--- a/LibDBlasLxF77.def.m2pp
+++ b/LibDBlasLxF77.def.m2pp
@@ -22,6 +22,9 @@
(* MOCKA : Definitionsmodul im Stil des Mocka Compilers *)
(* *)
(* ansonsten gibt es keine Aenderungen am Quellcode *)
+ (* *)
+ (* There are three version contained, one for the GNU, XDS and Mocka *)
+ (* compiler which can be produced by the m2pp command line given above *)
(*------------------------------------------------------------------------*)
(* Schnittstelle zu dblas level 2 & 3 FORTRAN 77 Subroutinen. *)
(* Low level interface to dblas level 2 & 3 FORTRAN 77 subroutines. *)
@@ -33,6 +36,7 @@
(* Last change: *)
(* *)
(* 29.10.17, MRi: Erstellen der ersten Version nur mit dgemm *)
+ (* 23.06.18, MRi: Hinzufuegen von zgemm *)
(*------------------------------------------------------------------------*)
(* Offene Punkte *)
(* *)
@@ -43,7 +47,7 @@
(* $Id: LibDBlasLxF77.def.m2pp,v 1.2 2018/01/16 09:19:51 mriedl Exp mriedl $ *)
-FROM LibDBlasL1F77 IMPORT CHAR1,INTEGER4,REAL4,DOUBLEPRECISION;
+FROM LibDBlasL1F77 IMPORT CHAR1,INTEGER4,REAL4,DOUBLEPRECISION,DOUBLECOMPLEX;
<* IF (__XDS__) THEN *>
CONST Version = "LibDBlasLxF77 for XDS Modula-2";
@@ -67,6 +71,11 @@
VAR C : (* ARRAY OF ARRAY OF *) REAL4;
VAR ldc : INTEGER4);
+ (*---------------------------------------------------------------*)
+ (* Matrix Matrix Multiplikatione / matrix matrix multiplication, *)
+ (* single precision version, see LibDBlas.def dgemm for details *)
+ (*---------------------------------------------------------------*)
+
PROCEDURE dgemm_(VAR TA : CHAR1;
VAR TB : CHAR1;
VAR M,N,K : INTEGER4;
@@ -79,4 +88,79 @@
VAR C : (* ARRAY OF ARRAY OF *) DOUBLEPRECISION;
VAR ldc : INTEGER4);
+ (*---------------------------------------------------------------*)
+ (* Matrix Matrix Multiplikatione / matrix matrix multiplication, *)
+ (* double precision version, see LibDBlas.def dgemm for details *)
+ (* See LibDBlas.def for details *)
+ (* *)
+ (* Hint: There are Fortran version with unrolled loops which *)
+ (* perform far better than the subroutines provided in the *)
+ (* standard source. If you use an optimized BLAS level3 library *)
+ (* (e.g. ATLAS) there is no need to take that into consideration *)
+ (*---------------------------------------------------------------*)
+
+PROCEDURE dgemmomp(VAR TA : CHAR1;
+ VAR TB : CHAR1;
+ VAR M,N,K : INTEGER4;
+ VAR Alpha : DOUBLEPRECISION;
+ VAR A : (* ARRAY OF ARRAY OF *) DOUBLEPRECISION;
+ VAR lda : INTEGER4;
+ VAR B : (* ARRAY OF ARRAY OF *) DOUBLEPRECISION;
+ VAR ldb : INTEGER4;
+ VAR Beta : DOUBLEPRECISION;
+ VAR C : (* ARRAY OF ARRAY OF *) DOUBLEPRECISION;
+ VAR ldc : INTEGER4);
+
+ (*---------------------------------------------------------------*)
+ (* Matrix Matrix Multiplikation / matrix matrix multiplication, *)
+ (* double precision version, see LibDBlas.def dgemm for details *)
+ (* *)
+ (* dgemmomp is an OpenMP based version of level 3 BLAS dgemmm. *)
+ (* dgemmomp is far from beeing optimal - please test if it is *)
+ (* really improving the performance within your sprecific *)
+ (* environment. On 32 bit systems the communication overhead *)
+ (* outperforms the potential gain in speed by using more than *)
+ (* one thread in many cases. If you set the number of threads to *)
+ (* one a version of dgemm using unrolled loops and blocking is *)
+ (* used if the dimension of the matriced involved exeed a *)
+ (* specific limit (see dgemmCbind.f90) so it still might be *)
+ (* worth calling dgemmomp. dgemmomp has an explicit Fortran 2003 *)
+ (* "C" interface so no need for "_" in the function name. *)
+ (* *)
+ (* HINT: The module OpenMPF77 permits controlling the number of *)
+ (* OpenMP threads used. *)
+ (*---------------------------------------------------------------*)
+
+PROCEDURE dgemmomp2(VAR TA : CHAR1;
+ VAR TB : CHAR1;
+ VAR M,N,K : INTEGER4;
+ VAR Alpha : DOUBLEPRECISION;
+ VAR A : (* ARRAY OF ARRAY OF *) DOUBLEPRECISION;
+ VAR lda : INTEGER4;
+ VAR B : (* ARRAY OF ARRAY OF *) DOUBLEPRECISION;
+ VAR ldb : INTEGER4;
+ VAR Beta : DOUBLEPRECISION;
+ VAR C : (* ARRAY OF ARRAY OF *) DOUBLEPRECISION;
+ VAR ldc : INTEGER4);
+ (*---------------------------------------------------------------*)
+ (* Another OMP parallel version of dgemm (experimantal) *)
+ (*---------------------------------------------------------------*)
+
+PROCEDURE zgemm_(VAR TA : CHAR1;
+ VAR TB : CHAR1;
+ VAR M,N,K : INTEGER4;
+ VAR Alpha : DOUBLECOMPLEX;
+ VAR A : (* ARRAY OF ARRAY OF *) DOUBLECOMPLEX;
+ VAR lda : INTEGER4;
+ VAR B : (* ARRAY OF ARRAY OF *) DOUBLECOMPLEX;
+ VAR ldb : INTEGER4;
+ VAR Beta : DOUBLECOMPLEX;
+ VAR C : (* ARRAY OF ARRAY OF *) DOUBLECOMPLEX;
+ VAR ldc : INTEGER4);
+
+ (*---------------------------------------------------------------*)
+ (* Matrix Matrix Multiplikatione / matrix matrix multiplication, *)
+ (* double complex version, see LibDBlas.def dgemm for details *)
+ (*---------------------------------------------------------------*)
+
END LibDBlasLxF77.