<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/63750>63750</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            [lld] xxHash performance - XXH_X64 vs XXH3_X64
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
            enhancement,
            lld,
            performance
      </td>
    </tr>

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

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

<pre>
    Currently for hashing (symbol names, ...) in lld we use llvm::xxHash64 that utilize https://github.com/Cyan4973/xxHash implementation.
In 2020 Yann Collet released stable version of XXH3 algorithm, and in 2021 it included additional performance optimizations.

This algorithm is described as:
`XXH3 (since v0.8.0): generates 64 or 128-bit hashes, using vectorized arithmetic.`

When linking debug version on clang binary on single thread we get:
```
 Time (mean ± σ):      7.715 s ±  0.143 s    [User: 5.685 s, System: 1.782 s]
 Range (min … max):    7.532 s …  8.375 s    50 runs
```
And xxHash64 according to perf takes around ~6.81%

When using XXH3_64 we get:
```
 Time (mean ± σ):      7.429 s ±  0.157 s    [User: 5.384 s, System: 1.810 s]
 Range (min … max):    7.256 s …  8.247 s    50 runs
```
and hash function according to perf takes ~1.32%.

This mean that there is +- cheap way to gain 4-5%.
Maybe it's time to switch ?
</pre>
<img width="1px" height="1px" alt="" src="http://email.email.llvm.org/o/eJyslM9u4zYQxp-GvgwsUENRfw46OE6NLdACRXeLpr0sKGkssUuRBkklcQ777AXlNM422xYF1jBsyyS_mfnN8FMh6NEStUzeMHm7UUucnG9_0i7639__sOnccG73i_dkoznD0XmYVJi0HYFhHc5z5wxYNVNguIcsyxg2oC0YM8ADwRIIjLmfmdgxsXt8fKfCVBYQJxVhidroJ4IpxlNIG_DA8DDqOC1d1ruZ4WF_VrZoKsHwcDkLej4ZmslGFbWzGeO3jO--t4AcOfymrIW9M4YieDKkAg0QouoMwT35oJ0Fd4S7u3cClBmd13GaU97KDilp5JiDjqBtb5aBBlDDoFMcZeBE_uj8rGxP4E5Rz_ppTSE853D5_DDpcJUGHWCg0HvdJbG1yMvmkq9JJIQ6Kd7zrM44w4aJHYxkyatIAcoCnIcc622n4wr-wnkJqQH31Efn9VPSXuNR1H3GSv46o18nsmC0_ZRODNQt4xWFhd4oO0KnrfLn9Jx0DUGcPKm1gSPF11k_v9dH-KBnSiXMpCywPbKbHNj-wGrxXMj6qrIqlxD-2gA8ywsBIS0xefNLIJ-2yqysJazFvT-HSGliIM-qGiEwefsc8Wdlx0tIbYF9h6zmbFfCrB6vEatMCkzxXpahzkQlLyElB7_Y8NWCdnaAlxFVfe_8kKBFtzYfovpEAZR3ix3gc5nVOUP5BvWlNam7H8viWxEssPmSoKy-QlDUxVuCdc7_P0GU5d8JYlH9N8F0j9KUwnGxfbod_4jxc54JZCjf3p6VxWoQcSJP6Q4xvNlCP5E6wYM6J6VRaQvFVl4VflTnjkBHhlWAmLhGB-FBx34CJg6boRVDIxq1oTYv67qqUCLfTG3XN7yjQfZ9pxpeH2WToyiOAx1VKSrZb3SLHAWveIUC64JnVMuGjuIoyrITUgpWcJqVNlkyusz5caNDWKhNx_nGqI5MWO0VkeyUDCT5F0NkuGeIxgwvv1-ZTPpP3m58m1S33TIGVnCjQwzXOFFHszp30pC3z8P7hVVt0yh-vCsLuA-Xqbwri83iTfsvtrs69uVre_LuD-ojw8NaVWB4WAv7MwAA__9h19bi">