single/double unification of elpa1_compute_complex_template.X90

Conflicts:
	src/elpa1_compute_complex_template.X90

Makefile.am

@@ -56,7 +56,8 @@ EXTRA_libelpa@SUFFIX@_private_la_DEPENDENCIES = \
 	src/elpa2_kernels/elpa2_kernels_complex_template.X90 \
 	src/elpa2_kernels/elpa2_kernels_simple_template.X90 \
         src/redist_band.X90 \
-        src/precision_macros.h
+        src/precision_macros.h \
+        src/precision_macros_complex.h
 
 lib_LTLIBRARIES = libelpa@SUFFIX@.la
 libelpa@SUFFIX@_la_LINK = $(FCLINK) $(AM_LDFLAGS) -version-info $(ELPA_SO_VERSION)
@@ -888,6 +889,7 @@ EXTRA_DIST = \
   src/elpa2_compute_real_template.X90 \
   src/elpa2_compute_complex_template.X90 \
   src/precision_macros.h \
+  src/precision_macros_complex.h \
   src/elpa2_kernels/elpa2_kernels_real_template.X90 \
   src/elpa2_kernels/elpa2_kernels_complex_template.X90 \
   src/elpa2_kernels/elpa2_kernels_simple_template.X90 \

generate/generate_precission.py

+#!/usr/bin/python
+simple_tokens = ["tridiag_complex_PRECISION",
+                 "trans_ev_complex_PRECISION",
+                 "solve_complex_PRECISION",
+                 "hh_transform_complex_PRECISION",
+                 "elpa_transpose_vectors_complex_PRECISION",
+                 "elpa_reduce_add_vectors_complex_PRECISION",
+                 ]
+
+blas_tokens = ["PRECISION_GEMV",
+               "PRECISION_TRMV",
+               "PRECISION_GEMM",
+               "PRECISION_TRMM",
+               "PRECISION_HERK",
+               ]
+
+explicit_tokens = [("PRECISION_SUFFIX", "\"_double\"", "\"_single\""),
+                   ("MPI_COMPLEX_PRECISION", "MPI_DOUBLE_COMPLEX", "MPI_COMPLEX"),
+                   ("MPI_REAL_PRECISION", "MPI_REAL8", "MPI_REAL4"),
+                   ("KIND_PRECISION", "rk8", "rk4"),
+                   ("PRECISION_CMPLX", "DCMPLX", "CMPLX"),
+                   ("PRECISION_IMAG", "DIMAG", "AIMAG"),
+                   ("CONST_REAL_0_0", "0.0_rk8", "0.0_rk4"),
+                   ("CONST_REAL_1_0", "1.0_rk8", "1.0_rk4"),
+                   ]
+
+print "#ifdef DOUBLE_PRECISION_COMPLEX"
+
+for token in simple_tokens:
+    print "#define ", token, token.replace("PRECISION", "double")
+for token in blas_tokens:
+    print "#define ", token, token.replace("PRECISION_", "Z")    
+for token in explicit_tokens:
+    print "#define ", token[0], token[1]
+    
+print "#else"
+
+for token in simple_tokens:
+    print "#undef ", token
+for token in blas_tokens:
+    print "#undef ", token
+for token in explicit_tokens:
+    print "#undef ", token[0]
+
+for token in simple_tokens:
+    print "#define ", token, token.replace("PRECISION", "single")
+for token in blas_tokens:
+    print "#define ", token, token.replace("PRECISION_", "C")
+for token in explicit_tokens:
+    print "#define ", token[0], token[2]
+
+print "#endif"
+    

src/precision_macros_complex.h

+#ifdef DOUBLE_PRECISION_COMPLEX
+#define  tridiag_complex_PRECISION tridiag_complex_double
+#define  trans_ev_complex_PRECISION trans_ev_complex_double
+#define  solve_complex_PRECISION solve_complex_double
+#define  hh_transform_complex_PRECISION hh_transform_complex_double
+#define  elpa_transpose_vectors_complex_PRECISION elpa_transpose_vectors_complex_double
+#define  elpa_reduce_add_vectors_complex_PRECISION elpa_reduce_add_vectors_complex_double
+#define  PRECISION_GEMV ZGEMV
+#define  PRECISION_TRMV ZTRMV
+#define  PRECISION_GEMM ZGEMM
+#define  PRECISION_TRMM ZTRMM
+#define  PRECISION_HERK ZHERK
+#define  PRECISION_SUFFIX "_double"
+#define  MPI_COMPLEX_PRECISION MPI_DOUBLE_COMPLEX
+#define  MPI_REAL_PRECISION MPI_REAL8
+#define  KIND_PRECISION rk8
+#define  PRECISION_CMPLX DCMPLX
+#define  PRECISION_IMAG DIMAG
+#define  CONST_REAL_0_0 0.0_rk8
+#define  CONST_REAL_1_0 1.0_rk8
+#else
+#undef  tridiag_complex_PRECISION
+#undef  trans_ev_complex_PRECISION
+#undef  solve_complex_PRECISION
+#undef  hh_transform_complex_PRECISION
+#undef  elpa_transpose_vectors_complex_PRECISION
+#undef  elpa_reduce_add_vectors_complex_PRECISION
+#undef  PRECISION_GEMV
+#undef  PRECISION_TRMV
+#undef  PRECISION_GEMM
+#undef  PRECISION_TRMM
+#undef  PRECISION_HERK
+#undef  PRECISION_SUFFIX
+#undef  MPI_COMPLEX_PRECISION
+#undef  MPI_REAL_PRECISION
+#undef  KIND_PRECISION
+#undef  PRECISION_CMPLX
+#undef  PRECISION_IMAG
+#undef  CONST_REAL_0_0
+#undef  CONST_REAL_1_0
+#define  tridiag_complex_PRECISION tridiag_complex_single
+#define  trans_ev_complex_PRECISION trans_ev_complex_single
+#define  solve_complex_PRECISION solve_complex_single
+#define  hh_transform_complex_PRECISION hh_transform_complex_single
+#define  elpa_transpose_vectors_complex_PRECISION elpa_transpose_vectors_complex_single
+#define  elpa_reduce_add_vectors_complex_PRECISION elpa_reduce_add_vectors_complex_single
+#define  PRECISION_GEMV CGEMV
+#define  PRECISION_TRMV CTRMV
+#define  PRECISION_GEMM CGEMM
+#define  PRECISION_TRMM CTRMM
+#define  PRECISION_HERK CHERK
+#define  PRECISION_SUFFIX "_single"
+#define  MPI_COMPLEX_PRECISION MPI_COMPLEX
+#define  MPI_REAL_PRECISION MPI_REAL4
+#define  KIND_PRECISION rk4
+#define  PRECISION_CMPLX CMPLX
+#define  PRECISION_IMAG AIMAG
+#define  CONST_REAL_0_0 0.0_rk4
+#define  CONST_REAL_1_0 1.0_rk4
+#endif

test/Fortran/test_real.F90

@@ -184,7 +184,13 @@ program test_real_double_precision
       print *
       print '(a)','Standard eigenvalue problem - ELPA1, REAL version'
       print *
-      print '(3(a,i0))','Matrix size=',na,', Number of eigenvectors=',nev,', Block size=',nblk
+      print '((a,i0))', 'Num MPI proc: ', nprocs 
+      print '((a,i0))', 'Matrix size: ', na 
+      print '((a,i0))', 'Num eigenvectors: ', nev
+      print '((a,i0))', 'Blocksize: ', nblk 
+      print '((a))', 'Using gpu: NO'
+      print *
+!       print '((a,i0))','Matrix size=',na,', Number of eigenvectors=',nev,', Block size=',nblk
       print '(3(a,i0))','Number of processor rows=',np_rows,', cols=',np_cols,', total=',nprocs
       print *
    endif

test/Fortran/test_real_gpu.F90

@@ -190,11 +190,21 @@ program test_real_gpu_version_double_precision
    if(myid==0) then
       print *
       print '(a)','Standard eigenvalue problem - REAL version'
+!       if (gpuAvailable) then
+!         print *,"with GPU Version"
+!       endif
+      print *
+      print '((a,i0))', 'Num MPI proc: ', nprocs 
+      print '((a,i0))', 'Matrix size: ', na 
+      print '((a,i0))', 'Num eigenvectors: ', nev
+      print '((a,i0))', 'Blocksize: ', nblk 
       if (gpuAvailable) then
-        print *,"with GPU Version"
+        print '((a))', 'Using gpu: YES'
+      else
+        print '((a))', 'Using gpu: NO, not available'
       endif
       print *
-      print '(3(a,i0))','Matrix size=',na,', Number of eigenvectors=',nev,', Block size=',nblk
+!      print '(3(a,i0))','Matrix size=',na,', Number of eigenvectors=',nev,', Block size=',nblk
       print '(3(a,i0))','Number of processor rows=',np_rows,', cols=',np_cols,', total=',nprocs
       print *
    endif