Test suite segfaults with gfortran 9.2.1

[mbanck]

Running the fast testsuite, all tests fail similarily in , like for t30:

[...]
 ETOT 13  -8.7264004259227    -1.599E-14 8.874E-25 9.869E-22
 prteigrs : about to open file t30o_EIG
 Fermi (or HOMO) energy (hartree) =   0.19736   Average Vxc (hartree)=  -0.35360
 Eigenvalues (hartree) for nkpt=   2  k points:
 kpt#   1, nband=  4, wtk=  0.25000, kpt=  0.2500  0.2500  0.2500 (reduced coord)
  -0.18526    0.07865    0.19736    0.19736
 kpt#   2, nband=  4, wtk=  0.75000, kpt=  0.2500  0.5000  0.5000 (reduced coord)
  -0.11309   -0.01075    0.09006    0.13464
 Fermi (or HOMO) energy (eV) =   5.37042   Average Vxc (eV)=  -9.62196
 Eigenvalues (   eV  ) for nkpt=   2  k points:
 kpt#   1, nband=  4, wtk=  0.25000, kpt=  0.2500  0.2500  0.2500 (reduced coord)
  -5.04123    2.14017    5.37042    5.37042
 kpt#   2, nband=  4, wtk=  0.75000, kpt=  0.2500  0.5000  0.5000 (reduced coord)
  -3.07722   -0.29262    2.45073    3.66369
 scprqt: <Vxc>= -3.5360049E-01 hartree

 At SCF step   13   max residual=  8.87E-25 < tolwfr=  1.00E-24 =>converged.

Program received signal SIGSEGV, Segmentation fault.
0x0000555556141276 in m_forces::forces (atindx1=..., diffor=0, dtefield=..., dtset=..., favg=..., fcart=..., fock=0x0, forold=..., fred=..., grchempottn=..., gresid=..., grewtn=..., grhf=..., grnl=..., 
    grvdw=..., grxc=..., gsqcut=0.4052847345693511, indsym=..., maxfor=0, mgfft=10, mpi_enreg=..., n1xccc=2501, n3xccc=1000, nattyp=..., nfft=1000, ngfft=..., ngrvdw=0, ntypat=1, pawrad=..., pawtab=..., 
    ph1d=..., psps=..., rhog=..., rhor=..., rprimd=..., symrec=..., synlgr=..., usefock=0, vresid=..., vxc=..., wvl=..., wvl_den=..., xred=..., electronpositron=0x0) at m_forces.F90:494
494	 if (usefock==1 .and. associated(fock).and.fock%fock_common%optfor) then
(gdb) bt full      
#0  0x0000555556141276 in m_forces::forces (atindx1=..., diffor=0, dtefield=..., dtset=..., favg=..., fcart=..., fock=0x0, forold=..., fred=..., grchempottn=..., gresid=..., grewtn=..., grhf=..., grnl=..., 
    grvdw=..., grxc=..., gsqcut=0.4052847345693511, indsym=..., maxfor=0, mgfft=10, mpi_enreg=..., n1xccc=2501, n3xccc=1000, nattyp=..., nfft=1000, ngfft=..., ngrvdw=0, ntypat=1, pawrad=..., pawtab=..., 
    ph1d=..., psps=..., rhog=..., rhor=..., rprimd=..., symrec=..., synlgr=..., usefock=0, vresid=..., vxc=..., wvl=..., wvl_den=..., xred=..., electronpositron=0x0) at m_forces.F90:494
        atmrho_dum = <not allocated>
        atmvloc_dum = <not allocated>
        calc_epaw3_forces = .FALSE.
        coredens_method = 2
        dummy6 = (0, 0, 0, 0, 0, 0)
        dyfrlo_dum = <not allocated>
        dyfrn_dum = <not allocated>
        dyfrv_dum = <not allocated>
        dyfrx2_dum = <not allocated>
        eei_dum1 = 0
        eei_dum2 = 9.1995023255640107e-321
        efield_flag = .FALSE.
        eltfrn_dum = <not allocated>
        ep3 = (4.6355756815252366e-310, 2.1219957913605248e-313, 4.0414569829813967e-321)
        epawf3red = <not allocated>
        fdir = -118248
        fin = (( 0, 0, 0) ( 0, 0, 0) )
        fioncart = (1.0185579797819065e-312, 6.9533558006145313e-310, 0)
        fionred = <not allocated>
        gauss_dum = <not allocated>
        gmet = (( 0.028481528154392026, -0.0094938427181306753, -0.0094938427181306753) ( -0.0094938427181306753, 0.028481528154392026, -0.0094938427181306736) ( -0.0094938427181306753, -0.0094938427181306736, 0.028481528154392026) )
        gprimd = (( -0.097436352138874097, 0.097436352138874097, 0.097436352138874097) ( 0.097436352138874097, -0.097436352138874097, 0.097436352138874097) ( 0.097436352138874097, 0.097436352138874097, -0.097436352138874097) )
        grl = (( -6.6613381477509392e-16, 2.2204460492503131e-16, 2.2204460492503131e-16) ( 8.8817841970012523e-16, -4.4408920985006262e-16, -4.4408920985006262e-16) )
        grnl_tmp = <not allocated>
        grtn = (( 1.5668782803424292e-05, 1.5668782823254329e-05, 1.5668782829866912e-05) ( -1.5668782758980094e-05, -1.5668782785622871e-05, -1.5668782779227677e-05) )
        grtn_indx = <not allocated>
        iatom = 3
        idir = -90200
        indx = 3
        ipositron = 21845
        is_hybrid_ncpp = .FALSE.
        itypat = 74
        mu = 1444228441
        optatm = 0
        optdyfr = 0
        opteltfr = 32767
        optgr = -146528
        option = 2
        optn = 1880027648
        optn2 = 32767
        optstr = -146528
        optv = -709758133
        qprtrb_dum = (-147118, 32767, 1880027648)
        rmet = (( 52.665713436049991, 26.332856718024996, 26.332856718024996) ( 26.332856718024996, 52.665713436049991, 26.332856718024996) ( 26.332856718024996, 26.332856718024996, 52.665713436049991) )
        strn_dummy6 = (-nan(0xfffffffffffff), 0, -nan(0xfffffffffffff), 3.9525251667299724e-323, 1.6360587548337309e-311, 6.9533485820324448e-310)
        strv_dummy6 = (2.1218986537067748e-314, 0, 6.9533557981212785e-310, 0, -nan(0xfffffffffffff), 0)
        tsec = (0, 0)
        ucvol = 270.25700511132948
        ucvol_local = 9.8813129168249309e-324
        v_dum = <not allocated>
        vloc_method = 2
        vol_element = 0
        vprtrb_dum = (6.9533522250891695e-310, 6.9533558005818242e-310)
        vxctotg = <not allocated>
        xccc3d_dum = <not allocated>
#1  0x0000555556153f75 in m_forstr::forstr (atindx1=..., cg=..., cprj=..., diffor=0, dtefield=..., dtset=..., eigen=..., electronpositron=0x0, energies=..., favg=..., fcart=..., fock=0x0, forold=..., 
    fred=..., grchempottn=..., gresid=..., grewtn=..., grhf=..., grvdw=..., grxc=..., gsqcut=0.4052847345693511, 
    indsym=<error reading variable: value requires 935902144 bytes, which is more than max-value-size>, kg=..., kxc=..., maxfor=0, mcg=296, mcprj=0, mgfftf=10, mpi_enreg=..., my_natom=2, n3xccc=1000, 
    nattyp=<error reading variable: value requires 4294606496 bytes, which is more than max-value-size>, nfftf=1000, ngfftf=..., ngrvdw=0, nhat=..., nkxc=0, npwarr=..., ntypat=1, nvresid=..., occ=..., 
    optfor=1, optres=0, paw_ij=..., pawang=..., pawfgr=..., pawfgrtab=..., pawrad=..., pawrhoij=..., pawtab=<error reading variable: value requires 1815569456 bytes, which is more than max-value-size>, 
    ph1d=<error reading variable: value requires 4292698624 bytes, which is more than max-value-size>, ph1df=..., psps=..., rhog=..., rhor=..., rprimd=..., stress_needed=1, strsxc=..., strten=..., symrec=..., 
    synlgr=..., ucvol=270.25700511132948, usecprj=0, vhartr=..., vpsp=..., vxc=..., wvl=..., xccc3d=..., xred=..., ylm=..., ylmgr=..., qvpotzero=0) at m_forstr.F90:485
[...]

It seems the fock pointer is invalid, though I am not sure whether that is the root of the problem:

(gdb) p fock
$1 = (PTR TO -> ( Type fock_type )) 0x0
(gdb) p *fock
Cannot access memory at address 0x58

The stacktrace as printed by the testsuite is as follows:

Backtrace for this error:
#0  0x7f0836659b90 in ???
#1  0x7f0836658dc5 in ???
#2  0x7f083631183f in ???
#3  0x55d9a4a23276 in __m_forces_MOD_forces
	at /<<PKGBUILDDIR>>/src/67_common/m_forces.F90:494
#4  0x55d9a4a35f74 in __m_forstr_MOD_forstr
	at /<<PKGBUILDDIR>>/src/67_common/m_forstr.F90:485
#5  0x55d9a40d6c81 in __m_afterscfloop_MOD_afterscfloop
	at /<<PKGBUILDDIR>>/src/94_scfcv/m_afterscfloop.F90:932
#6  0x55d9a40b9dbc in __m_scfcv_core_MOD_scfcv_core
	at /<<PKGBUILDDIR>>/src/94_scfcv/m_scfcv_core.F90:2108
#7  0x55d9a408f014 in scfcv_scfcv
	at /<<PKGBUILDDIR>>/src/94_scfcv/m_scfcv.F90:746
#8  0x55d9a408f5ef in __m_scfcv_MOD_scfcv_run
	at /<<PKGBUILDDIR>>/src/94_scfcv/m_scfcv.F90:536
#9  0x55d9a4077638 in __m_gstate_MOD_gstate
	at /<<PKGBUILDDIR>>/src/95_drive/m_gstate.F90:1330
#10  0x55d9a3e7eeb8 in __m_gstateimg_MOD_gstateimg
	at /<<PKGBUILDDIR>>/src/95_drive/m_gstateimg.F90:550
#11  0x55d9a3e5b142 in __m_driver_MOD_driver
	at /<<PKGBUILDDIR>>/src/95_drive/m_driver.F90:705
#12  0x55d9a3e49185 in abinit
	at /<<PKGBUILDDIR>>/src/98_main/abinit.F90:444
#13  0x55d9a3e4caaf in main
	at /<<PKGBUILDDIR>>/src/98_main/abinit.F90:94
Segmentation fault

[mbanck]

If I comment out the if (usefock==1 .and. associated(fock).and.fock%fock_common%optstr) then and similar in m_forces.F90, m_forstr.F90 and m_stress.F90, the test case runs fine (but t08-t12 are still failing).

If I step through those functions in gdb, I clearly see usefock = 0 though, so I am a bit baffled why that is needed.

[mbanck]

It runs fine with FCFLAGS=-ffree-line-length-none -g -O2 but I get the segfaults with FCFLAGS=-ffree-line-length-none -g

[mbanck]

So it appears that fock%fock_common%optstr cannot be guaranteed not to be evaluated after associated(fock) is false as the Fortran language apparently does not define short-circuit beahviour, see https://www.scivision.dev/fortran-short-circuit-logic/

That explains why it works fine at -O2 (and would as well at -O1 in gfortran), but crashed in -O0.

I guess the fix would be to have two ifs here, like:

 if (usefock==1 .and. associated(fock)) then
   if (fock%fock_common%optfor) then
     grtn(:,:)=grtn(:,:)+fock%fock_common%forces(:,:)
   end if
 end if

[jmbeuken]

Thank you for pointing out this problem.
We will integrate the correction in version 8.11.12.

In addition, how did you compile ABINIT ( which version? ) to "trigger" this error?
We have not detected this problem on our testfarm.
We compile ABINIT on a bot with gnu 9.2 and

FCFLAGS_EXTRA="-O2 -g -Wall -Wno-maybe-uninitialized -ffpe-trap=invalid,zero,overflow -fbacktrace -pedantic -fcheck=all"
thank you

[mbanck]

You need to build it with -O0 in order to crash it, at -O2 GFortran decides to use the short-circuit behavior and runs fine.

The Debian packaging was using -O0 (at least for a subset that included the problematic source file) so we saw this crash. Why it did not crash in earlier versions (-O0 was set for a long time) I do not know, maybe the GFortran behavior has changed, so I could not find any substantive information on that when tried to research it back in September.