void SetPrint(int argcIn, const char* argvIn[]) {
if( argcIn == 1 || (argcIn == 2 && argvIn[1][0] == 'm') )
{
cout << endl;
PrintSteps(true);
PrintSummary(true);
}
else if( argcIn > 1 && argcIn < 4 )
{
if( argvIn[1][0] == '-' && (argvIn[1][1] == 'n' || argvIn[1][9] == 'n') )
{
PrintSteps(false);
PrintSummary(false);
}
else if( argvIn[1][0] == '-' && (argvIn[1][1] == 's' || argvIn[1][9] == 's') )
{
PrintSteps(true);
PrintSummary(false);
}
else if( argvIn[1][0] == '-' && (argvIn[1][1] == 'a' || argvIn[1][9] == 'a') )
{
PrintSteps(false);
PrintSummary(true);
}
cout << endl;
}
else
{
PrintSteps(false);
PrintSummary(false);
}
}
void SingleEvent ()
{
real vvSum;
real sp;
int n;
NextEvent ();
if (evIdB < MOL_LIMIT) {
ProcessCollision ();
++ collCount;
} else if (evIdB < MOL_LIMIT + NDIM * 2 || evIdB >= MOL_LIMIT + 100) {
ProcessCellCrossing ();
++ crossCount;
} else if (evIdB == MOL_LIMIT + 6) {
UpdateSystem ();
nextSumTime += intervalSum;
ScheduleEvent (0, MOL_LIMIT + 6, nextSumTime);
VZero (vSum);
vvSum = 0.;
sp = 0.;
DO_MOL {
VVAdd (vSum, mol[n].rv);
vvSum += VLenSq (mol[n].rv);
sp += VDot (mol[n].r, gravField);
}
kinEnVal = vvSum * 0.5 / nMol;
totEnVal = kinEnVal - sp / nMol;
PrintSummary (stdout);
} else if (evIdB == MOL_LIMIT + 7) {
void SingleStep ()
{
++ stepCount;
timeNow = stepCount * deltaT;
PredictorStep ();
PredictorStepQ ();
GenSiteCoords ();
ComputeSiteForces ();
ComputeTorqs ();
ComputeAccelsQ ();
ApplyThermostat ();
CorrectorStep ();
CorrectorStepQ ();
AdjustQuat ();
ApplyBoundaryCond ();
EvalProps ();
if (stepCount % stepAdjustTemp == 0) AdjustTemp ();
AccumProps (1);
if (stepCount % stepAvg == 0) {
AccumProps (2);
PrintSummary (stdout);
AccumProps (0);
}
if (stepCount >= stepEquil && (stepCount - stepEquil) % stepRdf == 0)
EvalRdf ();
}
int myGetState(void * hndl, UserState * ustate, unsigned int privdata)
{
static int iter=0;
log_verbose("Reached myGetState with privdata %x\n", privdata);
/* Same state is being returned for both engines */
ustate->LinkState = LINK_UP;
ustate->Errors = 0;
ustate->MinPktSize = RawMinPktSize;
ustate->MaxPktSize = RawMaxPktSize;
ustate->TestMode = RawTestMode;
if(privdata == 0x54545454)
ustate->Buffers = TxBufs.TotalNum;
else
ustate->Buffers = RxBufs.TotalNum;
if(iter++ >= 4)
{
PrintSummary();
iter = 0;
}
return 0;
}
int main (int argc, char **argv)
{
int i, nocase = 0;
FILE *fd;
char filename[20];
ACSM_STRUCT * acsm;
if (argc < 3)
{
fprintf (stderr,"Usage: acsmx filename pattern1 pattern2 ... -nocase\n");
exit (0);
}
acsm = acsmNew ();
strcpy (filename, argv[1]);
fd = fopen(filename,"r");
if(fd == NULL)
{
fprintf(stderr,"Open file error!\n");
exit(1);
}
for (i = 1; i < argc; i++)
if (strcmp (argv[i], "-nocase") == 0)
nocase = 1;
for (i = 2; i < argc; i++)
{
if (argv[i][0] == '-')
continue;
printf("%s,%d\n",argv[i],strlen (argv[i]));
acsmAddPattern (acsm, argv[i], strlen (argv[i]), nocase,1);
}
fgets(text,MAXLEN,fd);
/* Generate GtoTo Table and Fail Table */
acsmCompile (acsm);
printf("--------------------------------\n");
NS_TIME(time);
NS_TIME_START(time);
/*Search Pattern*/
//while ( fgets(text,MAXLEN,fd) )
//{
acsmSearch (acsm, text, strlen (text), PrintMatch);
// nline++;
//}
NS_TIME_END(time);
PrintSummary(acsm->acsmPatterns);
int a[10]={45,45,45,4,1};
#ifdef __HAVE__LOAD__
printf("-------%d\n", getSummary (acsm->acsmPatterns,a));
#endif
acsmFree (acsm);
printf ("\n### AC Match Finished ###\n");
// system("pause");
return (0);
}
void SingleEvent ()
{
real vvSum;
int n;
NextEvent ();
if (evIdB < MOL_LIMIT) {
ProcessCollision ();
EvalFreePath ();
++ collCount;
} else if (evIdB >= MOL_LIMIT + 100) {
ProcessCellCrossing ();
++ crossCount;
} else if (evIdB == MOL_LIMIT + 6) {
UpdateSystem ();
nextSumTime += intervalSum;
ScheduleEvent (0, MOL_LIMIT + 6, nextSumTime);
VZero (vSum);
vvSum = 0.;
DO_MOL {
VVAdd (vSum, mol[n].rv);
vvSum += VLenSq (mol[n].rv);
}
kinEnVal = vvSum * 0.5 / nMol;
PrintSummary (stdout);
}
void PrintQueryResult(int ego, char **caption, char **result, int row, int col) {
if (ego)
VPrint(caption, result, row, col);
else
HPrint(caption, result, row, col);
PrintSummary(row);
}
void SingleStep ()
{
++ stepCount;
timeNow = stepCount * deltaT;
PredictorStep ();
PredictorStepS ();
ComputeForces ();
ComputeForcesDipoleR ();
ComputeForcesDipoleF ();
ComputeDipoleAccel ();
ApplyThermostat ();
CorrectorStep ();
CorrectorStepS ();
AdjustDipole ();
ApplyBoundaryCond ();
EvalProps ();
if (stepCount % stepAdjustTemp == 0) AdjustTemp ();
AccumProps (1);
if (stepCount % stepAvg == 0) {
AccumProps (2);
PrintSummary (stdout);
AccumProps (0);
}
if (stepCount >= stepEquil && (stepCount - stepEquil) % stepRdf == 0)
EvalRdf ();
}
static void __exit rawdata_cleanup(void)
{
int i;
/* Stop the polling routine */
del_timer_sync(&poll_timer);
//DriverState = CLOSED;
/* Stop any running tests, else the hardware's packet checker &
* generator will continue to run.
*/
XIo_Out32(TXbarbase+TX_CONFIG_ADDRESS, 0);
XIo_Out32(TXbarbase+RX_CONFIG_ADDRESS, 0);
printk(KERN_INFO "%s: Unregistering Xilinx driver from kernel.\n", MYNAME);
if (TxBufCnt != RxBufCnt)
{
printk("%s: Buffers Transmitted %u Received %u\n", MYNAME, TxBufCnt, RxBufCnt);
printk("TxSeqNo = %u, RxSeqNo = %u\n", TxSeqNo, RxSeqNo);
mdelay(1);
}
#ifdef FIFO_EMPTY_CHECK
DmaFifoEmptyWait(MYHANDLE,DIR_TYPE_S2C);
// wait for appropriate time to stabalize
mdelay(STABILITY_WAIT_TIME);
#endif
DmaUnregister(handle[0]);
#ifdef FIFO_EMPTY_CHECK
DmaFifoEmptyWait(MYHANDLE,DIR_TYPE_C2S);
// wait for appropriate time to stabalize
mdelay(STABILITY_WAIT_TIME);
#endif
DmaUnregister(handle[2]);
PrintSummary();
mdelay(1000);
/* Not sure if free_page() sleeps or not. */
spin_lock_bh(&RawLock);
printk("Freeing user buffers\n");
for(i=0; i<TxBufs.TotalNum; i++)
//kfree(TxBufs.origVA[i]);
free_page((unsigned long)(TxBufs.origVA[i]));
for(i=0; i<RxBufs.TotalNum; i++)
//kfree(RxBufs.origVA[i]);
free_page((unsigned long)(RxBufs.origVA[i]));
spin_unlock_bh(&RawLock);
}
void MCommandProfiler::Analysis()
{
FILE* fp = fopen("CmdProfile.txt", "wt");
if (fp == 0) return;
PrintTitle(fp);
PrintSummary(fp);
PrintCmdItems(fp);
fclose(fp);
Reset();
}
int main (int argc, char **argv)
{
GetNameList (argc, argv);
//PrintNameList (stdout);
SetParams ();
SetupJob ();
moreCycles = 1;
PrintSummary (stdout);
while (moreCycles) {
SingleStep ();
if (stepCount >= stepLimit) moreCycles = 0;
}
}
int main(int argc, char **argv)
{
int i, nocase = 0;
FILE *fd;
char filename[20];
ACSM_STRUCT * acsm;
// if (argc < 3)
// {
// fprintf(stderr, "Usage: acsmx filename pattern1 pattern2 ... -nocase\n");
// exit(0);
// }
acsm = acsmNew();
strcpy_s(filename, "test.txt");
fd = fopen(filename, "r");
if (fd == NULL)
{
fprintf(stderr, "Open file error!\n");
exit(1);
}
for (i = 1; i < argc; i++)
if (strcmp(argv[i], "-nocase") == 0)
nocase = 1;
char a[] = { "test" };
acsmAddPattern(acsm, (unsigned char *)a, strlen(a), nocase);
/* Generate GtoTo Table and Fail Table */
acsmCompile(acsm);
/*Search Pattern*/
while (fgets(( char*)text, MAXLEN, fd))
{
acsmSearch(acsm, text, strlen((char*)text), PrintMatch);
nline++;
}
PrintSummary(acsm->acsmPatterns);
acsmFree(acsm);
printf("\n### AC Match Finished ###\n");
system("pause");
return (0);
}
void SingleStep ()
{
++ stepCount;
timeNow = stepCount * deltaT;
LeapfrogStep (1);
ApplyBoundaryCond ();
ComputeForces ();
LeapfrogStep (2);
//EvalProps ();
//AccumProps (1);
if (stepCount % stepAvg == 0) {
//AccumProps (2);
PrintSummary (stdout);
//PrintVels (stdout);
//AccumProps (0);
}
}
static void __exit rawdata_cleanup(void)
{
int i;
/* Stop the polling routine */
del_timer_sync(&poll_timer);
//DriverState = CLOSED;
/* Stop any running tests, else the hardware's packet checker &
* generator will continue to run.
*/
log_verbose("TXbarbase = %p\n", TXbarbase);
XIo_Out32(TXbarbase+TX_CONFIG_ADDRESS, 0);
#ifndef XAUI
XIo_Out32(TXbarbase+RX_CONFIG_ADDRESS, 0);
#endif
printk(KERN_INFO "%s: Unregistering Xilinx driver from kernel.\n", MYNAME);
if (TxBufCnt != RxBufCnt)
{
printk("%s: Buffers Transmitted %u Received %u\n", MYNAME, TxBufCnt, RxBufCnt);
printk("TxSeqNo = %u, RxSeqNo = %u\n", TxSeqNo, RxSeqNo);
mdelay(1);
}
DmaUnregister(handle[0]);
DmaUnregister(handle[2]);
PrintSummary();
mdelay(1000);
/* Not sure if free_page() sleeps or not. */
spin_lock_bh(&RawLock);
printk("Freeing user buffers\n");
for(i=0; i<TxBufs.TotalNum; i++)
//kfree(TxBufs.origVA[i]);
free_page((unsigned long)(TxBufs.origVA[i]));
for(i=0; i<RxBufs.TotalNum; i++)
//kfree(RxBufs.origVA[i]);
free_page((unsigned long)(RxBufs.origVA[i]));
spin_unlock_bh(&RawLock);
}
int main(int argc, char* argv[])
{
int attrib_ct, line_ct;
int k; //loop counters
ListP *Attributes;
//open files
getFileHandles(argc, argv);
//get attributes and line counters
attrib_ct = readValue();
line_ct = readValue();
getc(inF); //eat newline
//Make array of attributes
Attributes = makeArrayofList(attrib_ct);
//read file
for (k = 0; k < line_ct; k++)
readAttributes(Attributes, attrib_ct);
PrintSummary(Attributes, attrib_ct);
}
void SingleStep ()
{
++ stepCount;
timeNow = stepCount * deltaT;
LeapfrogStep (1);
GenSiteCoords ();
ComputeSiteForces ();
ComputeTorqs ();
ApplyThermostat ();
LeapfrogStep (2);
ApplyBoundaryCond ();
EvalProps ();
if (stepCount % stepAdjustTemp == 0 || stepCount < stepEquil &&
stepCount % 100 == 0) AdjustTemp ();
AccumProps (1);
if (stepCount % stepAvg == 0) {
AccumProps (2);
PrintSummary (stdout);
AccumProps (0);
}
}
STATUS LNPUBLIC ReadSummaryData
(VOID far *optional_param,
SEARCH_MATCH far *search_info,
ITEM_TABLE far *summary_info)
{
SEARCH_MATCH SearchMatch;
STATUS error;
memcpy ((char*)(&SearchMatch), (char *)search_info, sizeof(SEARCH_MATCH));
if (!(SearchMatch.SERetFlags & SE_FMATCH))
return (NOERROR);
/* Print the note ID. */
printf ("\nNote ID is: %#010lx.\n", SearchMatch.ID.NoteID);
/* Print the summary data. */
if (error = PrintSummary( (char*)summary_info ))
return (error);
return (NOERROR);
}
////////////////////////////////////////////////////////////////////////////////
// Program Main
////////////////////////////////////////////////////////////////////////////////
int main(int argc, char *argv[])
{
int Nx, Ny, Nz, max_iters;
int blockX, blockY, blockZ;
if (argc == 8) {
Nx = atoi(argv[1]);
Ny = atoi(argv[2]);
Nz = atoi(argv[3]);
max_iters = atoi(argv[4]);
blockX = atoi(argv[5]);
blockY = atoi(argv[6]);
blockZ = atoi(argv[7]);
}
else
{
printf("Usage: %s nx ny nz i block_x block_y block_z number_of_threads\n",
argv[0]);
exit(1);
}
// Get the number of GPUS
int number_of_devices;
checkCuda(cudaGetDeviceCount(&number_of_devices));
if (number_of_devices < 2) {
printf("Less than two devices were found.\n");
printf("Exiting...\n");
return -1;
}
// Decompose along the Z-axis
int _Nz = Nz/number_of_devices;
// Define constants
const _DOUBLE_ L = 1.0;
const _DOUBLE_ h = L/(Nx+1);
const _DOUBLE_ dt = h*h/6.0;
const _DOUBLE_ beta = dt/(h*h);
const _DOUBLE_ c0 = beta;
const _DOUBLE_ c1 = (1-6*beta);
// Check if ECC is turned on
ECCCheck(number_of_devices);
// Set the number of OpenMP threads
omp_set_num_threads(number_of_devices);
#pragma omp parallel
{
unsigned int tid = omp_get_num_threads();
#pragma omp single
{
printf("Number of OpenMP threads: %d\n", tid);
}
}
// CPU memory operations
int dt_size = sizeof(_DOUBLE_);
_DOUBLE_ *u_new, *u_old;
u_new = (_DOUBLE_ *)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(Nz+2));
u_old = (_DOUBLE_ *)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(Nz+2));
init(u_old, u_new, h, Nx, Ny, Nz);
// Allocate and generate arrays on the host
size_t pitch_bytes;
size_t pitch_gc_bytes;
_DOUBLE_ *h_Unew, *h_Uold;
_DOUBLE_ *h_s_Uolds[number_of_devices], *h_s_Unews[number_of_devices];
_DOUBLE_ *left_send_buffer[number_of_devices], *left_receive_buffer[number_of_devices];
_DOUBLE_ *right_send_buffer[number_of_devices], *right_receive_buffer[number_of_devices];
h_Unew = (_DOUBLE_ *)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(Nz+2));
h_Uold = (_DOUBLE_ *)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(Nz+2));
init(h_Uold, h_Unew, h, Nx, Ny, Nz);
#pragma omp parallel
{
unsigned int tid = omp_get_thread_num();
h_s_Unews[tid] = (_DOUBLE_ *)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(_Nz+2));
h_s_Uolds[tid] = (_DOUBLE_ *)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(_Nz+2));
right_send_buffer[tid] = (_DOUBLE_ *)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(_GC_DEPTH));
left_send_buffer[tid] = (_DOUBLE_ *)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(_GC_DEPTH));
right_receive_buffer[tid] = (_DOUBLE_ *)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(_GC_DEPTH));
left_receive_buffer[tid] = (_DOUBLE_ *)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(_GC_DEPTH));
checkCuda(cudaHostAlloc((void**)&h_s_Unews[tid], dt_size*(Nx+2)*(Ny+2)*(_Nz+2), cudaHostAllocPortable));
checkCuda(cudaHostAlloc((void**)&h_s_Uolds[tid], dt_size*(Nx+2)*(Ny+2)*(_Nz+2), cudaHostAllocPortable));
checkCuda(cudaHostAlloc((void**)&right_send_buffer[tid], dt_size*(Nx+2)*(Ny+2)*(_GC_DEPTH), cudaHostAllocPortable));
checkCuda(cudaHostAlloc((void**)&left_send_buffer[tid], dt_size*(Nx+2)*(Ny+2)*(_GC_DEPTH), cudaHostAllocPortable));
checkCuda(cudaHostAlloc((void**)&right_receive_buffer[tid], dt_size*(Nx+2)*(Ny+2)*(_GC_DEPTH), cudaHostAllocPortable));
checkCuda(cudaHostAlloc((void**)&left_receive_buffer[tid], dt_size*(Nx+2)*(Ny+2)*(_GC_DEPTH), cudaHostAllocPortable));
init_subdomain(h_s_Uolds[tid], h_Uold, Nx, Ny, _Nz, tid);
}
// GPU memory operations
_DOUBLE_ *d_s_Unews[number_of_devices], *d_s_Uolds[number_of_devices];
_DOUBLE_ *d_right_send_buffer[number_of_devices], *d_left_send_buffer[number_of_devices];
_DOUBLE_ *d_right_receive_buffer[number_of_devices], *d_left_receive_buffer[number_of_devices];
#pragma omp parallel
{
unsigned int tid = omp_get_thread_num();
checkCuda(cudaSetDevice(tid));
CopyToConstantMemory(c0, c1);
checkCuda(cudaMallocPitch((void**)&d_s_Uolds[tid], &pitch_bytes, dt_size*(Nx+2), (Ny+2)*(_Nz+2)));
checkCuda(cudaMallocPitch((void**)&d_s_Unews[tid], &pitch_bytes, dt_size*(Nx+2), (Ny+2)*(_Nz+2)));
checkCuda(cudaMallocPitch((void**)&d_left_receive_buffer[tid], &pitch_gc_bytes, dt_size*(Nx+2), (Ny+2)*(_GC_DEPTH)));
checkCuda(cudaMallocPitch((void**)&d_right_receive_buffer[tid], &pitch_gc_bytes, dt_size*(Nx+2), (Ny+2)*(_GC_DEPTH)));
checkCuda(cudaMallocPitch((void**)&d_left_send_buffer[tid], &pitch_gc_bytes, dt_size*(Nx+2), (Ny+2)*(_GC_DEPTH)));
checkCuda(cudaMallocPitch((void**)&d_right_send_buffer[tid], &pitch_gc_bytes, dt_size*(Nx+2), (Ny+2)*(_GC_DEPTH)));
}
// Copy data from host to the device
double HtD_timer = 0.;
HtD_timer -= omp_get_wtime();
#pragma omp parallel
{
unsigned int tid = omp_get_thread_num();
checkCuda(cudaSetDevice(tid));
checkCuda(cudaMemcpy2D(d_s_Uolds[tid], pitch_bytes, h_s_Uolds[tid], dt_size*(Nx+2), dt_size*(Nx+2), ((Ny+2)*(_Nz+2)), cudaMemcpyDefault));
checkCuda(cudaMemcpy2D(d_s_Unews[tid], pitch_bytes, h_s_Unews[tid], dt_size*(Nx+2), dt_size*(Nx+2), ((Ny+2)*(_Nz+2)), cudaMemcpyDefault));
}
HtD_timer += omp_get_wtime();
int pitch = pitch_bytes/dt_size;
int gc_pitch = pitch_gc_bytes/dt_size;
// GPU kernel launch parameters
dim3 threads_per_block(blockX, blockY, blockZ);
unsigned int blocksInX = getBlock(Nx, blockX);
unsigned int blocksInY = getBlock(Ny, blockY);
unsigned int blocksInZ = getBlock(_Nz-2, k_loop);
dim3 thread_blocks(blocksInX, blocksInY, blocksInZ);
dim3 thread_blocks_halo(blocksInX, blocksInY);
double compute_timer = 0.;
compute_timer -= omp_get_wtime();
#pragma omp parallel
{
unsigned int tid = omp_get_thread_num();
for(int iterations = 0; iterations < max_iters; iterations++)
{
// Compute inner nodes
checkCuda(cudaSetDevice(tid));
ComputeInnerPoints(thread_blocks, threads_per_block, d_s_Unews[tid], d_s_Uolds[tid], pitch, Nx, Ny, _Nz);
// Copy right boundary data to host
if (tid == 0)
{
checkCuda(cudaSetDevice(tid));
CopyBoundaryRegionToGhostCell(thread_blocks_halo, threads_per_block, d_s_Unews[tid], d_right_send_buffer[tid], Nx, Ny, _Nz, pitch, gc_pitch, 0);
checkCuda(cudaMemcpy2D(right_send_buffer[tid], dt_size*(Nx+2), d_right_send_buffer[tid], pitch_gc_bytes, dt_size*(Nx+2), (Ny+2)*(_GC_DEPTH), cudaMemcpyDefault));
}
// Copy left boundary data to host
if (tid == 1)
{
checkCuda(cudaSetDevice(tid));
CopyBoundaryRegionToGhostCell(thread_blocks_halo, threads_per_block, d_s_Unews[tid], d_left_send_buffer[tid], Nx, Ny, _Nz, pitch, gc_pitch, 1);
checkCuda(cudaMemcpy2D(left_send_buffer[tid], dt_size*(Nx+2), d_left_send_buffer[tid], pitch_gc_bytes, dt_size*(Nx+2), (Ny+2)*(_GC_DEPTH), cudaMemcpyDefault));
}
#pragma omp barrier
// Copy right boundary data to device 1
if (tid == 1)
{
checkCuda(cudaSetDevice(tid));
checkCuda(cudaMemcpy2D(d_left_receive_buffer[tid], pitch_gc_bytes, right_send_buffer[tid-1], dt_size*(Nx+2), dt_size*(Nx+2), ((Ny+2)*(_GC_DEPTH)), cudaMemcpyDefault));
CopyGhostCellToBoundaryRegion(thread_blocks_halo, threads_per_block, d_s_Unews[tid], d_left_receive_buffer[tid], Nx, Ny, _Nz, pitch, gc_pitch, 1);
}
// Copy left boundary data to device 0
if (tid == 0)
{
checkCuda(cudaSetDevice(tid));
checkCuda(cudaMemcpy2D(d_right_receive_buffer[tid], pitch_gc_bytes, left_send_buffer[tid+1], dt_size*(Nx+2), dt_size*(Nx+2), ((Ny+2)*(_GC_DEPTH)), cudaMemcpyDefault));
CopyGhostCellToBoundaryRegion(thread_blocks_halo, threads_per_block, d_s_Unews[tid], d_right_receive_buffer[tid], Nx, Ny, _Nz, pitch, gc_pitch, 0);
}
// Swap pointers on the host
#pragma omp barrier
checkCuda(cudaSetDevice(tid));
checkCuda(cudaDeviceSynchronize());
swap(_DOUBLE_*, d_s_Unews[tid], d_s_Uolds[tid]);
}
}
compute_timer += omp_get_wtime();
// Copy data from device to host
double DtH_timer = 0;
DtH_timer -= omp_get_wtime();
#pragma omp parallel
{
unsigned int tid = omp_get_thread_num();
checkCuda(cudaSetDevice(tid));
checkCuda(cudaMemcpy2D(h_s_Uolds[tid], dt_size*(Nx+2), d_s_Uolds[tid], pitch_bytes, dt_size*(Nx+2), (Ny+2)*(_Nz+2), cudaMemcpyDeviceToHost));
}
DtH_timer += omp_get_wtime();
// Merge sub-domains into a one big domain
#pragma omp parallel
{
unsigned int tid = omp_get_thread_num();
merge_domains(h_s_Uolds[tid], h_Uold, Nx, Ny, _Nz, tid);
}
// Calculate on host
#if defined(DEBUG) || defined(_DEBUG)
cpu_heat3D(u_new, u_old, c0, c1, max_iters, Nx, Ny, Nz);
#endif
float gflops = CalcGflops(compute_timer, max_iters, Nx, Ny, Nz);
PrintSummary("3D Heat (7-pt)", "Plane sweeping", compute_timer, HtD_timer, DtH_timer, gflops, max_iters, Nx);
_DOUBLE_ t = max_iters * dt;
CalcError(h_Uold, u_old, t, h, Nx, Ny, Nz);
#if defined(DEBUG) || defined(_DEBUG)
//exportToVTK(h_Uold, h, "heat3D.vtk", Nx, Ny, Nz);
#endif
#pragma omp parallel
{
unsigned int tid = omp_get_thread_num();
checkCuda(cudaSetDevice(tid));
checkCuda(cudaFree(d_s_Unews[tid]));
checkCuda(cudaFree(d_s_Uolds[tid]));
checkCuda(cudaFree(d_right_send_buffer[tid]));
checkCuda(cudaFree(d_left_send_buffer[tid]));
checkCuda(cudaFree(d_right_receive_buffer[tid]));
checkCuda(cudaFree(d_left_receive_buffer[tid]));
checkCuda(cudaFreeHost(h_s_Unews[tid]));
checkCuda(cudaFreeHost(h_s_Uolds[tid]));
checkCuda(cudaFreeHost(left_send_buffer[tid]));
checkCuda(cudaFreeHost(right_send_buffer[tid]));
checkCuda(cudaFreeHost(left_receive_buffer[tid]));
checkCuda(cudaFreeHost(right_receive_buffer[tid]));
checkCuda(cudaDeviceReset());
}
free(u_old);
free(u_new);
return 0;
}
///////////////////////
// Main program entry
///////////////////////
int main(int argc, char** argv)
{
unsigned int max_iters, Nx, Ny, Nz, blockX, blockY, blockZ;
int rank, numberOfProcesses;
if (argc == 8)
{
Nx = atoi(argv[1]);
Ny = atoi(argv[2]);
Nz = atoi(argv[3]);
max_iters = atoi(argv[4]);
blockX = atoi(argv[5]);
blockY = atoi(argv[6]);
blockZ = atoi(argv[7]);
}
else
{
printf("Usage: %s nx ny nz i block_x block_y block_z\n", argv[0]);
exit(1);
}
InitializeMPI(&argc, &argv, &rank, &numberOfProcesses);
AssignDevices(rank);
ECCCheck(rank);
// Define constants
const _DOUBLE_ L = 1.0;
const _DOUBLE_ h = L/(Nx+1);
const _DOUBLE_ dt = h*h/6.0;
const _DOUBLE_ beta = dt/(h*h);
const _DOUBLE_ c0 = beta;
const _DOUBLE_ c1 = (1-6*beta);
// Copy constants to Constant Memory on the GPUs
CopyToConstantMemory(c0, c1);
// Decompose along the z-axis
const int _Nz = Nz/numberOfProcesses;
const int dt_size = sizeof(_DOUBLE_);
// Host memory allocations
_DOUBLE_ *u_new, *u_old;
_DOUBLE_ *h_Uold;
u_new = (_DOUBLE_*)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(Nz+2));
u_old = (_DOUBLE_*)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(Nz+2));
if (rank == 0)
{
h_Uold = (_DOUBLE_*)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(Nz+2));
}
init(u_old, u_new, h, Nx, Ny, Nz);
// Allocate and generate host subdomains
_DOUBLE_ *h_s_Uolds, *h_s_Unews, *h_s_rbuf[numberOfProcesses];
_DOUBLE_ *left_send_buffer, *left_receive_buffer;
_DOUBLE_ *right_send_buffer, *right_receive_buffer;
h_s_Unews = (_DOUBLE_*)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(_Nz+2));
h_s_Uolds = (_DOUBLE_*)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(_Nz+2));
#if defined(DEBUG) || defined(_DEBUG)
if (rank == 0)
{
for (int i = 0; i < numberOfProcesses; i++)
{
h_s_rbuf[i] = (_DOUBLE_*)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(_Nz+2));
checkCuda(cudaHostAlloc((void**)&h_s_rbuf[i], dt_size*(Nx+2)*(Ny+2)*(_Nz+2), cudaHostAllocPortable));
}
}
#endif
right_send_buffer = (_DOUBLE_*)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(_GC_DEPTH));
left_send_buffer = (_DOUBLE_*)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(_GC_DEPTH));
right_receive_buffer = (_DOUBLE_*)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(_GC_DEPTH));
left_receive_buffer = (_DOUBLE_*)malloc(sizeof(_DOUBLE_)*(Nx+2)*(Ny+2)*(_GC_DEPTH));
checkCuda(cudaHostAlloc((void**)&h_s_Unews, dt_size*(Nx+2)*(Ny+2)*(_Nz+2), cudaHostAllocPortable));
checkCuda(cudaHostAlloc((void**)&h_s_Uolds, dt_size*(Nx+2)*(Ny+2)*(_Nz+2), cudaHostAllocPortable));
checkCuda(cudaHostAlloc((void**)&right_send_buffer, dt_size*(Nx+2)*(Ny+2)*(_GC_DEPTH), cudaHostAllocPortable));
checkCuda(cudaHostAlloc((void**)&left_send_buffer, dt_size*(Nx+2)*(Ny+2)*(_GC_DEPTH), cudaHostAllocPortable));
checkCuda(cudaHostAlloc((void**)&right_receive_buffer, dt_size*(Nx+2)*(Ny+2)*(_GC_DEPTH), cudaHostAllocPortable));
checkCuda(cudaHostAlloc((void**)&left_receive_buffer, dt_size*(Nx+2)*(Ny+2)*(_GC_DEPTH), cudaHostAllocPortable));
init_subdomain(h_s_Uolds, u_old, Nx, Ny, _Nz, rank);
// GPU stream operations
cudaStream_t compute_stream;
cudaStream_t data_stream;
checkCuda(cudaStreamCreate(&compute_stream));
checkCuda(cudaStreamCreate(&data_stream));
// GPU Memory Operations
size_t pitch_bytes, pitch_gc_bytes;
_DOUBLE_ *d_s_Unews, *d_s_Uolds;
_DOUBLE_ *d_right_send_buffer, *d_left_send_buffer;
_DOUBLE_ *d_right_receive_buffer, *d_left_receive_buffer;
checkCuda(cudaMallocPitch((void**)&d_s_Uolds, &pitch_bytes, dt_size*(Nx+2), (Ny+2)*(_Nz+2)));
checkCuda(cudaMallocPitch((void**)&d_s_Unews, &pitch_bytes, dt_size*(Nx+2), (Ny+2)*(_Nz+2)));
checkCuda(cudaMallocPitch((void**)&d_left_send_buffer, &pitch_gc_bytes, dt_size*(Nx+2), (Ny+2)*(_GC_DEPTH)));
checkCuda(cudaMallocPitch((void**)&d_left_receive_buffer, &pitch_gc_bytes, dt_size*(Nx+2), (Ny+2)*(_GC_DEPTH)));
checkCuda(cudaMallocPitch((void**)&d_right_send_buffer, &pitch_gc_bytes, dt_size*(Nx+2), (Ny+2)*(_GC_DEPTH)));
checkCuda(cudaMallocPitch((void**)&d_right_receive_buffer, &pitch_gc_bytes, dt_size*(Nx+2), (Ny+2)*(_GC_DEPTH)));
// Copy subdomains from host to device and get walltime
double HtD_timer = 0.;
MPI_CHECK(MPI_Barrier(MPI_COMM_WORLD));
HtD_timer -= MPI_Wtime();
MPI_CHECK(MPI_Barrier(MPI_COMM_WORLD));
checkCuda(cudaMemcpy2D(d_s_Uolds, pitch_bytes, h_s_Uolds, dt_size*(Nx+2), dt_size*(Nx+2), ((Ny+2)*(_Nz+2)), cudaMemcpyDefault));
checkCuda(cudaMemcpy2D(d_s_Unews, pitch_bytes, h_s_Unews, dt_size*(Nx+2), dt_size*(Nx+2), ((Ny+2)*(_Nz+2)), cudaMemcpyDefault));
MPI_CHECK(MPI_Barrier(MPI_COMM_WORLD));
HtD_timer += MPI_Wtime();
MPI_CHECK(MPI_Barrier(MPI_COMM_WORLD));
unsigned int ghost_width = 1;
int pitch = pitch_bytes/dt_size;
int gc_pitch = pitch_gc_bytes/dt_size;
// GPU kernel launch parameters
dim3 threads_per_block(blockX, blockY, blockZ);
unsigned int blocksInX = getBlock(Nx, blockX);
unsigned int blocksInY = getBlock(Ny, blockY);
unsigned int blocksInZ = getBlock(_Nz-2, k_loop);
dim3 thread_blocks(blocksInX, blocksInY, blocksInZ);
dim3 thread_blocks_halo(blocksInX, blocksInY);
//MPI_Status status;
MPI_Status status[numberOfProcesses];
MPI_Request gather_send_request[numberOfProcesses];
MPI_Request right_send_request[numberOfProcesses], left_send_request[numberOfProcesses];
MPI_Request right_receive_request[numberOfProcesses], left_receive_request[numberOfProcesses];
double compute_timer = 0.;
MPI_CHECK(MPI_Barrier(MPI_COMM_WORLD));
compute_timer -= MPI_Wtime();
MPI_CHECK(MPI_Barrier(MPI_COMM_WORLD));
for(unsigned int iterations = 0; iterations < max_iters; iterations++)
{
// Compute right boundary data on device 0
if (rank == 0) {
int kstart = (_Nz+1)-ghost_width;
int kstop = _Nz+1;
ComputeInnerPointsAsync(thread_blocks_halo, threads_per_block, data_stream, d_s_Unews, d_s_Uolds, pitch, Nx, Ny, _Nz, kstart, kstop);
CopyBoundaryRegionToGhostCellAsync(thread_blocks_halo, threads_per_block, data_stream, d_s_Unews, d_right_send_buffer, Nx, Ny, _Nz, pitch, gc_pitch, 0);
checkCuda(cudaMemcpy2DAsync(right_send_buffer, dt_size*(Nx+2), d_right_send_buffer, pitch_gc_bytes, dt_size*(Nx+2), (Ny+2)*(_GC_DEPTH), cudaMemcpyDefault, data_stream));
checkCuda(cudaStreamSynchronize(data_stream));
MPI_CHECK(MPI_Isend(right_send_buffer, (Nx+2)*(Ny+2)*(_GC_DEPTH), MPI_DOUBLE, 1, 0, MPI_COMM_WORLD, &right_send_request[rank]));
}
else
{
int kstart = 1;
int kstop = 1+ghost_width;
ComputeInnerPointsAsync(thread_blocks_halo, threads_per_block, data_stream, d_s_Unews, d_s_Uolds, pitch, Nx, Ny, _Nz, kstart, kstop);
CopyBoundaryRegionToGhostCellAsync(thread_blocks_halo, threads_per_block, data_stream, d_s_Unews, d_left_send_buffer, Nx, Ny, _Nz, pitch, gc_pitch, 1);
checkCuda(cudaMemcpy2DAsync(left_send_buffer, dt_size*(Nx+2), d_left_send_buffer, pitch_gc_bytes, dt_size*(Nx+2), (Ny+2)*(_GC_DEPTH), cudaMemcpyDefault, data_stream));
checkCuda(cudaStreamSynchronize(data_stream));
MPI_CHECK(MPI_Isend(left_send_buffer, (Nx+2)*(Ny+2)*(_GC_DEPTH), MPI_DOUBLE, 0, 1, MPI_COMM_WORLD, &left_send_request[rank]));
}
// Compute inner nodes for device 0
if (rank == 0) {
int kstart = 1;
int kstop = (_Nz+1)-ghost_width;
ComputeInnerPointsAsync(thread_blocks, threads_per_block, compute_stream, d_s_Unews, d_s_Uolds, pitch, Nx, Ny, _Nz, kstart, kstop);
}
// Compute inner nodes for device 1
else
{
int kstart = 1+ghost_width;
int kstop = _Nz+1;
ComputeInnerPointsAsync(thread_blocks, threads_per_block, compute_stream, d_s_Unews, d_s_Uolds, pitch, Nx, Ny, _Nz, kstart, kstop);
}
// Receive data from device 1
if (rank == 0) {
MPI_CHECK(MPI_Irecv(left_send_buffer, (Nx+2)*(Ny+2)*(_GC_DEPTH), MPI_DOUBLE, 1, 1, MPI_COMM_WORLD, &right_receive_request[rank]));
}
else
{
MPI_CHECK(MPI_Irecv(right_send_buffer, (Nx+2)*(Ny+2)*(_GC_DEPTH), MPI_DOUBLE, 0, 0, MPI_COMM_WORLD, &left_receive_request[rank]));
}
if (rank == 0) {
MPI_CHECK(MPI_Wait(&right_receive_request[rank], &status[rank]));
checkCuda(cudaMemcpy2DAsync(d_right_receive_buffer, pitch_gc_bytes, left_send_buffer, dt_size*(Nx+2), dt_size*(Nx+2), ((Ny+2)*(_GC_DEPTH)), cudaMemcpyDefault, data_stream));
CopyGhostCellToBoundaryRegionAsync(thread_blocks_halo, threads_per_block, data_stream, d_s_Unews, d_right_receive_buffer, Nx, Ny, _Nz, pitch, gc_pitch, 0);
}
else
{
MPI_CHECK(MPI_Wait(&left_receive_request[rank], &status[rank]));
checkCuda(cudaMemcpy2DAsync(d_left_receive_buffer, pitch_gc_bytes, right_send_buffer, dt_size*(Nx+2), dt_size*(Nx+2), ((Ny+2)*(_GC_DEPTH)), cudaMemcpyDefault, data_stream));
CopyGhostCellToBoundaryRegionAsync(thread_blocks_halo, threads_per_block, data_stream, d_s_Unews, d_left_receive_buffer, Nx, Ny, _Nz, pitch, gc_pitch, 1);
}
if (rank == 0)
{
MPI_CHECK(MPI_Wait(&right_send_request[rank], MPI_STATUS_IGNORE));
}
else
{
MPI_CHECK(MPI_Wait(&left_send_request[rank], MPI_STATUS_IGNORE));
}
// Swap pointers on the host
checkCuda(cudaDeviceSynchronize());
swap(_DOUBLE_*, d_s_Unews, d_s_Uolds);
}
MPI_CHECK(MPI_Barrier(MPI_COMM_WORLD));
compute_timer += MPI_Wtime();
MPI_CHECK(MPI_Barrier(MPI_COMM_WORLD));
// Copy data from device to host
double DtH_timer = 0;
MPI_CHECK(MPI_Barrier(MPI_COMM_WORLD));
DtH_timer -= MPI_Wtime();
MPI_CHECK(MPI_Barrier(MPI_COMM_WORLD));
checkCuda(cudaMemcpy2D(h_s_Uolds, dt_size*(Nx+2), d_s_Uolds, pitch_bytes, dt_size*(Nx+2), (Ny+2)*(_Nz+2), cudaMemcpyDefault));
MPI_CHECK(MPI_Barrier(MPI_COMM_WORLD));
DtH_timer += MPI_Wtime();
MPI_CHECK(MPI_Barrier(MPI_COMM_WORLD));
// Gather results from subdomains
MPI_CHECK(MPI_Isend(h_s_Uolds, (Nx+2)*(Ny+2)*(_Nz+2), MPI_DOUBLE, 0, 0, MPI_COMM_WORLD, &gather_send_request[rank]));
if (rank == 0)
{
for (int i = 0; i < numberOfProcesses; i++)
{
MPI_CHECK(MPI_Recv(h_s_rbuf[i], (Nx+2)*(Ny+2)*(_Nz+2), MPI_DOUBLE, i, 0, MPI_COMM_WORLD, &status[rank]));
merge_domains(h_s_rbuf[i], h_Uold, Nx, Ny, _Nz, i);
}
}
// Calculate on host
#if defined(DEBUG) || defined(_DEBUG)
if (rank == 0)
{
cpu_heat3D(u_new, u_old, c0, c1, max_iters, Nx, Ny, Nz);
}
#endif
if (rank == 0)
{
float gflops = CalcGflops(compute_timer, max_iters, Nx, Ny, Nz);
PrintSummary("3D Heat (7-pt)", "Plane sweeping", compute_timer, HtD_timer, DtH_timer, gflops, max_iters, Nx);
_DOUBLE_ t = max_iters * dt;
CalcError(h_Uold, u_old, t, h, Nx, Ny, Nz);
}
Finalize();
// Free device memory
checkCuda(cudaFree(d_s_Unews));
checkCuda(cudaFree(d_s_Uolds));
checkCuda(cudaFree(d_right_send_buffer));
checkCuda(cudaFree(d_left_send_buffer));
checkCuda(cudaFree(d_right_receive_buffer));
checkCuda(cudaFree(d_left_receive_buffer));
// Free host memory
checkCuda(cudaFreeHost(h_s_Unews));
checkCuda(cudaFreeHost(h_s_Uolds));
#if defined(DEBUG) || defined(_DEBUG)
if (rank == 0)
{
for (int i = 0; i < numberOfProcesses; i++)
{
checkCuda(cudaFreeHost(h_s_rbuf[i]));
}
free(h_Uold);
}
#endif
checkCuda(cudaFreeHost(left_send_buffer));
checkCuda(cudaFreeHost(left_receive_buffer));
checkCuda(cudaFreeHost(right_send_buffer));
checkCuda(cudaFreeHost(right_receive_buffer));
checkCuda(cudaDeviceReset());
free(u_old);
free(u_new);
return 0;
}
void CG3::operator()(cudaColorSpinorField &x, cudaColorSpinorField &b)
{
// Check to see that we're not trying to invert on a zero-field source
const double b2 = norm2(b);
if(b2 == 0){
profile.TPSTOP(QUDA_PROFILE_INIT);
printfQuda("Warning: inverting on zero-field source\n");
x=b;
param.true_res = 0.0;
param.true_res_hq = 0.0;
return;
}
ColorSpinorParam csParam(x);
csParam.create = QUDA_ZERO_FIELD_CREATE;
cudaColorSpinorField x_prev(b, csParam);
cudaColorSpinorField r_prev(b, csParam);
cudaColorSpinorField temp(b, csParam);
cudaColorSpinorField r(b);
cudaColorSpinorField w(b);
mat(r, x, temp); // r = Mx
double r2 = xmyNormCuda(b,r); // r = b - Mx
PrintStats("CG3", 0, r2, b2, 0.0);
double stop = stopping(param.tol, b2, param.residual_type);
if(convergence(r2, 0.0, stop, 0.0)) return;
// First iteration
mat(w, r, temp);
double rAr = reDotProductCuda(r,w);
double rho = 1.0;
double gamma_prev = 0.0;
double gamma = r2/rAr;
cudaColorSpinorField x_new(x);
cudaColorSpinorField r_new(r);
axpyCuda(gamma, r, x_new); // x_new += gamma*r
axpyCuda(-gamma, w, r_new); // r_new -= gamma*w
// end of first iteration
// axpbyCuda(a,b,x,y) => y = a*x + b*y
int k = 1; // First iteration performed above
double r2_prev;
while(!convergence(r2, 0.0, stop, 0.0) && k<param.maxiter){
x_prev = x; x = x_new;
r_prev = r; r = r_new;
mat(w, r, temp);
rAr = reDotProductCuda(r,w);
r2_prev = r2;
r2 = norm2(r);
// Need to rearrange this!
PrintStats("CG3", k, r2, b2, 0.0);
gamma_prev = gamma;
gamma = r2/rAr;
rho = 1.0/(1. - (gamma/gamma_prev)*(r2/r2_prev)*(1.0/rho));
x_new = x;
axCuda(rho,x_new);
axpyCuda(rho*gamma,r,x_new);
axpyCuda((1. - rho),x_prev,x_new);
r_new = r;
axCuda(rho,r_new);
axpyCuda(-rho*gamma,w,r_new);
axpyCuda((1.-rho),r_prev,r_new);
double rr_old = reDotProductCuda(r_new,r);
printfQuda("rr_old = %1.14lf\n", rr_old);
k++;
}
if(k == param.maxiter)
warningQuda("Exceeded maximum iterations %d", param.maxiter);
// compute the true residual
mat(r, x, temp);
param.true_res = sqrt(xmyNormCuda(b, r)/b2);
PrintSummary("CG3", k, r2, b2);
return;
}
void CG::operator()(cudaColorSpinorField &x, cudaColorSpinorField &b)
{
profile.Start(QUDA_PROFILE_INIT);
// Check to see that we're not trying to invert on a zero-field source
const double b2 = norm2(b);
if(b2 == 0){
profile.Stop(QUDA_PROFILE_INIT);
printfQuda("Warning: inverting on zero-field source\n");
x=b;
param.true_res = 0.0;
param.true_res_hq = 0.0;
return;
}
cudaColorSpinorField r(b);
ColorSpinorParam csParam(x);
csParam.create = QUDA_ZERO_FIELD_CREATE;
cudaColorSpinorField y(b, csParam);
mat(r, x, y);
// zeroCuda(y);
double r2 = xmyNormCuda(b, r);
csParam.setPrecision(param.precision_sloppy);
cudaColorSpinorField Ap(x, csParam);
cudaColorSpinorField tmp(x, csParam);
cudaColorSpinorField *tmp2_p = &tmp;
// tmp only needed for multi-gpu Wilson-like kernels
if (mat.Type() != typeid(DiracStaggeredPC).name() &&
mat.Type() != typeid(DiracStaggered).name()) {
tmp2_p = new cudaColorSpinorField(x, csParam);
}
cudaColorSpinorField &tmp2 = *tmp2_p;
cudaColorSpinorField *x_sloppy, *r_sloppy;
if (param.precision_sloppy == x.Precision()) {
csParam.create = QUDA_REFERENCE_FIELD_CREATE;
x_sloppy = &x;
r_sloppy = &r;
} else {
csParam.create = QUDA_COPY_FIELD_CREATE;
x_sloppy = new cudaColorSpinorField(x, csParam);
r_sloppy = new cudaColorSpinorField(r, csParam);
}
cudaColorSpinorField &xSloppy = *x_sloppy;
cudaColorSpinorField &rSloppy = *r_sloppy;
cudaColorSpinorField p(rSloppy);
if(&x != &xSloppy){
copyCuda(y,x);
zeroCuda(xSloppy);
}else{
zeroCuda(y);
}
const bool use_heavy_quark_res =
(param.residual_type & QUDA_HEAVY_QUARK_RESIDUAL) ? true : false;
profile.Stop(QUDA_PROFILE_INIT);
profile.Start(QUDA_PROFILE_PREAMBLE);
double r2_old;
double stop = b2*param.tol*param.tol; // stopping condition of solver
double heavy_quark_res = 0.0; // heavy quark residual
if(use_heavy_quark_res) heavy_quark_res = sqrt(HeavyQuarkResidualNormCuda(x,r).z);
int heavy_quark_check = 10; // how often to check the heavy quark residual
double alpha=0.0, beta=0.0;
double pAp;
int rUpdate = 0;
double rNorm = sqrt(r2);
double r0Norm = rNorm;
double maxrx = rNorm;
double maxrr = rNorm;
double delta = param.delta;
// this parameter determines how many consective reliable update
// reisudal increases we tolerate before terminating the solver,
// i.e., how long do we want to keep trying to converge
int maxResIncrease = 0; // 0 means we have no tolerance
profile.Stop(QUDA_PROFILE_PREAMBLE);
profile.Start(QUDA_PROFILE_COMPUTE);
blas_flops = 0;
int k=0;
PrintStats("CG", k, r2, b2, heavy_quark_res);
int steps_since_reliable = 1;
while ( !convergence(r2, heavy_quark_res, stop, param.tol_hq) &&
k < param.maxiter) {
matSloppy(Ap, p, tmp, tmp2); // tmp as tmp
double sigma;
bool breakdown = false;
if (param.pipeline) {
double3 triplet = tripleCGReductionCuda(rSloppy, Ap, p);
r2 = triplet.x; double Ap2 = triplet.y; pAp = triplet.z;
r2_old = r2;
alpha = r2 / pAp;
sigma = alpha*(alpha * Ap2 - pAp);
if (sigma < 0.0 || steps_since_reliable==0) { // sigma condition has broken down
r2 = axpyNormCuda(-alpha, Ap, rSloppy);
sigma = r2;
breakdown = true;
}
r2 = sigma;
} else {
r2_old = r2;
pAp = reDotProductCuda(p, Ap);
alpha = r2 / pAp;
// here we are deploying the alternative beta computation
Complex cg_norm = axpyCGNormCuda(-alpha, Ap, rSloppy);
r2 = real(cg_norm); // (r_new, r_new)
sigma = imag(cg_norm) >= 0.0 ? imag(cg_norm) : r2; // use r2 if (r_k+1, r_k+1-r_k) breaks
}
// reliable update conditions
rNorm = sqrt(r2);
if (rNorm > maxrx) maxrx = rNorm;
if (rNorm > maxrr) maxrr = rNorm;
int updateX = (rNorm < delta*r0Norm && r0Norm <= maxrx) ? 1 : 0;
int updateR = ((rNorm < delta*maxrr && r0Norm <= maxrr) || updateX) ? 1 : 0;
// force a reliable update if we are within target tolerance (only if doing reliable updates)
if ( convergence(r2, heavy_quark_res, stop, param.tol_hq) && delta >= param.tol) updateX = 1;
if ( !(updateR || updateX)) {
//beta = r2 / r2_old;
beta = sigma / r2_old; // use the alternative beta computation
if (param.pipeline && !breakdown) tripleCGUpdateCuda(alpha, beta, Ap, rSloppy, xSloppy, p);
else axpyZpbxCuda(alpha, p, xSloppy, rSloppy, beta);
if (use_heavy_quark_res && k%heavy_quark_check==0) {
copyCuda(tmp,y);
heavy_quark_res = sqrt(xpyHeavyQuarkResidualNormCuda(xSloppy, tmp, rSloppy).z);
}
steps_since_reliable++;
} else {
axpyCuda(alpha, p, xSloppy);
if (x.Precision() != xSloppy.Precision()) copyCuda(x, xSloppy);
xpyCuda(x, y); // swap these around?
mat(r, y, x); // here we can use x as tmp
r2 = xmyNormCuda(b, r);
if (x.Precision() != rSloppy.Precision()) copyCuda(rSloppy, r);
zeroCuda(xSloppy);
// break-out check if we have reached the limit of the precision
static int resIncrease = 0;
if (sqrt(r2) > r0Norm && updateX) { // reuse r0Norm for this
warningQuda("CG: new reliable residual norm %e is greater than previous reliable residual norm %e", sqrt(r2), r0Norm);
k++;
rUpdate++;
if (++resIncrease > maxResIncrease) break;
} else {
resIncrease = 0;
}
rNorm = sqrt(r2);
maxrr = rNorm;
maxrx = rNorm;
r0Norm = rNorm;
rUpdate++;
// explicitly restore the orthogonality of the gradient vector
double rp = reDotProductCuda(rSloppy, p) / (r2);
axpyCuda(-rp, rSloppy, p);
beta = r2 / r2_old;
xpayCuda(rSloppy, beta, p);
if(use_heavy_quark_res) heavy_quark_res = sqrt(HeavyQuarkResidualNormCuda(y,r).z);
steps_since_reliable = 0;
}
breakdown = false;
k++;
PrintStats("CG", k, r2, b2, heavy_quark_res);
}
if (x.Precision() != xSloppy.Precision()) copyCuda(x, xSloppy);
xpyCuda(y, x);
profile.Stop(QUDA_PROFILE_COMPUTE);
profile.Start(QUDA_PROFILE_EPILOGUE);
param.secs = profile.Last(QUDA_PROFILE_COMPUTE);
double gflops = (quda::blas_flops + mat.flops() + matSloppy.flops())*1e-9;
reduceDouble(gflops);
param.gflops = gflops;
param.iter += k;
if (k==param.maxiter)
warningQuda("Exceeded maximum iterations %d", param.maxiter);
if (getVerbosity() >= QUDA_VERBOSE)
printfQuda("CG: Reliable updates = %d\n", rUpdate);
// compute the true residuals
mat(r, x, y);
param.true_res = sqrt(xmyNormCuda(b, r) / b2);
#if (__COMPUTE_CAPABILITY__ >= 200)
param.true_res_hq = sqrt(HeavyQuarkResidualNormCuda(x,r).z);
#else
param.true_res_hq = 0.0;
#endif
PrintSummary("CG", k, r2, b2);
// reset the flops counters
quda::blas_flops = 0;
mat.flops();
matSloppy.flops();
profile.Stop(QUDA_PROFILE_EPILOGUE);
profile.Start(QUDA_PROFILE_FREE);
if (&tmp2 != &tmp) delete tmp2_p;
if (param.precision_sloppy != x.Precision()) {
delete r_sloppy;
delete x_sloppy;
}
profile.Stop(QUDA_PROFILE_FREE);
return;
}
int mySetState(void * hndl, UserState * ustate, unsigned int privdata)
{
int val;
static unsigned int testmode;
log_verbose(KERN_INFO "Reached mySetState with privdata %x\n", privdata);
/* Check driver state */
if(DriverState != REGISTERED)
{
printk("Driver does not seem to be ready\n");
return EFAULT;
}
/* Check handle value */
if((hndl != handle[0]) && (hndl != handle[2]))
{
printk("Came with wrong handle\n");
return EBADF;
}
/* Valid only for TX engine */
if(privdata == 0x54545454)
{
spin_lock_bh(&RawLock);
/* Set up the value to be written into the register */
RawTestMode = ustate->TestMode;
if(RawTestMode & TEST_START)
{
testmode = 0;
if(RawTestMode & ENABLE_LOOPBACK) testmode |= LOOPBACK;
if(RawTestMode & ENABLE_PKTCHK) testmode |= PKTCHKR;
if(RawTestMode & ENABLE_PKTGEN) testmode |= PKTGENR;
}
else
{
/* Deliberately not clearing the loopback bit, incase a
* loopback test was going on - allows the loopback path
* to drain off packets. Just stopping the source of packets.
*/
if(RawTestMode & ENABLE_PKTCHK) testmode &= ~PKTCHKR;
if(RawTestMode & ENABLE_PKTGEN) testmode &= ~PKTGENR;
}
printk("SetState TX with RawTestMode %x, reg value %x\n",
RawTestMode, testmode);
/* Now write the registers */
if(RawTestMode & TEST_START)
{
if(!(RawTestMode & (ENABLE_PKTCHK|ENABLE_PKTGEN|ENABLE_LOOPBACK)))
{
printk("%s Driver: TX Test Start with wrong mode %x\n",
MYNAME, testmode);
RawTestMode = 0;
spin_unlock_bh(&RawLock);
return EBADRQC;
}
printk("%s Driver: Starting the test - mode %x, reg %x\n",
MYNAME, RawTestMode, testmode);
/* Next, set packet sizes. Ensure they don't exceed PKTSIZEs */
RawMinPktSize = ustate->MinPktSize;
RawMaxPktSize = ustate->MaxPktSize;
/* Set RX packet size for memory path */
val = RawMaxPktSize;
if(val % BYTEMULTIPLE)
{
printk("********** ODD PACKET SIZE **********\n");
//val -= (val % BYTEMULTIPLE);
}
printk("Reg %x = %x\n", PKT_SIZE_ADDRESS, val);
RawMinPktSize = RawMaxPktSize = val;
/* Now ensure the sizes remain within bounds */
if(RawMaxPktSize > MAXPKTSIZE)
RawMinPktSize = RawMaxPktSize = MAXPKTSIZE;
if(RawMinPktSize < MINPKTSIZE)
RawMinPktSize = RawMaxPktSize = MINPKTSIZE;
if(RawMinPktSize > RawMaxPktSize)
RawMinPktSize = RawMaxPktSize;
val = RawMaxPktSize;
printk("========Reg %x = %d\n", PKT_SIZE_ADDRESS, val);
XIo_Out32(TXbarbase+PKT_SIZE_ADDRESS, val);
printk("RxPktSize %d\n", val);
/*
#ifdef XRAWDATA0
XIo_Out32(TXbarbase+START_ADRS_0, 0x00000000);
XIo_Out32(TXbarbase+START_ADRS_1, 0x04000000);
XIo_Out32(TXbarbase+END_ADRS_0, 0x03000000);
XIo_Out32(TXbarbase+END_ADRS_1, 0x07000000);
XIo_Out32(TXbarbase+WRBURST_0, BURST_SIZE );
XIo_Out32(TXbarbase+RDBURST_0, BURST_SIZE );
XIo_Out32(TXbarbase+WRBURST_1, BURST_SIZE );
XIo_Out32(TXbarbase+RDBURST_1, BURST_SIZE );
#else
XIo_Out32(TXbarbase+START_ADRS_2, 0x08000000);
XIo_Out32(TXbarbase+START_ADRS_3, 0x0C000000);
XIo_Out32(TXbarbase+END_ADRS_2, 0x0B000000);
XIo_Out32(TXbarbase+END_ADRS_3, 0x0F000000);
XIo_Out32(TXbarbase+WRBURST_2, BURST_SIZE );
XIo_Out32(TXbarbase+RDBURST_2, BURST_SIZE );
XIo_Out32(TXbarbase+WRBURST_3, BURST_SIZE );
XIo_Out32(TXbarbase+RDBURST_3, BURST_SIZE );
#endif
*/
/* Incase the last test was a loopback test, that bit may not be cleared. */
XIo_Out32(TXbarbase+TX_CONFIG_ADDRESS, 0);
if(RawTestMode & (ENABLE_PKTCHK|ENABLE_LOOPBACK))
{
TxSeqNo = 0;
if(RawTestMode & ENABLE_LOOPBACK)
RxSeqNo = 0;
printk("========Reg %x = %x\n", TX_CONFIG_ADDRESS, testmode);
XIo_Out32(TXbarbase+TX_CONFIG_ADDRESS, testmode);
}
if(RawTestMode & ENABLE_PKTGEN)
{
RxSeqNo = 0;
printk("========Reg %x = %x\n", RX_CONFIG_ADDRESS, testmode);
XIo_Out32(TXbarbase+RX_CONFIG_ADDRESS, testmode);
}
}
/* Else, stop the test. Do not remove any loopback here because
* the DMA queues and hardware FIFOs must drain first.
*/
else
{
printk("%s Driver: Stopping the test, mode %x\n", MYNAME, testmode);
printk("========Reg %x = %x\n", TX_CONFIG_ADDRESS, testmode);
XIo_Out32(TXbarbase+TX_CONFIG_ADDRESS, testmode);
printk("========Reg %x = %x\n", RX_CONFIG_ADDRESS, testmode);
XIo_Out32(TXbarbase+RX_CONFIG_ADDRESS, testmode);
/* Not resetting sequence numbers here - causes problems
* in debugging. Instead, reset the sequence numbers when
* starting a test.
*/
}
PrintSummary();
spin_unlock_bh(&RawLock);
}
return 0;
}
int main (int argc, char **argv)
{
int i;
unsigned int total_len = 0;
struct timeval begtime,endtime;
FILE *sfd,*pfd;
char sfilename[20] = "string";
char pfilename[20] = "pattern";
//===============================================
if (argc < 4)
{
fprintf (stderr,"Usage: acsmx stringfile patternfile ... -nocase\n");
exit (0);
}
strcpy (sfilename, argv[1]);
sfd = fopen(sfilename,"r");
if(sfd == NULL)
{
fprintf(stderr,"Open file error!\n");
exit(1);
}
strcpy(pfilename,argv[2]);
pfd = fopen(pfilename,"r");
if(sfd == NULL)
{
fprintf(stderr,"Open file error!\n");
exit(1);
}
thread_num = atoi(argv[3]);
acsm = acsmNew (thread_num);
//read patterns
i = 0;
while(fgets(pattern,MAXPATTERNLEN,pfd))
{
int len = strlen(pattern);
acsmAddPattern (acsm, pattern, len-1);
i++;
}
fclose(pfd);
printf("\n\nread %d patterns\n\n===============================",i);
/* Generate GtoTo Table and Fail Table */
acsmCompile (acsm);
//=========================================================
/*read string*/
for(i = 0;i < MAXLINE;i++)
{
if(!fgets(text[i],MAXLEN,sfd))
break;
total_len += strlen(text[i]) - 1; //ignore the last char '\n'
}
line = i;
fclose(sfd);
printf("\n\nreading finished...\n=============================\n\n");
printf("%d lines\t%d bytes",line,total_len);
printf("\n\n=============================\n");
gettimeofday(&begtime,0);
//create multi_thread
thread_array = calloc(thread_num,sizeof(pthread_t));
valid_len_array = calloc(thread_num,sizeof(unsigned int));
pthread_barrier_init(&barrier_thread,NULL,thread_num);
pthread_barrier_init(&barrier_validation,NULL,thread_num);
for(i = 0;i < thread_num; i++)
{
pthread_create(&thread_array[i], NULL, SearchThread, (void*)i);
}
//===========================================================
int err;
for(i = 0;i < thread_num;i++)
{
err = pthread_join(thread_array[i],NULL);
if(err != 0)
{
printf("can not join with thread %d:%s\n", i,strerror(err));
}
}
gettimeofday(&endtime,0);
PrintSummary(acsm);
acsmFree (acsm);
printf ("\n### AC Match Finished ###\n");
printf("\nTime Cost: %lu us\n\n",(endtime.tv_sec - begtime.tv_sec)*1000000 + (endtime.tv_usec - begtime.tv_usec));
printf ("\n====================================\n\n");
printf ("Validation Stage Len:\n\n");
for(i = 0;i < thread_num;i++)
printf("rank%d\t%u\n",i,valid_len_array[i]);
printf ("\n====================================\n\n");
free(thread_array);
free(valid_len_array);
pthread_barrier_destroy(&barrier_thread);
pthread_barrier_destroy(&barrier_validation);
return 0;
}
int main( int argc, char **argv ) {
struct stat stat_buff;
stat_record_t sum_stat;
printer_t print_header, print_record;
nfprof_t profile_data;
char *rfile, *Rfile, *Mdirs, *wfile, *ffile, *filter, *tstring, *stat_type;
char *byte_limit_string, *packet_limit_string, *print_format, *record_header;
char *print_order, *query_file, *UnCompress_file, *nameserver, *aggr_fmt;
int c, ffd, ret, element_stat, fdump;
int i, user_format, quiet, flow_stat, topN, aggregate, aggregate_mask, bidir;
int print_stat, syntax_only, date_sorted, do_tag, compress, do_xstat;
int plain_numbers, GuessDir, pipe_output, csv_output;
time_t t_start, t_end;
uint32_t limitflows;
char Ident[IDENTLEN];
rfile = Rfile = Mdirs = wfile = ffile = filter = tstring = stat_type = NULL;
byte_limit_string = packet_limit_string = NULL;
fdump = aggregate = 0;
aggregate_mask = 0;
bidir = 0;
t_start = t_end = 0;
syntax_only = 0;
topN = -1;
flow_stat = 0;
print_stat = 0;
element_stat = 0;
do_xstat = 0;
limitflows = 0;
date_sorted = 0;
total_bytes = 0;
total_flows = 0;
skipped_blocks = 0;
do_tag = 0;
quiet = 0;
user_format = 0;
compress = 0;
plain_numbers = 0;
pipe_output = 0;
csv_output = 0;
is_anonymized = 0;
GuessDir = 0;
nameserver = NULL;
print_format = NULL;
print_header = NULL;
print_record = NULL;
print_order = NULL;
query_file = NULL;
UnCompress_file = NULL;
aggr_fmt = NULL;
record_header = NULL;
Ident[0] = '\0';
while ((c = getopt(argc, argv, "6aA:Bbc:D:E:s:hHn:i:j:f:qzr:v:w:K:M:NImO:R:XZt:TVv:x:l:L:o:")) != EOF) {
switch (c) {
case 'h':
usage(argv[0]);
exit(0);
break;
case 'a':
aggregate = 1;
break;
case 'A':
if ( !ParseAggregateMask(optarg, &aggr_fmt ) ) {
exit(255);
}
aggregate_mask = 1;
break;
case 'B':
GuessDir = 1;
case 'b':
if ( !SetBidirAggregation() ) {
exit(255);
}
bidir = 1;
// implies
aggregate = 1;
break;
case 'D':
nameserver = optarg;
if ( !set_nameserver(nameserver) ) {
exit(255);
}
break;
case 'E':
query_file = optarg;
if ( !InitExporterList() ) {
exit(255);
}
PrintExporters(query_file);
exit(0);
break;
case 'X':
fdump = 1;
break;
case 'Z':
syntax_only = 1;
break;
case 'q':
quiet = 1;
break;
case 'z':
compress = 1;
break;
case 'c':
limitflows = atoi(optarg);
if ( !limitflows ) {
LogError("Option -c needs a number > 0\n");
exit(255);
}
break;
case 's':
stat_type = optarg;
if ( !SetStat(stat_type, &element_stat, &flow_stat) ) {
exit(255);
}
break;
case 'V': {
char *e1, *e2;
e1 = "";
e2 = "";
#ifdef NSEL
e1 = "NSEL-NEL";
#endif
printf("%s: Version: %s%s%s\n",argv[0], e1, e2, nfdump_version);
exit(0);
} break;
case 'l':
packet_limit_string = optarg;
break;
case 'K':
LogError("*** Anonymisation moved! Use nfanon to anonymise flows!\n");
exit(255);
break;
case 'H':
do_xstat = 1;
break;
case 'L':
byte_limit_string = optarg;
break;
case 'N':
plain_numbers = 1;
break;
case 'f':
ffile = optarg;
break;
case 't':
tstring = optarg;
break;
case 'r':
rfile = optarg;
if ( strcmp(rfile, "-") == 0 )
rfile = NULL;
break;
case 'm':
print_order = "tstart";
Parse_PrintOrder(print_order);
date_sorted = 1;
LogError("Option -m depricated. Use '-O tstart' instead\n");
break;
case 'M':
Mdirs = optarg;
break;
case 'I':
print_stat++;
break;
case 'o': // output mode
print_format = optarg;
break;
case 'O': { // stat order by
int ret;
print_order = optarg;
ret = Parse_PrintOrder(print_order);
if ( ret < 0 ) {
LogError("Unknown print order '%s'\n", print_order);
exit(255);
}
date_sorted = ret == 6; // index into order_mode
} break;
case 'R':
Rfile = optarg;
break;
case 'w':
wfile = optarg;
break;
case 'n':
topN = atoi(optarg);
if ( topN < 0 ) {
LogError("TopnN number %i out of range\n", topN);
exit(255);
}
break;
case 'T':
do_tag = 1;
break;
case 'i':
strncpy(Ident, optarg, IDENT_SIZE);
Ident[IDENT_SIZE - 1] = 0;
if ( strchr(Ident, ' ') ) {
LogError("Ident must not contain spaces\n");
exit(255);
}
break;
case 'j':
UnCompress_file = optarg;
UnCompressFile(UnCompress_file);
exit(0);
break;
case 'x':
query_file = optarg;
InitExtensionMaps(NO_EXTENSION_LIST);
DumpExMaps(query_file);
exit(0);
break;
case 'v':
query_file = optarg;
QueryFile(query_file);
exit(0);
break;
case '6': // print long IPv6 addr
Setv6Mode(1);
break;
default:
usage(argv[0]);
exit(0);
}
}
if (argc - optind > 1) {
usage(argv[0]);
exit(255);
} else {
/* user specified a pcap filter */
filter = argv[optind];
FilterFilename = NULL;
}
// Change Ident only
if ( rfile && strlen(Ident) > 0 ) {
ChangeIdent(rfile, Ident);
exit(0);
}
if ( (element_stat || flow_stat) && (topN == -1) )
topN = 10;
if ( topN < 0 )
topN = 0;
if ( (element_stat && !flow_stat) && aggregate_mask ) {
LogError("Warning: Aggregation ignored for element statistics\n");
aggregate_mask = 0;
}
if ( !flow_stat && aggregate_mask ) {
aggregate = 1;
}
if ( rfile && Rfile ) {
LogError("-r and -R are mutually exclusive. Plase specify either -r or -R\n");
exit(255);
}
if ( Mdirs && !(rfile || Rfile) ) {
LogError("-M needs either -r or -R to specify the file or file list. Add '-R .' for all files in the directories.\n");
exit(255);
}
extension_map_list = InitExtensionMaps(NEEDS_EXTENSION_LIST);
if ( !InitExporterList() ) {
exit(255);
}
SetupInputFileSequence(Mdirs, rfile, Rfile);
if ( print_stat ) {
nffile_t *nffile;
if ( !rfile && !Rfile && !Mdirs) {
LogError("Expect data file(s).\n");
exit(255);
}
memset((void *)&sum_stat, 0, sizeof(stat_record_t));
sum_stat.first_seen = 0x7fffffff;
sum_stat.msec_first = 999;
nffile = GetNextFile(NULL, 0, 0);
if ( !nffile ) {
LogError("Error open file: %s\n", strerror(errno));
exit(250);
}
while ( nffile && nffile != EMPTY_LIST ) {
SumStatRecords(&sum_stat, nffile->stat_record);
nffile = GetNextFile(nffile, 0, 0);
}
PrintStat(&sum_stat);
exit(0);
}
// handle print mode
if ( !print_format ) {
// automatically select an appropriate output format for custom aggregation
// aggr_fmt is compiled by ParseAggregateMask
if ( aggr_fmt ) {
int len = strlen(AggrPrependFmt) + strlen(aggr_fmt) + strlen(AggrAppendFmt) + 7; // +7 for 'fmt:', 2 spaces and '\0'
print_format = malloc(len);
if ( !print_format ) {
LogError("malloc() error in %s line %d: %s\n", __FILE__, __LINE__, strerror(errno) );
exit(255);
}
snprintf(print_format, len, "fmt:%s %s %s",AggrPrependFmt, aggr_fmt, AggrAppendFmt );
print_format[len-1] = '\0';
} else if ( bidir ) {
print_format = "biline";
} else
print_format = DefaultMode;
}
if ( strncasecmp(print_format, "fmt:", 4) == 0 ) {
// special user defined output format
char *format = &print_format[4];
if ( strlen(format) ) {
if ( !ParseOutputFormat(format, plain_numbers, printmap) )
exit(255);
print_record = format_special;
record_header = get_record_header();
user_format = 1;
} else {
LogError("Missing format description for user defined output format!\n");
exit(255);
}
} else {
// predefined output format
// Check for long_v6 mode
i = strlen(print_format);
if ( i > 2 ) {
if ( print_format[i-1] == '6' ) {
Setv6Mode(1);
print_format[i-1] = '\0';
} else
Setv6Mode(0);
}
i = 0;
while ( printmap[i].printmode ) {
if ( strncasecmp(print_format, printmap[i].printmode, MAXMODELEN) == 0 ) {
if ( printmap[i].Format ) {
if ( !ParseOutputFormat(printmap[i].Format, plain_numbers, printmap) )
exit(255);
// predefined custom format
print_record = printmap[i].func;
record_header = get_record_header();
user_format = 1;
} else {
// To support the pipe output format for element stats - check for pipe, and remember this
if ( strncasecmp(print_format, "pipe", MAXMODELEN) == 0 ) {
pipe_output = 1;
}
if ( strncasecmp(print_format, "csv", MAXMODELEN) == 0 ) {
csv_output = 1;
set_record_header();
record_header = get_record_header();
}
// predefined static format
print_record = printmap[i].func;
user_format = 0;
}
break;
}
i++;
}
}
if ( !print_record ) {
LogError("Unknown output mode '%s'\n", print_format);
exit(255);
}
// this is the only case, where headers are printed.
if ( strncasecmp(print_format, "raw", 16) == 0 )
print_header = format_file_block_header;
if ( aggregate && (flow_stat || element_stat) ) {
aggregate = 0;
LogError("Command line switch -s overwrites -a\n");
}
if ( !filter && ffile ) {
if ( stat(ffile, &stat_buff) ) {
LogError("Can't stat filter file '%s': %s\n", ffile, strerror(errno));
exit(255);
}
filter = (char *)malloc(stat_buff.st_size+1);
if ( !filter ) {
LogError("malloc() error in %s line %d: %s\n", __FILE__, __LINE__, strerror(errno) );
exit(255);
}
ffd = open(ffile, O_RDONLY);
if ( ffd < 0 ) {
LogError("Can't open filter file '%s': %s\n", ffile, strerror(errno));
exit(255);
}
ret = read(ffd, (void *)filter, stat_buff.st_size);
if ( ret < 0 ) {
perror("Error reading filter file");
close(ffd);
exit(255);
}
total_bytes += ret;
filter[stat_buff.st_size] = 0;
close(ffd);
FilterFilename = ffile;
}
// if no filter is given, set the default ip filter which passes through every flow
if ( !filter || strlen(filter) == 0 )
filter = "any";
Engine = CompileFilter(filter);
if ( !Engine )
exit(254);
if ( fdump ) {
printf("StartNode: %i Engine: %s\n", Engine->StartNode, Engine->Extended ? "Extended" : "Fast");
DumpList(Engine);
exit(0);
}
if ( syntax_only )
exit(0);
if ( print_order && flow_stat ) {
printf("-s record and -O (-m) are mutually exclusive options\n");
exit(255);
}
if ((aggregate || flow_stat || print_order) && !Init_FlowTable() )
exit(250);
if (element_stat && !Init_StatTable(HashBits, NumPrealloc) )
exit(250);
SetLimits(element_stat || aggregate || flow_stat, packet_limit_string, byte_limit_string);
if ( tstring ) {
if ( !ScanTimeFrame(tstring, &t_start, &t_end) )
exit(255);
}
if ( !(flow_stat || element_stat || wfile || quiet ) && record_header ) {
if ( user_format ) {
printf("%s\n", record_header);
} else {
// static format - no static format with header any more, but keep code anyway
if ( Getv6Mode() ) {
printf("%s\n", record_header);
} else
printf("%s\n", record_header);
}
}
nfprof_start(&profile_data);
sum_stat = process_data(wfile, element_stat, aggregate || flow_stat, print_order != NULL,
print_header, print_record, t_start, t_end,
limitflows, do_tag, compress, do_xstat);
nfprof_end(&profile_data, total_flows);
if ( total_bytes == 0 ) {
printf("No matched flows\n");
exit(0);
}
if (aggregate || print_order) {
if ( wfile ) {
nffile_t *nffile = OpenNewFile(wfile, NULL, compress, is_anonymized, NULL);
if ( !nffile )
exit(255);
if ( ExportFlowTable(nffile, aggregate, bidir, date_sorted, extension_map_list) ) {
CloseUpdateFile(nffile, Ident );
} else {
CloseFile(nffile);
unlink(wfile);
}
DisposeFile(nffile);
} else {
PrintFlowTable(print_record, topN, do_tag, GuessDir, extension_map_list);
}
}
if (flow_stat) {
PrintFlowStat(record_header, print_record, topN, do_tag, quiet, csv_output, extension_map_list);
#ifdef DEVEL
printf("Loopcnt: %u\n", loopcnt);
#endif
}
if (element_stat) {
PrintElementStat(&sum_stat, plain_numbers, record_header, print_record, topN, do_tag, quiet, pipe_output, csv_output);
}
if ( !quiet ) {
if ( csv_output ) {
PrintSummary(&sum_stat, plain_numbers, csv_output);
} else if ( !wfile ) {
if (is_anonymized)
printf("IP addresses anonymised\n");
PrintSummary(&sum_stat, plain_numbers, csv_output);
if ( t_last_flow == 0 ) {
// in case of a pre 1.6.6 collected and empty flow file
printf("Time window: <unknown>\n");
} else {
printf("Time window: %s\n", TimeString(t_first_flow, t_last_flow));
}
printf("Total flows processed: %u, Blocks skipped: %u, Bytes read: %llu\n",
total_flows, skipped_blocks, (unsigned long long)total_bytes);
nfprof_print(&profile_data, stdout);
}
}
Dispose_FlowTable();
Dispose_StatTable();
FreeExtensionMaps(extension_map_list);
#ifdef DEVEL
if ( hash_hit || hash_miss )
printf("Hash hit: %i, miss: %i, skip: %i, ratio: %5.3f\n", hash_hit, hash_miss, hash_skip, (float)hash_hit/((float)(hash_hit+hash_miss)));
#endif
return 0;
}
//---------------------------------------------
//mining frequent itemsets from database
//---------------------------------------------
void MineAllPats()
{
ITEM_COUNTER* pfreqitems;
int *pitem_order_map;
double ntotal_occurrences;
clock_t start, start_mining;;
int i;
if(goparameters.bresult_name_given)
gpfsout = new FSout(goparameters.pat_filename);
start_mining = clock();
gntotal_patterns = 0;
gnmax_pattern_len = 0;
gntotal_singlepath = 0;
gndepth = 0;
gntotal_call = 0;
gpdata = new Data(goparameters.data_filename);
//count frequent items in original database
pfreqitems = NULL;
start = clock();
goDBMiner.ScanDBCountFreqItems(&pfreqitems);
//goTimeTracer.mdInitial_count_time = (double)(clock()-start)/CLOCKS_PER_SEC;
gppat_counters = new int[gnmax_trans_len];
//goMemTracer.IncBuffer(gnmax_trans_len*sizeof(int));
for(i=0;i<gnmax_trans_len;i++)
gppat_counters[i] = 0;
if(gntotal_freqitems==0)
delete gpdata;
else if(gntotal_freqitems==1)
{
gnmax_pattern_len = 1;
OutputOnePattern(pfreqitems[0].item, pfreqitems[0].support);
delete gpdata;
}
else if(gntotal_freqitems>1)
{
gpheader_itemset = new int[gntotal_freqitems];
//goMemTracer.IncBuffer(gntotal_freqitems*sizeof(int));
//an array which maps an item to its frequency order; the most infrequent item has order 0
pitem_order_map = new int[gnmax_item_id];
for(i=0;i<gnmax_item_id;i++)
pitem_order_map[i] = -1;
for(i=0;i<gntotal_freqitems;i++)
pitem_order_map[pfreqitems[i].item] = i;
//if the number of frequent items in a conditonal database is smaller than maximal bucket size, use bucket counting technique
if(gntotal_freqitems <= MAX_BUCKET_SIZE)
{
gpbuckets = new int[(1<<gntotal_freqitems)];
memset(gpbuckets, 0, sizeof(int)*(1<<gntotal_freqitems));
goDBMiner.ScanDBBucketCount(pitem_order_map, gpbuckets);
bucket_count(gpbuckets, gntotal_freqitems, pfreqitems);
delete []gpbuckets;
delete gpdata;
delete []pitem_order_map;
}
else
{
HEADER_TABLE pheader_table;
goAFOPTMiner.Init(MIN(gnmax_trans_len, gntotal_freqitems+1));
gpbuckets = new int[(1<<MAX_BUCKET_SIZE)];
//goMemTracer.IncBuffer(sizeof(int)*(1<<MAX_BUCKET_SIZE));
//initialize header table
start = clock();
pheader_table = new HEADER_NODE[gntotal_freqitems];
//goMemTracer.IncBuffer(sizeof(HEADER_NODE)*gntotal_freqitems);
ntotal_occurrences = 0;
for(i=0;i<gntotal_freqitems;i++)
{
pheader_table[i].item = pfreqitems[i].item;
pheader_table[i].nsupport = pfreqitems[i].support;
pheader_table[i].parray_conddb = NULL;
pheader_table[i].order = i;
ntotal_occurrences += pfreqitems[i].support;
}
//to choose a proper representation format for each conditional database
if((double)goparameters.nmin_sup/gndb_size>=BUILD_TREE_MINSUP_THRESHOLD ||
ntotal_occurrences/(gndb_size*gntotal_freqitems)>=BUILD_TREE_AVGSUP_THRESHOLD ||
gntotal_freqitems<=BUILD_TREE_ITEM_THRESHOLD )
{
for(i=0;i<gntotal_freqitems;i++)
pheader_table[i].flag = AFOPT_FLAG;
}
else
{
for(i=0;i<gntotal_freqitems-BUILD_TREE_ITEM_THRESHOLD;i++)
pheader_table[i].flag = 0;
for(i=MAX(0, gntotal_freqitems-BUILD_TREE_ITEM_THRESHOLD);i<gntotal_freqitems;i++)
pheader_table[i].flag = AFOPT_FLAG;
}
//scan database to construct conditional databases and do bucket counting
memset(gpbuckets, 0, sizeof(int)*(1<<MAX_BUCKET_SIZE));
goDBMiner.ScanDBBuildCondDB(pheader_table, pitem_order_map, gntotal_freqitems, gpbuckets);
//goTimeTracer.mdInitial_construct_time = (double)(clock()-start)/CLOCKS_PER_SEC;
//goMemTracer.mninitial_struct_size = //goMemTracer.mnArrayDB_size+//goMemTracer.mnAFOPTree_size+sizeof(int)*(1<<MAX_BUCKET_SIZE);
bucket_count(gpbuckets, MAX_BUCKET_SIZE, &(pfreqitems[gntotal_freqitems-MAX_BUCKET_SIZE]));
delete []pitem_order_map;
delete gpdata;
//mining frequent itemsets in depth first order
if((double)goparameters.nmin_sup/gndb_size>=BUILD_TREE_MINSUP_THRESHOLD ||
ntotal_occurrences/(gndb_size*gntotal_freqitems)>=BUILD_TREE_AVGSUP_THRESHOLD ||
gntotal_freqitems<=BUILD_TREE_ITEM_THRESHOLD)
goAFOPTMiner.DepthAFOPTGrowth(pheader_table, gntotal_freqitems, 0);
else
{
goArrayMiner.DepthArrayGrowth(pheader_table, gntotal_freqitems);
goAFOPTMiner.DepthAFOPTGrowth(pheader_table, gntotal_freqitems, gntotal_freqitems-BUILD_TREE_ITEM_THRESHOLD);
}
delete []pheader_table;
//goMemTracer.DecBuffer(gntotal_freqitems*sizeof(HEADER_NODE));
delete []gpbuckets;
//goMemTracer.DecBuffer(sizeof(int)*(1<<MAX_BUCKET_SIZE));
}
delete []gpheader_itemset;
//goMemTracer.DecBuffer(gntotal_freqitems*sizeof(int));
}
delete []pfreqitems;
//goMemTracer.DecBuffer(gntotal_freqitems*sizeof(ITEM_COUNTER));
if(goparameters.bresult_name_given)
delete gpfsout;
//goTimeTracer.mdtotal_running_time = (double)(clock()-start_mining)/CLOCKS_PER_SEC;
PrintSummary();
delete []gppat_counters;
//goMemTracer.DecBuffer(gnmax_trans_len*sizeof(int));
}
