#define _GNU_SOURCE
#include <stdio.h>
#include <sched.h>
#include <stdlib.h>
#include <string.h>
#include <signal.h>
#include <unistd.h>
#include <inttypes.h>
#include <errno.h>
#include <sys/types.h>
#include <sys/mman.h>
#include <sys/syscall.h>
#include <linux/perf_event.h>

#define TIMINGS 5

static struct perf_event_attr attr;
static int fdperf = -1;
static struct perf_event_mmap_page *buf;
static cpu_set_t cores;
static long long core;
static int64_t where[TIMINGS];
static int64_t offset[TIMINGS];
static int64_t result[TIMINGS];
static long long lastwhere;
static long long lastoffset;

void printresults(void)
{
  printf("core %lld",core);
  for (int i = 0;i < TIMINGS;++i) {
    if (where[i] != lastwhere) {
      lastwhere = where[i];
      printf(" [index 0x%llx]",lastwhere);
    }
    if (offset[i] != lastoffset) {
      lastoffset = offset[i];
      printf(" [offset 0x%llx]",lastoffset);
    }
    printf(" %lld",(long long) result[i]);
  }
  if (result[TIMINGS-1] == result[0]) printf(" stuck (perf bug 1)");
  printf("\n");
  fflush(stdout);
}

void sigsegv()
{
  printf("core %lld segmentation fault (perf bug 2)\n",core);
  exit(100);
}

int main()
{
  signal(SIGSEGV,sigsegv);
  attr.type = PERF_TYPE_HARDWARE;
  attr.size = sizeof(struct perf_event_attr);
  attr.config = PERF_COUNT_HW_CPU_CYCLES;
  attr.exclude_kernel = 1;
  attr.exclude_hv = 1;
  fdperf = syscall(__NR_perf_event_open,&attr,0,-1,-1,0);
  if (fdperf == -1) abort();
  for (int rdpmc = 0;rdpmc < 3;++rdpmc) {
    if (rdpmc) {
      buf = mmap(NULL,sysconf(_SC_PAGESIZE),PROT_READ,MAP_SHARED,fdperf,0);
      if (!buf) {
        printf("mmap failed for rdpmc; quitting\n");
        return 111;
      }
      if (rdpmc == 1)
        printf("will now use rdpmc in an intel-specific way (fails on amd)\n");
      else
        printf("will now use rdpmc according to the perf manual\n");
      fflush(stdout);
    }
    for (int loop = 0;loop < 3;++loop) {
      for (core = 0;;++core) {
        CPU_ZERO(&cores);
        CPU_SET(core,&cores);
        if (sched_setaffinity(0,sizeof cores,&cores) != 0) break;
        for (int i = 0;i < TIMINGS;++i) {
          if (rdpmc == 1) {
            asm volatile("rdpmc;shlq $32,%%rdx;orq %%rdx,%%rax"
              : "=a"(result[i]) : "c"((1<<30)|1) : "%rdx");
          } else if (rdpmc) {
            unsigned int seq;
            do {
              seq = buf->lock;
              asm volatile("" ::: "memory");
              where[i] = buf->index;
              offset[i] = buf->offset;
              asm volatile("rdpmc;shlq $32,%%rdx;orq %%rdx,%%rax"
                : "=a"(result[i]) : "c"(where[i]-1) : "%rdx");
              asm volatile("" ::: "memory");
            } while (buf->lock != seq);
            result[i] += offset[i];
            result[i] &= 0xffffffffffff;
          } else {
            int r = read(fdperf,&result[i],8);
            if (r < 8) result[i] = r;
            if (r == -1) result[i] = -errno;
          }
        }
        printresults();
      }
    }
  }
  return 0;
}