Simplified size class calculation

The rounding code can be integrated directly into the sizeclass code,
which means, the code paths can be simplified slightly.

The special case for zero mantissa bits is not required, as the compiler
generates the same code from the more complex path due to
eliminating dead code after folding the constant zero through.

src/ds/bits.h

@@ -396,52 +396,33 @@ namespace snmalloc
     template<size_t MANTISSA_BITS, size_t LOW_BITS = 0>
     static size_t to_exp_mant(size_t value)
     {
-      value += ((size_t)1 << (LOW_BITS)) - 1;
-      value >>= LOW_BITS;
+      size_t LEADING_BIT = ((size_t)1 << (MANTISSA_BITS + LOW_BITS)) >> 1;
+      size_t MANTISSA_MASK = ((size_t)1 << MANTISSA_BITS) - 1;
 
-      if (MANTISSA_BITS > 0)
-      {
-        size_t LEADING_BIT = ((size_t)1 << MANTISSA_BITS) >> 1;
-        size_t MANTISSA_MASK = ((size_t)1 << MANTISSA_BITS) - 1;
+      value = value - 1;
 
-        value = value - 1;
+      size_t e =
+        bits::BITS - MANTISSA_BITS - LOW_BITS - clz(value | LEADING_BIT) ;
+      size_t b = (e == 0) ? 0 : 1;
+      size_t m = (value >> (LOW_BITS + e - b)) & MANTISSA_MASK;
 
-        size_t e = (bits::BITS - clz(value | LEADING_BIT)) - MANTISSA_BITS;
-        size_t shift_e = (e == 0) ? 0 : e - 1;
-        size_t m = (value >> shift_e) & MANTISSA_MASK;
-
-        return (e << MANTISSA_BITS) + m;
-      }
-      else
-      {
-        return bits::next_pow2_bits(value);
-      }
+      return (e << MANTISSA_BITS) + m;
     }
 
     template<size_t MANTISSA_BITS, size_t LOW_BITS = 0>
     constexpr static size_t to_exp_mant_const(size_t value)
     {
-      value += ((size_t)1 << LOW_BITS) - 1;
-      value >>= LOW_BITS;
+      size_t LEADING_BIT = ((size_t)1 << (MANTISSA_BITS + LOW_BITS)) >> 1;
+      size_t MANTISSA_MASK = ((size_t)1 << MANTISSA_BITS) - 1;
 
-      if (MANTISSA_BITS > 0)
-      {
-        size_t LEADING_BIT = (size_t)1 << (MANTISSA_BITS - 1);
-        size_t MANTISSA_MASK = ((size_t)1 << MANTISSA_BITS) - 1;
+      value = value - 1;
 
-        value = value - 1;
+      size_t e = bits::BITS -
+        MANTISSA_BITS - LOW_BITS - clz_const(value | LEADING_BIT) ;
+      size_t b = (e == 0) ? 0 : 1;
+      size_t m = (value >> (LOW_BITS + e - b)) & MANTISSA_MASK;
 
-        size_t e =
-          (bits::BITS - clz_const(value | LEADING_BIT)) - MANTISSA_BITS;
-        size_t shift_e = (e == 0) ? 0 : e - 1;
-        size_t m = (value >> shift_e) & MANTISSA_MASK;
-
-        return (e << MANTISSA_BITS) + m;
-      }
-      else
-      {
-        return bits::next_pow2_bits_const(value);
-      }
+      return (e << MANTISSA_BITS) + m;
     }
 
     template<size_t MANTISSA_BITS, size_t LOW_BITS = 0>