[llvm-commits] [llvm] r67142 - in /llvm/trunk: lib/Target/X86/X86Instr64bit.td test/CodeGen/X86/call-imm.ll

[Chris Lattner]

On Mar 17, 2009, at 7:03 PM, Evan Cheng wrote:

> llvm only supports small code model fully. Medium and large code
> models are not fully baked. Perhaps this just need a
> Require<[SmallCode]>. It's true the call destination displacement can
> be beyond 4G. But in small code model that's undefined behavior
> anyway, right?

Why does "call <some random address>" imply that the address is in the  
text segment?  I don't see what mcmodel has to do with this, nor do I  
understand why the optimization is that important.  If we cared a lot  
about this for the JIT, then the client of the JIT should output an  
llvm::Function prototype object, then use EE->addGlobalMapping to set  
its address to the desired location.

-Chris

>
>
> Evan
>
> On Mar 17, 2009, at 6:31 PM, Mike Stump wrote:
>
>> On Mar 17, 2009, at 5:43 PM, Chris Lattner wrote:
>>> Disable the "call to immediate" optimization on x86-64.
>>
>> I think it would be valid for any model other than -mcmodel=large, if
>> people wanted to have -mcmodel support.
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits