#!/bin/csh # # 13 Mar 06 - script to compile one GAMESS source file # # Invoke this script by './comp MODULE >& MODULE.log &'. # # ----- The next two lines require your attention ----- # 1. Choose your machine, the legal values for TARGET are # amd64,axp64,compaq-sc,cray-pvp,cray-t3e,cray-x1,cray-xd1,cray-xt3, # fuji-pp32,fuji-pp64,hpux32,hpux64,ibm32,ibm64,ibm64-sp,ibm-bg, # linux-ia64,linux-pc,mac32,macG5,necsx,sgi32,sgi64,sun32,sun64 # 2. Choose your GAMESS directory tree. # set TARGET=ibm64 chdir /u1/mike/gamess # # ---- and now the script begins... # unset echo set MODULE=$1 set ACTONLY=$2 # this argument is almost always omitted if (null$MODULE == null) then echo No compile module name given... exit endif # echo ======================== $MODULE ============================== date # # let's be sure they typed the intended target correctly # set OK=false if ($TARGET == amd64) set OK=true if ($TARGET == axp64) set OK=true if ($TARGET == compaq-sc) set OK=true if ($TARGET == cray-pvp) set OK=true if ($TARGET == cray-t3e) set OK=true if ($TARGET == cray-x1) set OK=true if ($TARGET == cray-xd1) set OK=true if ($TARGET == cray-xt3) set OK=true if ($TARGET == fuji-pp32) set OK=true if ($TARGET == fuji-pp64) set OK=true if ($TARGET == hpux32) set OK=true if ($TARGET == hpux64) set OK=true if ($TARGET == ibm32) set OK=true if ($TARGET == ibm64) set OK=true if ($TARGET == ibm64-sp) set OK=true if ($TARGET == ibm64-bg) set OK=true if ($TARGET == linux-ia64) set OK=true if ($TARGET == linux-pc) set OK=true if ($TARGET == mac32) set OK=true if ($TARGET == macG5) set OK=true if ($TARGET == necsx) set OK=true if ($TARGET == sgi32) set OK=true if ($TARGET == sgi64) set OK=true if ($TARGET == sun32) set OK=true if ($TARGET == sun64) set OK=true if ($OK == false) then echo The comp script does not select a correct TARGET machine type. echo What you typed when editing this script was $TARGET exit 4 endif # # ----- Is this a vector machine? ----- # set VECTORIZE=false if ($TARGET == cray-pvp) set VECTORIZE=true if ($TARGET == cray-x1) set VECTORIZE=true if ($TARGET == necsx) set VECTORIZE=true # # ----- Are level 3 BLAS in a system library? ----- # If so, some matrix multiplies in MTHLIB.SRC can use DGEMM calls # If not, print a warning to try to get people to find a BLAS library. # # N.B. Machines that use vector.src accomplish this business in a # different manner, so ignore this part. # set BLAS3=false if ($MODULE == mthlib) then # if ($TARGET == amd64) set BLAS3=true if (($TARGET == axp64) && \ ((-e /usr/lib/libdxml.a)||(-e /usr/lib/libcxml.a))) set BLAS3=true if ($TARGET == compaq-sc) set BLAS3=true if ($TARGET == cray-t3e) set BLAS3=true if ($TARGET == cray-x1) set BLAS3=true if (($TARGET == cray-xd1)||($TARGET == cray-xt3)) then if (-e $ACML_DIR) set BLAS3=true if (-e $PGI/linux86-64/$PGI_VERS_STR/lib/libblas.a) set BLAS3=true endif if ($TARGET == fuji-pp32) set BLAS3=true if ($TARGET == fuji-pp64) set BLAS3=true if ($TARGET == hpux32) set BLAS3=true if ($TARGET == hpux64) set BLAS3=true if ($TARGET == ibm32) set BLAS3=true if ($TARGET == ibm64) set BLAS3=true if ($TARGET == ibm64-sp) set BLAS3=true if ($TARGET == ibm-bg) set BLAS3=true # math lib not actually used! # 1st is Fedora Core's, 2nd is ASCI download from ORNL, 3rd is Atlas # the names of the latter two files on your system might well be # different from these, edit here and in 'lked' to match your paths. if (($TARGET == linux-pc) && (-e /usr/lib/libblas.a)) set BLAS3=true if (($TARGET == linux-pc) && (-e /usr/local/bin/libblas-asci.a)) set BLAS3=true if (($TARGET == linux-pc) && (-e /usr/local/lib/libatlas.a)) set BLAS3=true # The standard BLAS library for Itanium2 is Intel's MKL, # but both NEC and SGI have their own custom libraries. if ($TARGET == linux-ia64) then if (-e /opt/intel) then if(q`cd /opt/intel ; ls | grep mkl` =~ qmkl* ) set BLAS3=true endif if (-e /opt/MathKeisan/lib/libblas.a) set BLAS3=true if (-e /opt/sgi) then if(q`cd /opt/sgi ; ls | grep scsl` =~ qscsl* ) set BLAS3=true endif endif if ($TARGET == mac32) set BLAS3=true if ($TARGET == macG5) set BLAS3=true if ($TARGET == sgi32) set BLAS3=true if ($TARGET == sgi64) set BLAS3=true if (($TARGET == sun32) && (-e /opt/SUNWspro/lib/libsunperf.so)) set BLAS3=true if (($TARGET == sun64) && (-e /opt/SUNWspro/lib/libsunperf.so)) set BLAS3=true # if ($BLAS3 == true) then echo Compiling matrix multiply routines to use a system BLAS library... else echo "--------------------------------------------------------------" echo ' Warning\! Warning\! Warning\! Warning\!' echo "No mathematical library containing the BLAS has been detected." echo "Use of a DGEMM from a good library will speed up computations" echo "at the MP2, CCSD, or similar correlated levels, perhaps a lot." echo "In the meantime, mthlib.src and all other calls to DGEMM will" echo "rely on the slower blas.src provided with GAMESS." echo "--------------------------------------------------------------" endif endif # # Normally we are sitting above the source code and objects, # and want to make sure they go in their normal directory. # # N.B. A few places below prepare temporary copies of the # code being compiled in the current directory level (.), # and these are always deleted at the end of this script. # These places are vector, valence bond, and sed hacking. # set SRCDIR=source set OBJDIR=object # # a few source files are kept in non-standard places, defined here # if ($MODULE == blaswrap) set SRCDIR=misc if ($MODULE == ddi) set SRCDIR=ddi/oldsrc if ($MODULE == ddishm) set SRCDIR=ddi/shmem if ($MODULE == ddio3k) set SRCDIR=ddi/shmem # # ----- delete scalar code before compiling on vector systems ----- # UNIX's 'ld' loader is incapable of loading two copies of the # object code for a routine, and simply using the first one. # So, if we are using VECTOR.SRC on this machine, we must split # up some of the modules, delete the scalar code for the routines # which are found in VECTOR.SRC, and then recombine the rest of # the code in that module. Ugh! # Note that some of Cray's fsplit's are no longer able to work # with any extension except .f. # if ($VECTORIZE == true) then rm -f *.f if ($TARGET == cray-x1) alias fsplit ftnsplit switch ($MODULE) case int2a: echo Deleting scalar code forms, genral cp source/int2a.src xxxxxx.f fsplit xxxxxx.f rm -f xxxxxx.f if (-e forms.f) rm forms.f genral.f if (-e FORMS.f) rm FORMS.f GENRAL.f cat *.f > int2a.src rm -f *.f set SRCDIR=. breaksw case grd2c: echo Deleting scalar code dspdfv, jdxyzv, jkgnmv, jkxyzv cp source/grd2c.src xxxxxx.f fsplit xxxxxx.f rm -f xxxxxx.f if (-e dspdfv.f) rm -f dspdfv.f jdxyzv.f jkgnmv.f jkxyzv.f if (-e DSPDFV.f) rm -f DSPDFV.f JDXYZV.f JKGNMV.f JKXYZV.f cat *.f > grd2c.src rm -f *.f set SRCDIR=. breaksw case hss2b: echo Deleting scalar code jddspd.f cp source/hss2b.src xxxxxx.f fsplit xxxxxx.f rm -f xxxxxx.f if (-e jddspd.f) rm -f jddspd.f if (-e JDDSPD.f) rm -f JDDSPD.f cat *.f > hss2b.src rm -f *.f set SRCDIR=. breaksw case eigen: echo Deleting scalar code gldiag.f cp source/eigen.src xxxxxx.f fsplit xxxxxx.f rm -f xxxxxx.f if (-e gldiag.f) rm -f gldiag.f if (-e GLDIAG.f) rm -f GLDIAG.f cat *.f > eigen.src rm -f *.f set SRCDIR=. breaksw case mthlib: echo Deleting scalar code mtarbr, tfsqb, tftri. cp source/mthlib.src xxxxxx.f fsplit xxxxxx.f rm -f xxxxxx.f if (-e mtarbr.f) rm -f mtarbr.f tfsqb.f tftri.f if (-e MTARBR.f) rm -f MTARBR.f TFSQB.f TFTRI.f cat *.f > mthlib.src rm -f *.f set SRCDIR=. breaksw case scflib: echo Deleting scalar code dirfck. cp source/scflib.src xxxxxx.f fsplit xxxxxx.f rm -f xxxxxx.f if (-e dirfck.f) rm -f dirfck.f if (-e DIRFCK.f) rm -f DIRFCK.f cat *.f > scflib.src rm -f *.f set SRCDIR=. breaksw default: breaksw endsw endif # # optional SIMOMM method using Tinker MM program # Do not edit this part, even if you want to use Tinker. # == === ==== ==== ==== # Note that Tinker compiles down inside its own directory. set TINKER=false if ($MODULE == Libtad) set TINKER=true if ($MODULE == Libteac) set TINKER=true if ($MODULE == Libtedl) set TINKER=true if ($MODULE == Libtemo) set TINKER=true if ($MODULE == Libterx) set TINKER=true if ($MODULE == Libtfi) set TINKER=true if ($MODULE == Libtjo) set TINKER=true if ($MODULE == Libtpr) set TINKER=true if ($MODULE == Libtsx) set TINKER=true if ($MODULE == Tdrive) set TINKER=true if ($MODULE == Tinkin) set TINKER=true if ($MODULE == Toys) set TINKER=true # We must compile directly out of the Tinker directory # so that its ghastly INCLUDE statements will work. # Note that Tinker must skip the code deletion at end! if ($TINKER == true) then chdir tinker set SRCDIR=. set OBJDIR=../object goto cmp endif # # optional Valence Band method using VB2000 # You may need to set the variable VBPATH just below. set VB2000=false if ($MODULE == vb2000) set VB2000=true if ($MODULE == vb2gms) set VB2000=true # compiling in the top level directory means code is erased later if ($VB2000 == true ) then set VBPATH=~/VB/SRC echo Copying $MODULE.src from VB2000 directory named $VBPATH set SRCDIR=. if ($MODULE == vb2000) then # Some messy sed stuff to set up. sed "s/^CTMAP/ /g" $VBPATH/vb2000.src | \ sed "s/^CGMS/ /g" > vb2000.src endif if ($MODULE == vb2gms) cp $VBPATH/vb2gms.src . endif # # ----- Does this module need to be activated? ----- # # Machines where "DOUBLE PRECISION" translates to more than a 64 # bit floating point quantity must be passed into the activation # to ensure the source is made single precision. # if ($TARGET == cray-pvp) goto act if ($TARGET == cray-t3e) goto act if ($TARGET == cray-x1) goto act if ($TARGET == necsx) goto act # # Machines needing the 64->32 bit BLAS wrappers must activate everything # This depends on whether or not the BLAS library is OK with 64 bit # integers in its argument, or not (in which case a sed hack to use # a 64 to 32 bit wrapper file is needed, by always branching here). # if ($TARGET == fuji-pp64) goto act if ($TARGET == ibm64) goto act if ($TARGET == ibm64-sp) goto act if ($TARGET == sgi64) goto act if ($TARGET == sun64) goto act # # Any module with machine dependent code must be activated... # if ($MODULE == aldeci) goto act if ($MODULE == ccsdt) goto act if ($MODULE == cisgrd) goto act if ($MODULE == cphf) goto act if ($MODULE == cpmchf) goto act if ($MODULE == cprohf) goto act if ($MODULE == ddi) goto act if ($MODULE == ddishm) goto act if ($MODULE == dft) goto act if ($MODULE == dftint) goto act if ($MODULE == grd2a) goto act if ($MODULE == gugdga) goto act if ($MODULE == gugdgb) goto act if ($MODULE == gugdm2) goto act if ($MODULE == gugem) goto act if ($MODULE == gugsrt) goto act if ($MODULE == gvb) goto act if ($MODULE == gvvpt2) goto act if ($MODULE == hss2a) goto act if ($MODULE == int2a) goto act if ($MODULE == iolib) goto act if ($MODULE == lagran) goto act if ($MODULE == local) goto act if ($MODULE == locpol) goto act if ($MODULE == mccas) goto act if ($MODULE == mcqdpt) goto act if ($MODULE == mcqud) goto act if ($MODULE == mcscf) goto act if ($MODULE == mctwo) goto act if ($MODULE == morokm) goto act if ($MODULE == mp2) goto act if ($MODULE == mp2grd) goto act if ($MODULE == nmr) goto act if ($MODULE == ordint) goto act if ($MODULE == rhfuhf) goto act if ($MODULE == tdhf) goto act if ($MODULE == trans) goto act if ($MODULE == trfdm2) goto act if ($MODULE == unport) goto act if ($MODULE == vb2gms) goto act if ($MODULE == vector) goto act # # We may need to do a hack below on the matrix multiply routines # if (($BLAS3 == true) && ($MODULE == mthlib)) goto act # # All other modules are straight FORTRAN, just copy them. # echo Copying source code, $MODULE.src does not require activation. cp $SRCDIR/$MODULE.src $MODULE.f goto cmp # # ----- Choose which version is to be activated ----- # The setting for 'ddi' applies to the possible use of the first version # of the Distributed Data Interface, and probably should be sockets only. # Most machines use the 2nd version of DDI, implemented entirely in C, # and compiled by the separate 'compddi' script. # In contrast, machines using the SHMEM library will compile ddishm.src # act: # if ($TARGET == amd64) then setenv MACHIN '*I64' if ($MODULE == ddi) setenv MACHIN '*SOC' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*AMD' endif # if ($TARGET == axp64) then setenv MACHIN '*I64' if ($MODULE == ddi) setenv MACHIN '*SOC' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*DEC' endif # SHMEM on Compaq has to activate its particular type if ($TARGET == compaq-sc) then setenv MACHIN '*I64' if ($MODULE == ddishm) setenv MACHIN '*DEC' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*DEC' endif # the old Cray vector systems would use TCP/IP sockets in DDI if ($TARGET == cray-pvp) then setenv MACHIN '*I64' if ($MODULE == ddi) setenv MACHIN '*SOC' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*CRY' if ($MODULE == vector) setenv MACHIN '*CRY' endif # the CRAY T3E has a special SHMEM implementation of DDI if ($TARGET == cray-t3e) then setenv MACHIN '*I64' if ($MODULE == ddishm) setenv MACHIN '*T3E' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*T3E' endif # the CRAY X1 has a special SHMEM implementation of DDI if ($TARGET == cray-x1) then setenv MACHIN '*I64' if ($MODULE == ddishm) setenv MACHIN '*CX1' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*CX1' if ($MODULE == vector) setenv MACHIN '*CRY' endif # the CRAY XD1 uses TCP/IP sockets, until GPSHMEM is working if ($TARGET == cray-xd1) then setenv MACHIN '*I64' if ($MODULE == ddishm) setenv MACHIN '*XD1' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*AMD' endif # the CRAY XT3 is using pure MPI, from the old version of DDI if ($TARGET == cray-xt3) then setenv MACHIN '*I64' if ($MODULE == ddi) setenv MACHIN '*MPI' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*XT3' endif if (($TARGET == fuji-pp32) || ($TARGET == fuji-pp64)) then setenv MACHIN '*I32' if ($TARGET == fuji-pp64) setenv MACHIN '*I64' if ($MODULE == ddi) setenv MACHIN '*SOC' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*SUN' endif if (($TARGET == hpux32) || ($TARGET == hpux64)) then setenv MACHIN '*I32' if ($TARGET == hpux64) setenv MACHIN '*I64' if ($MODULE == ddi) setenv MACHIN '*SOC' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*HP ' endif if (($TARGET == ibm32) || ($TARGET == ibm64) || ($TARGET == ibm64-sp)) then setenv MACHIN '*I64' if ($TARGET == ibm32) setenv MACHIN '*I32' if ($MODULE == ddi) setenv MACHIN '*SOC' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*AIX' endif # Blue Gene uses the old DDI toolkit, in pure MPI mode. if ($TARGET == ibm-bg) then setenv MACHIN '*I32' if ($MODULE == ddi) setenv MACHIN '*MPI' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*AIX' endif if ($TARGET == linux-ia64) then setenv MACHIN '*I64' if ($MODULE == ddi) setenv MACHIN '*SOC' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*ITA' endif if ($TARGET == linux-pc) then setenv MACHIN '*I32' if ($MODULE == ddi) setenv MACHIN '*SOC' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*F2C' endif if (($TARGET == mac32) || ($TARGET == macG5)) then setenv MACHIN '*I32' if ($MODULE == ddi) setenv MACHIN '*SOC' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*F2C' endif if ($TARGET == necsx) then setenv MACHIN '*I64' if ($MODULE == ddi) setenv MACHIN '*MPI' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*NEC' if ($MODULE == vector) setenv MACHIN '*NEC' endif if (($TARGET == sgi32) || ($TARGET == sgi64)) then setenv MACHIN '*I32' if ($TARGET == sgi64) setenv MACHIN '*I64' if ($MODULE == ddi) setenv MACHIN '*SOC' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*SGI' endif if (($TARGET == sun32) || ($TARGET == sun64)) then setenv MACHIN '*I32' if ($TARGET == sun64) setenv MACHIN '*I64' if ($MODULE == ddi) setenv MACHIN '*SOC' if ($MODULE == iolib) setenv MACHIN '*UNX' if ($MODULE == unport) setenv MACHIN '*SUN' endif # # ----- Activate the source code to produce FORTRAN ----- # set echo setenv SRCIN $SRCDIR/$MODULE.src setenv CODEOUT $MODULE.f # tools/actvte.x unset echo # # --------------- special machine "sed hacks" -------------------- # Not everything can be done with ACTVTE, so we use "sed" (ugh!). # # Direct access record length is measured in bytes on most Unix, # but some kinds insist in measuring 4-byte words. This changes # only one line in RAOPEN and one line in OPENDA in IOLIB.SRC. # SGI's f90 should not execute the following, but SGI's f77 must. # if (($MODULE == iolib) && \ ($TARGET == axp64 || $TARGET == compaq-sc || \ $TARGET == sgi32 || $TARGET == sgi64)) then mv iolib.f iolib.junk sed -e "s/ \* RECL=8\*IRECLN/ * RECL=2*IRECLN/" \ iolib.junk > iolib.f rm -f iolib.junk endif # # Use DGEMM matrix multiply if possible. # if (($MODULE == mthlib) && ($BLAS3 == true)) then mv mthlib.f mthlib.junk sed -e "s/\*BL3/ /" mthlib.junk > mthlib.f rm -f mthlib.junk endif # # For 64 bit machines which also have a traditional 32 bit product # line, the source code typically assumes 32 bits. This hack lets # the source code know it is going to be running as a 64 bit machine. # if (($MODULE == unport) && \ (($TARGET == fuji-pp64) || \ ($TARGET == hpux64) || \ ($TARGET == ibm64) || \ ($TARGET == ibm64-sp) || \ ($TARGET == sgi64) || \ ($TARGET == sun64))) then mv unport.f unport.junk sed -e "s/ NWDVAR = 2/ NWDVAR = 1/" unport.junk > unport.f rm -f unport.junk endif # # This hack is necessary only if compiler flags below request the # use of 64 bit integers in GAMESS, and if the vendor's BLAS library # insists that its integer arguments be 32 bit quantities. # There is a special wrapper file gamess/misc/blaswrap.src that deals # with this problem, provided we hack the source to call the wrappers. # Not a perfect strategy, imagine an unrelated routine named MDSCAL, # but there may not be spaces in front of all of these, especially DDOT. # if (($MODULE != blaswrap) && ($MODULE != blas) && \ (($TARGET == fuji-pp64) || ($TARGET == ibm64) || ($TARGET == ibm64-sp) || \ ($TARGET == sgi64) || ($TARGET == sun64))) then sed -e "s/DASUM/XASUM/g" -e "s/dasum/XASUM/g" \ -e "s/DAXPY/XAXPY/g" -e "s/daxpy/XAXPY/g" \ -e "s/DCOPY/XCOPY/g" -e "s/dcopy/XCOPY/g" \ -e "s/DDOT/XDOT/g" -e "s/ddot/XDOT/g" \ -e "s/DNRM2/XNRM2/g" -e "s/dnrm2/XNRM2/g" \ -e "s/DROT/XROT/g" -e "s/drot/XROT/g" \ -e "s/DROTG/XROTG/g" -e "s/drotg/XROTG/g" \ -e "s/DSCAL/XSCAL/g" -e "s/dscal/XSCAL/g" \ -e "s/DSWAP/XSWAP/g" -e "s/dswap/XSWAP/g" \ -e "s/IDAMAX/IXAMAX/g" -e "s/idamax/IXAMAX/g" \ -e "s/DGER/XGER/g" -e "s/dger/XGER/g" \ -e "s/DTRMM/XTRMM/g" -e "s/dtrmm/XTRMM/g" \ -e "s/DTRMV/XTRMV/g" -e "s/dtrmv/XTRMV/g" \ -e "s/DTRSM/XTRSM/g" -e "s/dtrsm/XTRSM/g" \ -e "s/DGEMV/XGEMV/g" -e "s/dgemv/XGEMV/g" \ -e "s/DGEMM/XGEMM/g" -e "s/dgemm/XGEMM/g" \ $MODULE.f > $MODULE.munged mv $MODULE.munged $MODULE.f endif # # The following hack makes an Apple version print a different banner. # The Apple version is source code identical to Linux, but it should # print something to keep it from looking like a PC! # if (($TARGET == mac32) || ($TARGET == macG5)) then if ($MODULE == unport) then mv unport.f unport.junk sed -e "s/PC-UNIX VERSION/ APPLE VERSION /" unport.junk > unport.f rm -f unport.junk endif endif # # The following hack selects AOINTS=DIST as the default on high end # Compaq and IBM systems, where we know we have a high quality network. # if (($MODULE == unport) && \ (($TARGET == compaq-sc) || ($TARGET == ibm64-sp))) then mv unport.f unport.junk sed -e "s/\*TRF/ /" unport.junk > unport.f rm -f unport.junk endif # # HP has a FORTRAN callable delay, we just have to change to HP's name # if (($TARGET == hpux32) || ($TARGET == hpux64)) then if ($MODULE == unport) then mv unport.f unport.junk sed -e "s/NAPTIME(IDELAY)/SLEEP(IDELAY)/" \ unport.junk > unport.f rm -f unport.junk endif endif # # ...and this is the end of the ugly "sed hack"s. # # --------- Now we are ready to compile on the target machine ---------- # cmp: # # Optionally, save pure FORTRAN in temporary directory for # syntax scan by the nifty FORTRAN analysis program FTNCHEK. # if ($ACTONLY == true) then mv $MODULE.f ~/scr exit endif # if (-e $OBJDIR/$MODULE.o) rm -f $OBJDIR/$MODULE.o # # Version for Opteron and other 64 bit Athlon chips made by AMD. # See 'linux-pc' if you have 32 bit Athlon chips made by AMD. # Typical software requirement: # Fedora Core 2/64 bit version, pgf77, pgcc, and ACML math library. # # The amd64 target was constructed by putting together advice from # Shiro Koseki, Hiroaki Umeda, and Ted Packwood. This version relies # upon commercially marketed compilers, and a freely available BLAS # library. This means that there is no "totally free" version based # on (for example) a g77 compiler. Please note that you should just # select a target of 'amd64' and ignore PGI's web page information # about porting (everything needed is now included in these scripts). # # The compiler is marketed by the Portland Group, see www.pgroup.com. # What GAMESS requires is the pgf77 and pgcc packages, not the pgf90 # products. A pgf77/pgcc bundle is priced at $499 for academic users, # as of November 2004. Be sure to set up the use of this compiler in # your .cshrc or other login script, for example # setenv PATH $PATH\:/usr/pgi/linux86-64/bin # setenv PGI /usr/pgi # # The BLAS library is the AMD Core Math Library, developed by AMD and # the NAG group. ACML is available most easily as an option while # installing pgf77, but it is also a free download, see # http://www.amd.com/us-en/Processors/DevelopWithAMD/0,,30_2252,00.html # so the amd64 version assumes that you have obtained this library. # if ($TARGET == amd64) then if (($MODULE == qeigen) || ($MODULE == int2c)) then mv $MODULE.f $MODULE.junk sed -e "s/Q-/D-/g" \ -e "s/Q+00/D+00/g" \ -e "s/REAL\*16/DOUBLE PRECISION/" $MODULE.junk > $MODULE.f rm -f $MODULE.junk endif set OPT = '-fastsse -Mipa=fast,safe -Mprof=func' if ($MODULE == int2b) set OPT = "$OPT -Mnounroll" if (($MODULE == mpcdat) || ($MODULE == mpcgrd) || ($MODULE == mpcint) || \ ($MODULE == mpcmol) || ($MODULE == mpcmsc)) set OPT = '-Mprof=func -O0' if ($MODULE == zheev) set OPT='-Mprof=func -O0' # defensive compiling set echo pgf77 -c -i8 -i8storage $OPT $MODULE.f unset echo endif # # Options for the Compaq AXP FORTRAN 5.2 compiler under Tru64 V4.0F are # -O0 turns off all optimization # -O1 does local optimizations # -O2 also does global optimizations # -O3 also does additional global optimizations # -O4 also does routine inlining (default) # -O5 also does software pipelining # -fast implies less accurate but faster libraries, etc. and -O4. # -c means compile only # -i8 selects use of 64 bit integers # -v gives compiler phase information, and timings # -automatic and -static use/avoid stack for local variables # -col72 chops source lines past 72 columns # -stand nosemantic means don't warn of nonANSI usage (default) # -stand nosource_form means don't warn of tabs/lowercase (default) # -stand nosyntax means don't warn of nonANSI syntax (default) # # Note that we've tried f77 3.3, 3.7, 3.8, and 5.2, but nothing in between. # This is an excellent compiler, and probably all versions work. # Under AXP Linux, it has a different name, but is the same great compiler. # if ($TARGET == axp64) then set OPT = '-O4' set FORT='f77' if (`uname` == Linux) set FORT='fort' set echo $FORT -c -i8 -v -automatic -col72 $OPT $MODULE.f unset echo endif # The SuperCluster assumes a specific AXP chip... if ($TARGET == compaq-sc) then set OPT = '-fast -tune ev67 -arch ev67' if ($MODULE == zheev) set OPT='-O0' set echo f77 -c -i8 -v -col72 $OPT $MODULE.f unset echo endif # # Cray parallel vector processor (e.g. T90, J90, SV1...) # Cray T90 tested in February of 2000 # Cray SV1 rev C with CF90 3.4.0.0.4 tested in March of 2000 # Note that all Cray clauses now use f90 compiler only. # if ($TARGET == cray-pvp) then set OPT = '-O2 ' if ($MODULE == int2a) set OPT='-Ovector0,scalar3 ' if ($MODULE == grd2a) set OPT='-Ovector0,scalar2 ' # Jerry says if ($MODULE == grd2b) set OPT='-Ovector0,scalar3 ' if ($MODULE == grd2c) set OPT='-Ovector0,scalar3 ' if ($MODULE == gugem) set OPT='-Ovector0,scalar3 ' if ($MODULE == vector) set OPT='-O1 ' if ($MODULE == gvb) set OPT='-O1 ' # use -O0 if exam15,16,21 fail if ($MODULE == gamess) set OPT='-O0 ' if ($MODULE == efinp) set OPT='-O0 ' if ($MODULE == efinta) set OPT='-Ovector0,scalar3 ' if ($MODULE == efintb) set OPT='-Ovector0,scalar3 ' if ($MODULE == efgrda) set OPT='-Ovector0,scalar3 ' if ($MODULE == efgrdb) set OPT='-Ovector0,scalar3 ' if ($MODULE == efgrdc) set OPT='-Ovector0,scalar3 ' if ($MODULE == efgrd2) set OPT='-Ovector0,scalar3 ' if ($MODULE == ormas1) set OPT='-O0 ' # also from Jerry set echo f90 -c -V $OPT $MODULE.f unset echo endif # # Compiler flags for Cray T3E under UNICOS/mk 2.0.3.18, cf90 3.1.0.3 # -Oscalarn is various scalar optimizations (n=0,1,2,3) # -Oaggress increases table sizes used during optimizations # -O[no]msgs,[no]negmsgs [disables]enables compiler optimization messages # -dp disables double precision # if ($TARGET == cray-t3e) then set OPT = '-Oscalar3,aggress' if ($MODULE == grd2c) set OPT='-O1' # for cf90 3.1.0.3 and others if ($MODULE == hss1b) set OPT='-O1' # for cf90 3.4.0.0 set echo env TARGET=cray-t3e f90 -c $OPT -dp -Onomsgs,nonegmsgs $MODULE.f unset echo endif # # Compiler flags for Cray X1 under UNICOS/mp 2.5+, ftn 5.3+ # -O[no]msgs,[no]negmsgs [disables]enables compiler optimization messages. # -Oscalarn is various scalar optimizations (n=0,1,2,3) # -Ovectorn is various vector optimizations (n=0,1,2,3) # -Oaggress causes the compiler to treat each subroutine as a single # optimization region. # -Ossp will build GAMESS in SSP (single-streaming processor) mode. # remove -Ossp to use multi-streaming processor mode (discouraged) # -dp disables double precision. # -rm generates listing files. # -xomp disables all OpenMP statements. # -sdefault64 uses default 64-bit integers and reals (among others). # if ($TARGET == cray-x1) then set OPT = '-O2' set LST = ' ' # change to '-rm -Omsgs,negmsgs' for listing file if ($MODULE == int2a) set OPT='-O1' if ($MODULE == vector) set OPT='-Oscalar3,vector3' set echo ftn -c -V $OPT $LST -Ossp,aggress -xomp -dp -sdefault64 $MODULE.f unset echo endif # # Compiler flags for Cray XD1/XT3 using PGI F90 6.0.* and MPICH 1.2.6. # For the XD1, make sure your "mpif90" script points to pgf90 and not a # different compiler. Note: F90 is required due to ddi/shmem/ddigpshm.src. # 'setenv PGI' should not be needed as this should be done automatically # by modules. If you do not have $PGI environment variable set, # uncomment these 2 lines (first change /usr/pgi to wherever the # PGI compiler is installed): # setenv PGI /usr/pgi # setenv PATH $PATH\:$PGI/linux86-64/bin # if (($TARGET == cray-xd1)||($TARGET == cray-xt3)) then if ($TARGET == cray-xt3) then set USE_FTN = ftn else # if ($TARGET == cray-xd1) then set USE_FTN = 'mpif90' endif if (($MODULE == qeigen) || ($MODULE == int2c)) then mv $MODULE.f $MODULE.junk sed -e "s/Q-/D-/g" \ -e "s/Q+00/D+00/g" \ -e "s/REAL\*16/DOUBLE PRECISION/" $MODULE.junk > $MODULE.f rm -f $MODULE.junk endif set echo set OPT = '-fastsse -Munroll' if ($MODULE == int2b) set OPT = '-fast -Mnounroll' if (($MODULE == mpcdat) || ($MODULE == mpcgrd) || ($MODULE == mpcint) || \ ($MODULE == mpcmol) || ($MODULE == mpcmsc)) set OPT = '-O0' set echo $USE_FTN -c $OPT -Mnosecond_underscore -tp amd64 -i8 -Minfo=all $MODULE.f unset echo endif # # Fujitsu PrimePower # Split into 32bit and 64bit versions, since the 32bit executable is # slightly (~5%) faster but cannot use larger than 2Gbyte memory. # From Roger Amos at ANU Supercomputer Facility, 1/2005 # # Compilation options: do not use -O5 or FMADD as code is slower. # More complicated cache optimisation does not improve performance. # # 64bit version tested with largepage=1, and largepage=2, the # program executes correctly but there is no performance advantage. # # tested with Solaris8 and Solaris9 and compiler # Fujitsu Fortran 5.5 (Feb 19 2004 17:23:48) # note: the compiler has two names, f90 and frt, which are the same. # if ($TARGET == fuji-pp32) then set OPT = '-Kfast_GP2=3,prefetch=5' set echo f90 -c -fs $OPT $MODULE.f unset echo endif # -KV9 allows large memory space # -CcdII8 causes all generic integer variables to be 64 bit # if ($TARGET == fuji-pp64) then set OPT = '-Kfast_GP2=3,prefetch=5,largepage=2' if ($MODULE == gamess) set OPT='-O0' # only fails in 64 bit mode set echo f90 -c -fs $OPT -KV9 -CcdII8 $MODULE.f unset echo endif # # HP-UX systems, # the latest tested version numbers are # HP-UX B.11.0, using f90 2.4.10 on PA-RISC 32 bit # HP-UX B.11.23, using f90 2.8 on Itanium2 64 bit # (f90 +version reports version number) # # The f90 options apply to both 32 and 64 bit compilations, and mean: # +On selects optimization level; n=0,1,2,3,4 (-O means +O2) # 0 is none # 1 is block transforms # 2 is full within subprograms # 3 crosses subprograms (includes inlining) also loop transforms # 4 is link time iptimizations, applies only to Itanium processors. # # +Oaggressive selects the following four additions to +O2: # +Oentrysched = instruction scheduling on entry/exit # +Olibcalls = special trig and power functions (lacking error hand.) # +Onofltacc = permits use of a faster single multiply/add instruction. # +Onoinitcheck = disable initialization of any unitialized datum # other optimizations include # +Onolimit = ignores time and memory limits when optimizing # +Ofastaccess = gives fast access to global data # # +Oaggressive is a decremented optimization, to be replaced in newer # systems by +Ofast. +Ofast selects the combination +O2, +Olibcalls, # +Onolimit, +Ofltacc=relaxed, +FPD, +DSnative (on IPF), +Oshortdata # and requires f90 2.5 or higher. # # +[no]save governs automatic variable allocation strategy # +[no]ppu governs postpend underscores on names of external refs. # +U77 gives access to BSD's 3F library # +DD64 uses the LP64 data model. # +i8 treats all integer or logical data as 64 bits # if ($TARGET == hpux32) then set OPT='+O2 +Oaggressive +Onolimit +Ofastaccess' if ($MODULE == unport) set OPT="$OPT +U77" set echo f90 -c $OPT +nosave +noppu $MODULE.f unset echo endif # if ($TARGET == hpux64) then # # the 64 bit libraries have a name conflict with ICOPY, XERBLA, and LSAME. # if ($MODULE == guess) then mv guess.f guess.junk sed -e "s/SUBROUTINE ICOPY/SUBROUTINE IYPOC/" \ guess.junk > guess.f rm -f guess.junk endif if ($MODULE == zheev) then mv zheev.f zheev.junk sed -e "s/SUBROUTINE XERBLA/SUBROUTINE XALBRE/" \ -e "s/FUNCTION LSAME/FUNCTION LEMAS/" \ -e "s/LSAME = CA.EQ.CB/LEMAS = CA.EQ.CB/" \ -e "s/IF( LSAME )/IF( LEMAS )/" \ -e "s/LSAME = INTA.EQ.INTB/LEMAS = INTA.EQ.INTB/" \ zheev.junk > zheev.f rm -f zheev.junk endif # +O3 failed to run 9 of 37 example jobs correctly, so +O2 is safer. set OPT='+O2' if ($MODULE == unport) set OPT="$OPT +U77" if ($MODULE == trfdm2) set OPT="+O1" # for CI gradients, exam05 set echo f90 -c $OPT +DD64 +i8 +nosave +noppu $MODULE.f unset echo endif # # For IBM 32 or 64 bit RS/6000 workstations or SP parallel systems, # the options for XL FORTRAN 7.1.0 and AIX 4.3.3 are # -c means compile only # -O2 and -O3 choose optimization levels (there is no -O1) # -qarch=com,601,pwr,pwr2,pwr3 (see also -qtune) # -qflag requests I,L,W,E,S,Q (information, language, # warning, error, severe, quiet) messages # -qflttrap=ov:zero:inv:inex:enable traps bad arithmetic # -qhalt stops compiler after certain error level # -qnosave forces non-static storage # -qinitauto=ff initializes non-static storage (useful if debugging) # -qextchk checks calling args and common lengths at link # -qsource requests a printable .lst file. # -qtune selects target chip # # The following clause requires XL FORTRAN 3.2 or higher. # Use of 64 bit integers requires XL FORTRAN 6.1 or higher. # The command to find your FORTRAN version level depends on your # AIX version: v3 is "lslpp -h xlfcmp.obj", v4/v5="lslpp -h xlfcmp". # # To look for XLF optimization problems in one routine, place # @PROCESS OPT(0) SAVE # above the suspect routine. Note that the directive remains in # force for only one subprogram, and must be above its first line. # # To profile code during development stages, # 1. "comp" desired modules with -pg added to the xlf flags # 2. "lked" after adding -p to the xlf flags. # 3. execute to completion, generating 'mon.out' file. # 4. generate results by 'prof gamess.00.x mon.out'. # # A possible way to look for memory allocation problems is VALGRIND. # Compile with the -p flag, and then be sure to look at child's # memory use: "valgrind --trace-children=yes rungms". This program # is available for Linux (not AIX) at valgrind.org. Untried by MWS. # if (($TARGET == ibm32) || ($TARGET == ibm64) || ($TARGET == ibm64-sp)) then set FORT='xlf' if ($TARGET == ibm64-sp) set FORT='mpxlf_r' # old 32 bit workstations, whether Power, Power2, or PowerPC # should all select plain vanilla common architecture. It is # faster, and runs on any combination of these chips you own. # under xlf 4.1.0 a few files won't fully optimize if ($TARGET == ibm32) then set OPT='-O3' if ($MODULE == grd2b) set OPT='-O3 -qspillsize=1500' if ($MODULE == gugdm) set OPT='-O2' if ($MODULE == gugem) set OPT='-O2' if ($MODULE == qrel) set OPT='-O2' if ($MODULE == trnstn) set OPT='-O2' set ARCH='com' set TUNE='604' # PowerPC tuning makes even Power1 run faster endif # 64 bit workstations or SP is sure to be Power3 or higher chip. # everything optimizes under xlf 5.1.1 and 6.1.0 and 7.1.0 # our AIX systems are a mix of Power3 and Power4, while # our Linux systems are all Power5. Please change the # architecture/tuning to match your case! if (($TARGET == ibm64) || ($TARGET == ibm64-sp)) then set OPT='-O3 -q64 -qintsize=8 -qspillsize=1500' if (`uname` == AIX) then set ARCH='pwr3' set TUNE='pwr3' else set ARCH='pwr5' set TUNE='pwr5' endif endif if ($MODULE == fmoint) set OPT="$OPT -qstrict" set echo $FORT -c $OPT -qarch=$ARCH -qtune=$TUNE \ -qflag=W:W -qhalt=W -qnosave $MODULE.f unset echo endif # # Blue Gene is a 32 bit processor based on the PPC440 embedded CPU # with the addition of a dual FPU. # Blue Gene uses MPI in our older DDI toolkit for parallelization. # Blue Gene is rather specialized, let's not mix with other IBM, # even though the xlf arguments are clearly rather similar. # Not sure why compiler path is hardcoded... # if ($TARGET == ibm-bg) then set FORT="/opt/ibmcmp/xlf/9.1/bin/blrts_xlf" set INCL="-I/bgl/BlueLight/ppcfloor/bglsys/include" set ARCH='440' set TUNE='440' set OPT='-O3 -qmaxmem=-1' if ($MODULE == grd2b) set OPT='-O3 -qspillsize=1500' if ($MODULE == gugdm) set OPT='-O2' if ($MODULE == gugem) set OPT='-O2' if ($MODULE == qrel) set OPT='-O2' if ($MODULE == trnstn) set OPT='-O2' if ($MODULE == ormas1) set OPT='-O2' if ($MODULE == zheev) set OPT='-O2' if ($MODULE == qeigen) set OPT="$OPT -qstrict" if ($MODULE == fmoint) set OPT="$OPT -qstrict" if ($MODULE == ddi) set OPT="$OPT $INCL" set echo $FORT -c $OPT -qarch=$ARCH -qtune=$TUNE \ -qflag=W:W -qhalt=W -qnosave $MODULE.f unset echo endif # # ------ RedHat Linux on "PC" (Pentium/Athlon chips) ----- # There are no less than four compilers for Linux included here! # They are g77, Intel's ifort, Portland's pgf77, and the obsolete # combination of f2c/gcc. The default compiler is g77, but many # people use the Intel compiler. It is a little faster, and it # supports quadruple precision, compared to the default of g77. # # If you want to use one of the other three, just make sure that the # one you prefer is uncommented. In addition, you will need to edit # the link-edit script 'lked' if you pick a different compiler here, # and possibly you will ned to change 'compddi' as well. # # Note that we have only used Redhat, so this is not necessarily # correct for Slackware, SUSE, FreeBSD, Debian, ..., although # many people have been able to get these others to work. # # 1. Using g77 compiler. compiler version query: rpm -q gcc-g77 # If your RedHat is older than 6.0, see the f2c/gcc clause below. # # There is an I/O library problem in g77, found in RedHat 6.0 -> 7.1, # which is not present in any of the 3.x GNU compilers. Thus, if # your RedHat is newer than 7.1, you can forget this, but if not, # you can find a fix for the I/O bug on our web page: # www.msg.ameslab.gov/GAMESS/GAMESS.html, # then check the bottom of the "how to get a copy of GAMESS" page. # It is crucial to install this, as it works around a very serious # error in the g77 2.x I/O library. # This bug does not exist in any of the 3.x GNU compiler releases. # # The meaning of g77's flags are # -O0, -O1, -O2, -O3 are the optimization levels, -O means -O1 # -Wno-globals -fno-globals together suppress argument type checking. # -fautomatic defeats use of static storage # -malign-double uses 64 bit data alignment # Tests on a representative GAMESS test suite (executing energy, gradient # hessian code, as well as various SCF, MCSCF, MP2 runs, show that each # of the more aggressive optimizations -ffast-math, -march=i586, and -O3 # gave speedups of less than 1% each. (-march=i686 slows a Celeron type # Pentium II fractionally). None of these optimizations are used below. # if ($TARGET == linux-pc) then set OPT = '-O2' if ($MODULE == ormas1) set OPT='-O2 -fno-move-all-movables' # RH9, g77 3.3 if ($MODULE == zheev) set OPT='-O0' # defensive compiling, for exam19 # # g77 does not support FORTRAN logical units above 99, or 128 bit math. # if (($MODULE == qeigen) || ($MODULE == int2c)) then mv $MODULE.f $MODULE.junk sed -e "s/Q-/D-/g" \ -e "s/Q+00/D+00/g" \ -e "s/REAL\*16/DOUBLE PRECISION/" $MODULE.junk > $MODULE.f rm -f $MODULE.junk endif if (($MODULE == iolib) || ($MODULE == inputa)) then mv $MODULE.f $MODULE.junk sed -e "s/MXUNIT=299/MXUNIT=99/g" $MODULE.junk > $MODULE.f rm -f $MODULE.junk endif set echo g77 -c $OPT -malign-double -fautomatic \ -Wno-globals -fno-globals $MODULE.f unset echo endif # # 2. Intel Fortran Compiler for Linux # This compiler can be downloaded for free for non-commercial purposes, # http://developer.intel.com/software/products/compilers/linux # and for-profit users can purchase this software for a reasonable price. # You need to get FORTRAN, C, and the MKL math library. # # Remember to set up using the compiler by executing the proper # setup script for your shell, e.g. # . /opt/intel/compiler50/ia32/bin/ifcvars.sh # # You must compile DDI to have only one trailing underscore in the # DDI routine names, namely edit 'compddi' to say -D_UNDERSCORES=1 # in its linux-pc definitions, rather than the 2 that g77 requires. # Otherwise 'lked' will get unresolved references to DDI_BCAST, etc. # # Prior to version 8.0, this compiler is spelled "ifc" rather than "ifort". # -O3 is the most aggressive optimization, the default is -O2. # -cm turn off printing of informational messages (f90 syntax, etc) # -w turn off all warnings # Dmitri likes to use "-O3 -xN -ssp -unroll8 -ip -pad -opt_report", # but that may cause certain ifort versions to fail on some files. # #--if ($TARGET == linux-pc) then #-- set OPT = '-O3' #-- if ($MODULE == delocl) set OPT='-O0' # from Cheol #-- if ($MODULE == zheev) set OPT='-O0' #-- set echo #-- ifort -c $OPT -cm -w $MODULE.f #-- unset echo #--endif # # 3. Portland Group compiler. # This is a synthesis of recommendations from Fred Arnold and Brian # Salter-Duke. -fast means -O2 -Munroll -Mnoframe, the latter option # seems to be the source of several incorrect results. It is said that # "-Munroll" and "-tp p6" gives no significant improvement in run times. # The -Mlfs flag is supposed to cure the 2 GB file size limit. # exam19 may hang when using this compiler. -Mlfs at link-time # enables large file support. Your login script must set # up the use of this compiler, along the lines of # setenv PATH $PATH\:/usr/pgi/linux86/bin # setenv PGI /usr/pgi # #--if ($TARGET == linux-pc) then #-- if (($MODULE == qfmm) || ($MODULE == solib) || ($MODULE == zheev)) then #-- mv $MODULE.f $MODULE.junk #-- sed -e s/DREAL/DBLE/g $MODULE.junk > $MODULE.f #-- rm -f $MODULE.junk #-- endif #-- if (($MODULE == qeigen) || ($MODULE == int2c)) then #-- mv $MODULE.f $MODULE.junk #-- sed -e "s/Q-/D-/g" \ #-- -e "s/Q+00/D+00/g" \ #-- -e "s/REAL\*16/DOUBLE PRECISION/" $MODULE.junk > $MODULE.f #-- rm -f $MODULE.junk #-- endif #-- set OPT = '-O2' #-- if ($MODULE == zheev) set OPT = '-O2 -Kieee' #-- if ($MODULE == bassto) set OPT = '-O0' # Takako #-- set echo #-- pgf77 -c $OPT -Msecond_underscore $MODULE.f #-- unset echo #--endif # # 4. RedHat 4.x and 5.x's FORTRAN compiler was f2c/gcc, and early # releases of g77 that were present in some of these distributions # do not run GAMESS correctly. If you have RedHat 6.0 or higher, # which do not contain the f2c interpreter, see just above. If your # RedHat is older than 6.0, use f2c/gcc instead of buggy old g77's. # This is a totally obsolete option, actually. # #--if ($TARGET == linux-pc) then #-- if (($MODULE == qfmm) || ($MODULE == solib) || ($MODULE == zheev)) then #-- mv $MODULE.f $MODULE.junk #-- sed -e s/DREAL/DBLE/g $MODULE.junk > $MODULE.f #-- rm -f $MODULE.junk #-- endif #-- if (($MODULE == qeigen) || ($MODULE == int2c)) then #-- mv $MODULE.f $MODULE.junk #-- sed -e "s/Q-/D-/g" \ #-- -e "s/Q+00/D+00/g" \ #-- -e "s/REAL\*16/DOUBLE PRECISION/" $MODULE.junk > $MODULE.f #-- rm -f $MODULE.junk #-- endif #-- if (($MODULE == iolib) || ($MODULE == inputa)) then #-- mv $MODULE.f $MODULE.junk #-- sed -e "s/MXUNIT=299/MXUNIT=99/g" $MODULE.junk > $MODULE.f #-- rm -f $MODULE.junk #-- endif #-- set OPT = '-O3 -malign-double' #-- set echo #-- f2c -w66 -a -Nn1604 -Nx800 $MODULE.f #-- gcc -c $OPT $MODULE.c #-- rm -f $MODULE.c #-- unset echo #--endif # # Linux on IA-64 (Itanium or Pentium 4/EM64T), using Intel FORTRAN and C. # By omitting chip specific compiler options like "-mcpu=itanium2 -tpp2", # this version is able to run on Pentium 4/EM64T as well as Itanium2. # # We run this on a SGI Altix system, using the following software: # linux kernel: 2.4.21-sgi230r3 (aka Propack 2.3) # Intel efc/ecc compilers: Version 7.1, Build 20030307 # libraries: libgcc-3.0.4-1 # We run this on HP rx2600 workstations, using the following software: # linux kernel: 2.4.18-e.12smp (aka RedHat 2.1AS (Derry)) # Intel ifort/icc compilers: Version 8.0, Build 20031017 # Fred Arnold and Jan Fredin have used various older compilers, # such as 6.0 and 7.0, and older kernels and glibc's as well. # # At version 8.0, Intel changed the names of these compilers # to ifort and icc. To deal with this, we have used the newer # names in the scripts, and placed softlinks in /bin pointing # from ifort to efc and icc to ecc on our old 7.0 machine. # Or, you could change the script names back to the originals. # # Intel's compilers are commercially licensed software, but Intel allows # downloads of a "non-commercial unsupported version" at no charge: # http://developer.intel.com/software/products/compilers # Oddly, the Intel Math Kernel Library (MKL) which contains the BLAS # is marketed at a reduced cost (e.g. about $39) to academic sites, # rather than being free, but you can download a 30-day trial version # from Intel's site while obtaining a permanent license. A suggested # source for this license is Programmer's Paradise, # http://www.pparadise.com # who also sell the Intel compilers if you are a commercial site. # Note that the MKL library should be used only serially, by adding # this to 'rungms': setenv MKL_SERIAL YES # # Like all other compilers, this script assumes that you have set up # its paths and library specifications externally, not here. # For example, your .cshrc might need to add to the path # /opt/intel_fc_80/bin and /opt/intel_cc_80/bin # and define library locations by # setenv LD_LIBRARY_PATH /opt/intel_fc_80/lib:/opt/intel_cc_80/lib:\ # /opt/intel/mkl61/lib/64 # Similarly, you need to be sure that 'lked' finds Intel's MKL library. # Intel changes these paths around, indeed they even changed the name # of the compilers from efc/ecc to ifort/icc, although the old namess # still work and are therefore kept here for backward compatibility. # # -cm means suppress comments # -w95 suppresses Hollerith initialization warnings # -WB means warn but don't fail on out-of-bounds array references # -On can have n=1,2,3. Use of 3 caused problems with earlier # compilers, so it isn't being used here. Try it if you like. # # We have heard that one of the 8.1 compilers required degradation of # optimization for cpmchf to -O0 to run exam36 correctly. # if ($TARGET == linux-ia64) then set OPT = '-O2' if ($MODULE == morokm) set OPT='-O0' # Jan Fredin if ($MODULE == zheev) set OPT='-O0' # Fred Arnold set echo ifort $OPT -w95 -cm -WB -i8 -c $MODULE.f unset echo endif # # Apple Macintosh G4 or G5 system using OS X (Tiger, 10.4) and g77 3.4. # # This will run on any G4 or G5 system under 10.1, 10.2, 10.3, or 10.4, # but under 10.1, you'll need to run the old DDI version in 'compddi'. # # Step 1 is install the "Xcode" software to get a C compiler, gcc. # This is available for download at Apple's web site, # http://developer.apple.com/tools/xcode/index.html # but Xcode can also be found on the Tiger Install DVD. # The compiler toolkit must match your OS X version (in 10.1 and in # 10.2, the compiler suite was called the Apple Developer's Toolkit). # # Step 2 is to download the Fink binary installer from the FINK project # http://fink.sourceforge.net # and install it using the directions on that web page. # # Step 3 is to install g77 as superuser: # su (requires that root account has been enabled in NetInfo Manager) # source /sw/bin/init.sh or init.csh or add /sw/bin,/sw/sbin to path # apt-get update # apt-get upgrade # apt-get install g77 # The installation of Atlas BLAS library is optional, Apple's library # is almost as good as Atlas, and the next step may be rather lengthy: # apt-get install atlas # # On a 10.2 system with g77 3.1 and gcc 3.1, grd2b needed to be -O1 # # if ($TARGET == mac32) then set OPT = '-O2' #--if ($MODULE == grd2b) set OPT='-O1' if (($MODULE == qeigen) || ($MODULE == int2c)) then mv $MODULE.f $MODULE.junk sed -e "s/Q-/D-/g" \ -e "s/Q+00/D+00/g" \ -e "s/REAL\*16/DOUBLE PRECISION/" $MODULE.junk > $MODULE.f rm -f $MODULE.junk endif if (($MODULE == iolib) || ($MODULE == inputa)) then mv $MODULE.f $MODULE.junk sed -e "s/MXUNIT=299/MXUNIT=99/g" $MODULE.junk > $MODULE.f rm -f $MODULE.junk endif set echo g77 -c $OPT -fautomatic -Wno-globals -fno-globals $MODULE.f unset echo endif # # Apple Macintosh G5 system using OS X, xlf 8.1, and gcc. # # This should work for Panther or Tiger, but it is a 32 bit version. # The xlf compiler does not yet support the 64 bit address space of # the new Tiger OS, sunt lacrimae rerum. # # Why would you want to buy xlf in spite of being blocked from full # 64 bit computing? Some representative tests show that it is good # good to use a compiler that knows the chip's instruction set, these # are from a 2.3 GHz G5 Xserve running Tiger, # g77 3.4 xlf 8.1 # RHF gradient 371 238 CPU seconds # MP2 gradient 622 523 # GVB hessian 667 572 # MCSCF gradient 890 523 # The xlf FORTRAN compiler is marketed by Absoft, at www.absoft.com. # Absoft's initial pricing (as of 2/2004) for this compiler was # $399 academic, $599 government, and $999 commercial. # Note that the macG5 version uses gcc, so you do not need to buy xlc. # See the mac32 target just above about obtaining Apple's gcc. # # The G5 version is essentially an IBM AIX version, due to the compiler # being the same one. See the 'ibm32' comments for compiler options. # -qextname lets us access the BLAS routines found in Apple's vecLib. # if ($TARGET == macG5) then set OPT='-O3 -qarch=g5 -qtune=g5' set echo xlf -c $OPT -qintsize=4 -qnosave -qspillsize=1500 -qextname \ -qflag=W:W -qhalt=W $MODULE.f unset echo endif # # Options for f90 under SUPER-UX on the NEC SX Series are # -ebw enable passing scalar arguments and set all word size 64bit # -size_t64 flags 64bit size_t and associated libraries # -pvtcl noassume vwork=stack sets vectorization temporaries parameters # -Nv vectorization is turned off # if ($TARGET == necsx) then set OPT = '-Wf"-v -pvctl noassume vwork=stack"' if ($MODULE == mcqdpt) \ set OPT='-Wf"-v -pvctl noassume vwork=stack noaltcode nodivloop -NO"' if ($MODULE == ddi) \ set OPT='-Wf"-v -pvctl noassume vwork=stack -I /usr/include"' # compiler character problem if ($MODULE == inputc) set OPT='-Wf"-Nv"' # next ones are faster in scalar if ($MODULE == int2a) set OPT='-Wf"-Nv"' if ($MODULE == int2b) set OPT='-Wf"-Nv"' if ($MODULE == grd2a) set OPT='-Wf"-Nv"' if ($MODULE == grd2b) set OPT='-Wf"-Nv"' if ($MODULE == grd2c) set OPT='-Wf"-Nv"' if ($MODULE == gugem) set OPT='-Wf"-Nv"' if ($MODULE == gugdrt) set OPT='-Wf"-Nv"' # set echo f90 -c -ebw -size_t64 $OPT $MODULE.f unset echo endif # # Options for SGI f77 under Irix, this pertains only to MIPS processors. # If you have an Altix from SGI, please use target 'linux-ia64'. # # Different versions of the SGI compiler may not work with the fairly # aggressive -O3 optimization used below. If you experience any numerical # problems, please try using -O2 rather than -O3 below. It is unsafe to # use very aggressive optimizations like "-OPT:IEEE_arith=3:roundoff=3". # # If you use f90, do not reset the units for RECL in the sed hacks above. # # -i8/-i4 selects 8/4 byte integer and logical data types. # -show would add verbose printing of compiler phase information. # -automatic is the opposite of -static storage allocation. # -OPT:Olimit=0 removes subroutine size optimization limits. # -woff suppresses warnings (2290=type checking on calling args). # if (($TARGET == sgi32) || ($TARGET == sgi64)) then set OPT='-O3' if ($MODULE == grd2c) set OPT='-O2' if ($MODULE == rxncrd) set OPT='-O2' if ($MODULE == gmsnbo) set OPT='-O2 -backslash' if ($TARGET == sgi64) set INTS=' -64 -i8' if ($TARGET == sgi32) set INTS='-n32 -i4' set FLAGS='-automatic -G0 -woff 2290 -OPT:Olimit=0' set echo f77 -c $OPT $INTS $FLAGS $MODULE.f unset echo endif # # There is a united clause for UltraSPARC and Opteron chips! # Options for UltraSPARC Sun f90 7.0 under Solaris 9, and # options for Opteron Sun f90 8.1 under Solaris 10. # See also a f77 compiling clause for older systems commented out below. # # -c compile only # -v print verbose information on compiler phases and options # -stackvar forces non-static storage # -fast is a smorgasbord of all the below, if you wish to # override one you must *follow* the -fast with it. # -xtarget=native (optimize for chip being compiled upon) # -O5 (see below) # -libmil (inline some math routines) # -fsimple=2 (0,1,2 control floating point simplifications) # -dalign (double word alignment and instructions) # -xlibmopt (link to optimized math library) # -depend (DO loop data dependency analysis) # -fns (fast underflow handling) # -ftrap=common (floating point error handling) # -xvector=yes (select vectorized math library) # -xprefetch=yes (auto generation of instruction prefetching) # -O1 = minimal optimization (must turn off -depend to use this) # you may want to set this a particular chip if you run # more times on a different system than you compile on. # -O2 = basic local and global optimizations (must turn off -depend) # -O3 = loop unrolling and additional global optimizations # -O4 = automatic inlining # -O5 = aggressive optimizations # Experience suggests that setting OPT to '-fast -O2 -nodepend' is # normally very safe, for any Solaris/f90 versions. # # Chip selection is by -xarch=nnn, where nnn can be (among other values), # v8plus any UltraSPARC, using 32 bit instructions # v9 any UltraSPARC, using 64 bit instruction set # pentium_pro most Intel chips, using 32 bits # amd64 Opteron, using 64 bits # "uname -p" says the processor is sparc or i386, as a generic reply, # allowing us to distinguish the families, e.g. Opteron replies "i386"! # N.B. We have not tested the "pentium_pro" 32 bit stuff below. # # If you change -xarch, change it in all scripts: compall/comp/compddi/lked # if ($TARGET == sun32) then set BITS=32 if (`uname -p` == sparc) set ARCH='-xarch=v8plus' if (`uname -p` == i386) set ARCH='-xarch=pentium_pro' endif if ($TARGET == sun64) then set BITS=64 if (`uname -p` == sparc) set ARCH='-xarch=v9' # the Opteron does not support quadruple precision. if (`uname -p` == i386) then set ARCH='-xarch=amd64' if (($MODULE == qeigen) || ($MODULE == int2c)) then mv $MODULE.f $MODULE.junk sed -e "s/Q-/D-/g" \ -e "s/Q+00/D+00/g" \ -e "s/REAL\*16/DOUBLE PRECISION/" $MODULE.junk > $MODULE.f rm -f $MODULE.junk endif endif endif if (($TARGET == sun32) || ($TARGET == sun64)) then set OPT='-fast -O4' # next are detunings for f90 7.5 under Solaris 9 on SPARC if ($MODULE == grd2c) set OPT='-fast -O4 -nodepend' if ($MODULE == gugdrt) set OPT='-fast -O2 -nodepend' if ($MODULE == grd2c) set OPT='-fast -O2 -nodepend' # prcivc if ($MODULE == trnstn) set OPT='-fast -O2 -nodepend' # next are detunings for f90 8.1 under Solaris 10 on Opteron if ($MODULE == mp2ddi) set OPT='-fast -O4 -nodepend' if ($MODULE == umpddi) set OPT='-fast -O4 -nodepend' if ($MODULE == zheev) set OPT='-fast -O2 -nodepend' set WARN='-erroff=WARNING_1278,WARNING_1279,WARNING_1615' set echo f90 -c -stackvar -xtypemap=integer:$BITS $OPT $WARN $ARCH $MODULE.f unset echo endif # # Following is a potted history of Solaris versions that support # UltraSPARC chips. 64 bit O/S support came with Solaris 7, and # support for the UltraSPARC III chip came with Forte/f90 6.1. # year Solaris compilers FORTRAN # 1997 2.6 WorkShop 3.0 f77 4.2 # 1998 7 WorkShop 5.0 f90 2.0 # 2000 8 Forte 6 f90 6.0 # 2002 9 Forte 7 f90 7.0 # 2005 10 Sun Studio 10 f90 8.1 # Version of Solaris: "uname -sr", reply of 5.6 means 2.6, 5.9 means 9. # Version runs 64 bit: "isainfo -v" # Version of FORTRAN: "f77 -V" or "f90 -V" # Version of processor: "fpversion", look at the -xtarget response. # # We ran a E450 for many years with f77 4.2 under Solaris 2.6, # never using Solaris 7 or 8, instead we jumped directly to 9. # It would appear that the compiling clause above should be used for # Solaris 7, 8, or 9, so the below is relevant only to Solaris 2.6. # # If you use the following old clause, change LDR to f77 in "lked". # # If you have a pre-ultraSPARC system, try OPT='-O2' or '-O3' while # eliminating the -xarch flag, and you might also need -Nl30. # #--if ($TARGET == sun32) then #-- set OPT='-fast -O4 -xarch=v8plus' #-- if ($MODULE == ddi) set OPT='-fast -O4 -nodepend -xarch=v8plus' #-- if ($MODULE == grd2c) set OPT='-fast -O4 -nodepend -xarch=v8plus' #-- if ($MODULE == ormas1) set OPT='-fast -O4 -nodepend -xarch=v8plus' #-- if ($MODULE == symslc) set OPT='-fast -O4 -nodepend -xarch=v8plus' #-- set echo #-- f77 -c $OPT -stackvar $MODULE.f #-- unset echo #--endif # # Store the generated object code, clean up, and quit # mv $MODULE.o $OBJDIR/$MODULE.o if ($TINKER == true) exit rm -f $MODULE.f if ($SRCDIR == .) rm -f $MODULE.src # delete copy with no scalar code unset echo date time