RegMapper: regmap(...) now takes BYTE addresses

If a device has configurable bus-width, we need a stable way of
enumerating registers. The byte offset stays unchanged.

This change also makes it possible to put an arbitrary number of RegFields
starting at some address which are then chopped up into appropriately bus-
sized registers.

src/main/scala/uncore/devices/Prci.scala

@@ -74,14 +74,11 @@ trait CoreplexLocalInterrupterModule extends Module with HasRegMap with MixCorep
    * bff8 mtime lo
    * bffc mtime hi
    */
-  val ipi_base     = 0
-  val timecmp_base = c.timecmpOffset(0) / c.beatBytes
-  val time_base    = c.timeOffset / c.beatBytes
 
-  regmap((
+  regmap(
-    RegField.split(makeRegFields(ipi),             ipi_base,     c.beatBytes) ++
+    0                  -> makeRegFields(ipi),
-    RegField.split(makeRegFields(timecmp.flatten), timecmp_base, c.beatBytes) ++
+    c.timecmpOffset(0) -> makeRegFields(timecmp.flatten),
-    RegField.split(makeRegFields(time),            time_base,    c.beatBytes)):_*)
+    c.timeOffset       -> makeRegFields(time))
 
   def makeRegFields(s: Seq[UInt]) = s.map(r => RegField(regWidth, r))
 }

src/main/scala/uncore/tilelink2/Example.scala

@@ -27,7 +27,7 @@ trait ExampleModule extends HasRegMap
   regmap(
     0 -> Seq(
       RegField(params.num, state)),
-    1 -> Seq(
+    4 -> Seq(
       RegField.w1ToClear(4, pending, state)))
 }
 

src/main/scala/uncore/tilelink2/RegField.scala

@@ -85,7 +85,9 @@ case class RegField(width: Int, read: RegReadFn, write: RegWriteFn)
 
 object RegField
 {
+  // Byte address => sequence of bitfields, lowest index => lowest address
   type Map = (Int, Seq[RegField])
+
   def apply(n: Int)            : RegField = apply(n, (), ())
   def apply(n: Int, rw: UInt)  : RegField = apply(n, rw, rw)
   def r(n: Int, r: RegReadFn)  : RegField = apply(n, r, ())

src/main/scala/uncore/tilelink2/RegMapper.scala

@@ -29,26 +29,35 @@ object RegMapper
 {
   // Create a generic register-based device
   def apply(bytes: Int, concurrency: Int, undefZero: Boolean, in: DecoupledIO[RegMapperInput], mapping: RegField.Map*) = {
-    val regmap = mapping.toList.filter(!_._2.isEmpty)
+    val bytemap = mapping.toList
-    require (!regmap.isEmpty)
-
-    // Ensure no register appears twice
-    regmap.combinations(2).foreach { case Seq((reg1, _), (reg2, _)) =>
-      require (reg1 != reg2)
-    }
     // Don't be an asshole...
-    regmap.foreach { reg => require (reg._1 >= 0) }
+    bytemap.foreach { byte => require (byte._1 >= 0) }
+
+    // Transform all fields into bit offsets Seq[(bit, field)]
+    val bitmap = bytemap.map { case (byte, fields) =>
+      val bits = fields.scanLeft(byte * 8)(_ + _.width).init
+      bits zip fields
+    }.flatten.sortBy(_._1)
+
+    // Detect overlaps
+    (bitmap.init zip bitmap.tail) foreach { case ((lbit, lfield), (rbit, rfield)) =>
+      require (lbit + lfield.width <= rbit, s"Register map overlaps at bit ${rbit}.")
+    }
+
+    // Group those fields into bus words Map[word, List[(bit, field)]]
+    val wordmap = bitmap.groupBy(_._1 / (8*bytes))
+
     // Make sure registers fit
     val inParams = in.bits.params
     val inBits = inParams.indexBits
-    assert (regmap.map(_._1).max < (1 << inBits))
+    assert (wordmap.keySet.max < (1 << inBits), "Register map does not fit in device")
 
     val out = Wire(Irrevocable(new RegMapperOutput(inParams)))
     val front = Wire(Irrevocable(new RegMapperInput(inParams)))
     front.bits := in.bits
 
     // Must this device pipeline the control channel?
-    val pipelined = regmap.map(_._2.map(_.pipelined)).flatten.reduce(_ || _)
+    val pipelined = wordmap.values.map(_.map(_._2.pipelined)).flatten.reduce(_ || _)
     val depth = concurrency
     require (depth >= 0)
     require (!pipelined || depth > 0, "Register-based device with request/response handshaking needs concurrency > 0")
@@ -60,7 +69,7 @@ object RegMapper
     def ofBits(bits: List[Boolean]) = bits.foldRight(0){ case (x,y) => (if (x) 1 else 0) | y << 1 }
 
     // Find the minimal mask that can decide the register map
-    val mask = AddressDecoder(regmap.map(_._1))
+    val mask = AddressDecoder(wordmap.keySet.toList)
     val maskMatch = ~UInt(mask, width = inBits)
     val maskFilter = toBits(mask)
     val maskBits = maskFilter.filter(x => x).size
@@ -75,17 +84,22 @@ object RegMapper
     val iRightReg = Array.fill(regSize) { Bool(true) }
     val oRightReg = Array.fill(regSize) { Bool(true) }
 
-    // Flatten the regmap into (RegIndex:Int, Offset:Int, field:RegField)
+    // Transform the wordmap into minimal decoded indexes, Seq[(index, bit, field)]
-    val flat = regmap.map { case (reg, fields) =>
+    val flat = wordmap.toList.map { case (word, fields) =>
-      val offsets = fields.scanLeft(0)(_ + _.width).init
+      val index = regIndexI(word)
-      val index = regIndexI(reg)
+      val uint = UInt(word, width = inBits)
-      val uint = UInt(reg, width = inBits)
       if (undefZero) {
         iRightReg(index) = ((front.bits.index ^ uint) & maskMatch) === UInt(0)
         oRightReg(index) = ((back .bits.index ^ uint) & maskMatch) === UInt(0)
       }
-      // println("mapping 0x%x -> 0x%x for 0x%x/%d".format(reg, index, mask, maskBits))
+      // Confirm that no field spans a word boundary
-      (offsets zip fields) map { case (o, f) => (index, o, f) }
+      fields foreach { case (bit, field) =>
+        val off = bit - 8*bytes*word
+        // println(s"Reg ${word}: [${off}, ${off+field.width})")
+        require (off + field.width <= bytes * 8, s"Field at word ${word}*(${bytes}B) has bits [${off}, ${off+field.width}), which exceeds word limit.")
+      }
+      // println("mapping 0x%x -> 0x%x for 0x%x/%d".format(word, index, mask, maskBits))
+      fields.map { case (bit, field) => (index, bit-8*bytes*word, field) }
     }.flatten
 
     // Forward declaration of all flow control signals

src/main/scala/uncore/tilelink2/RegisterRouterTest.scala

@@ -183,14 +183,14 @@ object RRTest0Map
 
   // All fields must respect byte alignment, or else it won't behave like an SRAM
   val map = Seq(
-    0 -> Seq(aa(8), ar(8), ad(8), ae(8)),
+    0  -> Seq(aa(8), ar(8), ad(8), ae(8)),
-    1 -> Seq(ra(8), rr(8), rd(8), re(8)),
+    4  -> Seq(ra(8), rr(8), rd(8), re(8)),
-    2 -> Seq(da(8), dr(8), dd(8), de(8)),
+    8  -> Seq(da(8), dr(8), dd(8), de(8)),
-    3 -> Seq(ea(8), er(8), ed(8), ee(8)),
+    12 -> Seq(ea(8), er(8), ed(8), ee(8)),
-    4 -> Seq(aa(3), ar(5), ad(1), ae(7), ra(2), rr(6), rd(4), re(4)),
+    16 -> Seq(aa(3), ar(5), ad(1), ae(7), ra(2), rr(6), rd(4), re(4)),
-    5 -> Seq(da(3), dr(5), dd(1), de(7), ea(2), er(6), ed(4), ee(4)),
+    20 -> Seq(da(3), dr(5), dd(1), de(7), ea(2), er(6), ed(4), ee(4)),
-    6 -> Seq(aa(8), rr(8), dd(8), ee(8)),
+    24 -> Seq(aa(8), rr(8), dd(8), ee(8)),
-    7 -> Seq(ar(8), rd(8), de(8), ea(8)))
+    28 -> Seq(ar(8), rd(8), de(8), ea(8)))
 }
 
 object RRTest1Map
@@ -203,8 +203,8 @@ object RRTest1Map
   def bb(bits: Int) = request(bits, busy, busy)
 
   val map = RRTest0Map.map.take(6) ++ Seq(
-    6 -> Seq(pp(8), pb(8), bp(8), bb(8)),
+    24 -> Seq(pp(8), pb(8), bp(8), bb(8)),
-    7 -> Seq(pp(3), pb(5), bp(1), bb(7), pb(5), bp(3), pp(4), bb(4)))
+    28 -> Seq(pp(3), pb(5), bp(1), bb(7), pb(5), bp(3), pp(4), bb(4)))
 }
 
 trait RRTest0Bundle