PATCH: More thorough cleanup of NVPTX-generated global names

[Eli Bendersky]

On Mon, Mar 31, 2014 at 9:03 AM, Eli Bendersky <eliben at google.com> wrote:

>
>
>
> On Mon, Mar 31, 2014 at 8:37 AM, Justin Holewinski <
> justin.holewinski at gmail.com> wrote:
>
>> No, this looks good to me!
>>
>
> Thanks, r205212.
>
>

I also reverted r203483 in r205216, since it's no longer necessary. I'll
close PR19099 too.

Eli


>
>>
>> On Mon, Mar 31, 2014 at 11:31 AM, Eli Bendersky <eliben at google.com>wrote:
>>
>>> Ping - any objections to get this in?
>>>
>>>
>>> On Wed, Mar 26, 2014 at 4:24 PM, Eli Bendersky <eliben at google.com>wrote:
>>>
>>>> Hello,
>>>>
>>>> As discussed in PR19099, the solution in r203483 was imperfect. While
>>>> we could fix global names when they are emitted in directives, this does
>>>> not work for their usage within instructions. Instruction emission is
>>>> hard-coded through MI and NVPTX does not control this.
>>>>
>>>> A more thorough solution is a target-specific IR pass that replaces all
>>>> invalid characters in symbol names before they go into codegen. This makes
>>>> sure that invalid characters will never appear in symbols in PTX.
>>>>
>>>> An added benefit is that this avoids collisions because Value->setName
>>>> does automatic collision resolution.
>>>>
>>>> Once this goes in, most of r203483 can be reverted (I'll take care of
>>>> that).
>>>>
>>>> PTAL,
>>>> Eli
>>>>
>>>>
>>>
>>
>>
>> --
>>
>> Thanks,
>>
>> Justin Holewinski
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140331/da76e02d/attachment.html>