void insieme_wi_startup_implementation(irt_work_item* wi) {
irt_work_item_range fullrange_wi = {0, WI_RANGE, 1};
uint64 start_time = irt_time_ms();
irt_work_group *wg1 = irt_wg_create(), *wg2 = irt_wg_create();
insieme_wi_test_params test_params = {1, wg1, wg2, 0, 0};
irt_work_item** test_wis = (irt_work_item**)malloc(NUM_WIS * sizeof(irt_work_item*));
for(int i = 0; i < NUM_WIS; ++i) {
test_wis[i] = irt_wi_create(fullrange_wi, &g_insieme_impl_table[1], (irt_lw_data_item*)&test_params);
irt_wg_insert(wg1, test_wis[i]);
}
for(int i = 0; i < NUM_WIS; ++i) {
ERR("Z1");
irt_scheduling_assign_wi(irt_g_workers[i % irt_g_worker_count], test_wis[i]);
ERR("Z2");
}
ERR("Z3 ---");
// for(int i=0; i<NUM_WIS; ++i) {
// ERR("Z3.5");
// irt_wi_join(test_wis[i]);
// ERR("Z4");
//}
irt_wg_join(wg1->id);
ERR("Z5 ---");
uint64 end_time = irt_time_ms();
printf("======================\n= manual irt test barrier done\n");
printf("= time taken: %lu\n", end_time - start_time);
printf("======================\n");
free(test_wis);
}
void insieme_wi_startup_implementation(irt_work_item* wi) {
uint64 start_time = irt_time_ms();
irt_work_group* wg1 = irt_wg_create();
insieme_wi_test_params* test_params = (insieme_wi_test_params*)malloc(sizeof(insieme_wi_test_params));
test_params->type_id = 1;
test_params->wg = wg1;
irt_work_item** test_wis = (irt_work_item**)malloc(NUM_WIS * sizeof(irt_work_item*));
irt_work_item_id* test_wi_ids = (irt_work_item_id*)malloc(NUM_WIS * sizeof(irt_work_item_id));
for(int i = 0; i < NUM_WIS; ++i) {
test_wis[i] = irt_wi_create(irt_g_wi_range_one_elem, &g_insieme_impl_table[1], (irt_lw_data_item*)test_params);
test_wi_ids[i] = test_wis[i]->id;
irt_wg_insert(wg1, test_wis[i]);
test_params->vals[i] = 0;
}
for(int i = 0; i < NUM_WIS; ++i) {
irt_scheduling_assign_wi(irt_g_workers[i % irt_g_worker_count], test_wis[i]);
}
for(int i = 0; i < NUM_WIS; ++i) {
irt_wi_join(test_wi_ids[i]);
}
uint64 end_time = irt_time_ms();
printf("======================\n= manual irt test redistribute done\n");
printf("= time taken: %lu\n", end_time - start_time);
bool check = true;
for(uint64 i = 0; i < NUM_WIS; ++i) {
if(test_params->vals[i] != NUM_WIS - 1) {
check = false;
printf("= fail at %lu, expected %d / actual %lu\n", i, NUM_WIS - 1, (test_params->vals[i]));
break;
}
}
printf("= result check: %s\n======================\n", check ? "OK" : "FAIL");
free(test_wis);
free(test_wi_ids);
free(test_params);
}
void insieme_wi_bench_implementation(irt_work_item* wi) {
insieme_wi_bench_params* params = (insieme_wi_bench_params*)wi->parameters;
if(params->count > 0) {
insieme_wi_bench_params bench_params = {1, params->count - 1, params->check};
irt_work_item_id* bench_wi_ids = (irt_work_item_id*)malloc(NUM_ITER * sizeof(irt_work_item_id));
for(int i = 0; i < NUM_ITER; ++i) {
irt_work_item* wi = irt_wi_create(irt_g_wi_range_one_elem, &g_insieme_impl_table[1], (irt_lw_data_item*)&bench_params);
bench_wi_ids[i] = wi->id;
irt_scheduling_assign_wi(irt_worker_get_current(), wi);
}
// irt_wi_multi_join(NUM_ITER, bench_wis);
for(int i = 0; i < NUM_ITER; ++i) {
irt_wi_join(bench_wi_ids[i]);
}
free(bench_wi_ids);
}
irt_atomic_inc(params->check, uint64);
}
void insieme_wi_startup_implementation(irt_work_item* wi) {
{
uint64 start_time = irt_time_ms();
uint64 check_val = 0;
insieme_wi_bench_params bench_params = {1, NUM_LEVELS, &check_val};
for(int i = 0; i < NUM_REPEATS; ++i) {
irt_work_item* bench_wi = irt_wi_create(irt_g_wi_range_one_elem, &g_insieme_impl_table[1], (irt_lw_data_item*)&bench_params);
irt_work_item_id wi_id = bench_wi->id;
irt_scheduling_assign_wi(irt_worker_get_current(), bench_wi);
irt_wi_join(wi_id);
}
uint64 total_time = irt_time_ms() - start_time;
uint64 total_wis = (uint64)pow((double)NUM_ITER, (double)NUM_LEVELS);
uint64 wis_per_sec = (uint64)(total_wis / ((double)total_time / 1000.0));
printf("======================\n= manual irt wi benchmark done\n");
printf("= number of wis executed: %lu\n", check_val);
printf("= time taken: %lu\n", total_time);
printf("= wis/s: %lu\n======================\n", wis_per_sec);
}
//{
// uint64 start_time = irt_time_ms();
// uint64 check_val = 0;
// insieme_wi_bench_params bench_params = { 1, NUM_LEVELS, &check_val };
// irt_work_item* bench_wi = irt_wi_create(irt_g_wi_range_one_elem, 2, (irt_lw_data_item*)&bench_params);
// irt_scheduling_assign_wi(irt_worker_get_current(), bench_wi);
// irt_wi_join(bench_wi);
// uint64 total_time = irt_time_ms() - start_time;
// uint64 total_wis = pow(NUM_ITER, NUM_LEVELS);
// uint64 wis_per_sec = (uint64)(total_wis/((double)total_time/1000.0));
// printf("======================\n= manual irt optional wi benchmark done\n");
// printf("= number of wis executed: %lu\n", check_val);
// printf("= time taken: %lu\n", total_time);
// printf("= optional wis/s: %lu\n======================\n", wis_per_sec);
//}
}
void insieme_wi_startup_implementation(irt_work_item* wi) {
// create data arrays
irt_data_range range[] = {{0,N,1},{0,N,1}};
irt_data_item* A = irt_di_create(INSIEME_DOUBLE_T_INDEX, 2, range);
irt_data_item* B = irt_di_create(INSIEME_DOUBLE_T_INDEX, 2, range);
irt_data_item* C = irt_di_create(INSIEME_DOUBLE_T_INDEX, 2, range);
// measure the time
uint64 start_time = irt_time_ms();
uint64 start_ticks = irt_time_ticks();
// create and run initialization job
insieme_wi_init_params init_params = { INSIEME_WI_INIT_PARAM_T_INDEX, A->id, B->id };
irt_work_item* init_wi = irt_wi_create((irt_work_item_range){ 0, N, 1 }, &g_insieme_impl_table[INSIEME_WI_INIT_INDEX], (irt_lw_data_item*)&init_params);
irt_work_item_id init_id = init_wi->id;
irt_scheduling_assign_wi(irt_worker_get_current(), init_wi);
// wait until finished
irt_wi_join(init_id);
// conduct the multiplication
insieme_wi_mul_params mul_params = { INSIEME_WI_MUL_PARAM_T_INDEX, A->id, B->id, C->id };
irt_work_item* mul_wi = irt_wi_create((irt_work_item_range){ 0, N, 1 }, &g_insieme_impl_table[INSIEME_WI_MUL_INDEX], (irt_lw_data_item*)&mul_params);
irt_work_item_id mul_id = mul_wi->id;
irt_scheduling_assign_wi(irt_worker_get_current(), mul_wi);
// wait until finished
irt_wi_join(mul_id);
// stop the time
uint64 end_ticks = irt_time_ticks();
uint64 end_time = irt_time_ms();
// check correctness
irt_data_range subrange[] = {{0,N,1},{0,N,1}};
irt_data_item* itemR = irt_di_create_sub(irt_data_item_table_lookup(C->id), subrange);
irt_data_block* blockR = irt_di_acquire(itemR, IRT_DMODE_READ_ONLY);
double** R = (double**)blockR->data;
printf("======================\n= manual irt test matrix multiplication\n");
printf("= time taken: %lu ms, %lu clock ticks\n", end_time - start_time, end_ticks - start_ticks);
bool check = true;
for (int i=0; i<N; i++) {
for (int j=0; j<N; j++) {
if (R[i][j] != i*j) {
check = false;
//printf("= fail at (%d,%d) - expected %d / actual %f\n", i, j, i*j, R[i][j]);
}
}
}
printf("= result check: %s\n======================\n", check ? "OK" : "FAIL");
irt_di_free(blockR);
irt_di_destroy(itemR);
// cleanup
irt_di_destroy(A);
irt_di_destroy(B);
irt_di_destroy(C);
}
