/* The main CPU accumulator thread */
void guppi_accum_thread(void *_args) {
float **accumulator; //indexed accumulator[accum_id][chan][subband][stokes]
char accum_dirty[NUM_SW_STATES];
struct sdfits_data_columns data_cols[NUM_SW_STATES];
int payload_type;
int i, j, k, rv;
/* Get arguments */
struct guppi_thread_args *args = (struct guppi_thread_args *)_args;
/* Set cpu affinity */
cpu_set_t cpuset, cpuset_orig;
sched_getaffinity(0, sizeof(cpu_set_t), &cpuset_orig);
//CPU_ZERO(&cpuset);
CPU_CLR(13, &cpuset);
CPU_SET(9, &cpuset);
rv = sched_setaffinity(0, sizeof(cpu_set_t), &cpuset);
if (rv<0) {
guppi_error("guppi_accum_thread", "Error setting cpu affinity.");
perror("sched_setaffinity");
}
/* Set priority */
rv = setpriority(PRIO_PROCESS, 0, args->priority);
if (rv<0) {
guppi_error("guppi_accum_thread", "Error setting priority level.");
perror("set_priority");
}
/* Attach to status shared mem area */
struct guppi_status st;
rv = guppi_status_attach(&st);
if (rv!=GUPPI_OK) {
guppi_error("guppi_accum_thread",
"Error attaching to status shared memory.");
pthread_exit(NULL);
}
pthread_cleanup_push((void *)guppi_status_detach, &st);
pthread_cleanup_push((void *)set_exit_status, &st);
pthread_cleanup_push((void *)guppi_thread_set_finished, args);
/* Init status */
guppi_status_lock_safe(&st);
hputs(st.buf, STATUS_KEY, "init");
guppi_status_unlock_safe(&st);
/* Read in general parameters */
struct guppi_params gp;
struct sdfits sf;
pthread_cleanup_push((void *)guppi_free_sdfits, &sf);
/* Attach to databuf shared mem */
struct guppi_databuf *db_in, *db_out;
db_in = guppi_databuf_attach(args->input_buffer);
char errmsg[256];
if (db_in==NULL) {
sprintf(errmsg,
"Error attaching to input databuf(%d) shared memory.",
args->input_buffer);
guppi_error("guppi_accum_thread", errmsg);
pthread_exit(NULL);
}
pthread_cleanup_push((void *)guppi_databuf_detach, db_in);
db_out = guppi_databuf_attach(args->output_buffer);
if (db_out==NULL) {
sprintf(errmsg,
"Error attaching to output databuf(%d) shared memory.",
args->output_buffer);
guppi_error("guppi_accum_thread", errmsg);
pthread_exit(NULL);
}
pthread_cleanup_push((void *)guppi_databuf_detach, db_out);
/* Determine high/low bandwidth mode */
char bw_mode[16];
if (hgets(st.buf, "BW_MODE", 16, bw_mode))
{
if(strncmp(bw_mode, "high", 4) == 0)
payload_type = INT_PAYLOAD;
else if(strncmp(bw_mode, "low", 3) == 0)
payload_type = FLOAT_PAYLOAD;
else
guppi_error("guppi_net_thread", "Unsupported bandwidth mode");
}
else
guppi_error("guppi_net_thread", "BW_MODE not set");
/* Read nchan and nsubband from status shared memory */
guppi_read_obs_params(st.buf, &gp, &sf);
/* Allocate memory for vector accumulators */
create_accumulators(&accumulator, sf.hdr.nchan, sf.hdr.nsubband);
pthread_cleanup_push((void *)destroy_accumulators, accumulator);
/* Clear the vector accumulators */
for(i = 0; i < NUM_SW_STATES; i++) accum_dirty[i] = 1;
reset_accumulators(accumulator, data_cols, accum_dirty, sf.hdr.nsubband, sf.hdr.nchan);
/* Loop */
int curblock_in=0, curblock_out=0;
int first=1;
float reqd_exposure=0;
double accum_time=0;
int integ_num;
float pfb_rate;
int heap, accumid, struct_offset, array_offset;
char *hdr_in=NULL, *hdr_out=NULL;
struct databuf_index *index_in, *index_out;
int nblock_int=0, npacket=0, n_pkt_drop=0, n_heap_drop=0;
signal(SIGINT,cc);
while (run) {
/* Note waiting status */
guppi_status_lock_safe(&st);
hputs(st.buf, STATUS_KEY, "waiting");
guppi_status_unlock_safe(&st);
/* Wait for buf to have data */
rv = guppi_databuf_wait_filled(db_in, curblock_in);
if (rv!=0) continue;
/* Note waiting status and current block*/
guppi_status_lock_safe(&st);
hputs(st.buf, STATUS_KEY, "accumulating");
hputi4(st.buf, "ACCBLKIN", curblock_in);
guppi_status_unlock_safe(&st);
/* Read param struct for this block */
hdr_in = guppi_databuf_header(db_in, curblock_in);
if (first)
guppi_read_obs_params(hdr_in, &gp, &sf);
else
guppi_read_subint_params(hdr_in, &gp, &sf);
/* Do any first time stuff: first time code runs, not first time process this block */
if (first) {
/* Set up first output header. This header is copied from block to block
each time a new block is created */
hdr_out = guppi_databuf_header(db_out, curblock_out);
memcpy(hdr_out, guppi_databuf_header(db_in, curblock_in),
GUPPI_STATUS_SIZE);
/* Read required exposure and PFB rate from status shared memory */
reqd_exposure = sf.data_columns.exposure;
pfb_rate = sf.hdr.efsampfr / (2 * sf.hdr.nchan);
/* Initialise the index in the output block */
index_out = (struct databuf_index*)guppi_databuf_index(db_out, curblock_out);
index_out->num_datasets = 0;
index_out->array_size = sf.hdr.nsubband * sf.hdr.nchan * NUM_STOKES * 4;
first=0;
}
/* Loop through each spectrum (heap) in input buffer */
index_in = (struct databuf_index*)guppi_databuf_index(db_in, curblock_in);
for(heap = 0; heap < index_in->num_heaps; heap++)
{
/* If invalid, record it and move to next heap */
if(!index_in->cpu_gpu_buf[heap].heap_valid)
{
n_heap_drop++;
continue;
}
/* Read in heap from buffer */
char* heap_addr = (char*)(guppi_databuf_data(db_in, curblock_in) +
sizeof(struct freq_spead_heap) * heap);
struct freq_spead_heap* freq_heap = (struct freq_spead_heap*)(heap_addr);
char* payload_addr = (char*)(guppi_databuf_data(db_in, curblock_in) +
sizeof(struct freq_spead_heap) * MAX_HEAPS_PER_BLK +
(index_in->heap_size - sizeof(struct freq_spead_heap)) * heap );
int *i_payload = (int*)(payload_addr);
float *f_payload = (float*)(payload_addr);
accumid = freq_heap->status_bits & 0x7;
/*Debug: print heap */
/* printf("%d, %d, %d, %d, %d, %d\n", freq_heap->time_cntr, freq_heap->spectrum_cntr,
freq_heap->integ_size, freq_heap->mode, freq_heap->status_bits,
freq_heap->payload_data_off);
*/
/* If we have accumulated for long enough, write vectors to output block */
if(accum_time >= reqd_exposure)
{
for(i = 0; i < NUM_SW_STATES; i++)
{
/*If a particular accumulator is dirty, write it to output buffer */
if(accum_dirty[i])
{
/*If insufficient space, first mark block as filled and request new block*/
index_out = (struct databuf_index*)(guppi_databuf_index(db_out, curblock_out));
if( (index_out->num_datasets+1) *
(index_out->array_size + sizeof(struct sdfits_data_columns)) >
db_out->block_size)
{
printf("Accumulator finished with output block %d\n", curblock_out);
/* Write block number to status buffer */
guppi_status_lock_safe(&st);
hputi4(st.buf, "ACCBLKOU", curblock_out);
guppi_status_unlock_safe(&st);
/* Update packet count and loss fields in output header */
hputi4(hdr_out, "NBLOCK", nblock_int);
hputi4(hdr_out, "NPKT", npacket);
hputi4(hdr_out, "NPKTDROP", n_pkt_drop);
hputi4(hdr_out, "NHPDROP", n_heap_drop);
/* Close out current integration */
guppi_databuf_set_filled(db_out, curblock_out);
/* Wait for next output buf */
curblock_out = (curblock_out + 1) % db_out->n_block;
guppi_databuf_wait_free(db_out, curblock_out);
while ((rv=guppi_databuf_wait_free(db_out, curblock_out)) != GUPPI_OK)
{
if (rv==GUPPI_TIMEOUT) {
guppi_warn("guppi_accum_thread", "timeout while waiting for output block");
continue;
} else {
guppi_error("guppi_accum_thread", "error waiting for free databuf");
run=0;
pthread_exit(NULL);
break;
}
}
hdr_out = guppi_databuf_header(db_out, curblock_out);
memcpy(hdr_out, guppi_databuf_header(db_in, curblock_in),
GUPPI_STATUS_SIZE);
/* Initialise the index in new output block */
index_out = (struct databuf_index*)guppi_databuf_index(db_out, curblock_out);
index_out->num_datasets = 0;
index_out->array_size = sf.hdr.nsubband * sf.hdr.nchan * NUM_STOKES * 4;
nblock_int=0;
npacket=0;
n_pkt_drop=0;
n_heap_drop=0;
}
/*Update index for output buffer*/
index_out = (struct databuf_index*)(guppi_databuf_index(db_out, curblock_out));
if(index_out->num_datasets == 0)
struct_offset = 0;
else
struct_offset = index_out->disk_buf[index_out->num_datasets-1].array_offset +
index_out->array_size;
array_offset = struct_offset + sizeof(struct sdfits_data_columns);
index_out->disk_buf[index_out->num_datasets].struct_offset = struct_offset;
index_out->disk_buf[index_out->num_datasets].array_offset = array_offset;
/*Copy sdfits_data_columns struct to disk buffer */
memcpy(guppi_databuf_data(db_out, curblock_out) + struct_offset,
&data_cols[i], sizeof(struct sdfits_data_columns));
/*Copy data array to disk buffer */
memcpy(guppi_databuf_data(db_out, curblock_out) + array_offset,
accumulator[i], index_out->array_size);
/*Update SDFITS data_columns pointer to data array */
((struct sdfits_data_columns*)
(guppi_databuf_data(db_out, curblock_out) + struct_offset))->data =
(unsigned char*)(guppi_databuf_data(db_out, curblock_out) + array_offset);
index_out->num_datasets = index_out->num_datasets + 1;
}
}
accum_time = 0;
integ_num += 1;
reset_accumulators(accumulator, data_cols, accum_dirty,
sf.hdr.nsubband, sf.hdr.nchan);
}
/* Only add spectrum to accumulator if blanking bit is low */
if((freq_heap->status_bits & 0x08) == 0)
{
/* Fill in data columns header fields */
if(!accum_dirty[accumid])
{
/*Record SPEAD header fields*/
data_cols[accumid].time = index_in->cpu_gpu_buf[heap].heap_rcvd_mjd;
data_cols[accumid].time_counter = freq_heap->time_cntr;
data_cols[accumid].integ_num = integ_num;
data_cols[accumid].sttspec = freq_heap->spectrum_cntr;
data_cols[accumid].accumid = accumid;
/* Fill in rest of fields from status buffer */
strcpy(data_cols[accumid].object, sf.data_columns.object);
data_cols[accumid].azimuth = sf.data_columns.azimuth;
data_cols[accumid].elevation = sf.data_columns.elevation;
data_cols[accumid].bmaj = sf.data_columns.bmaj;
data_cols[accumid].bmin = sf.data_columns.bmin;
data_cols[accumid].bpa = sf.data_columns.bpa;
data_cols[accumid].centre_freq_idx = sf.data_columns.centre_freq_idx;
data_cols[accumid].ra = sf.data_columns.ra;
data_cols[accumid].dec = sf.data_columns.dec;
data_cols[accumid].exposure = 0.0;
for(i = 0; i < NUM_SW_STATES; i++)
data_cols[accumid].centre_freq[i] = sf.data_columns.centre_freq[i];
accum_dirty[accumid] = 1;
}
data_cols[accumid].exposure += (float)(freq_heap->integ_size)/pfb_rate;
data_cols[accumid].stpspec = freq_heap->spectrum_cntr;
/* Add spectrum to appropriate vector accumulator (high-bw mode) */
if(payload_type == INT_PAYLOAD)
{
for(i = 0; i < sf.hdr.nchan; i++)
{
for(j = 0; j < sf.hdr.nsubband; j++)
{
for(k = 0; k < NUM_STOKES; k++)
{
accumulator[accumid]
[i*sf.hdr.nsubband*NUM_STOKES + j*NUM_STOKES + k] +=
(float)i_payload[i*sf.hdr.nsubband*NUM_STOKES + j*NUM_STOKES + k];
}
}
}
}
/* Add spectrum to appropriate vector accumulator (low-bw mode) */
else
{
for(i = 0; i < sf.hdr.nchan; i++)
{
for(j = 0; j < sf.hdr.nsubband; j++)
{
for(k = 0; k < NUM_STOKES; k++)
{
accumulator[accumid]
[i*sf.hdr.nsubband*NUM_STOKES + j*NUM_STOKES + k] +=
f_payload[i*sf.hdr.nsubband*NUM_STOKES + j*NUM_STOKES + k];
}
}
}
}
}
accum_time += (double)freq_heap->integ_size / pfb_rate;
}
/* Update packet count and loss fields from input header */
nblock_int++;
npacket += gp.num_pkts_rcvd;
n_pkt_drop += gp.num_pkts_dropped;
/* Done with current input block */
guppi_databuf_set_free(db_in, curblock_in);
curblock_in = (curblock_in + 1) % db_in->n_block;
/* Check for cancel */
pthread_testcancel();
}
pthread_exit(NULL);
pthread_cleanup_pop(0); /* Closes set_exit_status */
pthread_cleanup_pop(0); /* Closes set_finished */
pthread_cleanup_pop(0); /* Closes guppi_free_sdfits */
pthread_cleanup_pop(0); /* Closes ? */
pthread_cleanup_pop(0); /* Closes destroy_accumulators */
pthread_cleanup_pop(0); /* Closes guppi_status_detach */
pthread_cleanup_pop(0); /* Closes guppi_databuf_detach */
}
static void *run(void * _args)
{
// Cast _args
struct guppi_thread_args *args = (struct guppi_thread_args *)_args;
#ifdef DEBUG_SEMS
fprintf(stderr, "s/tid %lu/NET/' <<.\n", pthread_self());
#endif
THREAD_RUN_BEGIN(args);
THREAD_RUN_SET_AFFINITY_PRIORITY(args);
THREAD_RUN_ATTACH_STATUS(args->instance_id, st);
st_p = &st; // allow global (this source file) access to the status buffer
/* Attach to paper_input_databuf */
THREAD_RUN_ATTACH_DATABUF(args->instance_id, paper_input_databuf, db, args->output_buffer);
/* Read in general parameters */
struct guppi_params gp;
struct sdfits pf;
char status_buf[GUPPI_STATUS_SIZE];
guppi_status_lock_busywait_safe(st_p);
memcpy(status_buf, st_p->buf, GUPPI_STATUS_SIZE);
guppi_status_unlock_safe(st_p);
guppi_read_obs_params(status_buf, &gp, &pf);
pthread_cleanup_push((void *)guppi_free_sdfits, &pf);
/* Read network params */
struct guppi_udp_params up;
//guppi_read_net_params(status_buf, &up);
paper_read_net_params(status_buf, &up);
// Store bind host/port info etc in status buffer
guppi_status_lock_busywait_safe(&st);
hputs(st.buf, "BINDHOST", up.bindhost);
hputi4(st.buf, "BINDPORT", up.bindport);
hputu4(st.buf, "MISSEDFE", 0);
hputu4(st.buf, "MISSEDPK", 0);
hputs(st.buf, STATUS_KEY, "running");
guppi_status_unlock_safe(&st);
struct guppi_udp_packet p;
/* Give all the threads a chance to start before opening network socket */
sleep(1);
#ifndef TIMING_TEST
/* Set up UDP socket */
int rv = guppi_udp_init(&up);
if (rv!=GUPPI_OK) {
guppi_error("guppi_net_thread",
"Error opening UDP socket.");
pthread_exit(NULL);
}
pthread_cleanup_push((void *)guppi_udp_close, &up);
#endif
/* Main loop */
uint64_t packet_count = 0;
uint64_t elapsed_wait_ns = 0;
uint64_t elapsed_recv_ns = 0;
uint64_t elapsed_proc_ns = 0;
float ns_per_wait = 0.0;
float ns_per_recv = 0.0;
float ns_per_proc = 0.0;
struct timespec start, stop;
struct timespec recv_start, recv_stop;
signal(SIGINT,cc);
while (run_threads) {
#ifndef TIMING_TEST
/* Read packet */
clock_gettime(CLOCK_MONOTONIC, &recv_start);
do {
clock_gettime(CLOCK_MONOTONIC, &start);
p.packet_size = recv(up.sock, p.data, GUPPI_MAX_PACKET_SIZE, 0);
clock_gettime(CLOCK_MONOTONIC, &recv_stop);
} while (p.packet_size == -1 && (errno == EAGAIN || errno == EWOULDBLOCK) && run_threads);
if(!run_threads) break;
if (up.packet_size != p.packet_size) {
if (p.packet_size != -1) {
#ifdef DEBUG_NET
guppi_warn("guppi_net_thread", "Incorrect pkt size");
#endif
continue;
} else {
guppi_error("guppi_net_thread",
"guppi_udp_recv returned error");
perror("guppi_udp_recv");
pthread_exit(NULL);
}
}
#endif
packet_count++;
// Copy packet into any blocks where it belongs.
const uint64_t mcnt = write_paper_packet_to_blocks((paper_input_databuf_t *)db, &p);
clock_gettime(CLOCK_MONOTONIC, &stop);
elapsed_wait_ns += ELAPSED_NS(recv_start, start);
elapsed_recv_ns += ELAPSED_NS(start, recv_stop);
elapsed_proc_ns += ELAPSED_NS(recv_stop, stop);
if(mcnt != -1) {
// Update status
ns_per_wait = (float)elapsed_wait_ns / packet_count;
ns_per_recv = (float)elapsed_recv_ns / packet_count;
ns_per_proc = (float)elapsed_proc_ns / packet_count;
guppi_status_lock_busywait_safe(&st);
hputu8(st.buf, "NETMCNT", mcnt);
// Gbps = bits_per_packet / ns_per_packet
// (N_BYTES_PER_PACKET excludes header, so +8 for the header)
hputr4(st.buf, "NETGBPS", 8*(N_BYTES_PER_PACKET+8)/(ns_per_recv+ns_per_proc));
hputr4(st.buf, "NETWATNS", ns_per_wait);
hputr4(st.buf, "NETRECNS", ns_per_recv);
hputr4(st.buf, "NETPRCNS", ns_per_proc);
guppi_status_unlock_safe(&st);
// Start new average
elapsed_wait_ns = 0;
elapsed_recv_ns = 0;
elapsed_proc_ns = 0;
packet_count = 0;
}
#if defined TIMING_TEST || defined NET_TIMING_TEST
#define END_LOOP_COUNT (1*1000*1000)
static int loop_count=0;
static struct timespec tt_start, tt_stop;
if(loop_count == 0) {
clock_gettime(CLOCK_MONOTONIC, &tt_start);
}
//if(loop_count == 1000000) run_threads = 0;
if(loop_count == END_LOOP_COUNT) {
clock_gettime(CLOCK_MONOTONIC, &tt_stop);
int64_t elapsed = ELAPSED_NS(tt_start, tt_stop);
printf("processed %d packets in %.6f ms (%.3f us per packet)\n",
END_LOOP_COUNT, elapsed/1e6, elapsed/1e3/END_LOOP_COUNT);
exit(0);
}
loop_count++;
#endif
/* Will exit if thread has been cancelled */
pthread_testcancel();
}
/* Have to close all push's */
#ifndef TIMING_TEST
pthread_cleanup_pop(0); /* Closes push(guppi_udp_close) */
#endif
pthread_cleanup_pop(0); /* Closes guppi_free_psrfits */
THREAD_RUN_DETACH_DATAUF;
THREAD_RUN_DETACH_STATUS;
THREAD_RUN_END;
return NULL;
}
