|
|
 |
 |
 |
 |
Fortran Programming Language
|
 |
 |
 |
 |
 |
 |
 |
 |
Different result appear after codes compiled to DLL in DF6
hello, every one: A strange phenomena puzzled me, when i develop an algorithm software using Digital Visual Fortran 6(windows 2K). The outcoming shoule be a DLL file. At the first stage, i built a fortran windows console project for the conveniece of debuging. After I finished all the code, I checked it very carefully and made sure it is correct. Then I tried to build the last DLL file. However, when i got DLL file and tested it through the same data used in the first stage, I found its output is different slightly to those got by console project. Since the software is very important, these slight difference should not be ignored. Both console project and DLL project do not apply any optimizaion(I use /optimise:0 option to prevent optimising) and all other options keeping default. What's the matter resulting in those differences? Are there some operation affecting the computing output when build DLL using DF6? Thanks for your attention and waiting for your advice.
hello, every one: A strange phenomena puzzled me, when i develop an algorithm software using Digital Visual Fortran 6(windows 2K). The outcoming shoule be a DLL file. At the first stage, i built a fortran windows console project for the conveniece of debuging. After I finished all the code, I checked it very carefully and made sure it is correct. Then I tried to build the last DLL file. However, when i got DLL file and tested it through the same data used in the first stage, I found its output is different slightly to those got by console project. Since the software is very important, these slight difference should not be ignored. Both console project and DLL project do not apply any optimizaion(I use /optimise:0 option to prevent optimising) and all other options keeping default. What's the matter resulting in those differences? Are there some operation affecting the computing output when build DLL using DF6? Thanks for your attention and waiting for your advice.
why I cannot find my post after submit? hello, every one: A strange phenomena puzzled me, when i develop an algorithm software using Digital Visual Fortran 6(windows 2K). The outcoming shoule be a DLL file. At the first stage, i built a fortran windows console project for the conveniece of debuging. After I finished all the code, I checked it very carefully and made sure it is correct. Then I tried to build the last DLL file. However, when i got DLL file and tested it through the same data used in the first stage, I found its output is different slightly to those got by console project. Since the software is very important, these slight difference should not be ignored. Both console project and DLL project do not apply any optimizaion(I use /optimise:0 option to prevent optimising) and all other options keeping default. What's the matter resulting in those differences? Are there some operation affecting the computing output when build DLL using DF6? Thanks for your attention and waiting for your advice.
hello, every one: A strange phenomena puzzled me, when i develop an algorithm software using Digital Visual Fortran 6(windows 2K). The outcoming shoule be a DLL file. At the first stage, i built a fortran windows console project for the conveniece of debuging. After I finished all the code, I checked it very carefully and made sure it is correct. Then I tried to build the last DLL file. However, when i got DLL file and tested it through the same data used in the first stage, I found its output is different slightly to those got by console project. Since the software is very important, these slight difference should not be ignored. Both console project and DLL project do not apply any optimizaion(I use /optimise:0 option to prevent optimising) and all other options keeping default. What's the matter resulting in those differences? Are there some operation affecting the computing output when build DLL using DF6? Thanks for your attention and waiting for your advice.
dusheng @supcon.com wrote: > hello, every one: > A strange phenomena puzzled me, when i develop an algorithm > software using Digital Visual Fortran 6(windows 2K). The outcoming > shoule be a DLL file. At the first stage, i built a fortran windows > console project for the conveniece of debuging. After I finished all > the code, I checked it very carefully and made sure it is correct. > Then I tried to build the last DLL file. However, when i got DLL file > and tested it through the same data used in the first stage, I found > its output is different slightly to those got by console project. > Since the software is very important, these slight difference should > not be ignored. > Both console project and DLL project do not apply any optimizaion(I > use /optimise:0 option to prevent optimising) and all other options > keeping default. What's the matter resulting in those differences? > Are there some operation affecting the computing output when build DLL > using DF6? > Thanks for your attention and waiting for your advice.
What sort of software is it? You might expect very small differences due to different intermediate storage methods of floating point values between your two programs. But, if you've got decent algorithms, these should be well down in the noise and not causing significant differences. More significant differences could be caused if, for instance, you're using a variable whose value has never been set, and in the two programs it contains different random values. Catherine. -- Catherine Rees Lay Polyhedron Software Ltd. Registered Office: Linden House, 93 High St, Standlake, Witney, OX29 7RH, United Kingdom. Registered in England No.2541693. Vat Reg No. GB 537 3214 57
dusheng @supcon.com wrote: > why I cannot find my post after submit? You posted through Google Groups, which is having another one of its occasional meltdowns where it doesn't update for days. Steve
On Apr 25, 3:52 am, dusheng@supcon.com wrote: > why I cannot find my post after submit?
1. Looking at the original version of your message, it appears that you have posted through google groups. There is a time delay between when you submit a message and when it appears. In fact Google groups probably posts messages faster than actual net news does. (In the old days, messages propagated from system to system via newsfeeds. This took time.) 2. You have posted your original message 4 times. <frown>
> hello, every one: > A strange phenomena puzzled me, when i develop an algorithm > software using Digital Visual Fortran 6(windows 2K). The outcoming > shoule be a DLL file. At the first stage, i built a fortran windows > console project for the conveniece of debuging. After I finished all > the code, I checked it very carefully and made sure it is correct. > Then I tried to build the last DLL file. However, when i got DLL file > and tested it through the same data used in the first stage, I found > its output is different slightly to those got by console project. > Since the software is very important, these slight difference should > not be ignored. > Both console project and DLL project do not apply any optimizaion(I > use /optimise:0 option to prevent optimising) and all other options > keeping default. What's the matter resulting in those differences? > Are there some operation affecting the computing output when build DLL > using DF6? > Thanks for your attention and waiting for your advice.
Maybe you are seeing an effect of the x86 floating point processor architecture. It might be that some intermediate results are being stored to memory (at reduced precision) instead of being kept inside the FPP. See the recent discussion in this newsgroup about whether this behavior (observed in some GNU compilers) is actually a bug at all. Perhaps your algorithm is numericaly unstable. IIRC, this bug was seen in a C program which essentially compared two floating point numbers for equality. In almost all cases, this is not a good idea at all. -- e-mail: epc8 at juno dot com -- elliot
"e p chandler" <e@juno.com> wrote in message news:1177561834.363990.110280@r35g2000prh.googlegroups.com... > On Apr 25, 3:52 am, dusheng @supcon.com wrote: >> why I cannot find my post after submit? > 1. Looking at the original version of your message, it appears that > you have posted through google groups. There is a time delay between > when you submit a message and when it appears. In fact Google groups > probably posts messages faster than actual net news does. (In the old > days, messages propagated from system to system via newsfeeds. This > took time.)
I post both through google and with giganews. Although there is usually a considerable lag when posts go through google, they do show up. It has also occured that giganews has swallowed a post where I was OP. I still only see the replies. Because, for example, the google portal is free and accessible from, say, libraries where *so many* people come to use the net, we see a different type of person than those who post from their offices. They are less likely to be usenet savvy.
> 2. You have posted your original message 4 times. <frown> >> hello, every one: >> A strange phenomena puzzled me, when i develop an algorithm >> software using Digital Visual Fortran 6(windows 2K). The outcoming >> shoule be a DLL file. At the first stage, i built a fortran windows >> console project for the conveniece of debuging. After I finished all >> the code, I checked it very carefully and made sure it is correct. >> Then I tried to build the last DLL file. However, when i got DLL file >> and tested it through the same data used in the first stage, I found >> its output is different slightly to those got by console project. >> Since the software is very important, these slight difference should >> not be ignored. >> Both console project and DLL project do not apply any optimizaion(I >> use /optimise:0 option to prevent optimising) and all other options >> keeping default. What's the matter resulting in those differences? >> Are there some operation affecting the computing output when build DLL >> using DF6? >> Thanks for your attention and waiting for your advice. > Maybe you are seeing an effect of the x86 floating point processor > architecture. It might be that some intermediate results are being > stored to memory (at reduced precision) instead of being kept inside > the FPP. See the recent discussion in this newsgroup about whether > this behavior (observed in some GNU compilers) is actually a bug at > all. Perhaps your algorithm is numericaly unstable. > IIRC, this bug was seen in a C program which essentially compared two > floating point numbers for equality. In almost all cases, this is not > a good idea at all.
I think he's using an IDE that he doesn't have configured correctly. -- WW
|
 |
 |
 |
 |
|