<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/121859>121859</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            clang++ is unable to optimize what clang-cl and clang can
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
            clang
      </td>
    </tr>

    <tr>
      <th>Assignees</th>
      <td>
      </td>
    </tr>

    <tr>
      <th>Reporter</th>
      <td>
          SEt-t
      </td>
    </tr>
</table>

<pre>
    On Windows.

Compiling with
`clang++ -O3 -mavx2 -fopenmp -flto -fuse-ld=lld main.cpp hit.cpp`
results in loop not vectorized. LTO seems to work as `hit_test` is inlined.

Loop is properly vectorized when compiled with
`clang-cl /clang:-O3 /clang:-mavx2 /clang:-fopenmp -flto -fuse-ld=lld main.cpp hit.cpp`
or (rename files to *.c)
`clang -O3 -mavx2 -fopenmp -flto -fuse-ld=lld main.c hit.c`
or removing `-fopenmp` or removing marked function call in main.cpp.

main.cpp
```cpp
#include <stdio.h>
#include <stdint.h>
#include <stdbool.h>
#include <time.h>

bool hit_test(float x, float y);

volatile uint64_t k = 1000000000;

int main(int argc, char **argv)
{
        uint64_t n = 0;

        int ts = clock();
#pragma omp parallel
        {
                uint64_t nx = 0;
                clock(); // <- this frightens optimizer, any external function call
                uint64_t r = (uint64_t)__builtin_frame_address(0);
#pragma omp for nowait schedule(dynamic, 16384)
                for (uint64_t i = k; i > 0; i--)
                {
                        r += 0x9e3779b97f4a7c15;
                        float x = (int)(r & 0xFFFFFF)*(2.f/0xFFFFFF) - 1;
                        float y = (int)(r >> 40)*(2.f/0xFFFFFF) - 1;

                        if (hit_test(x, y))
                                nx++;
                }
#pragma omp atomic
                n += nx;
        }
        int te = clock();

        printf("%g\n", 4.*n/k);
        printf("Time: %i\n", te-ts);

        return 0;
}
```

hit.cpp
```cpp
#include <stdbool.h>

bool hit_test(float x, float y)
{
    return x*x + y*y <= 1;
}
```
</pre>
<img width="1" height="1" alt="" src="http://email.email.llvm.org/o/eJycVsuuozgQ_RpnUyIyJrwWLLiPrFrqxbQ0y8iBAjwxNrJNHv31I0PIJdP3anomQsKvOlXnVFEOt1a0CrEg8QuJ3zZ8dJ02xR_vLnCbo65vxXcFfwpV64vdEloSWr7qfhBSqBYuwnV-LaGV5Kol7IWwFwi-RxD0_HxlEDR6QNUPEDTSaQia0WIgaxK9SVlDz4XaVsMAnXD-TRJKaGnQjtJZEAqk1gMo7eCMldNG_MR6C99-fAeL2FtwGi7anIBbIAnthDs4tI4kFIQ3l0JhfY_5m0cSFgajBzTytkKES4cKqomUnz1zCioJhO1nflHpua2nM831yv9grA0QlhlUvEdohMSJGmHltiIsXwXzX6Wd3TycGOz12eeNJHQx92qtt3puTlhDM6rKCa2g4lL6VCyR3_VcpnNw83OfskioSo41AolerauF3nYkev90S7kv945ay883nejxsUNLfxAeyWdZIzV3cCXsFebhzYsYvcynz1pyJyTCKJRLdgcHJyDRG4R0-T2OCuUm2oRlfshNW3nQquM-XyVhJTft-Z6hdDbKH7Bqgv1AIzT3KM5O65XU1Ymw7CM0Fg2Gtz0H3Q8wcMOlRDnbPcCf8K_PDvzmM6ovS8L2XrIAXCcsNEa0nUNlQQ9O9OInGs-Iqxvg1aFRXD5n_le3ZvJKWLasEJYfDsdRSCfUoTG8xwOva4PWEpbRL_g12oDSFy4c2KrDepRIWFbfFO_FJHKYRNnuLu0UQDN_JI84xBTHydP0w_dJCBBBsDZaC0do7hFeJtGuOUZpmh_ztNnxtArjtYre21xDC1ehPE3_jQJhCdDrfvpNayVhGds2hO1XyxBA-Cnk7TPI6N0T2NHfw_uAFI1HWpX-VPRTua9FmB51nbvzU1Tp2y-Z4U77FCxH1CKZun5Yzmb3esYv6nk6MRihXDOtM8LilsSvahq-wm5LWKkI259WNk8GP0SPJCqBsFisDB0Gzv7Dj0E3GvX4GOYAl740n1o67r83rKfO87v9ZWkBAAD3aK6ElVevnz9R-sy_Tp3m0yA3dRHVeZTzDRZhGiWMZlmebrqC7vI0w_yYVg1GVbiLMaVpHPKKhcc6itONKBhlMQ1pSmm4i9NtniUpYnTMkiZiScjJjmLPhdxKee632rQbYe2IRcjCLM43kh9R2un-Z-x-jzP_V8AU3iA4jq0lOyqFdfYDwgknsVhf-8LCqPhRor-8lv4Cl447eNykXNXzBCquNqORRefcYElUzr2qFa4bj9tK94Ttva_7KxiM_gsrR9h-it0Str-Hfy7Y3wEAAP__0o-M2A">