Merge branch 'master_fix_up' into ELPA_KNL

Makefile.am

@@ -52,22 +52,19 @@ EXTRA_libelpa@SUFFIX@_private_la_DEPENDENCIES = \
         src/elpa1_compute_template.X90 \
         src/elpa2_compute_real_template.X90 \
         src/elpa2_compute_complex_template.X90 \
-	src/elpa2_bandred_real_template.X90 \
+		src/elpa2_template.X90 \
+	src/elpa2_bandred_template.X90 \
 	src/elpa2_symm_matrix_allreduce_real_template.X90 \
-	src/elpa2_trans_ev_band_to_full_real_template.X90 \
-	src/elpa2_tridiag_band_real_template.X90 \
+	src/elpa2_trans_ev_band_to_full_template.X90 \
+	src/elpa2_tridiag_band_template.X90 \
 	src/elpa2_trans_ev_tridi_to_band_real_template.X90 \
-	src/elpa2_bandred_complex_template.X90 \
 	src/elpa2_herm_matrix_allreduce_complex_template.X90 \
-	src/elpa2_trans_ev_band_to_full_complex_template.X90 \
-	src/elpa2_tridiag_band_complex_template.X90 \
 	src/elpa2_trans_ev_tridi_to_band_complex_template.X90 \
 	src/elpa2_kernels/elpa2_kernels_real_template.X90 \
 	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_complex.h
+        src/precision_macros.h
 
 lib_LTLIBRARIES = libelpa@SUFFIX@.la
 libelpa@SUFFIX@_la_LINK = $(FCLINK) $(AM_LDFLAGS) -version-info $(ELPA_SO_VERSION)
@@ -356,6 +353,7 @@ dist_files_DATA = \
   test/Fortran/test_real2_default.F90 \
   test/Fortran/test_real2_qr.F90 \
   test/Fortran/test_real2_api.F90 \
+  test/Fortran/test_real2_banded.F90 \
   test/Fortran/test_real.F90 \
   test/Fortran/test_real_with_c.F90 \
   test/Fortran/test_toeplitz.F90 \
@@ -386,6 +384,7 @@ noinst_PROGRAMS = \
   elpa2_test_real_default@SUFFIX@ \
   elpa2_test_real_qr@SUFFIX@ \
   elpa2_test_real_api@SUFFIX@ \
+  elpa2_test_real_banded@SUFFIX@ \
   elpa2_test_complex@SUFFIX@ \
   elpa2_test_complex_default@SUFFIX@ \
   elpa2_test_complex_api@SUFFIX@ \
@@ -573,6 +572,11 @@ elpa2_test_real_api@SUFFIX@_LDADD = $(build_lib)
 elpa2_test_real_api@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) @FC_MODOUT@private_modules @FC_MODINC@private_modules
 EXTRA_elpa2_test_real_api@SUFFIX@_DEPENDENCIES = test/Fortran/elpa_print_headers.X90
 
+elpa2_test_real_banded@SUFFIX@_SOURCES = test/Fortran/test_real2_banded.F90
+elpa2_test_real_banded@SUFFIX@_LDADD = $(build_lib)
+elpa2_test_real_banded@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) @FC_MODOUT@private_modules @FC_MODINC@private_modules
+EXTRA_elpa2_test_real_banded@SUFFIX@_DEPENDENCIES = test/Fortran/elpa_print_headers.X90
+
 elpa1_test_complex@SUFFIX@_SOURCES = test/Fortran/test_complex.F90
 elpa1_test_complex@SUFFIX@_LDADD = $(build_lib)
 elpa1_test_complex@SUFFIX@_FCFLAGS = $(AM_FCFLAGS) @FC_MODOUT@private_modules @FC_MODINC@private_modules
@@ -772,6 +776,7 @@ check_SCRIPTS = \
   elpa2_test_real_qr@SUFFIX@.sh \
   elpa2_test_complex_default@SUFFIX@.sh \
   elpa2_test_real_api@SUFFIX@.sh \
+  elpa2_test_real_banded@SUFFIX@.sh \
   elpa2_test_complex_api@SUFFIX@.sh \
   elpa_driver_real@SUFFIX@.sh \
   elpa_driver_complex@SUFFIX@.sh \
@@ -942,18 +947,15 @@ EXTRA_DIST = \
   src/elpa1_tridiag_template.X90 \
   src/elpa2_compute_real_template.X90 \
   src/elpa2_compute_complex_template.X90 \
-  src/elpa2_bandred_complex_template.X90 \
-  src/elpa2_bandred_real_template.X90 \
+  src/elpa2_bandred_template.X90 \
   src/elpa2_herm_matrix_allreduce_complex_template.X90 \
   src/elpa2_symm_matrix_allreduce_real_template.X90 \
-  src/elpa2_trans_ev_band_to_full_complex_template.X90 \
-  src/elpa2_trans_ev_band_to_full_real_template.X90 \
+  src/elpa2_template.X90 \
+  src/elpa2_tridiag_band_template.X90 \
+  src/elpa2_trans_ev_band_to_full_template.X90 \
   src/elpa2_trans_ev_tridi_to_band_complex_template.X90 \
   src/elpa2_trans_ev_tridi_to_band_real_template.X90 \
-  src/elpa2_tridiag_band_complex_template.X90 \
-  src/elpa2_tridiag_band_real_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_precision.py

@@ -2,9 +2,9 @@
 import sys
 
 simple_tokens = [
+    "PRECISION",
     "elpa_transpose_vectors_NUMBER_PRECISION",
     "elpa_reduce_add_vectors_NUMBER_PRECISION",
-
     "bandred_NUMBER_PRECISION",
     "trans_ev_band_to_full_NUMBER_PRECISION",
     "tridiag_band_NUMBER_PRECISION",
@@ -15,10 +15,11 @@ simple_tokens = [
     "solve_tridi_PRECISION",
     "solve_tridi_col_PRECISION",
     "solve_tridi_single_problem_PRECISION",
-
+    "solve_evp_NUMBER_2stage_PRECISION",
     "qr_pdgeqrf_2dcomm_PRECISION",
     "hh_transform_NUMBER_PRECISION",
     "symm_matrix_allreduce_PRECISION",
+    "herm_matrix_allreduce_PRECISION",
     "redist_band_NUMBER_PRECISION",
     "unpack_row_NUMBER_cpu_PRECISION",
     "unpack_row_NUMBER_cpu_openmp_PRECISION",
@@ -45,9 +46,19 @@ simple_tokens = [
     "global_product_PRECISION",
     "add_tmp_PRECISION",
     "v_add_s_PRECISION",
+    "launch_compute_hh_trafo_c_kernel_NUMBER_PRECISION",
+    "compute_hh_trafo_NUMBER_gpu_PRECISION", 
+    "launch_my_pack_c_kernel_NUMBER_PRECISION",
+    "launch_my_unpack_c_kernel_NUMBER_PRECISION",
+    "launch_compute_hh_dotp_c_kernel_NUMBER_PRECISION",    
+    "launch_extract_hh_tau_c_kernel_NUMBER_PRECISION",
+    "AVAILABLE_UPCASENUMBER_ELPA_KERNELS",
+    "UPCASENUMBER_ELPA_KERNEL_GENERIC",
+    "DEFAULT_UPCASENUMBER_ELPA_KERNEL",
+    "UPCASENUMBER_ELPA_KERNEL_NAMES",
+    "UPCASENUMBER_ELPA_KERNEL_GPU",
 ]
 
-
 blas_tokens = [
     "PRECISION_GEMV",
     "PRECISION_TRMV",
@@ -57,6 +68,8 @@ blas_tokens = [
     "PRECISION_SYRK",
     "PRECISION_SYMV",
     "PRECISION_SYMM",
+    "PRECISION_HEMV",
+    "PRECISION_HER2",
     "PRECISION_SYR2",
     "PRECISION_SYR2K",
     "PRECISION_GEQRF",
@@ -75,6 +88,7 @@ blas_tokens = [
 explicit_tokens_complex = [
     ("PRECISION_SUFFIX", "\"_double\"", "\"_single\""),
     ("MPI_COMPLEX_PRECISION", "MPI_DOUBLE_COMPLEX", "MPI_COMPLEX"),
+    ("MPI_COMPLEX_EXPLICIT_PRECISION", "MPI_COMPLEX16", "MPI_COMPLEX8"),
     ("MPI_REAL_PRECISION", "MPI_REAL8", "MPI_REAL4"),
     ("KIND_PRECISION", "rk8", "rk4"),
     ("PRECISION_CMPLX", "DCMPLX", "CMPLX"),
@@ -82,8 +96,15 @@ explicit_tokens_complex = [
     ("PRECISION_REAL", "DREAL", "REAL"),
     ("CONST_REAL_0_0", "0.0_rk8", "0.0_rk4"),
     ("CONST_REAL_1_0", "1.0_rk8", "1.0_rk4"),
+    ("CONST_REAL_0_5", "0.5_rk8", "0.5_rk4"),
+    ("CONST_COMPLEX_PAIR_0_0", "(0.0_rk8,0.0_rk8)", "(0.0_rk4,0.0_rk4)"),
+    ("CONST_COMPLEX_PAIR_1_0", "(1.0_rk8,0.0_rk8)", "(1.0_rk4,0.0_rk4)"),
+    ("CONST_COMPLEX_PAIR_NEGATIVE_1_0", "(-1.0_rk8,0.0_rk8)", "(-1.0_rk4,0.0_rk4)"),
+    ("CONST_COMPLEX_PAIR_NEGATIVE_0_5", "(-0.5_rk8,0.0_rk8)", "(-0.5_rk4,0.0_rk4)"),
     ("CONST_COMPLEX_0_0", "0.0_ck8", "0.0_ck4"),
+    ("CONST_COMPLEX_1_0", "1.0_ck8", "1.0_ck4"),
     ("size_of_PRECISION_complex", "size_of_double_complex_datatype", "size_of_single_complex_datatype"),
+    ("C_DATATYPE_KIND", "c_double", "c_float"),
 ]
 
 explicit_tokens_real = [
@@ -95,6 +116,7 @@ explicit_tokens_real = [
     ("CONST_8_0", "8.0_rk8", "8.0_rk4"),
     ("size_of_PRECISION_real",  "size_of_double_real_datatype",  "size_of_single_real_datatype"),
     ("MPI_REAL_PRECISION", "MPI_REAL8", "MPI_REAL4"),
+    ("C_DATATYPE_KIND", "c_double", "c_float"),
 ]
 
 
@@ -103,7 +125,10 @@ blas_prefixes = {("real","single") : "S", ("real","double") : "D", ("complex","s
 
 def print_variant(number, precision, explicit):
     for token in simple_tokens:
-        print "#define ", token.replace("NUMBER", number), token.replace("PRECISION", precision).replace("NUMBER", number)
+        print "#define ", token, token.replace("PRECISION", precision).replace("UPCASENUMBER", number.upper()).replace("NUMBER", number)
+        print "#define ", token + "_STR", "'" + token.replace("PRECISION", precision).replace("UPCASENUMBER", number.upper()).replace("NUMBER", number) + "'"
+        if("NUMBER" in token):
+            print "#define ", token.replace("NUMBER", number), token.replace("PRECISION", precision).replace("NUMBER", number)
     for token in blas_tokens:
         print "#define ", token, token.replace("PRECISION_", blas_prefixes[(number, precision)])    
     for token in explicit:
@@ -111,28 +136,51 @@ def print_variant(number, precision, explicit):
     
 def print_undefs(number, explicit):
     for token in simple_tokens:
-        print "#undef ", token.replace("NUMBER", number)
+        print "#undef ", token
+        print "#undef ", token + "_STR"
+        if("NUMBER" in token):
+            print "#undef ", token.replace("NUMBER", number)
     for token in blas_tokens:
         print "#undef ", token
     for token in explicit:
         print "#undef ", token[0]
 
 
-if(sys.argv[1] == "complex"):
-    print "#ifdef DOUBLE_PRECISION_COMPLEX"
-    print_undefs("complex", explicit_tokens_complex)
-    print_variant("complex", "double", explicit_tokens_complex)
-    print "#else"
-    print_undefs("complex", explicit_tokens_complex)
-    print_variant("complex", "single", explicit_tokens_complex)
-    print "#endif"
-elif(sys.argv[1] == "real"):    
-    print "#ifdef DOUBLE_PRECISION_REAL"
-    print_undefs("real", explicit_tokens_real)
-    print_variant("real", "double", explicit_tokens_real)
-    print "#else"
-    print_undefs("real", explicit_tokens_real)
-    print_variant("real", "single", explicit_tokens_real)
-    print "#endif"
-else:
-    assert(False)
+print "#ifdef REALCASE"
+print "#undef  MATH_DATATYPE"
+print "#define  MATH_DATATYPE real"
+print_undefs("real", explicit_tokens_real)
+#print_undefs("complex", explicit_tokens_complex)
+print "#ifdef DOUBLE_PRECISION"
+print_variant("real", "double", explicit_tokens_real)
+print "#endif"
+print "#ifdef SINGLE_PRECISION"
+print_variant("real", "single", explicit_tokens_real)
+print "#endif"
+print "#endif"
+
+print "#ifdef COMPLEXCASE"
+print "#undef  MATH_DATATYPE"
+print "#define  MATH_DATATYPE complex"
+#print_undefs("real", explicit_tokens_real)
+print_undefs("complex", explicit_tokens_complex)
\ No newline at end of file
+print "#ifdef DOUBLE_PRECISION"
+print_variant("complex", "double", explicit_tokens_complex)
+print "#endif"
+print "#ifdef SINGLE_PRECISION"
+print_variant("complex", "single", explicit_tokens_complex)
+print "#endif"
+print "#endif"
+
+#print "#elif MACROS_TYPE == COMPLEX_DOUBLE"
+#print "#undef  NUMBER"
+#print_undefs("complex", explicit_tokens_complex)
+#print "#define  NUMBER complex"
+#print_variant("complex", "double", explicit_tokens_complex)
+
+#print "#elif MACROS_TYPE == COMPLEX_SINGLE"
+#print "#undef  NUMBER"
+#print_undefs("complex", explicit_tokens_complex)
+#print "#define  NUMBER complex"
+#print_variant("complex", "single", explicit_tokens_complex)
+#print "#endif"

src/elpa1_compute_private.F90

@@ -159,12 +159,14 @@ module ELPA1_COMPUTE
 #define DATATYPE REAL(kind=rk8)
 #define BYTESIZE 8
 #define REALCASE 1
+#define DOUBLE_PRECISION 1
 #include "elpa_transpose_vectors.X90"
 #include "elpa_reduce_add_vectors.X90"
 #undef DOUBLE_PRECISION_REAL
 #undef DATATYPE
 #undef BYTESIZE
 #undef REALCASE
+#undef DOUBLE_PRECISION
 
 ! single precision
 #ifdef WANT_SINGLE_PRECISION_REAL
@@ -173,11 +175,13 @@ module ELPA1_COMPUTE
 #define DATATYPE REAL(kind=rk4)
 #define BYTESIZE 4
 #define REALCASE 1
+#define SINGLE_PRECISION 1
 #include "elpa_transpose_vectors.X90"
 #include "elpa_reduce_add_vectors.X90"
 #undef DATATYPE
 #undef BYTESIZE
 #undef REALCASE
+#undef SINGLE_PRECISION
 
 #endif
 
@@ -187,11 +191,13 @@ module ELPA1_COMPUTE
 #define DATATYPE COMPLEX(kind=ck8)
 #define BYTESIZE 16
 #define COMPLEXCASE 1
+#define DOUBLE_PRECISION 1
 #include "elpa_transpose_vectors.X90"
 #include "elpa_reduce_add_vectors.X90"
 #undef DATATYPE
 #undef BYTESIZE
 #undef COMPLEXCASE
+#undef DOUBLE_PRECISION
 #undef DOUBLE_PRECISION_COMPLEX
 
 #ifdef WANT_SINGLE_PRECISION_COMPLEX
@@ -200,11 +206,13 @@ module ELPA1_COMPUTE
 #undef DOUBLE_PRECISION_REAL
 #define DATATYPE COMPLEX(kind=ck4)
 #define COMPLEXCASE 1
+#define SINGLE_PRECISION 1
 #include "elpa_transpose_vectors.X90"
 #include "elpa_reduce_add_vectors.X90"
 #undef DATATYPE
 #undef BYTESIZE
 #undef COMPLEXCASE
+#undef SINGLE_PRECISION
 
 #endif /* WANT_SINGLE_PRECISION_COMPLEX */
 
@@ -213,6 +221,9 @@ module ELPA1_COMPUTE
 #define REAL_DATATYPE rk8
 
 #define REALCASE 1
+#define DOUBLE_PRECISION 1
+
+! remove? :
 #undef COMPLEXCASE
 
 #include "elpa1_compute_template.X90"
@@ -220,6 +231,7 @@ module ELPA1_COMPUTE
 #undef DOUBLE_PRECISION_REAL
 #undef REAL_DATATYPE
 #undef REALCASE
+#undef DOUBLE_PRECISION
 
 ! real single precision
 #if defined(WANT_SINGLE_PRECISION_REAL)
@@ -228,11 +240,15 @@ module ELPA1_COMPUTE
 #define REAL_DATATYPE rk4
 
 #define REALCASE 1
+#define SINGLE_PRECISION 1
+
+!remove? :
 #undef COMPLEXCASE
 
 #include "elpa1_compute_template.X90"
 
 #undef REALCASE
+#undef SINGLE_PRECISION
 
 #undef DOUBLE_PRECISION_REAL
 #undef REAL_DATATYPE
@@ -245,10 +261,14 @@ module ELPA1_COMPUTE
 #define COMPLEX_DATATYPE ck8
 
 #define COMPLEXCASE 1
+#define DOUBLE_PRECISION 1
+
+! remove? :
 #undef REALCASE
 #include "elpa1_compute_template.X90"
 
 #undef COMPLEXCASE
+#undef DOUBLE_PRECISION
 
 #undef DOUBLE_PRECISION_COMPLEX
 #undef REAL_DATATYPE
@@ -263,11 +283,14 @@ module ELPA1_COMPUTE
 #define COMPLEX_DATATYPE ck4
 
 #define COMPLEXCASE 1
+#define SINGLE_PRECISION 1
+!remove ? :
 #undef REALCASE
 
 #include "elpa1_compute_template.X90"
 
 #undef COMPLEXCASE
+#undef SINGLE_PRECISION
 
 #undef DOUBLE_PRECISION_COMPLEX
 #undef COMPLEX_DATATYPE

src/elpa1_compute_template.X90

@@ -52,13 +52,7 @@
 ! distributed along with the original code in the file "COPYING".
 #endif
 
-#if REALCASE == 1
 #include "precision_macros.h"
-#endif
-
-#if COMPLEXCASE == 1
-#include "precision_macros_complex.h"
-#endif
 
 #if REALCASE == 1
 

src/elpa2_compute.F90

@@ -167,8 +167,12 @@ module ELPA2_compute
 #define DOUBLE_PRECISION_REAL 1
 #define REAL_DATATYPE rk8
 
+#define REALCASE 1
+#define DOUBLE_PRECISION 1
 #include "precision_macros.h"
 #include "elpa2_compute_real_template.X90"
+#undef REALCASE
+#undef DOUBLE_PRECISION
 
 #undef DOUBLE_PRECISION_REAL
 #undef REAL_DATATYPE
@@ -179,8 +183,12 @@ module ELPA2_compute
 #undef DOUBLE_PRECISION_REAL
 #define REAL_DATATYPE rk4
 
+#define REALCASE 1
+#define SINGLE_PRECISION 1
 #include "precision_macros.h"
 #include "elpa2_compute_real_template.X90"
+#undef REALCASE
+#undef SINGLE_PRECISION
 
 #undef DOUBLE_PRECISION_REAL
 #undef REAL_DATATYPE
@@ -192,8 +200,12 @@ module ELPA2_compute
 #define REAL_DATATYPE rk8
 #define COMPLEX_DATATYPE ck8
 
-#include "precision_macros_complex.h"
+#define COMPLEXCASE 1
+#define DOUBLE_PRECISION 1
+#include "precision_macros.h"
 #include "elpa2_compute_complex_template.X90"
+#undef COMPLEXCASE
+#undef DOUBLE_PRECISION
 
 #undef DOUBLE_PRECISION_COMPLEX
 #undef REAL_DATATYPE
@@ -207,8 +219,12 @@ module ELPA2_compute
 #define REAL_DATATYPE rk4
 #define COMPLEX_DATATYPE ck4
 
-#include "precision_macros_complex.h"
+#define COMPLEXCASE 1
+#define SINGLE_PRECISION 1
+#include "precision_macros.h"
 #include "elpa2_compute_complex_template.X90"
+#undef COMPLEXCASE
+#undef SINGLE_PRECISION
 
 #undef DOUBLE_PRECISION_COMPLEX
 #undef COMPLEX_DATATYPE

src/elpa2_compute_complex_template.X90

@@ -61,35 +61,28 @@
 ! distributed along with the original code in the file "COPYING".
 #endif