llvm dev - Nov 2009 - [LLVMdev] Puzzled by "set" in target description file

Hello everyone,

I  try to write target description by reference to existing backends.

Target description declares an instruction as below:

def MOVi2pieces : AI1x2<(outs GPR:$dst), (ins so_imm2part:$src), Pseudo,
                         "mov", " $dst, $src",
                         [(set GPR:$dst, so_imm2part:$src)]>;

I find 'set' declared in file TargetSelectDAG.td as
     def set;

But if i modify it,
     def set_try;

def MOVi2pieces : AI1x2<(outs GPR:$dst), (ins so_imm2part:$src), Pseudo,
                         "mov", " $dst, $src",
                         [(set_try GPR:$dst, so_imm2part:$src)]>;

tblgen will give a message: Unrecognized node set_try.

Why 'set ' is a node without SDNode when define?

Thanks.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20091125/ecf9e992/attachment.html>

On Nov 24, 2009, at 10:33 PM, Heyu Zhu wrote:
> Hello everyone,
>  
> I  try to write target description by reference to existing backends.
>  
> Target description declares an instruction as below:
> 
> def MOVi2pieces : AI1x2<(outs GPR:$dst), (ins so_imm2part:$src), Pseudo,
>                          "mov", " $dst, $src",
>                          [(set GPR:$dst, so_imm2part:$src)]>;
>  
> I find 'set' declared in file TargetSelectDAG.td as
>      def set;
>  
> But if i modify it,
>      def set_try;
>  
> def MOVi2pieces : AI1x2<(outs GPR:$dst), (ins so_imm2part:$src), Pseudo,
>                          "mov", " $dst, $src",
>                          [(set_try GPR:$dst, so_imm2part:$src)]>;
>  
> tblgen will give a message: Unrecognized node set_try.
>  
> Why 'set ' is a node without SDNode when define?
>  
'set' in this case is used essentially as a keyword. The TableGen
program specifically looks for the word "set" when it does pattern
matching. If you look at utils/TableGen/CodeGenDAGPatterns.cpp, you'll see
what I mean.

If you need your own version of "set", then you'll have to modify
the TableGen program itself to recognize the new keyword. But it's probably
best to not create a new "set" if at all possible. :-)

-bw