PGI Guide to GAMESS

This guide is intended to help PGI customers build and run the General Atomic and Molecular Electronic Structure System (GAMESS) using PGI 5.1 and 5.2 compilers on a on a AMD64 or Operton 64-bit Linux system.

Version Information
 

This guide was created for the GAMESS December 12, 2003 version and PGI Release 5.1-3 pgf90 and pgcc 64-bit compilers on an AMD64 or Operton system running 64-bit Linux.  The guide was updated on July 14, 2004 for the 5.2 release.

Currently, GAMESS does not support AMD64 and needs to be ported.  In the Configuration Section of this document we describe what source code changes need to be made for the 32-bit x86 version of GAMESS to be ported to AMD64.  The port was developed by AMD and PGI.

Application Notes
  Information about GAMESS can be found at the GAMESS home page. GAMESS is maintained by the members of the Gordon research group at Iowa State University. From the GAMESS Home page:

"GAMESS is a program for ab initio quantum chemistry. Briefly, GAMESS can compute SCF wave functions ranging from RHF, ROHF, UHF, GVB, and MCSCF. Correlation corrections to these SCF wave functions include Configuration Interaction, second order perturbation theory, and Coupled-Cluster approaches, as well as the Density Functional Theory approximation. Analytic gradients are available, for automatic geometry optimization, transition state searches, or reaction path following. Computation of the energy hessian permits prediction of vibrational frequencies. The chart below summarizes the program's present capabilities for obtaining wave functions, applying correlation treatments, and computing derivatives. A variety of molecular properties, ranging from simple dipole moments to frequency dependent hyperpolarizabilities may be computed. Many basis sets are stored internally, and together with effective core potentials, all elements up to Radon may be included in molecules. Several graphics programs are available for viewing of the final results. Many of the computational functions can be performed using direct techniques, or in parallel on appropriate hardware."

Obtaining the Source Code
  Information on obtaining the GAMESS source code can be found at the US Government's Ames Laboratory. The Ames lab requires you to register before downloading the source code.  Note, you should download the 32-bit x86 Linux version and make the porting changes as described in the Configuration Section.
Dependencies
  No known dependencies.
Configuration and Set-up Information
  Once unpacked, please read the enclosed documentation (*.DOC) files and the "readme.unix" file in the "misc" directory.  To build and run GAMESS, you will first need to edit several scripts (compall, comp, lked, rungms, runall) and five source files (ddikick.c, ddisoc.c, trfdm2.src, unport.src, zunix.c).  Due to license restrictions on the source code, we cannot make available pre configured versions of these files. We have submitted these changes to the Ames Laboratory and hope to see them adopted in future GAMESS releases.  Note that the line numbers may change with each release of GAMESS.  Line numbering is progressive so as you add new lines, the next line number reference is for the modified file.

compall
  1. Line 15: set the "TARGET" to "linux-amd64". 
  2. Line 16: replace the "/u1/mike/gamess" with the root directory of your GAMESS package.
  3. Line 34: set "CCOMP" to "pgcc".
    Note: You may need to explicitly set your PATH environment variable as well.
       setenv PATH $PATH\:/usr/pgi/linux86-64/[5.1|5.2]/bin #Set to your path
  4. Line 56: Add a new "extraflag": 
    if ($TARGET == linux-amd64)  set extraflags='-DLINUXAMD64'
  5. Line 110: Add a new case for linux-amd64: 
    case linux-amd64:
  6. Line 124: Add a new "socketflags": 
    if ($TARGET == linux-amd64) set socketflags='-DLINUXAMD64'
  7. If using IPA, add the "-Mipa=[option]" flag to the compilation of zunix (line 67), ddikick (line 135), and ddisoc (line 136).
comp
  1. Line 17: set the "TARGET" to "linux-amd64".
  2. Line 18: replace the "/u1/mike/gamess" with the root directory of your GAMESS package.
  3. Line 44: If you wish to use a BLAS library, i.e. "-lblas" or '-lacml', set "BLAS3=true".
  4. Line 312: Add the following after the "linux-pc" target: 
    if ($TARGET == linux-amd64) then
                              setenv MACHIN '*I64'
       if ($MODULE == ddi)    setenv MACHIN '*SOC'
       if ($MODULE == iolib)  setenv MACHIN '*UNX'
       if ($MODULE == unport) setenv MACHIN '*AMD'
    endif
  5. Line 899: After the linux-ia64 target add the following: 
    if ($TARGET == linux-amd64) then
       setenv PATH $PATH\:/usr/pgi/linux86-64/5.1/bin #Set to your path
       set OPT = '-fast -i8'
       if ($MODULE == qeigen) then
          mv qeigen.f qeigen.junk
          sed -e "s/Q-/D-/g" \
              -e "s/Q+00/D+00/g" \
              -e "s/REAL\*16/DOUBLE PRECISION/" qeigen.junk > qeigen.f
          rm -f qeigen.junk
       endif
       set echo
       # Compile in the object directory when using IPA
       mv $MODULE.f object/$MODULE.f
       chdir object
       pgf90 $OPT -Mfixed -c $MODULE.f
       chdir ../
       unset echo
    endif
    Be sure to change your path to the location of the PGI compilers.
    Note that we have modified the script to build objects in the object directory to facilitate the use of IPA.  At line 1130, the script attempts to copy the object file from the base directory to the object directory.  Since the object file does not exist in the base directory, an error will be emitted.  You can either comment out line 1130 or ignore the error.
"-i8" must be used when compiling the Fortran source or the program will hang during execution.  You can modify the other option flags as needed but we recommend using "-fast -i8".  You can use either pgf77 or pgf90.  Remove "-Mfixed" if using pgf77.
lked
  1. Line 17: set the "TARGET" to "linux-amd64".
  2. Line 18: replace the "/u1/mike/gamess" with the root directory of your GAMESS package.
  3. Line 55: Add a new case for linux-amd64: 
    case linux-amd64:
  4. Line 319: After the linux-ia64 target add the following: 
    if ($TARGET == linux-amd64) then
       set LDR='pgf90'
       set LDOPTS='-fast -i8 -lblas -lpthread'   
       set VECTOR=' '
       set QUICHE='zunix.o'
       set BLAS=' '         # or set BLAS='blas.o' if BLAS3=false in comp
       set LIBRARIES=' '
    endif
  5. You may also need to explicitly set your path to the compiler binaries: 
    setenv PATH $PATH\:/usr/pgi/linux86/[5.1|5.2]
rungms
  1. Determine where your scratch directory is located. For our testing purposes, we created a "scr" directory in the local gamess directory. Set the "SCR" variable (line 33) to this directory. Note that this should be an absolute path.
  2. In the file assignment section, you should change files starting with "~/scr" to the actual directory location. For our tests, we used the local scr directory.
  3. Line 227: Set the "GMSPATH" to the location of the ddikick and gamess executables.
runall Change the "chdir" to the location of the ddikick and gamess executables.
source/ddikick.c Line 57: Add the following case to the the #if statement: 
        (defined LINUXAMD64) ||                 \
source/ddisoc.c After the LINUX definitions add the following (line 204) 
    #ifdef LINUXAMD64
    #include <sys/select.h>
    #include <sys/time.h>
    #define SOC_INIT      soc_init_
    #define SOC_NPROC     soc_nproc_
    #define SOC_SEND      soc_send_
    #define SOC_RECV      soc_recv_
    #define SOC_ISEND     soc_isend_
    #define SOC_IRECV     soc_irecv_
    #define SOC_WAIT      soc_wait_
    #define SOC_GSEND     soc_gsend_
    #define SOC_GRECV     soc_grecv_
    #define SOC_RCVANY    soc_rcvany_
    #define SOC_WAKEKICK  soc_wakekick_
    #define SOC_KILLKICK  soc_killkick_
    #define FORTINT long
    #define getsockcast int
    #define SOCK_BUFF_SIZE 32768  /* half RedHat 5.1 default of 64K is better */
    #endif
source/trfdm2.src
  1. Modify lines 585 to 588.  Change 
          IMOD=LOCI
          JMOD=LOCJ
          KMOD=LOCK
          LMOD=LOCL
    To: 
          IMODAMD=LOCI
          JMODAMD=LOCJ
          KMODAMD=LOCK
          LMODAMD=LOCL
  2. Modify lines 595 to 600.  Change 
         IJKLX = NJKL*(IMOD-MINI)+ NKL*(JMOD-MINJ)+
         *          NL*(KMOD-MINK)+     (LMOD-MINL)+1+IJKLAD
          XIJKL(LOC     ) = VAL
          XIJKL(LOC+LBIN) = IJKLX
          IF(DBUG) WRITE(IW,9995) I,J,K,L,ISH,JSH,KSH,LSH,IBOX,
         *                        IMOD,JMOD,KMOD,LMOD,IJKLX,VAL,IJKLAD
    To: 
         IJKLX = NJKL*(IMODAMD-MINI)+ NKL*(JMODAMD-MINJ)+
         *          NL*(KMODAMD-MINK)+     (LMODAMD-MINL)+1+IJKLAD
          XIJKL(LOC     ) = VAL
          XIJKL(LOC+LBIN) = IJKLX
          IF(DBUG) WRITE(IW,9995) I,J,K,L,ISH,JSH,KSH,LSH,IBOX,
         *         IMODAMD,JMODAMD,KMODAMD,LMODAMD,IJKLX,VAL,IJKLAD
  3. Modify lines 619 to 665 as follows: 
      141 IF(.NOT.IANDJ) GO TO 140
          IMODAMD=LOCJ
          JMODAMD=LOCI
          KMODAMD=LOCK
          LMODAMD=LOCL
          GO TO 130
    C
      142 IF(.NOT.KANDL) GO TO 140
          IMODAMD=LOCI
          JMODAMD=LOCJ
          KMODAMD=LOCL
          LMODAMD=LOCK
          GO TO 130
    C
      143 IF(.NOT.IANDJ.OR..NOT.KANDL) GO TO 140
          IMODAMD=LOCJ
          JMODAMD=LOCI
          KMODAMD=LOCL
          LMODAMD=LOCK
          GO TO 130
    C
      144 IF(.NOT.IJNDKL) GO TO 150
          IMODAMD=LOCK
          JMODAMD=LOCL
          KMODAMD=LOCI
          LMODAMD=LOCJ
          GO TO 130
    C
      145 IF(.NOT.KANDL) GO TO 140
          IMODAMD=LOCL
          JMODAMD=LOCK
          KMODAMD=LOCI
          LMODAMD=LOCJ
          GO TO 130
    C
      146 IF(.NOT.IANDJ) GO TO 140
          IMODAMD=LOCK
          JMODAMD=LOCL
          KMODAMD=LOCJ
          LMODAMD=LOCI
          GO TO 130
    C
      147 IF(.NOT.KANDL.OR..NOT.IANDJ) GO TO 150
          IMODAMD=LOCL
          JMODAMD=LOCK
          KMODAMD=LOCJ
          LMODAMD=LOCI
source/unport.src
  1. Add the following after the "PC-UNIX" section around line 202: 
    C-----------------------------------------------------------------------
    C                          AMD64-PGI
    *AMD  IF (ICORFL.EQ.1) THEN
    *AMD     WRITE(IW,1) 'GDB','BACKTRACE','QUIT'
    *AMD     CALL ABORT
    *AMD  ELSE
    *AMD     WRITE(IW,2)
    *AMD  END IF
  2. Add the following around line 340: 
    *AMD  NWDVAR = 1
  3. Add the following around line 382: 
    *AMD  VERSN = '********* AMD64-PGI VERSION ********'
  4. Add the following around line 495: 
    *AMD  LOGAND =  AND(IVAL,JVAL)
  5. Add the following around line 535: 
    *AMD  LOGOR =  OR(IVAL,JVAL)
  6. Add the following around line 580: 
    *AMD  LOGXOR =  XOR(IVAL,JVAL)
  7. Add the following around line 728: 
    C 
    *AMD  PARAMETER (MEMSIZ=   1 000 000)
    *AMD  COMMON /FMCOM / X(1)
  8. Add the following around line 800: 
    *AMD  LOCX = LADDRS(X)
  9. Add the following around line 845: 
    *AMD  LOCMEM = MEMGET(MEMLIM)
  10. Add the following around line 875: 
    *AMD  LOFFS = (LOFFS+7)/8 + 1
  11. Add the following around line 980: 
    *AMD  CALL MEMREL(LOCMEM)
  12. Add the following around line 1154 (after F2C): 
    C-----------------------------------------------------------------------
    C 
    *AMD  DOUBLE PRECISION TIMSTR(3)
    *AMD  CHARACTER*24 STAMP
    *AMD  CALL FDATE(STAMP)
    *AMD  READ(UNIT=STAMP,FMT='(3A8)') TIMSTR
  13. Add the following around line 1454 (after F2C): 
    C
    C-----------------------------------------------------------------------
    C           * * * *  AMD64-PGI SPECIFIC PORTION  * * * *
    C
    *AMD  DOUBLE PRECISION CPU0,WALL0,CPU,WALL,TIM
    *AMD  EXTERNAL ETIME
    *AMD  DOUBLE PRECISION ETIME
    *AMD  REAL TARRAY(2)
    *AMD  LOGICAL FIRST
    *AMD  COMMON /TIMING/ CPU,WALL
    *AMD  SAVE FIRST,CPU0,WALL0
    *AMD  DATA FIRST/.TRUE./
    C
    *AMD  IF(FIRST) THEN
    *AMD     FIRST=.FALSE.
    *AMD     WALL0 = ETIME(TARRAY)
    *AMD     CPU0 = TARRAY(1)+TARRAY(2)
    *AMD  END IF
    C
    *AMD  WALL = ETIME(TARRAY)
    *AMD  CPU = TARRAY(1)+TARRAY(2)
    *AMD  CPU = CPU - CPU0
    *AMD  WALL = WALL - WALL0
    *AMD  TIM = CPU
    *AMD  RETURN
source/zunix.c Add the following around line 260 
    /*----------------- AMD64 running Linux -----------------*/

    #ifdef LINUXAMD64

    #include <stdlib.h>
    #include <malloc.h>

    long memget_(nwords) long *nwords;
       { size_t nbytes;
         nbytes = (*nwords+2)*8;
         return (long) malloc(nbytes); }

    void memrel_(locmem) long *locmem;
       { free((void*)*locmem); }

    long laddrs_(void *arg)
      { return (long) arg; }

    #include <unistd.h>
    void naptime_(nptime) long *nptime;
       { sleep((unsigned int) *nptime); }

    #endif
Building GAMESS
 
Steps to build GAMESS
  1. Configure your installation as described in the Configuration Section.
  2. Build the activate utility in the tools directory:
    cd tools
    cp actvte.code actvte.f
    vi actvte.f
    :%s/^\*UNX/    / <--- 4 blanks between the strokes
    :wq
    pgf77 -o actvte.x actvte.f
    rm actvte.f
    cd ../
  3. Type "compall" from the command line to compile the source.
  4. Type "lked" from the command line to link the source.
  5. 5.1 only: Repeat steps 3 and 4 if using IPA.
Running GAMESS
  To run the 35 example data sets, use the "runall" command.  The output can be found in the "eN.log" files.
Verifying Correctness
  The "TESTS.DOC" file gives a full explanation on verifying the example program's output.  The abbreviated version is as follows.  The "INPUT CARD" portion of each log file contains some expected values.  You need to search the output to see if these values match the actual output. Note that there are difference between the log file's expected answer and the "TESTS.DOC" file's expected answers.  You should use the log file's expected values.  Also, there might be slight precision variances between the actual and expected values. Precision can change depending upon your system and compiler optimization. See the Precision FAQ for more information.
Known Issues and Limitations
  The following are a list of known issues when using the PGI compiler with GAMESS.
  • "-i8" Must be used
    GAMESS uses a mix of both Fortran and C and care must be taken to ensure that integer data type representation matches. "-i8" ensures 8 byte integer representation which is the same as a C "long int".
  • Using "-Minline"
    Using "-Minline" will cause several source files to grow too large for compilation.
  • Using "-Mbyteswapio" and "-Mchkptr"
    Using "-Mbyteswapio" and "-Mchkptr" can cause a few example tests to exit abnormally. They are not needed for GAMESS and should not be used.
  • Incorrect results with "-fastsse" and "-Mvect"
    A bug in the vectorizer can cause GAMESS to produce incorrect results.   Vectorization is enabled using the "-Mvect=[option]" flag and is part of "-fastsse".  You can set your options to "-fast -Mscalarsse -Mcache_align -Mflushz" which is equivalent to "-fastsse" less "-Mvect=sse".
Click me