update: Makefile for SNB-EP, alignment of a-matrix to 64 bytes

ELPA_2011.12.Intrinsics/src/elpa2.f90

@@ -1282,6 +1282,7 @@ subroutine trans_ev_tridi_to_band_real(na, nev, nblk, nbw, q, ldq, mpi_comm_rows
 
     a_dim2 = max_blk_size + nbw
 
+!DEC$ ATTRIBUTES ALIGN: 64:: a
     allocate(a(stripe_width,a_dim2,stripe_count))
     a(:,:,:) = 0
 

ELPA_2011.12.Intrinsics/test/Makefile

@@ -3,28 +3,28 @@
 # ------------------------------------------------------------------------------
 # Settings for Intel Fortran (Linux), Intel Composer XE 2011 (ifort 12.1) with AVX for Sandy Bridge:
 #
-#X86=1
-#F90=mpiifort -O3 -traceback -fpe0 -g
-#F90OPT=$(F90) -mavx
-##CC=mpiicc -O3
-##CCOPT=$(CC) -mavx
-#CC=gcc -O3
-#CCOPT=$(CC) -mavx -funsafe-loop-optimizations -funsafe-math-optimizations -ftree-vect-loop-version -ftree-vectorize
-#MKL_HOME=/opt/intel/mkl/lib/intel64
-#LIBS = -mkl -L$(MKL_HOME) -lmkl_scalapack_lp64 -lmkl_blacs_intelmpi_lp64
+X86=1
+F90=mpiifort -O3 -traceback -fpe0 -g
+F90OPT=$(F90) -mavx
+#CC=mpiicc -O3
+#CCOPT=$(CC) -mavx
+CC=gcc -O3
+CCOPT=$(CC) -mavx -funsafe-loop-optimizations -funsafe-math-optimizations -ftree-vect-loop-version -ftree-vectorize
+MKL_HOME=/opt/intel/mkl/lib/intel64
+LIBS = -mkl=sequential -L$(MKL_HOME) -lmkl_scalapack_lp64 -lmkl_blacs_intelmpi_lp64
 #
 # ------------------------------------------------------------------------------
 # Settings for Intel Fortran (Linux), Intel Composer XE 2011 (ifort 12.1) with SSE3:
 #
-X86=1
-F90=mpiifort -O3 -traceback -fpe0 -g
-F90OPT=$(F90) -msse3
-CC=gcc -O3
-CCOPT=$(CC) -msse3 -funsafe-loop-optimizations -funsafe-math-optimizations -ftree-vect-loop-version -ftree-vectorize
+#X86=1
+#F90=mpiifort -O3 -traceback -fpe0 -g
+#F90OPT=$(F90) -msse3
+#CC=gcc -O3
+#CCOPT=$(CC) -msse3 -funsafe-loop-optimizations -funsafe-math-optimizations -ftree-vect-loop-version -ftree-vectorize
 #CC=icc -O3
 #CCOPT=$(CC) -msse3
-MKL_HOME=/opt/intel/mkl/lib/intel64
-LIBS = -mkl -L$(MKL_HOME) -lmkl_scalapack_lp64 -lmkl_blacs_intelmpi_lp64
+#MKL_HOME=/opt/intel/mkl/lib/intel64
+#LIBS = -mkl=sequential -L$(MKL_HOME) -lmkl_scalapack_lp64 -lmkl_blacs_intelmpi_lp64
 #
 # ------------------------------------------------------------------------------
 # Settings for Intel Fortran (Linux), Intel Composer XE 2011 (ifort 12.1) and GCC 4.6 with FMA4 for AMD Bulldozer: