llvm dev - Jan 2010 - [LLVMdev] TableGen !eq() Operator Patch

Attached is a patch to implement an !eq() operator in TableGen.  We use this 
for the AVX specification to allow the user to control what kind of pattern 
should be used for a particular instruction def.  For example, we use it for 
reg-mem instructions to let the user choose between a built-in generic reg-mem 
pattern, the same pattern that was used for the reg-reg variant of the
instruction or a custom pattern provided by the user.

It only operates on strings.

An alternative to this would be to implement named parameters in TableGen but 
that seems like overkill for this case.  Named parameters could be useful in 
other areas to reduce template complexity but implementing it is non-trivial 
given the widespread assumption in the TableGen sources that template 
parameters exactly match up with template arguments.

Comments?  Ok to commit?

                            -Dave
-------------- next part --------------
A non-text attachment was scrubbed...
Name: eq.patch
Type: text/x-diff
Size: 4569 bytes
Desc: not available
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20100105/01e01702/attachment.patch>

On Jan 5, 2010, at 9:59 AM, David Greene wrote:
> Attached is a patch to implement an !eq() operator in TableGen.  We  
> use this
> for the AVX specification to allow the user to control what kind of  
> pattern
> should be used for a particular instruction def.  For example, we  
> use it for
> reg-mem instructions to let the user choose between a built-in  
> generic reg-mem
> pattern, the same pattern that was used for the reg-reg variant of the
> instruction or a custom pattern provided by the user.
>
> It only operates on strings.
>
> An alternative to this would be to implement named parameters in  
> TableGen but
> that seems like overkill for this case.  Named parameters could be  
> useful in
> other areas to reduce template complexity but implementing it is non- 
> trivial
> given the widespread assumption in the TableGen sources that template
> parameters exactly match up with template arguments.
I can't say if this is really the best answer for AVX, but  
independently of that the patch looks great, please commit.

-Chris

On Tuesday 05 January 2010 12:31, Chris Lattner wrote:
> I can't say if this is really the best answer for AVX, but
> independently of that the patch looks great, please commit.
Ok.  I'm sure we'll iterate on the AVX stuff quite a bit.

                         -Dave