[PATCH] Address Space Casting

[Michele Scandale]

  Someone can confirm me that is not a problem the change in the LLVMOpcode enumeration of C bindings?


================
Comment at: lib/Target/X86/X86ISelLowering.h:766
@@ -765,1 +765,3 @@
 
+    virtual bool isNoopAddrSpaceCast(unsigned SrcAS, unsigned DestAS) const;
+
----------------
Matt Arsenault wrote:
> Michele Scandale wrote:
> > Matt Arsenault wrote:
> > > LLVM_OVERRIDE
> > This is not used on other similar cases... I just kept the local style.
> It's not really a style thing. It prevents accidentally doing the wrong thing, like if you typoed it as virtual isNopoAddrSpaceCast, it wouldn't actually do what you expect and this catches that. Not enough places use it yet.
> 
> Also do you have a test for this x86 part?
> 
I have overriden this hook in order to have the 'test/CodeGen/X86/atomic-dagsched.ll' working correctly. This simply makes explicit the assumption that address spaces < 256 should be managed as address space zero, so the addrspace cast between these is a no-op cast. I haven't other tests.


http://llvm-reviews.chandlerc.com/D1401