Fortran Programming Language :: About Migrating from F77 to F95


	About Migrating from F77 to F95 Hi all, I used to write f77 codes and now I want to migrate to f95. Since there is a large set of old f77 styled codes in my own lib, I have several questions on the necessities for the migration: 1. if I keep the old codes unchanged and only re-compile them with compilers like ifort, will the efficiency be promoted a lot even without rewriting codes? and in what degree the efficiency will be promoted if I further rewrite all the codes in f95 style and recompile them with ifort? 2. is there any suggestion form you experienced guys about rewriting f77 codes complying with f95 styles? Please leave your comments, thanks in advance Zuying: You should consider trying out Mike Metcalf's convert program which will convert F77 fixed format syntax to f90 free format syntax. It will do other good things (like indent DO groups). Get it at: ftp://ftp.numerical.rl.ac.uk/pub/MandR/convert.f90 compile it and read the leading comments, and perhaps change any default options that you wiish. There are other possible changes you could make manually, like replacing COMMON with modules, declaring subprogram arguments with INTENT clauses. If you have questions about this, post the code and ask this group those questions. Since F77 is a subset of F90 you should be able to compile the F77 code with an F90/95 compiler (like Intel Fortran). Skip Knoble On 30 Apr 2007 04:57:36 -0700, "zuy@gmail.com" <zuy@gmail.com> wrote: -\|Hi all, -\| -\|I used to write f77 codes and now I want to migrate to f95. Since -\|there is a large set of old f77 styled codes in my own lib, I have -\|several questions on the necessities for the migration: -\| -\|1. if I keep the old codes unchanged and only re-compile them with -\|compilers like ifort, will the efficiency be promoted a lot even -\|without rewriting codes? and in what degree the efficiency will be -\|promoted if I further rewrite all the codes in f95 style and recompile -\|them with ifort? -\| -\|2. is there any suggestion form you experienced guys about rewriting -\|f77 codes complying with f95 styles? -\| -\|Please leave your comments, thanks in advance <zuy@gmail.com> wrote in message news:1177934256.001763.323830@o5g2000hsb.googlegroups.com... > 1. if I keep the old codes unchanged and only re-compile them with > compilers like ifort, will the efficiency be promoted a lot even > without rewriting codes? As you don't specify the compiler you're using for F77 it's hard to say, but it's unlikely that you'd see any material change. > and in what degree the efficiency will be > promoted if I further rewrite all the codes in f95 style and recompile > them with ifort? If it ain't broke, don't fix it. > 2. is there any suggestion from you experienced guys about rewriting > f77 codes complying with f95 styles? It can be a pain to mix Fortran 77 and Fortran 95 source forms and styles. Thus, it would be useful to make these minimal changes to your existing code: a) convert the fixed-form source to free-form source using a tool such as convert.90 (which will optionally add procedure names to the procedure end statements); b) organize your old code into modules so that the interfaces become explicit (which requires procedure names on the end statements). When that all works you might consider replacing fixed storage with dynamic storage, common with module variables etc. (but see comment above). HTH Mike Metcalf On Apr 30, 7:57 am, "zuy@gmail.com" <zuy@gmail.com> wrote: > Hi all, > I used to write f77 codes and now I want to migrate to f95. Since > there is a large set of old f77 styled codes in my own lib, I have > several questions on the necessities for the migration: > 1. if I keep the old codes unchanged and only re-compile them with > compilers like ifort, will the efficiency be promoted a lot even > without rewriting codes? That depends on what Fortran 77 compiler you have been using. There are some old benchmarks at http://www.polyhedron.com/pb04/win32/f77bench_p4.html showing Intel Visual Fortran to be faster than g77. > and in what degree the efficiency will be > promoted if I further rewrite all the codes in f95 style and recompile > them with ifort? I would NOT expect rewriting F77 codes in F95 style (modules, array operations, assumed shape arrays etc.) to increase the speed of a program, and it could make it slower. Fortran compilers are good at optimizing F77, based on a lot of experience. I have read that the speed penalty of using some of the modern features of Fortran has fallen since Fortran 90 compilers were first introduced. > 2. is there any suggestion form you experienced guys about rewriting > f77 codes complying with f95 styles? You can use Michael Metcalf's convert.f90 program at http://www.slac.stanford.edu/comp/fortran/convert.f90 to convert fixed to free source form. I think free source form is more readable and less error-prone. Alan Miller's to_f90.f90 program http://users.bigpond.net.au/amiller/to_f90.f90 is more ambitious and attempts to add some F90 features such as INTENT. Note the caveats under "Known problems". In general, spending your time on writing new code in F95 to do new things may be more productive than making stylistic changes to old code, unless you are going to modify the old code anyway to add functionality. Writing Fortran 95 wrappers for Fortran 77 code can be productive. Lapack95 http://www.netlib.org/lapack95/ is an example of this. zuy@gmail.com wrote: > Hi all, > I used to write f77 codes and now I want to migrate to f95. Since > there is a large set of old f77 styled codes in my own lib, I have > several questions on the necessities for the migration: > 1. if I keep the old codes unchanged and only re-compile them with > compilers like ifort, will the efficiency be promoted a lot even > without rewriting codes? and in what degree the efficiency will be > promoted if I further rewrite all the codes in f95 style and recompile > them with ifort? As others said, you ask for a comparison between an unknown compiler and an unknown version of ifort, so no answer makes much sense. ifort, particularly in the latest versions, is exceptionally good at vectorizing loops with many assignments, provided they meet the qualifications. It also necessarily works harder at vectorizing double precision than certain other compilers. ifort will gain efficiency from modern syntax only in those cases where your rewrite forces you to make better optimizable source code. ifort for 32-bit and 32-bit extended 64-bit depends heavily on vectorization of stride 1 arrays as its primary means for optimization. ifort still tends to be weak in optimization of array syntax with non-default stride. ifort is just beginning to address the issues of fusion for efficiency for the recent CPUs which care about it. So, you might not gain as much from such a move if you don't intend to keep your compiler up to date. gfortran is doing well in optimization of f95 syntax, short of dealing with fusion and the like. Michael Metcalf <michaelmetc@compuserve.com> wrote: > <zuy@gmail.com> wrote in message > news:1177934256.001763.323830@o5g2000hsb.googlegroups.com... > > and in what degree the efficiency will be > > promoted if I further rewrite all the codes in f95 style and recompile > > them with ifort? > If it ain't broke, don't fix it. In particular, if you do this naively, it might well hurt performance. Getting good performance is a bit of an art in any language. It cannot be achieved by just writing on one or another style changing to an f95 is not a "magic bullet" for performance. If you want to do some changes for other reasons, that's one thing. But if you convert to an f95 style solely in the hopes of getting better performance, without understanding what particular things you expect to improve performance and why, then you will likely be disappointed. -- Richard Maine \| Good judgement comes from experience; email: last name at domain . net \| experience comes from bad judgement. domain: summertriangle \| -- Mark Twain Thank you all, Then I gonna stay with my previous codes and only write f95 for new codes. On Apr 30, 11:29 pm, nos@see.signature (Richard Maine) wrote: - Hide quoted text - > Michael Metcalf <michaelmetc@compuserve.com> wrote: > > <zuy@gmail.com> wrote in message > >news:1177934256.001763.323830@o5g2000hsb.googlegroups.com... > > > and in what degree the efficiency will be > > > promoted if I further rewrite all the codes in f95 style and recompile > > > them with ifort? > > If it ain't broke, don't fix it. > In particular, if you do this naively, it might well hurt performance. > Getting good performance is a bit of an art in any language. It cannot > be achieved by just writing on one or another style changing to an f95 > is not a "magic bullet" for performance. If you want to do some changes > for other reasons, that's one thing. But if you convert to an f95 style > solely in the hopes of getting better performance, without understanding > what particular things you expect to improve performance and why, then > you will likely be disappointed. > -- > Richard Maine \| Good judgement comes from experience; > email: last name at domain . net \| experience comes from bad judgement. > domain: summertriangle \| -- Mark Twain In article <1177934256.001763.323@o5g2000hsb.googlegroups.com>, zuy@gmail.com <zuy@gmail.com> wrote: >Hi all, >2. is there any suggestion form you experienced guys about rewriting >f77 codes complying with f95 styles? You may find my notes on Fortran 95 for Fortran 77 users helpful. See http://www.mcs.vuw.ac.nz/math/papers/JFH10_f95_2006 -- John Harper, School of Mathematics, Statistics and Computer Science, Victoria University, PO Box 600, Wellington 6140, New Zealand e-mail john.har@vuw.ac.nz phone (+64)(4)463 5341 fax (+64)(4)463 5045