LIS_INT lis_solver_set_shadowresidual(LIS_SOLVER solver, LIS_VECTOR r0, LIS_VECTOR rs0)
{
unsigned long init[4]={0x123, 0x234, 0x345, 0x456}, length=4;
LIS_INT i,n,resid;
LIS_DEBUG_FUNC_IN;
resid = solver->options[LIS_OPTIONS_INIT_SHADOW_RESID];
if( resid==LIS_RANDOM )
{
n = solver->A->n;
init_by_array(init, length);
#ifdef USE_QUAD_PRECISION
if( solver->precision==LIS_PRECISION_DEFAULT )
#endif
{
#ifdef _OPENMP
#pragma omp parallel for private(i)
#endif
for(i=0;i<n;i++)
{
rs0->value[i] = genrand_real1();
}
}
#ifdef USE_QUAD_PRECISION
else
{
#ifdef _OPENMP
#pragma omp parallel for private(i)
#endif
for(i=0;i<n;i++)
{
rs0->value[i] = genrand_real1();
rs0->value_lo[i] = 0.0;
}
}
#endif
}
else
{
#ifdef USE_QUAD_PRECISION
if( solver->precision==LIS_PRECISION_DEFAULT )
#endif
{
lis_vector_copy(r0,rs0);
}
#ifdef USE_QUAD_PRECISION
else
{
lis_vector_copyex_mm(r0,rs0);
}
#endif
}
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
void gretl_rand_set_multi_seed (const gretl_matrix *seed)
{
guint32 useeds[4];
int i;
for (i=0; i<4; i++) {
useeds[i] = seed->val[i];
}
init_by_array(useeds, 4);
gretl_rand_octet(NULL);
}
int* makeArray( int size ) {
unsigned long init[4] = { 0x123, 0x234, 0x345, 0x456 };
unsigned long length = 10;
init_by_array( init, length );
int*array = ( int* )malloc( sizeof(int) * size );
if( !array )
return NULL;
int*p = array;
for( int i = 0; i < size; i++ )
*p++ = abs( (int) genrand_int32() >> 16 );
return array;
}
void init_triangular_table(void)
{
int i;
unsigned long init[4]={0x123, 0x234, 0x345, 0x456}, length=4;
init_by_array(init, length);
for (i = 0; i < 257; i++) {
triangular_table[i] = (double)(i/* - (genrand_int32() % 1)*/) / 256.0;
if(triangular_table[i] < 0) {triangular_table[i] = 0;}
else if(triangular_table[i] > 1.0) {triangular_table[i] = 1.0;}
}
triangular_table[0] = 0.0;
triangular_table[256] = 1.0;
}
main()
{
int i, j;
clock_t before, after;
// Synthesizable Random Number Initialization - mt19937ar
unsigned long init[4]={0x123, 0x234, 0x345, 0x456}, length=4; //***Mod
init_by_array(init, length); //***Mod
// printf( "Initialization...." );
char f = 'c';
double s = 3.0;
double x = 1.0;
double t = 1.0;
double r = 5.0;
double b = 1.5;
double a[NUM_PASSES];
// printf( "....done\nStarting test........\n" );
before = clock();
BlackScholesWrapper(f, s, x, t, r, b, a);
after = clock();
// Mean, Variance and Confidence Calculations
double sum = 0;
double mean = 0;
double variance = 0;
for (i = 0; i < NUM_PASSES; i++) {
sum += a[i];
}
mean = sum/NUM_PASSES;
for(i = 0; i < NUM_PASSES; i++){
variance += (a[i]-mean)*(a[i]-mean)/(double)NUM_PASSES;
}
double conf_width = 1.96 * sqrt(variance) / sqrt ((double) NUM_PASSES);
// printf ("Black-Scholes benchmark:\n"
// "------------------------\n"
// "S %.2f\n"
// "T %.2f\n"
// "r %.2f\n"
// "b %.2f\n",
// 3.0, 1, 5.0, 1.5);
printf("Run#: %d, Mean: %f, Variance: %f, Confidence Interval: (%f, %f)\n", NUM_PASSES, mean, variance, mean-conf_width, mean+conf_width);
// printf( "Simulation Time in secs: %f\n", ((double)(after-before))/CLOCKS_PER_SEC );
return(0);
}
int main ( int argc, char * const argv[] )
{
printf( "Testing output and speed of inventors' Mersenne Twister program\n" );
printf( "\nTest of random integer generation:\n" );
unsigned long oneSeed = 4357UL;
unsigned long bigSeed[4] = { 0x123UL, 0x234UL, 0x345UL, 0x456UL };
printf( "\nTest of random integer generation:\n" );
init_by_array( bigSeed, 4 );
unsigned long i;
for( i = 0; i < 1000UL; ++i )
{
printf( "%10lu ", genrand_int32() );
if( i % 5 == 4 ) printf("\n");
}
printf( "\nTest of random real number [0,1) generation:\n" );
for( i = 0; i < 1000UL; ++i )
{
printf( "%10.8f ", genrand_real2() );
if( i % 5 == 4 ) printf("\n");
}
printf( "\nTest of random real number [0,1] generation:\n" );
init_genrand( oneSeed );
for( i = 0; i < 2000UL; ++i )
{
printf( "%10.8f ", genrand_real1() );
if( i % 5 == 4 ) printf("\n");
}
printf( "\nTest of time to generate one billion random integers:\n" );
init_genrand( oneSeed );
unsigned long junk = 0;
clock_t start = clock();
for( i = 0; i < 1000000000UL; ++i )
{
junk ^= genrand_int32();
}
clock_t stop = clock();
if( junk == 0 ) printf( "jinx\n" );
printf( "Time elapsed = " );
printf( "%8.3f", double( stop - start ) / CLOCKS_PER_SEC );
printf( " s\n" );
return 0;
}
int test_rand_save_state()
{
const unsigned nrand=10;
unsigned long init[4]={0x123, 0x234, 0x345, 0x456}, length=4;
double* rands;
unsigned i;
ellipsis_mt19937_rng* rand;
rands=(double*)malloc(nrand*sizeof(double));
rand=(ellipsis_mt19937_rng*)malloc(sizeof(ellipsis_mt19937_rng));
init_by_array(rand,init, length);
/* store the first half */
for(i=0;i<nrand/2;++i)
{
rands[i]=genrand_uniform(rand);
}
/* save the state */
save_rand_state(rand,"myrand.dat");
/* store the 2nd half */
for(i=0;i<nrand/2;++i)
{
rands[nrand/2+i]=genrand_uniform(rand);
}
/* read the state */
read_rand_state(rand,"myrand.dat");
/* check if the 2nd half is the same */
for(i=0;i<nrand/2;++i)
{
if(rands[nrand/2+i]!=genrand_uniform(rand))
return EXIT_FAILURE;
}
free(rand);
return EXIT_SUCCESS;
}
int main(void)
{
unsigned long oneSeed = 4357UL;
unsigned long bigSeed[4] = { 0x123, 0x234, 0x345, 0x456 };
printf( "Testing output and speed of mt19937ar-c*k.c\n" );
printf( "\nTest of random integer generation:\n" );
init_by_array( bigSeed, 4 );
for( i = 0; i < 1000; ++i )
{
printf( "%10lu ", genrand_int32() );
if( i % 5 == 4 ) printf("\n");
}
printf( "\nTest of random real number [0,1) generation:\n" );
for( i = 0; i < 1000; ++i )
{
printf( "%10.8f ", genrand_real2() );
if( i % 5 == 4 ) printf("\n");
}
printf( "\nTest of random real number [0,1] generation:\n" );
init_genrand( oneSeed );
for( i = 0; i < 1000; ++i )
{
printf( "%10.8f ", genrand_real1() );
if( i % 5 == 4 ) printf("\n");
}
printf( "\nTest of time to generate 300 million random integers:\n" );
init_genrand( oneSeed );
start = clock();
for( i = 0; i < 300000000; ++i )
{
junk = genrand_int32();
}
stop = clock();
printf( "Time elapsed = " );
printf( "%8.3f", (double)( stop - start ) / CLOCKS_PER_SEC );
printf( " s\n" );
return 0;
}
static int seed_rng (lua_State *L) {
nl_RNG *r = getrng(L);
if (lua_isnoneornil(L, 1))
init_genrand(r, RNG_SEED);
else if (lua_isnumber(L, 1)) /* seed? */
init_genrand(r, lua_tointeger(L, 1));
else { /* vector */
unsigned long initkey[RNG_MAXSTATES];
int i, k;
lua_Number *e;
nl_Matrix *m = nl_checkmatrix(L, 1);
checkrvector(L, m, 1);
for (i = 0, e = m->data; i < m->size; i++, e += m->stride) {
lua_number2int(k, *e);
initkey[i] = (unsigned long) k;
}
init_by_array(r, initkey, m->size);
}
return 0;
}
int main(void)
{
int i;
unsigned long init[4]={0x123, 0x234, 0x345, 0x456}, length=4;
init_by_array(init, length);
/* This is an example of initializing by an array. */
/* You may use init_genrand(seed) with any 32bit integer */
/* as a seed for a simpler initialization */
printf("1000 outputs of genrand_int32()\n");
for (i=0; i<1000; i++) {
printf("%10lu ", genrand_int32());
if (i%5==4) printf("\n");
}
printf("\n1000 outputs of genrand_real2()\n");
for (i=0; i<1000; i++) {
printf("%10.8f ", genrand_real2());
if (i%5==4) printf("\n");
}
return 0;
}
int main( int argc, char ** argv ) {
std::string OutputFileName( argv[ 1 ] );
if( argc >= 4 ) {
unsigned long init[4]={0x123, 0x234, 0x345, 0x456}, length=4;
init_by_array(init, length);
for( int i = 0; i < 10; ++i ) {
Simulation Australia( atol( argv[ 2 ] ) );
Australia.Start( OutputFileName.c_str( ), atoi( argv[ 3 ] ) );
Australia.Stop( );
OutputFileName += "1";
}
}
else {
std::cout << "Error: missing arguments" << std::endl;
std::cout << "Reminder: Simulation OutputFileName MaxNumberOfRabbits LifeTimeInMonths" << std::endl;
}
return 0;
}
int test_random_uni()
{
unsigned long init[4]={0x123, 0x234, 0x345, 0x456}, length=4;
double s=0;
double s2=0;
double u;
double mu;
double var;
const unsigned long nrand=10000000;
unsigned long i;
ellipsis_mt19937_rng* rand;
rand=(ellipsis_mt19937_rng*)malloc(sizeof(ellipsis_mt19937_rng));
init_by_array(rand,init, length);
/* find the mean and the variance */
for(i=0;i<nrand;++i)
{
u=genrand_uniform(rand);
s+=u;
s2+=u*u;
}
mu=s/(double)nrand;
var=(s2-s*s/(double)nrand)/(double)(nrand-1);
/* mean should be 0.5 and the variance should be 1/12 */
if(fabs(mu - 0.5)>1e-3 || fabs(var-1./12.)>1e-3)
{
free(rand);
return EXIT_FAILURE;
}
free(rand);
return EXIT_SUCCESS;
}
int main(int argc, char **argv) {
tap_n((3+5)*4 + (2+5)*2 + (2+5));
if (argc > 1) {
long s = atol(argv[1]);
init_by_array((uint32_t*)&s, sizeof s / sizeof(uint32_t));
} else {
time_t s = time(NULL);
init_by_array((uint32_t*)&s, sizeof s / sizeof(uint32_t));
}
void *data = malloc(DATA_LEN);
random_fill(data, DATA_LEN);
const size_t codepoints = DATA_LEN * CHAR_BIT / 21;
const mushcell pos = MUSHCELL_MAX - WRAP_AFTER;
mushcoords beg = MUSHCOORDS(pos,pos,pos), end;
void *space_buf = malloc(mushspace_sizeof);
mushspace *space;
bool ok;
mushbounds bounds, expected_loose, expected_tight;
size_t spaces_beg, spaces_end;
codepoint_reader cp_reader;
#define BOUNDS_CHECK \
mushspace_get_loose_bounds(space, &bounds); \
\
tap_leqcos(bounds.beg, expected_loose.beg, \
"get_loose_bounds reports appropriate beg", \
"get_loose_bounds reports too large beg"); \
tap_geqcos(bounds.end, expected_loose.end, \
"get_loose_bounds reports appropriate end", \
"get_loose_bounds reports too small end"); \
\
ok = mushspace_get_tight_bounds(space, &bounds); \
tap_bool(ok, "get_tight_bounds says that the space is nonempty", \
"get_tight_bounds says that the space is empty"); \
\
tap_eqcos(bounds.beg, expected_tight.beg, \
"get_tight_bounds reports correct beg", \
"get_tight_bounds reports incorrect beg"); \
tap_eqcos(bounds.end, expected_tight.end, \
"get_tight_bounds reports correct end", \
"get_tight_bounds reports incorrect end"); \
#define LOAD_STRING(suf, T, ENCODER, BLOWUP, FOREACH_CP) \
space = mushspace_init(space_buf, NULL); \
if (!space) { \
tap_not_ok("init returned null"); \
tap_skip_remaining("init failed"); \
return 1; \
} \
tap_ok("init succeeded"); \
\
T *encoded_data##suf = (T*)data; \
size_t encoded_len##suf = DATA_LEN / sizeof *encoded_data##suf; \
if (BLOWUP) { \
encoded_data##suf = \
malloc((codepoints * BLOWUP) * sizeof *encoded_data##suf); \
encoded_len##suf = ENCODER(data, encoded_data##suf, codepoints); \
} \
\
mushspace_load_string##suf( \
space, encoded_data##suf, encoded_len##suf, &end, beg, true); \
\
if (BLOWUP) \
free(encoded_data##suf); \
\
size_t ii##suf = 0; \
mushcell cp##suf; \
\
spaces_beg = 0; \
spaces_end = 0; \
\
FOREACH_CP(suf) { \
if (ii##suf <= WRAP_AFTER) { \
if (cp##suf == ' ') \
++spaces_end; \
else \
spaces_end = 0; \
continue; \
} \
if (cp##suf == ' ') \
++spaces_beg; \
else \
break; \
} \
\
expected_loose.beg = expected_tight.beg = \
MUSHCOORDS(MUSHCELL_MIN, beg.y, beg.z); \
expected_loose.end = expected_tight.end = \
MUSHCOORDS(MUSHCELL_MAX, beg.y, beg.z); \
\
expected_loose.beg.x = expected_tight.beg.x += spaces_beg; \
expected_loose.end.x = expected_tight.end.x -= spaces_end; \
\
tap_eqcos(end, expected_tight.end, \
"load_string" #suf " reports correct end", \
"load_string" #suf " reports incorrect end"); \
\
ok = true; \
FOREACH_CP(suf) { \
mushcell gc = \
mushspace_get(space, mushcoords_add(beg, MUSHCOORDS(ii##suf,0,0))); \
if (gc == cp##suf) \
continue; \
ok = false; \
tap_not_ok("get doesn't match data given to load_string" #suf); \
printf(" ---\n" \
" failed index: %zu\n" \
" expected: %" MUSHCELL_PRIx "\n" \
" got: %" MUSHCELL_PRIx "\n" \
" ...\n", \
ii##suf, cp##suf, gc); \
break; \
} \
if (ok) \
tap_ok("get matches data given to load_string" #suf); \
\
BOUNDS_CHECK; \
mushspace_free(space);
#define DIRECT_FOREACH_CP(s) \
ii##s = 0; \
for (; ii##s < encoded_len##s && (cp##s = encoded_data##s[ii##s], true); \
++ii##s)
#define READER_FOREACH_CP(s) \
cp_reader = make_codepoint_reader(data, codepoints); \
for (ii##s = 0; (cp##s = next_codepoint(&cp_reader)) != UINT32_MAX; ++ii##s)
LOAD_STRING(, unsigned char, dummy, 0, DIRECT_FOREACH_CP);
LOAD_STRING(_cell, mushcell, dummy, 0, DIRECT_FOREACH_CP);
LOAD_STRING(_utf8, uint8_t, encode_utf8, 4, READER_FOREACH_CP);
LOAD_STRING(_utf16, uint16_t, encode_utf16, 2, READER_FOREACH_CP);
const mushcell *data_cell = (const mushcell*)data;
const size_t data_cell_count = DATA_LEN / sizeof *data_cell;
spaces_beg = 0;
spaces_end = 0;
for (size_t i = WRAP_AFTER+1; i < data_cell_count && data_cell[i++] == ' ';)
++spaces_beg;
for (size_t i = WRAP_AFTER+1; i-- > 0 && data_cell[i] == ' ';)
++spaces_end;
expected_loose.beg = expected_tight.beg =
MUSHCOORDS(MUSHCELL_MIN, beg.y, beg.z);
expected_loose.end = expected_tight.end =
MUSHCOORDS(MUSHCELL_MAX, beg.y, beg.z);
expected_loose.beg.x = expected_tight.beg.x += spaces_beg;
expected_loose.end.x = expected_tight.end.x -= spaces_beg;
#define PUT(FOREACH_CELL, S) \
space = mushspace_init(space_buf, NULL); \
if (!space) { \
tap_not_ok("init returned null"); \
tap_skip_remaining("init failed"); \
return 1; \
} \
tap_ok("init succeeded"); \
FOREACH_CELL { \
mushspace_put(space, mushcoords_add(beg, MUSHCOORDS(i, 0, 0)), \
data_cell[i]); \
} \
\
ok = true; \
for (size_t i = 0; i < data_cell_count; ++i) { \
mushcell dc = data_cell[i], \
gc = mushspace_get(space, \
mushcoords_add(beg, MUSHCOORDS(i, 0, 0))); \
if (gc == dc) \
continue; \
ok = false; \
tap_not_ok("get doesn't match what was put" S); \
printf(" ---\n" \
" failed index: %zu\n" \
" expected: %" MUSHCELL_PRIx "\n" \
" got: %" MUSHCELL_PRIx "\n" \
" ...\n", \
i, dc, gc); \
break; \
} \
if (ok) \
tap_ok("get matches what was put" S); \
\
BOUNDS_CHECK;
#define FORWARD for (size_t i = 0; i < data_cell_count; ++i)
#define REVERSE for (size_t i = data_cell_count; i--;)
PUT(FORWARD, " (forward order)");
mushspace_free(space);
PUT(REVERSE, " (reverse order)");
if (!ok) {
tap_skip_remaining("won't copy bad space");
return 1;
}
void *space_buf2 = malloc(mushspace_sizeof);
mushspace *space2 = mushspace_copy(space_buf2, space, NULL);
mushspace_free(space);
free(space_buf);
space = space2;
if (!space) {
tap_not_ok("copy returned null");
tap_skip_remaining("copy failed");
return 1;
}
tap_ok("copy succeeded");
ok = true;
for (size_t i = 0; i < DATA_LEN / sizeof *data_cell; ++i) {
mushcell dc = data_cell[i],
gc = mushspace_get(space,
mushcoords_add(beg, MUSHCOORDS(i, 0, 0)));
if (gc == dc)
continue;
ok = false;
tap_not_ok("get in copy doesn't match data");
printf(" ---\n"
" failed index: %zu\n"
" expected: %" MUSHCELL_PRIx "\n"
" got: %" MUSHCELL_PRIx "\n"
" ...\n",
i, dc, gc);
break;
}
if (ok)
tap_ok("get in copy matched data");
BOUNDS_CHECK;
free(data);
mushspace_free(space);
free(space_buf2);
}
LIS_INT lis_idr1(LIS_SOLVER solver)
{
LIS_MATRIX A;
LIS_VECTOR b,x;
LIS_VECTOR r,t,v,av,*dX,*dR,*P;
LIS_SCALAR om, h;
LIS_SCALAR M,m,c;
LIS_REAL bnrm2, nrm2, tol;
LIS_REAL angle;
LIS_INT i,j,k,s,oldest;
LIS_INT iter,maxiter,n,output,conv;
double times,ptimes,tim;
unsigned long init[4]={0x123, 0x234, 0x345, 0x456}, length=4;
LIS_DEBUG_FUNC_IN;
A = solver->A;
b = solver->b;
x = solver->x;
n = A->n;
maxiter = solver->options[LIS_OPTIONS_MAXITER];
output = solver->options[LIS_OPTIONS_OUTPUT];
conv = solver->options[LIS_OPTIONS_CONV_COND];
s = 1;
ptimes = 0.0;
r = solver->work[0];
t = solver->work[1];
v = solver->work[2];
av = solver->work[3];
P = &solver->work[4];
dX = &solver->work[4+s];
dR = &solver->work[4+2*s];
angle = 0.7;
/* Initial Residual */
if( lis_solver_get_initial_residual(solver,NULL,NULL,r,&bnrm2) )
{
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
tol = solver->tol;
init_by_array(init, length);
for(i=0;i<n;i++)
{
P[0]->value[i] = genrand_real1();
}
/*
lis_vector_copy(r,P[0]);
*/
lis_idrs_orth(s,P);
#ifdef PRE_RIGHT
times = lis_wtime();
lis_psolve(solver, r, dX[0]);
ptimes += lis_wtime()-times;
LIS_MATVEC(A,dX[0],dR[0]);
#else
#ifdef PRE_BOTH
times = lis_wtime();
lis_psolve_right(solver, r, t);
ptimes += lis_wtime()-times;
LIS_MATVEC(A,t,av);
lis_vector_print(av);
times = lis_wtime();
lis_psolve_left(solver, av, v);
ptimes += lis_wtime()-times;
#endif
#endif
/*
lis_idrs_omega(dR[k],r,angle,&om);
*/
lis_vector_dot(dR[0],dR[0],&h);
lis_vector_dot(dR[0],r,&om);
om = om / h;
lis_vector_scale(om,dX[0]);
lis_vector_scale(-om,dR[0]);
lis_vector_axpy(1.0,dX[0],x);
lis_vector_axpy(1.0,dR[0],r);
/* convergence check */
lis_solver_get_residual[conv](r,solver,&nrm2);
if( output )
{
if( output & LIS_PRINT_MEM ) solver->residual[1] =
nrm2;
if( output & LIS_PRINT_OUT && A->my_rank==0 )
printf("iter: %5d residual = %e\n", 1, nrm2);
}
if( tol >= nrm2 )
{
solver->retcode = LIS_SUCCESS;
solver->iter = 1;
solver->resid = nrm2;
solver->ptimes = ptimes;
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
lis_vector_dot(P[0],dR[0],&M);
iter = s;
oldest = 0;
lis_vector_dot(P[0],r,&m);
while( iter<=maxiter )
{
tim = lis_wtime();
/* solve Mc=m */
c = m/M;
for(i=0;i<n;i++)
{
v->value[i] = r->value[i] + -c*dR[0]->value[i];
}
/*
lis_vector_copy(r,v);
lis_vector_axpy(-c,dR[0],v);
*/
#ifdef PRE_RIGHT
times = lis_wtime();
lis_psolve(solver, v, av);
ptimes += lis_wtime()-times;
LIS_MATVEC(A,av,t);
#else
#ifdef PRE_BOTH
times = lis_wtime();
lis_psolve_right(solver, v, t);
ptimes += lis_wtime()-times;
LIS_MATVEC(A,t,av);
times = lis_wtime();
lis_psolve_left(solver, av, t);
ptimes += lis_wtime()-times;
#endif
#endif
/*
lis_idrs_omega(t,v,angle,&om);
lis_vector_dot(t,t,&h);
lis_vector_dot(t,v,&om);
*/
h = t->value[0]*t->value[0];
om = t->value[0]*v->value[0];
for(i=1;i<n;i++)
{
h += t->value[i]*t->value[i];
om += t->value[i]*v->value[i];
}
om = om / h;
/*
printf("i=%d om = %lf\n",iter,om);
*/
#if 0
lis_vector_scale(-om,t);
for(j=0;j<s;j++)
{
lis_vector_axpy(-c[j],dR[j],t);
}
lis_vector_copy(t,dR[oldest]);
lis_vector_scale(om,av);
for(j=0;j<s;j++)
{
lis_vector_axpy(-c[j],dX[j],av);
}
lis_vector_copy(av,dX[oldest]);
#else
for(i=0;i<n;i++)
{
h = om*av->value[i];
h -= dX[0]->value[i] * c;
dX[0]->value[i] = h;
h = -om*t->value[i];
h -= dR[0]->value[i] * c;
dR[0]->value[i] = h;
}
#endif
lis_vector_axpy(1.0,dR[0],r);
lis_vector_axpy(1.0,dX[0],x);
iter++;
/* convergence check */
lis_solver_get_residual[conv](r,solver,&nrm2);
if( output )
{
if( output & LIS_PRINT_MEM ) solver->residual[iter]
= nrm2;
if( output & LIS_PRINT_OUT && A->my_rank==0 )
printf("iter: %5d residual = %e\n", iter, nrm2);
}
if( tol >= nrm2 )
{
solver->retcode = LIS_SUCCESS;
solver->iter = iter;
solver->resid = nrm2;
solver->ptimes = ptimes;
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
lis_vector_dot(P[0],dR[0],&h);
m += h;
M = h;
/* solve Mc=m */
c = m/M;
for(i=0;i<n;i++)
{
v->value[i] = r->value[i] + -c*dR[0]->value[i];
}
/*
lis_vector_copy(r,v);
lis_vector_axpy(-c,dR[0],v);
*/
#ifdef PRE_RIGHT
times = lis_wtime();
lis_psolve(solver, v, av);
ptimes += lis_wtime()-times;
#endif
#if 0
lis_vector_scale(om,av);
for(j=0;j<s;j++)
{
lis_vector_axpy(-c[j],dX[j],av);
}
lis_vector_copy(av,dX[oldest]);
#else
for(i=0;i<n;i++)
{
h = om*av->value[i];
h -= dX[0]->value[i] * c;
dX[0]->value[i] = h;
}
#endif
LIS_MATVEC(A,dX[0],dR[0]);
lis_vector_scale(-1.0,dR[0]);
lis_vector_axpy(1.0,dR[0],r);
lis_vector_axpy(1.0,dX[0],x);
iter++;
/* convergence check */
lis_solver_get_residual[conv](r,solver,&nrm2);
if( output )
{
if( output & LIS_PRINT_MEM ) solver->residual[iter]
= nrm2;
if( output & LIS_PRINT_OUT && A->my_rank==0 )
printf("iter: %5d residual = %e\n", iter, nrm2);
}
if( tol >= nrm2 )
{
solver->retcode = LIS_SUCCESS;
solver->iter = iter;
solver->resid = nrm2;
solver->ptimes = ptimes;
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
lis_vector_dot(P[0],dR[0],&h);
m += h;
M = h;
tim = lis_wtime() - tim;
/*
printf("update m,M: %e\n",tim);
*/
}
solver->retcode = LIS_MAXITER;
solver->iter = iter;
solver->resid = nrm2;
LIS_DEBUG_FUNC_OUT;
return LIS_MAXITER;
}
LIS_INT lis_idrs(LIS_SOLVER solver)
{
LIS_MATRIX A;
LIS_VECTOR b,x;
LIS_VECTOR r,t,v,av,*dX,*dR,*P;
LIS_SCALAR om, h;
LIS_SCALAR *M,*m,*c,*MM;
LIS_REAL bnrm2, nrm2, tol;
LIS_REAL angle;
LIS_INT i,j,k,s,oldest;
LIS_INT iter,maxiter,n,output,conv;
double times,ptimes,tim;
unsigned long init[4]={0x123, 0x234, 0x345, 0x456}, length=4;
LIS_DEBUG_FUNC_IN;
A = solver->A;
b = solver->b;
x = solver->x;
n = A->n;
maxiter = solver->options[LIS_OPTIONS_MAXITER];
output = solver->options[LIS_OPTIONS_OUTPUT];
conv = solver->options[LIS_OPTIONS_CONV_COND];
s = solver->options[LIS_OPTIONS_IDRS_RESTART];
ptimes = 0.0;
r = solver->work[0];
t = solver->work[1];
v = solver->work[2];
av = solver->work[3];
dX = &solver->work[4];
P = &solver->work[4+s];
dR = &solver->work[4+2*s];
angle = 0.7;
m = (LIS_SCALAR *)lis_malloc(s*sizeof(LIS_SCALAR), "lis_idrs::m");
c = (LIS_SCALAR *)lis_malloc(s*sizeof(LIS_SCALAR), "lis_idrs::c");
M = (LIS_SCALAR *)lis_malloc(s*s*sizeof(LIS_SCALAR), "lis_idrs::M");
MM = (LIS_SCALAR *)lis_malloc(s*s*sizeof(LIS_SCALAR),
"lis_idrs::M");
/* Initial Residual */
if( lis_solver_get_initial_residual(solver,NULL,NULL,r,&bnrm2) )
{
lis_free2(4,m,c,M,MM);
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
tol = solver->tol;
init_by_array(init, length);
for(k=0;k<s;k++)
{
for(i=0;i<n;i++)
{
P[k]->value[i] = genrand_real1();
}
}
lis_idrs_orth(s,P);
for( k=0; k<s; k++ )
{
#ifdef PRE_RIGHT
times = lis_wtime();
lis_psolve(solver, r, dX[k]);
ptimes += lis_wtime()-times;
LIS_MATVEC(A,dX[k],dR[k]);
#endif
lis_vector_dot(dR[k],dR[k],&h);
lis_vector_dot(dR[k],r,&om);
om = om / h;
lis_vector_scale(om,dX[k]);
lis_vector_scale(-om,dR[k]);
lis_vector_axpy(1.0,dX[k],x);
lis_vector_axpy(1.0,dR[k],r);
/* convergence check */
lis_solver_get_residual[conv](r,solver,&nrm2);
if( output )
{
if( output & LIS_PRINT_MEM ) solver->residual[k+1] =
nrm2;
if( output & LIS_PRINT_OUT && A->my_rank==0 )
printf("iter: %5d residual = %e\n", k+1, nrm2);
}
if( tol >= nrm2 )
{
lis_free2(4,m,c,M,MM);
solver->retcode = LIS_SUCCESS;
solver->iter = k+1;
solver->resid = nrm2;
solver->ptimes = ptimes;
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
for(i=0;i<s;i++)
{
lis_vector_dot(P[i],dR[k],&M[k*s+i]);
}
}
iter = s;
oldest = 0;
for(i=0;i<s;i++)
{
lis_vector_dot(P[i],r,&m[i]);
}
while( iter<=maxiter )
{
tim = lis_wtime();
lis_array_solve(s,M,m,c,MM); /* solve Mc=m */
lis_vector_copy(r,v);
for(j=0;j<s;j++)
{
lis_vector_axpy(-c[j],dR[j],v);
}
if( (iter%(s+1))==s )
{
#ifdef PRE_RIGHT
times = lis_wtime();
lis_psolve(solver, v, av);
ptimes += lis_wtime()-times;
LIS_MATVEC(A,av,t);
#endif
lis_vector_dot(t,t,&h);
lis_vector_dot(t,v,&om);
om = om / h;
#if 0
lis_vector_scale(-om,t);
for(j=0;j<s;j++)
{
lis_vector_axpy(-c[j],dR[j],t);
}
lis_vector_copy(t,dR[oldest]);
lis_vector_scale(om,av);
for(j=0;j<s;j++)
{
lis_vector_axpy(-c[j],dX[j],av);
}
lis_vector_copy(av,dX[oldest]);
#else
for(i=0;i<n;i++)
{
h = om*av->value[i];
for(j=0;j<s;j++)
{
h -= dX[j]->value[i] * c[j];
}
dX[oldest]->value[i] = h;
}
for(i=0;i<n;i++)
{
h = -om*t->value[i];
for(j=0;j<s;j++)
{
h -= dR[j]->value[i] * c[j];
}
dR[oldest]->value[i] = h;
}
#endif
}
else
{
#ifdef PRE_RIGHT
times = lis_wtime();
lis_psolve(solver, v, av);
ptimes += lis_wtime()-times;
#endif
#if 0
lis_vector_scale(om,av);
for(j=0;j<s;j++)
{
lis_vector_axpy(-c[j],dX[j],av);
}
lis_vector_copy(av,dX[oldest]);
#else
for(i=0;i<n;i++)
{
h = om*av->value[i];
for(j=0;j<s;j++)
{
h -= dX[j]->value[i] * c[j];
}
dX[oldest]->value[i] = h;
}
#endif
LIS_MATVEC(A,dX[oldest],dR[oldest]);
lis_vector_scale(-1.0,dR[oldest]);
}
lis_vector_axpy(1.0,dR[oldest],r);
lis_vector_axpy(1.0,dX[oldest],x);
iter++;
/* convergence check */
lis_solver_get_residual[conv](r,solver,&nrm2);
if( output )
{
if( output & LIS_PRINT_MEM ) solver->residual[iter]
= nrm2;
if( output & LIS_PRINT_OUT && A->my_rank==0 )
printf("iter: %5d residual = %e\n", iter, nrm2);
}
if( tol >= nrm2 )
{
lis_free2(4,m,c,M,MM);
solver->retcode = LIS_SUCCESS;
solver->iter = iter;
solver->resid = nrm2;
solver->ptimes = ptimes;
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
for(i=0;i<s;i++)
{
lis_vector_dot(P[i],dR[oldest],&h);
m[i] += h;
M[oldest*s+i] = h;
}
oldest++;
if( oldest==s ) oldest = 0;
tim = lis_wtime() - tim;
/*
printf("update m,M: %e\n",tim);
*/
}
lis_free2(4,m,c,M,MM);
solver->retcode = LIS_MAXITER;
solver->iter = iter;
solver->resid = nrm2;
LIS_DEBUG_FUNC_OUT;
return LIS_MAXITER;
}
int main(int argc, char **argv) {
// We don't check bounds here: we'd end up essentially duplicating the logic
// from load_string, and there's little point in that.
tap_n(2*4 + 2*2 + 2);
if (argc > 1) {
long s = atol(argv[1]);
init_by_array((uint32_t*)&s, sizeof s / sizeof(uint32_t));
} else {
time_t s = time(NULL);
init_by_array((uint32_t*)&s, sizeof s / sizeof(uint32_t));
}
void *data = malloc(DATA_LEN);
random_fill(data, DATA_LEN);
const size_t codepoints = DATA_LEN * CHAR_BIT / 21;
mushcoords beg = MUSHCOORDS(MUSHCELL_MAX-WRAP_AFTER,
MUSHCELL_MAX-WRAP_AFTER,
MUSHCELL_MAX-WRAP_AFTER),
end;
void *space_buf = malloc(mushspace_sizeof);
mushspace *space;
bool ok;
codepoint_reader cp_reader;
mushcoords pos, pos_next;
bool cr;
#define POS_INIT \
pos_next = beg; \
cr = false;
#define CP_POS(cp) \
pos = pos_next; \
switch (cp) { \
default: \
if (MUSHSPACE_DIM > 1 && cr) { \
cr = false; \
pos = MUSHCOORDS(beg.x, pos.y + 1, pos.z); \
} \
pos_next = MUSHCOORDS(pos.x + 1, pos.y, pos.z); \
break; \
case '\r': \
if (MUSHSPACE_DIM > 1) { \
if (cr) \
pos_next = pos = MUSHCOORDS(beg.x, pos.y + 1, pos.z); \
cr = true; \
} \
continue; \
case '\n': \
if (MUSHSPACE_DIM > 1) { \
cr = false; \
pos_next = pos = MUSHCOORDS(beg.x, pos.y + 1, pos.z); \
} \
continue; \
case '\f': \
if (MUSHSPACE_DIM > 2) { \
cr = false; \
pos_next = pos = MUSHCOORDS(beg.x, beg.y, pos.z + 1); \
} else if (cr) { \
cr = false; \
pos_next = pos = MUSHCOORDS(beg.x, pos.y + 1, pos.z); \
} \
continue; \
}
#define LOAD_STRING(suf, T, ENCODER, BLOWUP, FOREACH_CP) \
space = mushspace_init(space_buf, NULL); \
if (!space) { \
tap_not_ok("init returned null"); \
tap_skip_remaining("init failed"); \
return 1; \
} \
tap_ok("init succeeded"); \
\
T *encoded_data##suf = (T*)data; \
size_t encoded_len##suf = DATA_LEN / sizeof *encoded_data##suf; \
if (BLOWUP) { \
encoded_data##suf = \
malloc((codepoints * BLOWUP) * sizeof *encoded_data##suf); \
encoded_len##suf = ENCODER(data, encoded_data##suf, codepoints); \
} \
\
mushspace_load_string##suf( \
space, encoded_data##suf, encoded_len##suf, &end, beg, false); \
\
if (BLOWUP) \
free(encoded_data##suf); \
\
ok = true; \
POS_INIT; \
FOREACH_CP(suf) { \
CP_POS(cp##suf); \
mushcell gc = mushspace_get(space, pos); \
if (gc == cp##suf) \
continue; \
ok = false; \
tap_not_ok("get doesn't match data given to load_string" #suf); \
printf(" ---\n" \
" expected: %" MUSHCELL_PRIx "\n" \
" got: %" MUSHCELL_PRIx "\n" \
" failed index: %zu\n", \
cp##suf, gc, ii##suf); \
printf(" failed pos relative to min: ("); \
for (uint8_t i = 0; i < MUSHSPACE_DIM; ++i) \
printf(" %" MUSHCELL_PRId, mushcell_sub(pos.v[i], MUSHCELL_MIN)); \
printf(" )\n" \
" ...\n"); \
break; \
} \
if (ok) \
tap_ok("get matches data given to load_string" #suf); \
\
mushspace_free(space);
#define DIRECT_FOREACH_CP(s) \
mushcell cp##s; \
size_t ii##s = 0; \
for (; ii##s < encoded_len##s && (cp##s = encoded_data##s[ii##s], true); \
++ii##s)
#define READER_FOREACH_CP(s) \
cp_reader = make_codepoint_reader(data, codepoints); \
size_t ii##s = 0; \
for (mushcell cp##s; (cp##s = next_codepoint(&cp_reader)) != UINT32_MAX; \
++ii##s)
LOAD_STRING(, unsigned char, dummy, 0, DIRECT_FOREACH_CP);
LOAD_STRING(_cell, mushcell, dummy, 0, DIRECT_FOREACH_CP);
LOAD_STRING(_utf8, uint8_t, encode_utf8, 4, READER_FOREACH_CP);
LOAD_STRING(_utf16, uint16_t, encode_utf16, 2, READER_FOREACH_CP);
const mushcell *data_cell = (const mushcell*)data;
const size_t data_cell_count = DATA_LEN / sizeof *data_cell;
mushcoords *saved_pos = malloc(data_cell_count * sizeof *saved_pos);
#define PUT(FOREACH_CELL, S, GET_POS, SAVE_POS) \
space = mushspace_init(space_buf, NULL); \
if (!space) { \
tap_not_ok("init returned null"); \
tap_skip_remaining("init failed"); \
free(saved_pos); \
return 1; \
} \
tap_ok("init succeeded"); \
POS_INIT; \
FOREACH_CELL { \
GET_POS(data_cell[i]); \
if (SAVE_POS) \
saved_pos[i] = pos; \
mushspace_put(space, pos, data_cell[i]); \
} \
\
ok = true; \
POS_INIT; \
for (size_t i = 0; i < data_cell_count; ++i) { \
mushcell dc = data_cell[i]; \
GET_POS(dc); \
mushcell gc = mushspace_get(space, pos); \
if (gc == dc) \
continue; \
ok = false; \
tap_not_ok("get doesn't match what was put" S); \
printf(" ---\n" \
" failed index: %zu\n" \
" expected: %" MUSHCELL_PRIx "\n" \
" got: %" MUSHCELL_PRIx "\n", \
i, dc, gc); \
printf(" failed pos relative to min: ("); \
for (uint8_t j = 0; j < MUSHSPACE_DIM; ++j) \
printf(" %" MUSHCELL_PRId, mushcell_sub(pos.v[j], MUSHCELL_MIN)); \
printf(" )\n" \
" ...\n"); \
break; \
} \
if (ok) \
tap_ok("get matches what was put" S);
#define FORWARD for (size_t i = 0; i < data_cell_count; ++i)
#define REVERSE for (size_t i = data_cell_count; i--;)
#define SAVED_POS(_) pos = saved_pos[i];
PUT(FORWARD, " (forward order)", CP_POS, true);
mushspace_free(space);
PUT(REVERSE, " (reverse order)", SAVED_POS, false);
free(saved_pos);
if (!ok) {
tap_skip_remaining("won't copy bad space");
return 1;
}
void *space_buf2 = malloc(mushspace_sizeof);
mushspace *space2 = mushspace_copy(space_buf2, space, NULL);
mushspace_free(space);
free(space_buf);
space = space2;
if (!space) {
tap_not_ok("copy returned null");
tap_skip_remaining("copy failed");
return 1;
}
tap_ok("copy succeeded");
ok = true;
POS_INIT;
for (size_t i = 0; i < DATA_LEN / sizeof *data_cell; ++i) {
mushcell dc = data_cell[i];
CP_POS(dc);
mushcell gc = mushspace_get(space, pos);
if (gc == dc)
continue;
ok = false;
tap_not_ok("get in copy doesn't match data");
printf(" ---\n"
" failed index: %zu\n"
" expected: %" MUSHCELL_PRIx "\n"
" got: %" MUSHCELL_PRIx "\n",
i, dc, gc);
printf(" failed pos relative to min: (");
for (uint8_t j = 0; j < MUSHSPACE_DIM; ++j)
printf(" %" MUSHCELL_PRId, mushcell_sub(pos.v[j], MUSHCELL_MIN)); \
printf(" )\n"
" ...\n");
break;
}
if (ok)
tap_ok("get in copy matched data");
free(data);
mushspace_free(space);
free(space_buf2);
}
void calc_post_equal_weights_samples(live_point* psamps,unsigned num_post_samps,double logz,unsigned num_dim,char* file_name)
{
unsigned i,k;
unsigned num_eff_posts;
ellipsis_mt19937_rng* rand;
unsigned long rand_init[4]= {0x123, 0x234, 0x345, 0x456};
int rand_length=4;
double uni_rand;
double S_k;
double eff_log_num_samples=0;
double* weight=(double*)malloc(num_post_samps*sizeof(double));
double sum_wt;
live_point* post_equal_weight_samples;
/* allocate memroy for random number generator*/
rand=(ellipsis_mt19937_rng*)malloc(sizeof(ellipsis_mt19937_rng));
init_by_array(rand,rand_init, rand_length);
/*calculate effective number of samples */
for(i=0; i<num_post_samps; ++i)
{
weight[i]=exp(psamps[i].log_weight-logz);
if(weight[i]>0)
{
eff_log_num_samples-=weight[i]*log(weight[i]);
}
}
num_eff_posts=(unsigned)floor((double)exp(eff_log_num_samples));
/*check if it is less than number of posterior samples*/
if(num_post_samps<num_eff_posts)
num_eff_posts=num_post_samps;
/*allocate eqully weighted posterior samples */
post_equal_weight_samples=(live_point*)malloc(num_eff_posts*sizeof(live_point));
for(i=0; i<num_eff_posts; ++i)
{
init_live_point(&post_equal_weight_samples[i],num_dim);
}
/*staircase sampling */
sum_wt=0;
k=0;
for(i=0; i<num_post_samps; ++i)
{
uni_rand=genrand_uniform(rand);
sum_wt+=weight[i];
S_k=uni_rand+num_eff_posts*sum_wt;
if(S_k-(double)k > 1.)
{
copy_live_point(&post_equal_weight_samples[k],&psamps[i]);
++k;
}
}
/* once again check if have the expected number of posterior samples */
if(k<num_eff_posts)
num_eff_posts=k;
/*write to txt file*/
write_live_points_to_txt_file(file_name,post_equal_weight_samples,num_eff_posts,num_dim);
/*clean up*/
free(rand);
free(weight);
free(post_equal_weight_samples);
}
