VALUE rb_blas_ixamax(int argc, VALUE *argv, VALUE self)
{
Matrix *dx;
int incx;
int incy;
int n;
//char error_msg[64];
VALUE n_value, incx_value;
rb_scan_args(argc, argv, "02", &incx_value, &n_value);
Data_Get_Struct(self, Matrix, dx);
if(incx_value == Qnil)
incx = 1;
else
incx = NUM2INT(incx_value);
if(n_value == Qnil)
n = dx->nrows;
else
n = NUM2INT(n_value);
if(dx == NULL || dx->ncols != 1)
{ //sprintf(error_msg, );
rb_raise(rb_eRuntimeError, "Self is not a Vector");
}
switch(dx->data_type)
{
case Single_t: //s
return INT2FIX(cblas_isamax(n , (float *)dx->data, incx));
case Double_t: //d
return INT2FIX(cblas_idamax(n , (double *)dx->data, incx));
case Complex_t: //c
return INT2FIX(cblas_icamax(n , dx->data, incx));
case Double_Complex_t: //z
return INT2FIX(cblas_izamax(n , dx->data, incx));
default:
//sprintf(error_msg, "Invalid data_type (%d) in Matrix", dx->data_type);
rb_raise(rb_eRuntimeError, "Invalid data_type (%d) in Matrix", dx->data_type);
return Qnil; //Never reaches here.
}
}
static void
CORE_cgetrf_reclap_rec(const int M, const int N,
PLASMA_Complex32_t *A, const int LDA,
int *IPIV, int *info,
const int thidx, const int thcnt, const int column)
{
int jp, n1, n2, lm, loff;
PLASMA_Complex32_t tmp1, tmp2, tmp3;
PLASMA_Complex32_t *Atop = A + column*LDA;
/* Assumption: N = min( M, N ); */
if (N > 1) {
int coff, ccnt;
n1 = N / 2;
n2 = N - n1;
CORE_cgetrf_reclap_rec( M, n1, A, LDA, IPIV, info,
thidx, thcnt, column );
if ( *info != 0 )
return;
CORE_cgetrf_reclap_update(M, column, n1, n2,
A, LDA, IPIV,
thidx, thcnt);
CORE_cgetrf_reclap_rec( M, n2, A, LDA, IPIV, info,
thidx, thcnt, column + n1 );
if ( *info != 0 )
return;
psplit( n1, thidx, thcnt, &coff, &ccnt );
if (ccnt > 0) {
CORE_claswap1( ccnt, Atop+coff*LDA, LDA, n1 + column, N + column, IPIV ); /* swap to the left */
}
} else {
int thrd;
CORE_cbarrier_thread( thidx, thcnt );
psplit( M, thidx, thcnt, &loff, &lm );
if (thidx == 0) {
loff = column;
lm -= column;
}
tmp2 = Atop[column]; /* all threads read the pivot element in case they need it */
jp = cblas_icamax( lm, Atop + loff, 1 );
tmp1 = Atop[loff + jp];
CORE_camax1_thread( tmp1, thidx, thcnt, &thrd,
&tmp3, loff + jp + 1, IPIV + column );
Atop[column] = tmp3; /* all threads set the pivot element: no need for synchronization */
if ( tmp3 != 0.0 ) {
if ( cabsf(tmp3) >= sfmin ) {
PLASMA_Complex32_t tmp = (PLASMA_Complex32_t)1.0 / tmp3;
n1 = (thidx == 0) ? 1 : 0;
cblas_cscal( lm - n1, CBLAS_SADDR(tmp), Atop + loff + n1, 1 );
} else {
int i;
PLASMA_Complex32_t *Atop2;
n1 = (thidx == 0) ? 1 : 0;
Atop2 = Atop + loff + n1;
for( i=0; i < lm-n1; i++, Atop2++)
*Atop2 = *Atop2 / tmp3;
}
if (thrd == thidx) { /* the thread that owns the best pivot */
if (loff + jp != column) /* if there is a need to exchange the pivot */
Atop[loff + jp] = tmp2 / tmp3;
}
} else {
*info = column + 1;
return;
}
CORE_cbarrier_thread( thidx, thcnt );
}
}
void
test_amax (void) {
{
int N = 1;
float X[] = { -0.388f };
int incX = -1;
int expected = 0;
int k;
k = cblas_isamax(N, X, incX);
gsl_test_int(k, expected, "samax(case 52)");
};
{
int N = 1;
double X[] = { 0.247 };
int incX = -1;
int expected = 0;
int k;
k = cblas_idamax(N, X, incX);
gsl_test_int(k, expected, "damax(case 53)");
};
{
int N = 1;
float X[] = { 0.704f, 0.665f };
int incX = -1;
int expected = 0;
int k;
k = cblas_icamax(N, X, incX);
gsl_test_int(k, expected, "camax(case 54)");
};
{
int N = 1;
double X[] = { -0.599, -0.758 };
int incX = -1;
int expected = 0;
int k;
k = cblas_izamax(N, X, incX);
gsl_test_int(k, expected, "zamax(case 55)");
};
{
int N = 2;
float X[] = { 0.909f, 0.037f };
int incX = 1;
int expected = 0;
int k;
k = cblas_isamax(N, X, incX);
gsl_test_int(k, expected, "samax(case 56)");
};
{
int N = 2;
double X[] = { 0.271, -0.426 };
int incX = 1;
int expected = 1;
int k;
k = cblas_idamax(N, X, incX);
gsl_test_int(k, expected, "damax(case 57)");
};
{
int N = 2;
float X[] = { -0.648f, 0.317f, 0.62f, 0.392f };
int incX = 1;
int expected = 1;
int k;
k = cblas_icamax(N, X, incX);
gsl_test_int(k, expected, "camax(case 58)");
};
{
int N = 2;
double X[] = { -0.789, 0.352, 0.562, 0.697 };
int incX = 1;
int expected = 1;
int k;
k = cblas_izamax(N, X, incX);
gsl_test_int(k, expected, "zamax(case 59)");
};
{
int N = 2;
float X[] = { 0.487f, 0.918f };
int incX = -1;
int expected = 0;
int k;
k = cblas_isamax(N, X, incX);
gsl_test_int(k, expected, "samax(case 60)");
};
{
int N = 2;
double X[] = { 0.537, 0.826 };
int incX = -1;
int expected = 0;
int k;
k = cblas_idamax(N, X, incX);
gsl_test_int(k, expected, "damax(case 61)");
};
{
int N = 2;
float X[] = { 0.993f, 0.172f, -0.825f, 0.873f };
int incX = -1;
int expected = 0;
int k;
k = cblas_icamax(N, X, incX);
gsl_test_int(k, expected, "camax(case 62)");
};
{
int N = 2;
double X[] = { 0.913, -0.436, -0.134, 0.129 };
int incX = -1;
int expected = 0;
int k;
k = cblas_izamax(N, X, incX);
gsl_test_int(k, expected, "zamax(case 63)");
};
}
int F77_icamax(const int *N, const void *X, const int *incX)
{
if (*N < 1 || *incX < 1) return(0);
return (cblas_icamax(*N, X, *incX)+1);
}
//
// Overloaded function for dispatching to
// * CBLAS backend, and
// * complex<float> value-type.
//
inline std::ptrdiff_t iamax( const int n, const std::complex<float>* x,
const int incx ) {
return cblas_icamax( n, x, incx );
}
