| Issue |
53147
|
| Summary |
clang++ generate aligned avx code cause coredump.
|
| Labels |
|
| Assignees |
|
| Reporter |
xsc351072127
|
```C
#include<stdio.h>
struct __attribute__ ((aligned (64)))
S
{
S()
{
A[0]=0;
A[1]=0;
A[2]=0;
A[3]=0;
A[4]=0;
A[5]=0;
A[6]=0;
A[7]=0;
};
long long A[8];
};
int main()
{
S* s1 = new S();
S* s2 = new S();
S* s3 = new S();
printf("s1=%p,s2=%p,s3=%p",s1,s2,s3);
return 0;
}
```
When use the command:
`clang++ -O1 -mavx512vl -mavx512bw align_test.cpp -o at`
or
`clang++ -O1 -mavx2 -mfma align_test.cpp -o at`
And run the at will get a coredump.
Because will got a simd instruction like the after to clean the memory.
```
vxorps %xmm0,%xmm0,%xmm0
vmovaps %zmm0,(%rdi)
```
But the memory address alloc from new, may not be 64 byte aligned.
I use the g++ to complie these code, g++ will not optimization the memory clean or use the vmovups.
The laster version gcc will use the c++17 standard new() operator with alignment parament.
`void* operator new[]( std::size_t count, std::align_val_t al );`
Use the https://godbolt.org/ can see more detail difference with gcc and clang.
I can reappearance the case in clang 7 and clang 11 in my mechine and some other version in https://godbolt.org/.
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
