[llvm-bugs] [Bug 39683] New: clang silently ignores trivial_abi in some cases

via llvm-bugs llvm-bugs at lists.llvm.org
Thu Nov 15 17:55:25 PST 2018


https://bugs.llvm.org/show_bug.cgi?id=39683

            Bug ID: 39683
           Summary: clang silently ignores trivial_abi in some cases
           Product: clang
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: C++
          Assignee: unassignedclangbugs at nondot.org
          Reporter: comexk at gmail.com
                CC: blitzrakete at gmail.com, dgregor at apple.com,
                    erik.pilkington at gmail.com, llvm-bugs at lists.llvm.org,
                    richard-llvm at metafoo.co.uk

Target: x86_64-apple-darwin18.2.0

In the following code, the class Test is marked trivial_abi (and is not a
template class), and Clang does not produce any warnings about trivial_abi when
compiling it, yet for the function test() returning a value of that class,
Clang's generated code returns it through memory rather than a register.

I'm not sure whether this class should actually be eligible for trivial_abi or
not, but if not, there definitely should be a warning.

If you change the definition  by adding an additional non-template move
constructor, then trivial_abi starts being respected:

+    Test(Test &&other) { x = other.x; }

Tested on latest trunk targeting macOS.

--

struct Test {
    int x;
    Test(int x) : x(x) {}
    template <typename T>
    Test(T &&other) { x = other.x; }
    Test(const Test &) = delete;
    template <typename T>
    Test &operator=(T &&other) { x = other.x; }
    Test &operator=(const Test &) = delete;
} __attribute__((trivial_abi));

Test test() { 
    return Test(42);
}

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20181116/518a8bca/attachment.html>


More information about the llvm-bugs mailing list