subroutine kscomp (work) implicit double precision (a-h,o-z) common /kscomm/ rdf ,adf ,fdl ,fdm ,rho ,drho ,rhomax, 1 fun0 ,slope ,delx ,alpha ,alpmax,a1 ,a2 , 2 a3 ,a4 ,f1 ,f2 ,f3 ,f4 ,alim , 3 atest ,ftest ,ifscl ,ifoff ,isx ,isx0 ,isxlb , 4 isxub ,iscl ,ig0 ,idf ,islp ,iobj0 ,iy , 5 ip ,ih ,ihess ,iside ,isact ,idobj ,idg , 6 itmp1 ,itmp2 ,inext ,jnext ,jsel ,itcnt ,icntr , 7 icnta ,isdflg,isdrst,ifncl ,nunit ,ndv ,ncon , 8 nobj ,nside ,nscale,iprnt ,itmax ,igrad ,limit dimension work(*) c c copy variables from common block kscomm to work array c c author - Gregory A. Wrenn c location - Lockheed Engineering and Sciences Co. c 144 Research Drive c Hampton, Va. 23666 c c last modification - 19 July 1996 c work(1) = rdf work(2) = adf work(3) = fdl work(4) = fdm work(5) = rho work(6) = drho work(7) = rhomax work(8) = fun0 work(9) = slope work(10) = delx work(11) = alpha work(12) = alpmax work(13) = a1 work(14) = a2 work(15) = a3 work(16) = a4 work(17) = f1 work(18) = f2 work(19) = f3 work(20) = f4 work(21) = alim work(22) = atest work(23) = ftest work(24) = float(ifscl) work(25) = float(ifoff) work(26) = float(isx) work(27) = float(isx0) work(28) = float(isxlb) work(29) = float(isxub) work(30) = float(iscl) work(31) = float(ig0) work(32) = float(idf) work(33) = float(islp) work(34) = float(iobj0) work(35) = float(iy) work(36) = float(ip) work(37) = float(ih) work(38) = float(ihess) work(39) = float(iside) work(40) = float(isact) work(41) = float(idobj) work(42) = float(idg) work(43) = float(itmp1) work(44) = float(itmp2) work(45) = float(inext) work(46) = float(jnext) work(47) = float(jsel) work(48) = float(itcnt) work(49) = float(icntr) work(50) = float(icnta) work(51) = float(isdflg) work(52) = float(isdrst) work(53) = float(ifncl) work(54) = float(nunit) work(55) = float(ndv) work(56) = float(ncon) work(57) = float(nobj) work(58) = float(nside) work(59) = float(nscale) work(60) = float(iprnt) work(61) = float(itmax) work(62) = float(igrad) work(63) = float(limit) c return end