<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/94419>94419</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            Math function vectorization failure with AVX-512
        </td>
    </tr>

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

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

    <tr>
      <th>Reporter</th>
      <td>
          m13253
      </td>
    </tr>
</table>

<pre>
    I am writing a machine learning software that needs to compute “Y = exp(a⋅X)”.

Sample code:
```c++
#include <cmath>
#include <cstddef>

void func(float a[]) {
    for(std::size_t i = 0; i != 16; i++) {
        a[i] = std::exp(a[i] * 2.0f);
    }
}
```
Expected output:
```asm
push    rbx
mov     rbx, rdi
vmovups zmm0, ZMMWORD PTR [rdi]
vaddps  zmm0, zmm0, zmm0
call    _ZGVeN16v_expf@PLT
vmovups ZMMWORD PTR [rbx], zmm0
pop     rbx
vzeroupper
ret
```
Actual output: Fails to use any vectorized math functions. (See <https://godbolt.org/z/975T6xbss>)

Clang version: 18.1.0

Compilation flags: `clang++ -Ofast -fopenmp -fveclib=libmvec -mprefer-vector-width=512 -march=skylake-avx512`

---

**Alternate 1:** without `* 2.0f`.
```c++
void func(float a[]) {
    for(std::size_t i = 0; i != 16; i++) {
        a[i] = std::exp(a[i]);
    }
}
```
Output: Calls the AVX2 math function, instead of the AVX-512 one.

**Alternate 2:** separate `* 2.0f` and `std::exp`.
```c++
void func(float a[]) {
    for(std::size_t i = 0; i != 16; i++) {
        a[i] *= 2.0f;
    }
    for(std::size_t i = 0; i != 16; i++) {
        a[i] = std::exp(a[i]);
    }
}
```
Output: Fails to use any vectorized math functions.
</pre>
<img width="1px" height="1px" alt="" src="http://email.email.llvm.org/o/eJzUVl9v2zYQ_zT0y0GCRFmy_aAHJ66HActStEHX9SU4iSebKyUSJKU4-fQDJduxs6BA3zqDoHTH4_358cez0Dm564hKlt-wfDPD3u-1Lds043k2q7R4Ln8HbOHJSi-7HSC0WO9lR6AIbRdUTjf-CS2B36OHjkg48Bpq3ZreE7APnC0Ttrr9G1i2AToYxpc4am_YMv_K-OpksolZsmHJepo_Y2sUQa0FseyoY0UyjZrxmzAmLc9kV6teELDstm7R71n24d0154Wg5nV1nActBTR9VzO-bJRGDzjBwfgK2OIYBQCg0ZbxpfMiZJStnXyhRw9yrCxh2U145WmQ0mIUj2m-9RN-IYZk-WbcfHZ5wue0xtfA46QJKGUXDthic6zg_HKCZhI_HAzVngTo3pve_xdBdO2kMb3bB5e2OkyKVg9wUvBbsEIeYWr10BsHL22bhIVvd3d_3X_awMeHT8Dym2CXH5MZUAjj4Gx69RwtalQqBHn89tsX-jMthkc6mIbNk49_PFyHexumOownc-nMaANXJQwvZHVvDNlJtuTfRWld-x7VK0awRalG-vaOALtnGKj22soXEhB4NdLES925GBhffqaRVnvvjQsQ8y3j250WlVY-1nbH-PaF8e1qkT8Uh8q5QDy-uuTercJuBwNZJ3UXMkiXcRonVya6NVJhiAqNwl2IBOEOhK0TwSC6b9B5iBptqGsNRM1AtZIVyzZKVu1ANUStsdSQjaaSoicpwj3Z5CmHqEVbB8F9f1b4nSIcDnnKzzhNcxRFlyLjYayVJ9uhJ0hHBMKAJ-n3uvchyzODiyT-0SX-dS_hT9-9-zOdblEFOu0J1l--8msGBQrLznlCAbo5GUXhOHRH8Y-Q5q9IOzJog-4aasBOBNVVSb_4CfB12DEW8C7c_7Oz_4lWMhNlJlbZCmdUpot0kefJsuCzfZnkS1wsRF6vOC6LKmsQ03nFs7SaU1OJdCZLnvB5UiR5kvLFnMeUU0H1PK0L4k1ByOYJtShVrNTQhpY0k871VK7m83Q1U1iRcuOfP-cdPcG4yDgP3wK2DHuiqt85Nk-UdN69evHSKyrvLus4F3hsVShVb2nsBSdmz3qryjfdUvp9X8W1bhnfBvfHR2Ss_odqz_h2TMqFPjomPZT83wAAAP__KH5oZw">