void *ppu_pthread_function(void *thread_arg) {
spe_context_ptr_t ctx;
struct package_t *arg = (struct package_t *) thread_arg;
/* Create SPE context */
if ((ctx = spe_context_create (0, NULL)) == NULL) {
perror ("Failed creating context");
exit (1);
}
/* Load SPE program into context */
if (spe_program_load (ctx, &lab8_spu)) {
perror ("Failed loading program");
exit (1);
}
/* Run SPE context */
unsigned int entry = SPE_DEFAULT_ENTRY;
/* transferul adresei structurii initiale */
if (spe_context_run(ctx, &entry, 0, (void *)arg, (void *)sizeof(struct package_t), NULL) < 0) {
perror ("Failed running context");
exit (1);
}
/* Destroy context */
if (spe_context_destroy (ctx) != 0) {
perror("Failed destroying context");
exit (1);
}
pthread_exit(NULL);
}
int main()
{
spe_context_ptr_t ctx;
unsigned int entry = SPE_DEFAULT_ENTRY;
init_matrix();
if((ctx = spe_context_create(0, NULL)) == NULL) {
perror ("Failed creating context");
exit (1);
}
if(spe_program_load(ctx, &lab10_spu)) {
perror ("Failed loading program");
exit (1);
}
printf("SPU:\n");
if(spe_context_run(ctx, &entry, 0, (void*)&v, (void*)M, NULL) < 0) {
perror ("Failed running context");
exit (1);
}
if(spe_context_destroy(ctx) != 0) {
perror("Failed destroying context");
exit (1);
}
printf("PPU:\n");
printf("received: %d %d \n", v[0][0].x, v[0][0].y);
printf("correct: %d\n", (destination.x == v[0][0].x) && (destination.y == v[0][0].y));
return 0;
}
int CreateSPEThread( PpuPthreadData_t *spedata, spe_program_handle_t *context, void *myarg )
{
// create SPE context
if ( ( spedata->spe_ctx = spe_context_create ( 0, NULL ) ) == NULL )
{
perror ( "Failed creating context" );
return -1;
}
// Load program into context
if ( spe_program_load ( spedata->spe_ctx, context ) )
{
perror ( "Failed loading program" );
return -1;
}
// Initialize context run data
spedata->entry = SPE_DEFAULT_ENTRY;
//speData[i].argp = mydata;
spedata->argp = myarg;
// Create pthread for each of the SPE conexts
if ( pthread_create ( &spedata->pthread, NULL, &PpuPthreadFunction, spedata ) )
{
perror ( "Failed creating thread" );
return -1;
}
return 1;
}
int main()
{
spe_context_ptr_t speid;
unsigned int flags = 0;
unsigned int entry = SPE_DEFAULT_ENTRY;
void *argp = NULL;
void *envp = NULL;
spe_stop_info_t stop_info;
int rc;
speid = spe_context_create(0,NULL);
if (speid==NULL)
{
perror("spe_context_create");
return -1;
}
//Load SPE executable object into the SPE context local store
if (spe_program_load(speid, &hello_spu))
{
perror("spe_program_load");
return -2;
}
//Run the SPE context
rc = spe_context_run(speid, &entry, flags, argp, envp, &stop_info);
if (rc<0) perror("spe_context_run");
//Destroy the SPE context
spe_context_destroy(speid);
return 0;
}
/**
* Create and start several threads on the SPEs
* @param nprocs Number of threads to start
*/
void create_spe_pthreads(fixedgrid_t* G)
{
uint32_t i;
for(i=0; i<G->nprocs; i++)
{
/* Configure environment */
G->threads[i].envv.speid = i;
G->threads[i].envv.nprocs = G->nprocs;
G->threads[i].envv.metptr = (uint32_t)(&G->threads[i].metrics);
/* Create context */
if((G->threads[i].speid = spe_context_create(0, NULL)) == NULL)
{
fprintf(stderr, "Failed spe_context_create (errno=%d)\n", errno);
exit(1);
}
/* Load program into context */
if(spe_program_load(G->threads[i].speid, &fixedgrid_spu))
{
fprintf(stderr, "Failed spe_program_load (errno=%d)\n", errno);
exit(1);
}
/* Create thread for each SPE context */
if(pthread_create(&G->threads[i].pthread, NULL, &spe_pthread_function, &G->threads[i]))
{
fprintf(stderr, "Failed pthread_create (errno=%d)\n", errno);
exit(1);
}
G->threads[i].status = SPE_STATUS_INIT;
}
}
int main(int argc, char **argv)
{
int i;
int ret;
spe_context_ptr_t spe;
spe_program_handle_t *prog;
unsigned int entry;
spe_stop_info_t stop_info;
prog = spe_image_open("vec_abs_spe.elf");
if (!prog) {
perror("spe_image_open");
exit(1);
}
spe = spe_context_create(0, NULL);
if (!spe) {
perror("spe_context_create");
exit(1);
}
ret = spe_program_load(spe, prog);
if (ret) {
perror("spe_program_load");
exit(1);
}
abs_params.ea_in = (unsigned long) in;
abs_params.ea_out = (unsigned long) out;
abs_params.size = SIZE;
entry = SPE_DEFAULT_ENTRY;
ret = spe_context_run(spe, &entry, 0, &abs_params, NULL, &stop_info);
if (ret < 0) {
perror("spe_context_run");
exit(1);
}
ret = spe_context_destroy(spe);
if (ret) {
perror("spe_context_destroy");
exit(1);
}
ret = spe_image_close(prog);
if (ret) {
perror("spe_image_close");
exit(1);
}
for (i = 0; i < SIZE; i++) {
printf("out[%02d]=%0.0f\n", i, out[i]);
}
return 0;
}
int main(int argc, char **argv)
{
int i;
int ret;
spe_context_ptr_t spe[NUM_SPE];
spe_program_handle_t *prog;
pthread_t thread[NUM_SPE];
prog = spe_image_open("increment_spe.elf");
if (!prog) {
perror("spe_image_open");
exit(1);
}
for (i = 0; i < NUM_SPE; i++) {
spe[i] = spe_context_create(0, NULL);
if (!spe) {
perror("spe_context_create");
exit(1);
}
ret = spe_program_load(spe[i], prog);
if (ret) {
perror("spe_program_load");
exit(1);
}
}
for (i = 0; i < NUM_SPE; i++) {
ret = pthread_create(&thread[i], NULL, run_increment_spe, &spe[i]);
if (ret) {
perror("pthread_create");
exit(1);
}
}
for (i = 0; i < NUM_SPE; i++) {
pthread_join(thread[i], NULL);
ret = spe_context_destroy(spe[i]);
if (ret < 0) {
perror("spe_context_destroy");
exit(1);
}
}
ret = spe_image_close(prog);
if (ret) {
perror("spe_image_close");
exit(1);
}
printf("result=%d\n", counter[0]);
return 0;
}
/**
* Create the SPU threads. This is done once during driver initialization.
* This involves setting the "init" message which is sent to each SPU.
* The init message specifies an SPU id, total number of SPUs, location
* and number of batch buffers, etc.
*/
void
cell_start_spus(struct cell_context *cell)
{
static boolean one_time_init = FALSE;
uint i, j;
uint timebase = get_timebase();
if (one_time_init) {
fprintf(stderr, "PPU: Multiple rendering contexts not yet supported "
"on Cell.\n");
abort();
}
one_time_init = TRUE;
assert(cell->num_spus <= CELL_MAX_SPUS);
ASSERT_ALIGN16(&cell_global.inits[0]);
ASSERT_ALIGN16(&cell_global.inits[1]);
/*
* Initialize the global 'inits' structure for each SPU.
* A pointer to the init struct will be passed to each SPU.
* The SPUs will then each grab their init info with mfc_get().
*/
for (i = 0; i < cell->num_spus; i++) {
cell_global.inits[i].id = i;
cell_global.inits[i].num_spus = cell->num_spus;
cell_global.inits[i].debug_flags = cell->debug_flags;
cell_global.inits[i].inv_timebase = 1000.0f / timebase;
for (j = 0; j < CELL_NUM_BUFFERS; j++) {
cell_global.inits[i].buffers[j] = cell->buffer[j];
}
cell_global.inits[i].buffer_status = &cell->buffer_status[0][0][0];
cell_global.inits[i].spu_functions = &cell->spu_functions;
cell_global.spe_contexts[i] = spe_context_create(0, NULL);
if (!cell_global.spe_contexts[i]) {
fprintf(stderr, "spe_context_create() failed\n");
exit(1);
}
if (spe_program_load(cell_global.spe_contexts[i], &g3d_spu)) {
fprintf(stderr, "spe_program_load() failed\n");
exit(1);
}
pthread_create(&cell_global.spe_threads[i], /* returned thread handle */
NULL, /* pthread attribs */
&cell_thread_function, /* start routine */
&cell_global.inits[i]); /* thread argument */
}
}
int main()
{
int i, spu_threads;
spe_context_ptr_t ctxs[MAX_SPU_THREADS];
pthread_t threads[MAX_SPU_THREADS];
/*
* Determine the number of SPE threads to create.
*/
spu_threads = spe_cpu_info_get(SPE_COUNT_USABLE_SPES, -1);
if (spu_threads > MAX_SPU_THREADS) spu_threads = MAX_SPU_THREADS;
/*
* Create several SPE-threads to execute 'simple_spu'.
*/
for(i=0; i<spu_threads; i++) {
/* Create context */
if ((ctxs[i] = spe_context_create (0, NULL)) == NULL) {
perror ("Failed creating context");
exit (1);
}
/* Load program into context */
if (spe_program_load (ctxs[i], &simple_spu)) {
perror ("Failed loading program");
exit (1);
}
/* Create thread for each SPE context */
if (pthread_create (&threads[i], NULL, &ppu_pthread_function, &ctxs[i])) {
perror ("Failed creating thread");
exit (1);
}
}
/* Wait for SPU-thread to complete execution. */
for (i=0; i<spu_threads; i++) {
if (pthread_join (threads[i], NULL)) {
perror("Failed pthread_join");
exit (1);
}
/* Destroy context */
if (spe_context_destroy (ctxs[i]) != 0) {
perror("Failed destroying context");
exit (1);
}
}
printf("\nThe program has successfully executed.\n");
return 0;
}
void *
spe_thread (void * arg)
{
int flags = 0;
unsigned int entry = SPE_DEFAULT_ENTRY;
spe_context_ptr_t *ctx = (spe_context_ptr_t *) arg;
spe_program_load (*ctx, &coremaker_spu);
spe_context_run (*ctx, &entry, flags, NULL, NULL, NULL);
pthread_exit (NULL);
}
int main(int argc, char **argv)
{
int ret;
spe_context_ptr_t spe;
spe_program_handle_t *prog;
unsigned int entry;
spe_stop_info_t stop_info;
unsigned long param;
prog = spe_image_open("print_param_spe.elf");
if (!prog) {
perror("spe_image_open");
exit(1);
}
spe = spe_context_create(0, NULL);
if (!spe) {
perror("spe_context_create");
exit(1);
}
ret = spe_program_load(spe, prog);
if (ret) {
perror("spe_program_load");
exit(1);
}
param = 12345678;
printf("[PPE] param=%ld\n", param);
entry = SPE_DEFAULT_ENTRY;
ret = spe_context_run(spe, &entry, 0, (void *) param, NULL, &stop_info);
if (ret < 0) {
perror("spe_context_run");
exit(1);
}
ret = spe_context_destroy(spe);
if (ret) {
perror("spe_context_destroy");
exit(1);
}
ret = spe_image_close(prog);
if (ret) {
perror("spe_image_close");
exit(1);
}
return 0;
}
int
indirect_handler (unsigned char *base, unsigned long offset)
{
int flags = 0;
unsigned int entry = SPE_DEFAULT_ENTRY;
spe_context_ptr_t ctx = spe_context_create (0, NULL);
spe_program_load (ctx, &bt2_spu);
spe_context_run (ctx, &entry, flags, NULL, NULL, NULL);
return 0;
}
/* Start the Spu threads */
void startSpuThreads(int spu_threads, SpuThreadData * spu_data) {
int i, no_spus;
/* Determine the number of SPE threads to create */
no_spus = spe_cpu_info_get(SPE_COUNT_USABLE_SPES, -1);
if (spu_threads < 0) {
spu_threads = no_spus;
} else if (no_spus < spu_threads) {
spu_threads = no_spus;
printf("Warning: Only %i Cell SPU processors available\n", spu_threads);
}
spu_data->no_spu_threads = spu_threads;
spu_data->spus = (SpuData *) malloc(sizeof(SpuData) * spu_threads);
if ((spu_data->spus == NULL)) {
perror("Failed to allocate SPU data for threads");
}
printf("Bringing up %i Cell SPU threads\n", spu_threads);
/* create the context gang */
if ((spu_data->gang = spe_gang_context_create(0)) == NULL) {
perror("Failed creating Cell SPU gang context");
exit(1);
}
for(i=0; i<spu_threads; i++) {
/* Create context */
if ((spu_data->spus[i].ctx = spe_context_create (CTX_FLAGS, spu_data->gang)) == NULL) {
perror ("Failed creating Cell SPU context");
exit (1);
}
/* load bootloader into spu's */
if (spe_program_load (spu_data->spus[i].ctx, &cellspu_bootloader)) {
perror ("Failed loading Cell SPU bootloader");
exit (1);
}
/* create a thread for each SPU */
if (pthread_create (&(spu_data->spus[i].boot_thread),
NULL,
&spu_bootstrap_thread,
&(spu_data->spus[i].ctx))) {
perror ("Failed creating Cell SPU thread");
exit (1);
}
}
}
/** Sends a SPECommand to the SPE.
*
* \param command A SPEcommand.
* \return Returns nonzero on error.
*/
static int submitSPECommand(SPECommand* command) {
#ifdef MAIL
/* Call the SPU Program*/
writeMailBox( (ppu_addr_t)command );
return readMailBox();
#else
if( !spe_context ){
spe_context = spe_context_create( 0, NULL );
spe_program_load( spe_context, &spe_dynprogr_handle );
}
unsigned int entry = SPE_DEFAULT_ENTRY;
return spe_context_run( spe_context, &entry, 0, command, NULL, NULL );
/* spe_context_destroy( spe_context ); */
#endif
}
static void * sws_spe_thread(void * arg)
{
struct yuvscaler_s * arg_ptr;
arg_ptr=(struct yuvscaler_s *) arg;
// spe_program_handle_t * program;
//
// program = spe_image_open("spu_yuvscaler");
if (spe_program_load(arg_ptr->ctx, &spu_yuvscaler_handle) < 0)
{
perror("error loading program");
pthread_exit(NULL);
}
spe_context_run(arg_ptr->ctx, &arg_ptr->entry, arg_ptr->runflags,arg_ptr->argp,arg_ptr->envp, NULL);
pthread_exit(NULL);
}
spe_context_ptr_t ps3_assign_context_to_program(spe_program_handle_t *program)
{
static spe_context_ptr_t cached_context;
static spe_program_handle_t *cached_program;
static int cached_pid;
int current_pid = getpid();
int thread_index = 99; /* Todo: get true cruncher index */
int retval;
if (cached_context)
{
if (cached_pid != current_pid)
{
Log("!!! FATAL !!! Cached SPE context forked from another pid (%d)\n", cached_pid);
abort();
}
if (cached_program != program)
{
// Log("Replacing SPE context because SPE program changed\n");
if (spe_context_destroy(cached_context))
Log("Alert SPE%d! spe_context_destroy() failed, errno=%d\n", thread_index, errno);
cached_context = NULL;
}
}
if (cached_context == NULL)
{
cached_context = spe_context_create(0, NULL);
if (cached_context == NULL)
{
Log("Alert SPE#%d! spe_context_create() failed\n", thread_index);
abort();
}
retval = spe_program_load(cached_context, program);
if (retval != 0)
{
Log("Alert SPE#%d: spe_program_load() returned %d\n", thread_index, retval);
abort();
}
cached_program = program;
cached_pid = current_pid;
}
return cached_context;
}
void *pthread_run_spe(void *arg){
spe_context_ptr_t spe_ctx;
context *data = (context *)arg;
void *argp;
unsigned int entry;
spe_ctx = spe_context_create(0, NULL);
spe_program_load (spe_ctx, &spu_pi);
entry=SPE_DEFAULT_ENTRY;
argp=data;
spe_context_run(spe_ctx, &entry,0,argp,NULL,NULL);
spe_context_destroy(spe_ctx);
pthread_exit(NULL);
}
int main(int argc, char **argv) {
int ret;
spe_context_ptr_t ctx;
unsigned int entry_point;
spe_stop_info_t stop_info;
/* Display the EA of the array */
printf("PPU array location: %#llx\n",
(unsigned long long)prime);
/* Create the SPE Context */
ctx = spe_context_create(0, NULL);
if (!ctx) {
perror("spe_context_create");
exit(1);
}
/* Load the program into the context */
ret = spe_program_load(ctx, &spu_prime_handle);
if (ret) {
perror("spe_program_load");
exit(1);
}
/* Run the program */
entry_point = SPE_DEFAULT_ENTRY;
ret = spe_context_run(ctx, &entry_point, 0,
NULL, NULL, &stop_info);
if (ret < 0) {
perror("spe_context_run");
exit(1);
}
/* Deallocate the context */
ret = spe_context_destroy(ctx);
if (ret) {
perror("spe_context_destroy");
exit(1);
}
return 0;
}
//PPU Code
int main(void){
int retval;
unsigned int entry_point = SPE_DEFAULT_ENTRY; // Required for continuing
//execution, SPE_DEFAULT_ENTRY is the standard starting offset.
spe_context_ptr_t my_context;
spe_stop_info_t stopinfo;
int stop_counter = 0;
spe_callback_handler_register(null_callback, 0x11, SPE_CALLBACK_NEW);
while(true) {
// Create the SPE Context
my_context = spe_context_create(SPE_EVENTS_ENABLE|SPE_MAP_PS, NULL);
// Load the embedded code into this context
spe_program_load(my_context, &spe_program_zero);
entry_point = SPE_DEFAULT_ENTRY;
do {
printf("before running the spu code\n");
retval = spe_context_run(my_context, &entry_point, 0, NULL, NULL, &stopinfo);
/* consume the stop info so we don't get the spu_stop in loop bug */
spe_stop_info_read(my_context, &stopinfo);
stop_counter++;
printf("after running the spu code (%d)\n", stop_counter);
printf("retval = %d\n", retval);
if(retval == 0x10) /* spu_stop(0x10) is sent from the spe when the loop is done */
{
break;
}
} while (retval > 0); // Run until exit or error
spe_context_destroy(my_context);
}
printf("finished with computation\n");
}
void *ppu_pthread_function(void *thread_arg) {
spe_context_ptr_t ctx;
pointers_t *arg = (pointers_t *) thread_arg;
/* Create SPE context */
if ((ctx = spe_context_create (0, NULL)) == NULL) {
perror ("Failed creating context");
exit (1);
}
/* Load SPE program into context */
if (spe_program_load (ctx, &ex1_spu)) {
perror ("Failed loading program");
exit (1);
}
pthread_t mbox_thread;
if (pthread_create (&mbox_thread, NULL, &mailbox_pthread_function, &ctx)) {
perror ("Failed creating thread");
exit (1);
}
/* Run SPE context */
unsigned int entry = SPE_DEFAULT_ENTRY;
if (spe_context_run(ctx, &entry, 0, arg, (void*)sizeof(pointers_t), NULL) < 0) {
perror ("Failed running context");
exit (1);
}
/* Destroy context */
if (spe_context_destroy (ctx) != 0) {
perror("Failed destroying context");
exit (1);
}
return NULL;
}
int CreateSPEContext( PpuPthreadData_t *spedata, spe_program_handle_t *context, void *myarg )
{
// create SPE context
if ( ( spedata->spe_ctx = spe_context_create ( 0, NULL ) ) == NULL )
{
perror ( "Failed creating context" );
return -1;
}
// Load program into context
if ( spe_program_load ( spedata->spe_ctx, context ) )
{
perror ( "Failed loading program" );
return -1;
}
// Initialize context run data
spedata->entry = SPE_DEFAULT_ENTRY;
//speData[i].argp = mydata;
spedata->argp = myarg;
return 1;
}
void *spe_code_launch_6(void *data)
{
// printf("inside of thread function\n");
int retval;
unsigned int entry_point = SPE_DEFAULT_ENTRY; /* Required for continuing
execution, SPE_DEFAULT_ENTRY is the standard starting offset. */
spe_context_ptr_t my_context;
// printf("before creating context\n");
/* Create the SPE Context */
my_context = spe_context_create(SPE_EVENTS_ENABLE|SPE_MAP_PS, NULL);
// printf("context created\n");
/* Load the embedded code into this context */
spe_program_load(my_context, &spe_code);
// printf("program loaded\n");
/* Run the SPE program until completion */
do
{
retval = spe_context_run(my_context, &entry_point, 0, spe6_Data, 6, NULL);
}
while (retval > 0); /* Run until exit or error */
spe_context_destroy(my_context);
pthread_exit(NULL);
}
int SPE_Boot(_THIS, spu_data_t * spe_data)
{
deprintf(2, "[PS3->SPU] Create SPE Context: %s\n", spe_data->program_name);
spe_data->ctx = spe_context_create(0, NULL);
if (spe_data->ctx == NULL) {
deprintf(2, "[PS3->SPU] Failed creating SPE context: %s\n", spe_data->program_name);
SDL_SetError("[PS3->SPU] Failed creating SPE context");
return -1;
}
deprintf(2, "[PS3->SPU] Load Program into SPE: %s\n", spe_data->program_name);
if (spe_program_load(spe_data->ctx, &spe_data->program)) {
deprintf(2, "[PS3->SPU] Failed loading program into SPE context: %s\n", spe_data->program_name);
SDL_SetError
("[PS3->SPU] Failed loading program into SPE context");
return -1;
}
spe_data->booted = 1;
deprintf(2, "[PS3->SPU] SPE boot successful\n");
return 0;
}
int main(int argc, char** argv)
{
double begin;
double end;
int errnum;
size_t nthread = P;
size_t i;
size_t nvertex;
unsigned int x; // sent to each SPU
int code; // status;
unsigned int reply; // from SPU
arg_t data[nthread];
param_t param[nthread] A16;
argc = argc; // to silence gcc...
progname = argv[0];
nvertex = atoi(argv[2]);
printf("nthread = %zu\n", nthread);
printf("nvertex = %zu\n", nvertex);
printf("ctx = %zu\n", sizeof(param_t));
printf("arg = %zu\n", sizeof(arg_t));
begin = sec();
for (i = 0; i < nthread; ++i) {
param[i].proc = i;
param[i].nvertex = nvertex;
if ((data[i].ctx = spe_context_create (0, NULL)) == NULL) {
perror ("Failed creating context");
exit(1);
}
if (spe_program_load (data[i].ctx, &dataflow)) {
perror ("Failed loading program");
exit(1);
}
data[i].arg = ¶m[i];
printf("i=%d param=%p\n", i, data[i].arg);
if (pthread_create (&data[i].pthread, NULL, work, &data[i])) {
perror ("Failed creating thread");
exit(1);
}
}
// send some data to each SPU and wait for a reply.
x = 42;
for (i = 0; i < nthread; ++i) {
reply = 0;
code = spe_out_mbox_read(data[i].ctx, &reply, 1);
printf("spu-%d reply-0: %u\tcode: %d\n",i, reply, code);
code = spe_in_mbox_write(data[i].ctx, &x, 1, 1);
code = spe_out_mbox_read(data[i].ctx, &reply, 1);
printf("spu-%d reply-1: %u\tcode: %d\n",i, reply, code);
code = spe_out_mbox_read(data[i].ctx, &reply, 1);
printf("spu-%d reply-2: %u\tcode: %d\n",i, reply, code);
}
end = sec();
printf("%1.3lf s\n", end-begin);
for (i = 0; i < nthread; ++i) {
printf("joining with PPU pthread %zu...\n", i);
errnum = pthread_join(data[i].pthread, NULL);
if (errnum != 0)
syserror(errnum, "pthread_join failed");
if (spe_context_destroy (data[i].ctx) != 0) {
perror("Failed destroying context");
exit(1);
}
}
return 0;
}
int main(int argc, char **argv)
{
int i;
int ret;
spe_context_ptr_t spe;
spe_program_handle_t *prog;
unsigned int entry;
spe_stop_info_t stop_info;
if (argc == 1) {
fprintf(stderr, "usage: %s <spu_image>\n", argv[0]);
return -1;
}
prog = spe_image_open(argv[1]);
if (!prog) {
perror("spe_image_open");
exit(1);
}
spe = spe_context_create(0, NULL);
if (!spe) {
perror("spe_context_create");
exit(1);
}
ret = spe_program_load(spe, prog);
if (ret) {
perror("spe_program_load");
exit(1);
}
abs_params.ea_in = (unsigned long) in;
abs_params.ea_out = (unsigned long) out;
abs_params.size = SIZE;
entry = SPE_DEFAULT_ENTRY;
ret = spe_context_run(spe, &entry, 0, &abs_params, NULL, &stop_info);
if (ret < 0) {
perror("spe_context_run");
exit(1);
}
ret = spe_context_destroy(spe);
if (ret) {
perror("spe_context_destroy");
exit(1);
}
ret = spe_image_close(prog);
if (ret) {
perror("spe_image_close");
exit(1);
}
for (i = 0; i < SIZE; i++) {
printf("%5.0f ", i, out[i]);
if ((i+1) % 4 == 0) printf("\n");
}
return 0;
}
///start the spus group (can be called at the beginning of each frame, to make sure that the right SPU program is loaded)
void SpuLibspe2Support::internal_startSPU()
{
m_activeSpuStatus.resize(numThreads);
for (int i=0; i < numThreads; i++)
{
if(data[i].context == NULL)
{
/* Create context */
if ((data[i].context = spe_context_create(0, NULL)) == NULL)
{
perror ("Failed creating context");
exit(1);
}
/* Load program into context */
if(spe_program_load(data[i].context, this->program))
{
perror ("Failed loading program");
exit(1);
}
m_activeSpuStatus[i].m_status = Spu_Status_Startup;
m_activeSpuStatus[i].m_taskId = i;
m_activeSpuStatus[i].m_commandId = 0;
m_activeSpuStatus[i].m_lsMemory.p = NULL;
data[i].entry = SPE_DEFAULT_ENTRY;
data[i].flags = 0;
data[i].argp.p = &m_activeSpuStatus[i];
data[i].envp.p = NULL;
/* Create thread for each SPE context */
if (pthread_create(&data[i].pthread, NULL, &ppu_pthread_function, &(data[i]) ))
{
perror ("Failed creating thread");
exit(1);
}
/*
else
{
printf("started thread %d\n",i);
}*/
}
}
for (int i=0; i < numThreads; i++)
{
if(data[i].context != NULL)
{
while( m_activeSpuStatus[i].m_status == Spu_Status_Startup)
{
// wait for spu to set up
sched_yield();
}
printf("Spu %d is ready\n", i);
}
}
}
int main(int argc, char** argv)
{
/* Iterators */
int i, j, k;
uint32_t block;
/* Time (seconds) */
long t_0;
long t_end;
long dt;
long steps;
long iter;
/* Emission control */
bool emflag = TRUE;
/* Start wall clock timer */
timer_start(TIMER_WALLCLOCK);
/* Initialize parallelization */
nprocs = spe_cpu_info_get(SPE_COUNT_USABLE_SPES, -1);
nprocs = nprocs > MAX_THREADS ? MAX_THREADS : nprocs;
if(argc > 1)
{
i = atoi(argv[1]);
if(i < 1)
{
fprintf(stderr, "Invalid number of SPUs: %d < 1.\n", i);
exit(1);
}
if(i < nprocs)
{
nprocs = i;
}
else
{
printf("%d SPUs unavailable. Using %d instead.\n", i, nprocs);
}
}
/* Create SPE threads */
for(i=0; i<nprocs; i++)
{
threads[i].argp = (void*)(&spe_argvs[i]);
/* Create context */
if((threads[i].speid = spe_context_create(0, NULL)) == NULL)
{
fprintf(stderr, "Failed spe_context_create(errno=%d strerror=%s)\n", errno, strerror(errno));
exit(1);
}
/* Load program into context */
if(spe_program_load(threads[i].speid, &fixedgrid_spu))
{
fprintf(stderr, "Failed spe_program_load(errno=%d strerror=%s)\n", errno, strerror(errno));
exit(1);
}
/* Create thread for each SPE context */
if(pthread_create(&threads[i].pthread, NULL, &ppu_pthread_function, &threads[i]))
{
fprintf(stderr, "Failed pthread_create(errno=%d strerror=%s)\n", errno, strerror(errno));
exit(1);
}
spe_set_status(i, SPE_STATUS_WAITING);
}
printf("\nRunning %d threads (%d SPU + 1 PPU).\n", (nprocs+1), nprocs);
/* Allocate concentration memory */
//conc = _malloc_align(NROWS*NCOLS*sizeof(double), 7);
//conc_buff = (double*)_malloc_align(MAX_THREADS*NY*sizeof(double), 7);
/* Allocation wind vector filed memory */
//wind_u = _malloc_align(NROWS*NCOLS*sizeof(double), 7);
//wind_v = _malloc_align(NROWS*NCOLS*sizeof(double), 7);
//wind_u_buff = (double*)_malloc_align(MAX_THREADS*NY*sizeof(double), 7);
//wind_v_buff = (double*)_malloc_align(MAX_THREADS*NY*sizeof(double), 7);
/* Allocation diffusion tensor memory */
//diff = _malloc_align(NROWS*NCOLS*sizeof(double), 7);
//diff_buff = (double*)_malloc_align(MAX_THREADS*NY*sizeof(double), 7);
/* Initialize concentration data */
double_array_init(NROWS*NCOLS, conc, O3_INIT);
/* Initialize wind field */
double_array_init(NROWS*NCOLS, wind_u, WIND_U_INIT);
double_array_init(NROWS*NCOLS, wind_v, WIND_V_INIT);
/* Initialize diffusion field */
double_array_init(NROWS*NCOLS, diff, DIFF_INIT);
/* Initialize time */
t_0 = 0.0;
t_end = year2sec(END_YEAR - START_YEAR) + day2sec(END_DOY - START_DOY) +
hour2sec(END_HOUR - START_HOUR) + minute2sec(END_MIN - START_MIN);
dt = STEP_SIZE;
steps = (long)( (t_end - t_0)/dt );
/* Print startup banner */
print_start_banner(NX*DX, NY*DY, 0.0, t_end, steps);
/* Store initial concentration */
write_conc(&(conc[0]), 0, 0);
/* BEGIN CALCULATIONS */
for(iter = 1; iter <= steps; iter++)
{
emflag = iter*dt < 6*3600.0 ? TRUE : FALSE;
timer_start(TIMER_ROW_DISCRET);
/* Discretize rows 1/2 timestep */
block = NROWS / nprocs;
for(i=0; i<nprocs; i++)
{
/* Configure SPE arguments */
spe_argvs[i].arg[0].u64 = (uint64_t)(&conc[i*block*NX]);
spe_argvs[i].arg[1].u64 = (uint64_t)(&wind_u[i*block*NX]);
spe_argvs[i].arg[2].u64 = (uint64_t)(&diff[i*block*NX]);
spe_argvs[i].arg[3].dbl = dt/2;
spe_argvs[i].arg[4].dbl = DX;
spe_argvs[i].arg[5].u32[0] = NX;
spe_argvs[i].arg[5].u32[1] = (i == nprocs - 1 ? block + NROWS % nprocs : block); //FIXME
/* Signal SPE */
spe_set_status(i, SPE_STATUS_WORKING);
}
/* Wait for SPEs to finish */
wait_all_spes();
timer_stop(TIMER_ROW_DISCRET);
timer_start(TIMER_COL_DISCRET);
/* Discretize colums 1 timestep */
for(i=0; i<NCOLS; i++)
{
k = i % nprocs;
while(spe_get_status(k) > 0) ; //intentional wait
if(i >= nprocs)
{
timer_start(TIMER_ARRAY_COPY);
for(j=0; j<NY; j++)
{
conc[i-nprocs + j*NX] = ccol[k*NY+j];
}
timer_stop(TIMER_ARRAY_COPY);
}
timer_start(TIMER_ARRAY_COPY);
for(j=0; j<NY; j++)
{
ccol[k*NY + j] = conc[i + j*NX];
wcol[k*NY + j] = wind_v[i + j*NX];
dcol[k*NY + j] = diff[i + j*NX];
}
timer_stop(TIMER_ARRAY_COPY);
// Configure SPE arguments
spe_argvs[k].arg[0].u64 = (uint64_t)(&ccol[k*NY]);
spe_argvs[k].arg[1].u64 = (uint64_t)(&wcol[k*NY]);
spe_argvs[k].arg[2].u64 = (uint64_t)(&dcol[k*NY]);
spe_argvs[k].arg[3].dbl = dt;
spe_argvs[k].arg[4].dbl = DY;
spe_argvs[k].arg[5].u32[0] = NY;
spe_argvs[k].arg[5].u32[1] = 1;
// Signal SPE
spe_set_status(k, SPE_STATUS_WORKING);
}
/* Wait for SPEs to finish */
wait_all_spes();
timer_stop(TIMER_COL_DISCRET);
timer_start(TIMER_ROW_DISCRET);
/* Discretize rows 1/2 timestep */
block = NROWS / nprocs;
for(i=0; i<nprocs; i++)
{
/* Configure SPE arguments */
spe_argvs[i].arg[0].u64 = (uint64_t)(&conc[i*block*NX]);
spe_argvs[i].arg[1].u64 = (uint64_t)(&wind_u[i*block*NX]);
spe_argvs[i].arg[2].u64 = (uint64_t)(&diff[i*block*NX]);
spe_argvs[i].arg[3].dbl = dt/2;
spe_argvs[i].arg[4].dbl = DX;
spe_argvs[i].arg[5].u32[0] = NX;
spe_argvs[i].arg[5].u32[1] = (i == nprocs - 1 ? block + NROWS % nprocs : block); //FIXME
/* Signal SPE */
spe_set_status(i, SPE_STATUS_WORKING);
}
/* Wait for SPEs to finish */
wait_all_spes();
timer_stop(TIMER_ROW_DISCRET);
/*
* Could update wind field here...
*/
/*
* Could update diffusion tensor here...
*/
/* Add emissions */
if(emflag)
{
conc[SOURCE_Y*NX + SOURCE_X] += dt * (SOURCE_RATE) / (DX * DY * 1000.0);
}
/* Store concentration */
#ifdef WRITE_EACH_ITER
write_conc(conc, iter, 0);
#endif
/* Indicate progress */
if(iter % 10 == 0)
{
printf("Iteration %ld of %ld. Time = %ld seconds.\n", iter, steps, iter*dt);
}
}
/* END CALCULATIONS */
/* Wait for SPU-thread to complete execution. */
for(i=0; i<nprocs; i++)
{
spe_set_status(i, SPE_STATUS_STOPPED);
if(pthread_join(threads[i].pthread, NULL))
{
perror("Failed pthread_join");
exit(1);
}
}
/* Store concentration */
write_conc(conc, iter-1, 0);
/* Show final time */
printf("Final time: %ld seconds.\n", (iter-1)*dt);
timer_stop(TIMER_WALLCLOCK);
print_timer_summary("===PPU Timers===");
/* Cleanup and exit */
return 0;
}
int main(int argc, char **argv) {
int i, retval, spus;
/* Determine number of available SPUs */
spus = spe_cpu_info_get(SPE_COUNT_USABLE_SPES, 0);
if (argc != 2) {
printf("Usage: 'ppu_threads <1-%u>'\n", spus);
exit(1);
}
else if ((atoi(argv[1]) < 1) ||
(atoi(argv[1]) > spus)) {
printf("Usage: 'ppu_threads <1-%u>'\n", spus);
exit(1);
}
else {
spus = atoi(argv[1]);
}
/* Create a context and thread for each SPU */
for (i=0; i<spus; i++) {
/* Create context */
if ((data[i].speid = spe_context_create(0, NULL)) == NULL)
{
perror("spe_context_create");
exit(1);
}
/* Load program into the context */
if ((retval =
spe_program_load(data[i].speid, &spu_threads)) != 0)
{
perror("spe_program_load");
exit (1);
}
/* Initialize control block and thread data */
control_block = i;
data[i].argp = (void*)control_block;
/* Create thread */
if ((retval =
pthread_create(
&data[i].pthread,
NULL,
&ppu_pthread_function,
&data[i])) != 0)
{
perror("pthread_create");
exit (1);
}
}
/* Wait for the threads to finish processing */
for (i = 0; i < spus; i++)
{
if ((retval = pthread_join(data[i].pthread, NULL)) != 0)
{
perror("pthread_join");
exit (1);
}
if ((retval = spe_context_destroy (data[i].speid)) != 0)
{
perror("spe_context_destroy");
exit (1);
}
}
return 0;
}
float calc_integral(float start, float end, float delta)
{
int i;
int ret;
float sum = 0.0f;
spe_program_handle_t *prog;
spe_context_ptr_t spe[NUM_SPE];
pthread_t thread[NUM_SPE];
thread_arg_t arg[NUM_SPE];
prog = spe_image_open("integral_spe.elf");
if (!prog) {
perror("spe_image_open");
exit(1);
}
for (i = 0; i < NUM_SPE; i++) {
spe[i] = spe_context_create(0, NULL);
if (!spe) {
perror("spe_context_create");
exit(1);
}
ret = spe_program_load(spe[i], prog);
if (ret) {
perror("spe_program_load");
exit(1);
}
}
for (i = 0; i < NUM_SPE; i++) {
integral_params[i].start = start + (end-start)/NUM_SPE * i;
integral_params[i].end = start + (end-start)/NUM_SPE * (i+1);
integral_params[i].delta = delta;
integral_params[i].sum = 0.0f;
arg[i].spe = spe[i];
arg[i].integral_params = &integral_params[i];
ret = pthread_create(&thread[i], NULL, run_integral_spe, &arg[i]);
if (ret) {
perror("pthread_create");
exit(1);
}
}
for (i = 0; i < NUM_SPE; i++) {
pthread_join(thread[i], NULL);
ret = spe_context_destroy(spe[i]);
if (ret) {
perror("spe_context_destroy");
exit(1);
}
}
ret = spe_image_close(prog);
if (ret) {
perror("spe_image_close");
exit(1);
}
for (i = 0; i < NUM_SPE; i++) {
printf("[PPE] sum = %f\n", integral_params[i].sum);
sum += integral_params[i].sum;
}
return sum;
}
int main( int argc, char *argv[] )
{
int i, j, dummy;
int tmi, tmj;
pthread_t threads [ NUM_THREADS ];
spe_context_ptr_t spe_contexts[ NUM_THREADS ];
thread_args_t thread_args [ NUM_THREADS ];
int rows;
dummy = argc; dummy = (int)argv;
// initialize initial & final matrix
for(i = 0; i < tsize; i++)
{
for(j = 0; j < tsize; j++)
{
tmi = tsize-i;
tmj = tsize-j;
Amatrix[i][j] = 3*tmi+tmj ;
Bmatrix[i][j] = 3*tmi+tmj ;
Cmatrix[i][j] = 0 ;
Dmatrix[i][j] = 0 ;
}
}
// perform multiply
printf( "SPE: Multiply \n");
gettimeofday( &time0, &tzone );
// start jobs
rows = ((tsize/32)+NUM_THREADS-1)/NUM_THREADS ;
// determine amount of work each spe should do
for (i = 0; i < NUM_THREADS; i++ )
{
// set arguments
args.Amat = (float (*)[tsize][tsize])Amatrix ;
args.Bmat = (float (*)[tsize][tsize])Bmatrix ;
args.Cmat = (float (*)[tsize][tsize])Cmatrix ;
args.i_initial = i*rows ;
spe_contexts[i] = spe_context_create( 0, NULL ); // (flags, gang)
spe_program_load( spe_contexts[i], &multiply_spu );
thread_args[i].spe_context = spe_contexts[i];
thread_args[i].argp = &args ;
thread_args[i].envp = NULL;
pthread_create( &threads[i], NULL, &spe_thread, &thread_args[i] );
}
// wait for tasks to complete
for (i = 0; i < NUM_THREADS; i++)
{
pthread_join( threads[i], NULL );
} // wait for threads
gettimeofday( &time1, &tzone );
// print time to complete
sec = time1.tv_sec - time0.tv_sec ;
usec = time1.tv_usec - time0.tv_usec ;
if ( usec < 0 )
{
sec--;
usec+=1000000 ;
}
printf(
"SPE: Multiply Done -- matrix[%d][%d]: time=%d.%06d\n",
tsize,
tsize,
sec,
usec);
for (i = 0; i < NUM_THREADS; i++)
{
spe_context_destroy( spe_contexts[i] );
} // destroy threads
// Check for correctness of final matrix
{
int error, i, j, k ;
for(i = 0; i < tsize; i++)
{
for(j = 0; j < tsize; j++)
{
for(k = 0; k < tsize; k++)
{
Dmatrix[i][j] += Amatrix[i][k] * Bmatrix[k][j] ;
}
}
}
error = 0;
for(i = 0; i < tsize; i++)
{
for(j = 0; j < tsize; j++)
{
if ( Cmatrix[i][j] != Dmatrix[i][j] )
error = 1 ;
}
}
if (error)
{
printf("Error in Multiply.\n");
}
else
{
printf("Multiply is correct.\n"); fflush(stdout);
}
}
return 0;
}