How to use MSVC intrinsics to get the equivalent o

Question

The following code calls the builtin functions for clz/ctz in GCC and, on other systems, has C versions. Obviously, the C versions are a bit suboptimal if the system has a builtin clz/ctz instruction, like x86 and ARM.

#ifdef __GNUC__
#define clz(x) __builtin_clz(x)
#define ctz(x) __builtin_ctz(x)
#else
static uint32_t ALWAYS_INLINE popcnt( uint32_t x )
{
    x -= ((x >> 1) & 0x55555555);
    x = (((x >> 2) & 0x33333333) + (x & 0x33333333));
    x = (((x >> 4) + x) & 0x0f0f0f0f);
    x += (x >> 8);
    x += (x >> 16);
    return x & 0x0000003f;
}
static uint32_t ALWAYS_INLINE clz( uint32_t x )
{
    x |= (x >> 1);
    x |= (x >> 2);
    x |= (x >> 4);
    x |= (x >> 8);
    x |= (x >> 16);
    return 32 - popcnt(x);
}
static uint32_t ALWAYS_INLINE ctz( uint32_t x )
{
    return popcnt((x & -x) - 1);
}

#endif

What functions do I need to call, which headers do I need to include, etc to add a proper ifdef for MSVC here? I've already looked at this page, but I'm not entirely sure what the #pragma is for (is it required?) and what restrictions it puts on MSVC version requirements for compilation. As someone who doesn't really use MSVC, I also don't know whether these intrinsics have C equivalents on other architectures, or whether I have to #ifdef x86/x86_64 as well when #defining them.

Answer 1

There are two intrinsics "_BitScanForward" and "_BitScanReverse", which suits the same purpose for MSVC. Include . The functions are:

#ifdef _MSC_VER
#include <intrin.h>

static uint32_t __inline ctz( uint32_t x )
{
   int r = 0;
   _BitScanReverse(&r, x);
   return r;
}

static uint32_t __inline clz( uint32_t x )
{
   int r = 0;
   _BitScanForward(&r, x);
   return r;
}
#endif

There are equivalent 64bit versions "_BitScanForward64" and "_BitScanReverse64".

Read more here:

x86 Intrinsics on MSDN

Answer 2

If MSVC has a compiler intrinsic for this, it'll be here:

Compiler Intrinsics on MSDN

Otherwise, you'll have to write it using __asm

Answer 3

Tested on linux and windows (x86) :

#ifdef WIN32
    #include <intrin.h>
    static uint32_t __inline __builtin_clz(uint32_t x) {
        unsigned long r = 0;
        _BitScanReverse(&r, x);
        return (31-r);
    }
#endif

uint32_t clz64(const uint64_t x)
{
    uint32_t u32 = (x >> 32);
    uint32_t result = u32 ? __builtin_clz(u32) : 32;
    if (result == 32) {
        u32 = x & 0xFFFFFFFFUL;
        result += (u32 ? __builtin_clz(u32) : 32);
    }
    return result;
}

Answer 4

Bouncing from sh0dan code, the implementation should be corrected like this :

#ifdef _MSC_VER
#include <intrin.h>

uint32_t __inline ctz( uint32_t value )
{
    DWORD trailing_zero = 0;

    if ( _BitScanForward( &trailing_zero, value ) )
    {
        return trailing_zero;
    }
    else
    {
        // This is undefined, I better choose 32 than 0
        return 32;
    }
}

uint32_t __inline clz( uint32_t value )
{
    DWORD leading_zero = 0;

    if ( _BitScanReverse( &leading_zero, value ) )
    {
       return 31 - leading_zero;
    }
    else
    {
         // Same remarks as above
         return 32;
    }
}
#endif

As commented in the code, both ctz and clz are undefined if value is 0. In our abstraction, we fixed __builtin_clz(value) as (value?__builtin_clz(value):32) but it's a choice