Source: https://github.com/joonicks/ptr64tohex

This is an experiment in assembly/C programming trying to
increase efficiency by reducing codesize or removing branches.

When converting a binary to hex string, I figured out that its
possible to do it with binary logic instead of if statements.

Normally the code would be;

	a = ... & 0xF
	if (a > 9)
		char = 'A' - 10 + a;
	else
		char = '0' + a;

	But the if statement creates an unpredictable branch
	that tanks performance.

With logic you can do the following instead
(and the gcc compiler does a good job with);

	a = ... & 0xF
	char = '0' + c + ((unsigned)(9 - c) >> 29);

--------------------------------------------------------------------------------------------------

The different implementations is as follows:

x_libc:
	~110-115 clock cycles
	Just a simple wrapper for libc sprintf("%016lX")

c_noob:
	~160 clock cycles
	What a noob might make without using any lookup table

c_luta:
	~22 clock cycles
	Based on a lookup table instead

c_simp:
	~25 clock cycles
	Simple loop based conversion using logic instead of lookup
	a = nibble & 15
	char = ((((a + 6) & 16) >> 4) * 7) + a + '0';

c_nobr:
	~9.5 clock cycles
	Branchless, all low nibbles then all high nibbles done at once
	gcc compiler uses extra instructions to improve speed

c_simd: (by dzaima)
	~4.5 clock cycles
	SIMD version using SSE instructions

a_noob:
	~160 clock cycles
	Equivalent to c_noob slightly tweaked non lookup

a_bext:
	~27 clock cycles
	Loop 1 nibble at a time using bextr instruction

a_nobr:
	~8 clock cycles
	Assembly version of c_nobr
	Better ordering of instructions could improve speed