llvm dev - Jul 2008 - [LLVMdev] GEP::getIndexValid() with other iterators

On Jul 16, 2008, at 9:58 AM, Matthijs Kooijman wrote:
> Hi all,
>
> once more with the patch inline for easy review. I did not include the
> argpromotion pass here, since it's not the main topic of this post.
Hi Matthijs,

I'd prefer to not turn this into a template.  Why not just define a  
version that takes an array of uint64_t's or something like that?

-Chris

Hi Chris,

> I'd prefer to not turn this into a template.  Why not just define a  
> version that takes an array of uint64_t's or something like that?
because I want to be able to pass in iterators. I could define a version that
takes std<uint64_t>::iterators, but next thing we know, we also need them
for
lists, SmallVectors, etc. That's why one of the original getIndexedType
methods is a template, and that's why I think it makes sense to make another
one a template.

Any particular objections to this? Is the code size increase a problem?
AFAICS, in cases where you need this method, it will be a tradeoff between
speed (having to iterate all your indices and create a new list with the
Value* versions) and code size (having a version of getIndexedPointer that can
handle your particular flavour of iterator).

Gr.

Matthijs
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20080723/51e4e3b6/attachment.sig>

On Jul 22, 2008, at 11:54 PM, Matthijs Kooijman wrote:
> Hi Chris,
>
>
>> I'd prefer to not turn this into a template.  Why not just define a
>> version that takes an array of uint64_t's or something like that?
> because I want to be able to pass in iterators. I could define a  
> version that
> takes std<uint64_t>::iterators, but next thing we know, we also need
> them for
> lists, SmallVectors, etc. That's why one of the original  
> getIndexedType
> methods is a template, and that's why I think it makes sense to make  
> another
> one a template.
What flavor of iterators do you want to pass in?  vector or  
smallvector?  If so, a pointer to the first element + extents is fine.
> Any particular objections to this? Is the code size increase a  
> problem?
> AFAICS, in cases where you need this method, it will be a tradeoff  
> between
> speed (having to iterate all your indices and create a new list with  
> the
> Value* versions) and code size (having a version of  
> getIndexedPointer that can
> handle your particular flavour of iterator).
My basic objection is that I don't like tons of code in header files.   
It obfuscates the header and slows down compile times (of llvm itself)

-Chris