<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/139274>139274</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            548.exchange2_r has 30% performance regression between LLVM 18/19 and  LLVM 20 on amd64
        </td>
    </tr>

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

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

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

<pre>
    548.exchange2_r built using flang-new-18:

```shell
548.exchange2_r: time (ms) = 181666
548.exchange2_r: clock freq (MHz) = 5701.6834
548.exchange2_r: instructions = 4.05239e+12
548.exchange2_r: branch instructions = 5.80567e+11
548.exchange2_r: ipc = 3.9123
548.exchange2_r: misprediction rate (%) = 0.8200
548.exchange2_r: mpki = 1.1747
```

548.exchange2_r built using flang-new-19:

```shell
548.exchange2_r: time (ms) = 182113
548.exchange2_r: clock freq (MHz) = 5701.3884
548.exchange2_r: instructions = 4.03508e+12
548.exchange2_r: branch instructions = 5.50304e+11
548.exchange2_r: ipc = 3.8863
548.exchange2_r: misprediction rate (%) = 0.7708
548.exchange2_r: mpki = 1.0513
```

548.exchange2_r built using flang-new-20:

```shell
548.exchange2_r: time (ms) = 240537
548.exchange2_r: clock freq (MHz) = 5700.9669
548.exchange2_r: instructions = 5.15502e+12
548.exchange2_r: branch instructions = 5.42299e+11
548.exchange2_r: ipc = 3.7592
548.exchange2_r: misprediction rate (%) = 0.7719
548.exchange2_r: mpki = 0.8121
```

A performance regression of ~30% is observed.

Hardware: Intel Core i9-14900K 5.7 GHz

Full results and compilation flags are available at https://jia.je/benchmark/ (in chinese).
</pre>
<img width="1" height="1" alt="" src="http://email.email.llvm.org/o/eJyklU-P5CYQxT8NvqBGUBgDBx9ms-pMlN1rrhG2q21mMO4Ankn2kM8e2Z3Nv5nO9s5KSJbselW897NULmc_RsSWqHdEva_cWqYltQ8eR-yrbhl-a1VtGP7aTy6OCD8n2q0-FLpmH0d6Ci6Oh4jPB2GIvCN8Pw2_nDxhCITf_acDkXe0-BkpATNnApYS-Z4KI5qmeb26D0v_SE8Jf9k0H-8_fRYpzQVrjKxf1_mYS1r74peY9_qacQXSIoF3Al7XdMnFfnopVcxw1ehdKq6MO_d7qWRWgHy9Zvb5nHDwe2eaXNljIKA-W-LMAOdXxOdHfwmLCV3rf2Z9if5GVvabWYEQVwz-PytpzFewkoqbt7FSXPL6NlbGNG9npTU3X2TF1R7Wm1gB_0ZWUHMl9dez4sw2jb2RlWJCKQ5vY1UDWHsbK63slfY3sRJX_PzFijMjQLxkdUfPmE5Lml3skSYcE-a8DVpO9HfJCSjqM126jOkJB3YR3bs0PLuE24QfYsFAv1sSUm8Porac_0gV0_T7-0-X6uMaAk2Y11AydXGg_TKffXC7n1NwY6YuIXVPzgfXBaSu0KmUc97-DzgSOD54xx6QwLHD2E-zS48EjlsKPtJ-8hEzErCsGlo5WGldha3QdWO4UFxXU2t728m-GbTVDhsntK2NtOJ0Es42J4TKt8BBccWtqKUBzXp0clNpLVU3CEFqjrPzgYXwNLMljZXPecVWSAu6roLrMOR91QBEfKb7VwKwbZ7UbqJDt46Z1Dz4XPLfbYovAV9sosllegn_Cp0OyzNipB8-_PSRCkPgKOye7eUNcLpE6uahqas1hfbfcY6-TGvH-mUmcNxu8ufjcE7LA_aFwHG_f97aXgw-tfBHAAAA__9uzSgc">