https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125133
Bug ID: 125133
Summary: Optimization: v = *(*p)++; can make different code
from v = **p, p++;
Product: gcc
Version: 15.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: other
Assignee: unassigned at gcc dot gnu.org
Reporter: Explorer09 at gmail dot com
Target Milestone: ---
Compile Explorer link:
https://godbolt.org/z/G3sj3jbY4
```c
// <https://gitlab.com/Explorer09/utf_convert/-/blob/main/utf9.c>
#include <stdint.h>
uint_least32_t utf9_to_code_point_simp(const uint_least16_t **sequence)
{
unsigned int nonet;
uint_least32_t value = 0;
do {
nonet = *(*sequence)++;
value = (value << 8) | (nonet & 0xFF);
} while (nonet >= 0x100);
return value;
}
uint_least32_t utf9_to_code_point_simp_2(const uint_least16_t **sequence)
{
unsigned int nonet;
uint_least32_t value = 0;
do {
nonet = **sequence;
(*sequence)++;
value = (value << 8) | (nonet & 0xFF);
} while (nonet >= 0x100);
return value;
}
```
The expected result is both functions compile to same code, because there is no
semantic difference.
I tested on x86-64, AArch64 and RV64, and the generated code is slightly
different without a good reason.
(Both -O2 and -Os optimization modes are tested.)
