<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/98306>98306</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            Buggy optimization of `vfmaddcsh` intrinsics
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
            new issue
      </td>
    </tr>

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

    <tr>
      <th>Reporter</th>
      <td>
          sayantn
      </td>
    </tr>
</table>

<pre>
    The `llvm.x86.avx512fp16.maskz.vfmadd.csh` intrinsic (and due to that, `_mm_maskz_fmadd_sch`) is being incorrectly optimized. This code snippet

```C
#include<immintrin.h>
#include<stdio.h>

int main() {
    __m128h a, b, c, r;
    _Float16 array[8];

    a = _mm_setr_ph(0.0, 1.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0);
    b = _mm_setr_ph(0.0, 2.0, 16.0, 17.0, 18.0, 19.0, 20.0, 21.0);
    c = _mm_setr_ph(0.0, 3.0, 22.0, 23.0, 24.0, 25.0, 26.0, 27.0);

    r = _mm_maskz_fmadd_sch(0, a, b, c);
    _mm_storeu_ph(array, r);

 for (int i = 0; i < 8; i++){
        printf("%f\n", (float) array[i]);
    }

    return 0;
}
```

In `clang`, the unoptimized and optimized output is different. The unoptimized output is the correct one according to Intel. `gcc` gives the correct output in both.


System specification:
 - `mingw-w64-x86_64-gcc 14.1.0-3`
 - `mingw-w64-x86_64-clang 18.1.8-1`
 - Intel Software Development Emulator v9.33.0

</pre>
<img width="1px" height="1px" alt="" src="http://email.email.llvm.org/o/eJyUVUuToyoU_jW4obR4RKILF5NkUjXrmX0KEJV7FSzAdKd__S2IeXXfXkzK4kg8jw_O9wH3XvdGqQaUO1AeMr6EwbrG8ws3wWTCtpfmz6AgYGgcz1PxXrGCn99LTLoZs2Li_t-P4txNvG0L6QfAENQmOG28lhCQipsWtouCwcIw8ADIPqY6TdMphZ5S5MnLGAlIDbWHQmnTQ22kdU7JMF6gnYOe9IdqC_hn0B5K2yrojZ5nFQA6APRjHRm6Pvt1Tqg2clxaBeheT9MVWTEA-vN_HHxotX3-mEZtApy4NoBUER_Y7q7_Qwjh6TRhUg2Qx2WJOMg4OECfnY6j5QEzyJ3jF1DuKlAe7h4PPw4BPcC4M14Fd5oHQCpUoJgQr-Y2vc3JaulqN6sto61fUIjvs9-ysNVuV1uttl7dbu74a3b5ffYVG1mrkNt8xUrK1a7VyfY1-6OGu9f4zBxSpdCXLrziS8iCdWq5Yru2IrXqa63Ousjc2HidaiJAd-l1D6v0CsguPfULG-JvdtqELnGFAFJ2oNyb9LqPKbvIhMiiGxV0pMInrGB7-LJ0FRZnEo71093nxvjnkF8makyO3PRJVXsYBgUXc5cRjKp8zOwS5iVE6bW665RTJkShvYY8nGKyVZvQGgW5lNa1UbPBwl8mqLGI9Xsp42nQ67P6FLNmMlDYMBQvAiY47srEe5V2phpCmD2gPwA5AnLsdRgWUUg7AXKMx9Fq8tnZf5QMgBy59yp4QI54Q2rEaIkAOVK1LVnNVN4JRPINVyjngnQ5KTvcIka3Um5jG56Q_L74oCboZyV1pyUP2pqI49qTPC5w0qZ_y9_YJn-v2Ilt8l7KqEFcoJzeW_KNb-pO1Bguqhw_e6cdhL9tF964U_Cgzmq086RMgD-nZeTBOniuCxqFlGKytqFtTWueqQZvCWIYo5pkQ1MxJmrKurhIIWiLW4a4IEIIXNdtWWe6IYhs0BYjjCmmpGhbQREXpaSEVoJhsEFq4nos0tlvXZ9p7xfV1BVFLBu5UKNPNwchRr3B9DHSvTxkrkmdEUvvwQaN2gf_yBJ0GFWzW_r-frinHYa2i7t1vU8-Xyc-W9zY_DUjEqjIiCvoc0P-CwAA__8Tavg6">