cacheutils.h

#ifndef _CACHEUTILS_H_
#define _CACHEUTILS_H_

#include <unistd.h>
#include <stdio.h>
#include <stdint.h>

/* ============================================================
 *                    User configuration
 * ============================================================ */
size_t CACHE_MISS = 150;

#define USE_RDTSC_BEGIN_END     0

#define USE_RDTSCP              1

/* ============================================================
 *                  User configuration End
 * ============================================================ */

// ---------------------------------------------------------------------------
uint64_t rdtsc() {
  uint64_t a, d;
  asm volatile("mfence");
#if USE_RDTSCP
  asm volatile("rdtscp" : "=a"(a), "=d"(d) :: "rcx");
#else
  asm volatile("rdtsc" : "=a"(a), "=d"(d));
#endif
  a = (d << 32) | a;
  asm volatile("mfence");
  return a;
}

// ---------------------------------------------------------------------------
void flush(void *p) { asm volatile("clflush 0(%0)\n" : : "c"(p) : "rax"); }

// ---------------------------------------------------------------------------
void maccess(void *p) { asm volatile("movq (%0), %%rax\n" : : "c"(p) : "rax"); }

// ---------------------------------------------------------------------------
void mfence() { asm volatile("mfence"); }

// --------------------------------------------------------------------------
void delayloop(uint32_t cycles) {
  uint64_t start = rdtsc();
  while ((rdtsc()-start) < cycles)
    ;
}

// ---------------------------------------------------------------------------
int flush_reload(void *ptr) {
  uint64_t start = 0, end = 0;

#if USE_RDTSC_BEGIN_END
  start = rdtsc_begin();
#else
  start = rdtsc();
#endif
  maccess(ptr);
#if USE_RDTSC_BEGIN_END
  end = rdtsc_end();
#else
  end = rdtsc();
#endif

  mfence();

  flush(ptr);

  if (end - start < CACHE_MISS) {
    return 1;
  }
  return 0;
}

// ---------------------------------------------------------------------------
int flush_reload_t(void *ptr) {
  uint64_t start = 0, end = 0;

#if USE_RDTSC_BEGIN_END
  start = rdtsc_begin();
#else
  start = rdtsc();
#endif
  maccess(ptr);
#if USE_RDTSC_BEGIN_END
  end = rdtsc_end();
#else
  end = rdtsc();
#endif

  mfence();

  flush(ptr);

  return (int)(end - start);
}

// ---------------------------------------------------------------------------
int reload_t(void *ptr) {
  uint64_t start = 0, end = 0;

#if USE_RDTSC_BEGIN_END
  start = rdtsc_begin();
#else
  start = rdtsc();
#endif
  maccess(ptr);
#if USE_RDTSC_BEGIN_END
  end = rdtsc_end();
#else
  end = rdtsc();
#endif

  mfence();

  return (int)(end - start);
}


// ---------------------------------------------------------------------------
size_t detect_flush_reload_threshold() {
  size_t reload_time = 0, flush_reload_time = 0, i, count = 1000000;
  size_t dummy[16];
  size_t *ptr = dummy + 8;

  maccess(ptr);
  for (i = 0; i < count; i++) {
    reload_time += reload_t(ptr);
  }
  for (i = 0; i < count; i++) {
    flush_reload_time += flush_reload_t(ptr);
  }
  reload_time /= count;
  flush_reload_time /= count;

  return (flush_reload_time + reload_time * 2) / 3;
}


#endif