diff --git a/src/private/bithacks.nim b/src/private/bithacks.nim
index 8470845..bf50436 100644
--- a/src/private/bithacks.nim
+++ b/src/private/bithacks.nim
@@ -53,3 +53,10 @@ proc bit_length*(n: MpUintImpl): int {.noSideEffect.}=
   let hi_bitlen = n.hi.bit_length
   result = if hi_bitlen == 0: n.lo.bit_length
            else: hi_bitlen + maxHalfRepr
+
+
+proc countLeadingZeroBits*(x: MpUintImpl): int {.inline, nosideeffect.} =
+  ## Returns the number of leading zero bits in integer.
+  let hi_clz = x.hi.countLeadingZeroBits
+  result = if hi_clz == 0: x.lo.countLeadingZeroBits
+           else: hi_clz
diff --git a/src/private/conversion.nim b/src/private/conversion.nim
index 53ffc57..8b90193 100644
--- a/src/private/conversion.nim
+++ b/src/private/conversion.nim
@@ -7,14 +7,24 @@
 #
 # at your option. This file may not be copied, modified, or distributed except according to those terms.
 
-import  ./uint_type,
+import  ./uint_type, ./size_mpuintimpl,
         macros
 
-# converter mpUint64*(x: MpUintImpl[uint32]): uint64 =
-#   cast[ptr uint64](unsafeAddr x)[]
+proc initMpUintImpl*[InType, OutType](x: InType, _: typedesc[OutType]): OutType {.noSideEffect.} =
 
-proc initMpUintImpl*[T: BaseUint](x: T): MpUintImpl[T] {.inline,noSideEffect.} =
-  result.lo = x
+  const
+    size_in = size_mpuintimpl(x)
+    size_out = size_mpuintimpl(result)
+
+  static:
+    assert size_out >= size_in, "The result type size should be equal or bigger than the input type size"
+
+  when OutType is SomeUnsignedInt:
+    result = x.OutType
+  elif size_in == size_out:
+    result = cast[type result](x)
+  else:
+    result.lo = initMpUintImpl(x, type result.lo)
 
 proc toSubtype*[T: SomeInteger](b: bool, _: typedesc[T]): T {.noSideEffect, inline.}=
   b.T
@@ -24,7 +34,7 @@ proc toSubtype*[T: MpUintImpl](b: bool, _: typedesc[T]): T {.noSideEffect, inlin
   result.lo = toSubtype(b, SubTy)
 
 proc zero*[T: BaseUint](_: typedesc[T]): T {.noSideEffect, inline.}=
-  result = T(0)
+  discard
 
 proc one*[T: BaseUint](_: typedesc[T]): T {.noSideEffect, inline.}=
   when T is SomeUnsignedInt:
@@ -73,3 +83,7 @@ proc toMpUintImpl*(n: uint16|uint32|uint64): auto {.noSideEffect, inline.} =
     return (cast[ptr [MpUintImpl[uint16]]](unsafeAddr n))[]
   elif n is uint16:
     return (cast[ptr [MpUintImpl[uint8]]](unsafeAddr n))[]
+
+proc toMpUintImpl*(n: MpUintImpl): MpUintImpl {.noSideEffect, inline.} =
+  ## No op
+  n
diff --git a/src/private/uint_binary_ops.nim b/src/private/uint_binary_ops.nim
index 8c5b5f2..ee48a71 100644
--- a/src/private/uint_binary_ops.nim
+++ b/src/private/uint_binary_ops.nim
@@ -52,16 +52,22 @@ proc `-`*(x, y: MpUintImpl): MpUintImpl {.noSideEffect, noInit, inline.}=
 
 proc naiveMulImpl[T: MpUintImpl](x, y: T): MpUintImpl[T] {.noSideEffect, noInit, inline.}
   # Forward declaration
-
+import typetraits
 proc naiveMul[T: BaseUint](x, y: T): MpUintImpl[T] {.noSideEffect, noInit, inline.}=
   ## Naive multiplication algorithm with extended precision
 
-  when T.sizeof in {1, 2, 4}:
+  const size = size_mpuintimpl(x)
+
+  when size in {8, 16, 32}:
     # Use types twice bigger to do the multiplication
     cast[type result](x.asDoubleUint * y.asDoubleUint)
 
-  elif T.sizeof == 8: # uint64 or MpUint[uint32]
+  elif size == 64: # uint64 or MpUint[uint32]
     # We cannot double uint64 to uint128
+    static:
+      echo "####"
+      echo x.type.name
+      echo size
     cast[type result](naiveMulImpl(x.toMpUintImpl, y.toMpUintImpl))
   else:
     # Case: at least uint128 * uint128 --> uint256
@@ -92,10 +98,10 @@ proc naiveMulImpl[T: MpUintImpl](x, y: T): MpUintImpl[T] {.noSideEffect, noInit,
   z1 += naiveMul(x.hi, y.lo)
   let z2 = (z1 < tmp).toSubtype(T) + naiveMul(x.hi, y.hi)
 
-  let tmp2  = initMpUintImpl(z1.lo shl halfSize)
+  let tmp2  = initMpUintImpl(z1.lo shl halfSize, T)
   result.lo = tmp2
   result.lo += z0
-  result.hi = (result.lo < tmp2).toSubtype(T) + z2 + initMpUintImpl(z1.hi)
+  result.hi = (result.lo < tmp2).toSubtype(T) + z2 + initMpUintImpl(z1.hi, type result.hi)
 
 proc `*`*(x, y: MpUintImpl): MpUintImpl {.noSideEffect, noInit.}=
   ## Multiplication for multi-precision unsigned uint
diff --git a/src/private/uint_bitwise_ops.nim b/src/private/uint_bitwise_ops.nim
index d1302d1..9918cc1 100644
--- a/src/private/uint_bitwise_ops.nim
+++ b/src/private/uint_bitwise_ops.nim
@@ -7,7 +7,7 @@
 #
 # at your option. This file may not be copied, modified, or distributed except according to those terms.
 
-import  ./uint_type, size_mpuintimpl
+import  ./uint_type, ./size_mpuintimpl, ./conversion
 
 
 proc `not`*(x: MpUintImpl): MpUintImpl {.noInit, noSideEffect, inline.}=
@@ -30,25 +30,23 @@ proc `xor`*(x, y: MpUintImpl): MpUintImpl {.noInit, noSideEffect, inline.}=
   result.lo = x.lo xor y.lo
   result.hi = x.hi xor y.hi
 
-proc `shl`*(x: MpUintImpl, y: SomeInteger): MpUintImpl {.noInit, inline, noSideEffect.}=
+proc `shl`*(x: MpUintImpl, y: SomeInteger): MpUintImpl {.inline, noSideEffect.}=
   ## Compute the `shift left` operation of x and y
   # Note: inlining this poses codegen/aliasing issue when doing `x = x shl 1`
   const halfSize = size_mpuintimpl(x) div 2
 
-  type SubTy = type x.lo
-
   result.hi = (x.hi shl y) or (x.lo shl (y - halfSize))
-  result.lo = if y < halfSize: x.lo shl y
-              else: 0.SubTy
+  if y < halfSize:
+    result.lo = x.lo shl y
 
-
-proc `shr`*(x: MpUintImpl, y: SomeInteger): MpUintImpl {.noInit, inline, noSideEffect.}=
+proc `shr`*(x: MpUintImpl, y: SomeInteger): MpUintImpl {.inline, noSideEffect.}=
   ## Compute the `shift right` operation of x and y
-  # Note: inlining this poses codegen/aliasing issue when doing `x = x shl 1`
   const halfSize = size_mpuintimpl(x) div 2
 
-  type SubTy = type x.lo
+  let overflow = y < halfSize
 
-  result.lo = (x.lo shr y) or (x.hi shl (y - halfSize)) # the shl is not a mistake
-  result.hi = if y < halfSize: x.hi shr y
-              else: 0.SubTy
+  result.lo = (x.lo shr y) or (
+    if overflow: x.hi shl (halfSize - y) else: x.hi shr (y - halfSize)
+  )
+  if overflow:
+    result.hi = x.hi shr y
diff --git a/src/uint_init.nim b/src/uint_init.nim
index 9aa8986..c563372 100644
--- a/src/uint_init.nim
+++ b/src/uint_init.nim
@@ -14,18 +14,17 @@ import  ./private/bithacks, ./private/conversion,
 
 import typetraits
 
-proc initMpUint*(n: SomeInteger, bits: static[int]): MpUint[bits] {.noSideEffect.} =
-  assert n >= 0
+proc initMpUint*[T: SomeInteger](n: T, bits: static[int]): MpUint[bits] {.noSideEffect.} =
+  assert n >= 0.T
   when result.data is MpuintImpl:
-    type SubTy = type result.data.lo
-
+    static: echo result.data.type.name
     let len = n.bit_length
     if len > bits:
       raise newException(ValueError, "Input " & $n & " cannot be stored in a multi-precision " &
                                     $bits & "-bit integer." &
                                     "\nIt requires at least " & $len & " bits of precision")
     elif len < bits div 2:
-      result.data.lo = SubTy(n)
+      result.data.lo = initMpUintImpl(n, type result.data.lo)
     else: # Both have the same size and memory representation
       when bits == 16:
         # TODO: If n is int it's not properly converted at the input
diff --git a/tests/all_tests.nim b/tests/all_tests.nim
index 785a027..bf8fa19 100644
--- a/tests/all_tests.nim
+++ b/tests/all_tests.nim
@@ -9,5 +9,6 @@
 
 import  test_endianness,
         test_comparison,
+        test_bitwise,
         test_addsub,
         test_muldiv
diff --git a/tests/test_bitwise.nim b/tests/test_bitwise.nim
index bf5ceb9..0929492 100644
--- a/tests/test_bitwise.nim
+++ b/tests/test_bitwise.nim
@@ -14,6 +14,7 @@ suite "Testing bitwise operations":
 
   let b = a * a
   let z = 10000'u16
+  assert cast[uint16](b) == z, "Test cannot proceed, something is wrong with the multiplication implementation"
 
   test "Shift left - by less than half the size of the integer":
     check: cast[uint16](b) == z # Sanity check