8-bit math using -O3, compared to -O2, gives different results.

test_uint8.c

/* 8-bit math using -O3, compared to -O2, gives different results.
 * Using both `gcc` and `clang` and both -Ox, i've see three (four)
 * different values. Which one is correct?
 *
 * NOTE code is extremely fragile: Even uncomment printf()
 * below, hides the "bug", makes all results same. Same is when
 * trying shrink the code even a bit.
 *
 *     gcc -O2 ./test_uint8.c     # prints ... ffffd565
 *     gcc -O3 ./test_uint8.c     # prints ... ff80d565
 *     clang -O2 ./test_uint8.c   # prints ... ffffd565
 *     clang -O3 ./test_uint8.c   # prints ... ff01d765
 *                                  or ff81d765, depends if printf() is uncommented.
 *
 * gcc (GCC) 16.1.1 20260430, clang version 22.1.5, Archlinux x86_64
 *
 * By jpka, 2026. Std: C99. License: GPL 2+ */

#include <stdio.h>
#include <stdint.h>
#include "sys/param.h" /* MIN(), MAX() */

#define MSG(S,...) if (v) printf(S"\n", ##__VA_ARGS__);

int w = 0;
int v = 2;

uint16_t foo (uint8_t n, uint8_t h)
{
    uint32_t result = (n * 255 + (h-1) * 12) % (12*255);
    return (uint16_t) result;
}

void func(uint8_t a, uint8_t h, uint8_t s, uint8_t l)
{
    if (h == 0)
        s = 0;

    if (w & 128)
        l = 255 - l;

    uint16_t A = (s * MIN(l, 255 - l) * 259) >> 8;

    uint32_t r0 = ((l * (257*255) - A * MAX(MIN(MIN(foo(0, h) - 3*255, 9*255 - foo(0, h)), 255), -255)) * a / (255*255*255));
    uint32_t r8 = ((l * (257*255) - A * MAX(MIN(MIN(foo(8, h) - 3*255, 9*255 - foo(8, h)), 255), -255)) * a / (255*255*255));
    uint32_t r4 = ((l * (257*255) - A * MAX(MIN(MIN(foo(4, h) - 3*255, 9*255 - foo(4, h)), 255), -255)) * a / (255*255*255));

    uint32_t result = (((((a << 8) + (uint8_t)r0) << 8) + (uint8_t)r8) << 8) + (uint8_t)r4;

    MSG("%d, %x %x %x %x, %d %d %d %d, %x", !!(w & 128), a, h, s, l, A, r0, r8, r4, result);
    /* printf("%d %d %d %d %x\n", A, r0, r8, r4, result); */
}

int main(void)
{
    uint8_t a = 0xff;
    uint8_t h = 0x20;
    uint8_t s = 0xff;
    uint8_t l = 0xb2;

    func(a, h, s, l);

    return 0;
}