Connect smem core IO to TL with translation

src/main/scala/rocket/VortexCore.scala

@@ -47,6 +47,8 @@ class VortexBundle(tile: VortexTile)(implicit p: Parameters) extends CoreBundle
   val imemTagWidth = UUID_WIDTH + NW_WIDTH
   val LSUQ_TAG_BITS = 4
   val dmemTagWidth = UUID_WIDTH + LSUQ_TAG_BITS
+  // dmem and smem shares the same tag width, DCACHE_NOSM_TAG_WIDTH
+  val smemTagWidth = dmemTagWidth
 
   // conditionally instantiate ports depending on whether we want to use VX_cache or not
   val imem = if (!tile.vortexParams.useVxCache) Some(Vec(1, new Bundle {
@@ -57,6 +59,10 @@ class VortexBundle(tile: VortexTile)(implicit p: Parameters) extends CoreBundle
     val a = Decoupled(new VortexBundleA(tagWidth = dmemTagWidth, dataWidth = 32))
     val d = Flipped(Decoupled(new VortexBundleD(tagWidth = dmemTagWidth, dataWidth = 32)))
   })) else None
+  val smem = if (!tile.vortexParams.useVxCache) Some(Vec(tile.numLanes, new Bundle {
+    val a = Decoupled(new VortexBundleA(tagWidth = smemTagWidth, dataWidth = 32))
+    val d = Flipped(Decoupled(new VortexBundleD(tagWidth = smemTagWidth, dataWidth = 32)))
+  })) else None
   val mem = if (tile.vortexParams.useVxCache) Some(new Bundle { 
     val a = Decoupled(new VortexBundleA(tagWidth = 15, dataWidth = 128))
     val d = Flipped(Decoupled(new VortexBundleD(tagWidth = 15, dataWidth = 128)))
@@ -103,7 +109,6 @@ class Vortex(tile: VortexTile)(implicit p: Parameters)
   // addResource("/vsrc/vortex/hw/syn/synopsys/models/memory/cln28hpc/rf2_32x128_wm1/vsim/rf2_32x128_wm1_tb.v")
   // addResource("/vsrc/vortex/hw/syn/modelsim/vortex_tb.v")
 
-
   addResource("/vsrc/vortex/hw/rtl/VX_gpu_pkg.sv")
 
   // addResource("/vsrc/vortex/hw/rtl/VX_cluster.sv")
@@ -341,6 +346,5 @@ class Vortex(tile: VortexTile)(implicit p: Parameters)
   }
 
   val nTotalRoCCCSRs = 0
-  val coreBundle = new VortexBundle(tile)
+  val io = IO(new VortexBundle(tile))
-  val io = IO(coreBundle)
 }

src/main/scala/tile/VortexTile.scala

@@ -259,9 +259,8 @@ class VortexTile private (
   // NOTE: We need TLWidthWidget here because there might be a data width
   // mismatch between Vortex's per-lane response and the system bus when we
   // don't instantiate either L1 or the coalescer.  This _should_ be optimized
-  // out when we instantiate coalescer which should handle data width conversion
+  // out when we instantiate either which should handle data width conversion
-  // internally (which it does by... using TLWidthWidget), but probably not
+  // internally (which it does by... using TLWidthWidget).
-  // the cleanest way to do this.
   val dmemAggregateNode = TLIdentityNode()
   dmemNodes.foreach { dmemAggregateNode := TLWidthWidget(4) := _ }
 
@@ -326,7 +325,8 @@ class VortexTile private (
 
   // Instantiate sharedmem
   // TODO: parametrize
-  val sharedmem = LazyModule(new TLRAM(AddressSet(0xff000000L, 0x00ffffffL), beatBytes = 4 /*FIXME*/))
+  // FIXME: beatBytes should be wordSize
+  val sharedmem = LazyModule(new TLRAM(AddressSet(0xff000000L, 0x00ffffffL), beatBytes = 4))
   val smemXbar = LazyModule(new TLXbar)
   smemNodes.foreach(smemXbar.node := _)
   sharedmem.node :=* smemXbar.node
@@ -492,6 +492,7 @@ class VortexTileModuleImp(outer: VortexTile) extends BaseTileModuleImp(outer) {
     outer.memNode.out(0)._1.a <> memTLAdapter.io.outReq
     memTLAdapter.io.outResp <> outer.memNode.out(0)._1.d
   } else {
+    def connectImem = {
       val imemTLAdapter = Module(
         new VortexTLAdapter(
           outer.imemSourceWidth,
@@ -505,7 +506,9 @@ class VortexTileModuleImp(outer: VortexTile) extends BaseTileModuleImp(outer) {
       core.io.imem.get(0).d <> imemTLAdapter.io.inResp
       outer.imemNodes(0).out(0)._1.a <> imemTLAdapter.io.outReq
       imemTLAdapter.io.outResp <> outer.imemNodes(0).out(0)._1.d
+    }
 
+    def connectDmem = {
       // @perf: this would duplicate SourceGenerator table for every lane and eat
       // up some area
       val dmemTLBundles = outer.dmemNodes.map(_.out.head._1)
@@ -567,6 +570,7 @@ class VortexTileModuleImp(outer: VortexTile) extends BaseTileModuleImp(outer) {
         tlAdapter.io.inReq <> coreMem.a
         coreMem.d <> tlAdapter.io.inResp
       }
+      // override response channel with matchingSources
       (core.io.dmem.get zip dmemTLAdapters).zipWithIndex.foreach {
         case ((coreMem, tlAdapter), i) =>
           coreMem.d.valid := tlAdapter.io.inResp.valid && matchingSources(i)
@@ -583,6 +587,35 @@ class VortexTileModuleImp(outer: VortexTile) extends BaseTileModuleImp(outer) {
       }
     }
 
+    def connectSmem = {
+      // @perf: this would duplicate SourceGenerator table for every lane and eat
+      // up some area
+      val smemTLBundles = outer.smemNodes.map(_.out.head._1)
+      val smemTLAdapters = Seq.tabulate(outer.numLanes) { _ =>
+        Module(
+          new VortexTLAdapter(
+            outer.smemSourceWidth,
+            chiselTypeOf(core.io.smem.get(0).a.bits),
+            chiselTypeOf(core.io.smem.get(0).d.bits),
+            outer.smemNodes(0).out.head
+          )
+        )
+      }
+      (core.io.smem.get zip smemTLAdapters) foreach { case (coreMem, tlAdapter) =>
+        tlAdapter.io.inReq <> coreMem.a
+        coreMem.d <> tlAdapter.io.inResp
+      }
+      (smemTLAdapters zip smemTLBundles) foreach { case (tlAdapter, tlOut) =>
+        tlOut.a <> tlAdapter.io.outReq
+        tlAdapter.io.outResp <> tlOut.d
+      }
+    }
+
+    connectImem
+    connectDmem
+    connectSmem
+  }
+
   // TODO: generalize for useVxCache
   if (!outer.vortexParams.useVxCache) {}
 }