[llvm-commits] Win64 pseudo instruction patch

[Cameron Esfahani]

Any new comments?

On Aug 13, 2010, at 4:44 PM, Cameron Esfahani wrote:

> New patch, based on svn 111057:
> 
> <pseudo2.patch>
> 
>> Hello, Cameron
>> 
>> > Attached is a patch which fixes some more of the Win64 ABI code gen.  Added support for the Windows 64 pseudo instructions like WINTAILJMPd64 and WINCALL64pcrel32.
>> >
>> > Also, updated a little bit of the coff-dump.py tool to dump the AMD64 relocations.  Updated COFF.h to list the AMD64 relocations.
>> >
>> > Still a bit more to do before Win64 COFFs work...
>> Please next time submit unrelated changes (e.g. MC changes vs tail
>> call stuff changes) as separate patches for the sake of easier review.
>> 
>> >+  bool AfterFPPop = Opc == X86::TAILJMPm64 || Opc == X86::TAILJMPm ||
>> >+                    Opc == X86::WINTAILJMPm64;
>> Is it possible to factor out the predicates of such sort into some
>> external predicate function? Something like "IsTAILJUMPMemOpcode()" or
>> whatever?
>> 
> Done.  I've created some helper routines to manage the different flavors of the TAILJUMP AND TCRETURN pseudo opcodes.
> 
>> >-  } else if (Subtarget->isTargetWin64()) {
>> >-    // We need to always allocate 32 bytes as register spill area.
>> >-    // FIXME: We might reuse these 32 bytes for leaf functions.
>> >-    StackSize += 32;
>> >-    MFI->setStackSize(StackSize);
>> >   }
>> As I already mentioned, this part is invalid. Consider the following
>> case: you have two functions A and B with A calling B. B does regs
>> dump into register save area. Since you haven't allocated the space on
>> stack for register save area then B will clobber 16 bytes on local
>> frame of A. This is definitely not what you will want.
>> 
> Ok, I've backed this out.  There were some weird things I saw when generating Win64 code, which is why I took it out in the first place.  I'll try and figure it out in the future.
> 
>> >-             RetOpcode == X86::TCRETURNmi64) {
>> >-    bool isMem = RetOpcode == X86::TCRETURNmi || RetOpcode == X86::TCRETURNmi64;
>> >+             RetOpcode == X86::TCRETURNmi64 ||
>> >+             RetOpcode == X86::WINTCRETURNri64 || RetOpcode == X86::WINTCRETURNdi64 ||
>> >+             RetOpcode == X86::WINTCRETURNmi64) {
>> >+    bool isMem = RetOpcode == X86::TCRETURNmi || RetOpcode == X86::TCRETURNmi64 ||
>> >+                 RetOpcode == X86::WINTCRETURNmi64;
>> Same as above.
>> 
>> > +      if (RetOpcode == X86::TCRETURNdi)
>> > +        Opcode = X86::TAILJMPd;
>> > +      else
>> > +        Opcode = (IsWin64 ? X86::WINTAILJMPd64 : X86::TAILJMPd64);
>> Also deserves some helper for such mapping? E.g. "TailCallIMapImmOpcode" ?
>> 
> This should be resolved with the new helper routines.
> 
>> -- 
>> With best regards, Anton Korobeynikov
>> Faculty of Mathematics and Mechanics, Saint Petersburg State University
> On Aug 12, 2010, at 6:47 PM, Cameron Esfahani wrote:
> 
>> New patch, based on SVN revision 110995:
>> 
>> <pseudo.patch>
> 
> Cameron Esfahani
> dirty at apple.com
> 
> "All that is necessary for the triumph of evil is that good men do nothing."
> 
> Edmund Burke
> 
> 
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits

Cameron Esfahani
dirty at apple.com

"It is the spirit and not the form of law that keeps justice alive."

Earl Warren



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20100816/59099861/attachment.html>