PNL_BEGIN
void pnlSeed(int s)
{
vslDeleteStream(&g_RNG.m_vslStream);
vslNewStream(&g_RNG.m_vslStream, _VSL_UNI_METHOD_, s);
}
void MainWindow::randomizeSigma_1()
{ double x1;
VSLStreamStatePtr stream;
vslNewStream( &stream, VSL_BRNG_MCG31, this->seed );
vdRngUniform( 0, stream, model->nI(), &model->Sigma[0], 0.0, 1.0 );
vslDeleteStream( &stream );
for (int i=0; i < model->nI(); ++i)
{
std::pair<int,int> ends = model->ends(i);
int from = ends.first;
int to = ends.second;
std::pair<double,double> xy0 = model->xy(from);
std::pair<double,double> xy1 = model->xy(to);
if (xy0.first==0 && xy1.first==0
|| xy0.first==0 && xy1.first==1
|| xy0.first==1 && xy1.first==0
|| xy0.first==1 && xy1.first==1
|| xy0.first==model->xmax() && xy1.first==model->xmax()
|| xy0.first==model->xmax()-1 && xy1.first==model->xmax()
|| xy0.first==model->xmax() && xy1.first==model->xmax()-1
|| xy0.first==model->xmax()-1 && xy1.first==model->xmax()-1
)
{
model->Sigma[i]=this->sigmaU;
}
else {x1=model->Sigma[i];
if (x1 < this->fraction) model->Sigma[i] = CUTOFF_SIGMA;
else model->Sigma[i] =1;
}
}
}
static void bernoulli_generate(int n, double p, int* r) {
int seed = 17 + caffe_rng_rand() % 4096;
#ifdef _OPENMP
int nthr = omp_get_max_threads();
int threshold = nthr * caffe::cpu::OpenMpManager::getProcessorSpeedMHz() / 3;
bool run_parallel =
(Caffe::mode() != Caffe::GPU) &&
(omp_in_parallel() == 0) &&
(n >= threshold);
if (!run_parallel) nthr = 1;
# pragma omp parallel num_threads(nthr)
{
const int ithr = omp_get_thread_num();
const int avg_amount = (n + nthr - 1) / nthr;
const int my_offset = ithr * avg_amount;
const int my_amount = std::min(my_offset + avg_amount, n) - my_offset;
#else
{
const int my_amount = n;
const int my_offset = 0;
#endif
VSLStreamStatePtr stream;
vslNewStream(&stream, VSL_BRNG_MCG31, seed);
vslSkipAheadStream(stream, my_offset);
viRngBernoulli(VSL_RNG_METHOD_BERNOULLI_ICDF, stream, my_amount,
r + my_offset, p);
vslDeleteStream(&stream);
}
}
void Caffe::set_random_seed(unsigned int seed)
{
CURAND_CHECK(curandDestroyGenerator(Get().curand_generator_));
CURAND_CHECK(curandCreateGenerator(&Get().curand_generator_, CURAND_RNG_PSEUDO_DEFAULT));
CURAND_CHECK(curandSetPseudoRandomGeneratorSeed(Get().curand_generator_, seed));
VSL_CHECK(vslDeleteStream(&Get().vsl_stream_));
VSL_CHECK(vslNewStream(&Get().vsl_stream_, VSL_BRNG_MT19937, seed));
}
JNIEXPORT jint JNICALL Java_edu_berkeley_bid_VSL_vslDeleteStream
(JNIEnv *env, jclass clazz, jobject jstream)
{
VSLStreamStatePtr streamp = getStream(env, clazz, jstream);
int status = vslDeleteStream(&streamp);
setStream(env, clazz, jstream, streamp);
return (jint)status;
}
Caffe::~Caffe()
{
if (cublas_handle_)
CUBLAS_CHECK(cublasDestroy(cublas_handle_));
if (curand_generator_)
CURAND_CHECK(curandDestroyGenerator(curand_generator_));
if (vsl_stream_)
VSL_CHECK(vslDeleteStream(&vsl_stream_));
}
double * get_vector(int size, int i)
{
double *vec;
VSLStreamStatePtr stream;
vslNewStream( &stream, VSL_BRNG_MT19937, i*time(0) );
vec = (double *)calloc(size, sizeof(double));
vdRngGaussian( VSL_RNG_METHOD_GAUSSIAN_BOXMULLER, stream,
size, vec, 1.0, 3.0 );
vslDeleteStream( &stream );
return vec;
}
void MainWindow::randRcr()
{
int i_Rcr = model->index_of_Rcr();
elementCr = fabs((model->I[ i_Rcr ]));
this->sigmaMin=model->Sigma[i_Rcr ];
VSLStreamStatePtr stream;
vslNewStream( &stream, VSL_BRNG_MCG31, this->seed );
vdRngUniform( 0, stream, model->nI(), &model->Sigma[0], 0.0, 1.0 );
vslDeleteStream( &stream );
double x1=model->Sigma[i_Rcr];
this->randc=x1;
this->rand=x1;
}
int main(){
unsigned int iter=200000000;
int i,j;
double x, y;
double dUnderCurve=0.0;
double pi=0.0;
VSLStreamStatePtr stream; //You need one stream for each thread
double end_time,start_time;
start_time=clock();
#pragma omp parallel private(stream,x,y,i) reduction(+:dUnderCurve)
{
double r[BLOCK_SIZE*2]; //Careful!!!
//you need a private copy of whole array for each thread
vslNewStream( &stream, BRNG, (int)clock() );
#pragma omp for
for(j=0;j<iter/BLOCK_SIZE;j++) {
vdRngUniform( METHOD, stream, BLOCK_SIZE*2, r, 0.0, 1.0 );
//Create random numbers into array r
for (i=0;i<BLOCK_SIZE;i++) {
x=r[i]; //X Coordinate
y=r[i+BLOCK_SIZE]; //Y Coordinate
if (x*x + y*y <= 1.0) { //is distance from Origin under Curve
dUnderCurve++;
}
}
}
vslDeleteStream( &stream );
}
pi = dUnderCurve / (double) iter * 4 ;
end_time=clock();
printf ("pi = %10.9f\n", pi);
printf ("Seconds = %10.9f\n",(double)((end_time-start_time)/1000.0));
return 0;
}
void hard_mkl()
{
/*char *results_file = "hard_mkl.txt";
FILE *res;
if((res=fopen(results_file, "w"))==NULL)
{
printf("Can't open file %s.\n", results_file);
exit(1);
}*/
for(int i = 10; i <= ARRAY_SIZE; i*=10)
{
VSLStreamStatePtr stream;
vslNewStream( &stream, VSL_BRNG_MT19937, i*time(0) );
double *ar1, *ar2, *ar3, *ar4, *ar5, *ar6;
ar1 = (double *)malloc(i*sizeof(double));
ar2 = (double *)malloc(i*sizeof(double));
ar3 = (double *)malloc(i*sizeof(double));
ar4 = (double *)malloc(i*sizeof(double));
ar5 = (double *)malloc(i*sizeof(double));
ar6 = (double *)malloc(i*sizeof(double));
vdRngGaussian( VSL_RNG_METHOD_GAUSSIAN_BOXMULLER, stream,
i, ar1, 1.0, 3.0 );
vdRngGaussian( VSL_RNG_METHOD_GAUSSIAN_BOXMULLER, stream,
i, ar2, 1.0, 3.0 );
double start = omp_get_wtime();
for(int j = 0; j < EXPERIMENTS_NUM; j++)
{
vdCos (i, ar1, ar3);
vdLn (i, ar1, ar4);
vdPow (i, ar1, ar2, ar5);
vdCosh(i, ar2, ar6);
}
double end = omp_get_wtime();
free(ar1); free(ar2); free(ar3); free(ar4); free(ar5); free(ar6);
//fprintf(res, "%lf\n", end-start);
printf("%lf, i=%d\n", end-start, i);
vslDeleteStream( &stream );
}
//fclose(res);
}
void Random<CPU>::set_seed (int seed)
{ rand_check (vslDeleteStream (&vStream_));
rand_check (vslNewStream (&vStream_, VSL_BRNG_MT19937, seed));
}
Random<CPU>::~Random ()
{ rand_check (vslDeleteStream (&vStream_));
}
static void
deinit_random_bit ()
{
(void) vslDeleteStream (&stream);
stream = NULL;
}
void rngDestroy(RngEngine* rng, RngErrorType* info) {
*info = (rng->m_stream != NULL)?(vslDeleteStream(&(rng->m_stream))):(0);
}
~CRNG()
{
if(m_vslStream) vslDeleteStream(&m_vslStream);
}
int main(int argc, char* argv[])
{
double
sTime, eTime;
double sum_delta = 0.0;
double sum_ref = 0.0;
double max_delta = 0.0;
double sumReserve = 0.0;
printf("Monte Carlo European Option Pricing Single Precision\n\n");
printf("Compiler Version = %d\n", __INTEL_COMPILER/100);
printf("Release Update = %d\n", __INTEL_COMPILER_UPDATE);
printf("Build Time = %s %s\n", __DATE__, __TIME__);
printf("Path Length = %d\n", RAND_N);
printf("Number of Options = %d\n", OPT_N);
printf("Block Size = %d\n", RAND_BLOCK_LENGTH);
printf("Worker Threads = %d\n\n", NTHREADS);
const int mem_size = sizeof(float)*OPT_PER_THREAD;
#ifndef _OPENMP
NTHREADS = 1;
#endif
float *samples[MAX_THREADS];
VSLStreamStatePtr Streams[MAX_THREADS];
const int nblocks = RAND_N/RAND_BLOCK_LENGTH;
#pragma omp parallel reduction(+ : sum_delta) reduction(+ : sum_ref) reduction(+ : sumReserve) reduction(max : max_delta)
{
#ifdef _OPENMP
int threadID = omp_get_thread_num();
#else
int threadID = 0;
#endif
unsigned int randseed = RANDSEED + threadID;
srand(randseed);
float *CallResultList = (float *)scalable_aligned_malloc(mem_size, SIMDALIGN);
float *CallConfidenceList = (float *)scalable_aligned_malloc(mem_size, SIMDALIGN);
float *StockPriceList = (float *)scalable_aligned_malloc(mem_size, SIMDALIGN);
float *OptionStrikeList = (float *)scalable_aligned_malloc(mem_size, SIMDALIGN);
float *OptionYearsList = (float *)scalable_aligned_malloc(mem_size, SIMDALIGN);
for(int i = 0; i < OPT_PER_THREAD; i++)
{
CallResultList[i] = 0.0f;
CallConfidenceList[i] = 0.0f;
StockPriceList[i] = RandFloat_T(5.0f, 50.0f, &randseed);
OptionStrikeList[i] = RandFloat_T(10.0f, 25.0f, &randseed);
OptionYearsList[i] = RandFloat_T(1.0f, 5.0f, &randseed);
}
samples[threadID] = (float *)scalable_aligned_malloc(RAND_BLOCK_LENGTH * sizeof(float), SIMDALIGN);
vslNewStream(&(Streams[threadID]), VSL_BRNG_MT2203 + threadID, RANDSEED);
#pragma omp barrier
if (threadID == 0)
{
printf("Starting options pricing...\n");
sTime = second();
start_cyc = _rdtsc();
}
for(int opt = 0; opt < OPT_PER_THREAD; opt++)
{
const float VBySqrtT = VLog2E * sqrtf(OptionYearsList[opt]);
const float MuByT = MuLog2E * OptionYearsList[opt];
const float Y = StockPriceList[opt];
const float Z = OptionStrikeList[opt];
float v0 = 0.0f;
float v1 = 0.0f;
for(int block = 0; block < nblocks; ++block)
{
float *rand = samples[threadID];
vsRngGaussian (VSL_RNG_METHOD_GAUSSIAN_ICDF, Streams[threadID], RAND_BLOCK_LENGTH, rand, MuByT, VBySqrtT);
#pragma vector aligned
#pragma simd reduction(+:v0) reduction(+:v1)
#pragma unroll(4)
for(int i=0; i < RAND_BLOCK_LENGTH; i++)
{
float callValue = Y * exp2f(rand[i]) - Z;
callValue = (callValue > 0.0) ? callValue : 0.0;
v0 += callValue;
v1 += callValue * callValue;
}
}
const float exprt = exp2f(RLog2E*OptionYearsList[opt]);
CallResultList[opt] = exprt * v0 * INV_RAND_N;
const float stdDev = sqrtf((F_RAND_N * v1 - v0 * v0) * STDDEV_DENOM);
CallConfidenceList[opt] = (float)(exprt * stdDev * CONFIDENCE_DENOM);
} //end of opt
#pragma omp barrier
if (threadID == 0) {
end_cyc = _rdtsc();
eTime = second();
printf("Parallel simulation completed in %f seconds.\n", eTime-sTime);
printf("Validating the result...\n");
}
double delta = 0.0, ref = 0.0, L1norm = 0.0;
int max_index = 0;
double max_local = 0.0;
for(int i = 0; i < OPT_PER_THREAD; i++)
{
double callReference, putReference;
BlackScholesBodyCPU(
callReference,
putReference,
StockPriceList[i],
OptionStrikeList[i], OptionYearsList[i], RISKFREE, VOLATILITY );
ref = callReference;
delta = fabs(callReference - CallResultList[i]);
sum_delta += delta;
sum_ref += fabs(ref);
if(delta > 1e-6)
sumReserve += CallConfidenceList[i] / delta;
max_local = delta>max_local? delta: max_local;
}
max_delta = max_local>max_delta? max_local: max_delta;
vslDeleteStream(&(Streams[threadID]));
scalable_aligned_free(samples[threadID]);
scalable_aligned_free(CallResultList);
scalable_aligned_free(CallConfidenceList);
scalable_aligned_free(StockPriceList);
scalable_aligned_free(OptionStrikeList);
scalable_aligned_free(OptionYearsList);
}//end of parallel block
sumReserve /= (double)OPT_N;
const double L1norm = sum_delta / sum_ref;
printf("L1_Norm = %4.3E\n", L1norm);
printf("Average RESERVE = %4.3f\n", sumReserve);
printf("Max Error = %4.3E\n", max_delta);
const unsigned long long cyc = end_cyc - start_cyc;
const double optcyc = (double)cyc/(double)OPT_N;
printf("==========================================\n");
printf("Total Cycles = %lld\n", cyc);
printf("Cyc/opt = %8.3f\n", optcyc);
printf("Time Elapsed = %8.3f\n", eTime-sTime);
printf("Options/sec = %8.3f\n", OPT_N/(eTime-sTime));
printf("==========================================\n");
return 0;
}
int main(int argc, char *argv[])
{
unsigned long long count = 0;
double EPSILON = X0*1.0E-2;
double err;
double PXend;
const double dt = T/N;
const double rootdt = sqrt((double)T/N);
int nCal = N/Ncache;
const int left = N%Ncache;
VSLStreamStatePtr stream;
int errcode = vslNewStream(&stream, VSL_BRNG_MT2203, 0);//seed=0
start_timer();
for (unsigned long long m = 0; m < M; ++m){
// one-time MC simulation
err = 0.0;
vdRngGaussian(VSL_RNG_METHOD_GAUSSIAN_BOXMULLER, stream, Ncache, NRV, 0.0f, 1.0f);// leaves the rest of random numbers generated by the idle thread
BM[0] = rootdt*NRV[0];
PX[0] = X0;
for (int k = 0; k < nCal; ++k){
//rootdt:firstprivate???
#pragma omp parallel default(none) shared(NRV, BM, PX, stream, err, PXend, rootdt, dt, k)
{
double errloc = 0.0;
double upbd, tmp;
//GUIDED_CHUNK too large: load imbalance
//GUIDED_CHUNK too small: scheduling overhead
//#pragma omp for schedule(guided, GUIDED_CHUNK)
#pragma omp for schedule(guided) //tunable
for (int i = 1; i < Ncache; ++i){
//tmp = BM[0];
tmp = 0.0;
#pragma simd reduction(+:tmp) vectorlengthfor(double) assert
for (int j = 1; j <= i; ++j){
//tmp += rootdt*NRV[j];
tmp += NRV[j];
}
//BM[i] = tmp;
BM[i] = BM[0] + tmp*rootdt;
//PX[i+1] = X0*exp(-0.5*SIGMA*SIGMA*(k*Ncache+i+1)*dt+SIGMA*tmp);
PX[i+1] = X0*exp(-0.5*SIGMA*SIGMA*(k*Ncache+i+1)*dt+SIGMA*BM[i]);
}
#pragma omp single
{
PX[1] = X0*exp(-0.5*SIGMA*SIGMA*(k*Ncache+1)*dt+SIGMA*BM[0]);
}
//maybe vary the scheduling strategy?
#pragma omp for reduction(+:err) nowait
for (int i = 0; i < Ncache; ++i){
int j = k*Ncache+i;
double Tj = j*(double)T/N;
upbd = (log(PX[i]/K)+0.5*SIGMA*SIGMA*(T-Tj))/(SIGMA*sqrt(T-Tj));
//errloc -= 1/(sqrt(2*PI))*(PX[i+1]-PX[i])*vNormalIntegral(upbd);
err += -1/(sqrt(2*PI))*(PX[i+1]-PX[i])*vNormalIntegral(upbd);
}
#pragma omp single
{
vdRngGaussian(VSL_RNG_METHOD_GAUSSIAN_BOXMULLER, stream, Ncache, NRV, 0.0f, 1.0f);// leaves the rest of random numbers generated by the idle thread
}//single
}//parallel
BM[0] = BM[Ncache-1] + rootdt*NRV[0];
PX[0] = PX[Ncache];
}//for nCal
PXend = PX[Ncache];
#pragma omp parallel default(none) shared(NRV, BM, PX, err, rootdt, dt, nCal, left, PXend)
{
double errloc = 0.0;
double upbd, tmp;
if(left!=0){
//GUIDED_CHUNK too large: load imbalance
//GUIDED_CHUNK too small: scheduling overhead
//#pragma omp for schedule(guided, GUIDED_CHUNK)
#pragma omp for schedule(guided) //tunable
for (int i = 1; i < left; ++i){
//tmp = BM[0];
tmp = 0.0;
#pragma simd reduction(+:tmp) vectorlengthfor(double) assert
for (int j = 1; j <= i; ++j){
//tmp += rootdt*NRV[j];
tmp += NRV[j];
}
//BM[i] = tmp;
BM[i] = BM[0] + tmp*rootdt;
//PX[i+1] = X0*exp(-0.5*SIGMA*SIGMA*(nCal*Ncache+i+1)*dt+SIGMA*BM[i]);
PX[i+1] = X0*exp(-0.5*SIGMA*SIGMA*(nCal*Ncache+i+1)*dt+SIGMA*BM[i]);
}
#pragma omp single
{
PX[1] = X0*exp(-0.5*SIGMA*SIGMA*(nCal*Ncache+1)*dt+SIGMA*BM[0]);
PXend = PX[left];
}
//maybe vary the scheduling strategy?
#pragma omp for reduction(+:err) nowait
for (int i = 0; i < left; ++i){
int j = nCal*Ncache+i;
double Tj = j*(double)T/N;
upbd = (log(PX[i]/K)+0.5*SIGMA*SIGMA*(T-Tj))/(SIGMA*sqrt(T-Tj));
err += -1/sqrt((2*PI))*(PX[i+1]-PX[i])*vNormalIntegral(upbd);
}
}//if
#pragma omp single nowait
{
upbd = (log(X0/K) + 0.5*SIGMA*SIGMA*T)/(SIGMA*sqrt(T));
errloc -= X0/(sqrt(2*PI))*vNormalIntegral(upbd);
#pragma omp atomic
err += errloc;
}
#pragma omp single nowait
{
upbd = (log(X0/K) - 0.5*SIGMA*SIGMA*T)/(SIGMA*sqrt(T));
errloc += K/(sqrt(2*PI))*vNormalIntegral(upbd);
#pragma omp atomic
err += errloc;
}
#pragma omp single nowait
{
if(PXend > K)
errloc += PXend - K;
#pragma omp atomic
err += errloc;
}
}//parallel
err = fabs(err);
if(err < EPSILON)
count++;
//printf("err=%.10lf\n",err);
}//MC simulation
printf ("time %g ms\n", stop_timer());
printf("err=%.20lf\n",err);
printf("count=%llu, M=%llu\n", count, M);
printf("%.5g\n", (double)count/(double)M);
vslDeleteStream(&stream);
return 0;
}
