jlg@a.UUCP (Jim Giles) (03/05/88)
> The criterion for Fortran 8x features (in my opinion) should be to > certify existing practice or to extend the language in a significant > way. I will use this criterion throughout this discussion. Generalized precision is not a feature that should be included in Fortran. No Fortran compiler I'm aware of uses this feature in any form, so the 'existing practice' test fails. Generalized precision does not add any new functionality either. Most implementations will continue to support only SINGLE and DOUBLE precision and will map the generalized precision variables to one or the other. This is true because most hardware will continue to possess only one or two precisions, and implementors will not wish to build more than one extended format in software. The use of generalized precision will only lead to confusion among unsophisticated users as to the inconsistent behaviour of their codes (with respect to speed) when they port between machines. The new intrinsic procedures for getting environmental information about the arithmetic should be left in the new standard. This is really an example of certifying existing practice, since I've never seen a Fortran support library without some 'machine epsilon' type calls available. J. Giles Los Alamos
reeder@ut-emx.UUCP (William P. Reeder) (03/08/88)
In article <507@a.UUCP>, jlg@a.UUCP (Jim Giles) writes: > Generalized precision is not a feature that should be included in Fortran. > No Fortran compiler I'm aware of uses this feature in any form, so the > 'existing practice' test fails. > I agree whole-heartedly!!! > Most > implementations will continue to support only SINGLE and DOUBLE precision > and will map the generalized precision variables to one or the other. Actually, S8 says that the compiler should use the physical format which meets exactly or exceeds by the least amount the qualities requested by the user for floating point quantities. If no such format exists, what happens is the infamous "processor dependent". (see page 4-3, section 4.3.1.2, lines 20 - 28) I also agree with what Jim says about portability problems with generalized precision -- many others have also brought them up with different examples. My comments to X3 (throught the University of Texas) also mentions this problem. I hope those X3J3 members who are reading this group recognize this as a serious problem. > J. Giles > Los Alamos -- William {Wills,Card,Weekly,Virtual} Reeder reeder@emx.utexas.edu The Looniversity of TexMex at Autism, Consternation Central, Austin TX 78712 DISCLAIMER: I speak only for myself, and usually only to myself.