https://bugs.llvm.org/show_bug.cgi?id=37449
Bug ID: 37449
Summary: clang++ fails to optimize variadic template
constructor with fold expression
Product: clang
Version: trunk
Hardware: PC
OS: Windows NT
Status: NEW
Severity: enhancement
Priority: P
Component: -New Bugs
Assignee: unassignedclangb...@nondot.org
Reporter: vittorio.ro...@outlook.com
CC: llvm-bugs@lists.llvm.org
Consider the following wrapper around `std::vector`:
#include <vector>
struct init { };
template <typename T>
struct vec : std::vector<T>
{
using std::vector<T>::vector;
template <typename... Xs>
vec(init, Xs&&... xs)
{
this->reserve(sizeof...(Xs));
(this->emplace_back(std::forward<Xs>(xs)), ...);
}
};
When compiling the following code `clang++-trunk -Ofast -std=c++2a`...
void foo()
{
#ifndef VARIADIC
vec<std::string> v;
v.reserve(2);
v.emplace_back("hello");
v.emplace_back("world");
#else
vec<std::string> v{init{}, "hello", "world"};
#endif
}
...the variadic version produces roughly 5x more assembly instructions and is
around 46% slower at run-time.
Marking `vec`'s constructor as `__attribute__((always_inline))` makes both
versions produce the same assembly.
Live assembly comparison on godbolt.org:
https://godbolt.org/g/YwM86C
Live run-time benchmark on quick-bench.com:
http://quick-bench.com/C5_4Bm5Crsqg1ykE0YmoHFOwmLs
I think that this is a very reasonable/common usage of a variadic template
constructor + fold expression - it is surprising to me as a developer that
clang++ doesn't optimize this. This *should* be a better alternative in terms
of run-time performance to `std::initalizer_list`, which prevents moves and has
other significant drawbacks - it is disappointing to see `always_inline` being
required here.
--
You are receiving this mail because:
You are on the CC list for the bug._______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
