The problem is that the Linux assembler (2.9-ia64-000216) apparently does
not support the mnemonics
used with the fclass instructions (@inf, @pos, @neg, @zero, @qnan, @snan,
@norm, @unorm).  
When we replace these mnemonics with their hexadecimal equivalents all is
well. A test case is
attached as shar file:

 <<problem.report>> 
Can the IA-64 assembler support folks let me know if this is a known problem
or if  they need any 
further information?

Thanks,
Shane Story ( 503- 696-3229)