[PATCH] D48342: [libcxx] Optimize vectors construction of trivial types from an iterator range with const-ness mismatch.
Volodymyr Sapsai via Phabricator via cfe-commits
cfe-commits at lists.llvm.org
Wed Jul 4 16:45:41 PDT 2018
vsapsai added a comment.
In https://reviews.llvm.org/D48342#1152063, @EricWF wrote:
> Are there any tests which actually exercise the new behavior?
Added tests only verify we don't use memcpy erroneously. And existing tests make sure there are no functionality regressions. But there is nothing to test the performance improvement. Are there any recommendations for that?
Comment at: libcxx/include/memory:1665
+ (is_same<allocator_type, allocator<typename _VSTD::remove_const<_SourceTp>::type> >::value
+ || is_same<allocator_type, allocator<const _SourceTp> >::value
+ || !__has_construct<allocator_type, _DestTp*, const _SourceTp&>::value) &&
> I'm not sure we should care about allocators for `T const`. The're all but an abomination.
My main goal was to avoid performance regression for
std::vector<const int> v(const_raw, const_raw + SIZE);
I'm not protecting allocators for `T const`, it just seems cheap to support them anyway.
Comment at: libcxx/test/std/containers/sequences/vector/vector.cons/construct_iter_iter_different_value_type.pass.cpp:13
+// template <class InputIter> vector(InputIter first, InputIter last);
+// Initialize a vector with a different value type. Make sure initialization
> Can this be folded into an existing test file for the constructor it's targeting?
Will move to construct_iter_iter.pass.cpp. Which reminded me that I need to make construct_iter_iter_alloc.pass.cpp and construct_iter_iter.pass.cpp more in sync.
More information about the cfe-commits