sk_ptr_t
sk_event_module_init(sk_cycle_t *cycle)
{
sk_shm_t shm;
size_t size,cl;
u_char *shared;
int limit;
struct rlimit rlmt;
if(getrlimit(RLIMIT_NOFILE,&rlmt) == -1){
//ERR LOG
}
if(sk_accept_mutex_ptr){
return SK_OK;
}
cl = 128;
size = cl /*sk_accept_mutex*/
+cl /*sk_connection_counter*/
+cl;/*sk_temp_number*/
shm.size = size;
shm.name.len = sizeof("sk_share_zone");
shm.name.data = (u_char*)"sk_share_zone";
shm.log = cycle->log;
if(sk_shm_alloc(&shm) != SK_OK){
FATAL_ERR("malloc shmtx error");
return SK_ERR;
}
shared = shm.addr;
sk_accept_mutex_ptr = (sk_atomic_t *)shared;
sk_accept_mutex.spin = (sk_uptr_t) -1;
if(sk_shmtx_create(&sk_accept_mutex,shared,cycle->lock_file.data)!= SK_OK){
FATAL_ERR("malloc shmtx error");
return SK_ERR;
}
sk_mempool_mutex.spin = (sk_uptr_t) -1;
if(sk_shmtx_create(&sk_mempool_mutex,shared,cycle->lock_file.data)!= SK_OK){
FATAL_ERR("malloc shmtx error");
return SK_ERR;
}
sk_connection_counter = (sk_atomic_t *)(shared +1 * cl);
(void)sk_atomic_cmp_set(sk_connection_counter,0,1);
//sk_temp_number = (sk_atomic_t *)(shared+2 * cl);
cycle->connection_n = 1024;
return SK_OK;
}
/**
* @brief
*
* @param s
* @param dims
* @param size
*/
static void parse_dims(const char* s, size_t** dims, size_t* size) {
if(s == NULL){
return;
}
const char* seps = ":";
char sep = seps[0];
size_t len = strlen(s);
char* s_copy = malloc(sizeof(char*) * len + 1);
strcpy(s_copy, s);
// Count separators
(*size) = 0;
for (size_t i = 0; i < len; ++i) {
if (s_copy[i] == sep) {
++(*size);
}
}
++(*size);
*dims = (size_t*) malloc(*size * sizeof(*dims));
// Get values
char *pch = strtok(s_copy, seps);
char *end = NULL;
size_t i = 0;
while (pch != NULL) {
(*dims)[i] = (size_t) strtol(pch, &end, 10);
pch=strtok(NULL, seps);
++i;
}
free(s_copy);
// Sanity checks
if (i != *size) {
FATAL_ERR("Couldn't parse dimensions from %s. Correct format is t:x:y:z\n", s);
}
for (size_t i = 0; i < *size; ++i) {
if ((*dims)[i] <= 0) {
FATAL_ERR("Bad dimension specifications in %s (parsed value is <0)\n", s);
}
}
}
ssize_t cstp_send_file(worker_st *ws, const char *file)
{
FILE* fp;
char buf[512];
ssize_t len, total = 0;
int ret;
fp = fopen(file, "r");
if (fp == NULL)
return GNUTLS_E_FILE_ERROR;
while ( (len = fread( buf, 1, sizeof(buf), fp)) > 0) {
ret = cstp_send(ws, buf, len);
FATAL_ERR(ws, ret);
total += ret;
}
fclose(fp);
return total;
}
int main(int argc, char ** argv){
MPI_Init(&argc, &argv);
// Default values
struct args args;
args.procs.nn = 0;
args.procs.ppn = 0;
args.dgeom_size = 0;
args.dgeom = NULL;
args.bgeom_size = 0;
args.bgeom = NULL;
args.cgeom_size = 0;
args.cgeom = NULL;
args.testfn = NULL;
args.write_test = 0;
args.read_test = 0;
args.report_type = REPORT_HUMAN;
args.par_access = NC_INDEPENDENT;
args.is_unlimited = 0;
args.verify = 0;
args.fill_value = 0;
char * dg = NULL, * bg = NULL, *cg = NULL, *iot = "ind", *xf = "human";
option_help options [] = {
{'n' , "nn" , "Number of nodes" , OPTION_OPTIONAL_ARGUMENT , 'd' , & args.procs.nn} ,
{'p' , "ppn" , "Number of processes" , OPTION_OPTIONAL_ARGUMENT , 'd' , & args.procs.ppn} ,
{'d' , "data-geometry" , "Data geometry (t:x:y:z)" , OPTION_OPTIONAL_ARGUMENT , 's' , & dg} ,
{'b' , "block-geometry" , "Block geometry (t:x:y:z)" , OPTION_OPTIONAL_ARGUMENT , 's' , & bg} ,
{'c' , "chunk-geometry" , "Chunk geometry (t:x:y:z|auto)" , OPTION_OPTIONAL_ARGUMENT , 's' , & cg} ,
{'r' , "read" , "Enable read benchmark" , OPTION_FLAG , 'd' , & args.read_test} ,
{'w' , "write" , "Enable write benchmark" , OPTION_FLAG , 'd' , & args.write_test} ,
{'t' , "io-type" , "Independent / Collective I/O (ind|coll)" , OPTION_OPTIONAL_ARGUMENT , 's' , & iot} ,
{'u' , "unlimited" , "Enable unlimited time dimension" , OPTION_FLAG , 'd' , & args.is_unlimited} ,
{'f' , "testfile" , "Filename of the testfile" , OPTION_OPTIONAL_ARGUMENT , 's' , & args.testfn} ,
{'x' , "output-format" , "Output-Format (parser|human)" , OPTION_OPTIONAL_ARGUMENT , 's' , & xf} ,
{'F' , "use-fill-value" , "Write a fill value" , OPTION_FLAG , 'd', & args.fill_value} ,
{0 , "verify" , "Verify that the data read is correct (reads the data again)", OPTION_FLAG , 'd' , & args.verify} ,
LAST_OPTION
};
int rank;
MPI_Comm_rank(MPI_COMM_WORLD, & rank);
// check the correctness of the options only for rank 0
if (rank == 0){
printf("Benchtool (datatype: %s) \n", xstr(DATATYPE));
parseOptions(argc, argv, options);
}
MPI_Barrier(MPI_COMM_WORLD);
if (rank != 0){
parseOptions(argc, argv, options);
}
parse_dims(dg, &args.dgeom, &args.dgeom_size);
parse_dims(bg, &args.bgeom, &args.bgeom_size);
if ((0 == strcmp(iot, "c")) | (0 == strcmp(iot, "coll")) | (0 == strcmp(iot,"collective"))) {
args.par_access = NC_COLLECTIVE;
}
else if ((0 == strcmp(iot, "i")) | (0 == strcmp(iot, "ind")) | (0 == strcmp(iot, "independent"))) {
args.par_access = NC_INDEPENDENT;
}
else {
FATAL_ERR("Unsupported parallel access type %s\n", xf);
}
if (0 == strcmp(xf, "parser")) {
args.report_type = REPORT_PARSER;
}else if (0 == strcmp(xf, "human")) {
args.report_type = REPORT_HUMAN;
}else{
FATAL_ERR("Unsupported report type %s\n", xf);
}
if (0 == args.procs.nn) {
char *end = NULL;
const char* env = getenv("SLURM_NNODES");
if (NULL != env) {
args.procs.nn = strtol(env, &end, 10);
}
if (0 == args.procs.nn) {
args.procs.nn = 1;
}
}
if (0 == args.procs.ppn) {
char *end = NULL;
const char* env = getenv("SLURM_NTASKS_PER_NODE");
if (NULL != env) {
args.procs.ppn = strtol(env, &end, 10);
}
if (0 == args.procs.ppn) {
args.procs.ppn = 1;
}
}
if (NULL == args.testfn) {
const char* testfn = "./testfn.nc";
args.testfn = (char*)malloc(sizeof(*args.testfn) * strlen(testfn) + 1);
strcpy(args.testfn, testfn);
}
if (NULL == args.dgeom) {
args.dgeom_size = NDIMS;
args.dgeom = (size_t*)malloc(sizeof(*args.dgeom) * args.dgeom_size);
args.dgeom[DT] = 100;
args.dgeom[DX] = args.procs.nn * 100;
args.dgeom[DY] = args.procs.ppn * 100;
args.dgeom[DZ] = 10;
}
if (NDIMS != args.dgeom_size) {
FATAL_ERR("Found %zu dimensions (expected %d).\n", args.dgeom_size, NDIMS);
}
// Automatic block layout
if (NULL == args.bgeom) {
args.bgeom_size = args.dgeom_size;
args.bgeom = (size_t*)malloc(sizeof(*args.bgeom) * args.bgeom_size);
args.bgeom[DT] = 1;
args.bgeom[DX] = args.dgeom[DX] / args.procs.nn;
args.bgeom[DY] = args.dgeom[DY] / args.procs.ppn;
args.bgeom[DZ] = args.dgeom[DZ];
}
if (cg != NULL && 0 == strcmp(cg, "auto")) {
args.cgeom_size = args.bgeom_size;
args.cgeom = (size_t*)malloc(sizeof(*args.cgeom) * args.cgeom_size);
args.cgeom[DT] = 1;
args.cgeom[DX] = args.bgeom[DX];
args.cgeom[DY] = args.bgeom[DY];
args.cgeom[DZ] = args.bgeom[DZ];
}
else {
parse_dims(cg, &args.cgeom, &args.cgeom_size);
}
if (NDIMS != args.bgeom_size) {
FATAL_ERR("Found %zu dimensions (expected %d).\n", args.bgeom_size, NDIMS);
}
if (NULL != args.cgeom) {
if (NDIMS != args.cgeom_size) {
FATAL_ERR("Found %zu dimensions (expected %d).\n", args.cgeom_size, NDIMS);
}
}
DEBUG_MESSAGE("dgeom (%zu:%zu:%zu:%zu)\n", args.dgeom[DT], args.dgeom[DX], args.dgeom[DY], args.dgeom[DZ]);
DEBUG_MESSAGE("bgeom (%zu:%zu:%zu:%zu)\n", args.bgeom[DT], args.bgeom[DX], args.bgeom[DY], args.bgeom[DZ]);
if (NULL != args.cgeom) {
DEBUG_MESSAGE("cgeom (%zu:%zu:%zu:%zu)\n", args.cgeom[DT], args.cgeom[DX], args.cgeom[DY], args.cgeom[DZ]);
}
DEBUG_MESSAGE("(nn %zu, ppn %zu)\n", args.procs.nn, args.procs.ppn);
DEBUG_MESSAGE("test filename %s\n", args.testfn);
if (args.dgeom[DX] % args.procs.nn != 0) {
FATAL_ERR("x must be a multiple of number of nodes.\n");
}
if (args.dgeom[DY] % args.procs.ppn != 0) {
FATAL_ERR("y must be a multiple of number of processes.\n");
}
if (NULL != args.cgeom) {
if (args.dgeom[DT] % args.cgeom[DT] != 0) {
FATAL_ERR("Time range must be a multiple of time slice (range=%zu; slice=%zu)\n", args.dgeom[DT], args.cgeom[DT]);
}
}
int nranks = 0;
MPI_Comm_size(MPI_COMM_WORLD, &nranks);
if (nranks != args.procs.nn * args.procs.ppn){
FATAL_ERR("Bad environment: np != nn * ppn; np(size of MPI_COMM_WORLD)=%d, nodes(nn)=%zu, ppn(procs per node)=%zu\n",
nranks, args.procs.nn, args.procs.ppn);
}
if ((args.read_test == false) & (args.write_test == false) & (args.verify == false)) {
args.write_test = true;
}
benchmark_t wbm;
benchmark_init(&wbm);
int header_printed = 0;
benchmark_t rbm;
benchmark_init(&rbm);
if (args.write_test || args.verify) {
benchmark_setup(&wbm, args.procs, NDIMS, args.dgeom, args.bgeom, args.cgeom, args.testfn, IO_MODE_WRITE, args.par_access, args.is_unlimited, args.fill_value);
if(rank == 0){
print_header(& wbm);
header_printed = 1;
}
}
if (args.write_test) {
benchmark_run(&wbm, NULL);
report_t report;
report_init(&report);
report_setup(&report, &wbm);
report_print(&report, args.report_type);
report_destroy(&report);
}
if (args.read_test) {
int ret;
benchmark_setup(&rbm, args.procs, NDIMS, args.dgeom, args.bgeom, args.cgeom, args.testfn, IO_MODE_READ, args.par_access, args.is_unlimited, 0);
if(rank == 0 && ! header_printed){
print_header(& rbm);
header_printed = 1;
}
ret = benchmark_run(&rbm, args.verify ? wbm.block : NULL );
report_t report;
report_init(&report);
report_setup(&report, &rbm);
report_print(&report, args.report_type);
report_destroy(&report);
}else if (args.verify) {
int ret;
benchmark_setup(& rbm, args.procs, NDIMS, args.dgeom, args.bgeom, args.cgeom, args.testfn, IO_MODE_READ, args.par_access, args.is_unlimited, 0);
if(rank == 0 && ! header_printed){
print_header(& rbm);
header_printed = 1;
}
ret = benchmark_run(& rbm, wbm.block);
if (args.verify){
if (ret) {
printf("TEST PASSED [%u]\n", wbm.rank);
}
else {
printf("TEST FAILED [%u]\n", wbm.rank);
}
}
}
MPI_Finalize();
benchmark_destroy(&wbm);
benchmark_destroy(&rbm);
free(args.dgeom);
free(args.bgeom);
free(args.cgeom);
free(args.testfn);
return 0;
}
int main(int argc, char** argv) {
xm_context_t* ctx;
FILE* out;
uint32_t num_samples = 0;
float buffer[buffer_size];
if(argc != 3)
FATAL("Usage: %s <xm-file-input> <wav-file-out>\n", argv[0]);
create_context_from_file(&ctx, rate, argv[1]);
if(ctx == NULL) exit(1);
xm_set_max_loop_count(ctx, 1);
out = fopen(argv[2], "w");
if(out == NULL) FATAL_ERR("could not open output file for writing");
/* WAVE format info taken from
* http://www-mmsp.ece.mcgill.ca/Documents/AudioFormats/WAVE/WAVE.html */
/* Unlike AU, WAVE files cannot have an unknown length. This
* is why we can't write directly to stdout (we need to rewind
* because module length is hard to know. */
fputs("RIFF", out);
puts_uint32_le(0, out); /* Chunk size. Will be filled later. */
fputs("WAVE", out);
fputs("fmt ", out); /* Start format chunk */
puts_uint32_le(16, out); /* Format chunk size */
puts_uint16_le(3, out); /* IEEE float sample data */
puts_uint16_le(channels, out); /* Number of channels */
puts_uint32_le(rate, out); /* Frames/sec (sampling rate) */
puts_uint32_le(rate * channels * sizeof(float), out); /* nAvgBytesPerSec ? */
puts_uint16_le(channels * sizeof(float), out); /* nBlockAlign ? */
puts_uint16_le(8 * sizeof(float), out); /* wBitsPerSample ? */
fputs("data", out); /* Start data chunk */
puts_uint32_le(0, out); /* Data chunk size. Will be filled later. */
while(xm_get_loop_count(ctx) == 0) {
xm_generate_samples(ctx, buffer, buffer_size / channels);
num_samples += buffer_size;
for(size_t k = 0; k < buffer_size; ++k) {
union {
float f;
uint32_t i;
} u;
u.f = buffer[k];
puts_uint32_le(u.i, out);
}
}
fseek(out, 4, SEEK_SET);
puts_uint32_le(36 + num_samples * sizeof(float), out);
fseek(out, 32, SEEK_SET);
puts_uint32_le(num_samples * sizeof(float), out);
fclose(out);
xm_free_context(ctx);
return 0;
}