IR: Add distinct MDNodes as a first-class concept

[Pete Cooper]

> On Jan 7, 2015, at 4:04 PM, Duncan P. N. Exon Smith <dexonsmith at apple.com> wrote:
> 
> Actually, no -- see below:
> 
>> On 2015-Jan-07, at 16:02, Pete Cooper <peter_cooper at apple.com> wrote:
>> 
>> Hi Duncan
>> 
>> I assume that prior to these commits clang is already using MDNode::getDistinct?  Otherwise I can’t see how your clang patch would work here.
>> 
>> Otherwise LGTM.
>> 
>> Thanks,
>> Pete
>>> On Jan 7, 2015, at 3:27 PM, Duncan P. N. Exon Smith <dexonsmith at apple.com> wrote:
>>> 
>>> In order to "defeat" `MDNode` uniquing, metadata schemas resort to
>>> using self-references to prevent merging.
>>> 
>>> Introduce a `distinct` designator which explicitly requests this
>>> behaviour, without the need for a self-reference.  These nodes are
>>> still stored in the `LLVMContext`, but are not uniqued based on
>>> their operands (well, not uniqued at all).
>>> 
>>> - This concept *already exists*.  Before the metadata/value split,
>>>  it was used whenever operands went to null (to prevent teardown
>>>  madness, but it happened more frequently).  It's still in use
>>>  for self-referencing `MDNode`s, as well as when
>>>  `MDNode::replaceOperandWith()` causes a uniquing collision (see
>>>  module flags behaviour before r225397 for an example).
> 
> ^ Distinct `MDNode`s are implicitly created for self-references and on
> uniquing collisions.
Cool.  Still LGTM then :)
> 
>>> 
>>>  Some recent commits have already exposed `getDistinct()` and
>>>  `isDistinct()` as API (see r225401 and r225406).
>>> 
>>> - The first patch adds assembly/bitcode support.  I chose the
>>>  'distinct' keyword.  It has an accompanying clang patch to
>>>  update testcases.
>>> 
>>> - The second patch adds support to `MapMetadata()` to maintain the
>>>  'distinct'-ness.
>>> 
>>> - The third patch just clears out the TODO.
>>> 
>>> I haven't updated any metadata schemas here; I figure the owners of
>>> the schemas can update those in their own time.
>>> 
>>> My main concern here was about how to deal with `MapMetadata()` --
>>> it feels wrong to just duplicate everything.  However, I looked
>>> through all the callers.  In *most* cases RF_NoModuleLevelChanges
>>> is passed, and in the others this naive solution seems correct (or
>>> at least harmless).
>>> 
>>> If this becomes an issue (i.e., there's a place with module-level
>>> changes where distinct metadata should only be duplicated when
>>> operands change), we can add another flag to decide whether to
>>> duplicate it.  For now, this just matches the behaviour we already
>>> get out of self-references.
>>> 
>>> <0001-IR-Add-distinct-MDNodes-to-bitcode-and-assembly.patch><clang.patch><0002-Utils-Keep-distinct-MDNodes-distinct-in-MapMetadata.patch><0003-IR-Drop-TODO-now-that-PR22111-is-finished.patch>_______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150107/533f27c9/attachment.html>