| Issue |
86738
|
| Summary |
`va_arg()` code is suboptimal for AArch64 and x86-64
|
| Labels |
backend:AArch64,
backend:X86,
missed-optimization
|
| Assignees |
|
| Reporter |
Kmeakin
|
https://godbolt.org/z/hE6Maszq1
```c
#include <stdarg.h>
#include <stddef.h>
#include <stdint.h>
uint32_t get_u32(size_t count, ...) {
uint32_t ret = 0;
va_list args;
va_start(args, count);
ret = va_arg(args, uint32_t);
va_end(args);
return ret;
}
```
This simple function to get the first variadic argument to a function compiles to way more instructions for AArch64 and x86-64 under Clang than GCC. Clang generates slightly fewer instructions for RISCV than GCC does though
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
