complex case handled by cannon algorithm as well

src/elpa_generalized/cannon_forw.c

@@ -2,6 +2,7 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <math.h>
+#include <complex.h>
 
 // most of the file is not compiled if not using MPI
 #ifdef WITH_MPI
@@ -58,6 +59,54 @@ void cannons_reduction_c_d(double* A, double* U, int local_rows, int local_cols,
 void cannons_reduction_c_f(float* A, float* U, int local_rows, int local_cols, int* a_desc,
                          float *Res, int ToStore, int row_comm, int col_comm);
 
+#define COMPLEXCASE 1
+#define DOUBLE_PRECISION 1
+#include "../general/precision_macros.h"
+#include "cannon_forw_template.c"
+#undef DOUBLE_PRECISION
+#undef COMPLEXCASE
+
+//***********************************************************************************************************
+/*
+!f> interface
+!f>   subroutine cannons_reduction_dc(A, U, local_rows, local_cols, a_desc, Res, toStore, row_comm, col_comm) &
+!f>                             bind(C, name="cannons_reduction_c_dc")
+!f>     use, intrinsic :: iso_c_binding
+!f>     complex(c_double)                     :: A(local_rows, local_cols), U(local_rows, local_cols), Res(local_rows, local_cols)
+!f>     !type(c_ptr), value                   :: A, U, Res
+!f>     integer(kind=c_int)                   :: a_desc(9)
+!f>     integer(kind=c_int),value             :: local_rows, local_cols
+!f>     integer(kind=c_int),value     ::  row_comm, col_comm, ToStore
+!f>   end subroutine
+!f> end interface
+*/
+void cannons_reduction_c_dc(double complex* A, double complex* U, int local_rows, int local_cols, int* a_desc,
+                         double complex *Res, int ToStore, int row_comm, int col_comm);
+
+#define COMPLEXCASE 1
+#define SINGLE_PRECISION 1
+#include "../general/precision_macros.h"
+#include "cannon_forw_template.c"
+#undef SINGLE_PRECISION
+#undef COMPLEXCASE
+
+//***********************************************************************************************************
+/*
+!f> interface
+!f>   subroutine cannons_reduction_fc(A, U, local_rows, local_cols, a_desc, Res, toStore, row_comm, col_comm) &
+!f>                             bind(C, name="cannons_reduction_c_fc")
+!f>     use, intrinsic :: iso_c_binding
+!f>     complex(c_float)                      :: A(local_rows, local_cols), U(local_rows, local_cols), Res(local_rows, local_cols)
+!f>     !type(c_ptr), value                   :: A, U, Res
+!f>     integer(kind=c_int)                   :: a_desc(9)
+!f>     integer(kind=c_int),value             :: local_rows, local_cols
+!f>     integer(kind=c_int),value     ::  row_comm, col_comm, ToStore
+!f>   end subroutine
+!f> end interface
+*/
+void cannons_reduction_c_fc(float complex* A, float complex* U, int local_rows, int local_cols, int* a_desc,
+                         float complex *Res, int ToStore, int row_comm, int col_comm);
+
 #else
 // Just because of the Intel preprocessor
 // TODO do something with it
@@ -70,4 +119,14 @@ void cannons_reduction_c_f(float* A, float* U, int local_rows, int local_cols, i
                          float *Res, int ToStore, int row_comm, int col_comm)
 {
 }
+
+void cannons_reduction_c_dc(double complex* A, double complex* U, int local_rows, int local_cols, int* a_desc,
+                         double complex *Res, int ToStore, int row_comm, int col_comm)
+{
+}
+
+void cannons_reduction_c_fc(float complex* A, float complex* U, int local_rows, int local_cols, int* a_desc,
+                         float complex *Res, int ToStore, int row_comm, int col_comm)
+{
+}
 #endif

src/elpa_impl_generalized_transform_template.F90

@@ -34,13 +34,6 @@
 
      call self%timer_start("transform_generalized()")
      call self%get("cannon_for_generalized",use_cannon,error)
-#if !defined(REALCASE)
-     if(my_p == 0) then
-       write(*,*) "Cannons algorithm can be used only for real at the moment"
-       write(*,*) "Switching to elpa Hermitian and scalapack"
-     end if
-     use_cannon = 0
-#endif
 
 #if !defined(WITH_MPI)
      if(my_p == 0) then
@@ -81,14 +74,12 @@
 
 
        call self%timer_start("cannons_reduction")
-#if defined(REALCASE)
        ! BEWARE! even though tmp is output from the routine, it has to be zero on input!
        tmp = 0.0_rck
        call cannons_reduction_&
          &ELPA_IMPL_SUFFIX&
          &(a, b, self%local_nrows, self%local_ncols, &
                               sc_desc, tmp, BuffLevelInt, mpi_comm_rows, mpi_comm_cols)
-#endif
        call self%timer_stop("cannons_reduction")
 
        a(1:self%local_nrows, 1:self%local_ncols) = tmp(1:self%local_nrows, 1:self%local_ncols)

src/general/precision_macros.h

@@ -271,7 +271,7 @@
 #define C_GEMM zgemm_
 #define C_LACPY zlacpy_
 #define C_PLACPY pzlacpy_
-#define C_PTRAN pztran_
+#define C_PTRAN pztranc_
 
 #define  PRECISION_TRTRI ZTRTRI
 #define  PRECISION_POTRF ZPOTRF
@@ -335,7 +335,7 @@
 #define C_GEMM cgemm_
 #define C_LACPY clacpy_
 #define C_PLACPY pclacpy_
-#define C_PTRAN pctran_
+#define C_PTRAN pctranc_
 
 #define  PRECISION_TRTRI CTRTRI
 #define  PRECISION_POTRF CPOTRF

src/general/precision_typedefs.h

 #ifdef REALCASE
 #ifdef DOUBLE_PRECISION
 //typedef double math_type;
+#undef math_type
 #define math_type double
 #endif
 #ifdef SINGLE_PRECISION
 //typedef float math_type;
+#undef math_type
 #define math_type float
 #endif
 #endif
 
 #ifdef COMPLEXCASE
 #ifdef DOUBLE_PRECISION
+#undef math_type
+#define math_type double complex
 #endif
 #ifdef SINGLE_PRECISION
+#undef math_type
+#define math_type float complex
 #endif
 #endif