Division/modulo 2x faster by removing shift in MpUint bit_length

src/private/utils.nim

@@ -11,10 +11,10 @@ macro getSubType*(T: typedesc): untyped =
 
 proc bit_length*[T: SomeUnsignedInt](n: T): int {.noSideEffect.}=
   ## Calculates how many bits are necessary to represent the number
-  result = 1
+
-  var y: T = n shr 1
+  var n = n
-  while y != 0.T:
+  while n != 0.T:
-    y = y shr 1
+    n = n shr 1
     inc(result)
 
 proc bit_length*[T: Natural](n: T): int {.noSideEffect.}=
@@ -22,20 +22,21 @@ proc bit_length*[T: Natural](n: T): int {.noSideEffect.}=
   #
   # For some reason using "SomeUnsignedInt or Natural" directly makes Nim compiler
   # throw a type mismatch
-  result = 1
+
-  var y: T = n shr 1
+  var n = n
-  while y != 0.T:
+  while n != 0.T:
-    y = y shr 1
+    n = n shr 1
     inc(result)
 
 proc bit_length*[T: MpUint](n: T): int {.noSideEffect.}=
   ## Calculates how many bits are necessary to represent the number
-  const zero = T()
+
-  result = 1
+  const maxHalfRepr = n.lo.type.sizeof * 8 - 1
-  var y: T = n shr 1
+
-  while y != zero:
+  if n.hi.bit_length == 0:
-    y = y shr 1 # TODO: shr is slow!
+    n.lo.bit_length
-    inc(result)
+  else:
+    n.hi.bit_length + maxHalfRepr
 
 proc asUint*[T: MpUInt](n: T): auto {.noSideEffect, inline.}=
   ## Casts a multiprecision integer to an uint of the same size

src/uint_logical_ops.nim

@@ -0,0 +1,24 @@
+# Copyright (c) 2018 Status Research & Development GmbH
+# Distributed under the MIT License (license terms are at http://opensource.org/licenses/MIT).
+
+import  uint_type
+
+proc `not`*(x: MpUint): MpUint {.noInit, noSideEffect, inline.}=
+  ## Bitwise complement of unsigned integer x
+  result.lo = not x.lo
+  result.hi = not x.hi
+
+proc `or`*(x, y: MpUint): MpUint {.noInit, noSideEffect, inline.}=
+  ## `Bitwise or` of numbers x and y
+  result.lo = x.lo or y.lo
+  result.hi = x.hi or y.hi
+
+proc `and`*(x, y: MpUint): MpUint {.noInit, noSideEffect, inline.}=
+  ## `Bitwise and` of numbers x and y
+  result.lo = x.lo and y.lo
+  result.hi = x.hi and y.hi
+
+proc `xor`*(x, y: MpUint): MpUint {.noInit, noSideEffect, inline.}=
+  ## `Bitwise xor` of numbers x and y
+  result.lo = x.lo xor y.lo
+  result.hi = x.hi xor y.hi