TestDriver(CkArgMsg* args) {
N = atoi(args->argv[1]);
numElementsPerPe = atoll(args->argv[2]);
localTableSize = (1l << N) / numElementsPerPe;
if (!localTableSize)
CkAbort("Table size is too small, or number of chares is too large\n");
tableSize = localTableSize * CkNumPes() * numElementsPerPe;
CkPrintf("Global table size = 2^%d * %d = %lld words\n",
N, CkNumPes(), tableSize);
CkPrintf("Number of processors = %d\nNumber of updates = %lld\n",
CkNumPes(), 4 * tableSize);
driverProxy = thishandle;
// Create the chares storing and updating the global table
updater_array = CProxy_Updater::ckNew(CkNumPes() * numElementsPerPe);
int dims[2] = {CkNumNodes(), CkNumPes() / CkNumNodes()};
CkPrintf("Aggregation topology: %d %d\n", dims[0], dims[1]);
// Instantiate communication library group with a handle to the client
//aggregator =
// CProxy_ArrayMeshStreamer<dtype, int, Updater, SimpleMeshRouter>
// ::ckNew(numMsgsBuffered, 2, dims, updater_array, 1);
delete args;
}
BFSGraph(CmiUInt8 nVertex, bool directed = true) :
directed(directed),
GraphLib::Graph<
BFSMultiVertex,
BFSEdge,
CProxy_BFSMultiVertex,
GraphLib::TransportType::Charm >(CmiNumPes()) {
int dims[2] = {CkNumNodes(), CkNumPes() / CkNumNodes()};
CkPrintf("Aggregation topology: %d %d\n", dims[0], dims[1]);
aggregator =
CProxy_ArrayMeshStreamer<dtype, long long, BFSMultiVertex, SimpleMeshRouter>
::ckNew(numMsgsBuffered, 2, dims, g, 1);
}
main::main(CkArgMsg *msg)
{
CmiPrintf("Megatest is running on %d nodes %d processors. \n", CkNumNodes(), CkNumPes());
int argc = msg->argc;
char **argv = msg->argv;
int numtests, i;
delete msg;
mainhandle = thishandle;
if (nTests<=0)
CkAbort("Megatest: No tests registered-- is MEGATEST_REGISTER_TEST malfunctioning?");
for (i=0; i<nTests; i++)
(tests[i].initializer)();
test_bank_size = nTests;
next_test_index = 0;
next_test_number = 0;
test_negate_skip=0;
test_repeat = 0;
for (i=1; i<argc; i++) {
if (strcmp(argv[i],"-only")==0)
test_negate_skip = 1;
if (strcmp(argv[i],"-repeat")==0)
test_repeat = 1;
}
num_tests_to_skip = argc;
tests_to_skip = argv;
CProxy_main(thishandle).start();
}
void maindone(void)
{
niter++;
if(niter == START_TRACE_ITER)
_traceControl.startTrace();
if(niter == END_TRACE_ITER)
_traceControl.endTrace();
if(niter == iterations)
{
double pingTimer = CkWallTimer() - start_time;
CkPrintf("Pingping %d\t\t %d \t\t%d \t\t%d \t\t%d \t\t%.1f\n",
nodeIndex, CharesPerPE, PEsPerNode, PEsPerNode * CharesPerPE, payload, 1000*1000*pingTimer/(iterations-WARM_UP));
if(nodeIndex == CkNumNodes() -1)
CkExit();
else
{
niter = 0;
for(int i=0; i<PEsPerNode * CharesPerPE; i++)
arr1[i].start(nodeIndex);
}
nodeIndex++;
}else
{
if(niter == WARM_UP)
start_time = CkWallTimer();
for(int i=0; i<PEsPerNode * CharesPerPE; i++)
arr1[i].start(nodeIndex);
}
};
main(CkArgMsg* m)
{
if(m->argc < 5)
CkPrintf("Usage: payload PEs CharesPerPE iteration\n");
niter = 0;
iterations=NITER;
payload=PAYLOAD;
if(m->argc>1)
totalPayload=atoi(m->argv[1]);
if(m->argc>2)
PEsPerNode = atoi(m->argv[2]);
if(m->argc>3)
CharesPerPE = atoi(m->argv[3]);
if(m->argc>4)
iterations=atoi(m->argv[4]);
payload = totalPayload/PEsPerNode/CharesPerPE;
mainProxy = thishandle;
arr1 = CProxy_Ping1::ckNew(CkNumNodes()* PEsPerNode * CharesPerPE );
start_time = CkWallTimer();
nodeIndex = 1;
int x,y,z,t;
TopoManager tmgr;
for(int i=0; i<CmiNumPes(); i+=CmiMyNodeSize())
{
tmgr.rankToCoordinates(i, x,y, z, t);
CkPrintf(" %d [%d:%d:%d:%d]\n", i, x, y, z, t);
}
CkPrintf("NodeIndex Chares Workers NoOfMsgs Bytes Total Time(us)\n");
_traceControl = CProxy_TraceControl::ckNew();
for(int i=0; i<PEsPerNode * CharesPerPE; i++)
arr1[i].start(nodeIndex);
delete m;
};
int ScriptTcl::Tcl_numNodes(ClientData, Tcl_Interp *interp, int argc, char **) {
if ( argc > 1 ) {
Tcl_SetResult(interp,"no arguments needed",TCL_VOLATILE);
return TCL_ERROR;
}
Tcl_SetObjResult(interp, Tcl_NewIntObj(CkNumNodes()));
return TCL_OK;
}
// Converse immediate handler
void immediatering_startHandler(void *vmsg)
{
const int maxRings = 1000;
immediateMsg *msg=(immediateMsg *)vmsg;
if(0!=strcmp(msg->data,"Array!")) {
CkAbort("Message corrupted");
}
if(CkMyNode()==0)
msg->iter++;
if (msg->iter < maxRings) { /* Keep passing message around the ring */
/* This is questionable: */
sendImmediate((CkMyNode()+1) % CkNumNodes(),msg->iter);
} else /* msg->iter>=maxRings, so stop: */ {
/* megatest_finish isn't safe from an immediate, so send a
regular non-immediate message out: */
int size=CmiMsgHeaderSizeBytes;
void *msg=CmiAlloc(size);
CmiSetHandler(msg,immediatering_finishHandlerIdx);
CmiSyncSendAndFree(0,size,(char *)msg);
}
CmiFree(msg);
}
void immRing_nodegroup::start(immMessage *msg)
{
const int maxRings = 50;
//CkPrintf("[%d] start %d\n", thisIndex, msg->iter);
if(!msg->check()) {
CkError("Message corrupted!\n");
megatest_finish();
return;
}
if(CkMyNode()==0)
msg->iter++;
if (msg->iter < maxRings) {
thisProxy[(CkMyNode()+1) % CkNumNodes()].start(msg);
} else {
delete msg;
//megatest_finish();
int size=CmiMsgHeaderSizeBytes;
void *msg=CmiAlloc(size);
CmiSetHandler(msg,immediatering_finishHandlerIdx);
CmiSyncSendAndFree(0,size,(char *)msg);
}
}
FVectorCache::FVectorCache() {
eps_chares_x = 7;
eps_chares_y = 7;
totalSize = 0;
GWBSE *gwbse = GWBSE::get();
L = gwbse->gw_parallel.L;
int total_eps_chares = eps_chares_x*eps_chares_y;
my_chare_count = total_eps_chares/CkNumNodes();
my_chare_start = CkMyNode()*my_chare_count;
int remaining = total_eps_chares%CkNumNodes();
if(CkMyNode()>0)
my_chare_start += remaining;
if(CkMyNode()==0)
my_chare_count += remaining;
my_eps_chare_indices_x = new int[my_chare_count];
my_eps_chare_indices_y = new int[my_chare_count];
findIndices();
int count = 0;
for(int i=eps_start_chare_x;i<=eps_end_chare_x;i++){
int j = 0;
if(i==eps_start_chare_x)
j = eps_start_chare_y;
int j_end = eps_chares_y-1;
if(i==eps_end_chare_x)
j_end = eps_end_chare_y;
while(j<=j_end){
my_eps_chare_indices_x[count] = i;
my_eps_chare_indices_y[count++] = j;
j++;
}
}
ndata = gwbse->gw_parallel.n_elems;
data_size_x = ndata/eps_chares_x;
if(ndata%eps_chares_x > 0)
data_size_x += 2;
data_size_y = ndata/eps_chares_y;
if(ndata%eps_chares_y > 0)
data_size_y += 2;
data_offset_x = new int[my_chare_count];
data_offset_y = new int[my_chare_count];
for(int i=0;i<my_chare_count;i++){
data_offset_x[i] = my_eps_chare_indices_x[i]*data_size_x;
data_offset_y[i] = my_eps_chare_indices_y[i]*data_size_y;
}
int size_x = data_size_x;
int size_y = data_size_y;
local_offset = new int[my_chare_count*2];
global_offset = new int[my_chare_count*2];
for(int i=0;i<my_chare_count;i++){
global_offset[2*i] = data_offset_x[i];//totalSize;
local_offset[2*i] = totalSize;
totalSize += size_x;
global_offset[2*i+1] = data_offset_y[i];//totalSize;
local_offset[2*i+1] = totalSize;
totalSize += size_y;
}
fs = new complex[NSIZE*L*totalSize];
contribute(CkCallback(CkReductionTarget(Controller,fCacheReady), controller_proxy));
}
Main(CkArgMsg *args) {
if (args->argc >= 3) {
dataSizeMin = atoi(args->argv[1]);
dataSizeMax = atoi(args->argv[2]);
}
else {
dataSizeMin = 32;
dataSizeMax = 16384;
}
bufferSize =
args->argc == 4 ? atoi(args->argv[3]) : TRAM_BUFFER_SIZE;
CkPrintf("size of envelope: %d\n\n", sizeof(envelope));
delete args;
iters = dataSizeMin / DATA_ITEM_SIZE;
allToAllGroup = CProxy_Participant::ckNew();
#if !CMK_BLUEGENEQ
int nDims = 2;
int dims[2] = {CkNumNodes(), CkNumPes() / CkNumNodes()};
CkPrintf("TEST 1: Using %dD TRAM Topology: %d %d\n",
nDims, dims[0], dims[1]);
// Alternative 3D topology
// int nDims = 3;
// int dim1 = CkNumNodes();
// int dim2 = 1;
// if (dim1 != 1) {
// while (dim2 < dim1) {
// dim2 *= 2;
// dim1 /= 2;
// }
// }
// int dims[3] = {dim1, dim2, CkNumPes() / CkNumNodes()};
// CkPrintf("Topology: %d %d %d\n", dims[0], dims[1], dims[2]);
#else
TopoManager tmgr;
int nDims = 3;
int dims[3] = {tmgr.getDimNA() * tmgr.getDimNB(),
tmgr.getDimNC() * tmgr.getDimND() * tmgr.getDimNE(),
tmgr.getDimNT()};
CkPrintf("TEST 1: Using %dD TRAM Topology: %d %d %d\n",
nDims, dims[0], dims[1], dims[2]);
// Alternative TRAM topologies for Blue Gene/Q using Topology Manager
// int nDims = 4;
// int dims[4] = {tmgr.getDimNA() * tmgr.getDimNB(), tmgr.getDimNC(),
// tmgr.getDimND() * tmgr.getDimNE(), tmgr.getDimNT()};
// int nDims = 6;
// int dims[6] = {tmgr.getDimNA(), tmgr.getDimNB(), tmgr.getDimNC(),
// tmgr.getDimND() * tmgr.getDimNE(),
// tmgr.getDimNT() / 8, 8};
#endif
mainProxy = thisProxy;
aggregator = CProxy_GroupMeshStreamer<DataItem, Participant,
SimpleMeshRouter>::
ckNew(nDims, dims, allToAllGroup, bufferSize, 1, 0.1);
testType = usingTram;
}
