void bli_zdot( conj_t conj, int n, dcomplex* x, int incx, dcomplex* y, int incy, dcomplex* rho )
{
#ifdef BLIS_ENABLE_CBLAS_INTERFACES
if ( bli_is_conj( conj ) )
{
cblas_zdotc_sub( n,
x, incx,
y, incy,
rho );
}
else // if ( !bli_is_conj( conj ) )
{
cblas_zdotu_sub( n,
x, incx,
y, incy,
rho );
}
#else
bli_zdot_in( conj,
n,
x, incx,
y, incy,
rho );
#endif
}
double wig20rot(complx *vec, complx *d20)
{
complx res;
const int len=5;
cblas_zdotu_sub(len,vec,INTONE,d20,INTONE,&res);
if ( fabs(res.im) > 1e-8 ) {
fprintf(stderr,"wig20rot error: result is not pure real\n");
exit(1);
}
return res.re;
}
static void
CDOUBLE_dot(void *a, intp stridea, void *b, intp strideb, void *res,
intp n, void *tmp)
{
register int na = stridea / sizeof(cdouble);
register int nb = strideb / sizeof(cdouble);
if ((sizeof(cdouble) * na == stridea) &&
(sizeof(cdouble) * nb == strideb) &&
(na >= 0) && (nb >= 0))
cblas_zdotu_sub((int)n, (double *)a, na, (double *)b, nb, (double *)res);
else
oldFunctions[PyArray_CDOUBLE](a, stridea, b, strideb, res, n, tmp);
}
void phi_dotu_sub(const int N, const Complex *X, const int incX,
const Complex *Y, const int incY, Complex *dotu){
#ifndef NOBLAS
#ifdef SINGLEPRECISION
cblas_cdotu_sub(N,X,incX,Y,incY,dotu);
#else
cblas_zdotu_sub(N,X,incX,Y,incY,dotu);
#endif
#else
int i;
*dotu = 0;
for(i = 0; i < N; ++i, X+=incX, Y+=incY){
*dotu += (*X)*(*Y);
}
#endif
}
std::complex<double> HostVector<std::complex<double> >::DotNonConj(const BaseVector<std::complex<double> > &x) const {
assert(&x != NULL);
const HostVector<std::complex<double> > *cast_x = dynamic_cast<const HostVector<std::complex<double> >*> (&x);
assert(cast_x != NULL);
assert(this->size_ == cast_x->size_);
std::complex<double> res;
cblas_zdotu_sub(this->size_, this->vec_, 1, cast_x->vec_, 1, &res);
return res;
}
// ----------------------------------------
int main( int argc, char** argv )
{
TESTING_INIT();
//real_Double_t t_m, t_c, t_f;
magma_int_t ione = 1;
magmaDoubleComplex *A, *B;
double diff, error;
magma_int_t ISEED[4] = {0,0,0,1};
magma_int_t m, n, k, size, maxn, ld;
magmaDoubleComplex x2_m, x2_c; // complex x for magma, cblas/fortran blas respectively
double x_m, x_c; // x for magma, cblas/fortran blas respectively
magma_opts opts;
parse_opts( argc, argv, &opts );
opts.tolerance = max( 100., opts.tolerance );
double tol = opts.tolerance * lapackf77_dlamch("E");
gTol = tol;
printf( "!! Calling these CBLAS and Fortran BLAS sometimes crashes (segfault), which !!\n"
"!! is why we use wrappers. It does not necesarily indicate a bug in MAGMA. !!\n"
"\n"
"Diff compares MAGMA wrapper to CBLAS and BLAS function; should be exactly 0.\n"
"Error compares MAGMA implementation to CBLAS and BLAS function; should be ~ machine epsilon.\n"
"\n" );
double total_diff = 0.;
double total_error = 0.;
int inc[] = { 1 }; //{ -2, -1, 1, 2 }; //{ 1 }; //{ -1, 1 };
int ninc = sizeof(inc)/sizeof(*inc);
for( int itest = 0; itest < opts.ntest; ++itest ) {
m = opts.msize[itest];
n = opts.nsize[itest];
k = opts.ksize[itest];
for( int iincx = 0; iincx < ninc; ++iincx ) {
magma_int_t incx = inc[iincx];
for( int iincy = 0; iincy < ninc; ++iincy ) {
magma_int_t incy = inc[iincy];
printf("=========================================================================\n");
printf( "m=%d, n=%d, k=%d, incx = %d, incy = %d\n",
(int) m, (int) n, (int) k, (int) incx, (int) incy );
printf( "Function MAGMA CBLAS BLAS Diff Error\n"
" msec msec msec\n" );
// allocate matrices
// over-allocate so they can be any combination of
// {m,n,k} * {abs(incx), abs(incy)} by
// {m,n,k} * {abs(incx), abs(incy)}
maxn = max( max( m, n ), k ) * max( abs(incx), abs(incy) );
ld = max( 1, maxn );
size = ld*maxn;
magma_zmalloc_pinned( &A, size ); assert( A != NULL );
magma_zmalloc_pinned( &B, size ); assert( B != NULL );
// initialize matrices
lapackf77_zlarnv( &ione, ISEED, &size, A );
lapackf77_zlarnv( &ione, ISEED, &size, B );
printf( "Level 1 BLAS ----------------------------------------------------------\n" );
// ----- test DZASUM
// get one-norm of column j of A
if ( incx > 0 && incx == incy ) { // positive, no incy
diff = 0;
error = 0;
for( int j = 0; j < k; ++j ) {
x_m = magma_cblas_dzasum( m, A(0,j), incx );
x_c = cblas_dzasum( m, A(0,j), incx );
diff += fabs( x_m - x_c );
x_c = blasf77_dzasum( &m, A(0,j), &incx );
error += fabs( (x_m - x_c) / (m*x_c) );
}
output( "dzasum", diff, error );
total_diff += diff;
total_error += error;
}
// ----- test DZNRM2
// get two-norm of column j of A
if ( incx > 0 && incx == incy ) { // positive, no incy
diff = 0;
error = 0;
for( int j = 0; j < k; ++j ) {
x_m = magma_cblas_dznrm2( m, A(0,j), incx );
x_c = cblas_dznrm2( m, A(0,j), incx );
diff += fabs( x_m - x_c );
x_c = blasf77_dznrm2( &m, A(0,j), &incx );
error += fabs( (x_m - x_c) / (m*x_c) );
}
output( "dznrm2", diff, error );
total_diff += diff;
total_error += error;
}
// ----- test ZDOTC
// dot columns, Aj^H Bj
diff = 0;
error = 0;
for( int j = 0; j < k; ++j ) {
// MAGMA implementation, not just wrapper
x2_m = magma_cblas_zdotc( m, A(0,j), incx, B(0,j), incy );
// crashes on MKL 11.1.2, ILP64
#if ! defined( MAGMA_WITH_MKL )
#ifdef COMPLEX
cblas_zdotc_sub( m, A(0,j), incx, B(0,j), incy, &x2_c );
#else
x2_c = cblas_zdotc( m, A(0,j), incx, B(0,j), incy );
#endif
error += fabs( x2_m - x2_c ) / fabs( m*x2_c );
#endif
// crashes on MacOS 10.9
#if ! defined( __APPLE__ )
x2_c = blasf77_zdotc( &m, A(0,j), &incx, B(0,j), &incy );
error += fabs( x2_m - x2_c ) / fabs( m*x2_c );
#endif
}
output( "zdotc", diff, error );
total_diff += diff;
total_error += error;
total_error += error;
// ----- test ZDOTU
// dot columns, Aj^T * Bj
diff = 0;
error = 0;
for( int j = 0; j < k; ++j ) {
// MAGMA implementation, not just wrapper
x2_m = magma_cblas_zdotu( m, A(0,j), incx, B(0,j), incy );
// crashes on MKL 11.1.2, ILP64
#if ! defined( MAGMA_WITH_MKL )
#ifdef COMPLEX
cblas_zdotu_sub( m, A(0,j), incx, B(0,j), incy, &x2_c );
#else
x2_c = cblas_zdotu( m, A(0,j), incx, B(0,j), incy );
#endif
error += fabs( x2_m - x2_c ) / fabs( m*x2_c );
#endif
// crashes on MacOS 10.9
#if ! defined( __APPLE__ )
x2_c = blasf77_zdotu( &m, A(0,j), &incx, B(0,j), &incy );
error += fabs( x2_m - x2_c ) / fabs( m*x2_c );
#endif
}
output( "zdotu", diff, error );
total_diff += diff;
total_error += error;
// tell user about disabled functions
#if defined( MAGMA_WITH_MKL )
printf( "cblas_zdotc and cblas_zdotu disabled with MKL (segfaults)\n" );
#endif
#if defined( __APPLE__ )
printf( "blasf77_zdotc and blasf77_zdotu disabled on MacOS (segfaults)\n" );
#endif
// cleanup
magma_free_pinned( A );
magma_free_pinned( B );
fflush( stdout );
}}} // itest, incx, incy
// TODO use average error?
printf( "sum diffs = %8.2g, MAGMA wrapper compared to CBLAS and Fortran BLAS; should be exactly 0.\n"
"sum errors = %8.2e, MAGMA implementation compared to CBLAS and Fortran BLAS; should be ~ machine epsilon.\n\n",
total_diff, total_error );
if ( total_diff != 0. ) {
printf( "some tests failed diff == 0.; see above.\n" );
}
else {
printf( "all tests passed diff == 0.\n" );
}
TESTING_FINALIZE();
int status = (total_diff != 0.);
return status;
}
// ----------------------------------------
int main( int argc, char** argv )
{
TESTING_INIT();
//real_Double_t t_m, t_c, t_f;
magma_int_t ione = 1;
magmaDoubleComplex *A, *B;
double error_cblas, error_fblas, error_inline;
magma_int_t ISEED[4] = {0,0,0,1};
magma_int_t i, j, k, m, n, size, maxn, ld;
// complex x for magma, cblas, fortran, inline blas respectively
magmaDoubleComplex x2_m, x2_c, x2_f, x2_i;
// real x for magma, cblas, fortran, inline blas respectively
double x_m, x_c, x_f, x_i;
MAGMA_UNUSED( x_c );
MAGMA_UNUSED( x_f );
MAGMA_UNUSED( x2_c );
MAGMA_UNUSED( x2_f );
MAGMA_UNUSED( x2_m );
magma_opts opts;
opts.parse_opts( argc, argv );
opts.tolerance = max( 100., opts.tolerance );
double tol = opts.tolerance * lapackf77_dlamch("E");
gTol = tol;
magma_int_t inc[] = { -2, -1, 1, 2 }; //{ 1 }; //{ -1, 1 };
magma_int_t ninc = sizeof(inc)/sizeof(*inc);
magma_int_t maxinc = 0;
for( i=0; i < ninc; ++i ) {
maxinc = max( maxinc, abs(inc[i]) );
}
printf( "!! Calling these CBLAS and Fortran BLAS sometimes crashes (segfaults), which !!\n"
"!! is why we use wrappers. It does not necesarily indicate a bug in MAGMA. !!\n"
"!! If MAGMA_WITH_MKL or __APPLE__ are defined, known failures are skipped. !!\n"
"\n" );
// tell user about disabled functions
#ifndef HAVE_CBLAS
printf( "n/a: HAVE_CBLAS not defined, so no cblas functions tested.\n\n" );
#endif
#if defined(MAGMA_WITH_MKL)
printf( "n/a: cblas_zdotc, cblas_zdotu, blasf77_zdotc, and blasf77_zdotu are disabled with MKL, due to segfaults.\n\n" );
#endif
#if defined(__APPLE__)
printf( "n/a: blasf77_zdotc and blasf77_zdotu are disabled on MacOS, due to segfaults.\n\n" );
#endif
printf( "%% Error w.r.t. Error w.r.t. Error w.r.t.\n"
"%% M N K incx incy Function CBLAS Fortran BLAS inline\n"
"%%====================================================================================\n" );
for( int itest = 0; itest < opts.ntest; ++itest ) {
if ( itest > 0 ) {
printf( "%%----------------------------------------------------------------------\n" );
}
m = opts.msize[itest];
n = opts.nsize[itest];
k = opts.ksize[itest];
// allocate matrices
// over-allocate so they can be any combination of
// {m,n,k} * {abs(incx), abs(incy)} by
// {m,n,k} * {abs(incx), abs(incy)}
maxn = max( max( m, n ), k ) * maxinc;
ld = max( 1, maxn );
size = ld*maxn;
TESTING_MALLOC_CPU( A, magmaDoubleComplex, size );
TESTING_MALLOC_CPU( B, magmaDoubleComplex, size );
// initialize matrices
lapackf77_zlarnv( &ione, ISEED, &size, A );
lapackf77_zlarnv( &ione, ISEED, &size, B );
// ----- test DZASUM
for( int iincx = 0; iincx < ninc; ++iincx ) {
magma_int_t incx = inc[iincx];
for( int iincy = 0; iincy < ninc; ++iincy ) {
magma_int_t incy = inc[iincy];
// get one-norm of column j of A
if ( incx > 0 && incx == incy ) { // positive, no incy
error_cblas = 0;
error_fblas = 0;
error_inline = 0;
for( j=0; j < k; ++j ) {
x_m = magma_cblas_dzasum( m, A(0,j), incx );
#ifdef HAVE_CBLAS
x_c = cblas_dzasum( m, A(0,j), incx );
error_cblas = max( error_cblas, fabs(x_m - x_c) / fabs(m*x_c) );
#else
x_c = 0;
error_cblas = SKIPPED_FLAG;
#endif
x_f = blasf77_dzasum( &m, A(0,j), &incx );
error_fblas = max( error_fblas, fabs(x_m - x_f) / fabs(m*x_f) );
// inline implementation
x_i = 0;
for( i=0; i < m; ++i ) {
x_i += MAGMA_Z_ABS1( *A(i*incx,j) ); // |real(Aij)| + |imag(Aij)|
}
error_inline = max( error_inline, fabs(x_m - x_i) / fabs(m*x_i) );
//printf( "dzasum xm %.8e, xc %.8e, xf %.8e, xi %.8e\n", x_m, x_c, x_f, x_i );
}
output( "dzasum", m, n, k, incx, incy, error_cblas, error_fblas, error_inline );
}
}
}
printf( "\n" );
// ----- test DZNRM2
// get two-norm of column j of A
for( int iincx = 0; iincx < ninc; ++iincx ) {
magma_int_t incx = inc[iincx];
for( int iincy = 0; iincy < ninc; ++iincy ) {
magma_int_t incy = inc[iincy];
if ( incx > 0 && incx == incy ) { // positive, no incy
error_cblas = 0;
error_fblas = 0;
error_inline = 0;
for( j=0; j < k; ++j ) {
x_m = magma_cblas_dznrm2( m, A(0,j), incx );
#ifdef HAVE_CBLAS
x_c = cblas_dznrm2( m, A(0,j), incx );
error_cblas = max( error_cblas, fabs(x_m - x_c) / fabs(m*x_c) );
#else
x_c = 0;
error_cblas = SKIPPED_FLAG;
#endif
x_f = blasf77_dznrm2( &m, A(0,j), &incx );
error_fblas = max( error_fblas, fabs(x_m - x_f) / fabs(m*x_f) );
// inline implementation (poor -- doesn't scale)
x_i = 0;
for( i=0; i < m; ++i ) {
x_i += real( *A(i*incx,j) ) * real( *A(i*incx,j) )
+ imag( *A(i*incx,j) ) * imag( *A(i*incx,j) );
// same: real( conj( *A(i*incx,j) ) * *A(i*incx,j) );
}
x_i = sqrt( x_i );
error_inline = max( error_inline, fabs(x_m - x_i) / fabs(m*x_i) );
//printf( "dznrm2 xm %.8e, xc %.8e, xf %.8e, xi %.8e\n", x_m, x_c, x_f, x_i );
}
output( "dznrm2", m, n, k, incx, incy, error_cblas, error_fblas, error_inline );
}
}
}
printf( "\n" );
// ----- test ZDOTC
// dot columns, Aj^H Bj
for( int iincx = 0; iincx < ninc; ++iincx ) {
magma_int_t incx = inc[iincx];
for( int iincy = 0; iincy < ninc; ++iincy ) {
magma_int_t incy = inc[iincy];
error_cblas = 0;
error_fblas = 0;
error_inline = 0;
for( j=0; j < k; ++j ) {
// MAGMA implementation, not just wrapper
x2_m = magma_cblas_zdotc( m, A(0,j), incx, B(0,j), incy );
// crashes with MKL 11.1.2, ILP64
#if defined(HAVE_CBLAS) && ! defined(MAGMA_WITH_MKL)
#ifdef COMPLEX
cblas_zdotc_sub( m, A(0,j), incx, B(0,j), incy, &x2_c );
#else
x2_c = cblas_zdotc( m, A(0,j), incx, B(0,j), incy );
#endif
error_cblas = max( error_cblas, fabs(x2_m - x2_c) / fabs(m*x2_c) );
#else
x2_c = MAGMA_Z_ZERO;
error_cblas = SKIPPED_FLAG;
#endif
// crashes with MKL 11.2.3 and MacOS 10.9
#if (! defined(COMPLEX) || ! defined(MAGMA_WITH_MKL)) && ! defined(__APPLE__)
x2_f = blasf77_zdotc( &m, A(0,j), &incx, B(0,j), &incy );
error_fblas = max( error_fblas, fabs(x2_m - x2_f) / fabs(m*x2_f) );
#else
x2_f = MAGMA_Z_ZERO;
error_fblas = SKIPPED_FLAG;
#endif
// inline implementation
x2_i = MAGMA_Z_ZERO;
magma_int_t A_offset = (incx > 0 ? 0 : (-n + 1)*incx);
magma_int_t B_offset = (incy > 0 ? 0 : (-n + 1)*incy);
for( i=0; i < m; ++i ) {
x2_i += conj( *A(A_offset + i*incx,j) ) * *B(B_offset + i*incy,j);
}
error_inline = max( error_inline, fabs(x2_m - x2_i) / fabs(m*x2_i) );
//printf( "zdotc xm %.8e + %.8ei, xc %.8e + %.8ei, xf %.8e + %.8ei, xi %.8e + %.8ei\n",
// real(x2_m), imag(x2_m),
// real(x2_c), imag(x2_c),
// real(x2_f), imag(x2_f),
// real(x2_i), imag(x2_i) );
}
output( "zdotc", m, n, k, incx, incy, error_cblas, error_fblas, error_inline );
}
}
printf( "\n" );
// ----- test ZDOTU
// dot columns, Aj^T * Bj
for( int iincx = 0; iincx < ninc; ++iincx ) {
magma_int_t incx = inc[iincx];
for( int iincy = 0; iincy < ninc; ++iincy ) {
magma_int_t incy = inc[iincy];
error_cblas = 0;
error_fblas = 0;
error_inline = 0;
for( j=0; j < k; ++j ) {
// MAGMA implementation, not just wrapper
x2_m = magma_cblas_zdotu( m, A(0,j), incx, B(0,j), incy );
// crashes with MKL 11.1.2, ILP64
#if defined(HAVE_CBLAS) && ! defined(MAGMA_WITH_MKL)
#ifdef COMPLEX
cblas_zdotu_sub( m, A(0,j), incx, B(0,j), incy, &x2_c );
#else
x2_c = cblas_zdotu( m, A(0,j), incx, B(0,j), incy );
#endif
error_cblas = max( error_cblas, fabs(x2_m - x2_c) / fabs(m*x2_c) );
#else
x2_c = MAGMA_Z_ZERO;
error_cblas = SKIPPED_FLAG;
#endif
// crashes with MKL 11.2.3 and MacOS 10.9
#if (! defined(COMPLEX) || ! defined(MAGMA_WITH_MKL)) && ! defined(__APPLE__)
x2_f = blasf77_zdotu( &m, A(0,j), &incx, B(0,j), &incy );
error_fblas = max( error_fblas, fabs(x2_m - x2_f) / fabs(m*x2_f) );
#else
x2_f = MAGMA_Z_ZERO;
error_fblas = SKIPPED_FLAG;
#endif
// inline implementation
x2_i = MAGMA_Z_ZERO;
magma_int_t A_offset = (incx > 0 ? 0 : (-n + 1)*incx);
magma_int_t B_offset = (incy > 0 ? 0 : (-n + 1)*incy);
for( i=0; i < m; ++i ) {
x2_i += *A(A_offset + i*incx,j) * *B(B_offset + i*incy,j);
}
error_inline = max( error_inline, fabs(x2_m - x2_i) / fabs(m*x2_i) );
//printf( "zdotu xm %.8e + %.8ei, xc %.8e + %.8ei, xf %.8e + %.8ei, xi %.8e + %.8ei\n",
// real(x2_m), imag(x2_m),
// real(x2_c), imag(x2_c),
// real(x2_f), imag(x2_f),
// real(x2_i), imag(x2_i) );
}
output( "zdotu", m, n, k, incx, incy, error_cblas, error_fblas, error_inline );
}
}
// cleanup
TESTING_FREE_CPU( A );
TESTING_FREE_CPU( B );
fflush( stdout );
} // itest, incx, incy
opts.cleanup();
TESTING_FINALIZE();
return gStatus;
}
void
test_dot (void) {
const double flteps = 1e-4, dbleps = 1e-6;
{
int N = 1;
float alpha = 0.0f;
float X[] = { 0.733f };
float Y[] = { 0.825f };
int incX = 1;
int incY = -1;
float expected = 0.604725f;
float f;
f = cblas_sdsdot (N, alpha, X, incX, Y, incY);
gsl_test_rel(f, expected, flteps, "sdsdot(case 1)");
};
{
int N = 1;
float alpha = 0.1f;
float X[] = { 0.733f };
float Y[] = { 0.825f };
int incX = 1;
int incY = -1;
float expected = 0.704725f;
float f;
f = cblas_sdsdot (N, alpha, X, incX, Y, incY);
gsl_test_rel(f, expected, flteps, "sdsdot(case 2)");
};
{
int N = 1;
float alpha = 1.0f;
float X[] = { 0.733f };
float Y[] = { 0.825f };
int incX = 1;
int incY = -1;
float expected = 1.604725f;
float f;
f = cblas_sdsdot (N, alpha, X, incX, Y, incY);
gsl_test_rel(f, expected, flteps, "sdsdot(case 3)");
};
{
int N = 1;
float alpha = 0.0f;
float X[] = { -0.812f };
float Y[] = { -0.667f };
int incX = -1;
int incY = 1;
float expected = 0.541604f;
float f;
f = cblas_sdsdot (N, alpha, X, incX, Y, incY);
gsl_test_rel(f, expected, flteps, "sdsdot(case 4)");
};
{
int N = 1;
float alpha = 0.1f;
float X[] = { -0.812f };
float Y[] = { -0.667f };
int incX = -1;
int incY = 1;
float expected = 0.641604f;
float f;
f = cblas_sdsdot (N, alpha, X, incX, Y, incY);
gsl_test_rel(f, expected, flteps, "sdsdot(case 5)");
};
{
int N = 1;
float alpha = 1.0f;
float X[] = { -0.812f };
float Y[] = { -0.667f };
int incX = -1;
int incY = 1;
float expected = 1.541604f;
float f;
f = cblas_sdsdot (N, alpha, X, incX, Y, incY);
gsl_test_rel(f, expected, flteps, "sdsdot(case 6)");
};
{
int N = 1;
float alpha = 0.0f;
float X[] = { 0.481f };
float Y[] = { 0.523f };
int incX = -1;
int incY = -1;
float expected = 0.251563f;
float f;
f = cblas_sdsdot (N, alpha, X, incX, Y, incY);
gsl_test_rel(f, expected, flteps, "sdsdot(case 7)");
};
{
int N = 1;
float alpha = 0.1f;
float X[] = { 0.481f };
float Y[] = { 0.523f };
int incX = -1;
int incY = -1;
float expected = 0.351563f;
float f;
f = cblas_sdsdot (N, alpha, X, incX, Y, incY);
gsl_test_rel(f, expected, flteps, "sdsdot(case 8)");
};
{
int N = 1;
float alpha = 1.0f;
float X[] = { 0.481f };
float Y[] = { 0.523f };
int incX = -1;
int incY = -1;
float expected = 1.251563f;
float f;
f = cblas_sdsdot (N, alpha, X, incX, Y, incY);
gsl_test_rel(f, expected, flteps, "sdsdot(case 9)");
};
{
int N = 1;
float X[] = { 0.785f };
float Y[] = { -0.7f };
int incX = 1;
int incY = -1;
float expected = -0.5495f;
float f;
f = cblas_sdot(N, X, incX, Y, incY);
gsl_test_rel(f, expected, flteps, "sdot(case 10)");
};
{
int N = 1;
double X[] = { 0.79 };
double Y[] = { -0.679 };
int incX = 1;
int incY = -1;
double expected = -0.53641;
double f;
f = cblas_ddot(N, X, incX, Y, incY);
gsl_test_rel(f, expected, dbleps, "ddot(case 11)");
};
{
int N = 1;
float X[] = { 0.474f, -0.27f };
float Y[] = { -0.144f, -0.392f };
int incX = 1;
int incY = -1;
float expected[2] = {-0.174096f, -0.146928f};
float f[2];
cblas_cdotu_sub(N, X, incX, Y, incY, &f);
gsl_test_rel(f[0], expected[0], flteps, "cdotu(case 12) real");
gsl_test_rel(f[1], expected[1], flteps, "cdotu(case 12) imag");
};
{
int N = 1;
float X[] = { 0.474f, -0.27f };
float Y[] = { -0.144f, -0.392f };
int incX = 1;
int incY = -1;
float expected[2] = {0.037584f, -0.224688f};
float f[2];
cblas_cdotc_sub(N, X, incX, Y, incY, &f);
gsl_test_rel(f[0], expected[0], flteps, "cdotc(case 13) real");
gsl_test_rel(f[1], expected[1], flteps, "cdotc(case 13) imag");
};
{
int N = 1;
double X[] = { -0.87, -0.631 };
double Y[] = { -0.7, -0.224 };
int incX = 1;
int incY = -1;
double expected[2] = {0.467656, 0.63658};
double f[2];
cblas_zdotu_sub(N, X, incX, Y, incY, &f);
gsl_test_rel(f[0], expected[0], dbleps, "zdotu(case 14) real");
gsl_test_rel(f[1], expected[1], dbleps, "zdotu(case 14) imag");
};
{
int N = 1;
double X[] = { -0.87, -0.631 };
double Y[] = { -0.7, -0.224 };
int incX = 1;
int incY = -1;
double expected[2] = {0.750344, -0.24682};
double f[2];
cblas_zdotc_sub(N, X, incX, Y, incY, &f);
gsl_test_rel(f[0], expected[0], dbleps, "zdotc(case 15) real");
gsl_test_rel(f[1], expected[1], dbleps, "zdotc(case 15) imag");
};
{
int N = 1;
float X[] = { -0.457f };
float Y[] = { 0.839f };
int incX = -1;
int incY = 1;
float expected = -0.383423f;
float f;
f = cblas_sdot(N, X, incX, Y, incY);
gsl_test_rel(f, expected, flteps, "sdot(case 16)");
};
{
int N = 1;
double X[] = { 0.949 };
double Y[] = { -0.873 };
int incX = -1;
int incY = 1;
double expected = -0.828477;
double f;
f = cblas_ddot(N, X, incX, Y, incY);
gsl_test_rel(f, expected, dbleps, "ddot(case 17)");
};
{
int N = 1;
float X[] = { 0.852f, -0.045f };
float Y[] = { 0.626f, -0.164f };
int incX = -1;
int incY = 1;
float expected[2] = {0.525972f, -0.167898f};
float f[2];
cblas_cdotu_sub(N, X, incX, Y, incY, &f);
gsl_test_rel(f[0], expected[0], flteps, "cdotu(case 18) real");
gsl_test_rel(f[1], expected[1], flteps, "cdotu(case 18) imag");
};
{
int N = 1;
float X[] = { 0.852f, -0.045f };
float Y[] = { 0.626f, -0.164f };
int incX = -1;
int incY = 1;
float expected[2] = {0.540732f, -0.111558f};
float f[2];
cblas_cdotc_sub(N, X, incX, Y, incY, &f);
gsl_test_rel(f[0], expected[0], flteps, "cdotc(case 19) real");
gsl_test_rel(f[1], expected[1], flteps, "cdotc(case 19) imag");
};
{
int N = 1;
double X[] = { -0.786, -0.341 };
double Y[] = { -0.271, -0.896 };
int incX = -1;
int incY = 1;
double expected[2] = {-0.09253, 0.796667};
double f[2];
cblas_zdotu_sub(N, X, incX, Y, incY, &f);
gsl_test_rel(f[0], expected[0], dbleps, "zdotu(case 20) real");
gsl_test_rel(f[1], expected[1], dbleps, "zdotu(case 20) imag");
};
{
int N = 1;
double X[] = { -0.786, -0.341 };
double Y[] = { -0.271, -0.896 };
int incX = -1;
int incY = 1;
double expected[2] = {0.518542, 0.611845};
double f[2];
cblas_zdotc_sub(N, X, incX, Y, incY, &f);
gsl_test_rel(f[0], expected[0], dbleps, "zdotc(case 21) real");
gsl_test_rel(f[1], expected[1], dbleps, "zdotc(case 21) imag");
};
{
int N = 1;
float X[] = { -0.088f };
float Y[] = { -0.165f };
int incX = -1;
int incY = -1;
float expected = 0.01452f;
float f;
f = cblas_sdot(N, X, incX, Y, incY);
gsl_test_rel(f, expected, flteps, "sdot(case 22)");
};
{
int N = 1;
double X[] = { -0.434 };
double Y[] = { -0.402 };
int incX = -1;
int incY = -1;
double expected = 0.174468;
double f;
f = cblas_ddot(N, X, incX, Y, incY);
gsl_test_rel(f, expected, dbleps, "ddot(case 23)");
};
{
int N = 1;
float X[] = { -0.347f, 0.899f };
float Y[] = { -0.113f, -0.858f };
int incX = -1;
int incY = -1;
float expected[2] = {0.810553f, 0.196139f};
float f[2];
cblas_cdotu_sub(N, X, incX, Y, incY, &f);
gsl_test_rel(f[0], expected[0], flteps, "cdotu(case 24) real");
gsl_test_rel(f[1], expected[1], flteps, "cdotu(case 24) imag");
};
{
int N = 1;
float X[] = { -0.347f, 0.899f };
float Y[] = { -0.113f, -0.858f };
int incX = -1;
int incY = -1;
float expected[2] = {-0.732131f, 0.399313f};
float f[2];
cblas_cdotc_sub(N, X, incX, Y, incY, &f);
gsl_test_rel(f[0], expected[0], flteps, "cdotc(case 25) real");
gsl_test_rel(f[1], expected[1], flteps, "cdotc(case 25) imag");
};
{
int N = 1;
double X[] = { -0.897, -0.204 };
double Y[] = { -0.759, 0.557 };
int incX = -1;
int incY = -1;
double expected[2] = {0.794451, -0.344793};
double f[2];
cblas_zdotu_sub(N, X, incX, Y, incY, &f);
gsl_test_rel(f[0], expected[0], dbleps, "zdotu(case 26) real");
gsl_test_rel(f[1], expected[1], dbleps, "zdotu(case 26) imag");
};
{
int N = 1;
double X[] = { -0.897, -0.204 };
double Y[] = { -0.759, 0.557 };
int incX = -1;
int incY = -1;
double expected[2] = {0.567195, -0.654465};
double f[2];
cblas_zdotc_sub(N, X, incX, Y, incY, &f);
gsl_test_rel(f[0], expected[0], dbleps, "zdotc(case 27) real");
gsl_test_rel(f[1], expected[1], dbleps, "zdotc(case 27) imag");
};
}
void wrapper_cblas_zdotu_sub(const int N, const void *X, const int incX, const void *Y, const int incY, void *dotu)
{
cblas_zdotu_sub(N, X, incX, Y, incY, dotu);
}
int main()
{
double time;
int size = N;
int one = 1;
//
double* da = (double*) malloc(2*N*sizeof(double));
double* db = (double*) malloc(2*N*sizeof(double));
double* cc = (double*) malloc(2* sizeof(double));
//
for (int ii = 0; ii < 2*N; ii = ii + 2)
{
da[ii + 0] = 0; //rand()%10; // real
da[ii + 1] = 1; //rand()%10; // imaginary
//
db[ii + 0] = 2; //rand()%10; // real
db[ii + 1] = 3; //rand()%10; // imaginary
}
//
cc[0] = 0.0; // real
cc[1] = 0.0; // imaginary
//
time = -myseconds();
asm volatile("# mkl");
zdotu_aos(N, da, one, db, one, cc);
time += myseconds();
//
assert(cc[1] != (double) N);
printf( "AOS: The complex dot product is: (%6.2f, %6.2f), bandwidth = %f GB/s, perf = %f Gflops/s (%f s.)\n", cc[0], cc[1], 2*N*8/1024./1024/1024/time, 2.*N/1e9/time, time );
//
time = -myseconds();
cblas_zdotu_sub(N, da, one, db, one, cc);
time += myseconds();
printf( "MKL: The complex dot product is: (%6.2f, %6.2f), bandwidth = %f GB/s, perf = %f Gflops/s (%f s.)\n", cc[0], cc[1], 2*N*8/1024./1024/1024/time, 2.*N/1e9/time, time );
free(da);
free(db);
//
da = (double*) malloc(N*sizeof(double));
db = (double*) malloc(N*sizeof(double));
double* dc = (double*) malloc(N*sizeof(double));
double* dd = (double*) malloc(N*sizeof(double));
//
for (int ii = 0; ii < N; ++ii)
{
da[ii] = 0; //rand()%10; // real
db[ii] = 1; //rand()%10; // imaginary
//
dc[ii] = 2; //rand()%10; // real
dd[ii] = 3; //rand()%10; // imaginary
}
//
time = -myseconds();
asm volatile("# mkl");
zdotu_soa(N, da, db, one, dc, dd, one, cc);
time += myseconds();
//
assert(cc[1] != (double) N);
printf( "SOA: The complex dot product is: (%6.2f, %6.2f), bandwidth = %f GB/s, perf = %f Gflops/s (%f s.)\n", cc[0], cc[1], 2*N*8/1024./1024/1024/time, 2.*N/1e9/time, time );
//
free(da);
free(db);
free(dc);
free(dd);
//
free(cc);
}
void WRAP_F77(acc_zdotu_sub)(const int *N, const void *X, const int *incX,
const void *Y, const int *incY, void *dotu)
{
cblas_zdotu_sub(*N, X, *incX, Y, *incY, dotu);
}
complex double zdotu_(int *N, void *CX, int *INCX, void *CY, int *INCY) {
complex double dotu;
cblas_zdotu_sub(*N, CX, *INCX, CY, *INCY, &dotu);
return dotu;
}
void F77_zdotu(const int *N, void *X, const int *incX,
void *Y, const int *incY,void *dotu)
{
cblas_zdotu_sub(*N, X, *incX, Y, *incY, dotu);
return;
}
//
// Overloaded function for dispatching to
// * CBLAS backend, and
// * complex<double> value-type.
//
inline std::complex<double> dot( const int n, const std::complex<double>* x,
const int incx, const std::complex<double>* y, const int incy ) {
std::complex<double> result;
cblas_zdotu_sub( n, x, incx, y, incy, &result );
return result;
}
DLLEXPORT openblas_complex_double z_dot_product(const blasint n, const openblas_complex_double x[], const openblas_complex_double y[]) {
openblas_complex_double ret;
cblas_zdotu_sub(n, (double*)x, 1, (double*)y, 1, &ret);
return ret;
}
void WRAP_F77(veclib_zdotu)(const int *N, const double complex *X, const int
*incX, const double complex *Y, const int *incY, double complex *dotu)
{
cblas_zdotu_sub(*N, X, *incX, Y, *incY, dotu);
}
