Bug report about GMP Kernel

[Franck MICHEL <Franck.Michel@univ.u-3mrs.fr>]

I'm doing computations with series of floating point complex coefficients,
4500 digits. I've installed Pari 2.2.6 with GMP Kernel and I'm very
satisfied by its  speed (about 3 times faster than Pari 2.1.0)

With simple programs, all works fine.
But with more complicated programs, the values obtained are wrong.
Unfortunately, I wasn't able to identify a simple operation (division,
power...) generating a false result. Therefore, I've simply shortened a
program to keep the first wrong result obtained.

The program uses relatively large data files in input, so I preferred to
use a web page for the bug report, rather than to send it by mail. You'll
find at

http://www.medicis.polytechnique.fr/~fmichel/buggpgmp.html

the program, the result with Pari 2.1.0 (which seems to be correct) and the
result with Pari 2.2.6 with GMP Kernel (which is wrong). More precisely:
the program computes a matrix vum[i,j] which is correct as long as i<=9,
and which is false if i=10.

Thanks in advance for your attention.
 
 Franck Michel