[llvm-bugs] [Bug 47346] New: MS ABI: wrong ABI for returning struct with explicitly defaulted default constructor

via llvm-bugs llvm-bugs at lists.llvm.org
Fri Aug 28 12:29:12 PDT 2020


https://bugs.llvm.org/show_bug.cgi?id=47346

            Bug ID: 47346
           Summary: MS ABI: wrong ABI for returning struct with explicitly
                    defaulted default constructor
           Product: clang
           Version: 10.0
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: C++
          Assignee: unassignedclangbugs at nondot.org
          Reporter: mizvekov 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

Consider this example reproduction:

```
extern void bar() noexcept;

struct ret1 {
    bool val;
};

struct ret2 {
    ret2() = default;
    bool val;
};

template<class T> extern T fooret() noexcept;

template <class T> void test() noexcept {
    auto ret = fooret<T>();
    if (ret.val) bar();
}

template void test<ret1>() noexcept;
template void test<ret2>() noexcept;
```

MSVC generates exactly the same code for both `test<ret1>` and `test<ret2>`,
expecting the struct to be returned by register from `fooret<>`.

clang generates different code for each, on ret1 case it expects by register,
on ret2 case it passes by implicit pointer parameter.

Workspace for convenience: https://godbolt.org/z/q3194d

Also, and sorry perhaps if this should be a separate bug, but notice on clang
the ret2 case is also suppressed from tail calling into bar.
This should not happen either way, and may be related to
https://bugs.llvm.org/show_bug.cgi?id=46126
But perhaps it's because the implicit parameter does not have the
`[[clang::noescape]]` attribute?

The following code with an explicit reference parameter shows the noescape
attribute makes a difference (even though it should not, but that is probably
related to #46126)

The following code shows this:
```
extern void bar() noexcept;

struct ret_t {
    unsigned int val;
};

extern void fooret1(ret_t&) noexcept;
extern void fooret2([[clang::noescape]] ret_t &) noexcept;

void test1() noexcept {
    bool val;
    {
        ret_t ret;
        fooret1(ret);
        val = ret.val;
    }
    if (val) bar();
}

void test2() noexcept {
    bool val;
    {
        ret_t ret;
        fooret2(ret);
        val = ret.val;
    }
    if (val) bar();
}
```

Another workspace for convenience: https://godbolt.org/z/r3qPsc

And there you can see test1 is suppressed from tail-calling into bar, but not
test2.

-- 
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/20200828/c33edf5a/attachment.html>


More information about the llvm-bugs mailing list