Static memory alignment benchmark on M3

hyperfine.out

# i +=17

$ hyperfine -w2 ./main # aligned(1)
Benchmark 1: ./main
  Time (mean ± σ):      52.4 ms ±   3.3 ms    [User: 16.9 ms, System: 31.7 ms]
  Range (min … max):    48.5 ms …  65.0 ms    47 runs

$ hyperfine -w2 ./main # aligned(4), ~+10%
Benchmark 1: ./main
  Time (mean ± σ):      57.6 ms ±   3.0 ms    [User: 18.8 ms, System: 34.5 ms]
  Range (min … max):    53.8 ms …  66.4 ms    43 runs

# i += 1

$ hyperfine -w2 ./main # aligned(1)
Benchmark 1: ./main
  Time (mean ± σ):      63.5 ms ±   1.4 ms    [User: 32.9 ms, System: 27.3 ms]
  Range (min … max):    61.6 ms …  70.1 ms    41 runs
  
$ hyperfine -w2 ./main # aligned(4), ~+11%
Benchmark 1: ./main
  Time (mean ± σ):      70.7 ms ±   1.9 ms    [User: 34.5 ms, System: 32.3 ms]
  Range (min … max):    68.3 ms …  76.1 ms    40 runs

main.c

#include <stdint.h>
#include <mach/mach_time.h>
#include <stdio.h>
#include <stdalign.h>

#define COUNT 100000000

// 7 bytes, a on the cache line boundary
struct __attribute__((packed, aligned(1))) {
  uint32_t a;
  uint8_t b;
  uint16_t c;
} test_mem[COUNT];

static inline uint64_t read_cntvct(void) {
    uint64_t cnt;
    asm volatile("mrs %0, cntvct_el0" : "=r"(cnt));
    return cnt;
}

int main() {
  uint64_t tsc1 = mach_absolute_time();

  // do memory stuff here
  long sum = 0;
  for (int i = 0; i < COUNT; i += 17) {
    test_mem[i].a = i % (2<<16);
    test_mem[i].b = i % (2<<16) + 1;
    sum += test_mem[i].a - test_mem[i].b;
  }

  uint64_t tsc2 = mach_absolute_time();
  printf("tsc diff: %lld\n", tsc2 - tsc1);
  printf("sum: %lu\n", sum);
}