fix mask selection

colors.nim

@@ -1,17 +0,0 @@
-type
-  Color* = enum
-    cRed, cGreen, cBlue, cYellow, cPurple
-
-
-const 
-  colorNames: array[Color, string] =
-    ["Red", "Green", "Blue", "Yellow", "Purple"]
-  colorAbbrevs: array[Color, char] = ['R', 'G', 'B', 'Y', 'P']
-
-
-proc `$`*(c: Color): string =
-  result = colorNames[c]
-
-
-proc abbrev*(c: Color): char =
-  result = colorAbbrevs[c]

combinators.nim

@@ -1,5 +1,5 @@
 import algorithm, random, sugar
-import fixedseq, game
+import faststack, fixedseq, game
 
 
 proc nextPermutation(x: var FixedSeq): bool =

config.nims

@@ -1,5 +1,5 @@
---threads: on
---d: release
---opt: speed
---passC: -flto
---passL: -flto
+# --threads: on
+# --d: release
+# --opt: speed
+# --passC: -flto
+# --passL: -flto

faststack.nim

@@ -0,0 +1,204 @@
+import bitops, strutils, random
+
+
+proc show(i: SomeInteger, bitlength = 16) =
+  echo BiggestInt(i).toBin(bitlength)
+
+
+type
+  Color* = enum
+    cRed, cGreen, cBlue, cYellow, cPurple
+
+  ColorStack* = object
+    data: uint16
+    len*: uint8
+
+
+const
+  masks = [
+    0'u16, # dummy value just to get the indices right
+    0b0_000_000_000_000_111,
+    0b0_000_000_000_111_111,
+    0b0_000_000_111_111_111,
+    0b0_000_111_111_111_111,
+    0b0_111_111_111_111_111,
+  ]
+  allColors* = ColorStack(len: 5, data: 0b0_000_001_010_011_100)
+
+
+template offset(s: ColorStack, idx: Natural): uint8 =
+  # Compute the bit offset for a given index. 
+  # Dependent on the stack's length.
+  (s.len - 1 - idx.uint8) * 3
+  # items are stored from left to right but right-aligned, so we
+  # need to shift right to access anything other than the last
+
+
+template offset(s: ColorStack, idx: BackwardsIndex): uint8 =
+  # for backwards index, we are still shifting right but
+  # the lower the index the less we have to shift
+  (idx.uint8 - 1) * 3
+
+
+proc add*(s: var ColorStack, c: Color) =
+  # e.g. if stack is 0b0000000000000100:
+  # and color is 0b00000011
+  # shift: 0b0000000000100000
+  # bitor: 0b0000000000100000 and 0b00000011
+  # results in 0b0000000000100011
+  s.data = (s.data shl 3).bitor(cast[uint8](c))
+  inc s.len
+
+
+proc high*(s: ColorStack): uint8 = s.len - 1
+
+proc low*(s: ColorStack): uint8 = 0 # just... always 0, I guess
+
+
+proc `[]`*(s: ColorStack, i: uint8 | BackwardsIndex): Color =
+  # shift, then mask everything but the three rightmost bits
+  result = Color(
+    (s.data shr s.offset(i)) and masks[1]
+  )
+
+
+proc `[]=`*(s: var ColorStack, i: uint8 | BackwardsIndex, c: Color) =
+  let offset = s.offset(i)
+  s.data = (s.data and bitnot(masks[1] shl offset)) or (c.uint16 shl offset)
+
+
+iterator items*(s: ColorStack): Color =
+  # s.len is unsigned so it will wrap around if we do s.len - 1 in that case
+  if s.len != 0:
+    for i in countdown(s.len - 1, 0'u8):
+      yield Color((s.data shr (i * 3)) and masks[1])
+
+
+iterator pairs*(s: ColorStack): (uint8, Color) =
+  var count = 0'u8
+  for color in s:
+    yield (count, color)
+    inc count
+
+
+proc find*(s: ColorStack, needle: Color): int8 =
+  for i in 0'u8 .. s.high:
+    if s[i] == needle:
+      return i.int8
+  return -1
+
+
+proc moveSubstack*(src, dst: var ColorStack, startIdx: uint8) =
+  if startIdx >= src.len:
+    raise newException(IndexDefect, "index " & $startIdx & " is out of bounds.")
+  # Moves a sub-stack from the top of src to the top of dst
+  # shift the dst stack by the length of the substack to make room
+  let nToMove = src.len - startIdx
+  let shift = nToMove * 3
+  dst.data = dst.data shl shift
+  # then we mask the source data to present only the items 
+  # being moved, and OR that with the shifted dst data
+  dst.data = dst.data or (src.data and masks[nToMove])
+  dst.len += nToMove
+  # then we shift the source to get rid of the moved items
+  src.data = src.data shr shift
+  src.len -= nToMove
+
+
+proc moveSubstackPre*(src, dst: var ColorStack, startIdx: uint8) =
+  if startIdx >= src.len:
+    raise newException(IndexDefect, "index " & $startIdx & " is out of bounds.")
+  # Moves a sub-stack from the top of src to the bottom of dst
+  let nToMove = src.len - startIdx
+  # shift src to position the substack above its destination,
+  # get rid of everything to the left of the substack,
+  # and OR that with the existing dst data
+  let newLen = dst.len + nToMove
+  dst.data = dst.data or ( (src.data shl (dst.len * 3)) and masks[newLen] )
+  dst.len = newLen
+  # get rid of the substack we just moved
+  src.data = src.data shr (nToMove * 3)
+  src.len -= nToMove
+
+
+proc swap*(s: var ColorStack, i1, i2: uint8) =
+  # Swap the values at two indices in the stack
+  if i1 == i2: return
+
+  # i1 and i2 are unsigned, so we have to watch out for underflows
+  let diff = if i1 > i2:
+    (i1 - i2) * 3
+  else:
+    (i2 - i1) * 3
+
+  # take masks[1] from above (rightmost position) and shift to position of i1.
+  # then do the same for i2, and OR them together.
+  let mask = (masks[1] shl s.offset(i1)) or (masks[1] shl s.offset(i2))
+  # get rid of everything but the two values we're swapping
+  let masked = s.data and mask
+  # shift by the distance between values in both directions, combine, then mask
+  let swapped = ((masked shl diff) or (masked shr diff)) and mask
+  # finally, AND with the inverse of mask so that only the values being 
+  # swapped are erased, and combine that with the swapped values
+  s.data = (s.data and mask.bitnot) or swapped
+
+
+proc shuffle*(r: var Rand, s: var ColorStack) =
+  # Fisher-Yates shuffle
+  for i in countdown(s.high, 1'u8):
+    let j = r.rand(i).uint8
+    if j != i:
+      s.swap(i, j)
+
+
+proc reverse*(s: var ColorStack, first, last: uint8) =
+  var x = first
+  var y = last
+  while x < y:
+    s.swap(x, y)
+    inc x
+    dec y
+
+
+iterator asInt*(s: ColorStack): int8 =
+  for i in 0'u8 .. s.high:
+    yield int8(s[i]) # now we do have to convert
+
+
+proc `$`*(s: ColorStack): string =
+  result = "St@["
+  for c in s:
+    if result[^1] != '[':
+      result.add(", ")
+    result.add($c)
+  result.add("]")
+
+
+proc check(s: ColorStack) =
+  # ensure length is accurate
+  var d = s.data
+  for i in 0'u8 .. 4'u8:
+    if (d and masks[1]) > 4:
+      raise newException(RangeDefect, "Value out of range.")
+    if d > 0 and i >= s.len:
+      raise newException(RangeDefect, "Invalid length.")
+    else:
+      d = d shr 3
+
+
+when isMainModule:
+  var one: ColorStack
+  one.add(cRed)
+  one.add(cGreen)
+  one.add(cBlue)
+  one.add(cYellow)
+  one.add(cPurple)
+
+  var two: ColorStack
+
+  one.moveSubstack(two, 2)
+
+
+  echo one, " ", one.len
+  echo two, " ", two.len
+  echo two.find(cRed)

fixedseq.nim

@@ -25,16 +25,6 @@ proc `$`*(s: FixedSeq): string =
   result.add("]")
 
 
-proc `==`*[T1: FixedSeq, T2: FixedSeq](a: T1, b: T2): bool =
-  # generics are so that we can compare ShiftStack vs regular FixedSeq
-  if a.len != b.len:
-    return false
-  for i in 0 ..< a.len:
-    if a.data[i] != b.data[i]:
-      return false
-  return true
-
-
 proc `[]`*(s: FixedSeq, i: Natural): FixedSeq.Contents =
   if i > s.last:
     raise newException(IndexDefect, "index " & $i & " is out of bounds.")
@@ -134,7 +124,7 @@ proc shuffle*(s: var FixedSeq, r: var Rand) =
 
 
 proc moveSubstack*(src, dst: var FixedSeq; start: Natural) =
-  var count: FixedSeq.Pointer = 0 # have to track this separately apparently
+  var count: typeof(src.last) = 0 # have to track this separately apparently
   for idx in start .. src.last:
     swap(src.data[idx], dst.data[dst.last + 1 + count])
     inc count
@@ -143,7 +133,7 @@ proc moveSubstack*(src, dst: var FixedSeq; start: Natural) =
 
 
 proc moveSubstackPre*(src, dst: var FixedSeq; start: Natural) =
-  let ssLen = FixedSeq.Pointer(src.last - start + 1) # length of substack
+  let ssLen = typeof(src.last)(src.last - start + 1) # length of substack
   for i in countdown(dst.last, 0):
     swap(dst.data[i], dst.data[i + ssLen])
 
@@ -154,5 +144,3 @@ proc moveSubstackPre*(src, dst: var FixedSeq; start: Natural) =
 
   dst.last += ssLen
   src.last -= ssLen
-
-include shiftstack

game.nim

@@ -1,17 +1,40 @@
 import hashes, options
-import fixedseq, colors
-export colors
+import fixedseq, faststack
+
+export faststack.Color, faststack.ColorStack, faststack.allColors
 
 
-type
-  ColorStack* = FixedSeq[5, Color, int8]
+# type
+#   Color* = enum
+#     cRed, cGreen, cBlue, cYellow, cPurple
+
+#   ColorStack* = FixedSeq[5, Color, int8]
 
 
-proc initColorStack*: ColorStack =
-  result.initFixedSeq
+# proc initColorStack*: ColorStack =
+#   result.initFixedSeq
 
 
-proc `$`*[T](s: FixedSeq[T, Color, int8]): string =
+# proc getAllColors: ColorStack = 
+#   var i = 0'u8
+#   for c in Color.low .. Color.high:
+#     result[i] = c
+
+const
+  colorNames: array[Color, string] =
+    ["Red", "Green", "Blue", "Yellow", "Purple"]
+  colorAbbrevs: array[Color, char] = ['R', 'G', 'B', 'Y', 'P']
+
+
+proc `$`*(c: Color): string =
+  result = colorNames[c]
+
+
+proc abbrev*(c: Color): char =
+  result = colorAbbrevs[c]
+
+
+proc `$`*(s: ColorStack): string =
   result.add("St@[")
   for i, color in s:
     result.add($color)
@@ -28,7 +51,7 @@ type
     tForward = 1
 
   Square* = object
-    camels*: ShiftStack
+    camels*: ColorStack
     tile*: Option[Tile]
 
   Board* = object
@@ -60,8 +83,8 @@ proc hash*(b: Board): Hash =
 
 
 proc init*(b: var Board) =
-  for sq in b.squares.mitems:
-    sq.camels.initFixedSeq
+  # for sq in b.squares.mitems:
+  #   sq.camels.initFixedSeq
   b.initialized = true
 
 
@@ -90,7 +113,7 @@ proc setState*(b: var Board;
   b.leader = some(leadCamel)
 
 
-proc diceRemaining*(b: Board): ColorStack =
+proc diceRemaining*(b: Board): FixedSeq[5, Color, int8] =
   result.initFixedSeq
   for color, isRolled in b.diceRolled:
     if not isRolled: result.add(color)
@@ -118,11 +141,11 @@ proc advance*(b: var Board, die: Die) =
     endPos += int(t)
     if t == tBackward: prepend = true
 
-  let stackStart = b[startPos].camels.find(color)
+  let stackStart = b[startPos].camels.find(color).uint8
   if prepend:
     b[startPos].camels.moveSubstackPre(b[endPos].camels, stackStart)
     let stackLen = b[startPos].camels.len - stackStart
-    for i in 0 ..< stackLen:
+    for i in 0'u8 ..< stackLen:
       # we know how many camels we added to the bottom, so set the position for each of those
       b.camels[b[endPos].camels[i]] = endPos
   else:
@@ -136,4 +159,4 @@ proc advance*(b: var Board, die: Die) =
   if endPos >= b.camels[b.leader.get]:
     b.leader = some(b[endPos].camels[^1])
   
-  b.diceRolled[color] = true
+  b.diceRolled[color] = true

shiftstack.nim

@@ -1,191 +0,0 @@
-# optimized bit-shifting versions of the FixedSequence substack operations
-import bitops, macros
-import colors
-
-
-macro show(expr: untyped) =
-  let node = expr.toStrLit
-  quote do:
-    echo `node`, " => ", `expr`
-
-
-proc getMasks(): (array[9, uint64], array[9, uint64]) =
-  # on little-endian architectures, casting an array[8, Color] to uint64 effectively
-  # reverses it. So we switch these masks so that we can refer to them consistently.
-  let 
-    left = [
-      0'u64,
-      0xff_00_00_00_00_00_00_00'u64,
-      0xff_ff_00_00_00_00_00_00'u64,
-      0xff_ff_ff_00_00_00_00_00'u64,
-      0xff_ff_ff_ff_00_00_00_00'u64,
-      0xff_ff_ff_ff_ff_00_00_00'u64,
-      0xff_ff_ff_ff_ff_ff_00_00'u64,
-      0xff_ff_ff_ff_ff_ff_ff_00'u64,
-      0xff_ff_ff_ff_ff_ff_ff_ff'u64,
-    ]
-    right = [
-      0'u64,
-      0x00_00_00_00_00_00_00_ff'u64,
-      0x00_00_00_00_00_00_ff_ff'u64,
-      0x00_00_00_00_00_ff_ff_ff'u64,
-      0x00_00_00_00_ff_ff_ff_ff'u64,
-      0x00_00_00_ff_ff_ff_ff_ff'u64,
-      0x00_00_ff_ff_ff_ff_ff_ff'u64,
-      0x00_ff_ff_ff_ff_ff_ff_ff'u64,
-      0xff_ff_ff_ff_ff_ff_ff_ff'u64,
-    ]
-  when cpuEndian == bigEndian:
-    result = (left, right)
-  when cpuEndian == littleEndian:
-    result = (right, left)
-
-
-type ShiftStack* = FixedSeq[8, Color, int8]
-const (masksLeft, masksRight) = getMasks()
-
-
-template `shl`(a: array[8, Color], offset: Natural): array[8, Color] =
-  when cpuEndian == bigEndian:
-    cast[array[8, Color]](cast[uint64](a) shl (offset * 8))
-  when cpuEndian == littleEndian: # direction is reversed
-    cast[array[8, Color]](cast[uint64](a) shr (offset * 8))
-
-
-template `shr`(a: array[8, Color], offset: Natural): array[8, Color] =
-  when cpuEndian == bigEndian:
-    cast[array[8, Color]](cast[uint64](a) shr (offset * 8))
-  when cpuEndian == littleEndian:
-    cast[array[8, Color]](cast[uint64](a) shl (offset * 8))
-
-
-template `and`(a: array[8, Color], mask: uint64): array[8, Color] =
-  cast[array[8, Color]](cast[uint64](a) and mask)
-
-
-template `or`(a: array[8, Color], mask: uint64): array[8, Color] =
-  cast[array[8, Color]](cast[uint64](a) or mask)
-
-
-template `or`(a: array[8, Color], mask: array[8, Color]): array[8, Color] =
-  cast[array[8, Color]](cast[uint64](a) or cast[uint64](mask))
-
-
-import strutils # remove later
-proc moveSubstack*(src, dst: var ShiftStack; start: Natural) =
-  # shift the source stack to position the substack above its final resting place
-  # offset is the length of the destination stack, minus the number of items NOT being moved
-  # number of items not being moved is the same as the start index
-  var substack: array[8, Color] 
-  if dst.len == start: # no shift necessary in this case
-    substack = src.data
-  elif dst.len > start:
-    substack = src.data shr (dst.len - start)
-  elif dst.len < start:
-    substack = src.data shl (start - dst.len)
-  # next, mask the source data to present only the items being moved
-  # dst.len of 0 corresponds to last mask in masksRight, aka masksRight[^1]
-  substack = substack and masksRight[^(dst.len + 1)]
-  # then combine
-  dst.data = dst.data or substack
-  # then git rid of the moved items from the source stack
-  src.data = src.data and masksLeft[start]
-  # a little bookkeeping
-  let ssLen = int8(src.len - start)
-  src.last -= ssLen
-  dst.last += ssLen
-
-
-proc moveSubstackPre*(src, dst: var ShiftStack; start: Natural) =
-  let ssLen = int8(src.len - start)
-  # shift the destination stack to make room for the new items
-  dst.data = dst.data shr ssLen
-  # shift source stack to line up the substack with its final resting place
-  let substack = src.data shl start
-  # combine
-  dst.data = dst.data or substack
-  # get rid of the moved items
-  src.data = src.data and masksLeft[start]
-  # more bookkeeping
-  src.last -= ssLen
-  dst.last += ssLen
-
-
-proc testMove[T1, T2: FixedSeq](a1, a2: var T1; b1, b2: var T2; i: Natural): bool =
-  let (orig_a1, orig_a2) = (a1, a2)
-  let (orig_b1, orig_b2) = (b1, b2)
-  a1.moveSubstack(a2, i)
-  b1.moveSubstack(b2, i)
-  if a1 != b1 or a2 != b2:
-    echo "Failed!"
-    show orig_b1
-    show orig_b2
-    echo "<<move ", i, ">>"
-    show b1
-    show b2
-    return false
-  return true
-
-
-when isMainModule:
-  var c1 = initFixedSeq(5, Color, int8)
-  var c2 = initFixedSeq(5, Color, int8)
-
-  var s1: ShiftStack
-  s1.initFixedSeq
-  var s2: ShiftStack
-  s2.initFixedSeq
-
-  c1.add(cPurple)
-  c1.add(cRed)
-  c1.add(cYellow)
-  c1.add(cBlue)
-  c1.add(cGreen)
-
-  s1.add(cPurple)
-  s1.add(cRed)
-  s1.add(cYellow)
-  s1.add(cBlue)
-  s1.add(cGreen)
-
-  # show s1
-  # show s2
-  # echo "<<move 2>>"
-  # s1.moveSubstack(s2, 2)
-  # show s1
-  # show s2
-
-  import random
-  randomize()
-  var r = initRand(rand(int64))
-  var success = true
-
-  for n in 1 .. 1_000_000:
-    var ranFirst, ranSecond: bool
-
-    if c1.len > 0:
-      let i = r.rand(c1.high)
-      ranFirst = true
-      if not testMove(c1, c2, s1, s2, i):
-        success = false
-        echo "Failed after ", n, " iterations."
-        break
-    else:
-      ranFirst = false
-
-    if c2.len > 0:
-      let j = r.rand(c2.high)
-      ranSecond = true
-      if not testMove(c2, c1, s2, s1, j):
-        success = false
-        echo "Failed after ", n, " iterations."
-        break
-    else:
-      ranSecond = false
-
-    if (not ranFirst) and (not ranSecond):
-      echo "Ran neither first nor second move."
-      break
-
-  if success:
-    echo "Success."

simulation.nim

@@ -1,5 +1,5 @@
 import cpuinfo, math, options, random, tables
-import combinators, game, fixedseq
+import combinators, game, faststack, fixedseq
 
 
 type
@@ -38,7 +38,7 @@ proc percents*(scores: ScoreSet): WinPercents =
 # ======================
 
 iterator legEndStates(b: Board): Board =
-  var diceRemaining: ColorStack
+  var diceRemaining: FixedSeq[5, Color, int8]
   diceRemaining.initFixedSeq
   for i, c in b.diceRolled:
     if not c: diceRemaining.add(i)
@@ -59,6 +59,39 @@ proc getLegScores*(b: Board): ScoreSet =
 # Full-game simulations
 # =====================
 
+# get rid of this later
+import strutils
+proc showSpaces*(b: Board; start, stop: Natural): string =
+  let numSpaces = stop - start + 1
+  let width = 4 * numSpaces - 1
+  var lines: array[7, string]
+  # start by building up an empty board
+  for i in 0 .. 6: # gotta initialize the strings
+    lines[i] = newString(width)
+    for c in lines[i].mitems:
+      c = ' '
+  # fill in the dividers
+  lines[5] = repeat("=== ", numSpaces - 1)
+  lines[5].add("===")
+
+  # now populate the board
+  for sp in 0 ..< numSpaces:
+    # fill in the square numbers
+    let squareNum = sp + start
+    let cellMid = 4 * sp + 1
+    for i, chr in $squareNum:
+      lines[6][cellMid + i] = chr
+
+    # fill in the camel stacks
+    for i, color in b.squares[squareNum].camels:
+      let lineNum = 4 - i # lines go to 6, but bottom 2 are reserved
+      let repr = '|' & color.abbrev & '|'
+      for j, chr in repr:
+        lines[lineNum][cellMid - 1 + j] = chr
+
+  result = lines.join("\n")
+# get rid of this later
+
 proc randomGame*(b: Board, r: var Rand): Color =
   var projection = b
   while true:

test.nim

@@ -1,7 +1,32 @@
-import math, random, strformat, times
+import math, random, strformat, times, std/monotimes
 import fixedseq, game, simulation, ui
 
 
+type
+  TestResults = object
+    ops: int
+    time: Duration
+
+
+proc summarize(tr: TestResults) =
+  let secs = tr.time.inMilliseconds.float / 1000
+  stdout.write("Test completed:\n")
+  stdout.write("  " & $tr.ops, " operations in " & $round(secs, 2) & " seconds\n")
+  stdout.write("  " & $round(tr.ops.float / secs, 2) & " operations per second")
+  stdout.flushFile()
+
+
+template executionTime(body: untyped): Duration =
+  let start = getMonoTime()
+  body
+  getMonoTime() - start
+
+
+proc getRand(): Rand =
+  randomize()
+  result = initRand(rand(int64))
+
+
 proc randomDice(r: var Rand): seq[tuple[c: Color, p: int]] = 
   for c in Color:
     let v = r.rand(1..3)
@@ -18,18 +43,15 @@ proc newRandomGame(r: var Rand): Board =
   result.setState(dice, [])
 
 
-proc testGames(n: SomeInteger = 100): auto =
-  var r = initRand(rand(int64))
-  let dice = randomDice(r)
-  var b: Board
-  b.init
-  b.setState(dice, [])
-  b.display(1, 5)
-
-  let startTime = cpuTime()
-  let scores = b.randomGames(n, parallel = true)
-  result = cpuTime() - startTime
-  scores.display()
+proc games(nTests, nSamples: SomeInteger, parallel = true): TestResults =
+  var r = getRand()
+  var scores: ScoreSet
+  for i in 1 .. nTests:
+    let b = newRandomGame(r)
+    let dur = executionTime:
+        let s = b.randomGames(nSamples, parallel = parallel)
+    result.ops += s.sum()
+    result.time += dur
 
 
 proc testLegs(n: Natural = 100): auto =
@@ -72,24 +94,4 @@ proc testSpread(nTests, nSamples: Natural) =
 
 
 when isMainModule:
-  randomize()
-  # var r = initRand(rand(int64))
-  # let b = newRandomGame(r)
-  # b.display(1, 5)
-  # echo b.showSpaces(1, 16)
-
-  # let scores = b.getLegScores
-  # echo scores.showPercents
-  # let start_states = 2_000
-  # let executionTime = testLegs(start_states)
-  # echo "Execution time: ", executionTime
-  # echo "Leg simulations per second: ", float(start_states * 29_160) / executionTime
-
-  for i in 1 .. 1:
-    let num_games = 100_000_005
-    let executionTime = testGames(num_games)
-    echo "Execution time: ", executionTime
-    echo "Full-game simulations per second: ", float(num_games) / executionTime
-    echo ""
-
-  # testSpread(100, 1_000_000)
+  games(10, 10_000_000).summarize()

ui.nim

@@ -1,5 +1,5 @@
 import os, math, strutils, strformat
-import fixedseq, game, simulation
+import faststack, fixedseq, game, simulation
 
 
 const help =