelink2/LazyModule.scala

@@ -9,18 +9,13 @@ abstract class LazyModule
 {
   protected[tilelink2] var bindings = List[() => Unit]()
   protected[tilelink2] var children = List[LazyModule]()
-  protected[tilelink2] var nodes = List[RootNode]()
+  protected[tilelink2] var nodes = List[BaseNode]()
   protected[tilelink2] var info: SourceInfo = UnlocatableSourceInfo
   protected[tilelink2] val parent = LazyModule.stack.headOption
 
   LazyModule.stack = this :: LazyModule.stack
   parent.foreach(p => p.children = this :: p.children)
 
-  // Use as: connect(source -> sink, source2 -> sink2, ...)
-  def connect[PO, PI, EO, EI, B <: Data](edges: (BaseNode[PO, PI, EO, EI, B], BaseNode[PO, PI, EO, EI, B])*)(implicit sourceInfo: SourceInfo) = {
-    edges.foreach { case (source, sink) => sink := source }
-  }
-
   def name = getClass.getName.split('.').last
   def line = sourceLine(info)
 

src/main/scala/uncore/tilelink2/Nodes.scala

@@ -6,23 +6,33 @@ import Chisel._
 import scala.collection.mutable.ListBuffer
 import chisel3.internal.sourceinfo.SourceInfo
 
-// PI = PortInputParameters
+// DI = Downwards flowing Parameters received on the inner side of the node
-// PO = PortOutputParameters
+// UI = Upwards   flowing Parameters generated by the inner side of the node
-// EI = EdgeInput
+// EI = Edge Parameters describing a connection on the inner side of the node
-// EO = EdgeOutput
+// BI = Bundle type used when connecting to the inner side of the node
-abstract class NodeImp[PO, PI, EO, EI, B <: Data]
+trait InwardNodeImp[DI, UI, EI, BI <: Data]
 {
-  def edgeO(po: PO, pi: PI): EO
+  def edgeI(pd: DI, pu: UI): EI
-  def edgeI(po: PO, pi: PI): EI
+  def bundleI(ei: Seq[EI]): Vec[BI]
-  def bundleO(eo: Seq[EO]): Vec[B]
+  def mixI(pu: UI, node: InwardNode[DI, UI, BI]): UI = pu
-  def bundleI(ei: Seq[EI]): Vec[B]
+  def connect(bo: => BI, bi: => BI, e: => EI)(implicit sourceInfo: SourceInfo): (Option[LazyModule], () => Unit)
-  def connect(bo: => B, eo: => EO, bi: => B, ei: => EI)(implicit sourceInfo: SourceInfo): (Option[LazyModule], () => Unit)
-  // If you want to track parameters as they flow through nodes, overload these:
-  def mixO(po: PO, node: BaseNode[PO, PI, EO, EI, B]): PO = po
-  def mixI(pi: PI, node: BaseNode[PO, PI, EO, EI, B]): PI = pi
 }
 
-abstract class RootNode
+// DO = Downwards flowing Parameters generated by the outner side of the node
+// UO = Upwards   flowing Parameters received on the outner side of the node
+// EO = Edge Parameters describing a connection on the outer side of the node
+// BO = Bundle type used when connecting to the outer side of the node
+trait OutwardNodeImp[DO, UO, EO, BO <: Data]
+{
+  def edgeO(pd: DO, pu: UO): EO
+  def bundleO(eo: Seq[EO]): Vec[BO]
+  def mixO(pd: DO, node: OutwardNode[DO, UO, BO]): DO = pd
+}
+
+abstract class NodeImp[D, U, EO, EI, B <: Data]
+  extends Object with InwardNodeImp[D, U, EI, B] with OutwardNodeImp[D, U, EO, B]
+
+abstract class BaseNode
 {
   // You cannot create a Node outside a LazyModule!
   require (!LazyModule.stack.isEmpty)
@@ -35,108 +45,146 @@ abstract class RootNode
   def colour = "blue"
   def omitGraphML = outputs.isEmpty && inputs.isEmpty
 
-  protected[tilelink2] def outputs: Seq[RootNode]
+  protected[tilelink2] def outputs: Seq[BaseNode]
-  protected[tilelink2] def inputs:  Seq[RootNode]
+  protected[tilelink2] def inputs:  Seq[BaseNode]
 }
 
-class BaseNode[PO, PI, EO, EI, B <: Data](imp: NodeImp[PO, PI, EO, EI, B])(
+trait InwardNode[DI, UI, BI <: Data] extends BaseNode
-  private val oFn: (Int, Seq[PO]) => Seq[PO],
-  private val iFn: (Int, Seq[PI]) => Seq[PI],
-  private val numPO: Range.Inclusive,
-  private val numPI: Range.Inclusive) extends RootNode
 {
-  // At least 0 ports must be supported
+  protected[tilelink2] val numPI: Range.Inclusive
-  require (!numPO.isEmpty, s"No number of outputs would be acceptable to ${name}${lazyModule.line}")
   require (!numPI.isEmpty, s"No number of inputs would be acceptable to ${name}${lazyModule.line}")
-  require (numPO.start >= 0, s"${name} accepts a negative number of outputs${lazyModule.line}")
   require (numPI.start >= 0, s"${name} accepts a negative number of inputs${lazyModule.line}")
 
-  val noOs = numPO.size == 1 && numPO.contains(0)
+  private val accPI = ListBuffer[(Int, OutwardNode[DI, UI, BI])]()
-  val noIs = numPI.size == 1 && numPI.contains(0)
-
-  private val accPO = ListBuffer[(Int, BaseNode[PO, PI, EO, EI, B])]()
-  private val accPI = ListBuffer[(Int, BaseNode[PO, PI, EO, EI, B])]()
-  private var oRealized  = false
   private var iRealized = false
 
-  private def reqO() = require(numPO.contains(accPO.size), s"${name} has ${accPO.size} outputs, expected ${numPO}${lazyModule.line}")
+  protected[tilelink2] def iPushed = accPI.size
+  protected[tilelink2] def iPush(index: Int, node: OutwardNode[DI, UI, BI])(implicit sourceInfo: SourceInfo) {
+    val info = sourceLine(sourceInfo, " at ", "")
+    val noIs = numPI.size == 1 && numPI.contains(0)
+    require (!noIs, s"${name}${lazyModule.line} was incorrectly connected as a sink" + info)
+    require (!iRealized, s"${name}${lazyModule.line} was incorrectly connected as a sink after it's .module was used" + info)
+    accPI += ((index, node))
+  }
+
   private def reqI() = require(numPI.contains(accPI.size), s"${name} has ${accPI.size} inputs, expected ${numPI}${lazyModule.line}")
-  protected def reqE(o: Int, i: Int) = require(i == o, s"${name} has ${i} inputs and ${o} outputs; they must match${lazyModule.line}")
+  protected[tilelink2] lazy val iPorts = { iRealized = true; reqI(); accPI.result() }
 
-  private lazy val oPorts = { oRealized = true; reqO(); accPO.result() }
+  protected[tilelink2] val iParams: Seq[UI]
-  private lazy val iPorts = { iRealized = true; reqI(); accPI.result() }
+  protected[tilelink2] def iConnect: Vec[BI]
+}
 
+trait OutwardNode[DO, UO, BO <: Data] extends BaseNode
+{
+  protected[tilelink2] val numPO: Range.Inclusive
+  require (!numPO.isEmpty, s"No number of outputs would be acceptable to ${name}${lazyModule.line}")
+  require (numPO.start >= 0, s"${name} accepts a negative number of outputs${lazyModule.line}")
+
+  private val accPO = ListBuffer[(Int, InwardNode [DO, UO, BO])]()
+  private var oRealized = false
+
+  protected[tilelink2] def oPushed = accPO.size
+  protected[tilelink2] def oPush(index: Int, node: InwardNode [DO, UO, BO])(implicit sourceInfo: SourceInfo) {
+    val info = sourceLine(sourceInfo, " at ", "")
+    val noOs = numPO.size == 1 && numPO.contains(0)
+    require (!noOs, s"${name}${lazyModule.line} was incorrectly connected as a source" + info)
+    require (!oRealized, s"${name}${lazyModule.line} was incorrectly connected as a source after it's .module was used" + info)
+    accPO += ((index, node))
+  }
+
+  private def reqO() = require(numPO.contains(accPO.size), s"${name} has ${accPO.size} outputs, expected ${numPO}${lazyModule.line}")
+  protected[tilelink2] lazy val oPorts = { oRealized = true; reqO(); accPO.result() }
+
+  protected[tilelink2] val oParams: Seq[DO]
+  protected[tilelink2] def oConnect: Vec[BO]
+}
+
+class MixedNode[DI, UI, EI, BI <: Data, DO, UO, EO, BO <: Data](
+  inner: InwardNodeImp [DI, UI, EI, BI],
+  outer: OutwardNodeImp[DO, UO, EO, BO])(
+  private val dFn: (Int, Seq[DI]) => Seq[DO],
+  private val uFn: (Int, Seq[UO]) => Seq[UI],
+  protected[tilelink2] val numPO: Range.Inclusive,
+  protected[tilelink2] val numPI: Range.Inclusive)
+  extends BaseNode with InwardNode[DI, UI, BI] with OutwardNode[DO, UO, BO]
+{
+  // meta-data for printing the node graph
   protected[tilelink2] def outputs = oPorts.map(_._2)
   protected[tilelink2] def inputs  = iPorts.map(_._2)
 
-  private lazy val oParams : Seq[PO] = {
+  private def reqE(o: Int, i: Int) = require(i == o, s"${name} has ${i} inputs and ${o} outputs; they must match${lazyModule.line}")
-    val o = oFn(oPorts.size, iPorts.map{ case (i, n) => n.oParams(i) })
+  protected[tilelink2] lazy val oParams: Seq[DO] = {
+    val o = dFn(oPorts.size, iPorts.map { case (i, n) => n.oParams(i) })
     reqE(oPorts.size, o.size)
-    o.map(imp.mixO(_, this))
+    o.map(outer.mixO(_, this))
   }
-  private lazy val iParams : Seq[PI] = {
+  protected[tilelink2] lazy val iParams: Seq[UI] = {
-    val i = iFn(iPorts.size, oPorts.map{ case (o, n) => n.iParams(o) })
+    val i = uFn(iPorts.size, oPorts.map { case (o, n) => n.iParams(o) })
     reqE(i.size, iPorts.size)
-    i.map(imp.mixI(_, this))
+    i.map(inner.mixI(_, this))
   }
 
-  lazy val edgesOut = (oPorts zip oParams).map { case ((i, n), o) => imp.edgeO(o, n.iParams(i)) }
+  lazy val edgesOut = (oPorts zip oParams).map { case ((i, n), o) => outer.edgeO(o, n.iParams(i)) }
-  lazy val edgesIn  = (iPorts zip iParams).map { case ((o, n), i) => imp.edgeI(n.oParams(o), i) }
+  lazy val edgesIn  = (iPorts zip iParams).map { case ((o, n), i) => inner.edgeI(n.oParams(o), i) }
 
-  lazy val bundleOut = imp.bundleO(edgesOut)
+  lazy val bundleOut = outer.bundleO(edgesOut)
-  lazy val bundleIn  = imp.bundleI(edgesIn)
+  lazy val bundleIn  = inner.bundleI(edgesIn)
 
-  def connectOut = bundleOut
+  def oConnect = bundleOut
-  def connectIn = bundleIn
+  def iConnect = bundleIn
 
-  def := (y: BaseNode[PO, PI, EO, EI, B])(implicit sourceInfo: SourceInfo): Option[LazyModule] = {
+  // connects the outward part of a node with the inward part of this node
+  def := (y: OutwardNode[DI, UI, BI])(implicit sourceInfo: SourceInfo): Option[LazyModule] = {
     val x = this // x := y
     val info = sourceLine(sourceInfo, " at ", "")
     require (!LazyModule.stack.isEmpty, s"${y.name} cannot be connected to ${x.name} outside of LazyModule scope" + info)
-    require (!y.noOs, s"${y.name}${y.lazyModule.line} was incorrectly connected as a source" + info)
+    val i = x.iPushed
-    require (!y.oRealized, s"${y.name}${y.lazyModule.line} was incorrectly connected as a source after it's .module was used" + info)
+    val o = y.oPushed
-    require (!x.noIs, s"${x.name}${x.lazyModule.line} was incorrectly connected as a sink" + info)
+    y.oPush(i, x)
-    require (!x.iRealized, s"${x.name}${x.lazyModule.line} was incorrectly connected as a sink after it's .module was used" + info)
+    x.iPush(o, y)
-    val i = x.accPI.size
+    val (out, binding) = inner.connect(y.oConnect(o), x.iConnect(i), x.edgesIn(i))
-    val o = y.accPO.size
-    y.accPO += ((i, x))
-    x.accPI += ((o, y))
-    val (out, binding) = imp.connect(y.connectOut(o), y.edgesOut(o), x.connectIn(i), x.edgesIn(i))
     LazyModule.stack.head.bindings = binding :: LazyModule.stack.head.bindings
     out
   }
 }
 
+class SimpleNode[D, U, EO, EI, B <: Data](imp: NodeImp[D, U, EO, EI, B])(
+  oFn: (Int, Seq[D]) => Seq[D],
+  iFn: (Int, Seq[U]) => Seq[U],
+  numPO: Range.Inclusive,
+  numPI: Range.Inclusive)
+    extends MixedNode[D, U, EI, B, D, U, EO, B](imp, imp)(oFn, iFn, numPO, numPI)
+
 class IdentityNode[PO, PI, EO, EI, B <: Data](imp: NodeImp[PO, PI, EO, EI, B])
-  extends BaseNode(imp)({case (_, s) => s}, {case (_, s) => s}, 0 to 999, 0 to 999)
+  extends SimpleNode(imp)({case (_, s) => s}, {case (_, s) => s}, 0 to 999, 0 to 999)
 
 class OutputNode[PO, PI, EO, EI, B <: Data](imp: NodeImp[PO, PI, EO, EI, B]) extends IdentityNode(imp)
 {
-  override def connectOut = bundleOut
+  override def oConnect = bundleOut
-  override def connectIn  = bundleOut
+  override def iConnect = bundleOut
 }
 
 class InputNode[PO, PI, EO, EI, B <: Data](imp: NodeImp[PO, PI, EO, EI, B]) extends IdentityNode(imp)
 {
-  override def connectOut = bundleIn
+  override def oConnect = bundleIn
-  override def connectIn  = bundleIn
+  override def iConnect = bundleIn
 }
 
 class SourceNode[PO, PI, EO, EI, B <: Data](imp: NodeImp[PO, PI, EO, EI, B])(po: PO, num: Range.Inclusive = 1 to 1)
-  extends BaseNode(imp)({case (n, Seq()) => Seq.fill(n)(po)}, {case (0, _) => Seq()}, num, 0 to 0)
+  extends SimpleNode(imp)({case (n, Seq()) => Seq.fill(n)(po)}, {case (0, _) => Seq()}, num, 0 to 0)
 {
   require (num.end >= 1, s"${name} is a source which does not accept outputs${lazyModule.line}")
 }
 
 class SinkNode[PO, PI, EO, EI, B <: Data](imp: NodeImp[PO, PI, EO, EI, B])(pi: PI, num: Range.Inclusive = 1 to 1)
-  extends BaseNode(imp)({case (0, _) => Seq()}, {case (n, Seq()) => Seq.fill(n)(pi)}, 0 to 0, num)
+  extends SimpleNode(imp)({case (0, _) => Seq()}, {case (n, Seq()) => Seq.fill(n)(pi)}, 0 to 0, num)
 {
   require (num.end >= 1, s"${name} is a sink which does not accept inputs${lazyModule.line}")
 }
 
 class InteriorNode[PO, PI, EO, EI, B <: Data](imp: NodeImp[PO, PI, EO, EI, B])
   (oFn: Seq[PO] => PO, iFn: Seq[PI] => PI, numPO: Range.Inclusive, numPI: Range.Inclusive)
-  extends BaseNode(imp)({case (n,s) => Seq.fill(n)(oFn(s))}, {case (n,s) => Seq.fill(n)(iFn(s))}, numPO, numPI)
+  extends SimpleNode(imp)({case (n,s) => Seq.fill(n)(oFn(s))}, {case (n,s) => Seq.fill(n)(iFn(s))}, numPO, numPI)
 {
   require (numPO.end >= 1, s"${name} is an adapter which does not accept outputs${lazyModule.line}")
   require (numPI.end >= 1, s"${name} is an adapter which does not accept inputs${lazyModule.line}")

src/main/scala/uncore/tilelink2/Parameters.scala

@@ -149,7 +149,7 @@ case class TLManagerParameters(
   sinkId:             IdRange       = IdRange(0, 1),
   regionType:         RegionType.T  = RegionType.GET_EFFECTS,
   executable:         Boolean       = false, // processor can execute from this memory
-  nodePath:           Seq[TLBaseNode] = Seq(),
+  nodePath:           Seq[BaseNode] = Seq(),
   // Supports both Acquire+Release+Finish of these sizes
   supportsAcquire:    TransferSizes = TransferSizes.none,
   supportsArithmetic: TransferSizes = TransferSizes.none,
@@ -293,7 +293,7 @@ case class TLManagerPortParameters(
 
 case class TLClientParameters(
   sourceId:            IdRange       = IdRange(0,1),
-  nodePath:            Seq[TLBaseNode] = Seq(),
+  nodePath:            Seq[BaseNode] = Seq(),
   // Supports both Probe+Grant of these sizes
   supportsProbe:       TransferSizes = TransferSizes.none,
   supportsArithmetic:  TransferSizes = TransferSizes.none,
@@ -401,6 +401,17 @@ case class TLBundleParameters(
       max(sizeBits,    x.sizeBits))
 }
 
+object TLBundleParameters
+{
+  def apply(client: TLClientPortParameters, manager: TLManagerPortParameters) =
+    new TLBundleParameters(
+      addrHiBits  = log2Up(manager.maxAddress + 1) - log2Ceil(manager.beatBytes),
+      dataBits    = manager.beatBytes * 8,
+      sourceBits  = log2Up(client.endSourceId),
+      sinkBits    = log2Up(manager.endSinkId),
+      sizeBits    = log2Up(log2Ceil(max(client.maxTransfer, manager.maxTransfer))+1))
+}
+
 case class TLEdgeParameters(
   client:  TLClientPortParameters,
   manager: TLManagerPortParameters)
@@ -411,10 +422,13 @@ case class TLEdgeParameters(
   // Sanity check the link...
   require (maxTransfer >= manager.beatBytes)
 
-  val bundle = TLBundleParameters(
+  val bundle = TLBundleParameters(client, manager)
-    addrHiBits  = log2Up(manager.maxAddress + 1) - log2Ceil(manager.beatBytes),
+}
-    dataBits    = manager.beatBytes * 8,
+
-    sourceBits  = log2Up(client.endSourceId),
+case class TLAsyncManagerPortParameters(depth: Int, base: TLManagerPortParameters) { require (isPow2(depth)) }
-    sinkBits    = log2Up(manager.endSinkId),
+case class TLAsyncClientPortParameters(base: TLClientPortParameters)
-    sizeBits    = log2Up(maxLgSize+1))
+case class TLAsyncBundleParameters(depth: Int, base: TLBundleParameters) { require (isPow2(depth)) }
+case class TLAsyncEdgeParameters(client: TLAsyncClientPortParameters, manager: TLAsyncManagerPortParameters)
+{
+  val bundle = TLAsyncBundleParameters(manager.depth, TLBundleParameters(client.base, manager.base))
 }

src/main/scala/uncore/tilelink2/TLNodes.scala

@@ -8,8 +8,8 @@ import chisel3.internal.sourceinfo.SourceInfo
 
 object TLImp extends NodeImp[TLClientPortParameters, TLManagerPortParameters, TLEdgeOut, TLEdgeIn, TLBundle]
 {
-  def edgeO(po: TLClientPortParameters, pi: TLManagerPortParameters): TLEdgeOut = new TLEdgeOut(po, pi)
+  def edgeO(pd: TLClientPortParameters, pu: TLManagerPortParameters): TLEdgeOut = new TLEdgeOut(pd, pu)
-  def edgeI(po: TLClientPortParameters, pi: TLManagerPortParameters): TLEdgeIn  = new TLEdgeIn(po, pi)
+  def edgeI(pd: TLClientPortParameters, pu: TLManagerPortParameters): TLEdgeIn  = new TLEdgeIn(pd, pu)
   def bundleO(eo: Seq[TLEdgeOut]): Vec[TLBundle] = {
     require (!eo.isEmpty)
     Vec(eo.size, TLBundle(eo.map(_.bundle).reduce(_.union(_))))
@@ -19,19 +19,18 @@ object TLImp extends NodeImp[TLClientPortParameters, TLManagerPortParameters, TL
     Vec(ei.size, TLBundle(ei.map(_.bundle).reduce(_.union(_)))).flip
   }
 
-  def connect(bo: => TLBundle, eo: => TLEdgeOut, bi: => TLBundle, ei: => TLEdgeIn)(implicit sourceInfo: SourceInfo): (Option[LazyModule], () => Unit) = {
+  def connect(bo: => TLBundle, bi: => TLBundle, ei: => TLEdgeIn)(implicit sourceInfo: SourceInfo): (Option[LazyModule], () => Unit) = {
-    val monitor = LazyModule(new TLMonitor(() => new TLBundleSnoop(bo.params), () => eo, sourceInfo))
+    val monitor = LazyModule(new TLMonitor(() => new TLBundleSnoop(bo.params), () => ei, sourceInfo))
     (Some(monitor), () => {
-      require (eo.asInstanceOf[TLEdgeParameters] == ei.asInstanceOf[TLEdgeParameters])
       bi <> bo
       monitor.module.io.in := TLBundleSnoop(bo)
     })
   }
 
-  override def mixO(po: TLClientPortParameters,  node: TLBaseNode): TLClientPortParameters  =
+  override def mixO(pd: TLClientPortParameters, node: OutwardNode[TLClientPortParameters, TLManagerPortParameters, TLBundle]): TLClientPortParameters  =
-   po.copy(clients  = po.clients.map  { c => c.copy (nodePath = node +: c.nodePath) })
+   pd.copy(clients  = pd.clients.map  { c => c.copy (nodePath = node +: c.nodePath) })
-  override def mixI(pi: TLManagerPortParameters, node: TLBaseNode): TLManagerPortParameters =
+  override def mixI(pu: TLManagerPortParameters, node: InwardNode[TLClientPortParameters, TLManagerPortParameters, TLBundle]): TLManagerPortParameters =
-   pi.copy(managers = pi.managers.map { m => m.copy (nodePath = node +: m.nodePath) })
+   pu.copy(managers = pu.managers.map { m => m.copy (nodePath = node +: m.nodePath) })
 }
 
 case class TLIdentityNode() extends IdentityNode(TLImp)
@@ -72,3 +71,42 @@ class TLInputNodeTest extends UnitTest(500000) {
 
   io.finished := Module(fuzzer.module).io.finished
 }
+
+object TLAsyncImp extends NodeImp[TLAsyncClientPortParameters, TLAsyncManagerPortParameters, TLAsyncEdgeParameters, TLAsyncEdgeParameters, TLAsyncBundle]
+{
+  def edgeO(pd: TLAsyncClientPortParameters, pu: TLAsyncManagerPortParameters): TLAsyncEdgeParameters = TLAsyncEdgeParameters(pd, pu)
+  def edgeI(pd: TLAsyncClientPortParameters, pu: TLAsyncManagerPortParameters): TLAsyncEdgeParameters = TLAsyncEdgeParameters(pd, pu)
+  def bundleO(eo: Seq[TLAsyncEdgeParameters]): Vec[TLAsyncBundle] = {
+    require (eo.size == 1)
+    Vec(eo.size, new TLAsyncBundle(eo(0).bundle))
+  }
+  def bundleI(ei: Seq[TLAsyncEdgeParameters]): Vec[TLAsyncBundle] = {
+    require (ei.size == 1)
+    Vec(ei.size, new TLAsyncBundle(ei(0).bundle)).flip
+  }
+
+  def connect(bo: => TLAsyncBundle, bi: => TLAsyncBundle, ei: => TLAsyncEdgeParameters)(implicit sourceInfo: SourceInfo): (Option[LazyModule], () => Unit) = {
+    (None, () => { bi <> bo })
+  }
+
+  override def mixO(pd: TLAsyncClientPortParameters, node: OutwardNode[TLAsyncClientPortParameters, TLAsyncManagerPortParameters, TLAsyncBundle]): TLAsyncClientPortParameters  =
+   pd.copy(base = pd.base.copy(clients  = pd.base.clients.map  { c => c.copy (nodePath = node +: c.nodePath) }))
+  override def mixI(pu: TLAsyncManagerPortParameters, node: InwardNode[TLAsyncClientPortParameters, TLAsyncManagerPortParameters, TLAsyncBundle]): TLAsyncManagerPortParameters =
+   pu.copy(base = pu.base.copy(managers = pu.base.managers.map { m => m.copy (nodePath = node +: m.nodePath) }))
+}
+
+case class TLAsyncIdentityNode() extends IdentityNode(TLAsyncImp)
+case class TLAsyncOutputNode() extends OutputNode(TLAsyncImp)
+case class TLAsyncInputNode() extends InputNode(TLAsyncImp)
+
+case class TLAsyncSourceNode() extends MixedNode(TLImp, TLAsyncImp)(
+  dFn = { case (1, s) => s.map(TLAsyncClientPortParameters(_)) },
+  uFn = { case (1, s) => s.map(_.base) },
+  numPO = 1 to 1,
+  numPI = 1 to 1)
+
+case class TLAsyncSinkNode(depth: Int) extends MixedNode(TLAsyncImp, TLImp)(
+  dFn = { case (1, s) => s.map(_.base) },
+  uFn = { case (1, s) => s.map(TLAsyncManagerPortParameters(depth, _)) },
+  numPO = 1 to 1,
+  numPI = 1 to 1)

src/main/scala/uncore/tilelink2/package.scala

@@ -5,8 +5,8 @@ import chisel3.internal.sourceinfo.{SourceInfo, SourceLine, UnlocatableSourceInf
 
 package object tilelink2
 {
-  type TLBaseNode = BaseNode[TLClientPortParameters, TLManagerPortParameters, TLEdgeOut, TLEdgeIn, TLBundle]
+  type TLBaseNode = SimpleNode[TLClientPortParameters, TLManagerPortParameters, TLEdgeOut, TLEdgeIn, TLBundle]
-  type IntBaseNode = BaseNode[IntSourcePortParameters, IntSinkPortParameters, IntEdge, IntEdge, Vec[Bool]]
+  type IntBaseNode = SimpleNode[IntSourcePortParameters, IntSinkPortParameters, IntEdge, IntEdge, Vec[Bool]]
   def OH1ToUInt(x: UInt) = OHToUInt((x << 1 | UInt(1)) ^ x)
   def UIntToOH1(x: UInt, width: Int) = ~(SInt(-1, width=width).asUInt << x)(width-1, 0)
   def trailingZeros(x: Int) = if (x > 0) Some(log2Ceil(x & -x)) else None

src/main/scala/util/AsyncQueue.scala

@@ -25,8 +25,7 @@ object AsyncGrayCounter {
   }
 }
 
-class AsyncQueueSource[T <: Data](gen: T, depth: Int, sync: Int, clockIn: Clock, resetIn: Bool)
+class AsyncQueueSource[T <: Data](gen: T, depth: Int, sync: Int) extends Module {
-    extends Module(_clock = clockIn, _reset = resetIn) {
   val bits = log2Ceil(depth)
   val io = new Bundle {
     // These come from the source domain
@@ -53,8 +52,7 @@ class AsyncQueueSource[T <: Data](gen: T, depth: Int, sync: Int, clockIn: Clock,
   io.mem := mem
 }
 
-class AsyncQueueSink[T <: Data](gen: T, depth: Int, sync: Int, clockIn: Clock, resetIn: Bool)
+class AsyncQueueSink[T <: Data](gen: T, depth: Int, sync: Int) extends Module {
-    extends Module(_clock = clockIn, _reset = resetIn) {
   val bits = log2Ceil(depth)
   val io = new Bundle {
     // These come from the sink domain
@@ -88,8 +86,13 @@ class AsyncQueue[T <: Data](gen: T, depth: Int = 8, sync: Int = 3) extends Cross
   require (depth > 0 && isPow2(depth))
 
   val io = new CrossingIO(gen)
-  val source = Module(new AsyncQueueSource(gen, depth, sync, io.enq_clock, io.enq_reset))
+  val source = Module(new AsyncQueueSource(gen, depth, sync))
-  val sink   = Module(new AsyncQueueSink  (gen, depth, sync, io.deq_clock, io.deq_reset))
+  val sink   = Module(new AsyncQueueSink  (gen, depth, sync))
+
+  source.clock := io.enq_clock
+  source.reset := io.enq_reset
+  sink.clock := io.deq_clock
+  sink.reset := io.deq_reset
 
   source.io.enq <> io.enq
   io.deq <> sink.io.deq