[PATCH]es: algorithm: Avoid moving initial subsequences in std::inplace_merge()

[Duncan P. N. Exon Smith]

> On 2015-Aug-05, at 16:43, Duncan P. N. Exon Smith <dexonsmith at apple.com> wrote:
> 
> 
>> On 2015-Jul-29, at 14:34, Duncan P. N. Exon Smith <dexonsmith at apple.com> wrote:
>> 
>> r227808 (and r227811) changed `std::inplace_merge()` to meet the
>> complexity guidelines of the standard, which strictly restricts the
>> number of calls to the predicate.  Unfortunately, this removed an
>> optimization that avoided moving any elements if the sequence was
>> already fully sorted.  Moroever, any initial/final subsequence that was
>> in the right spot would previously never be touched.
>> 
>> r243530, whose main purpose was to fix a latent self-move bug made more
>> common by r227808/r227811, already brought half of this optimization
>> back.  It avoids unnecessarily moving the final subsequence.
>> 
>> These patches close the loop, bringing back the other half of the
>> optimization without re-introducing the extra predicate call that
>> r227808 avoided.
>> 
>> 0001 is an NFC refactor, while 0002 has the actual change.
>> 
>> (The change in r227811 was to *always* malloc a temporary buffer.  I
>> wonder if something like this could safely optimize that away as well?)
> 
> ping.
> 
> <0001-algorithm-Refactor-__buffered_inplace_merge-NFC.patch><0002-algorithm-Avoid-moving-initial-subsequences-in-std-i.patch>

(sorry for the double-ping; this time I've CC'ed the new mailing list)

-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-algorithm-Refactor-__buffered_inplace_merge-NFC.patch
Type: application/octet-stream
Size: 3046 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20150805/11a24f0b/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0002-algorithm-Avoid-moving-initial-subsequences-in-std-i.patch
Type: application/octet-stream
Size: 7225 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20150805/11a24f0b/attachment-0001.obj>