gcc/testsuite/gfortran.dg/vect/pr32380.f

   1 ! { dg-do compile }
   2 ! { dg-require-effective-target vect_float }
   3 ! { dg-additional-options "-O3 -fcray-pointer" }
   4 ! PR 32380 - loops were not vectorized due to unaligned store.
   5       subroutine trnfbt(e,f,qs,mte,gm,ihgenf,hgener,lft,llt,sthick,
   6      .                  fibl,istupd,ies,hoff)
   7       parameter (nlq=96)
   8       integer nnlq
   9       common/newnlq/nnlq
  10 c   ... implicit common ...
  11       integer imauto,iteopt,lauto,mthsol,ilimit,maxref,icnvrg,
  12      & igdiv,nwebuf,neql,neqt,imterm,imphas,nbfgs,
  13      & numupd,istif,itrlas,imerr,imdof,neqtgl,lsmtd,lsdir
  14       common/bki01i/imauto,iteopt,lauto,mthsol,ilimit,maxref,icnvrg,
  15      & igdiv,nwebuf,neql,neqt,imterm,imphas,nbfgs,
  16      & numupd,istif,itrlas,imerr,imdof,neqtgl,lsmtd,lsdir
  17       REAL dtimp,dtimp0,timeim,dtmnim,dtmxim,cvtl,ectl,rctl,
  18      & tolls,dnorm2,dtprnt,dtplot,dtiter,dtrefm
  19       common/bki01r/dtimp,dtimp0,timeim,dtmnim,dtmxim,cvtl,ectl,rctl,
  20      & tolls,dnorm2,dtprnt(2),dtplot(2),dtiter(2),dtrefm(2)
  21       REAL ascntl
  22       common/bki02r/ascntl(150)
  23       logical lsensw
  24       common/bki01l/lsensw(20)
  25       integer imip,isolvr,icwrb
  26       common/bki02i/imip(100),isolvr(200),icwrb(50)
  27 c   ... implicit common ...
  28 c
  29 c
  30 c
  31       integer lnodim,ndofpn,nnpke,melemt,imlft,imllt,is17loc
  32       common/bki03iloc/lnodim(nlq,16),ndofpn,nnpke,melemt,imlft,imllt,
  33      &                 is17loc
  34       real*4 ske
  35       common/bki03rloc/ske(nlq,1176)
  36       integer lmke
  37       common/bki04iloc/lmke(nlq,48)
  38 c******************************************************************
  39 c|  livermore software technology corporation  (lstc)             |
  40 c|  ------------------------------------------------------------  |
  41 c|  copyright 1987,1988,1989 john o. hallquist, lstc              |
  42 c|  all rights reserved                                           |
  43 c******************************************************************
  44 c
  45 c
  46 c
  47 c
  48 c
  49 c
  50 c
  51 c
  52 c
  53 c
  54 c
  55 c
  56 c
  57 c
  58 c
  59 c
  60 c
  61       common/bk12loc/b12,b2,qhg,qhgm,qhgb,qhgw
  62       common/aux00loc/
  63      & sig1m(nlq),sig2m(nlq),sig4m(nlq),sig1n(nlq),sig2n(nlq),
  64      & sig4n(nlq),sig5n(nlq),sig6n(nlq),sig5l(nlq),sig6l(nlq),
  65      & str33(nlq),enginc(nlq)
  66       common/aux01loc/
  67      &ft11(nlq),ft12(nlq),ft13(nlq),ft21(nlq),ft22(nlq),ft23(nlq),
  68      &fm11(nlq),fm12(nlq),fm21(nlq),fm22(nlq),
  69      &fm31(nlq),fm32(nlq),fm41(nlq),fm42(nlq),
  70      &fmr11(nlq),fmr12(nlq),fmr21(nlq),fmr22(nlq),fmr31(nlq),
  71      &fmr32(nlq),fmr41(nlq),fmr42(nlq),sg5(nlq),sg6(nlq)
  72       common/aux7loc/
  73      1 vx1(nlq),vx2(nlq),vx3(nlq),vx4(nlq),
  74      2 vx5(nlq),vx6(nlq),vx7(nlq),vx8(nlq),
  75      3 vy1(nlq),vy2(nlq),vy3(nlq),vy4(nlq),
  76      4 vy5(nlq),vy6(nlq),vy7(nlq),vy8(nlq),
  77      5 vz1(nlq),vz2(nlq),vz3(nlq),vz4(nlq),
  78      6 vz5(nlq),vz6(nlq),vz7(nlq),vz8(nlq)
  79       common/aux10loc/area(nlq),
  80      1 px1(nlq),px2(nlq),px3(nlq),px4(nlq),
  81      & px5(nlq),px6(nlq),px7(nlq),px8(nlq),
  82      2 py1(nlq),py2(nlq),py3(nlq),py4(nlq),
  83      & py5(nlq),py6(nlq),py7(nlq),py8(nlq),
  84      3 pz1(nlq),pz2(nlq),pz3(nlq),pz4(nlq),
  85      & pz5(nlq),pz6(nlq),pz7(nlq),pz8(nlq),
  86      4 dx1(nlq),dx2(nlq),dx3(nlq),dx4(nlq),
  87      5 dx5(nlq),dx6(nlq),dx7(nlq),dx8(nlq),
  88      6 dy1(nlq),dy2(nlq),dy3(nlq),dy4(nlq),
  89      7 dy5(nlq),dy6(nlq),dy7(nlq),dy8(nlq),
  90      8 dz1(nlq),dz2(nlq),dz3(nlq),dz4(nlq),
  91      9 dz5(nlq),dz6(nlq),dz7(nlq),dz8(nlq)
  92       common/aux11loc/
  93      &ft31(nlq),ft32(nlq),ft33(nlq),ft41(nlq),ft42(nlq),ft43(nlq),
  94      &htx(nlq),hty(nlq),gm1(nlq),gm2(nlq),gm3(nlq),gm4(nlq),
  95      &bsum(nlq),qhx(nlq),qhy(nlq),qwz(nlq),qtx(nlq),qty(nlq)
  96       real*4 mx1,my1,mz1,mx2,my2,mz2,mx3,my3,mz3,mx4,my4,mz4
  97       common/aux13loc/
  98      &zeta(nlq),thick(nlq),fga(nlq),fgb(nlq),fgc(nlq),
  99      &gl11(nlq),gl12(nlq),gl13(nlq),gl21(nlq),gl22(nlq),gl23(nlq),
 100      &gl31(nlq),gl32(nlq),gl33(nlq),
 101      &x1(nlq),y1(nlq),z1(nlq),x2(nlq),y2(nlq),z2(nlq),
 102      &x3(nlq),y3(nlq),z3(nlq),x4(nlq),y4(nlq),z4(nlq),
 103      &fx1(nlq),fy1(nlq),fz1(nlq),fx2(nlq),fy2(nlq),fz2(nlq),
 104      &fx3(nlq),fy3(nlq),fz3(nlq),fx4(nlq),fy4(nlq),fz4(nlq),
 105      &mx1(nlq),my1(nlq),mz1(nlq),mx2(nlq),my2(nlq),mz2(nlq),
 106      &mx3(nlq),my3(nlq),mz3(nlq),mx4(nlq),my4(nlq),mz4(nlq)
 107       common/aux33loc/
 108      1 ix1(nlq),ix2(nlq),ix3(nlq),ix4(nlq),ixs(nlq,4),mxt(nlq)
 109       common/aux35loc/rhoa(nlq),cxx(nlq),fcl(nlq),fcq(nlq)
 110       common/hourgloc/ymod(nlq),gmod(nlq),ifsv(nlq)
 111       common/soundloc/sndspd(nlq),sndsp(nlq),diagm(nlq),sarea(nlq),
 112      . dxl(nlq)
 113       common/bel6loc/bm(nlq,3,8),bb(nlq,3,8),bs(nlq,2,12),bhg(nlq,4),
 114      1 ex(nlq,3,8),dp0(nlq,3,3),dp1(nlq,3,3),dp2(nlq,3,3),
 115      2 ds(nlq),dhg(nlq,5)
 116 c
 117       common/shlioc/ioshl(60)
 118       common/failuloc/sieu(nlq),fail(nlq),ifaili(nlq)
 119       logical output,slnew
 120       common/csforc/ncs1,ncs2,ncs3,ncs4,ncs5,ncs6,ncs7,ncs8,ncs9,
 121      1 ncs10,ncs11,ncs12,ncs13,ncs14,ncs15,
 122      1 numcsd,csdinc,csdout,output,slnew,future(8)
 123       common/csfsavloc/savfrc(nlq,24),svfail(nlq),ndof,ifail
 124       common/sorterloc/nnc,lczc
 125       common/sorter/znnc,zlczc,
 126      & ns11,ns12,ns13,ns14,ns15,ns16,
 127      & nh11,nh12,nh13,nh14,nh15,nh16,
 128      & nt11,nt12,nt13,nt14,nt15,nt16,
 129      & nb11,nb12,nb13,nb14,nb15,nb16,
 130      & nu11,nu12,nu13,nu14,nu15,nu16,
 131      & nd11,nd12,nd13,nd14,nd15,nd16
 132       common/subtssloc/dt1siz(nlq)
 133       common/matflr/mtfail(200)
 134       common/berwcmloc/xll(nlq),rigx(nlq),rigy(nlq)
 135       common /mem/ mp
 136       integer ia(1)
 137       pointer(mp,ia)
 138       real*4 mmode,ies
 139       dimension e(3,1),f(3,1),qs(9,1),gm(4,*),hgener(*)
 140       dimension qs1(nlq),qs2(nlq),qs3(nlq),qs4(nlq),qs5(nlq)
 141       dimension fibl(9,1),sthick(*),ies(*),hoff(*)
 142 c
 143       ifail=0
 144       if (qhgb+qhgw+qhgm.gt.1.e-04) then
 145       tmode=qhgb*ymod(lft)/1920.0
 146       wmode=qhgw*gmod(lft)/120.00
 147       mmode=qhgm*ymod(lft)/80.000
 148 c
 149       hgfac=rhoa(lft)*sndspd(lft)
 150 c
 151       do i=lft,llt
 152       htxi  =area(i)*(x3(i)-x2(i)-x4(i))
 153       htyi  =area(i)*(y3(i)-y2(i)-y4(i))
 154       gm1(i)= 1.-px1(i)*htxi-py1(i)*htyi
 155       gm2(i)=-1.-px2(i)*htxi-py2(i)*htyi
 156       gm3(i)= 2.-gm1(i)
 157       gm4(i)=-2.-gm2(i)
 158       qhx(i)=gm2(i)*vx2(i)+gm3(i)*vx3(i)+gm4(i)*vx4(i)
 159       qhy(i)=gm2(i)*vy2(i)+gm3(i)*vy3(i)+gm4(i)*vy4(i)
 160       qwz(i)=gm2(i)*vz2(i)+gm3(i)*vz3(i)+gm4(i)*vz4(i)
 161       enddo
 162       do i=lft,llt
 163       c3=      sqrt(abs(sarea(i)))*thick(i)/(dt1siz(i)+1.e-16)
 164       c2=(hgfac*qhgw)*c3
 165       c1=(hgfac*qhgb*.01)*c3*thick(i)*thick(i)
 166       c3=(hgfac*qhgm)*c3
 167       qtx(i)=gm2(i)*vx6(i)+gm3(i)*vx7(i)+gm4(i)*vx8(i)
 168       qty(i)=gm2(i)*vy6(i)+gm3(i)*vy7(i)+gm4(i)*vy8(i)
 169       xll2  =2.*xll(i)
 170       qhxi  =qhx(i)+xll2*rigy(i)
 171       qhyi  =qhy(i)-xll2*rigx(i)
 172       qs1(i)=c3*qhxi
 173       qs2(i)=c3*qhyi
 174       qs3(i)=c2*qwz(i)
 175       qs4(i)=c1*qtx(i)
 176       qs5(i)=c1*qty(i)
 177       enddo
 178 c
 179 c
 180 c
 181       if (isolvr(18).eq.0) then
 182 c
 183       do i=lft,llt
 184       fm11(i)= fm11(i)+gm1(i)*qs4(i)
 185       fm12(i)= fm12(i)+gm1(i)*qs5(i)
 186       fm21(i)= fm21(i)+gm2(i)*qs4(i)
 187       fm22(i)= fm22(i)+gm2(i)*qs5(i)
 188       fm31(i)= fm31(i)+gm3(i)*qs4(i)
 189       fm32(i)= fm32(i)+gm3(i)*qs5(i)
 190       fm41(i)= fm41(i)+gm4(i)*qs4(i)
 191       fm42(i)= fm42(i)+gm4(i)*qs5(i)
 192       enddo
 193 C
 194       else
 195 c
 196       do 45 i=lft,llt
 197       ft31(i)=-ft11(i)+gm3(i)*qs1(i)
 198       ft32(i)=-ft12(i)+gm3(i)*qs2(i)
 199       ft33(i)=-ft13(i)+gm3(i)*qs3(i)
 200       ft41(i)=-ft21(i)+gm4(i)*qs1(i)
 201       ft42(i)=-ft22(i)+gm4(i)*qs2(i)
 202       ft43(i)=-ft23(i)+gm4(i)*qs3(i)
 203       ft11(i)= ft11(i)+gm1(i)*qs1(i)
 204       ft12(i)= ft12(i)+gm1(i)*qs2(i)
 205       ft13(i)= ft13(i)+gm1(i)*qs3(i)
 206       ft21(i)= ft21(i)+gm2(i)*qs1(i)
 207       ft22(i)= ft22(i)+gm2(i)*qs2(i)
 208       ft23(i)= ft23(i)+gm2(i)*qs3(i)
 209       fm11(i)= fm11(i)+gm1(i)*qs4(i)
 210       fm12(i)= fm12(i)+gm1(i)*qs5(i)
 211       fm21(i)= fm21(i)+gm2(i)*qs4(i)
 212       fm22(i)= fm22(i)+gm2(i)*qs5(i)
 213       fm31(i)= fm31(i)+gm3(i)*qs4(i)
 214       fm32(i)= fm32(i)+gm3(i)*qs5(i)
 215       fm41(i)= fm41(i)+gm4(i)*qs4(i)
 216       fm42(i)= fm42(i)+gm4(i)*qs5(i)
 217    45 continue
 218       endif
 219 c
 220       else
 221 c
 222       do 40 i=lft,llt
 223       ft31(i)=-ft11(i)
 224       ft32(i)=-ft12(i)
 225       ft33(i)=-ft13(i)
 226       ft41(i)=-ft21(i)
 227       ft42(i)=-ft22(i)
 228       ft43(i)=-ft23(i)
 229    40 continue
 230       endif
 231 c
 232 c
 233       do i=lft,llt
 234       mz1(i)=gl31(i)*fm11(i)+gl32(i)*fm12(i)
 235       mz2(i)=gl31(i)*fm21(i)+gl32(i)*fm22(i)
 236       fz1(i)=gl31(i)*ft11(i)+gl32(i)*ft12(i)+gl33(i)*ft13(i)
 237       fz2(i)=gl31(i)*ft21(i)+gl32(i)*ft22(i)+gl33(i)*ft23(i)
 238       mz3(i)=gl31(i)*fm31(i)+gl32(i)*fm32(i)
 239       mz4(i)=gl31(i)*fm41(i)+gl32(i)*fm42(i)
 240       fz3(i)=gl31(i)*ft31(i)+gl32(i)*ft32(i)+gl33(i)*ft33(i)
 241       fz4(i)=gl31(i)*ft41(i)+gl32(i)*ft42(i)+gl33(i)*ft43(i)
 242       enddo
 243    90 continue
 244 c
 245       if (output) then
 246       do i=lft,llt
 247       savfrc(i, 1)= fx1(i)
 248       savfrc(i, 2)= fy1(i)
 249       enddo
 250 c
 251       ndof=4
 252       if (ifail.eq.1) then
 253         do i=lft,llt
 254         svfail(i)=fail(i)
 255         enddo
 256       endif
 257       endif
 258 c
 259       return
 260       end
 261
 262 ! { dg-final { scan-tree-dump-times "vectorized 6 loops" 1 "vect" { xfail powerpc*-*-* ia64-*-*-* } } }
 263 ! { dg-final { cleanup-tree-dump "vect" } }