Problem with DTGSYL

[Tobias]

Hello Lapack Forum,

I am currently implementing an algorithm for solving the projected generalized lyapunov equation. This algorithm uses the LAPACK routine DTGSYL for solving the generalized Sylvester equation. While testing my code I noticed an error in DTGSYL which only occurs when I extend the "leading dimension" - parameters. Since I work in my University I cannot recompile my LAPACK/BLAS library without optimization (I am not sure if it is compiled with optimization anyway) and thus I am not sure whether it is an compiler optimization error or not. I wrote this short test example in pure Fortran77 code, which should compile pretty easy. Could someone who definitely has unoptimized LAPACK/BALS please download

http://www.math.tu-berlin.de/~bruell/dtgsyl_fail.zip

and follow the (very) short instructions in the file README.txt and then tell me if this is an error in LAPACK or not.

Greetings,
Tobias[/url]

[sven]

Dear Tobias,

Thank you for your zip file, much appreciated. It enabled me to find a bug in DTGSY2. If you send me an email to 'sven at nag.co.uk' then I can send you the fixed version. (Which gives me the correct results for your program.)

Best wishes,

Sven Hammarling.

[Julie]

Thank you Tobias for pointing that out.
The correct version of STGSY2 and DTGSY2 can be found at http://www.netlib.org/lapack/release_notes.html
They will be included in the next "bug fix" release of LAPACK.
Sincerely
Julie