/** @class PAPI_stop_counters
* @brief Stop counting hardware events and reset values to zero.
*
* @par C Interface:
* \#include <papi.h> @n
* int PAPI_stop_counters( long long *values, int array_len );
*
* @param *values
* an array where to put the counter values
* @param array_len
* the number of items in the *values array
*
* @post
* After this function is called, the values are reset to zero.
*
* @retval PAPI_EINVAL
* One or more of the arguments is invalid.
* @retval PAPI_ENOTRUN
* The EventSet is not started yet.
* @retval PAPI_ENOEVST
* The EventSet has not been added yet.
*
* The PAPI_stop_counters() function stops the counters and copies the counts
* into the *values array.
* The counters must have been started by a previous call to PAPI_start_counters().
*
* \code
int Events[2] = { PAPI_TOT_CYC, PAPI_TOT_INS };
long long values[2];
if ( PAPI_start_counters( Events, 2 ) != PAPI_OK )
handle_error(1);
your_slow_code();
if ( PAPI_stop_counters( values, 2 ) != PAPI_OK )
handle_error(1);
* \endcode
*
* @see PAPI_read_counters() PAPI_start_counters() PAPI_set_opt()
*/
int
PAPI_stop_counters( long long *values, int array_len )
{
int retval;
HighLevelInfo *state = NULL;
if ( ( retval = _internal_check_state( &state ) ) != PAPI_OK )
return ( retval );
if ( state->running == 0 )
return ( PAPI_ENOTRUN );
if ( state->running == HL_START ) {
if ( array_len < state->num_evts || values == NULL) {
return ( PAPI_EINVAL );
} else {
retval = PAPI_stop( state->EventSet, values );
}
}
if ( state->running > HL_START ) {
long long tmp_values[3];
retval = PAPI_stop( state->EventSet, tmp_values );
}
if ( retval == PAPI_OK ) {
_internal_cleanup_hl_info( state );
PAPI_cleanup_eventset( state->EventSet );
}
APIDBG( "PAPI_stop_counters returns %d\n", retval );
return retval;
}
int PAPI_stop_counters(long long * values, int array_len)
{
int retval;
HighLevelInfo *state = NULL;
if ((retval = _internal_check_state(&state)) != PAPI_OK)
return (retval);
if (state->running == 0)
return (PAPI_ENOTRUN);
if (state->running == HL_FLOPS || state->running == HL_FLIPS || state->running == HL_IPC) {
long long tmp_values[2];
retval = PAPI_stop(state->EventSet, tmp_values);
}
else if(state->running != HL_START_COUNTERS || array_len < state->num_evts)
return (PAPI_EINVAL);
else
retval = PAPI_stop(state->EventSet, values);
if (retval==PAPI_OK) {
_internal_cleanup_hl_info(state);
PAPI_cleanup_eventset(state->EventSet);
}
APIDBG("PAPI_stop_counters returns %d\n", retval);
return retval;
}
int PAPI_start_counters(int *events, int array_len)
{
int i, retval;
HighLevelInfo *state = NULL;
if ((retval = _internal_check_state(&state)) != PAPI_OK)
return (retval);
if(state->running != 0)
return(PAPI_EINVAL);
/* load events to the new EventSet */
for (i = 0; i < array_len; i++) {
retval = PAPI_add_event(state->EventSet, events[i]);
if (retval == PAPI_EISRUN)
return (retval);
if (retval) {
/* remove any prior events that may have been added
* and cleanup the high level information
*/
_internal_cleanup_hl_info(state);
PAPI_cleanup_eventset(state->EventSet);
return (retval);
}
}
/* start the EventSet */
if ((retval = _internal_start_hl_counters(state)) == PAPI_OK) {
state->running = HL_START_COUNTERS;
state->num_evts = array_len;
}
return (retval);
}
int PAPI_ipc(float *rtime, float *ptime, long long * ins, float *ipc)
{
HighLevelInfo *state = NULL;
int retval;
if ((retval = _internal_check_state(&state)) != PAPI_OK)
return (retval);
return(_hl_rate_calls(rtime,ptime,ins,ipc,PAPI_TOT_INS,state));
}
/*
* How many hardware counters does this platform support?
*/
int PAPI_num_counters(void)
{
int retval;
HighLevelInfo *tmp = NULL;
/* Make sure the Library is initialized, etc... */
if ((retval = _internal_check_state(&tmp)) != PAPI_OK)
return (retval);
return (PAPI_get_opt(PAPI_MAX_HWCTRS, NULL));
}
int PAPI_flops(float *rtime, float *ptime, long long * flpops, float *mflops)
{
HighLevelInfo *state = NULL;
int retval;
if ((retval = _internal_check_state(&state)) != PAPI_OK)
return (retval);
if ((retval =
_hl_rate_calls(rtime, ptime, flpops, mflops, PAPI_FP_OPS, state)) != PAPI_OK)
return (retval);
return (PAPI_OK);
}
/** @class PAPI_ipc
* @brief Get instructions per cycle, real and processor time.
*
* @par C Interface:
* \#include <papi.h> @n
* int PAPI_ipc( float *rtime, float *ptime, long long *ins, float *ipc );
*
* @param *rtime
* total realtime since the first PAPI_flops() call
* @param *ptime
* total process time since the first PAPI_flops() call
* @param *ins
* total instructions since the first call
* @param *ipc
* instructions per cycle achieved since the previous call
*
* @retval PAPI_EINVAL
* The counters were already started by something other than: PAPI_ipc()
* @retval PAPI_ENOEVNT
* The total instructions or total cycles event does not exist.
* @retval PAPI_ENOMEM
* Insufficient memory to complete the operation.
*
* The first call to PAPI_ipc() will initialize the PAPI High Level interface,
* set up the counters to monitor PAPI_TOT_INS and PAPI_TOT_CYC events
* and start the counters.
* Subsequent calls will read the counters and return total real time,
* total process time, total instructions since the start of the measurement
* and the instructions per cycle rate since latest call to PAPI_ipc().
* A call to PAPI_stop_counters() will stop the counters from running and then
* calls such as PAPI_start_counters() can safely be used.
*
* @see PAPI_flops() PAPI_stop_counters() PAPI_set_opt() PAPI_flips()
*/
int
PAPI_ipc( float *rtime, float *ptime, long long *ins, float *ipc )
{
if ( rtime == NULL || ptime == NULL || ins == NULL || ipc == NULL )
return PAPI_EINVAL;
HighLevelInfo *state = NULL;
int retval;
if ( ( retval = _internal_check_state( &state ) ) != PAPI_OK )
return ( retval );
return _hl_rate_calls( rtime, ptime, ins, ipc,
( unsigned int ) PAPI_TOT_INS, state );
}
/** @class PAPI_flops
* @brief Simplified call to get Mflops/s (floating point instruction rate), real and processor time.
*
* @par C Interface:
* \#include <papi.h> @n
* int PAPI_flops( float *rtime, float *ptime, long long *flpops, float *mflops );
*
* @param *rtime
* total realtime since the first PAPI_flops() call
* @param *ptime
* total process time since the first PAPI_flops() call
* @param *flpins
* total floating point instructions since the first call
* @param *rtime
* total realtime since the first PAPI_flops() call
* @param *ptime
* total process time since the first PAPI_flops() call
* @param *flpops
* total floating point instructions since the first call
*
* @retval PAPI_EINVAL
* The counters were already started by something other than: PAPI_flips() or PAPI_flops().
* @retval PAPI_ENOEVNT
* The floating point operations, floating point instructions or total cycles
* event does not exist.
* @retval PAPI_ENOMEM
* Insufficient memory to complete the operation.
*
* The first call to PAPI_flops() will initialize the PAPI High Level interface,
* set up the counters to monitor PAPI_FP_OPS and PAPI_TOT_CYC events
* and start the counters.
* Subsequent calls will read the counters and return total real time,
* total process time, total floating point instructions since the start of the
* measurement and the Mflop/s rate since latest call to PAPI_flops().
* A call to PAPI_stop_counters() will stop the counters from running and then
* calls such as PAPI_start_counters() can safely be used.
*
* @internal
* The next three calls all use _hl_rate_calls() to return an instruction rate value.
* PAPI_flops returns information related to floating point instructions using
* the PAPI_FP_INS event. This is intended to measure instruction rate through the
* floating point pipe with no massaging.
* PAPI_flops return information related to theoretical floating point operations
* rather than simple instructions. It uses the PAPI_FP_OPS event which attempts to
* 'correctly' account for, e.g., FMA undercounts and FP Store overcounts, etc.
*
* @see PAPI_stop_counters() PAPI_ipc() PAPI_set_opt()
*/
int
PAPI_flops( float *rtime, float *ptime, long long *flpops, float *mflops )
{
if ( rtime == NULL || ptime == NULL || flpops == NULL || mflops == NULL )
return PAPI_EINVAL;
HighLevelInfo *state = NULL;
int retval;
if ( ( retval = _internal_check_state( &state ) ) != PAPI_OK )
return ( retval );
if ( ( retval =
_hl_rate_calls( rtime, ptime, flpops, mflops,
( unsigned int ) PAPI_FP_OPS, state ) ) != PAPI_OK )
return ( retval );
return ( PAPI_OK );
}
int _internal_hl_read_cnts(long long * values, int array_len, int flag)
{
int retval;
HighLevelInfo *state = NULL;
if ((retval = _internal_check_state(&state)) != PAPI_OK)
return (retval);
if (state->running != HL_START_COUNTERS || array_len < state->num_evts)
return (PAPI_EINVAL);
if (flag == PAPI_HL_ACCUM)
return (PAPI_accum(state->EventSet, values));
else if (flag == PAPI_HL_READ) {
if ((retval = PAPI_read(state->EventSet, values)) != PAPI_OK)
return (retval);
return (PAPI_reset(state->EventSet));
}
/* Invalid flag passed in */
return (PAPI_EINVAL);
}
int
_hl_rate_calls( float *real_time, float *proc_time, int *events,
long long *values, long long *ins, float *rate, int mode )
{
long long rt, pt; // current elapsed real and process times in usec
int num_events = 2;
int retval = 0;
HighLevelInfo *state = NULL;
if ( ( retval = _internal_check_state( &state ) ) != PAPI_OK ) {
return ( retval );
}
if ( state->running != HL_STOP && state->running != mode ) {
return PAPI_EINVAL;
}
if ( state->running == HL_STOP ) {
switch (mode) {
case HL_FLOP:
case HL_FLIP:
num_events = 1;
break;
case HL_IPC:
break;
case HL_EPC:
if ( events[2] != 0 ) num_events = 3;
break;
default:
return PAPI_EINVAL;
}
if (( retval = PAPI_add_events( state->EventSet, events, num_events )) != PAPI_OK ) {
_internal_cleanup_hl_info( state );
PAPI_cleanup_eventset( state->EventSet );
return retval;
}
state->total_ins = 0;
state->initial_real_time = state->last_real_time = PAPI_get_real_usec( );
state->initial_proc_time = state->last_proc_time = PAPI_get_virt_usec( );
if ( ( retval = PAPI_start( state->EventSet ) ) != PAPI_OK ) {
return retval;
}
/* Initialize the interface */
state->running = mode;
*real_time = 0.0;
*proc_time = 0.0;
*rate = 0.0;
} else {
if ( ( retval = PAPI_stop( state->EventSet, values ) ) != PAPI_OK ) {
state->running = HL_STOP;
return retval;
}
/* Read elapsed real and process times */
rt = PAPI_get_real_usec();
pt = PAPI_get_virt_usec();
/* Convert to seconds with multiplication because it is much faster */
*real_time = ((float)( rt - state->initial_real_time )) * .000001;
*proc_time = ((float)( pt - state->initial_proc_time )) * .000001;
state->total_ins += values[0];
switch (mode) {
case HL_FLOP:
case HL_FLIP:
/* Calculate MFLOP and MFLIP rates */
if ( pt > 0 ) {
*rate = (float)values[0] / (pt - state->last_proc_time);
} else *rate = 0;
break;
case HL_IPC:
case HL_EPC:
/* Calculate IPC */
if (values[1]!=0) {
*rate = (float) ((float)values[0] / (float) ( values[1]));
}
break;
default:
return PAPI_EINVAL;
}
state->last_real_time = rt;
state->last_proc_time = pt;
if ( ( retval = PAPI_start( state->EventSet ) ) != PAPI_OK ) {
state->running = HL_STOP;
return retval;
}
}
*ins = state->total_ins;
return PAPI_OK;
}
