Atomically increment two integers with CAS

Question

Apparently, it is possible to atomically increment two integers with compare-and-swap instructions. This talk claims that such an algorithm exists but it does not detail wha

Answer 1

If sse2 is available, you can use paddq to add 2 64 bit integers to two other 64 bit integers in one instruction.

#include "emmintrin.h"
//initialize your values somewhere:
//const __m128i ones = _mm_set1_epi64x(1);
//volatile register __m128i vars = 
//    _mm_set_epi64x(24,7);
static inline __m128i inc_both(__m128i vars, __m128i ones){
  return _mm_add_epi64(vars,ones);
}

This should compile to

    paddq  %xmm0, %xmm1

Since it is static inline, it may use other xmm registers though. If there is significant register pressure the ones operands may become ones(℅rip)

Note: this can be used for adding values other than 1 and there are similar operations for most other math, bitwise and compare instructions, should you need them.

So you can use the lock prefix and make it into an inline asm macro

#define inc64x2(vars) asm volatile( \
    "paddq %0, %1\n":"+x"(vars):"x"(ones) \
  );

The arm neon equivalent is something like: vaddq_s64(...), but there is a great article about arm/x86 equivalents here.