The first C result is absurd, not sure how the author could have gotten it. First of all, the code as written will just optimize to nothing, so we need to add

The first C result is absurd, not sure how the author could have gotten it. First of all, the code as written will just optimize to nothing, so we need to add an asm("" : "=g" (s) : "0" (s)) in the loop to stop strength reduction and autovectorization, and we need to return the final value to stop dead code elimination. Once that is done, the result is more than 2 billion iterations per second on a ~3 GHz Intel desktop CPU, while the author gives an absurd value of 500m iterations which could not have been possibly obtained with any recent Intel Xeon/Core i5/i7 CPU. BTW, the assembly code produced is this: 1: add $0x1,%edx add $0x1,%esi cmp %eax,%edx jne 1b Which is unlikely to take more than 1/2 cycles to execute on any reasonable CPU as my test data in fact shows.