Add "MergedCreditedCrossing" for TileLink channels

src/main/scala/prci/ClockCrossingType.scala

@@ -16,22 +16,29 @@ trait HasClockDomainCrossing extends HasDomainCrossing { this: LazyModule =>
 }
 
 /** Enumerates the types of clock crossings generally supported by Diplomatic bus protocols  */
-sealed trait ClockCrossingType extends CrossingType
+trait ClockCrossingType extends CrossingType
 {
-  def sameClock = this match {
-    case _: SynchronousCrossing | _: CreditedCrossing => true
-    case _ => false
-  }
+  def sameClock: Boolean
 }
 
 case object NoCrossing // converts to SynchronousCrossing(BufferParams.none) via implicit def in package
 case class SynchronousCrossing(params: BufferParams = BufferParams.default) extends ClockCrossingType
+{
+  def sameClock = true
+}
 case class RationalCrossing(direction: RationalDirection = FastToSlow) extends ClockCrossingType
+{
+  def sameClock = false
+}
 case class AsynchronousCrossing(depth: Int = 8, sourceSync: Int = 3, sinkSync: Int = 3, safe: Boolean = true, narrow: Boolean = false) extends ClockCrossingType
 {
+  def sameClock = false
   def asSinkParams = AsyncQueueParams(depth, sinkSync, safe, narrow)
 }
 case class CreditedCrossing(sourceDelay: CreditedDelay, sinkDelay: CreditedDelay) extends ClockCrossingType
+{
+  def sameClock = true
+}
 
 object CreditedCrossing {
   def apply(delay: CreditedDelay): CreditedCrossing = CreditedCrossing(delay, delay.flip)

src/main/scala/subsystem/Configs.scala

@@ -12,7 +12,7 @@ import freechips.rocketchip.devices.debug.{DebugModuleKey, DefaultDebugModulePar
 import freechips.rocketchip.devices.tilelink.{
   BuiltInErrorDeviceParams, BootROMLocated, BootROMParams, CLINTKey, DevNullDevice, CLINTParams, PLICKey, PLICParams, DevNullParams
 }
-import freechips.rocketchip.prci.{SynchronousCrossing, AsynchronousCrossing, RationalCrossing, ClockCrossingType}
+import freechips.rocketchip.prci.{SynchronousCrossing, AsynchronousCrossing, RationalCrossing, ClockCrossingType, CreditedCrossing}
 import freechips.rocketchip.diplomacy.{
   AddressSet, MonitorsEnabled,
 }
@@ -517,6 +517,16 @@ class WithRationalRocketTiles extends Config((site, here, up) => {
   }
 })
 
+class WithCreditedRocketTiles(mergedCredited: Boolean = false) extends Config((site, here, up) => {
+  case TilesLocated(location) => up(TilesLocated(location), site) map {
+    case tp: RocketTileAttachParams => tp.copy(crossingParams = tp.crossingParams.copy(
+      crossingType = CreditedCrossing(),
+      forceMergedCreditedTLCrossings = mergedCredited
+    ))
+    case t => t
+  }
+})
+
 class WithEdgeDataBits(dataBits: Int) extends Config((site, here, up) => {
   case MemoryBusKey => up(MemoryBusKey, site).copy(beatBytes = dataBits/8)
   case ExtIn => up(ExtIn, site).map(_.copy(beatBytes = dataBits/8))

src/main/scala/subsystem/HierarchicalElement.scala

@@ -35,6 +35,8 @@ trait HierarchicalElementCrossingParamsLike {
   def resetCrossingType: ResetCrossingType
   /** Keep the element clock separate from the interconnect clock (e.g. even if they are synchronous to one another) */
   def forceSeparateClockReset: Boolean
+  /** Used a MergedCreditedTLCrossing for credited TL crossings to save pins */
+  def forceMergedCreditedTLCrossings: Boolean
 }
 
 /** An interface for describing the parameterization of how a particular element port is connected to an interconnect */

src/main/scala/subsystem/HierarchicalElementPRCIDomain.scala

@@ -11,7 +11,7 @@ import freechips.rocketchip.diplomacy.{DisableMonitors, FlipRendering}
 import freechips.rocketchip.interrupts.{IntInwardNode, IntOutwardNode}
 import freechips.rocketchip.prci.{ClockCrossingType, ResetCrossingType, ResetDomain, ClockSinkNode, ClockSinkParameters, ClockIdentityNode, FixedClockBroadcast, ClockDomain}
 import freechips.rocketchip.tile.{RocketTile, TraceBundle}
-import freechips.rocketchip.tilelink.{TLInwardNode, TLOutwardNode}
+import freechips.rocketchip.tilelink.{TLInwardNode, TLOutwardNode, UseTLMergedCreditedCrossing}
 import freechips.rocketchip.util.TraceCoreInterface
 
 import freechips.rocketchip.tilelink.TLClockDomainCrossing
@@ -83,7 +83,9 @@ abstract class HierarchicalElementPRCIDomain[T <: BaseHierarchicalElement](
         element { element.makeSlaveBoundaryBuffers(crossingType) }
     }
     val tlSlaveClockXing = this.crossIn(tlSlaveResetXing)
-    tlSlaveClockXing(crossingType)
+    tlSlaveClockXing(crossingType)(p.alterPartial {
+      case UseTLMergedCreditedCrossing => crossingParams.forceMergedCreditedTLCrossings
+    })
   } } }
 
   /** External code looking to connect the ports where this tile masters an interconnect
@@ -95,6 +97,8 @@ abstract class HierarchicalElementPRCIDomain[T <: BaseHierarchicalElement](
         element_reset_domain.crossTLOut(element.masterNode)
     } }
     val tlMasterClockXing = this.crossOut(tlMasterResetXing)
-    tlMasterClockXing(crossingType)
+    tlMasterClockXing(crossingType)(p.alterPartial {
+      case UseTLMergedCreditedCrossing => crossingParams.forceMergedCreditedTLCrossings
+    })
   }
 }

src/main/scala/subsystem/RocketSubsystem.scala

@@ -16,7 +16,8 @@ case class RocketCrossingParams(
   slave: HierarchicalElementSlavePortParams = HierarchicalElementSlavePortParams(),
   mmioBaseAddressPrefixWhere: TLBusWrapperLocation = CBUS,
   resetCrossingType: ResetCrossingType = NoResetCrossing(),
-  forceSeparateClockReset: Boolean = false
+  forceSeparateClockReset: Boolean = false,
+  forceMergedCreditedTLCrossings: Boolean = false
 ) extends HierarchicalElementCrossingParamsLike
 
 case class RocketTileAttachParams(

src/main/scala/tilelink/Arbiter.scala

@@ -123,7 +123,7 @@ abstract class DecoupledArbiterTest(
     case (z, i) => (beatsLeftFromIdx(i), z)
   }:_*)
 
-  count := count + 1.U
+  when (!io.finished) { count := count + 1.U }
   io.finished := count >= txns.U
 }
 

src/main/scala/tilelink/Bundles.scala

@@ -5,8 +5,7 @@ package freechips.rocketchip.tilelink
 import chisel3._
 import freechips.rocketchip.util._
 import scala.collection.immutable.ListMap
-import chisel3.util.Decoupled
-import chisel3.util.DecoupledIO
+import chisel3.util.{Decoupled, DecoupledIO, Valid}
 import chisel3.reflect.DataMirror
 
 abstract class TLBundleBase(val params: TLBundleParameters) extends Bundle
@@ -317,3 +316,44 @@ class TLCreditedBundle(params: TLBundleParameters) extends TLBundleBase(params)
   val d = Flipped(CreditedIO(new TLBundleD(params)))
   val e = CreditedIO(new TLBundleE(params))
 }
+
+class TLBundleACE(params: TLBundleParameters) extends TLBundleBase(params)
+{
+  // fixed fields during multibeat:
+  val opcode  = UInt(3.W)
+  val param   = UInt(List(TLAtomics.width, TLPermissions.aWidth, TLPermissions.cWidth, TLHints.width).max.W)
+  val size    = UInt(params.sizeBits.W)
+  val source  = UInt(params.sourceBits.W) // from
+  val address = UInt(params.addressBits.W) // to
+  val sink    = UInt(params.sinkBits.W) // to
+  val user    = BundleMap(params.requestFields)
+  val echo    = BundleMap(params.echoFields)
+  // variable fields during multibeat:
+  val mask    = UInt((params.dataBits/8).W)
+  val data    = UInt(params.dataBits.W)
+  val corrupt = Bool() // only applies to *Data messages
+}
+
+class TLBundleBD(params: TLBundleParameters) extends TLBundleBase(params)
+{
+  // fixed fields during multibeat:
+  val opcode  = UInt(3.W)
+  val param   = UInt(TLPermissions.bdWidth.W) // cap perms
+  val size    = UInt(params.sizeBits.W)
+  val source  = UInt(params.sourceBits.W) // to
+  val address = UInt(params.addressBits.W) // from
+  val sink    = UInt(params.sinkBits.W)   // from
+  val denied  = Bool() // implies corrupt iff *Data
+  val user    = BundleMap(params.responseFields)
+  val echo    = BundleMap(params.echoFields)
+  // variable fields during multibeat:
+  val mask    = UInt((params.dataBits/8).W)
+  val data    = UInt(params.dataBits.W)
+  val corrupt = Bool() // only applies to *Data messages
+}
+
+class TLMergedCreditedBundle(params: TLBundleParameters) extends TLBundleBase(params)
+{
+  val eca = MultiCreditedIO(new TLBundleACE(params), 3)
+  val db = Flipped(MultiCreditedIO(new TLBundleBD(params), 2))
+}

src/main/scala/tilelink/CrossingHelper.scala

@@ -32,8 +32,11 @@ case class TLInwardClockCrossingHelper(name: String, scope: LazyScope, node: TLI
         node :*=* scope { TLRationalCrossingSink(direction.flip) :*=* TLRationalNameNode(name) } :*=* TLRationalNameNode(name) :*=* TLRationalCrossingSource()
       case SynchronousCrossing(buffer) =>
         node :*=* scope { TLBuffer(buffer) :*=* TLNameNode(name) } :*=* TLNameNode(name)
-      case CreditedCrossing(sourceDelay, sinkDelay) =>
+      case CreditedCrossing(sourceDelay, sinkDelay) => if (p(UseTLMergedCreditedCrossing)) {
+        node :*=* scope { TLMergedCreditedSink(sinkDelay) :*=* TLMergedCreditedNameNode(name) } :*=* TLMergedCreditedNameNode(name) :*=* TLMergedCreditedSource(sourceDelay)
+      } else {
         node :*=* scope { TLCreditedSink(sinkDelay) :*=* TLCreditedNameNode(name) } :*=* TLCreditedNameNode(name) :*=* TLCreditedSource(sourceDelay)
+      }
     }
   }
 }
@@ -63,8 +66,11 @@ case class TLOutwardClockCrossingHelper(name: String, scope: LazyScope, node: TL
         TLRationalCrossingSink(direction) :*=* TLRationalNameNode(name) :*=* scope { TLRationalNameNode(name) :*=* TLRationalCrossingSource() } :*=* node
       case SynchronousCrossing(buffer) =>
         TLNameNode(name) :*=* scope { TLNameNode(name) :*=* TLBuffer(buffer) } :*=* node
-      case CreditedCrossing(sourceDelay, sinkDelay) =>
+      case CreditedCrossing(sourceDelay, sinkDelay) => if (p(UseTLMergedCreditedCrossing)) {
+        TLMergedCreditedSink(sinkDelay) :*=* TLMergedCreditedNameNode(name) :*=* scope { TLMergedCreditedNameNode(name) :*=* TLMergedCreditedSource(sourceDelay) } :*=* node
+      } else {
         TLCreditedSink(sinkDelay) :*=* TLCreditedNameNode(name) :*=* scope { TLCreditedNameNode(name) :*=* TLCreditedSource(sourceDelay) } :*=* node
+      }
     }
   }
 }

src/main/scala/tilelink/MergedCredited.scala

@@ -0,0 +1,128 @@
+package freechips.rocketchip.tilelink
+
+import chisel3._
+import chisel3.util._
+
+import org.chipsalliance.cde.config._
+import org.chipsalliance.diplomacy.lazymodule._
+
+import freechips.rocketchip.diplomacy.{AddressSet}
+import freechips.rocketchip.prci.{ClockCrossingType, CreditedCrossing}
+import freechips.rocketchip.subsystem.CrossingWrapper
+import freechips.rocketchip.util.{CreditedDelay, MultiCreditedIO}
+
+case object UseTLMergedCreditedCrossing extends Field[Boolean](false)
+
+class TLMergedCreditedSource(delay: TLMergedCreditedDelay)(implicit p: Parameters) extends LazyModule
+{
+  val node = TLMergedCreditedSourceNode(delay)
+  override lazy val module = new Impl
+  class Impl extends LazyModuleImp(this) {
+    (node.in zip node.out) foreach { case ((in, edgeIn), (out, edgeOut)) =>
+      val tld = edgeOut.delay
+
+      val to_credited = Wire(Vec(3, Decoupled(new TLBundleACE(edgeIn.bundle))))
+      val from_credited = Wire(Vec(2, Decoupled(new TLBundleBD(edgeIn.bundle))))
+
+      out.eca <> MultiCreditedIO.fromSenders(to_credited, tld.ace.total).pipeline(delay.ace)
+      from_credited <> VecInit(out.db.pipeline(delay.bd).toReceivers(tld.bd.total))
+
+      Seq(in.e, in.c, in.a).zipWithIndex.foreach { case (channel, i) =>
+        val out = to_credited(i)
+        channel.ready := out.ready
+        out.valid := channel.valid
+        out.bits := DontCare
+        (out.bits: Data).waiveAll :<>= (channel.bits: Data).waiveAll
+      }
+
+      Seq(in.d, in.b).zipWithIndex.foreach { case (channel, i) =>
+        val out = from_credited(i)
+        out.ready := channel.ready
+        channel.valid := out.valid
+        (channel.bits: Data).waiveAll :<>= (out.bits: Data).waiveAll
+      }
+    }
+  }
+}
+
+object TLMergedCreditedSource {
+  def apply(delay: TLMergedCreditedDelay)(implicit p: Parameters): TLMergedCreditedSourceNode = {
+    val source = LazyModule(new TLMergedCreditedSource(delay))
+    source.node
+  }
+  def apply(delay: CreditedDelay)(implicit p: Parameters): TLMergedCreditedSourceNode = apply(TLMergedCreditedDelay(delay))
+  def apply()(implicit p: Parameters): TLMergedCreditedSourceNode = apply(CreditedDelay(1, 1))
+}
+
+
+class TLMergedCreditedSink(delay: TLMergedCreditedDelay)(implicit p: Parameters) extends LazyModule
+{
+  val node = TLMergedCreditedSinkNode(delay)
+  override lazy val module = new Impl
+  class Impl extends LazyModuleImp(this) {
+    (node.in zip node.out) foreach { case ((in, edgeIn), (out, edgeOut)) =>
+      val tld = edgeIn.delay
+
+      val to_credited = Wire(Vec(2, Decoupled(new TLBundleBD(edgeOut.bundle))))
+      val from_credited = Wire(Vec(3, Decoupled(new TLBundleACE(edgeOut.bundle))))
+
+      in.db <> MultiCreditedIO.fromSenders(to_credited, tld.bd.total).pipeline(delay.bd)
+      from_credited <> in.eca.pipeline(delay.ace).toReceivers(tld.ace.total)
+
+      Seq(out.d, out.b).zipWithIndex.foreach { case (channel, i) =>
+        val in = to_credited(i)
+        channel.ready := in.ready
+        in.valid := channel.valid
+        in.bits := DontCare
+        (in.bits: Data).waiveAll :<>= (channel.bits: Data).waiveAll
+      }
+
+      Seq(out.e, out.c, out.a).zipWithIndex.foreach { case (channel, i) =>
+        val in = from_credited(i)
+        in.ready := channel.ready
+        channel.valid := in.valid
+        (channel.bits: Data).waiveAll :<>= (in.bits: Data).waiveAll
+      }
+    }
+  }
+}
+
+object TLMergedCreditedSink {
+  def apply(delay: TLMergedCreditedDelay)(implicit p: Parameters): TLMergedCreditedSinkNode = {
+    val sink = LazyModule(new TLMergedCreditedSink(delay))
+    sink.node
+  }
+  def apply(delay: CreditedDelay)(implicit p: Parameters): TLMergedCreditedSinkNode = apply(TLMergedCreditedDelay(delay))
+  def apply()(implicit p: Parameters): TLMergedCreditedSinkNode = apply(CreditedDelay(1, 1))
+}
+
+// Synthesizable unit tests
+import freechips.rocketchip.unittest._
+
+class TLRAMMergedCreditedCrossing(txns: Int, params: CreditedCrossing)(implicit p: Parameters) extends LazyModule {
+  val model = LazyModule(new TLRAMModel("MergedCreditedCrossing"))
+  val fuzz = LazyModule(new TLFuzzer(txns))
+  val island = LazyModule(new CrossingWrapper(params))
+  val ram  = island { LazyModule(new TLRAM(AddressSet(0x0, 0x3ff))) }
+
+  island.crossTLIn(ram.node) := TLFragmenter(4, 256) := TLDelayer(0.1) := model.node := fuzz.node
+
+  lazy val module = new Impl
+  class Impl extends LazyModuleImp(this) with UnitTestModule {
+    io.finished := fuzz.module.io.finished
+  }
+}
+
+class TLRAMMergedCreditedCrossingTest(txns: Int = 5000, timeout: Int = 500000)(implicit p: Parameters) extends UnitTest(timeout) {
+  val u = p.alterPartial { case UseTLMergedCreditedCrossing => true }
+
+  val dut_1000 = Module(LazyModule(new TLRAMMergedCreditedCrossing(txns, CreditedCrossing(CreditedDelay(1, 0), CreditedDelay(0, 0)))(u)).module)
+  val dut_0100 = Module(LazyModule(new TLRAMMergedCreditedCrossing(txns, CreditedCrossing(CreditedDelay(0, 1), CreditedDelay(0, 0)))(u)).module)
+  val dut_0010 = Module(LazyModule(new TLRAMMergedCreditedCrossing(txns, CreditedCrossing(CreditedDelay(0, 0), CreditedDelay(1, 0)))(u)).module)
+  val dut_0001 = Module(LazyModule(new TLRAMMergedCreditedCrossing(txns, CreditedCrossing(CreditedDelay(0, 0), CreditedDelay(0, 1)))(u)).module)
+  val dut_1111 = Module(LazyModule(new TLRAMMergedCreditedCrossing(txns, CreditedCrossing(CreditedDelay(1, 1), CreditedDelay(1, 1)))(u)).module)
+
+  val duts = Seq(dut_1000, dut_0100, dut_0010, dut_0001, dut_1111)
+  duts.foreach { _.io.start := true.B }
+  io.finished := duts.map(_.io.finished).reduce(_ && _)
+}

src/main/scala/tilelink/Nodes.scala

@@ -192,3 +192,42 @@ case class TLCreditedSinkNode(delay: TLCreditedDelay)(implicit valName: ValName)
   extends MixedAdapterNode(TLCreditedImp, TLImp)(
     dFn = { p => p.base.v1copy(minLatency = 1) },
     uFn = { p => TLCreditedManagerPortParameters(delay, p) }) with FormatNode[TLCreditedEdgeParameters, TLEdgeOut]
+
+// Merged Credited version of TileLink channels
+trait TLMergedCreditedFormatNode extends FormatNode[TLMergedCreditedEdgeParameters, TLMergedCreditedEdgeParameters]
+
+object TLMergedCreditedImp extends SimpleNodeImp[TLMergedCreditedClientPortParameters, TLMergedCreditedManagerPortParameters, TLMergedCreditedEdgeParameters, TLMergedCreditedBundle]
+{
+  def edge(pd: TLMergedCreditedClientPortParameters, pu: TLMergedCreditedManagerPortParameters, p: Parameters, sourceInfo: SourceInfo) = TLMergedCreditedEdgeParameters(pd, pu, p, sourceInfo)
+  def bundle(e: TLMergedCreditedEdgeParameters) = new TLMergedCreditedBundle(e.bundle)
+  def render(e: TLMergedCreditedEdgeParameters) = RenderedEdge(colour = "#ffff00" /* yellow */, e.delay.toString)
+
+  override def mixO(pd: TLMergedCreditedClientPortParameters, node: OutwardNode[TLMergedCreditedClientPortParameters, TLMergedCreditedManagerPortParameters, TLMergedCreditedBundle]): TLMergedCreditedClientPortParameters  =
+   pd.copy(base = pd.base.v1copy(clients  = pd.base.clients.map  { c => c.v1copy (nodePath = node +: c.nodePath) }))
+  override def mixI(pu: TLMergedCreditedManagerPortParameters, node: InwardNode[TLMergedCreditedClientPortParameters, TLMergedCreditedManagerPortParameters, TLMergedCreditedBundle]): TLMergedCreditedManagerPortParameters =
+   pu.copy(base = pu.base.v1copy(managers = pu.base.managers.map { m => m.v1copy (nodePath = node +: m.nodePath) }))
+}
+
+case class TLMergedCreditedAdapterNode(
+  clientFn:  TLMergedCreditedClientPortParameters  => TLMergedCreditedClientPortParameters  = { s => s },
+  managerFn: TLMergedCreditedManagerPortParameters => TLMergedCreditedManagerPortParameters = { s => s })(
+  implicit valName: ValName)
+  extends AdapterNode(TLMergedCreditedImp)(clientFn, managerFn) with TLMergedCreditedFormatNode
+
+case class TLMergedCreditedIdentityNode()(implicit valName: ValName) extends IdentityNode(TLMergedCreditedImp)() with TLMergedCreditedFormatNode
+
+object TLMergedCreditedNameNode {
+  def apply(name: ValName) = TLMergedCreditedIdentityNode()(name)
+  def apply(name: Option[String]): TLMergedCreditedIdentityNode = apply(ValName(name.getOrElse("with_no_name")))
+  def apply(name: String): TLMergedCreditedIdentityNode = apply(Some(name))
+}
+
+case class TLMergedCreditedSourceNode(delay: TLMergedCreditedDelay)(implicit valName: ValName)
+  extends MixedAdapterNode(TLImp, TLMergedCreditedImp)(
+    dFn = { p => TLMergedCreditedClientPortParameters(delay, p) },
+    uFn = { p => p.base.v1copy(minLatency = 1) }) with FormatNode[TLEdgeIn, TLMergedCreditedEdgeParameters] // discard cycles from other clock domain
+
+case class TLMergedCreditedSinkNode(delay: TLMergedCreditedDelay)(implicit valName: ValName)
+  extends MixedAdapterNode(TLMergedCreditedImp, TLImp)(
+    dFn = { p => p.base.v1copy(minLatency = 1) },
+    uFn = { p => TLMergedCreditedManagerPortParameters(delay, p) }) with FormatNode[TLMergedCreditedEdgeParameters, TLEdgeOut]