ugly testing harness

build.sh

rm *.o test
as -a --gstabs --64 -march=core2 -o funcs64.o funcs64.S
gcc test.c -Wall -m64 -march=core2 -c
gcc funcs64.o test.o -Wall -m64 -march=core2 -o test

funcs64.S

.text
        .align 8
.globl test1
        .type test1,@function
.globl test2
        .type test2,@function
test1:
        pushq %rbp
        movq %rsp,%rbp
        leave
        ret

test2:
        sub $0x8, %rsp
        movq %rbp, (%rsp)
        movq (%rsp), %rbp
        add $0x8, %rsp
        ret

go.sh

for i in {1..2500}                                       
do               
./test >> results.txt
done

test.c

#define _GNU_SOURCE
#include <errno.h>
#include <sched.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>

#include <sys/time.h>

#define TEST_RUNS 500000000

extern void
test1();

extern void
test2();

static __inline__ unsigned long long
rdtsc(void)
{
  unsigned long hi = 0, lo = 0;
  __asm__ __volatile__ ("lfence\n\trdtsc" : "=a"(lo), "=d"(hi));
  return ( (unsigned long long)lo)|( ((unsigned long long)hi)<<32 );
}

static void
run_test(void (*fn)(void), const char *id)
{
  size_t i = 0;
  long long before_cycles = 0, after_cycles = 0;
  struct timeval before_tv, after_tv;

  memset(&before_tv, 0, sizeof before_tv);
  memset(&after_tv, 0, sizeof after_tv);

  gettimeofday(&before_tv, NULL);
  before_cycles = rdtsc();
  for (; i < TEST_RUNS; i++) {
    fn();
  }
  after_cycles = rdtsc();
  gettimeofday(&after_tv, NULL);

  printf("Result %s: %lld cycles\n", id, after_cycles - before_cycles);
  printf("Result %s: %ld microsecs\n",
      id,
      ((after_tv.tv_sec  - before_tv.tv_sec) * 1000000) +
      (after_tv.tv_usec - before_tv.tv_usec));
}

int
main(int argc, char *argv[])
{
  int ret = 0;
  cpu_set_t mask;

  CPU_ZERO(&mask);
  CPU_SET((int)3, &mask);
  ret = sched_setaffinity(0, sizeof mask, &mask);

  if (ret == -1) {
    perror("setting cpu affinity failed");
    exit(EXIT_FAILURE);
  }

  sleep(2);

  run_test(test1, "test1 (frame pointer)");
  run_test(test2, "test2 (no frame pointer)");

  return 0;
}