[Bug tree-optimization/123225] [16 Regression] Overly-aggressive vectorization of uncounted loops

[lili.cui at intel dot com via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123225

--- Comment #5 from cuilili <lili.cui at intel dot com> ---

Here is another small test case extracted from a performance regression,
compiled with "-march=skylake-avx512 -Ofast".

int scan_until_bar(const char *p)
{
    do {
        ++p;

        char c = *p;
        if (c == ' ' || c == '\0')
            return -1;
        if (c == '|')
            break;
    } while (true);

    return (int)(uintptr_t)p;
}

See the non-vectorized vs vectorized assembly comparisons here:
https://godbolt.org/z/MrTchn7sc