bpo-29782: Consolidate _Py_Bit_Length()

In pythonGH-2866, _Py_Bit_Length() was added to pymath.h for lack of a better
location. pythonGH-20518 added a more appropriate header file for bit utilities. It
also shows how to properly use intrinsics. This allows reconsidering bpo-29782.

* Move the function to the new header.
* Changed return type to match __builtin_clzl and reviewed usage.
* Use intrinsics where available.
* Pick a (mostly theoretical) fallback implementation suitable for
  inlining.

Author: niklasf

Include/internal/pycore_bitutils.h

@@ -7,8 +7,8 @@
    - _Py_bswap64(uint64_t)
 */
 
-#ifndef Py_INTERNAL_BSWAP_H
-#define Py_INTERNAL_BSWAP_H
+#ifndef Py_INTERNAL_BITUTILS_H
+#define Py_INTERNAL_BITUTILS_H
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -131,8 +131,32 @@ _Py_popcount32(uint32_t x)
 }
 
 
+// Return the index of the most significant 1 bit in 'x'. This is the smallest
+// integer k such that x < 2**k. Equivalent to floor(log2(x)) + 1 for x != 0.
+static inline int
+_Py_bit_length(unsigned long x) {
+    if (!x) {
+        return 0;
+    }
+#if (defined(__clang__) || defined(__GNUC__))
+    return sizeof(unsigned long) * 8 - __builtin_clzl(x);
+#elif defined(_MSC_VER)
+    Py_BUILD_ASSERT(4 == sizeof(unsigned long));
+    unsigned long msb;
+    _BitScanReverse(&msb, x);
+    return 32 - msb;
+#else
+    int msb = 0;
+    while (x != 0) {
+        msb += 1;
+        x >>= 1;
+    }
+    return msb;
+#endif
+}
+
+
 #ifdef __cplusplus
 }
 #endif
-#endif /* !Py_INTERNAL_BSWAP_H */
-
+#endif /* !Py_INTERNAL_BITUTILS_H */

Include/pymath.h

@@ -227,12 +227,4 @@ PyAPI_FUNC(void) _Py_set_387controlword(unsigned short);
  * behavior. */
 #define _Py_InIntegralTypeRange(type, v) (_Py_IntegralTypeMin(type) <= v && v <= _Py_IntegralTypeMax(type))
 
-/* Return the smallest integer k such that n < 2**k, or 0 if n == 0.
- * Equivalent to floor(log2(x))+1.  Also equivalent to: bitwidth_of_type -
- * count_leading_zero_bits(x)
- */
-#ifndef Py_LIMITED_API
-PyAPI_FUNC(unsigned int) _Py_bit_length(unsigned long d);
-#endif
-
 #endif /* Py_PYMATH_H */

Modules/mathmodule.c

@@ -53,6 +53,7 @@ raised for division by zero and mod by zero.
  */
 
 #include "Python.h"
+#include "pycore_bitutils.h"
 #include "pycore_dtoa.h"
 #include "_math.h"
 

Python/pymath.c

@@ -79,18 +79,3 @@ round(double x)
     return copysign(y, x);
 }
 #endif /* HAVE_ROUND */
-
-static const unsigned int BitLengthTable[32] = {
-    0, 1, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4,
-    5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5
-};
-
-unsigned int _Py_bit_length(unsigned long d) {
-   unsigned int d_bits = 0;
-   while (d >= 32) {
-       d_bits += 6;
-       d >>= 6;
-   }
-   d_bits += BitLengthTable[d];
-   return d_bits;
-}