using cannons algorithm for real singe as well

src/elpa_generalized/cannon_forw.c

@@ -38,7 +38,7 @@ int numroc_(int*, int*, int*, int*, int*);
 //***********************************************************************************************************
 /*
 !f> interface
-!f>   subroutine cannons_reduction(A, U, local_rows, local_cols, a_desc, Res, toStore, row_comm, col_comm) &
+!f>   subroutine cannons_reduction_d(A, U, local_rows, local_cols, a_desc, Res, toStore, row_comm, col_comm) &
 !f>                             bind(C, name="cannons_reduction_c_d")
 !f>     use, intrinsic :: iso_c_binding
 !f>     real(c_double)                        :: A(local_rows, local_cols), U(local_rows, local_cols), Res(local_rows, local_cols)
@@ -52,10 +52,40 @@ int numroc_(int*, int*, int*, int*, int*);
 void cannons_reduction_c_d(double* A, double* U, int local_rows, int local_cols, int* a_desc,
                          double *Res, int ToStore, int row_comm, int col_comm);
 
+#define REALCASE 1
+#define SINGLE_PRECISION 1
+#include "../general/precision_macros.h"
+#include "cannon_forw_template.c"
+#undef SINGLE_PRECISION
+#undef REALCASE
+
+//***********************************************************************************************************
+/*
+!f> interface
+!f>   subroutine cannons_reduction_f(A, U, local_rows, local_cols, a_desc, Res, toStore, row_comm, col_comm) &
+!f>                             bind(C, name="cannons_reduction_c_f")
+!f>     use, intrinsic :: iso_c_binding
+!f>     real(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_f(float* A, float* U, int local_rows, int local_cols, int* a_desc,
+                         float *Res, int ToStore, int row_comm, int col_comm);
+
 #else
+// Just because of the Intel preprocessor
+// TODO do something with it
 void cannons_reduction_c_d(double* A, double* U, int local_rows, int local_cols, int* a_desc,
                          double *Res, int ToStore, int row_comm, int col_comm)
 {
 }
 
+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)
+{
+}
 #endif

src/elpa_impl_generalized_transform_template.F90

@@ -34,9 +34,9 @@
 
      call self%timer_start("transform_generalized()")
      call self%get("cannon_for_generalized",use_cannon,error)
-#if !defined(REALCASE) || !defined(DOUBLE_PRECISION)
+#if !defined(REALCASE)
      if(my_p == 0) then
-       write(*,*) "Cannons algorithm can be used only for real double at the moment"
+       write(*,*) "Cannons algorithm can be used only for real at the moment"
        write(*,*) "Switching to elpa Hermitian and scalapack"
      end if
      use_cannon = 0
@@ -81,10 +81,12 @@
 
 
        call self%timer_start("cannons_reduction")
-#if defined(REALCASE) && defined(DOUBLE_PRECISION)
+#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(a, b, self%local_nrows, self%local_ncols, &
+       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")

src/general/precision_macros.h

@@ -11,6 +11,11 @@
 #undef  REAL_DATATYPE
 #undef  C_REAL_DATATYPE
 
+#undef C_GEMM
+#undef C_LACPY
+#undef C_PLACPY
+#undef C_PTRAN
+
 #undef  PRECISION_TRTRI
 #undef  PRECISION_POTRF
 #undef  PRECISION_TRSM
@@ -48,6 +53,7 @@
 
 #undef  MPI_REAL_PRECISION
 #undef  MPI_MATH_DATATYPE_PRECISION
+#undef  MPI_MATH_DATATYPE_PRECISION_C
 #undef  MPI_MATH_DATATYPE_PRECISION_EXPL
 #undef  C_DATATYPE_KIND
 #undef THRESHOLD
@@ -103,9 +109,15 @@
 #define  scal_PRECISION_LASET PDLASET
 #define  MPI_REAL_PRECISION MPI_REAL8
 #define  MPI_MATH_DATATYPE_PRECISION MPI_REAL8
+#define  MPI_MATH_DATATYPE_PRECISION_C MPI_DOUBLE
 #define  MPI_MATH_DATATYPE_PRECISION_EXPL MPI_REAL8
 #define  C_DATATYPE_KIND c_double
 
+#define C_GEMM dgemm_
+#define C_LACPY dlacpy_
+#define C_PLACPY pdlacpy_
+#define C_PTRAN pdtran_
+
 #define THRESHOLD 1e-11_rk8
 #endif /* DOUBLE_PRECISION */
 
@@ -157,9 +169,15 @@
 #define  scal_PRECISION_LASET PSLASET
 #define  MPI_REAL_PRECISION MPI_REAL4
 #define  MPI_MATH_DATATYPE_PRECISION MPI_REAL4
+#define  MPI_MATH_DATATYPE_PRECISION_C MPI_FLOAT
 #define  MPI_MATH_DATATYPE_PRECISION_EXPL MPI_REAL4
 #define  C_DATATYPE_KIND c_float
 
+#define C_GEMM sgemm_
+#define C_LACPY slacpy_
+#define C_PLACPY pslacpy_
+#define C_PTRAN pstran_
+
 #define THRESHOLD 1e-4_rk4
 #endif /* SINGLE_PRECISION */
 
@@ -179,6 +197,11 @@
 #undef REAL_DATATYPE
 #undef C_REAL_DATATYPE
 
+#undef C_GEMM
+#undef C_LACPY
+#undef C_PLACPY
+#undef C_PTRAN
+
 #undef  PRECISION_TRTRI
 #undef  PRECISION_POTRF
 #undef  PRECISION_TRSM
@@ -216,6 +239,7 @@
 #undef  ELPA_IMPL_SUFFIX
 #undef  MPI_COMPLEX_PRECISION
 #undef  MPI_MATH_DATATYPE_PRECISION
+#undef  MPI_MATH_DATATYPE_PRECISION_C
 #undef  MPI_MATH_DATATYPE_PRECISION_EXPL
 #undef  MPI_COMPLEX_EXPLICIT_PRECISION
 #undef  MPI_REAL_PRECISION
@@ -244,6 +268,11 @@
 #define REAL_DATATYPE RK8
 #define C_REAL_DATATYPE c_double
 
+#define C_GEMM zgemm_
+#define C_LACPY zlacpy_
+#define C_PLACPY pzlacpy_
+#define C_PTRAN pztran_
+
 #define  PRECISION_TRTRI ZTRTRI
 #define  PRECISION_POTRF ZPOTRF
 #define  PRECISION_TRSM ZTRSM
@@ -278,6 +307,7 @@
 #define  scal_PRECISION_LASET PZLASET
 #define  MPI_COMPLEX_PRECISION MPI_DOUBLE_COMPLEX
 #define  MPI_MATH_DATATYPE_PRECISION MPI_DOUBLE_COMPLEX
+#define  MPI_MATH_DATATYPE_PRECISION_C MPI_DOUBLE_COMPLEX
 #define  MPI_MATH_DATATYPE_PRECISION_EXPL MPI_COMPLEX16
 #define  MPI_COMPLEX_EXPLICIT_PRECISION MPI_COMPLEX16
 #define  MPI_REAL_PRECISION MPI_REAL8
@@ -302,6 +332,11 @@
 #define REAL_DATATYPE RK4
 #define C_REAL_DATATYPE c_float
 
+#define C_GEMM cgemm_
+#define C_LACPY clacpy_
+#define C_PLACPY pclacpy_
+#define C_PTRAN pctran_
+
 #define  PRECISION_TRTRI CTRTRI
 #define  PRECISION_POTRF CPOTRF
 #define  PRECISION_TRSM CTRSM
@@ -336,6 +371,7 @@
 #define  scal_PRECISION_LASET PCLASET
 #define  MPI_COMPLEX_PRECISION MPI_COMPLEX
 #define  MPI_MATH_DATATYPE_PRECISION MPI_COMPLEX
+#define  MPI_MATH_DATATYPE_PRECISION_C MPI_COMPLEX
 #define  MPI_MATH_DATATYPE_PRECISION_EXPL MPI_COMPLEX8
 #define  MPI_COMPLEX_EXPLICIT_PRECISION MPI_COMPLEX8
 #define  MPI_REAL_PRECISION MPI_REAL4

src/general/precision_typedefs.h

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