#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <math.h>
#include "papi.h"

#define error(func, ret) \
	{ fprintf(stderr, "%s failed.\n", func); if (ret) exit(ret);}

// static PAPI_hw_info_t *info;
static int events;
static long long counters[128];
static char *events_desc[128];

static void
cleanup(void)
{
    if (PAPI_is_initialized()) PAPI_shutdown();
}

static void
add_event(int code, const char *name, const char *msg)
{
    static int nevents = 0;

    if (PAPI_query_event(code) == PAPI_OK) {
	if (PAPI_add_event(events, code) != PAPI_OK) {
	    fprintf(stderr, "PAPI_add_event failed for %s (%s).\n", name, msg);
	} else {
	    fprintf(stderr, "Successfully added PAPI_FP_OPS.\n");
	    events_desc[nevents] = malloc(strlen(msg) + strlen(name) + 5);
	    sprintf(events_desc[nevents], "%s (%s):", msg, name);
	    nevents++;
	}
    } else {
	fprintf(stderr, "No event counter for %s.\n", msg);
    }
}

int
main(int argc, char **argv)
{
    int i, n;
    PAPI_option_t options;

    if (PAPI_library_init(PAPI_VER_CURRENT) != PAPI_VER_CURRENT)
	error("PAPI_library_init", 1);

    atexit(cleanup);

//    info = PAPI_get_hardware_info();
//    if (info == NULL)
//	error("PAPI_get_hardware_info", 1);

    if (PAPI_create_eventset(&events) != PAPI_OK)
	error("PAPI_create_eventset", 1);
#if 0	/* seems multiplexing does not work on PIII */
    if (PAPI_multiplex_init() != PAPI_OK)
	error("PAPI_multiplex_init", 1);
    if (PAPI_set_multiplex(events) != PAPI_OK)
	error("PAPI_set_multiplex", 1);
#endif
    options.domain.eventset = events;
    options.domain.domain = PAPI_DOM_ALL;
    PAPI_set_opt(PAPI_DOMAIN, &options);

    //add_event(PAPI_FP_OPS,  "PAPI_FP_OPS",  "floating point operations");
    //add_event(PAPI_FP_INS,  "PAPI_FP_INS",  "floating point instructions");

    add_event(PAPI_L1_TCM,  "PAPI_L1_TCM",  "L1 total cache misses");
    add_event(PAPI_L2_TCM,  "PAPI_L2_TCM",  "L2 total cache misses");

    //add_event(PAPI_L1_TCM,  "PAPI_L1_TCH",  "L1 total cache hits");
    //add_event(PAPI_L2_TCM,  "PAPI_L2_TCH",  "L2 total cache hits");

    /* Translation Lookaside Buffer */
    //add_event(PAPI_TLB_TL,  "PAPI_TLB_TL",  "TLB buffer misses");

    n = PAPI_num_events(events);
    fprintf(stderr, "%d counters initialized.\n", n);

    {
	double a[10000], b[sizeof(a)/sizeof(*a)];
	PAPI_start(events);
	PAPI_reset(events);
	for (i = 0; i < sizeof(a)/sizeof(*a); i++) a[i] = b[i] * 3.0;
	PAPI_stop(events, counters);
	for (i = 0; i < n; i++) {
	    fprintf(stderr, "%-40s %lld\n", events_desc[i], counters[i]);
	}
    }

    return 0;
}