Merge pull request #4 from RSE-Cambridge/fix-build-process

Fix build process: autoconf deprecated and templates provided

Makefile

@@ -49,11 +49,11 @@ default :
 # If "|| exit 1" is not present, the error code from make in subdirectories
 # is not returned and make goes on even if compilation has failed
 
-pw : bindir libfft libla mods liblapack libs libiotk 
+pw : bindir libfft libla mods libs libiotk 
 	if test -d PW ; then \
 	( cd PW ; $(MAKE) TLDEPS= all || exit 1) ; fi
 
-pw-lib : bindir libfft libla mods liblapack libs libiotk
+pw-lib : bindir libfft libla mods libs libiotk
 	if test -d PW ; then \
 	( cd PW ; $(MAKE) TLDEPS= pw-lib || exit 1) ; fi
 
@@ -117,7 +117,7 @@ all	:	pw
 # compile modules, libraries, directory for binaries, etc
 ###########################################################
 
-libla : touch-dummy libeigensolver_gpu liblapack
+libla : touch-dummy libeigensolver_gpu
 	( cd LAXlib ; $(MAKE) TLDEPS= all || exit 1 )
 
 libfft : touch-dummy
@@ -139,10 +139,10 @@ bindir :
 # Targets for external libraries
 ############################################################
 
-libblas : liblapack
+#libblas : liblapack
 
-liblapack: touch-dummy
-	cd install ; $(MAKE) -f extlibs_makefile $@
+#liblapack: touch-dummy
+#	cd install ; $(MAKE) -f extlibs_makefile $@
 
 libeigensolver_gpu: touch-dummy
 	cd install ; $(MAKE) -f extlibs_makefile $@
@@ -213,19 +213,12 @@ clean :
 veryclean : clean
 	- @(cd install ; $(MAKE) -f extlibs_makefile veryclean)
 	- rm -rf install/patch-plumed
-	- cd install ; rm -f config.log configure.msg config.status \
-		CPV/version.h ChangeLog* intel.pcl */intel.pcl
-	- rm -rf include/configure.h install/make_wannier90.inc
-	- cd install ; rm -fr autom4te.cache
-	- cd pseudo; ./clean_ps ; cd -
+	- cd install ; rm -f CPV/version.h ChangeLog* intel.pcl */intel.pcl
 	- cd install; ./clean.sh ; cd -
-	- cd include; ./clean.sh ; cd -
 	- rm -f espresso.tar.gz
-	- rm -rf make.inc
+#	- rm -rf make.inc
 
-# remove everything not in the original distribution
 distclean : veryclean
-#	( cd install ; $(MAKE) -f plugins_makefile $@ || exit 1 )
 
 tar :
 	@if test -f espresso.tar.gz ; then /bin/rm espresso.tar.gz ; fi

Modules/environment.f90

@@ -207,6 +207,10 @@ SUBROUTINE parallel_info ( )
     WRITE( stdout, '(/5X,"Parallel version (MPI), running on ",&
          &I5," processors")' ) nproc 
 #endif
+#if defined(USE_CUDA)
+    WRITE( stdout, '(/5X,"GPU accelerated version",/5x)')
+#endif 
+    !
 #if defined(USE_CUDA)
     WRITE( stdout, '(/5X,"GPU accelerated version",/5x)')
 #endif 

Modules/make.depend

@@ -489,4 +489,3 @@ xsf.o : kind.o
 ylmr2.o : constants.o
 ylmr2.o : kind.o
 zdotc_wrapper.o : kind.o
-environment.o : ../include/configure.h

PW/src/Makefile

@@ -253,10 +253,18 @@ wannier_occ.o \
 wannier_enrg.o \
 wyckoff.o       
 
+
 QEMODS=../../Modules/libqemod.a ../../FFTXlib/libqefft.a ../../LAXlib/libqela.a
 
 TLDEPS=bindir mods libs liblapack libblas
 
+# If F90FLAGS_WORKAROUND is defined some files are compiled differently
+# # as workaround for compiler bugs
+ifdef F90FLAGS_WORKAROUND
+read_file_new.o pw_restart.o :
+	$(MPIF90) $(F90FLAGS_WORKAROUND) -c $(*).f90 -o $(*).o
+endif
+
 all : tldeps pw.x manypw.x generate_vdW_kernel_table.x generate_rVV10_kernel_table.x
 
 pw.x : $(PWOBJS) libpw.a $(LIBOBJS) $(QEMODS)

README.md

@@ -32,21 +32,18 @@ strongly recommend for its memory capacity and performance.
 This version of QE-GPU runs **exclusively** in parallel, Open MPI is required
 and also Intel MKL.
 
-### Installation
 
-The installation process follows the same procedure as Quantum ESPRESSO suite:
+### Installation
 
-```
-./configure --enable-gpu
-make pw
-```
+To compile QE-GPU, copy a `make.inc` template from "install/" directory into the main directory and run make.
 
-"make" alone prints a list of acceptable targets with various level of GPU
-support. Binaries go in "bin/". Additional configure options are made
-available to customize the building process:
+These make.inc templates are available:
+* `make.inc_x86-64` to compile on any x86-64 machines with NVIDIA GPU (`GPU_ARCH={35, 60}`)
+* `make.inc_x86-64_CPU-only` to compile on x86-64 without GPU support
+* `make.inc_CRAY_PizDaint` to compile on Piz Daint at CSCS, CRAY XC30 with P100 PCI GPU support (`GPU_ARCH=60`)
+* `make.inc_POWER_DAVIDE` to compile on PRACE D.A.V.I.D.E. at E4, POWER8 system with GPU support (`GPU_ARCH=60`)
 
-* `--enable-gpu=<kepler|pascal>` to enable GPU support (default: "pascal"
-architecture selected)
+By invoking _make_ alone a list of acceptable targets will be displayed. Binaries go in "bin/". Read comments in the `make.inc` templates to customize it further based on your ebvironment and where math libraries are located. The architectures/environments supported are x86-64, POWER and CRAY.
 
 The QE-GPU package has been reduced in size to the minimum essential. For more
 information, please refer to the general documentation provided with the full

include/c_defs.h

@@ -0,0 +1,22 @@
+/* include/c_defs.h.  Generated from c_defs.h.in by configure.  */
+/*
+Copyright (C) 2006 Quantum-ESPRESSO group
+This file is distributed under the terms of the
+GNU General Public License. See the file `License'
+in the root directory of the present distribution,
+or http://www.gnu.org/copyleft/gpl.txt .
+*/
+
+/* File c_defs.h.in is used by configure to generate c_defs.h
+   Variables that configure defines should be #undef-ined in
+   include/c_defs.h.in !!! */
+
+/* fortran-to-C naming convention, for functions with and without
+   underscores in the name (some compilers treat them differently) */ 
+
+#define F77_FUNC(name,NAME) name ## _
+#define F77_FUNC_(name,NAME) name ## _
+
+/* do we have the mallinfo structure (see clib/memstat.c) ? */
+
+#define HAVE_MALLINFO 1

include/fft_defs.h

@@ -0,0 +1,2 @@
+
+#define C_POINTER integer*8

install/Makefile.lib_eigsolve.in → install/Makefile.lib_eigsolve

@@ -1,6 +1,8 @@
-FLAGS = -O3 -mp -pgf90libs -Mcuda=cc@gpu_arch_num@,cuda8.0,ptxinfo,maxregcount:64 -Mlarge_arrays
-CUFLAGS = -O3 -arch=sm_@gpu_arch_num@
-FLAGS2 = -O3 -mp -pgf90libs -Mcuda=cc@gpu_arch_num@,cuda8.0,ptxinfo,maxregcount:64,nollvm -Mlarge_arrays
+include ../../make.inc
+
+FLAGS = -O3 -mp -pgf90libs -Mcuda=cc$(GPU_ARCH),cuda8.0,ptxinfo,maxregcount:64 -Mlarge_arrays
+CUFLAGS = -O3 -arch=sm_$(GPU_ARCH)
+FLAGS2 = -O3 -mp -pgf90libs -Mcuda=cc$(GPU_ARCH),cuda8.0,ptxinfo,maxregcount:64,nollvm -Mlarge_arrays
 #OPTFLAGS = -DUSE_NVTX
 
 .NOTPARALLEL: all
@@ -20,4 +22,4 @@ lib_eigsolve.a: $(OBJS)
 	ar -cr lib_eigsolve.a $(OBJS)
 
 clean:
-	rm lib_eigsolve.a *.mod *.o
+	rm -f lib_eigsolve.a *.mod *.o