[llvm-bugs] [Bug 47577] New: Atomic operation emits an __atomic_* call rather than an instruction for trivial structure

via llvm-bugs llvm-bugs at lists.llvm.org
Fri Sep 18 04:29:05 PDT 2020


https://bugs.llvm.org/show_bug.cgi?id=47577

            Bug ID: 47577
           Summary: Atomic operation emits an __atomic_* call rather than
                    an instruction for trivial structure
           Product: clang
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: C++
          Assignee: unassignedclangbugs at nondot.org
          Reporter: antoshkka 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 the example:


#include <atomic>

struct SleepState {
    unsigned short flags;
    unsigned short epoch;
};

void test(std::atomic<SleepState>& i) {
    SleepState ss{2, 4};
    i.compare_exchange_weak(ss, SleepState{4,2});
}


Function test() could be compiled to the following:

test(std::atomic<SleepState>&):
  mov eax, 262146
  mov edx, 131076
  mov DWORD PTR [rsp-4], 262146
  lock cmpxchg DWORD PTR [rdi], edx
  ret


However clang calls __atomic_compare_exchange when libstdc++ is used:

test(std::atomic<SleepState>&)
  push rax
  mov rsi, rdi
  mov dword ptr [rsp], 262146
  mov dword ptr [rsp + 4], 131076
  mov rdx, rsp
  lea rcx, [rsp + 4]
  mov edi, 4
  mov r8d, 5
  mov r9d, 5
  call __atomic_compare_exchange
  pop rax
  ret


Godbolt playground: https://godbolt.org/z/Eajn57

-- 
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/20200918/67b90286/attachment.html>


More information about the llvm-bugs mailing list