llvm dev - Dec 2004 - [LLVMdev] JIT stubs for functions with external linkage

Hi,

Before I go ahead and remove the generation of stubs for external 
functions, is there any good reason why they get generated? I can 
understand that the JIT wants to make stubs for the functions it 
compiles itself - it enables selective recompilation and replacement - 
but functions which are defined in a library (or in my case by the 
application using the JIT) can't be changed anyway. For me the extra jmp 
instruction for each call is significant, so I want to remove it...

m.

On Fri, 3 Dec 2004, Morten Ofstad wrote:
> Before I go ahead and remove the generation of stubs for external
> functions, is there any good reason why they get generated?
It depends on the target.  On PPC, for example, calls have limited
displacement.  By using stubs, we can ensure it's possible to reach the
external function.

They should not be generated on X86.
> I can understand that the JIT wants to make stubs for the functions it
> compiles itself - it enables selective recompilation and replacement -
And lazy compilation.
> but functions which are defined in a library (or in my case by the
> application using the JIT) can't be changed anyway. For me the extra
jmp
> instruction for each call is significant, so I want to remove it...
Sounds good.

-Chris

-- 
http://nondot.org/sabre/
http://llvm.cs.uiuc.edu/