Merge branch 'master' into vectored-stvec

src/main/scala/uncore/axi4/Deinterleaver.scala

@@ -38,8 +38,8 @@ class AXI4Deinterleaver(maxReadBytes: Int)(implicit p: Parameters) extends LazyM
       out.w <> in.w
       in.b <> out.b
 
-      if (queues == 1) {
-        // Gracefully do nothing
+      if (beats <= 1) {
+        // Nothing to do if only single-beat R
         in.r <> out.r
       } else {
         // Buffer R response
@@ -48,7 +48,7 @@ class AXI4Deinterleaver(maxReadBytes: Int)(implicit p: Parameters) extends LazyM
 
         // Which ID is being enqueued and dequeued?
         val locked = RegInit(Bool(false))
-        val deq_id = Reg(UInt(width=log2Ceil(queues)))
+        val deq_id = Reg(UInt(width=log2Up(queues)))
         val enq_id = out.r.bits.id
         val deq_OH = UIntToOH(deq_id, queues)
         val enq_OH = UIntToOH(enq_id, queues)

src/main/scala/uncore/axi4/IdIndexer.scala

@@ -34,9 +34,7 @@ class AXI4IdIndexer(idBits: Int)(implicit p: Parameters) extends LazyModule
         userBits = mp.userBits + max(0, log2Ceil(mp.endId) - idBits),
         masters  = masters)
     },
-    slaveFn = { sp => sp.copy(
-      slaves = sp.slaves.map(s => s.copy(
-        interleavedId = if (idBits == 0) Some(0) else s.interleavedId)))
+    slaveFn = { sp => sp
     })
 
   lazy val module = new LazyModuleImp(this) {
@@ -56,12 +54,22 @@ class AXI4IdIndexer(idBits: Int)(implicit p: Parameters) extends LazyModule
 
       val bits = log2Ceil(edgeIn.master.endId) - idBits
       if (bits > 0) {
-       out.ar.bits.user.get := Cat(in.ar.bits.user.toList ++ Seq(in.ar.bits.id >> idBits))
-       out.aw.bits.user.get := Cat(in.aw.bits.user.toList ++ Seq(in.aw.bits.id >> idBits))
-       in.r.bits.user.foreach { _ := out.r.bits.user.get >> bits }
-       in.b.bits.user.foreach { _ := out.b.bits.user.get >> bits }
-       in.r.bits.id := Cat(out.r.bits.user.get, out.r.bits.id)
-       in.b.bits.id := Cat(out.b.bits.user.get, out.b.bits.id)
+        // (in.aX.bits.id >> idBits).width = bits > 0
+        out.ar.bits.user.get := Cat(in.ar.bits.user.toList ++ Seq(in.ar.bits.id >> idBits))
+        out.aw.bits.user.get := Cat(in.aw.bits.user.toList ++ Seq(in.aw.bits.id >> idBits))
+        // user.isDefined => width > 0
+        in.r.bits.user.foreach { _ := out.r.bits.user.get >> bits }
+        in.b.bits.user.foreach { _ := out.b.bits.user.get >> bits }
+        // Special care is needed in case of 0 idBits, b/c .id has width 1 still
+        if (idBits == 0) {
+          out.ar.bits.id := UInt(0)
+          out.aw.bits.id := UInt(0)
+          in.r.bits.id := out.r.bits.user.get
+          in.b.bits.id := out.b.bits.user.get
+        } else {
+          in.r.bits.id := Cat(out.r.bits.user.get, out.r.bits.id)
+          in.b.bits.id := Cat(out.b.bits.user.get, out.b.bits.id)
+        }
       }
     }
   }

src/main/scala/uncore/axi4/Parameters.scala

@@ -15,7 +15,7 @@ case class AXI4SlaveParameters(
   nodePath:      Seq[BaseNode] = Seq(),
   supportsWrite: TransferSizes = TransferSizes.none,
   supportsRead:  TransferSizes = TransferSizes.none,
-  interleavedId: Option[Int]   = None) // The device will not interleave read responses
+  interleavedId: Option[Int]   = None) // The device will not interleave responses (R+B)
 {
   address.foreach { a => require (a.finite) }
   address.combinations(2).foreach { case Seq(x,y) => require (!x.overlaps(y), s"$x and $y overlap") }

src/main/scala/uncore/axi4/SRAM.scala

@@ -40,7 +40,7 @@ class AXI4RAM(address: AddressSet, executable: Boolean = true, beatBytes: Int =
 
     val w_full = RegInit(Bool(false))
     val w_id   = Reg(UInt())
-    val w_user = Reg(UInt())
+    val w_user = Reg(UInt(width = 1 max in.params.userBits))
 
     when (in. b.fire()) { w_full := Bool(false) }
     when (in.aw.fire()) { w_full := Bool(true) }
@@ -65,7 +65,7 @@ class AXI4RAM(address: AddressSet, executable: Boolean = true, beatBytes: Int =
 
     val r_full = RegInit(Bool(false))
     val r_id   = Reg(UInt())
-    val r_user = Reg(UInt())
+    val r_user = Reg(UInt(width = 1 max in.params.userBits))
 
     when (in. r.fire()) { r_full := Bool(false) }
     when (in.ar.fire()) { r_full := Bool(true) }

src/main/scala/uncore/axi4/Test.scala

@@ -26,8 +26,8 @@ class AXI4LiteFuzzRAM()(implicit p: Parameters) extends LazyModule
 
   model.node := fuzz.node
   xbar.node  := TLDelayer(0.1)(TLBuffer(BufferParams.flow)(TLDelayer(0.2)(model.node)))
-  ram.node   := AXI4Fragmenter()(AXI4Deinterleaver(16)(TLToAXI4(4, true )(xbar.node)))
-  gpio.node  := AXI4Fragmenter()(AXI4Deinterleaver(16)(TLToAXI4(4, false)(xbar.node)))
+  ram.node   := AXI4UserYanker()(AXI4IdIndexer(0)(TLToAXI4(4, true )(TLFragmenter(4, 16)(xbar.node))))
+  gpio.node  := AXI4UserYanker()(AXI4IdIndexer(0)(TLToAXI4(4, false)(TLFragmenter(4, 16)(xbar.node))))
 
   lazy val module = new LazyModuleImp(this) with HasUnitTestIO {
     io.finished := fuzz.module.io.finished

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

@@ -37,11 +37,13 @@ class TLFragmenter(val minSize: Int, val maxSize: Int, val alwaysMin: Boolean =
     supportsPutFull    = expandTransfer(m.supportsPutFull),
     supportsPutPartial = expandTransfer(m.supportsPutPartial),
     supportsHint       = expandTransfer(m.supportsHint))
-  def mapClient(c: TLClientParameters) = c.copy(
-    sourceId = IdRange(c.sourceId.start << fragmentBits, c.sourceId.end << fragmentBits))
 
   val node = TLAdapterNode(
-    clientFn  = { c => c.copy(clients = c.clients.map(mapClient)) },
+    // We require that all the responses are mutually FIFO
+    // Thus we need to compact all of the masters into one big master
+    clientFn  = { c => c.copy(clients = Seq(TLClientParameters(
+      sourceId = IdRange(0, c.endSourceId << fragmentBits),
+      requestFifo = true))) },
     managerFn = { m => m.copy(managers = m.managers.map(mapManager)) })
 
   lazy val module = new LazyModuleImp(this) {

src/main/scala/uncore/tilelink2/ToAXI4.scala

@@ -175,16 +175,13 @@ class TLToAXI4(beatBytes: Int, combinational: Boolean = true)(implicit p: Parame
       val a_sel = UIntToOH(arw.id, edgeOut.master.endId).toBools
       val d_sel = UIntToOH(Mux(r_wins, out.r.bits.id, out.b.bits.id), edgeOut.master.endId).toBools
       val d_last = Mux(r_wins, out.r.bits.last, Bool(true))
-      val d_first = RegInit(Bool(true))
-      when (in.d.fire()) { d_first := d_last }
       val stalls = ((a_sel zip d_sel) zip idCount) filter { case (_, n) => n > 1 } map { case ((as, ds), n) =>
         val count = RegInit(UInt(0, width = log2Ceil(n + 1)))
         val write = Reg(Bool())
         val idle = count === UInt(0)
 
-        // Once we start getting the response, it's safe to already switch R/W
         val inc = as && out_arw.fire()
-        val dec = ds && d_first && in.d.fire()
+        val dec = ds && d_last && in.d.fire()
         count := count + inc.asUInt - dec.asUInt
 
         assert (!dec || count =/= UInt(0)) // underflow