Variable size problem (NCOL>=34) with DCDIST()

[jianni]

Hello,

I am using the IMSL version that came with Compaq Fortran Version 6.6 with the same compiler and Visual Studio 6.0. I am having problems running DCDIST() ( REAL*8 version of CDIST() ) with 34 or more variables.
Everything works fine with 33 or less variables (NCOL<34) but when I use the subroutine with 34 or more DCDIST() hangs. I get no errors back. I have to press control-C to stop it. I have the reserve stack at 250 megabytes (for both Debug and release) so I don't think that's the problem. I tried to read any info/warns/errors that may be returned with

CALL ERSET (0, 1, -1)

but, again, the system hangs at the DCDIST() call. I've been using many IMSL rountines the past several years and this is the first strange hang bug I've encountered. Is it a bug?

[ed]

The problem was actually with an underlying routine where there was a limit that involved the product of NCOL and NROW. Specifically, if (NRA*NCA) is greater than IMACH(5)/NRA or IMACH(5)/NCA then the algorithm will fail due to an integer overflow.

This issue was identified in IMSL Fortran v4 and fixed for future releases.

[jianni]

Ed,

Are there any patches I can apply for the CDIST routine?
I hope I don't have to re-purchase the entire Fortran IMSL library as that would really be laime.

-james


BTW

I see you have the Eagles logo? I'm from West Philly. Eagles Rock, glad to see TO off to Dallas...

[jianni]

The problem was actually with an underlying routine where there was a limit that involved the product of NCOL and NROW. Specifically, if (NRA*NCA) is greater than IMACH(5)/NRA or IMACH(5)/NCA then the algorithm will fail due to an integer overflow.

This issue was identified in IMSL Fortran v4 and fixed for future releases.

I have only about 60 something rows and 36 columns, but my Leading dimension on the DIST matrix is 8001.
I'm assuming IMACH(5) is INT*4 equals about 2e6
so 2e6/8001=2135 which is about where my error is.
So my question is: do you mean when NRA*NCA is greater than IMACH(5)/LDX, IMSL will hang?

[ed]

Unfortunately, there is no patch fix available. The issue was with an underlying subroutine that affected 3 or 4 functions and was only fixed with the release of IMSL Fortran v5.

From reading notes in the database, if NCA*NRA*NCA or NCA*NRA*NRA is greater than about 2e6, the underlying function hangs. The fix was to change a variable from INT*4 to INT*8.

And yes, I've been an Eagles fan for 20 years or so through all those thin years (I grew up in Delaware). And it was a great day to see TO go to Dallas, but I fear those 2 games next year. It would have been better to see him go to the AFC....