Über fast thread-local storage on OS X

fast-tls.c

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <pthread.h>

/** Snipped from pthread_machdep.h */
#define __PTK_FRAMEWORK_JAVASCRIPTCORE_KEY4 94

extern __inline__ void *
_pthread_getspecific_direct(unsigned long slot) {
  void* ret;
#if defined(__i386__) || defined(__x86_64__)
  __asm__("mov %%gs:%1, %0" : "=r" (ret) : "m" (*(void **)(slot * sizeof(void *))));
#else
#error "No definition of pthread_getspecific_direct!"
#endif
  return ret;
}


/* To be used with static constant keys only */
extern __inline__ int
_pthread_setspecific_direct(unsigned long slot, void * val)
{
#if defined(__x86_64__)
  /* PIC is free and cannot be disabled, even with: gcc -mdynamic-no-pic ... */
  __asm__("movq %1,%%gs:%0" : "=m" (*(void **)(slot * sizeof(void *))) : "rn" (val));
#else
#error "No definition of pthread_setspecific_direct!"
#endif
  return 0;
}
/** End snippets */ 

static const pthread_key_t fooKey =
  __PTK_FRAMEWORK_JAVASCRIPTCORE_KEY4;
#define GET_FOO() ((void*)(_pthread_getspecific_direct(fooKey)))
#define SET_FOO(to) (_pthread_setspecific_direct(fooKey, to))

void* f1(void* x) {
  int* tmp;

  SET_FOO(x);
  sleep(10);

  tmp = GET_FOO();
  printf("thread #1: foo = %p/%d\n", tmp, *tmp);

  /* el fin */
  free(x);
  pthread_exit(0);
}

void* f2(void* x) {
  int* tmp;

  sleep(5);
  SET_FOO(x);

  tmp = GET_FOO();
  printf("thread #2: foo = %p/%d\n", tmp, *tmp);

  /* el fin */
  free(x);
  pthread_exit(0);
}

int main(int ac, char* av[]) {
  int foobar = 0;
  if (ac < 2) foobar = 10;
  else        foobar = atoi(av[1]);
  printf("[MAIN] default value = %d\n", foobar);

  /* initialize thread-local data structures */
  int* i1 = malloc(sizeof(*i1));
  int* i2 = malloc(sizeof(*i2));  
  *i1 = foobar+1;
  *i2 = foobar+2;
  printf("[MAIN] mallocs: thr1 = %p; thr2 = %p\n", i1, i2);

  /* initialize threads */
  pthread_t thr1, thr2;

  pthread_create(&thr1, NULL, &f1, i1);
  pthread_create(&thr2, NULL, &f2, i2);
  pthread_join(thr1, NULL);
  pthread_join(thr2, NULL);
  return 0;
}