Merge branch 'master' into ELPA_GPU

.gitlab-ci.yml

@@ -370,3 +370,15 @@ single-precision-mpi-ftimings-redirect-real-avx_block6-jobs:
     - make -j 8
     - export LD_LIBRARY_PATH=$MKL_HOME/lib/intel64:$LD_LIBRARY_PATH
     - make check TEST_FLAGS='1500 50 16'
+
+set-kernel-via-environment-variable-job:
+  script:
+    - export LANG=C
+    - module load impi intel gcc mkl autotools
+    - ./autogen.sh
+    - ./configure CFLAGS="-O3 -mavx" CXXFLAGS="-O3 -mavx" FCFLAGS="-O3 -mavx" SCALAPACK_LDFLAGS="-L/afs/@cell/common/soft/intel/ics2015/15.0/mkl/lib/intel64 -lmkl_scalapack_lp64 -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lmkl_blacs_intelmpi_lp64 -lpthread -lm -Wl,-rpath,$MKL_HOME/lib/intel64" SCALAPACK_FCFLAGS="-L/afs/@cell/common/soft/intel/ics2015/15.0/mkl/lib/intel64 -lmkl_scalapack_lp64 -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lmkl_blacs_intelmpi_lp64 -lpthread -lm -I$MKLROOT/include/intel64/lp64" --enable-openmp
+    - make -j 8
+    - export LD_LIBRARY_PATH=$MKL_HOME/lib/intel64:$LD_LIBRARY_PATH
+    - export REAL_ELPA_KERNEL=REAL_ELPA_KERNEL_GENERIC
+    - export COMPLEX_ELPA_KERNEL=COMPLEX_ELPA_KERNEL_GENERIC
+    - make check TEST_FLAGS='1500 50 16'

configure.ac

@@ -881,9 +881,9 @@ if test x"${install_real_bgq}" = x"yes" ; then
 fi
 
 if test x"${use_specific_complex_kernel}" = x"no" ; then
-	AC_DEFINE([WITH_NO_SPECIFIC_COMPLEX_KERNEL],[1],[do not use only one specific complex kernel (set at compile time)])
+  AC_DEFINE([WITH_NO_SPECIFIC_COMPLEX_KERNEL],[1],[do not use only one specific complex kernel (set at compile time)])
 else
-	AC_DEFINE([WITH_ONE_SPECIFIC_COMPLEX_KERNEL],[1],[use only one specific complex kernel (set at compile time)])
+  AC_DEFINE([WITH_ONE_SPECIFIC_COMPLEX_KERNEL],[1],[use only one specific complex kernel (set at compile time)])
 fi
 
 if test x"${use_specific_real_kernel}" = x"no" ; then

m4/ax_elpa_specific_kernels.m4

@@ -86,6 +86,30 @@ AC_DEFUN([DEFINE_OPTION_SPECIFIC_REAL_KERNEL],[
      fi
     fi
 
+    if test x"${install_real_sse_block2}" = x"yes" ; then
+     if test x"${can_compile_sse_intrinsics}" = x"no" ; then
+       AC_MSG_ERROR([$2 kernel was set, but cannot be compiled!])
+     else
+       want_sse=yes
+     fi
+    fi
+
+    if test x"${install_real_sse_block4}" = x"yes" ; then
+     if test x"${can_compile_sse_intrinsics}" = x"no" ; then
+       AC_MSG_ERROR([$2 kernel was set, but cannot be compiled!])
+     else
+       want_sse=yes
+     fi
+    fi
+
+    if test x"${install_real_sse_block6}" = x"yes" ; then
+     if test x"${can_compile_sse_inrinsics}" = x"no" ; then
+       AC_MSG_ERROR([$2 kernel was set, but cannot be compiled!])
+     else
+       want_sse=yes
+     fi
+    fi
+
     if test x"${install_real_avx_block2}" = x"yes" ; then
      if test x"${can_compile_avx}" = x"no" ; then
        AC_MSG_ERROR([$2 kernel was set, but cannot be compiled!])
@@ -111,10 +135,10 @@ AC_DEFUN([DEFINE_OPTION_SPECIFIC_REAL_KERNEL],[
     fi
 
     AC_MSG_NOTICE([$1 will be the only compiled kernel for real case])
-    if test x"${want_gpu}" = x"yes" ; then
-      AC_MSG_WARN([At the moment this disables GPU support!])
-      AC_MSG_WARN([IF GPU support is wanted do NOT specify a specific real kernel])
-    fi
+#    if test x"${want_gpu}" = x"yes" ; then
+#      AC_MSG_WARN([At the moment this disables GPU support!])
+#      AC_MSG_WARN([IF GPU support is wanted do NOT specify a specific real kernel])
+#    fi
    else
     AC_MSG_FAILURE([$1 failed; A specific kernel for real case has already been defined before!])
    fi
@@ -145,7 +169,7 @@ AC_DEFUN([DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL],[
     want_avx=no
     want_avx2=no
 
-    install_gpu=no
+#    install_gpu=no
     use_specific_complex_kernel=yes
     dnl now set the specific kernel
     $3=yes
@@ -161,6 +185,8 @@ AC_DEFUN([DEFINE_OPTION_SPECIFIC_COMPLEX_KERNEL],[
     if test x"${install_complex_sse_assembly}" = x"yes" ; then
      if test x"${can_compile_sse_assembly}" = x"no" ; then
        AC_MSG_ERROR([$2 kernel was set, but cannot be compiled!])
+     else
+       want_sse=yes
      fi
     fi
 

src/elpa2_utilities.F90

@@ -81,7 +81,6 @@ module ELPA2_utilities
             REAL_ELPA_KERNEL_AVX2_BLOCK4, REAL_ELPA_KERNEL_AVX2_BLOCK6, &
             REAL_ELPA_KERNEL_GPU,  DEFAULT_REAL_ELPA_KERNEL
 
-
   public :: COMPLEX_ELPA_KERNEL_GENERIC, COMPLEX_ELPA_KERNEL_GENERIC_SIMPLE, &
             COMPLEX_ELPA_KERNEL_BGP, COMPLEX_ELPA_KERNEL_BGQ,                &
             COMPLEX_ELPA_KERNEL_SSE, COMPLEX_ELPA_KERNEL_SSE_BLOCK1,         &

src/mod_compute_hh_trafo_complex.F90

@@ -333,7 +333,6 @@ module compute_hh_trafo_complex
 #endif /* WITH_NO_SPECIFIC_COMPLEX_KERNEL */
 
 #if defined(WITH_NO_SPECIFIC_COMPLEX_KERNEL) || (defined(WITH_ONE_SPECIFIC_COMPLEX_KERNEL) && !defined(WITH_COMPLEX_AVX_BLOCK2_KERNEL) && !defined(WITH_COMPLEX_AVX2_BLOCK2_KERNEL))
-
             ttt = mpi_wtime()
             do j = ncols, 1, -1
 #ifdef WITH_OPENMP

src/mod_mpi_stubs.F90

@@ -121,6 +121,6 @@ module elpa_mpi_stubs
       new_comm = mpi_communicator
       mpierr = 0
       return
-    end subroutine
+    end subroutine mpi_comm_split
 
 end module

test/fortran_test_programs/elpa_test_programs_print_headers.X90

@@ -168,7 +168,7 @@
 #ifdef WITH_REAL_GENERIC_SIMPLE_KERNEL
      print *,"GENERIC SIMPLE kernel for real matrices"
 #endif
-#ifdef WITH_REAL_SSE_KERNEL
+#ifdef WITH_REAL_SSE_ASSEMBLY_KERNEL
      print *,"SSE ASSEMBLER kernel for real matrices"
 #endif
 #ifdef WITH_REAL_BGP_KERNEL
@@ -213,7 +213,7 @@
 #ifdef WITH_COMPLEX_GENERIC_SIMPLE_KERNEL
      print *,"GENERIC SIMPLE kernel for complex matrices"
 #endif
-#ifdef WITH_COMPLEX_SSE_KERNEL
+#ifdef WITH_COMPLEX_SSE_ASSEMBLY_KERNEL
      print *,"SSE ASSEMBLER kernel for complex matrices"
 #endif
 

test/fortran_test_programs/test_complex2_choose_kernel_with_api.F90

@@ -349,8 +349,6 @@ program test_complex2_choose_kernel_with_api_double_precision
                                  COMPLEX_ELPA_KERNEL_SSE)
 #endif
 
-#ifdef WITH_ONE_SPECIFIC_COMPLEX_KERNEL
-
 #ifdef  WITH_COMPLEX_SSE_BLOCK2_KERNEL
                                  COMPLEX_ELPA_KERNEL_SSE_BLOCK2)
 #else
@@ -367,33 +365,12 @@ program test_complex2_choose_kernel_with_api_double_precision
 #endif
 #endif
 
-#else /* WITH_ONE_SPECIFIC_COMPLEX_KERNEL */
-
-#ifdef  WITH_COMPLEX_SSE_BLOCK1_KERNEL
-                                 COMPLEX_ELPA_KERNEL_SSE_BLOCK1)
-#endif
-
-#ifdef  WITH_COMPLEX_SSE_BLOCK2_KERNEL
-                                 COMPLEX_ELPA_KERNEL_SSE_BLOCK2)
-#endif
-
-#ifdef  WITH_COMPLEX_AVX_BLOCK1_KERNEL
-                                 COMPLEX_ELPA_KERNEL_AVX_BLOCK1)
-#endif
-
-#ifdef  WITH_COMPLEX_AVX_BLOCK2_KERNEL
-                                 COMPLEX_ELPA_KERNEL_AVX_BLOCK2)
-#endif
-
-#endif  /*   WITH_ONE_SPECIFIC_COMPLEX_KERNEL */
-
 #ifdef  WITH_GPU_VERSION
                                  COMPLEX_ELPA_KERNEL_GPU)
 #endif
 
 #endif /* WITH_ONE_SPECIFIC_COMPLEX_KERNEL */
 
-
    if (.not.(successELPA)) then
       write(error_unit,*) "solve_evp_complex_2stage produced an error! Aborting..."
 #ifdef WITH_MPI

test/fortran_test_programs/test_real2_choose_kernel_with_api.F90

@@ -394,7 +394,6 @@ program test_real2_choose_kernel_with_api_double_precision
                               REAL_ELPA_KERNEL_AVX_BLOCK6)
 #endif
 
-
 #endif /* WITH_ONE_SPECIFIC_REAL_KERNEL */
 
 #ifdef WITH_REAL_BGP_KERNEL