Patch fixing 3.3 bug PR 9745 and PR 10021

[Franz Sirl]

At 04:12 10.07.2003, Jim Wilson wrote:
>Marcus Meissner wrote:
>>>The loop patch attached to your mail compiled with -O2 does not change
>>>the assembler output and does not fix the testcase in 10021.
>>Err, I mean the 9745 testcase, sorry.
>
>This contradicts what I have seen with a cross-compiler, and what David 
>Edelsohn has reported.  However, I should point out that there are 3 
>testcases in 9745.  Maybe my patch doesn't fix all of them?  Which one(s) 
>did you try?  Make sure you don't have any other patches applied that 
>might skew the results, in particular my earlier alias.c patch.

I can confirm that your patch fixes libmcrypt, where 'make check' now 
passes. What isn't fixed is the minimized testcase pr11319.c. But even 
compiling this testcase with -fno-strict-aliasing still lets it fail, so I 
rather think your patch uncovers YALB (yet another loop bug)?

Franz.