int main(int argc, char **argv){
unsigned int runs = 400;
seed_rand();
std::vector<double> xvalues(runs), yvalues(runs), zvalues(runs);
for ( unsigned int i = 0; i < runs ; i++ )
{
xvalues[i] = 1.0 * ((double) rand() - (double)RAND_MAX /2.0) /(double)RAND_MAX;
yvalues[i] = 1.0 * ((double) rand() - (double)RAND_MAX /2.0) /(double)RAND_MAX;
zvalues[i] = 1.0 * ((double) rand() - (double)RAND_MAX /2.0) /(double)RAND_MAX;
}
for ( unsigned int i = 0; i < runs ; i++ )
{
btMatrix3x3 mat;
btQuaternion q_baseline(xvalues[i],yvalues[i],zvalues[i]);
mat.setRotation(q_baseline);
btQuaternion q_from_m;
mat.getRotation(q_from_m);
std::printf("%f, angle between quaternions\n", q_from_m.angle(q_baseline));
}
return 0;
}
static void test_part(){/**Compute the covariance of 4 points with various separation.*/
rand_t rstat;
int seed=4;
double r0=0.2;
double dx=1./64.;
long N=1+2;
long ofx=0;
long ofy=0;
long nx=N;
long ny=N;
long nframe=1000000;
seed_rand(&rstat, seed);
map_t *atm=mapnew(nx, ny, dx,dx, NULL);
dmat *vec=dnew(4,1);
dmat *cov=NULL;
dmat* pp=(dmat*)atm;
for(long i=0; i<nframe; i++){
info("%ld of %ld\n", i, nframe);
for(long j=0; j<nx*ny; j++){
atm->p[j]=randn(&rstat);
}
fractal_do((dmat*)atm, dx, r0,L0,ninit);
vec->p[0]=IND(pp,ofx+0,ofy+0);
vec->p[1]=IND(pp,ofx+1,ofy+0);
vec->p[2]=IND(pp,ofx+0,ofy+1);
vec->p[3]=IND(pp,ofx+1,ofy+1);
dmm(&cov, 1, vec, vec, "nt", 1);
}
dscale(cov, 1./nframe);
writebin(cov,"cov.bin");
}
int main(){
int N=4;
dmat *A=dnew(N,N);
dcell *B=cellnew(2,2);
B->p[0]=dnew(N,N);
B->p[2]=dnew(N,N);
rand_t rstat;
seed_rand(&rstat, 1);
drandn(A, 1, &rstat);
drandn(B->p[0], 1, &rstat);
drandn(B->p[2], 1, &rstat);
A->header=strdup("dx=1/64\ndy=1/64\nLong=kaflasdjfl ksflksjlfjasdflkj asldfj als fals fl asfjalsdf jalsf djasldf jal fsalfkjasdflkj asldfkj asldf \nadf\nwvf=1.e-4");
writebin(A, "A.bin");
writebin(A, "A.fits");
writebin(A, "A.fits.gz");
writebin(B, "B.bin");
writebin(B, "B.fits");
dmat *C=dread("A.fits");
writebin(C, "C.fits");
dcell *D=dcellread("B.fits");
writebin(D, "D.fits");
dcell *E=dcellread("A.fits");
writebin(E, "E.fits");
writebin(E, "E.fits.gz");
}
int main(int argv, char **argc)
{
Vi2c* V = new Vi2c;
v->scl = 1;
v->sca = 1;
int byte_in, bit_out;
long int clock = 0;
seed_rand();
printf("%d\n", rand());
while(true)
{
switch(clock%8)
{
case 0 :
byte_in = read_bit_vector();
V->nois = rand()&15;
V->in = byte_in;
break;
case 1 :
case 2 : break;
case 3 : V->clk = 1; break;
case 4 :
case 5 : break;
case 6 : V->clk = 0; break;
case 7 : bit_out = V->out; putc(bit_out + '0', stdout); break;
}
V->eval();
if(byte_in == EOF) {break;}
clock++;
}
delete V;
exit(0);
}
int
main(int argc, char *argv[])
{
int i, j;
QDP_initialize(&argc, &argv);
QDP_profcontrol(0);
seed = time(NULL);
j = 0;
for(i=1; i<argc; i++) {
switch(argv[i][0]) {
case 'c' : cgtype=atoi(&argv[i][1]); break;
case 'k' : kappa=atof(&argv[i][1]); break;
case 'n' : nit=atoi(&argv[i][1]); break;
case 's' : seed=atoi(&argv[i][1]); break;
case 'S' : style=atoi(&argv[i][1]); break;
case 'x' : j=i; while((i+1<argc)&&(isdigit(argv[i+1][0]))) ++i; break;
default : usage(argv[0]);
}
}
lattice_size = (int *) malloc(ndim*sizeof(int));
if(j==0) {
for(i=0; i<ndim; ++i) lattice_size[i] = 8;
} else {
if(!isdigit(argv[j][1])) usage(argv[0]);
lattice_size[0] = atoi(&argv[j][1]);
for(i=1; i<ndim; ++i) {
if((++j<argc)&&(isdigit(argv[j][0]))) {
lattice_size[i] = atoi(&argv[j][0]);
} else {
lattice_size[i] = lattice_size[i-1];
}
}
}
if(QDP_this_node==0) {
printf("size = %i", lattice_size[0]);
for(i=1; i<ndim; i++) {
printf(" %i", lattice_size[i]);
}
printf("\n");
printf("kappa = %g\n", kappa);
printf("seed = %i\n", seed);
}
QDP_set_latsize(ndim, lattice_size);
QDP_create_layout();
rs = QDP_create_S();
seed_rand(rs, seed);
start();
QDP_finalize();
return 0;
}
int init_spinlock_local(uint32_t thread_num, uint32_t* limit)
{
//assign the thread to the correct core
set_cpu(thread_num);
*limit = 1;
spinlock_seeds = seed_rand();
MEM_BARRIER;
return 0;
}
static void test_ints(){
rand_t init;
seed_rand(&init,1);
const int nwfs=6;
lrand(&init);/*atm */
rand_t wfs_rand[nwfs];
for(int iwfs=0; iwfs<nwfs; iwfs++){
seed_rand(wfs_rand+iwfs,lrand(&init));
}
dcell *mtche=dcellread("powfs0_mtche.bin");
int nsim=500;
int nsa=2582;
dmat *nea=dnew(nsim,nsa*2);
double(*pnea)[nsim]=(void*)nea->p;
double rne=3;
double bkgrnd=0;
double siglev=1000;
for(int iwfs=0; iwfs<nwfs; iwfs++){
for(int isim=0; isim<nsim; isim++){
dbg("iwfs %d isim=%d\n",iwfs,isim);
/*dcell *ints=dcellread("ints_%d_wfs%d.bin",isim,iwfs); */
dcell *ints=dcellread("ints_%d_wfs%d.bin",isim,iwfs);
/*dcell *i0=dcellread("powfs0_i0.bin"); */
dmat *im=NULL, *imy=NULL;
double gnf[2], gny[2];
for(int isa=0; isa<nsa; isa++){
dcp(&im,ints->p[isa]);
dscale(im,siglev);
gnf[0]=0; gnf[1]=0;
dmulvec(gnf, mtche->p[isa], im->p,1.);
gny[0]=0; gny[1]=0;
dcp(&imy, im);
addnoise(imy, &wfs_rand[iwfs], bkgrnd, bkgrnd, 0,0,rne);
dmulvec(gny, mtche->p[isa], imy->p,1.);
P(pnea,isim,isa)=gny[0]-gnf[0];
P(pnea,isim,isa+nsa)=gny[1]-gnf[1];
}
}
writebin(nea,"test_sanea_wfs%d.bin",iwfs);
}
}
static void test_cov(){/*not good */
rand_t rstat;
int seed=4;
double r0=0.2;
double dx=1./64;
long N=1+1024;
long nx=N;
long ny=N;
long nframe=1;
seed_rand(&rstat, seed);
map_t *atm=mapnew(nx, ny, dx,dx, NULL);
cmat *atmhat=cnew((N+1)*3,(N+1)*3);
dmat *atmhattot=dnew((N+1)*3,(N+1)*3);
//cfft2plan(atmhat,-1);
//cfft2plan(atmhat, 1);
dset((dmat*)atm,1);
cembedd(atmhat, (dmat*)atm, 0);
cfft2(atmhat, -1);
cabs22d(&atmhattot, 1, atmhat, 1);
ccpd(&atmhat, atmhattot);
cfft2i(atmhat, 1);
cfftshift(atmhat);
dmat *denom=dnew((N+1)*3,(N+1)*3);
dmat *cov=dnew((N+1)*3,(N+1)*3);
creal2d(&denom, 0, atmhat, 1);
writebin(denom, "denom.bin");
dzero(atmhattot);
for(long i=0; i<nframe; i++){
info("%ld of %ld\n", i, nframe);
for(long j=0; j<nx*ny; j++){
atm->p[j]=randn(&rstat);
}
fractal_do((dmat*)atm, dx, r0,L0,ninit);
/*mapwrite(atm, "atm_%ld.bin", i); */
cembedd(atmhat, (dmat*)atm, 0);
cfft2(atmhat, -1);
cabs22d(&atmhattot, 1, atmhat, 1);
if(i==0 || (i+1)%10==0){
dscale(atmhattot, 1./(i+1));
ccpd(&atmhat, atmhattot);
writebin(atmhattot, "atm_psf_%ld.bin",i+1);
cfft2i(atmhat, 1);
cfftshift(atmhat);
creal2d(&cov, 0, atmhat, 1);
for(long k=0; k<cov->nx*cov->ny; k++){
cov->p[k]/=denom->p[k];
}
writebin(cov, "atm_cov_%ld.bin",i+1);
}
}
}
void *test(void *data)
{
int rand_max;
thread_data_t *d = (thread_data_t *)data;
unsigned short seed[3];
seeds = seed_rand();
//#ifdef __sparc__
phys_id = the_cores[d->id];
cluster_id = get_cluster(phys_id);
//#else
// phys_id = d->id;
//#endif
/* Initialize seed (use rand48 as rand is poor) */
seed[0] = (unsigned short)rand_r(&d->seed);
seed[1] = (unsigned short)rand_r(&d->seed);
seed[2] = (unsigned short)rand_r(&d->seed);
rand_max = d->bank->size - 1;
/* local initialization of locks */
local_th_data[d->id] = init_lock_array_local(phys_id, d->bank->size, the_locks);
/* Wait on barrier */
barrier_cross(d->barrier);
while (stop == 0)
{
uint32_t nb = (uint32_t) (my_random(&(seeds[0]),&(seeds[1]),&(seeds[2])) % 100);
uint32_t acc = (uint32_t) my_random(&(seeds[0]),&(seeds[1]),&(seeds[2])) & rand_max;
account_t* accp = &d->bank->accounts[acc];
if (nb < d->deposit_perc)
{
deposit(accp, 1, d->id);
d->nb_deposit++;
}
else if (nb < d->withdraw_perc)
{
withdraw(accp, 1, d->id);
d->nb_withdraw++;
}
else /* nb < balance_perc */
{
check(accp, d->id);
d->nb_balance++;
}
}
/* Free locks */
//free_local(local_th_data[d->id], d->bank->size);
return NULL;
}
uint32_t* init_rw_ttas_array_local(uint32_t thread_num, uint32_t size){
set_cpu(thread_num);
rw_seeds = seed_rand();
uint32_t* limits;
limits = (uint32_t*)malloc(size * sizeof(uint32_t));
uint32_t i;
for (i = 0; i < size; i++) {
limits[i]=1;
}
return limits;
}
/**
* Seed a random number.
* @param arg eh?
* @return Void... Nothing returned.
*/
void
seed (char *arg)
{
if (arg == NULL)
seed_rand ();
else if (arg[0] == '.')
old_seed ();
else
srand (atoi (arg));
return;
}
uint32_t* init_ttas_array_local(uint32_t thread_num, uint32_t size){
//assign the thread to the correct core
set_cpu(thread_num);
ttas_seeds = seed_rand();
uint32_t* limits;
limits = (uint32_t*)malloc(size * sizeof(uint32_t));
uint32_t i;
for (i = 0; i < size; i++) {
limits[i]=1;
}
return limits;
}
bool fpt(unsigned long long n, unsigned long long k)
{
if(n == 2 || n == 3 ) return true;
if(n < 2 || (n != 2 && n%2==0) ) return false;
seed_rand();
for(int i=0; i<k; i++)
{
unsigned long long a = (rand()%(n-1))+1;
if(congruent_mod(a,n-1,n)!=1)
return false;
}
return true;
}
/**
test type I/II filter with ideal measurement to make sure it is implemented correctly.
*/
dmat* servo_test(dmat *input, double dt, int dtrat, dmat *sigma2n, dmat *gain){
if(input->ny==1){/*single mode. each column is for a mode.*/
input->ny=input->nx;
input->nx=1;
}
int nmod=input->nx;
PDMAT(input,pinput);
dmat *merr=dnew(nmod,1);
dcell *mreal=cellnew(1,1);
dmat *mres=dnew(nmod,input->ny);
dmat *sigman=NULL;
if(dnorm(sigma2n)>0){
sigman=dchol(sigma2n);
}
dcell *meas=cellnew(1,1);
dmat *noise=dnew(nmod, 1);
SERVO_T *st2t=servo_new(NULL, NULL, 0, dt*dtrat, gain);
rand_t rstat;
seed_rand(&rstat, 1);
PDMAT(mres,pmres);
/*two step delay is ensured with the order of using, copy, acc*/
for(int istep=0; istep<input->ny; istep++){
memcpy(merr->p, pinput[istep], nmod*sizeof(double));
dadd(&merr, 1, mreal->p[0], -1);
memcpy(pmres[istep],merr->p,sizeof(double)*nmod);
if(istep % dtrat == 0){
dzero(meas->p[0]);
}
dadd(&meas->p[0], 1, merr, 1);/*average the error. */
dcellcp(&mreal, st2t->mint->p[0]);
if((istep+1) % dtrat == 0){
if(dtrat!=1) dscale(meas->p[0], 1./dtrat);
if(sigman){
drandn(noise, 1, &rstat);
if(sigman->nx>0){
dmm(&meas->p[0], 1, sigman, noise, "nn", 1);
}else{
dadd(&meas->p[0], 1, noise, sigman->p[0]);
}
}
servo_filter(st2t, meas);
}
}
dfree(sigman);
dfree(merr);
dcellfree(mreal);
dcellfree(meas);
servo_free(st2t);
return mres;
}
void *test(void *data)
{
int rand_max;
thread_data_t *d = (thread_data_t *)data;
uint64_t res;
phys_id= the_cores[d->id];
set_cpu(phys_id);
seeds = seed_rand();
rand_max = num_entries - 1;
/* Init of local data if necessary */
/* Wait on barrier */
barrier_cross(d->barrier);
int entry;
while (stop == 0) {
if (num_entries==1) {
entry=0;
} else {
entry =(int) my_random(&(seeds[0]),&(seeds[1]),&(seeds[2])) & rand_max;
}
// entry = (int)(erand48(seed) * rand_max) + rand_min;
#ifdef TEST_CAS
do {
res = CAS_U8(&(the_data[entry].data),0,1);
} while(res!=0);
#elif defined(TEST_TAS)
do {
res = TAS_U8(&(the_data[entry].data));
} while (res!=0);
#elif defined(TEST_FAI)
FAI_U8(&(the_data[entry].data));
#else
perror("No test primitive specified");
#endif
MEM_BARRIER;
the_data[entry].data = 0;
d->num_operations++;
if (op_pause>0) {
cpause(op_pause);
}
}
/* Free any local data if necessary */
return NULL;
}
static void test_accuracy(){
rand_t rstat;
int seed=4;
double r0=0.2;
double dx=1./64.;
long N=1+64;
long nx=N;
long ny=N;
seed_rand(&rstat, seed);
map_t *atm=mapnew(nx, ny, dx, dx,NULL);
for(long i=0; i<nx*ny; i++){
atm->p[i]=randn(&rstat);
}
map_t *atm2=mapnew(nx, ny, dx, dx,NULL);
for(long i=0; i<nx*ny; i++){
atm2->p[i]=randn(&rstat);
}
mapwrite(atm, "atm_rand.bin");
mapwrite(atm2, "atm2_rand.bin");
fractal_do((dmat*)atm, dx, r0,L0,ninit);
mapwrite(atm, "atm_frac.bin");
fractal_inv((dmat*)atm, dx, r0,L0,ninit);
mapwrite(atm, "atm_frac_inv.bin");
fractal_trans((dmat*)atm2, dx, r0,L0,ninit);
mapwrite(atm2, "atm2_frac_trans.bin");
fractal_inv_trans((dmat*)atm2, dx, r0,L0,ninit);
mapwrite(atm2, "atm2_frac_inv_trans.bin");
/*
atm2 is u, atm is v, now comparing u'Av against v'A'u. they should be the same.
A^-1Av should be the same as v
A'^-1A'u should be the same as u.
*/
dmat *u=dread("atm2_rand.bin");
dmat *Atu=dread("atm2_frac_trans.bin");
dmat *Av=dread("atm_frac.bin");
dmat *v=dread("atm_rand.bin");
double dot1=dotdbl(u->p, Av->p, NULL, nx*ny);
double dot2=dotdbl(v->p, Atu->p, NULL, nx*ny);
info("dot1=%g, dot2=%g, diff=%g\n", dot1, dot2, dot1-dot2);
dmat *u2=dread("atm2_frac_inv_trans.bin");
dmat *v2=dread("atm_frac_inv.bin");
double d1=ddiff(u, u2);
double d2=ddiff(v, v2);
info("d1=%g, d2=%g\n", d1, d2);
}
void
SprayList::init_thread(const size_t nthreads)
{
if (!initialized) {
seeds = seed_rand();
ssalloc_init();
d = new thread_data_t;
d->seed = rand();
d->seed2 = rand();
initialized = true;
}
d->nb_threads = nthreads;
}
/*
static int test_fft_speed_small(){
int nis=128;
int *is=mycalloc(nis,int);
dmat *tim=dnew(nis,1);
for(int ii=0; ii<nis; ii++){
is[ii]=ii+1;
}
ccell *ac=cellnew(nis,1);
rand_t stat;
seed_rand(&stat,1);
for(int ii=0; ii<nis; ii++){
ac->p[ii]=cnew(is[ii],is[ii]);
//cfft2plan(ac->p[ii],-1);
crandn(ac->p[ii],20,&stat);
}
TIC;
for(int ii=0; ii<nis; ii++){
info2("size %4d: ",is[ii]);
tic;
for(int i=0; i<1000; i++){
cfft2(ac->p[ii],-1);
}
toc2("fft");
tim->p[ii]=toc3;
}
writebin(tim,"fft_timing");
}
static void test_fft_speed(){
int nis=2048;
int *is=mycalloc(nis,int);
dmat *tim=dnew(nis,1);
for(int ii=0; ii<nis; ii++){
is[ii]=(ii+1)*2;
}
ccell *ac=cellnew(nis,1);
rand_t stat;
seed_rand(&stat,1);
TIC;
for(int ii=0; ii<nis; ii++){
info2("size %4d: ",is[ii]);
tic;
ac->p[ii]=cnew(is[ii],is[ii]);
//cfft2plan(ac->p[ii],-1);
crandn(ac->p[ii],20,&stat);
toc("plan");
}
toc("plan");
for(int ii=0; ii<nis; ii++){
info2("size %4d: ",is[ii]);
tic;
int nrepeat;
if(is[ii]<300)
nrepeat=100;
else if(is[ii]<1000)
nrepeat=10;
else
nrepeat=1;
for(int i=0; i<nrepeat; i++){
cfft2(ac->p[ii],-1);
}
toc2("fft");
tim->p[ii]=toc3/nrepeat;
}
writebin(tim,"fft_timing");
}*/
static void test_fft_special(){
int nis=2;
int *is=mycalloc(nis,int);
dmat *tim=dnew(nis,1);
is[0]=3824;
is[1]=4096;
ccell *ac=ccellnew(nis,1);
rand_t rstat;
seed_rand(&rstat,1);
TIC;
for(int ii=0; ii<nis; ii++){
info2("size %4d: ",is[ii]);
tic;
ac->p[ii]=cnew(is[ii],is[ii]);
//cfft2plan(ac->p[ii],-1);
//cfft2partialplan(ac->p[ii],512,-1);
crandn(ac->p[ii],20,&rstat);
toc("plan");
}
for(int ii=0; ii<nis; ii++){
info2("size %4d: ",is[ii]);
tic;
int nrepeat;
if(is[ii]<300)
nrepeat=100;
else if(is[ii]<1000)
nrepeat=10;
else
nrepeat=4;
for(int i=0; i<nrepeat; i++){
cfft2(ac->p[ii],-1);
}
toc2("fft");
for(int i=0; i<nrepeat; i++){
cfft2partial(ac->p[ii],512,-1);
}
toc2("fft2partial");
tim->p[ii]=toc3/nrepeat;
}
writebin(tim,"fft_timing");
}
int main(int argc, char* argv[]){
/*dsp *RLMc1=dspread("RLMc_old.bin"); */
if(argc!=2){
error("Need 1 argument\n");
}
dspcell *RLM=dspcellread("%s",argv[1]);
dsp *RLMc=dspcell2sp(RLM);
tic;info2("chol ...");
spchol *R1=chol_factorize(RLMc);
toc("done");
rand_t rstat;
seed_rand(&rstat,1);
dmat *y=dnew(RLMc->m, 1);
drandn(y, 1, &rstat);
dmat *x=NULL, *x2=NULL, *x3=NULL;
chol_convert(R1, 1);
tic;
chol_solve(&x, R1, y);
toc("cholmod");tic;
chol_solve(&x, R1, y);
toc("cholmod");tic;
chol_solve_upper(&x3, R1, y);
toc("upper");tic;
chol_solve_upper(&x3, R1, y);
toc("upper");tic;
chol_solve_lower(&x2, R1,y);
toc("lower");tic;
chol_solve_lower(&x2, R1,y);
toc("lower");tic;
chol_solve(&x, R1, y);
toc("cholmod");tic;
chol_solve(&x, R1, y);
toc("cholmod");tic;
writebin(y,"y");
writebin(x,"x");
writebin(x2,"x2");
writebin(x3,"x3");
chol_free(R1);
dspfree(RLMc);
dspcellfree(RLM);
}
int main(int argc, char **argv){
unsigned int runs = 400;
seed_rand();
std::vector<double> xvalues(runs), yvalues(runs), zvalues(runs);
for ( unsigned int i = 0; i < runs ; i++ )
{
xvalues[i] = 1.0 * ((double) rand() - (double)RAND_MAX /2.0) /(double)RAND_MAX;
yvalues[i] = 1.0 * ((double) rand() - (double)RAND_MAX /2.0) /(double)RAND_MAX;
zvalues[i] = 1.0 * ((double) rand() - (double)RAND_MAX /2.0) /(double)RAND_MAX;
}
//Useful Operator Overload
for ( unsigned int i = 0; i < runs ; i++ )
{
btTransform transform(btQuaternion(0,0,0), btVector3(xvalues[i],yvalues[i],zvalues[i]));
btQuaternion initial(xvalues[i],yvalues[i],zvalues[i]);
btQuaternion final(xvalues[i],yvalues[i],zvalues[i]);
final = transform * initial;
std::printf("Useful Operator Overload: %f, angle between quaternions\n", initial.angle(final));
}
TEST(Bullet, Slerp)
{
unsigned int runs = 100;
seed_rand();
btQuaternion q1, q2;
q1.setEuler(0,0,0);
for (unsigned int i = 0 ; i < runs ; i++)
{
q2.setEuler(1.0 * ((double) rand() - (double)RAND_MAX /2.0) /(double)RAND_MAX,
1.0 * ((double) rand() - (double)RAND_MAX /2.0) /(double)RAND_MAX,
1.0 * ((double) rand() - (double)RAND_MAX /2.0) /(double)RAND_MAX);
btQuaternion q3 = slerp(q1,q2,0.5);
EXPECT_NEAR(q3.angle(q1), q2.angle(q3), 1e-5);
}
}
bool fpt_p(unsigned long long n, unsigned long long k)
{
if(n == 2 || n == 3 ) return true;
if(n < 2 || (n != 2 && n%2==0) ) return false;
seed_rand();
bool ans = true;
#pragma omp parallel for
for(int i=0; i<k; i++)
{
#pragma omp flush (ans)
if(ans)
{
unsigned long long a = (rand()%(n-1))+1;
if(congruent_mod(a,n-1,n)!=1)
{
ans=false;
#pragma omp flush (ans)
}
}
}
return ans;
}
static void test_corner(){/*Compute the covariance of 4 corner points*/
rand_t rstat;
int seed=4;
double r0=0.2;
double dx=32;
long N=1+1;
long nx=N;
long ny=N;
long nframe=1000000;
seed_rand(&rstat, seed);
map_t *atm=mapnew(nx, ny, dx, dx,NULL);
dmat *vec=dref_reshape((dmat*)atm, N*N, 1);
dmat *cov=NULL;
for(long i=0; i<nframe; i++){
info("%ld of %ld\n", i, nframe);
for(long j=0; j<nx*ny; j++){
atm->p[j]=randn(&rstat);
}
fractal_do((dmat*)atm, dx, r0,L0,ninit);
dmm(&cov, 1, vec, vec, "nt", 1);
}
dscale(cov, 1./nframe);
writebin(cov,"cov.bin");
}
uint32_t init_rw_ttas_local(uint32_t thread_num){
set_cpu(thread_num);
rw_seeds = seed_rand();
return 1;
}
/* Threads will start here */
int worker_listen(worker *worker) {
int i, b;
monitor *read_mon; /* temp monitor storage */
monitor *write_mon;
int in_file = -1;
int out_file = -1; /* the files */
char *in_buffer, *out_buffer; /* i/o buffers */
job *read_msg, *write_msg; /* temp job storage */
int res = -1;
int what_disc = 0;
int what_block = 0;
double rand_result;
struct drand48_data work_space;
#if 0
char in_buff1[BLOCK_SIZE]; /* read-ahead buffers */
char in_buff2[BLOCK_SIZE];
char in_buff3[BLOCK_SIZE];
char in_buff4[BLOCK_SIZE];
char out_buff1[BLOCK_SIZE]; /* write-behind buffers */
char out_buff2[BLOCK_SIZE];
char out_buff3[BLOCK_SIZE];
char out_buff4[BLOCK_SIZE];
#endif
seed_rand(&work_space);
/* Get in/out file for each repetition desired */
for (i = 0; i < worker->repetition; i++) {
in_file = -1;
out_file= -1;
while (out_file == in_file) {
drand48_r(&work_space, &rand_result);
in_file = (int)(rand_result * 1000);
drand48_r(&work_space, &rand_result);
out_file = (int)(rand_result * 1000);
}
// make made up buffers
in_buffer = emalloc(BLOCK_SIZE * sizeof(char));
memset(in_buffer, 0, BLOCK_SIZE);
out_buffer = emalloc(BLOCK_SIZE * sizeof (char));
memset(out_buffer, 0, BLOCK_SIZE);
for (b=0; b < BLOCKS_PER_FILE; b++) {
/* Create read monitor */
read_mon = emalloc(sizeof(monitor));
read_mon->buffer = in_buffer;
read_mon->request_time = worker->time;
/* Create read job. mainly used for original implementation */
read_msg = emalloc(sizeof(job));
read_msg->message = READ;
read_msg->communication_monitor = read_mon;
/* Create write monitor */
write_mon = emalloc(sizeof(monitor));
write_mon->buffer = out_buffer;
write_mon->request_time = worker->time;
/* Create write job. mainly used for original implementation */
write_msg = emalloc(sizeof(job));
write_msg->message = WRITE;
write_msg->communication_monitor = write_mon;
compute_physical_address(in_file, b, &what_disc, &what_block, worker->number_of_discs);
read_mon->block_number = what_block;
compute_physical_address(out_file, b, &what_disc, &what_block, worker->number_of_discs);
write_mon->block_number = what_block;
if (in_file < out_file) {
// read
pthread_mutex_lock(&worker->all_discs[what_disc].read_lock);
cbuffer_add(read_msg,&worker->all_discs[what_disc].read_cbuf);
pthread_mutex_unlock(&worker->all_discs[what_disc].read_lock);
// write
pthread_mutex_lock(&worker->all_discs[what_disc].write_lock);
cbuffer_add(write_msg,&worker->all_discs[what_disc].write_cbuf);
pthread_mutex_unlock(&worker->all_discs[what_disc].write_lock);
} else {
// write
pthread_mutex_lock(&worker->all_discs[what_disc].write_lock);
while (is_cb_full(&worker->all_discs[what_disc].write_cbuf) == 0) {
sched_yield();
}
cbuffer_add(write_msg,&worker->all_discs[what_disc].write_cbuf);
pthread_mutex_unlock(&worker->all_discs[what_disc].write_lock);
// read
pthread_mutex_lock(&worker->all_discs[what_disc].read_lock);
cbuffer_add(read_msg,&worker->all_discs[what_disc].read_cbuf);
pthread_mutex_unlock(&worker->all_discs[what_disc].read_lock);
}
}
printf("Sent write requests for file: %d\n",out_file);
printf("Sent read requests for file: %d\n",in_file);
// TODO TODO TODO - maybe not the below solution. may have to do it for each block
// checks replies after all blocks for the file have been queued
// Figured its better to send the requests to disc first as it is the choke point
// check reply queue and read replies here
}
printf("Worker took: %ld\n", worker->time);
// printf("Worker took %ld\n", );
return 0;
}
int main(int argc, char **argv)
{
set_cpu(the_cores[0]);
ssalloc_init();
seeds = seed_rand();
#ifdef PAPI
if (PAPI_VER_CURRENT != PAPI_library_init(PAPI_VER_CURRENT))
{
printf("PAPI_library_init error.\n");
return 0;
}
else
{
printf("PAPI_library_init success.\n");
}
if (PAPI_OK != PAPI_query_event(PAPI_L1_DCM))
{
printf("Cannot count PAPI_L1_DCM.");
}
printf("PAPI_query_event: PAPI_L1_DCM OK.\n");
if (PAPI_OK != PAPI_query_event(PAPI_L2_DCM))
{
printf("Cannot count PAPI_L2_DCM.");
}
printf("PAPI_query_event: PAPI_L2_DCM OK.\n");
#endif
struct option long_options[] = {
// These options don't set a flag
{"help", no_argument, NULL, 'h'},
{"duration", required_argument, NULL, 'd'},
{"priority-queue", required_argument, NULL, 'p'},
{"linden", required_argument, NULL, 'L'},
{"spray-list", required_argument, NULL, 'l'},
{"event-simulator", required_argument, NULL, 'e'},
{"initial-size", required_argument, NULL, 'i'},
{"num-threads", required_argument, NULL, 'n'},
{"range", required_argument, NULL, 'r'},
{"seed", required_argument, NULL, 's'},
{"update-rate", required_argument, NULL, 'u'},
{"elasticity", required_argument, NULL, 'x'},
{"nothing", required_argument, NULL, 'l'},
{NULL, 0, NULL, 0}
};
sl_intset_t *set;
pq_t *linden_set;
int i, c, size;
val_t last = 0;
val_t val = 0;
pval_t pval = 0;
unsigned long reads, effreads, updates, collisions, effupds, aborts, aborts_locked_read, aborts_locked_write,
aborts_validate_read, aborts_validate_write, aborts_validate_commit, add, added, remove, removed,
aborts_invalid_memory, aborts_double_write, max_retries, failures_because_contention, depdist;
thread_data_t *data;
pthread_t *threads;
pthread_attr_t attr;
barrier_t barrier;
struct timeval start, end;
struct timespec timeout;
int duration = DEFAULT_DURATION;
int initial = DEFAULT_INITIAL;
int nb_threads = DEFAULT_NB_THREADS;
long range = DEFAULT_RANGE;
int seed = DEFAULT_SEED;
int seed2 = DEFAULT_SEED;
int update = DEFAULT_UPDATE;
int unit_tx = DEFAULT_ELASTICITY;
int alternate = DEFAULT_ALTERNATE;
int pq = DEFAULT_PQ;
int sl = DEFAULT_SL;
int es = DEFAULT_ES;
int lin = DEFAULT_LIN;
int effective = DEFAULT_EFFECTIVE;
sigset_t block_set;
while(1) {
i = 0;
c = getopt_long(argc, argv, "hAplLe:f:d:i:n:r:s:u:x:l:", long_options, &i);
if(c == -1)
break;
if(c == 0 && long_options[i].flag == 0)
c = long_options[i].val;
switch(c) {
case 0:
break;
case 'h':
printf("intset -- STM stress test "
"(skip list)\n"
"\n"
"Usage:\n"
" intset [options...]\n"
"\n"
"Options:\n"
" -h, --help\n"
" Print this message\n"
" -A, --Alternate\n"
" Consecutive insert/remove target the same value\n"
" -l, --spray-list\n"
" Remove via delete_min operations using a spray list\n"
" -p, --priority-queue\n"
" Remove via delete_min operations using a skip list\n"
" -e, --event-simulator\n"
" Descrete event simulator experiment, parameter = dependency distance\n"
" -L, --linden\n"
" Use Linden's priority queue\n"
" -f, --effective <int>\n"
" update txs must effectively write (0=trial, 1=effective, default=" XSTR(DEFAULT_EFFECTIVE) ")\n"
" -d, --duration <int>\n"
" Test duration in milliseconds (0=infinite, default=" XSTR(DEFAULT_DURATION) ")\n"
" -i, --initial-size <int>\n"
" Number of elements to insert before test (default=" XSTR(DEFAULT_INITIAL) ")\n"
" -n, --num-threads <int>\n"
" Number of threads (default=" XSTR(DEFAULT_NB_THREADS) ")\n"
" -r, --range <int>\n"
" Range of integer values inserted in set (default=" XSTR(DEFAULT_RANGE) ")\n"
" -s, --seed <int>\n"
" RNG seed (0=time-based, default=" XSTR(DEFAULT_SEED) ")\n"
" -u, --update-rate <int>\n"
" Percentage of update transactions (default=" XSTR(DEFAULT_UPDATE) ")\n"
" -x, --elasticity (default=4)\n"
" Use elastic transactions\n"
" 0 = non-protected,\n"
" 1 = normal transaction,\n"
" 2 = read elastic-tx,\n"
" 3 = read/add elastic-tx,\n"
" 4 = read/add/rem elastic-tx,\n"
" 5 = fraser lock-free\n"
);
exit(0);
case 'A':
alternate = 1;
break;
case 'l':
sl = 1;
break;
case 'p':
pq = 1;
break;
case 'e':
es = 1;
depdist = atoi(optarg);
break;
case 'L':
lin = 1;
break;
case 'f':
effective = atoi(optarg);
break;
case 'd':
duration = atoi(optarg);
break;
case 'i':
initial = atoi(optarg);
break;
case 'n':
nb_threads = atoi(optarg);
break;
case 'r':
range = atol(optarg);
break;
case 's':
seed = atoi(optarg);
break;
case 'u':
update = atoi(optarg);
break;
case 'x':
unit_tx = atoi(optarg);
break;
case '?':
printf("Use -h or --help for help\n");
exit(0);
default:
exit(1);
}
}
assert(duration >= 0);
assert(initial >= 0);
assert(nb_threads > 0);
assert(range > 0);
assert(update >= 0 && update <= 100);
// if (range < initial)
// {
range = 100000000;
// }
printf("Set type : skip list\n");
printf("Duration : %d\n", duration);
printf("Initial size : %u\n", initial);
printf("Nb threads : %d\n", nb_threads);
printf("Value range : %ld\n", range);
printf("Seed : %d\n", seed);
printf("Update rate : %d\n", update);
printf("Elasticity : %d\n", unit_tx);
printf("Alternate : %d\n", alternate);
printf("Priority Q : %d\n", pq);
printf("Spray List : %d\n", sl);
printf("Linden : %d\n", lin);
printf("Efffective : %d\n", effective);
printf("Type sizes : int=%d/long=%d/ptr=%d/word=%d\n",
(int)sizeof(int),
(int)sizeof(long),
(int)sizeof(void *),
(int)sizeof(uintptr_t));
timeout.tv_sec = duration / 1000;
timeout.tv_nsec = (duration % 1000) * 1000000;
if ((data = (thread_data_t *)malloc(nb_threads * sizeof(thread_data_t))) == NULL) {
perror("malloc");
exit(1);
}
if ((threads = (pthread_t *)malloc(nb_threads * sizeof(pthread_t))) == NULL) {
perror("malloc");
exit(1);
}
if (seed == 0)
srand((int)time(0));
else
srand(seed);
*levelmax = floor_log_2((unsigned int) initial);
set = sl_set_new();
/* stop = 0; */
*running = 1;
// Init STM
printf("Initializing STM\n");
TM_STARTUP();
// Populate set
printf("Adding %d entries to set\n", initial);
i = 0;
if (lin) {
int offset = 32; // not sure what this does
_init_gc_subsystem();
linden_set = pq_init(offset);
}
if (es) { // event simulator has event ids 1..m
// no timeout in ES, finishes when list is empty
// timeout.tv_sec = 0;
// timeout.tv_nsec = 0;
if ((nb_deps = (int *)malloc(initial * sizeof(int))) == NULL) {
perror("malloc");
exit(1);
}
if ((deps = (val_t **)malloc(initial * sizeof(val_t*))) == NULL) {
perror("malloc");
exit(1);
}
while (i < initial)
{
if ((deps[i] = (val_t*)malloc(MAX_DEPS * sizeof(val_t))) == NULL) {
perror("malloc");
exit(1);
}
int num_deps = 0;
nb_deps[i] = 0;
if (lin) {
insert(linden_set, i, i);
} else {
sl_add(set, (val_t)i, 0);
}
while (i < initial-1 && num_deps < MAX_DEPS &&
rand_range_re(NULL, 3) % 2) { // Add geometrically distributed # of deps TODO: parametrize '2'
val_t dep = ((val_t)i)+1;
int dep_var = sqrt(depdist);
dep += depdist + rand_range_re(NULL,2*dep_var) - dep_var;
if (dep >= initial) dep = initial-1;
// while (dep < initial-1 && rand_range_re(NULL, 11) % 10) { // dep should be i+GEO(10) TODO: parametrize '10'
// dep++;
// }
deps[i][num_deps] = dep;
num_deps++;
}
nb_deps[i] = num_deps;
if (num_deps < MAX_DEPS) {
deps[i][num_deps] = -1; // marks last dep
}
i++;
}
} else if (lin) {
while (i < initial)
{
#ifdef DISTRIBUTION_EXPERIMENT
pval = i;
#else
pval = rand_range_re(NULL, range);
#endif
insert(linden_set, pval, pval);
last = pval;
i++;
}
} else {
while (i < initial)
{
#ifdef DISTRIBUTION_EXPERIMENT
val = i;
#else
val = rand_range_re(NULL, range);
#endif
if (sl_add(set, val, 0))
{
last = val;
i++;
}
}
}
#ifdef PRINT_LIST
print_skiplist(set);
#endif
size = sl_set_size(set);
printf("Set size : %d\n", size);
printf("Level max : %d\n", *levelmax);
// Access set from all threads
barrier_init(&barrier, nb_threads + 1);
pthread_attr_init(&attr);
pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE);
printf("Creating threads: ");
for (i = 0; i < nb_threads; i++)
{
printf("%d, ", i);
data[i].first = last;
data[i].range = range;
data[i].update = update;
data[i].unit_tx = unit_tx;
data[i].alternate = alternate;
data[i].pq = pq;
data[i].sl = sl;
data[i].es = es;
data[i].effective = effective;
data[i].first_remove = -1;
data[i].nb_collisions = 0;
data[i].nb_add = 0;
data[i].nb_clean = 0;
data[i].nb_added = 0;
data[i].nb_remove = 0;
data[i].nb_removed = 0;
data[i].nb_contains = 0;
data[i].nb_found = 0;
data[i].nb_aborts = 0;
data[i].nb_aborts_locked_read = 0;
data[i].nb_aborts_locked_write = 0;
data[i].nb_aborts_validate_read = 0;
data[i].nb_aborts_validate_write = 0;
data[i].nb_aborts_validate_commit = 0;
data[i].nb_aborts_invalid_memory = 0;
data[i].nb_aborts_double_write = 0;
data[i].max_retries = 0;
data[i].nb_threads = nb_threads;
data[i].seed = rand();
data[i].seed2 = rand();
data[i].set = set;
data[i].barrier = &barrier;
data[i].failures_because_contention = 0;
data[i].id = i;
/* LINDEN */
data[i].lin = lin;
data[i].linden_set = linden_set;
if (pthread_create(&threads[i], &attr, test, (void *)(&data[i])) != 0) {
fprintf(stderr, "Error creating thread\n");
exit(1);
}
}
pthread_attr_destroy(&attr);
// Catch some signals
if (signal(SIGHUP, catcher) == SIG_ERR ||
//signal(SIGINT, catcher) == SIG_ERR ||
signal(SIGTERM, catcher) == SIG_ERR) {
perror("signal");
exit(1);
}
// Start threads
barrier_cross(&barrier);
printf("STARTING...\n");
gettimeofday(&start, NULL);
#ifndef DISTRIBUTION_EXPERIMENT // don't sleep if doing distro experiment
if (duration > 0) {
nanosleep(&timeout, NULL);
} else {
sigemptyset(&block_set);
sigsuspend(&block_set);
}
#endif
/* AO_store_full(&stop, 1); */
*running = 0;
// if (!es) {
gettimeofday(&end, NULL);
// }
printf("STOPPING...\n");
// Wait for thread completion
for (i = 0; i < nb_threads; i++) {
if (pthread_join(threads[i], NULL) != 0) {
fprintf(stderr, "Error waiting for thread completion\n");
exit(1);
}
}
// if (es) {
// gettimeofday(&end, NULL); // time when all threads finish
// }
printf ("duration = %d\n", duration);
duration = (end.tv_sec * 1000 + end.tv_usec / 1000) - (start.tv_sec * 1000 + start.tv_usec / 1000);
printf ("duration = %d\n", duration);
aborts = 0;
aborts_locked_read = 0;
aborts_locked_write = 0;
aborts_validate_read = 0;
aborts_validate_write = 0;
aborts_validate_commit = 0;
aborts_invalid_memory = 0;
aborts_double_write = 0;
failures_because_contention = 0;
reads = 0;
effreads = 0;
updates = 0;
collisions = 0;
add = 0;
added = 0;
remove = 0;
removed = 0;
effupds = 0;
max_retries = 0;
for (i = 0; i < nb_threads; i++) {
printf("Thread %d\n", i);
printf(" #add : %lu\n", data[i].nb_add);
printf(" #added : %lu\n", data[i].nb_added);
printf(" #remove : %lu\n", data[i].nb_remove);
printf(" #removed : %lu\n", data[i].nb_removed);
printf(" #cleaned : %lu\n", data[i].nb_clean);
printf("first remove : %d\n", data[i].first_remove);
printf(" #collisions : %lu\n", data[i].nb_collisions);
printf(" #contains : %lu\n", data[i].nb_contains);
printf(" #found : %lu\n", data[i].nb_found);
printf(" #aborts : %lu\n", data[i].nb_aborts);
printf(" #lock-r : %lu\n", data[i].nb_aborts_locked_read);
printf(" #lock-w : %lu\n", data[i].nb_aborts_locked_write);
printf(" #val-r : %lu\n", data[i].nb_aborts_validate_read);
printf(" #val-w : %lu\n", data[i].nb_aborts_validate_write);
printf(" #val-c : %lu\n", data[i].nb_aborts_validate_commit);
printf(" #inv-mem : %lu\n", data[i].nb_aborts_invalid_memory);
printf(" #dup-w : %lu\n", data[i].nb_aborts_double_write);
printf(" #failures : %lu\n", data[i].failures_because_contention);
printf(" Max retries : %lu\n", data[i].max_retries);
aborts += data[i].nb_aborts;
aborts_locked_read += data[i].nb_aborts_locked_read;
aborts_locked_write += data[i].nb_aborts_locked_write;
aborts_validate_read += data[i].nb_aborts_validate_read;
aborts_validate_write += data[i].nb_aborts_validate_write;
aborts_validate_commit += data[i].nb_aborts_validate_commit;
aborts_invalid_memory += data[i].nb_aborts_invalid_memory;
aborts_double_write += data[i].nb_aborts_double_write;
failures_because_contention += data[i].failures_because_contention;
reads += data[i].nb_contains;
effreads += data[i].nb_contains +
(data[i].nb_add - data[i].nb_added) +
(data[i].nb_remove - data[i].nb_removed);
updates += (data[i].nb_add + data[i].nb_remove);
collisions += data[i].nb_collisions;
add += data[i].nb_add;
added += data[i].nb_added;
remove += data[i].nb_remove;
removed += data[i].nb_removed;
effupds += data[i].nb_removed + data[i].nb_added;
size += data[i].nb_added - data[i].nb_removed;
if (max_retries < data[i].max_retries)
max_retries = data[i].max_retries;
}
printf("Set size : %d (expected: %d)\n", sl_set_size(set), size);
printf("Duration : %d (ms)\n", duration);
printf("#txs : %lu (%f / s)\n", reads + updates, (reads + updates) * 1000.0 / duration);
printf("#read txs : ");
if (effective) {
printf("%lu (%f / s)\n", effreads, effreads * 1000.0 / duration);
printf(" #contains : %lu (%f / s)\n", reads, reads * 1000.0 / duration);
} else printf("%lu (%f / s)\n", reads, reads * 1000.0 / duration);
printf("#eff. upd rate: %f \n", 100.0 * effupds / (effupds + effreads));
printf("#update txs : ");
if (effective) {
printf("%lu (%f / s)\n", effupds, effupds * 1000.0 / duration);
printf(" #upd trials : %lu (%f / s)\n", updates, updates * 1000.0 /
duration);
} else printf("%lu (%f / s)\n", updates, updates * 1000.0 / duration);
printf("#total_remove : %lu\n", remove);
printf("#total_removed: %lu\n", removed);
printf("#total_add : %lu\n", add);
printf("#total_added : %lu\n", added);
printf("#net (rem-add): %lu\n", removed-added);
printf("#total_collide: %lu\n", collisions);
printf("#norm_collide : %f\n", ((double)collisions)/removed);
printf("#aborts : %lu (%f / s)\n", aborts, aborts * 1000.0 / duration);
printf(" #lock-r : %lu (%f / s)\n", aborts_locked_read, aborts_locked_read * 1000.0 / duration);
printf(" #lock-w : %lu (%f / s)\n", aborts_locked_write, aborts_locked_write * 1000.0 / duration);
printf(" #val-r : %lu (%f / s)\n", aborts_validate_read, aborts_validate_read * 1000.0 / duration);
printf(" #val-w : %lu (%f / s)\n", aborts_validate_write, aborts_validate_write * 1000.0 / duration);
printf(" #val-c : %lu (%f / s)\n", aborts_validate_commit, aborts_validate_commit * 1000.0 / duration);
printf(" #inv-mem : %lu (%f / s)\n", aborts_invalid_memory, aborts_invalid_memory * 1000.0 / duration);
printf(" #dup-w : %lu (%f / s)\n", aborts_double_write, aborts_double_write * 1000.0 / duration);
printf(" #failures : %lu\n", failures_because_contention);
printf("Max retries : %lu\n", max_retries);
#ifdef PRINT_END
print_skiplist(set);
#endif
#ifdef PAPI
long total_L1_miss = 0;
unsigned k = 0;
for (k = 0; k < nb_threads; k++) {
total_L1_miss += g_values[k][0];
//printf("[Thread %d] L1_DCM: %lld\n", i, g_values[i][0]);
//printf("[Thread %d] L2_DCM: %lld\n", i, g_values[i][1]);
}
printf("\n#L1 Cache Misses: %lld\n", total_L1_miss);
printf("#Normalized Cache Misses: %f\n", ((double)total_L1_miss)/(reads+updates));
#endif
// Delete set
sl_set_delete(set);
// Cleanup STM
TM_SHUTDOWN();
free(threads);
free(data);
return 0;
}
int main(int argc, char **argv)
{
set_cpu(the_cores[0]);
ssalloc_init();
seeds = seed_rand();
pin(pthread_self(), 0);
#ifdef PAPI
if (PAPI_VER_CURRENT != PAPI_library_init(PAPI_VER_CURRENT))
{
printf("PAPI_library_init error.\n");
return 0;
}
else
{
printf("PAPI_library_init success.\n");
}
if (PAPI_OK != PAPI_query_event(PAPI_L1_DCM))
{
printf("Cannot count PAPI_L1_DCM.");
}
printf("PAPI_query_event: PAPI_L1_DCM OK.\n");
if (PAPI_OK != PAPI_query_event(PAPI_L2_DCM))
{
printf("Cannot count PAPI_L2_DCM.");
}
printf("PAPI_query_event: PAPI_L2_DCM OK.\n");
#endif
struct option long_options[] = {
// These options don't set a flag
{"help", no_argument, NULL, 'h'},
{"num-threads", required_argument, NULL, 'n'},
{"seed", required_argument, NULL, 's'},
{"input-file", required_argument, NULL, 'i'},
{"output-file", required_argument, NULL, 'o'},
{"max-size", required_argument, NULL, 'm'},
{"nothing", required_argument, NULL, 'l'},
{NULL, 0, NULL, 0}
};
sl_intset_t *set;
pq_t *linden_set;
int i, c, size, edges;
unsigned long reads, effreads, updates, collisions, effupds,
add, added, remove, removed;
thread_data_t *data;
pthread_t *threads;
pthread_attr_t attr;
barrier_t barrier;
struct timeval start, end;
int nb_threads = DEFAULT_NB_THREADS;
int seed = DEFAULT_SEED;
int seed2 = DEFAULT_SEED;
int pq = DEFAULT_PQ;
int sl = DEFAULT_SL;
int lin = DEFAULT_LIN;
char *input = "";
char *output = "";
int src = 0;
int max = -1;
int weighted = 0;
int bimodal = 0;
sigset_t block_set;
while(1) {
i = 0;
c = getopt_long(argc, argv, "hplLwbn:s:i:o:m:", long_options, &i);
if(c == -1)
break;
if(c == 0 && long_options[i].flag == 0)
c = long_options[i].val;
switch(c) {
case 0:
break;
case 'h':
printf("SSSP "
"(priority queue)\n"
"\n"
"Usage:\n"
" sssp [options...]\n"
"\n"
"Options:\n"
" -h, --help\n"
" Print this message\n"
" -l, --spray-list\n"
" Remove via delete_min operations using a spray list\n"
" -p, --priority-queue\n"
" Remove via delete_min operations using a skip list\n"
" -L, --linden\n"
" Use Linden's priority queue\n"
" -n, --num-threads <int>\n"
" Number of threads (default=" XSTR(DEFAULT_NB_THREADS) ")\n"
" -s, --seed <int>\n"
" RNG seed (0=time-based, default=" XSTR(DEFAULT_SEED) ")\n"
" -i, --input-file <string>\n"
" file to read the graph from (required) \n"
" -o, --output-file <string>\n"
" file to write the resulting shortest paths to\n"
" -m, --max-size <int>\n"
" if input graph exceeds max-size, use only first max-size nodes\n"
" -w, --weighted\n"
" use random edge weights uniformly chosen in [0,1]; fixed between trials given fixed seed\n"
" -b, --bimodal\n"
" use random edge weights chosen in [20,30]U[70,80]; fixed between trials given fixed seed\n"
);
exit(0);
case 'l':
sl = 1;
break;
case 'p':
pq = 1;
break;
case 'L':
lin = 1;
break;
case 'w':
weighted = 1;
break;
case 'b':
bimodal = 1;
break;
case 'n':
nb_threads = atoi(optarg);
break;
case 's':
seed = atoi(optarg);
break;
case 'i':
input = optarg;
break;
case 'o':
output = optarg;
break;
case 'm':
max = atoi(optarg);
break;
case '?':
printf("Use -h or --help for help\n");
exit(0);
default:
exit(1);
}
}
assert(nb_threads > 0);
if (seed == 0)
srand((int)time(0));
else
srand(seed);
printf("Set type : skip list\n");
printf("Nb threads : %d\n", nb_threads);
printf("Seed : %d\n", seed);
printf("Priority Q : %d\n", pq);
printf("Spray List : %d\n", sl);
printf("Linden : %d\n", lin);
printf("Type sizes : int=%d/long=%d/ptr=%d/word=%d\n",
(int)sizeof(int),
(int)sizeof(long),
(int)sizeof(void *),
(int)sizeof(uintptr_t));
if ((data = (thread_data_t *)malloc(nb_threads * sizeof(thread_data_t))) == NULL) {
perror("malloc");
exit(1);
}
if ((threads = (pthread_t *)malloc(nb_threads * sizeof(pthread_t))) == NULL) {
perror("malloc");
exit(1);
}
// TODO: Build graph here
FILE* fp = fopen(input, "r");
fscanf(fp, "# Nodes: %d Edges: %d\n", &size, &edges);
if (size > max && max != -1) size = max;
#ifndef STATIC
if ((nodes = (graph_node_t*)malloc(size * sizeof(graph_node_t))) == NULL) {
perror("malloc");
exit(1);
}
#endif
for (i = 0;i < size;i++) {
nodes[i].deg = 0;
nodes[i].dist = -1;
nodes[i].times_processed = 0;
}
int u,v;
int cur = 0, count = 0;
while (fscanf(fp, "%d %d\n", &u, &v) == 2) {
if (u >= size) continue;
if (v >= size) continue;
nodes[u].deg++;
}
#ifndef STATIC
for (i = 0;i < size;i++) {
if ((nodes[i].adj = (int*)malloc(nodes[i].deg * sizeof(int))) == NULL) {
perror("malloc");
exit(1);
}
if ((nodes[i].weights = (int*)malloc(nodes[i].deg * sizeof(int))) == NULL) {
perror("malloc");
exit(1);
}
}
#endif
fclose(fp);
nodes[src].dist = 0;
fp = fopen(input, "r");
int tmp;
fscanf(fp, "# Nodes: %d Edges: %d\n", &tmp, &edges);
int *idx;
if ((idx= (int*)malloc(size * sizeof(int))) == NULL) {
perror("malloc");
exit(1);
}
for (i = 0;i < size;i++) {
idx[i] = 0;
}
while (fscanf(fp, "%d %d\n", &u, &v) == 2) {
if (u >= size) continue;
if (v >= size) continue;
assert(idx[u] < nodes[u].deg);
nodes[u].adj[idx[u]] = v;
if (weighted) {
nodes[u].weights[idx[u]] = rand() % 100;
} else if (bimodal) {
if (rand() % 2) {
nodes[u].weights[idx[u]] = (rand() % 11) + 20;
} else {
nodes[u].weights[idx[u]] = (rand() % 11) + 70;
}
} else {
nodes[u].weights[idx[u]] = 1;
}
idx[u]++;
}
free(idx);
// for (u = 0; u < size; u++) {
// for (count = 0; count < nodes[u].deg; count++) {
// printf("%d %d\n", u, nodes[u].adj[count]);
// }
// }
// pq/sl
*levelmax = floor_log_2(size)+2;
set = sl_set_new();
sl_add_val(set, 0, src, TRANSACTIONAL);
// linden
if (lin) {
int offset = 32; // not sure what this does
_init_gc_subsystem();
linden_set = pq_init(offset);
insert(linden_set, 1, src);
nodes[src].dist = 1; // account for the fact that keys must be positive
}
printf("Graph size : %d\n", size);
printf("Level max : %d\n", *levelmax);
// Access set from all threads
barrier_init(&barrier, nb_threads + 1);
pthread_attr_init(&attr);
pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE);
printf("Creating threads: ");
for (i = 0; i < nb_threads; i++)
{
printf("%d, ", i);
data[i].pq = pq;
data[i].sl = sl;
data[i].first_remove = -1;
data[i].nb_collisions = 0;
data[i].nb_add = 0;
data[i].nb_clean = 0;
data[i].nb_added = 0;
data[i].nb_remove = 0;
data[i].nb_removed = 0;
data[i].nb_contains = 0;
data[i].nb_found = 0;
data[i].nb_aborts = 0;
data[i].nb_aborts_locked_read = 0;
data[i].nb_aborts_locked_write = 0;
data[i].nb_aborts_validate_read = 0;
data[i].nb_aborts_validate_write = 0;
data[i].nb_aborts_validate_commit = 0;
data[i].nb_aborts_invalid_memory = 0;
data[i].nb_aborts_double_write = 0;
data[i].max_retries = 0;
data[i].nb_threads = nb_threads;
data[i].seed = rand();
data[i].seed2 = rand();
data[i].set = set;
data[i].barrier = &barrier;
data[i].failures_because_contention = 0;
data[i].id = i;
/* LINDEN */
data[i].lin = lin;
data[i].linden_set = linden_set;
if (pthread_create(&threads[i], &attr, sssp, (void *)(&data[i])) != 0) {
fprintf(stderr, "Error creating thread\n");
exit(1);
}
}
pthread_attr_destroy(&attr);
// Catch some signals
if (signal(SIGHUP, catcher) == SIG_ERR ||
//signal(SIGINT, catcher) == SIG_ERR ||
signal(SIGTERM, catcher) == SIG_ERR) {
perror("signal");
exit(1);
}
/* stop = 0; */
*running = 1;
// Start threads
barrier_cross(&barrier);
printf("STARTING...\n");
gettimeofday(&start, NULL);
// Wait for thread completion
for (i = 0; i < nb_threads; i++) {
if (pthread_join(threads[i], NULL) != 0) {
fprintf(stderr, "Error waiting for thread completion\n");
exit(1);
}
}
gettimeofday(&end, NULL);
printf("STOPPING...\n");
long nb_processed = 0;
long unreachable = 0;
if (strcmp(output,"")) {
FILE *out = fopen(output, "w");
for (i = 0;i < size;i++) {
fprintf(out, "%d %d\n", i, nodes[i].dist);
}
fclose(out);
} else {
for (i = 0;i < size;i++) {
printf("%d %d\n", i, nodes[i].dist);
}
}
for (i = 0;i < size;i++) {
nb_processed += nodes[i].times_processed;
if (nodes[i].times_processed == 0) {
unreachable++;
}
}
int duration = (end.tv_sec * 1000 + end.tv_usec / 1000) - (start.tv_sec * 1000 + start.tv_usec / 1000);
printf ("duration = %d\n", duration);
reads = 0;
effreads = 0;
updates = 0;
collisions = 0;
add = 0;
added = 0;
remove = 0;
removed = 0;
effupds = 0;
for (i = 0; i < nb_threads; i++) {
printf("Thread %d\n", i);
printf(" #add : %lu\n", data[i].nb_add);
printf(" #added : %lu\n", data[i].nb_added);
printf(" #remove : %lu\n", data[i].nb_remove);
printf(" #removed : %lu\n", data[i].nb_removed);
printf(" #cleaned : %lu\n", data[i].nb_clean);
printf(" #collisions : %lu\n", data[i].nb_collisions);
printf(" #contains : %lu\n", data[i].nb_contains);
printf(" #found : %lu\n", data[i].nb_found);
reads += data[i].nb_contains;
effreads += data[i].nb_contains +
(data[i].nb_add - data[i].nb_added) +
(data[i].nb_remove - data[i].nb_removed);
updates += (data[i].nb_add + data[i].nb_remove);
collisions += data[i].nb_collisions;
add += data[i].nb_add;
added += data[i].nb_added;
remove += data[i].nb_remove;
removed += data[i].nb_removed;
effupds += data[i].nb_removed + data[i].nb_added;
size += data[i].nb_added - data[i].nb_removed;
}
printf("Set size : %d (expected: %d)\n", sl_set_size(set), size);
printf("nodes processed:%d\n", nb_processed);
printf("unreachable : %d\n", unreachable);
printf("wasted work : %d\n", nb_processed - (size - unreachable));
printf("Duration : %d (ms)\n", duration);
printf("#ops : %lu (%f / s)\n", reads + updates, (reads + updates) * 1000.0 / duration);
printf("#read ops : ");
printf("%lu (%f / s)\n", reads, reads * 1000.0 / duration);
printf("#eff. upd rate: %f \n", 100.0 * effupds / (effupds + effreads));
printf("#update ops : ");
printf("%lu (%f / s)\n", updates, updates * 1000.0 / duration);
printf("#total_remove : %lu\n", remove);
printf("#total_removed: %lu\n", removed);
printf("#total_add : %lu\n", add);
printf("#total_added : %lu\n", added);
printf("#net (rem-add): %lu\n", removed-added);
printf("#total_collide: %lu\n", collisions);
printf("#norm_collide : %f\n", ((double)collisions)/removed);
#ifdef PRINT_END
print_skiplist(set);
#endif
#ifdef PAPI
long total_L1_miss = 0;
unsigned k = 0;
for (k = 0; k < nb_threads; k++) {
total_L1_miss += g_values[k][0];
total_L2_miss += g_values[k][1];
//printf("[Thread %d] L1_DCM: %lld\n", i, g_values[i][0]);
//printf("[Thread %d] L2_DCM: %lld\n", i, g_values[i][1]);
}
printf("\n#L1 Cache Misses: %lld\n", total_L1_miss);
printf("#Normalized Cache Misses: %f\n", ((double)total_L1_miss)/(reads+updates));
printf("\n#L2 Cache Misses: %lld\n", total_L2_miss);
printf("#Normalized Cache Misses: %f\n", ((double)total_L1_miss)/(reads+updates));
#endif
// Delete set
if (pq || sl) {
sl_set_delete(set);
}
free(threads);
free(data);
return 0;
}
void* sssp(void *data) {
thread_data_t *d = (thread_data_t *)data;
/* Create transaction */
set_cpu(the_cores[d->id]);
/* Wait on barrier */
ssalloc_init();
PF_CORRECTION;
seeds = seed_rand();
#ifdef PIN
int id = d->id;
// int cpu = 40*(id/40) + 4*(id%10) + (id%40)/10;
int cpu = 4*(id%20) + id/20;
// printf("Pinning %d to %d\n",id,cpu);
pin(pthread_self(), cpu);
// pin(pthread_self(), id);
#endif
#ifdef PAPI
if (PAPI_OK != PAPI_start_counters(g_events, G_EVENT_COUNT))
{
printf("Problem starting counters 1.");
}
#endif
barrier_cross(d->barrier);
// Begin SSSP
int fail = 0;
// int radius = 0;
while (1) {
val_t node;
slkey_t dist_node;
// print_skiplist(d->set);
while (1) {
if (d->sl) {
if (spray_delete_min_key(d->set, &dist_node, &node, d)) break; // keep trying until get a node
} else if (d->pq) {
if (lotan_shavit_delete_min_key(d->set, &dist_node, &node, d)) break;
} else if (d->lin) {
node = (val_t) deletemin_key(d->linden_set, &dist_node, d); break;
} else {
printf("error: no queue selected\n");
exit(1); // TODO: grace
}
if (dist_node == -1) { // flag that list is empty
break;
}
dist_node = 0;
}
if (dist_node == -1) { // list is empty; TODO make sure threads don't quit early
fail++;
if (fail > 20*d->nb_threads) { // TODO: really need a better break condition...
break;
}
continue;
}
fail = 0;
if (dist_node != nodes[node].dist) continue; // dead node
nodes[node].times_processed++;
int i;
for (i = 0;i < nodes[node].deg;i++) {
int v = nodes[node].adj[i];
int w = nodes[node].weights[i];
slkey_t dist_v = nodes[v].dist;
// printf("v=%d dist_v=%d\n", v, dist_v);
if (dist_v == -1 || dist_node + w < dist_v) { // found better path to v
// printf("attempting cas...\n");
// printf("nodes[v].dist=%d dist_v=%d dist_node=%d\n", nodes[v].dist, dist_v, dist_node);
int res = ATOMIC_CAS_MB(&nodes[v].dist, dist_v, dist_node+w);
// printf("%d nodes[%d].dist=%d\n", res, v, nodes[v].dist);
if (res) {
if (d->pq || d->sl) {
sl_add_val(d->set, dist_node+w, v, TRANSACTIONAL); // add to queue only if CAS is successful
} else if (d->lin) {
insert(d->linden_set, dist_node+w, v);
}
d->nb_add++;
// if (dist_node+1 > radius) {
// radius = dist_node+1;
// printf("radius %d\n", radius);
// }
}
}
}
}
// End SSSP
#ifdef PAPI
if (PAPI_OK != PAPI_read_counters(g_values[d->id], G_EVENT_COUNT))
{
printf("Problem reading counters 2.");
}
#endif
PF_PRINT;
return NULL;
}
void *test(void *data)
{
DDPRINT("starting test\n",NULL);
//get the per-thread data
thread_data_t *d = (thread_data_t *)data;
//scale percentages of the various operations to the range 0..255
//this saves us a floating point operation during the benchmark
//e.g instead of random()%100 to determine the next operation we will do, we can simply do random()&256
//this saves time on some platfroms
uint32_t read_thresh = 256 * finds / 100;
//uint32_t write_thresh = 256 * (finds + inserts) / 100;
//place the thread on the apropriate cpu
set_cpu(d->id);
//initialize the custom memeory allocator for this thread (we do not use malloc due to concurrency bottleneck issues)
ssalloc_init();
// ssalloc_align();
bst_init_local(d->id);
//for fine-grain latency measurements, we need to get the lenght of a getticks() function call, which is also counted
//by default when we do getticks(); //code... getticks(); PF_START and PF_STOP use this when fine grain measurements are enabled
PF_CORRECTION;
uint32_t rand_max;
//seed the custom random number generator
seeds = seed_rand();
rand_max = max_key;
uint32_t op;
skey_t key;
int i;
int last = -1;
DDPRINT("staring initial insert\n",NULL);
DDPRINT("number of inserts: %u up to %u\n",d->num_add,rand_max);
//before starting the test, we insert a number of elements in the data structure
//we do this at each thread to avoid the situation where the entire data structure
//resides in the same memory node
// int num_elem = 0;
// if (num_threads == 1){
// num_elem = max_key/2;
// } else{
// num_elem = max_key/4;
// }
// pthread_mutex_lock(d->init_lock);
// fprintf(stderr, "Starting critical section %d\n", d->id);
for (i=0;i<max_key/4;++i) {
key = my_random(&seeds[0],&seeds[1],&seeds[2]) & rand_max;
DDPRINT("key is %u\n",key);
//we make sure the insert was effective (as opposed to just updating an existing entry)
if (d->id < 2) {
if (bst_add(key,root, d->id)!=TRUE) {
i--;
} }
}
// fprintf(stderr, "Exiting critical section %d\n", d->id);
// pthread_mutex_unlock(d->init_lock);
DDPRINT("added initial data\n",NULL);
bool_t res;
/* Init of local data if necessary */
ticks t1,t2;
/* Wait on barrier */
// fprintf(stderr, "Waiting on barrier; thread %d\n", d->id);
barrier_cross(d->barrier);
//start the test
while (*running) {
//generate a key (node that rand_max is expected to be a power of 2)
key = my_random(&seeds[0],&seeds[1],&seeds[2]) & rand_max;
//generate the operation
op = my_random(&seeds[0],&seeds[1],&seeds[2]) & 0xff;
if (op < read_thresh) {
//do a find operation
//PF_START and PF_STOP can be used to do latency measurements of the operation
//to enable them, DO_TIMINGS must be defined at compile time, otherwise they do nothing
//PF_START(2);
bst_contains(key,root, d->id);
//PF_STOP(2);
} else if (last == -1) {
//do a write operation
if (bst_add(key,root, d->id)) {
d->num_insert++;
last=1;
}
} else {
//do a delete operation
if (bst_remove(key,root, d->id)) {
d->num_remove++;
last=-1;
}
}
d->num_operations++;
//memory barrier to ensure no unwanted reporderings are happening
//MEM_BARRIER;
}
//summary of the fine grain measurements if enabled
PF_PRINT;
return NULL;
}
void* test(void *data) {
int unext, last = -1;
val_t val = 0;
pval_t pval = 0;
thread_data_t *d = (thread_data_t *)data;
/* Create transaction */
TM_THREAD_ENTER(d->id);
set_cpu(the_cores[d->id]);
/* Wait on barrier */
ssalloc_init();
PF_CORRECTION;
seeds = seed_rand();
#ifdef PIN
int id = d->id;
int cpu = 40*(id/40) + 4*(id%10) + (id%40)/10;
// printf("Pinning %d to %d\n",id,cpu);
pin(pthread_self(), cpu);
// pin(pthread_self(), id);
#endif
#ifdef PAPI
if (PAPI_OK != PAPI_start_counters(g_events, G_EVENT_COUNT))
{
printf("Problem starting counters 1.");
}
#endif
barrier_cross(d->barrier);
/* Is the first op an update? */
unext = (rand_range_re(&d->seed, 100) - 1 < d->update);
#ifdef DISTRIBUTION_EXPERIMENT
while (1)
#else
while (*running)
#endif
{
if (d->es) { // event simulator experiment
if (d->lin) {
if (!empty(d->linden_set)) {
d->nb_remove++;
pval_t pval = deletemin(d->linden_set, d);
d->nb_removed++;
// printf("%d %d\n", pval, deps[pval][0]);
int i = 0;
val_t dep;
while ((dep = deps[pval][i]) != -1 && i < MAX_DEPS) {
d->nb_add++;
if (insert(d->linden_set, dep, dep)) {
d->nb_added++;
}
i++;
}
}
} else {
if (d->set->head->next[0]->next[0] != NULL) {// set not empty
d->nb_remove++;
if (d->sl) { // spray list
if (spray_delete_min(d->set, &val, d)) {
d->nb_removed++;
} else {
continue;
}
} else if (d->pq) { // lotan_shavit pq
if (lotan_shavit_delete_min(d->set, &val, d)) {
d->nb_removed++;
// continue; // TODO: maybe try remove this to simulate task handling (dependency checks still occur)
} else {
continue;
}
}
// struct timespec ten_usec;
// ten_usec.tv_sec = 0;
// ten_usec.tv_nsec = 10000;
// nanosleep(&ten_usec, NULL);
// dependency handling
int i = 0;
val_t dep;
while ((dep = deps[val][i]) != -1 && i < MAX_DEPS) {
if (!sl_contains(d->set, dep, TRANSACTIONAL)) { // dependent has been removed, need to add it again
if (sl_add(d->set, dep, TRANSACTIONAL)) { // check if insert actually succeeded (otherwise someone else did it first)
d->nb_added++;
}
d->nb_add++;
}
i++;
}
}
}
} else { // not event simulator
if (unext) { // update
if (last < 0) { // add
val = rand_range_re(&d->seed, d->range);
if (d->lin) {
pval = val;
insert(d->linden_set, pval, pval);
d->nb_added++;
last = pval;
} else { // not linden
if (sl_add(d->set, val, TRANSACTIONAL)) {
d->nb_added++;
last = val;
}
}
d->nb_add++;
} else { // remove
if (d->pq) {
if (lotan_shavit_delete_min(d->set, &val, d)) {
d->nb_removed++;
if (d->first_remove == -1) {
d->first_remove = val;
}
}
last = -1;
}
else if (d->sl) {
if (spray_delete_min(d->set, &val, d)) {
d->nb_removed++;
if (d->first_remove == -1) {
d->first_remove = val;
}
last = -1;
}
}
else if (d->lin) {
if ((pval = deletemin(d->linden_set, d))) {
d->nb_removed++;
if (d->first_remove == -1) {
d->first_remove = pval;
}
last = -1;
}
}
else if (d->alternate) { // alternate mode (default)
if (sl_remove(d->set, last, TRANSACTIONAL)) {
d->nb_removed++;
if (d->first_remove == -1) {
d->first_remove = val;
}
}
last = -1;
} else {
/* Random computation only in non-alternated cases */
val = rand_range_re(&d->seed, d->range);
/* Remove one random value */
if (sl_remove_succ(d->set, val, TRANSACTIONAL)) {
d->nb_removed++;
if (d->first_remove == -1) {
d->first_remove = val;
}
/* Repeat until successful, to avoid size variations */
last = -1;
}
}
d->nb_remove++;
}
} else { // read
if (d->alternate) {
if (d->update == 0) {
if (last < 0) {
val = d->first;
last = val;
} else { // last >= 0
val = rand_range_re(&d->seed, d->range);
last = -1;
}
} else { // update != 0
if (last < 0) {
val = rand_range_re(&d->seed, d->range);
//last = val;
} else {
val = last;
}
}
} else val = rand_range_re(&d->seed, d->range);
PF_START(2);
if (sl_contains(d->set, val, TRANSACTIONAL))
d->nb_found++;
PF_STOP(2);
d->nb_contains++;
}
/* Is the next op an update? */
if (d->effective) { // a failed remove/add is a read-only tx
unext = ((100 * (d->nb_added + d->nb_removed))
< (d->update * (d->nb_add + d->nb_remove + d->nb_contains)));
} else { // remove/add (even failed) is considered as an update
unext = (rand_range_re(&d->seed, 100) - 1 < d->update);
}
}
#ifdef DISTRIBUTION_EXPERIMENT
if (d->first_remove != -1) {
break; //only one run
}
#endif
}
#ifdef PAPI
if (PAPI_OK != PAPI_read_counters(g_values[d->id], G_EVENT_COUNT))
{
printf("Problem reading counters 2.");
}
#endif
/* Free transaction */
TM_THREAD_EXIT();
PF_PRINT;
return NULL;
}
