pow2_ceil.h

// Rounds up to the next nearest power-of-two value
static inline unsigned int pow2_ceil( unsigned int x ) {
	#if defined( __clang__ ) || defined( __GNUC__ )
		return 1 + ( ( x >= 2 ) * ( ( ( 1u << 31 ) >> ( __builtin_clz( ( x - 1 ) | 1 ) - 1 ) ) - 1 ) );
	#elif defined( _MSC_VER ) && ( defined( _M_IX86 ) || defined( _M_X64 ) )
		return ( 1u << 31 ) >> ( __lzcnt( ( x - 1 ) | ( x == 0 ) ) - 1 );
	#elif defined( _MSC_VER ) && ( defined( _M_ARM ) || defined( _M_ARM64 ) )
		return ( 1u << 31 ) >> ( __clz( ( x - 1 ) | ( x == 0 ) ) - 1 );
	#else
	    --x; // if x is already pow2, we don't want the next pow2
		// bit propagation
		x |= x >> 1;
		x |= x >> 2;
		x |= x >> 4;
		x |= x >> 8;
		x |= x >> 16;
		x = x + 1 // bring x from one-less-than-pow2 to pow2
			+ ( x >> 31 ); // we want 1, not 0, when x is 0
		return x;
	#endif
}

// Macro version of pow2_ceil, which rounds a constant (at compile time) up to its nearest power-of-two value
#define _pow2c01( x ) ( (x) | ( (x) >> 1 ) )
#define _pow2c02( x ) ( _pow2c01( x ) | (_pow2c01( x ) >>  2 ) )
#define _pow2c04( x ) ( _pow2c02( x ) | (_pow2c02( x ) >>  4 ) )
#define _pow2c08( x ) ( _pow2c04( x ) | (_pow2c04( x ) >>  8 ) )
#define _pow2c16( x ) ( _pow2c08( x ) | (_pow2c08( x ) >> 16 ) )
#define POW2_CEIL(x) ( ( (x) == 0 ) ? 1 : ( _pow2c16( (x) - 1 ) + 1 ) )

I found out that on the latest compiler versions, my previous intrinsics-based implementations tripped some undefined behavior optimizations. fixed it now. Thanks to @aganm for help with testing this