.RI "int \fBTHIS_ELPA_COMPLEX_KERNEL\fp: choose the compute kernel for 2-stage solver"
.br
.RI "char *\fBmethod\fP: use 1stage solver if "1stage", use 2stage solver if "2stage", (at the moment) use 2stage solver if "auto" "
.RI "int \fBsuccess\fP: return value indicating success (1) or failure (0)
.RI "int \fBuseGPU\fP: decide whether GPUs should be used or not"
.RI "int \fBsuccess\fP: return value indicating success (1) or failure (0)
.SH DESCRIPTION
Solve the complex eigenvalue problem with the 2-stage solver. The ELPA communicators \fBmpi_comm_rows\fP and \fBmpi_comm_cols\fP are obtained with the \fBelpa_get_communicators\fP(3) function. The distributed quadratic marix \fBa\fP has global dimensions \fBna\fP x \fBna\fP, and a local size \fBlda\fP x \fBmatrixCols\fP. The solver will compute the first \fBnev\fP eigenvalues, which will be stored on exit in \fBev\fP. The eigenvectors corresponding to the eigenvalues will be stored in \fBq\fP. All memory of the arguments must be allocated outside the call to the solver.
...
...
@@ -96,4 +97,4 @@ Solve the complex eigenvalue problem with the 2-stage solver. The ELPA communica
The interface \fBelpa_solve_evp_complex\fP(3) is a more flexible alternative.
.RI "With the definintions of the input and output variables:"
.br
.RI "int \fBna\fP: global dimension of quadratic matrix \fBa\fP to solve"
.br
.RI "int \fBnev\fP: number of eigenvalues to be computed; the first \fBnev\fP eigenvalules are calculated"
.br
.RI "complex *\fBa\fP: pointer to locally distributed part of the matrix \fBa\fP. The local dimensions are \fBlda\fP x \fBmatrixCols\fP"
.br
.RI "int \fBlda\fP: leading dimension of locally distributed matrix \fBa\fP"
.br
.RI "float *\fBev\fP: pointer to memory containing on output the first \fBnev\fP computed eigenvalues"
.br
.RI "complex *\fBq\fP: pointer to memory containing on output the first \fBnev\fP computed eigenvectors"
.br
.RI "int \fBldq\fP: leading dimension of matrix \fBq\fP which stores the eigenvectors"
.br
.RI "int \fBnblk\fP: blocksize of block cyclic distributin, must be the same in both directions"
.br
.RI "int \fBmatrixCols\fP: number of columns of locally distributed matrices \fBa\fP and \fBq\fP"
.br
.RI "int \fBmpi_comm_rows\fP: communicator for communication in rows. Constructed with \fBelpa_get_communicators\fP(3)"
.br
.RI "int \fBmpi_comm_cols\fP: communicator for communication in colums. Constructed with \fBelpa_get_communicators\fP(3)"
.br
.RI "int \fBmpi_comm_all\fP: communicator for all processes in the processor set involved in ELPA"
.br
.RI "int \fBTHIS_ELPA_COMPLEX_KERNEL\fp: choose the compute kernel for 2-stage solver"
.br
.RI "int \fBuseGPU\fP: decide whether GPUs should be used or not"
.RI "int \fBsuccess\fP: return value indicating success (1) or failure (0)
.SH DESCRIPTION
Solve the complex eigenvalue problem with the 2-stage solver. The ELPA communicators \fBmpi_comm_rows\fP and \fBmpi_comm_cols\fP are obtained with the \fBelpa_get_communicators\fP(3) function. The distributed quadratic marix \fBa\fP has global dimensions \fBna\fP x \fBna\fP, and a local size \fBlda\fP x \fBmatrixCols\fP. The solver will compute the first \fBnev\fP eigenvalues, which will be stored on exit in \fBev\fP. The eigenvectors corresponding to the eigenvalues will be stored in \fBq\fP. All memory of the arguments must be allocated outside the call to the solver.
.br
The interface \fBelpa_solve_evp_complex\fP(3) is a more flexible alternative.
.RI "With the definintions of the input and output variables:"
.br
.RI "int \fBna\fP: global dimension of quadratic matrix \fBa\fP to solve"
.br
.RI "int \fBnev\fP: number of eigenvalues to be computed; the first \fBnev\fP eigenvalules are calculated"
.br
.RI "float *\fBa\fP: pointer to locally distributed part of the matrix \fBa\fP. The local dimensions are \fBlda\fP x \fBmatrixCols\fP"
.br
.RI "int \fBlda\fP: leading dimension of locally distributed matrix \fBa\fP"
.br
.RI "float *\fBev\fP: pointer to memory containing on output the first \fBnev\fP computed eigenvalues"
.br
.RI "float *\fBq\fP: pointer to memory containing on output the first \fBnev\fP computed eigenvectors"
.br
.RI "int \fBldq\fP: leading dimension of matrix \fBq\fP which stores the eigenvectors"
.br
.RI "int \fBnblk\fP: blocksize of block cyclic distributin, must be the same in both directions"
.br
.RI "int \fBmatrixCols\fP: number of columns of locally distributed matrices \fBa\fP and \fBq\fP"
.br
.RI "int \fBmpi_comm_rows\fP: communicator for communication in rows. Constructed with \fBelpa_get_communicators\fP(3)"
.br
.RI "int \fBmpi_comm_cols\fP: communicator for communication in colums. Constructed with \fBelpa_get_communicators\fP(3)"
.br
.RI "int \fBmpi_comm_all\fP: communicator for all processes in the processor set involved in ELPA"
.br
.RI "int \fBTHIS_ELPA_REAL_KERNEL\fp: choose the compute kernel for 2-stage solver"
.br
.RI "int \fBuseQR\fP: if set to 1 switch to QR-decomposition"
.br
.RI "int \fBuseGPU\fP: decide whether GPUs should be used or not"
.br
.RI "int \fBsuccess\fP: return value indicating success (1) or failure (0)
.SH DESCRIPTION
Solve the real eigenvalue problem with the 2-stage solver. The ELPA communicators \fBmpi_comm_rows\fP and \fBmpi_comm_cols\fP are obtained with the \fBelpa_get_communicators\fP(3) function. The distributed quadratic marix \fBa\fP has global dimensions \fBna\fP x \fBna\fP, and a local size \fBlda\fP x \fBmatrixCols\fP. The solver will compute the first \fBnev\fP eigenvalues, which will be stored on exit in \fBev\fP. The eigenvectors corresponding to the eigenvalues will be stored in \fBq\fP. All memory of the arguments must be allocated outside the call to the solver.
.br
The interface \fBelpa_solve_evp_real\fP(3) is a more flexible alternative.