forums.silverfrost.com

[Kevin]

Lahey Computer Systems will permanently close on
December 31, 2022

[Robert]

Haven't they effectively been closed for a while though?

[mecej4]

Yes, Lahey closed down three months ago.

They provided two different compiler products. The first was a superb Windows 32-bit compiler (LF95) from Fujitsu with customisations for use with Visual Studio and Windows libraries. The second was an older version of Gfortran with Visual Studio integrations (LGF). Included in the packages were some rather useful libraries: BLAS, Lapack, Fujitsu SSL2 (all for LF95), and BLAS, Lapack and SLATEC (all for LGF).

Absoft, which closed down in October, also provided prebuilt SLATEC libraries with their Fortran compiler product.

SLATEC is a fine, if dated, mathematical routine library. You can obtain the sources at Netlib (www.netlib.org/slatec). To get an idea of what the library contains, see http://www.lahey.com/docs/lgf14help/LFUGSLATEC.htm .

I spent some time attempting to build a SLATEC DLL for use with FTN95. I found and fixed a number of errors (uninitialised variables, subscript overrun, name clashes between single-precision and double-precision versions of routines and their common blocks). SLATEC is a large library (about 2,000 source files), not easy to use by a new user who is not familiar with F77.

The DLL that I produced is about 6.7 MB, and works with FTN95-64 or Intel Fortran 64-bit. All the 54 test programs included in the SLATEC distributions ran correctly when built with FTN95-64.

If someone is interested, I can upload the DLL to the cloud and provide a link here.

[Robert]

SLATEC sounds interesting, maybe we could make it a download from our main site.

[mecej4]

Robert, even though the SLATEC DLL that I built works fine with FTN95-compiled driver programs, it has one drawback. The DLL was compiled with the Intel OneAPI compiler, and depends on the the Intel RTL package being installed. There is a free download for the RTL, but it is a chore to download, install and configure.

In an attempt to avoid requiring the Intel RTL, I tried to build a DLL using FTN95. I ran into a surprising problem, which I have reported in a separate thread:

http://forums.silverfrost.com/viewtopic.php?p=33842#33842

In short, the SLATEC package has a substantial error reporting subpackage, and the gateway routine, XERMSG, has a namesake already lurking in SALFLIBC.DLL/SALFLIBC64.DLL. If this namesake gets linked in instead of the SLATEC routine, unbeknown to the user, mysterious program crashes occur.

I hope that this problem can be overcome.

[mecej4]

Robert, here is an update on my attempt to build a SLATEC DLL using FTN95 (rather than other compilers, which would drag in their runtime libraries).

Paul has provided me updated DLLs with fixes for two of the three issues in my post http://forums.silverfrost.com/viewtopic.php?t=4746 . I am now able to build a DLL for SLATEC using FTN95. Of the 54 tests in the SLATEC test suite, 53 pass.

The remaining test (TEST04) exercises the Fullerton special function routines (Bessel, Airy, etc., with real or complex arguments), and it fails because of the complex variables/arithmetic bug that I reported (in the same thread as the one quoted above). When Paul fixes that bug I should be able to release my updated SLATEC source files.

Have you decided how you want to test the new SLATEC distribution and make the DLLs (32- and 64-bit) and sources available on www.silverfrost.com? I ask because I probably have to document my fixes so that users can become aware of the changes, although I have tried not to touch the API to the SLATEC routines.

We are all probably aware that the SLATEC source code was placed in the public domain by the US Govt., decades ago.

I am working on preparing a few example programs that use the SLATEC DLL to solve standard numerical analysis problems. Here is one, for solving a nonlinear equation in one variable (for details of the physical derivation, see https://www.diva-portal.org/smash/get/diva2:647481/FULLTEXT01.pdf . Stefan's eq. is eq. (2.57) ).

[mecej4]

The compiler has been updated to fix the bug that caused TEST04 of SLATEC to fail when compiled with /64. Updated SLATEC DLLs, and more examples, are available here:
https://drive.google.com/file/d/10spK3G9RHFcNevrJgDtbBONqyZshJSKs/view?usp=sharing