Software Archive
Read-only legacy content
17061 Discussions

Try the CVF 6.5A update

Steven_L_Intel1
Employee
1,251 Views
I'm giving forum readers first crack at trying the CVF 6.5A update (for X86 right now - Alpha version will follow). The software itself is "frozen", this is just to make sure that the update process itself works for most people. Naturally, this is only for people already running CVF 6.5.

Download and run CVF-X86-65A.exe Please report any problems here, or let me know that it worked ok. After installing, try rebuilding your application and viewing the online documentation.

This update is suitable for Standard and Pro edition users, and will update Pro-only bits if you have them.

If all goes well, I'll release the final updates next week.

Steve
0 Kudos
32 Replies
Steven_L_Intel1
Employee
348 Views
Jack,

You didn't send source with your problem report.

The upgrades don't touch the start menu. I agree they should.

Looks like I never had the upgrade replace BUGREPRT.TXT. The old address still works.

Steve
0 Kudos
jackosullivan
Beginner
348 Views
Steve,
Thanks so much for the speedy vf-support solution/workaround to specify STATUS='SCRATCH' on that OPEN() problem.
Strange I was the only one who noticed -- maybe everyone else is too professional to leave anything unspecified any more, but I'm just an old engineer who happens to need to program to get my work done. The "old" is getting more sincere all the time: been at this computing thing for 35 years, now .. and at this same simulator program for 34 of those.

Thanks again for your great support, as usual...and of course,
God bless!
jack
Jack
0 Kudos
Steven_L_Intel1
Employee
348 Views
The Alpha version of the update is now available.

  1. Create a temporary folder and download CVF-ALPHA-65A.exe into that folder.

  2. Run CVF-ALPHA-65A.exe to unpack the files.
  3. From a command prompt in that folder, use the command "upgrade -l" to start the update.


You can delete the temporary folder afterwards if desired.

As usual, let me know if there are any problems. This will get announced on our web site in the next day or two.

Steve
0 Kudos
Intel_C_Intel
Employee
348 Views
Couldn't find it in ftp.compaq.com/pub/products/fortran/vf nor in the supp directory.
0 Kudos
Steven_L_Intel1
Employee
348 Views
Grumble! Sorry about that - will see what's going wrong with the FTP site publishing process. This has usually worked well. I'll respond here later when it's really there.

Steve
0 Kudos
Steven_L_Intel1
Employee
348 Views
Ok - problem fixed - file is there now. Enjoy!

Steve
0 Kudos
Intel_C_Intel
Employee
348 Views
When compiling a free form (.f90) source file that uses tabs, I get the following warning message:
Warning: Extension to FORTRAN-90: tab formatting

Is this supposed to be a warning condition? The documentation would not lead me to believe so.
Is there a work around for suppressing this specific message? The other occasional warnings are lost in a sea of the tab formatting warnings.

Thanks for your consideration
0 Kudos
Steven_L_Intel1
Employee
348 Views
If you ask for Fortran 90 standards checking, then a tab character in the source IS an extension, though the message it is giving is not the one I would have chosen. A tab in the statement part, not part of a comment or character literal, IS an extension.

If you don't like it, turn off standards checking or don't use tabs.

Steve
0 Kudos
Intel_C_Intel
Employee
348 Views
OK, I got it. First time the earliest version of the Alpha downlaod didn't do anything funny. Looks like you fixed more instances of unnecessary copies of contiguous arrays; copytest1.f90 runs a little better than 5 orders of magnitude faster now. Still misses cases where an array section with bounds unknown at compile time is passed or when a pointer component of a structure is passed. Still doesn't optimize tensor contractions to anywhere near the extent I would have liked (scheme.f90). Sad that you didn't get it to optimize right logical shifts (perftest.f90).
 
"And now that it's all over 
All I can do is wish you well." 
              - B. B. King 
0 Kudos
Steven_L_Intel1
Employee
348 Views
James,

The shift optimization you suggested has been added to the next "generation" of our code generator, but as you know, it won't be seen on Alpha NT. We've JUST finished work on eliminating unnecessary copies of derived type components and some more cases, though we know of a few others not done yet. Your suggestions and examples have been very helpful - we appreciate your taking the time to send them to us.

Steve
0 Kudos
davidgraham
Beginner
348 Views
Steve,
I haven't seen any posting about what is in the 6.5A update.
I would like to know what is in the update before updating.
Thanks,
David
0 Kudos
Steven_L_Intel1
Employee
348 Views
It's primarily bug fixes, with some performance improvements (especially for Athlon). If you have the Professional Edition, the COM Server Wizard can now create EXE servers. Here's the gist of it from the release notes:


  • The Fortran COM Server now allows you to select an out-of-process
    (.EXE) COM server. Users now have a choice of the type of server:
    either a DLL (in-process) COM server or an EXE (out-of-process) COM
    server.
  • New compiler options and their descriptions:


    • /assume:[no]protect_constants
      This option specifies whether
      constant actual arguments can be changed. By default, actual arguments
      that are constants are read-only (/assume:protect_constants). To allow
      changes to actual arguments that are constants, specify
      /assume:noprotect_constants.
    • /check:[no]arg_temp_created
      Requests that a run-time
      informational message appear if actual arguments are copied into
      temporary storage before routine calls. Specifying
      /check:noarg_temp_created (the default) does not generate the extra
      code needed to display a message run-time when this condition occurs.
    • /warn:[no]ignore_loc
      Requests that the compiler issues
      warnings when %LOC is stripped from an argument. The default is
      /warn:noignore_loc (does not issue a warning for this condition).

  • The hardware selected by the /ARCH command line option is the basis
    for the /TUNE option, i.e., you cannot use /TUNE to tune generated code
    for an architecture older than the one selected by /ARCH.
  • New data ATTRIBUTES directive keywords are now available:

    ALLOW_NULL, DECORATE, IGNORE_LOC, and NOMIXED_STR_LEN_ARG


I'll get the release notes up on the web site soon.

Steve
0 Kudos
Reply