[PATCH][IA64] Fix speculation support.

[Vladimir Makarov]

Maxim Kuvyrkov wrote:
> Hello,
>
> This patch fixes yet another issue of speculation support in 
> scheduler.  The bug is in initialization of new instructions emitted 
> during scheduling: in some cases we initialize the data for new 
> instructions twice, which causes sanity checks to fail.
>
> The particular case is that a new basic block with a single jump 
> instruction is emitted at the end of the function to make a "pocket" 
> for speculation recovery code.  For that jump instruction we 
> initialize dependency data structures twice: when emitting the 
> instruction and later, when analyzing dependencies for that 
> 1-instruction region.
>
> The patch cleans up extend_global() and extend_all() functions giving 
> them better names and comments.  All uses of these functions are 
> updated to call the appropriate extender.
>
> I don't have access to ia64 box at the moment, so I only checked it on 
> a cross-compiler.  Andrey, can you, possibly, test the patch?
>
> OK for trunk and 4.3 branch?
>
The patch is OK.  Thanks for the patch.