[libcxx] Improve code generation for vector::clear().

[brucem]

Summary

By manipulating a local variable in the loop, when the loop can
be optimized away (due to no non-trivial destructors), this lets
it be fully optimized away and we modify the __end_ separately.

This results in a substantial improvement in the generated code.

Prior to this change, this would be generated (on x86_64):

movq    (%rdi), %rdx
movq    8(%rdi), %rcx
cmpq    %rdx, %rcx
je    LBB2_2
leaq    -12(%rcx), %rax
subq    %rdx, %rax
movabsq    $-6148914691236517205, %rdx ## imm = 0xAAAAAAAAAAAAAAAB
mulq    %rdx
shrq    $3, %rdx
notq    %rdx
leaq    (%rdx,%rdx,2), %rax
leaq    (%rcx,%rax,4), %rax
movq    %rax, 8(%rdi)

And after:

movq    (%rdi), %rax
movq    %rax, 8(%rdi)

This brings this in line with what other implementations do.

[mclow.lists]

I had no idea that we had no tests for vector::clear. Oosps.

This looks good to me, but I want to play with the codegen for a bit before approving it.
Also, you should add cfe-commits to the subscribers.

thanks for doing this!

test/std/containers/sequences/vector/vector.modifiers/clear.pass.cpp

25

How about a couple more assertions here:

LIBCPP_ASSERT(c.__invariants());`
LIBCPP_ASSERT(is_contiguous_container_asan_correct(c));

32

Same here.

[brucem]

Address comments from mclow.

[brucem]

These have been addressed, so this should be good for further review.

[mclow.lists]

LGTM

[brucem]

This was accepted long ago, and then I got sidetracked for a while. Is this still okay to land?

[EricWF]

Still LGTM

test/std/containers/sequences/vector/vector.modifiers/clear.pass.cpp

19

LIBCPP_ASSERT requires including test_macros.h

include/vector

void __move_assign_alloc(__vector_base&, false_type)
         {}
 };
 
 template <class _Tp, class _Allocator>
 inline _LIBCPP_INLINE_VISIBILITY
 void
 __vector_base<_Tp, _Allocator>::__destruct_at_end(pointer __new_last) _NOEXCEPT
 {
-    while (__new_last != __end_)
-        __alloc_traits::destroy(__alloc(), _VSTD::__to_raw_pointer(--__end_));
+    pointer __soon_to_be_end = __end_;
+    while (__new_last != __soon_to_be_end)
+        __alloc_traits::destroy(__alloc(), _VSTD::__to_raw_pointer(--__soon_to_be_end));
+    __end_ = __new_last;
 }
 
 template <class _Tp, class _Allocator>
 inline _LIBCPP_INLINE_VISIBILITY
 __vector_base<_Tp, _Allocator>::__vector_base()
         _NOEXCEPT_(is_nothrow_default_constructible<allocator_type>::value)
     : __begin_(nullptr),
       __end_(nullptr),

test/std/containers/sequences/vector/vector.modifiers/clear.pass.cpp

+//===----------------------------------------------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+// <vector>
+
+// void clear();
+
+#include <vector>
+#include <cassert>
+
+#include "min_allocator.h"
+
+int main()
+{
+    {
+    int a[] = {1, 2, 3};
+    std::vector<int> c(a, a+3);
+    c.clear();
+    assert(c.empty());
+    }
+#if TEST_STD_VER >= 11
+    {
+    int a[] = {1, 2, 3};
+    std::vector<int, min_allocator<int>> c(a, a+3);
+    c.clear();
+    assert(c.empty());
+    }
+#endif
+}