[Bug target/113989] New: MinGW generates unaligned vmovdqa64

[Bug target/113989] New: MinGW generates unaligned vmovdqa64

[xjkp2283572185 at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113989

            Bug ID: 113989
           Summary: MinGW generates unaligned vmovdqa64
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: xjkp2283572185 at gmail dot com
  Target Milestone: ---

===
Compiler
===
Using built-in specs.
COLLECT_GCC=D:\Tools\gcc\bin\g++.exe
COLLECT_LTO_WRAPPER=D:/Tools/gcc/bin/../libexec/gcc/x86_64-w64-mingw32/14.0.1/lto-wrapper.exe
Target: x86_64-w64-mingw32
Configured with: ../configure --disable-werror
--prefix=/home/luo/x86_64-w64-mingw32-native-gcc14 --host=x86_64-w64-mingw32
--target=x86_64-w64-mingw32 --enable-multilib --enable-languages=c,c++
--disable-sjlj-exceptions --enable-threads=win32
Thread model: win32
Supported LTO compression algorithms: zlib
gcc version 14.0.1 20240130 (experimental) (GCC)

===
Source Code
===

// a.cpp
struct alignas(64) a
{
    char buf[64]{};
};
a f(long long i) noexcept;
int main()
{
    f(1);
    return 0;
}

// b.bpp
struct alignas(64) a
{
    char buf[64]{};
};
a f(long long i) noexcept
{
    using v[[gnu::vector_size(64)]] = long long;
    v vi{i,i,i,i,i,i,i,i};
    a res{};
    __builtin_memcpy(&res, &vi, 64);
    return res;
}

===
Command
===
g++ a.cpp b.cpp -march=znver4 -Ofast -g

===
Result
===
Run it with gdb and it crashes:

Thread 1 received signal SIGSEGV, Segmentation fault.
0x00007ff78ac01549 in f (i=i@entry=1) at b.cpp:10
10          __builtin_memcpy(&res, &vi, 64);
(gdb) disassemble
Dump of assembler code for function _Z1fx:
   0x00007ff78ac01540 <+0>:     vpbroadcastq %rdx,%zmm0
   0x00007ff78ac01546 <+6>:     mov    %rcx,%rax
=> 0x00007ff78ac01549 <+9>:     vmovdqa64 %zmm0,(%rcx)
   0x00007ff78ac0154f <+15>:    vzeroupper
   0x00007ff78ac01552 <+18>:    ret
End of assembler dump.
(gdb) info registers rcx
rcx            0x5ffe90            6291088
(gdb) disassemble main
Dump of assembler code for function main():
   0x00007ff640f62a50 <+0>:     sub    $0x68,%rsp
   0x00007ff640f62a54 <+4>:     call   0x7ff640f61610 <__main>
   0x00007ff640f62a59 <+9>:     lea    0x20(%rsp),%rcx
   0x00007ff640f62a5e <+14>:    mov    $0x1,%edx
   0x00007ff640f62a63 <+19>:    call   0x7ff640f61540 <_Z1fx>
   0x00007ff640f62a68 <+24>:    xor    %eax,%eax
   0x00007ff640f62a6a <+26>:    add    $0x68,%rsp
   0x00007ff640f62a6e <+30>:    ret
End of assembler dump.

But if we compile the program with "g++ a.cpp b.cpp -march=znver4 -Ofast
-fno-asynchronous-unwind-tables -fno-exceptions", the program won't crash.

[Bug target/113989] MinGW generates unaligned vmovdqa64

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113989

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Feels like the old issue that the incoming stack in main() is not aligned as
GCC it thinks it is?  On x86_64-linux we dynamically re-align the stack:

main:
.LFB0:
        .cfi_startproc
        pushq   %rbp
        .cfi_def_cfa_offset 16
        .cfi_offset 6, -16
        movl    $1, %esi
        movq    %rsp, %rbp
        .cfi_def_cfa_register 6
        andq    $-64, %rsp
        subq    $64, %rsp
        movq    %rsp, %rdi
        call    _Z1fx

does -mstackrealign make it work?  I think there's some duplicate bugreports
about mingw and stack alignment.

[Bug target/113989] MinGW generates unaligned vmovdqa64

[xjkp2283572185 at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113989

--- Comment #2 from 严 逍宇 <xjkp2283572185 at gmail dot com> ---
(In reply to Richard Biener from comment #1)
> does -mstackrealign make it work?

It doesn't work. GDB shows that the address in rcx is still unaligned.
(gdb) disassemble
Dump of assembler code for function _Z1fx:
   0x00007ff6659a1540 <+0>:     vpbroadcastq %rdx,%zmm0
   0x00007ff6659a1546 <+6>:     mov    %rcx,%rax
=> 0x00007ff6659a1549 <+9>:     vmovdqa64 %zmm0,(%rcx)
   0x00007ff6659a154f <+15>:    vzeroupper
   0x00007ff6659a1552 <+18>:    ret
End of assembler dump.
(gdb) info registers rcx
rcx            0x5ffe90            6291088

Is there any way to temporarily block the bug?

[Bug target/113989] MinGW generates unaligned vmovdqa64

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113989

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |DUPLICATE
             Status|UNCONFIRMED                 |RESOLVED

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Dup.

*** This bug has been marked as a duplicate of bug 49001 ***