[llvm-commits] [llvm] r141567 - /llvm/trunk/include/llvm/Target/Target.td

[Jim Grosbach]

On Oct 18, 2011, at 12:24 PM, Andrew Trick wrote:

> 
> On Oct 10, 2011, at 1:56 PM, Jim Grosbach wrote:
> 
>> 
>> On Oct 10, 2011, at 11:51 AM, Jakob Stoklund Olesen wrote:
>> 
>>> Author: stoklund
>>> Date: Mon Oct 10 13:51:33 2011
>>> New Revision: 141567
>>> 
>>> URL: http://llvm.org/viewvc/llvm-project?rev=141567&view=rev
>>> Log:
>>> Mark the standard pseudos as isPseudo = 1.
>>> 
>>> The difference between isPseudo and isCodeGenOnly is a bit murky, but
>>> isCodeGenOnly should eventually go away.  It is used for instructions
>>> that are clones of real instructions with slightly different properties.
>> 
>> Generally, isPseudo denotes pseudo-instructions from the perspective of code-gen. isCodeGenOnly denotes pseudos from the perspective of the assembler instruction matcher.
>> 
>> Specifically:
>> isPseudo: Does not have encoding information and should be expanded, at the latest, during lowering to MCInst.
>> isCodeGenOnly: Does have encoding information and can go through to the CodeEmitter unchanged, but duplicates a canonical instruction definition's encoding and should be ignored when constructing the assembler match tables.
>> 
>> And yes, isCodeGenOnly should eventually go away. It's necessary for now until we convert the duplicate definitions into pseudos for code-gen.
>> 
>> -Jim
> 
> 
> That makes sense. However, I always thought it was a subset relation given the definition of PseudoInst:
> 
> class PseudoInst<dag oops, dag iops, InstrItinClass itin, list<dag> pattern>
> ...
>  let isCodeGenOnly = 1;
>  let isPseudo = 1;

Sorta.  It should be sufficient to just mark them as isPseudo, but it just seems odd to have "isCodeGenOnly" not always be true for pseudos. This is a purely pragmatic thing, IIRC, to keep the predicate functions returning consistent results.

-Jim