#include "common.h"
#define PROFILE 1
#define CHECK_CU_ERROR(err, cufunc) \
if (err != CUDA_SUCCESS) { printf ("Error %d for CUDA Driver API function '%s'\n", err, cufunc); return -1; }
// tbb::task_scheduler_init init(1);
int main() {
cudaSetDevice(0);
CUdevice device;
cuDeviceGet(&device, 0);
// bool verbose = 0;
// srand(123);
// unsigned int size = 52428800; //200 MB
// double alpha = 2.5;
// bool custom = true;
// bool skipping = true;
//2, 4, 8, 12, 16, 24, 32, 40
//4, 10, 20, 30, 40
//TODO: make it support cache size > 8 GB (there are lots of integer overflow resulting in negative offset to the gpuCache) should have used unsigned int everywhere
// CPUGPUProcessing* cgp = new CPUGPUProcessing(size * 12, 0, 52428800 * 15, 52428800 * 20, verbose, custom, skipping);
// QueryProcessing* qp;
cout << endl;
// CacheManager* cm = cgp->cm;
// bool exit = 0;
// string input, query, many, policy;
// int many_query;
// ReplacementPolicy repl_policy;
// double time = 0;
// long long cpu_traffic = 0, gpu_traffic = 0, pcie_traffic = 0, malloc_time_total = 0;
// double time1 = 0, time2 = 0;
// long long cpu_traffic1 = 0, cpu_traffic2 = 0;
// long long gpu_traffic1 = 0, gpu_traffic2 = 0;
// long long pcie_traffic1 = 0, pcie_traffic2 = 0;
// double malloc_time_total1 = 0, malloc_time_total2 = 0;
// Distribution dist = None;
// int processed_segment = 0;
// int skipped_segment = 0;
// double mean = 0;
// qp = new QueryProcessing(cgp, verbose, dist);
// if (dist == Zipf) {
// qp->qo->setDistributionZipfian(alpha);
// } else if (dist == Norm) {
// qp->qo->setDistributionNormal(mean, 0.5);
// }
// #ifdef PROFILE
PAPI_library_init( PAPI_VER_CURRENT );
// InitMonitor();
// #endif
// #ifdef PROFILE
cout << endl;
cout << "We are profiling" << endl;
int retval;
char const *EventName[] = { "cuda:::dram__bytes_read.sum:device=0", "cuda:::dram__bytes_write.sum:device=0"}; // CUPTI_11 event.
int* events = new int [NUM_EVENTS];
int EventSet = PAPI_NULL;
for(int i = 0; i < NUM_EVENTS; i++ ){
retval = PAPI_event_name_to_code( (char *)EventName[i], &events[i] );
assert(retval == PAPI_OK);
}
retval = PAPI_create_eventset( &EventSet ); assert(retval == PAPI_OK);
retval = PAPI_add_events( EventSet, events, NUM_EVENTS ); assert(retval == PAPI_OK);
// #endif
CUcontext sessionCtx = NULL;
CUcontext poppedCtx, curCtx;
CHECK_CU_ERROR( cuCtxCreate(&sessionCtx, 0, device), "cuCtxCreate");
cuCtxGetCurrent(&curCtx); cout << curCtx << endl;
CHECK_CU_ERROR( cuCtxPopCurrent(&poppedCtx), "cuCtxPopCurrent" );
cuCtxGetCurrent(&curCtx); cout << curCtx << endl;
// while (!exit) {
// cout << "Select Options:" << endl;
// cout << "1. Run Specific Query" << endl;
// cout << "2. Run Random Queries" << endl;
// cout << "3. Run Experiment" << endl;
// cout << "4. Replacement" << endl;
// cout << "5. Dump Trace" << endl;
// cout << "6. Exit" << endl;
// cout << "cache. Cache Specific Column" << endl;
// cout << "clear. Delete Columns from GPU" << endl;
// cout << "custom. Toggle custom malloc" << endl;
// cout << "skipping. Toggle segment skipping" << endl;
// cout << "pipelining. Toggle operator pipelining" << endl;
// cout << "Your Input: ";
// cin >> input;
// if (input.compare("1") == 0) {
// time = 0; cpu_traffic = 0; gpu_traffic = 0; pcie_traffic = 0; malloc_time_total = 0;
// cgp->resetTime();
// cout << "Input Query: ";
// cin >> query;
// qp->setQuery(stoi(query));
// #ifdef PROFILE
long long p_values[NUM_EVENTS];
long long read = 0, write = 0;
for(int i = 0; i < NUM_EVENTS; i++ ){
p_values[i] = 0;
}
// CHECK_CU_ERROR(cuCtxGetCurrent(&curCtx), "cuCtxGetCurrent"); cout << curCtx << endl;
// CHECK_CU_ERROR(cuCtxSetCurrent(sessionCtx), "cuCtxSetCurrent");
CHECK_CU_ERROR(cuCtxPushCurrent(sessionCtx), "cuCtxPushCurrent");
retval = PAPI_start( EventSet ); assert(retval == PAPI_OK);
CHECK_CU_ERROR( cuCtxPopCurrent(&sessionCtx), "cuCtxPopCurrent" );
// CHECK_CU_ERROR(cuCtxSetCurrent(curCtx), "cuCtxSetCurrent");
// StartMonitor();
// #endif
parallel_for(int(0), 64, [=](int j){
CUcontext poppedCtx;
cuCtxPushCurrent(sessionCtx);
cudaStream_t stream;
cudaStreamCreate(&stream);
cudaStreamSynchronize(stream);
cudaStreamDestroy(stream);
cuCtxPopCurrent(&poppedCtx);
});
// time1 = qp->processQuery(sessionCtx);
// cpu_traffic1 = cgp->cpu_traffic_total;
// gpu_traffic1 = cgp->gpu_traffic_total;
// pcie_traffic1 = cgp->pcie_traffic_total;
// malloc_time_total1 = cgp->malloc_time_total;
// #ifdef PROFILE
// EndMonitor(read, write);
cout << endl;
CHECK_CU_ERROR(cuCtxPushCurrent(sessionCtx), "cuCtxPushCurrent");
CHECK_CU_ERROR( cuCtxSynchronize( ), "cuCtxSynchronize" );
CHECK_CU_ERROR( cuCtxPopCurrent(&sessionCtx), "cuCtxPopCurrent" );
retval = PAPI_stop( EventSet, p_values ); assert(retval == PAPI_OK);
retval = PAPI_reset(EventSet); assert(retval == PAPI_OK);
// gpu_traffic = 0;
for(int i = 0; i < NUM_EVENTS; i++ ) {
printf( "stop: %12lld \t=0X%016llX \t\t --> %s \n", p_values[i], p_values[i], EventName[i] );
// gpu_traffic += p_values[i];
}
// cpu_traffic = read + write;
// cout << "Read: " << read << " Write: " << write << endl;
// cout << "GPU traffic: " << gpu_traffic << " CPU traffic: " << cpu_traffic << endl;
// #endif
// cgp->resetTime();
// cout << endl;
// cout << endl;
// if (time1 <= time2) {
// time += time1; cpu_traffic += cpu_traffic1; gpu_traffic += gpu_traffic1; pcie_traffic += pcie_traffic1; malloc_time_total += malloc_time_total1;
// } else {
// time += time2; cpu_traffic += cpu_traffic2; gpu_traffic += gpu_traffic2; pcie_traffic += pcie_traffic2; malloc_time_total += malloc_time_total2;
// }
// } else if (input.compare("2") == 0) {
// time = 0; cpu_traffic = 0; gpu_traffic = 0; pcie_traffic = 0; malloc_time_total = 0;
// cout << "How many queries: ";
// cin >> many;
// many_query = stoi(many);
// cgp->resetTime();
// cgp->qo->processed_segment = 0;
// cgp->qo->skipped_segment = 0;
// cout << "Executing Random Query" << endl;
// for (int i = 0; i < many_query; i++) {
// qp->generate_rand_query();
// time1 = qp->processQuery(sessionCtx);
// cpu_traffic1 = cgp->cpu_traffic_total;
// gpu_traffic1 = cgp->gpu_traffic_total;
// pcie_traffic1 = cgp->pcie_traffic_total;
// malloc_time_total1 = cgp->malloc_time_total;
// cgp->resetTime();
// // time2 = qp->processQuery2();
// // cpu_traffic2 = cgp->cpu_traffic_total;
// // gpu_traffic2 = cgp->gpu_traffic_total;
// // pcie_traffic2 = cgp->pcie_traffic_total;
// // malloc_time_total2 = cgp->malloc_time_total;
// // cgp->resetTime();
// // if (time1 <= time2) {
// time += time1; cpu_traffic += cpu_traffic1; gpu_traffic += gpu_traffic1; pcie_traffic += pcie_traffic1; malloc_time_total += malloc_time_total1;
// // } else {
// // time += time2; cpu_traffic += cpu_traffic2; gpu_traffic += gpu_traffic2; pcie_traffic += pcie_traffic2; malloc_time_total += malloc_time_total2;
// // }
// }
// processed_segment = cgp->qo->processed_segment;
// skipped_segment = cgp->qo->skipped_segment;
// srand(123);
// } else if (input.compare("3") == 0) {
// time = 0; cpu_traffic = 0; gpu_traffic = 0; pcie_traffic = 0; malloc_time_total = 0;
// cout << "How many queries: ";
// cin >> many;
// many_query = stoi(many);
// cout << "Replacement Policy: ";
// cin >> policy;
// if (policy == "LRU") {
// repl_policy = LRU;
// } else if (policy == "LFU") {
// repl_policy = LFU;
// } else if (policy == "LRUSegmented") {
// repl_policy = LRUSegmented;
// } else if (policy == "LFUSegmented") {
// repl_policy = LFUSegmented;
// } else if (policy == "LRU2") {
// repl_policy = LRU2;
// } else if (policy == "LRU2Segmented") {
// repl_policy = LRU2Segmented;
// } else if (policy == "SemanticAware") {
// repl_policy = Segmented;
// }
// cuCtxGetCurrent(&curCtx); cout << curCtx << endl; cout <<sessionCtx << endl;
// cgp->resetTime();
// cgp->qo->processed_segment = 0;
// cgp->qo->skipped_segment = 0;
// cout << "Warmup" << endl;
// for (int i = 0; i < 100; i++) {
// qp->generate_rand_query();
// time1 = qp->processQuery(sessionCtx);
// cgp->resetTime();
// }
// cgp->cm->runReplacement(repl_policy);
// cout << "Run Experiment" << endl;
// #ifdef PROFILE
// cout << endl;
// cout << "We are profiling" << endl;
// long long p_values[NUM_EVENTS];
// long long read = 0, write = 0;
// for(int i = 0; i < NUM_EVENTS; i++ ){
// p_values[i] = 0;
// }
// #endif
// for (int iter = 0; iter < 24; iter++) {
// #ifdef PROFILE
// CHECK_CU_ERROR(cuCtxPushCurrent(sessionCtx), "cuCtxPushCurrent");
// retval = PAPI_start( EventSet ); assert(retval == PAPI_OK);
// CHECK_CU_ERROR( cuCtxPopCurrent(&sessionCtx), "cuCtxPopCurrent" );
// // StartMonitor();
// #endif
// for (int i = 0; i < many_query; i++) {
// qp->generate_rand_query();
// time1 = qp->processQuery(sessionCtx);
// cpu_traffic1 = cgp->cpu_traffic_total;
// gpu_traffic1 = cgp->gpu_traffic_total;
// pcie_traffic1 = cgp->pcie_traffic_total;
// malloc_time_total1 = cgp->malloc_time_total;
// cgp->resetTime();
// // time2 = qp->processQuery2();
// // cpu_traffic2 = cgp->cpu_traffic_total;
// // gpu_traffic2 = cgp->gpu_traffic_total;
// // pcie_traffic2 = cgp->pcie_traffic_total;
// // malloc_time_total2 = cgp->malloc_time_total;
// // cgp->resetTime();
// // if (time1 <= time2) {
// time += time1; cpu_traffic += cpu_traffic1; gpu_traffic += gpu_traffic1; pcie_traffic += pcie_traffic1; malloc_time_total += malloc_time_total1;
// // } else {
// // time += time2; cpu_traffic += cpu_traffic2; gpu_traffic += gpu_traffic2; pcie_traffic += pcie_traffic2; malloc_time_total += malloc_time_total2;
// // }
// }
// #ifdef PROFILE
// // EndMonitor(read, write);
// cout << endl;
// CHECK_CU_ERROR(cuCtxPushCurrent(sessionCtx), "cuCtxPushCurrent");
// CHECK_CU_ERROR( cuCtxSynchronize( ), "cuCtxSynchronize" );
// CHECK_CU_ERROR( cuCtxPopCurrent(&sessionCtx), "cuCtxPopCurrent" );
// retval = PAPI_stop( EventSet, p_values ); assert(retval == PAPI_OK);
// for(int i = 0; i < NUM_EVENTS; i++ ) {
// printf( "stop: %12lld \t=0X%016llX \t\t --> %s \n", p_values[i], p_values[i], EventName[i] );
// }
// retval = PAPI_reset(EventSet); assert(retval == PAPI_OK);
// #endif
// cgp->cm->runReplacement(repl_policy);
// qp->percentageData();
// if (dist == Norm && iter % 3 == 0 && iter > 0) {
// mean = ((int) mean + 1) % 7;
// qp->qo->setDistributionNormal(mean, 1);
// }
// }
// #ifdef PROFILE
// gpu_traffic = 0;
// for(int i = 0; i < NUM_EVENTS; i++ ) {
// printf( "stop: %12lld \t=0X%016llX \t\t --> %s \n", p_values[i], p_values[i], EventName[i] );
// gpu_traffic += p_values[i];
// }
// cpu_traffic = read + write;
// cout << "Read: " << read << " Write: " << write << endl;
// cout << "GPU traffic: " << gpu_traffic << " CPU traffic: " << cpu_traffic << endl;
// #endif
// processed_segment = cgp->qo->processed_segment;
// skipped_segment = cgp->qo->skipped_segment;
// srand(123);
// string runs = "logs/runs";
// FILE *fptr = fopen(runs.c_str(), "a");
// if (fptr == NULL)
// {
// printf("Could not open file\n");
// assert(0);
// }
// fprintf(fptr, "{\"cache_size\":%u,\"cumulated_time\":%.2f}\n", size, time);
// fclose(fptr);
// } else if (input.compare("4") == 0) {
// cout << "Replacement Policy: ";
// cin >> policy;
// if (policy == "LRU") {
// repl_policy = LRU;
// } else if (policy == "LFU") {
// repl_policy = LFU;
// } else if (policy == "LRUSegmented") {
// repl_policy = LRUSegmented;
// } else if (policy == "LFUSegmented") {
// repl_policy = LFUSegmented;
// } else if (policy == "LRU2") {
// repl_policy = LRU2;
// } else if (policy == "LRU2Segmented") {
// repl_policy = LRU2Segmented;
// } else if (policy == "SemanticAware") {
// repl_policy = Segmented;
// }
// cgp->cm->runReplacement(repl_policy);
// qp->percentageData();
// srand(123);
// } else if (input.compare("5") == 0) {
// string filename;
// cout << "File name: ";
// cin >> filename;
// qp->dumpTrace("logs/"+filename);
// cout << "Dumped Trace" << endl;
// } else if (input.compare("cache") == 0) {
// string column_name;
// int ret;
// do {
// cout << " Column to cache: ";
// cin >> column_name;
// ret = cgp->cm->cacheSpecificColumn(column_name);
// } while (ret != 0);
// } else if (input.compare("clear") == 0) {
// cgp->cm->deleteAll();
// } else if (input.compare("skipping") == 0) {
// skipping = !skipping;
// cgp->skipping = skipping;
// cgp->qo->skipping = skipping;
// qp->skipping = skipping;
// if (skipping) cout << "Segment skipping is enabled" << endl;
// else cout << "Segment skipping is disabled" << endl;
// } else if (input.compare("custom") == 0) {
// custom = !custom;
// cgp->custom = custom;
// cgp->qo->custom = custom;
// qp->custom = custom;
// if (custom) cout << "Custom malloc is enabled" << endl;
// else cout << "Custom malloc is disabled" << endl;
// } else {
// exit = true;
// }
// cout << endl;
// cout << "Cumulated Time: " << time << endl;
// cout << "CPU traffic: " << cpu_traffic << endl;
// cout << "GPU traffic: " << gpu_traffic << endl;
// cout << "PCIe traffic: " << pcie_traffic << endl;
// cout << "Malloc time: " << malloc_time_total << endl;
// cout << "Fraction Skipped Segment: " << skipped_segment * 1.0 /(processed_segment + skipped_segment) << endl;
// cout << endl;
// // cout << endl;
// }
if (sessionCtx != NULL) {
cuCtxDestroy(sessionCtx);
}
// #ifdef PROFILE
retval = PAPI_cleanup_eventset(EventSet); assert(retval == PAPI_OK);
retval = PAPI_destroy_eventset(&EventSet); assert(retval == PAPI_OK);
// #endif
// #ifdef PROFILE
PAPI_shutdown();
// #endif
}