<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/150197>150197</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            Unoptimized header masks mixed with VP intrinsics may have different lengths during EVL tail folding
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
            backend:RISC-V,
            vectorizers
      </td>
    </tr>

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

    <tr>
      <th>Reporter</th>
      <td>
          lukel97
      </td>
    </tr>
</table>

<pre>
    As spotted by @Mel-Chen in this review comment: https://github.com/llvm/llvm-project/pull/149981#discussion_r2224826250

Consider an EVL tail folded loop with a VF of 4 and a trip count of 5. With EVL tail folding, it's possible that this will take place with two iterations, one with EVL=3, and one with EVL=2.

A header mask will come in with the form `icmp ule wide-canonical-iv, backedge-tc`.

Most recipes will be converted to a VP intrinsic to use EVL in `optimizeMaskToEVL`. This should really be thought of as an optimisation, but consider a recipe that isn't handled yet or slips through, and so still uses the header mask.

The header mask is generated as `icmp ule wide-canonical-iv, backedge-tc`.

On the first iteration, the mask will look like:

`[0, 1, 2, 3] <= 4 = [T, T, T, T]` 

However for the recipes which were optimized to VP intrinsics, they will have an EVL of 3, so basically a mask of `[T, T, T, F]`. 

On the second iteration, the mask will look like:

`[4, 5, 6, 7] <= 4 = [T, F, F, F]`

But for the VP intrinsics, they will have an EVL of 2 so a mask of `[T, T, F, F]`.

We need to convert the header masks to something of the form `icmp ult step-vector, EVL`, otherwise we end up processing a different number of elements per iteration depending on whether or not it was converted to a VP intrinsic.
</pre>
<img width="1" height="1" alt="" src="http://email.email.llvm.org/o/eJycVUGP4zYP_TXKhUhgy3E8PviQmdng-4AdtGins8dCtphIjSwZopw0--sL2pnNzKBbtHuhAVoiH98jRUVkDx6xEeW9KB8XakwmxMaNR3R1tWiDvjRbAhpCSqihvYBYZ0_olg8GPVgPyViCiCeLZ-hC36NPotiCSWkgUWyF3Am5O9hkxnbVhV7InXOn189yiOEP7JKQu2F0Tshdvq7ru1zIQlvqRiIb_O9RSrm-kxtZZiLbimz7EDxZjRGUh08vnyEp62AfnEYNLoQBzjYZUPCyg7CHNSivQUGKdoAujD6xt1zBFz717r71ByEfwCYhK4IhENnWISSj0lzo2ToHSR0RBqc6nBOlcwCbMKpkgycOEPz116eXz6J4LNjHID745WquZwsGFdfTKzrOObrQI9M7JzAI-xB7EJvMdv0Ao-M4Gped8sHbTrmlPXGSVnVH1Adcpk5ssmv4p0AJInZ2wGsFLUIX_Akja5oCU_UzWJ-i9WQ79oyEEzXWc9IwJNvbr_ik6PgcGPwmW8EzM0ImjE5DROXchQMnE8aDmThWxArNl2liZ8I4Js5-VfAKbObYkheySmCU1w41XDBBiEDODgTJRI78yiUFoMTFjIQ0UfSGxGvlz--9YAkO6Fkp1Izux_j8yc-K2EjpJjwfZ_dNQxfCEZw9Is_BdFNsMlHeZ3w0ZyPZFKJ8BFE8iOIR1sBWlPfP_OetKR_FJoM5zP_CGU8YuSemlN-0NbYzcMaI8KrYJO9bcemK8zJjNOqEr2MU9jB1KgVoFTEL7gJqLijsYQb_AdhuBraCd9QQdsHrH-FmzUdLNhs21Xe52b0xE4Q5yv2YvtHy78uWXPP3K32b5toDXxA8zuxeJ-ljCxL_o9BjMtYfOOzfjHECSjgsT9ilEDnLPFvTG5IMxrMlhDMCeg3jAEMMHRJxPAXa7vcY0SfwY99i5BTokF9gggHjjX7QOKDXEwwPZ4McmgfLB25gOCv6p_dgtdBNoeuiVgts8qos5KbI87uFaYq6KrO8vWvlum1LVWaYlVJWuZatzNYbvbCNzGSZVbLI6iIv71ZFUXe6rtQey6Kr95VYZ9gr61a8DlYhHhaWaMQmL7O8rhZOtehoWk5STsPotSi2v_z_14fli5A8P0LKmT77FSOxr3xcxGZaL-14ILHOnKVEtwzJJofNb_42I-9k6-2fqOeH910HQa8uc-PcmHfoD8kQ6DEyux-XyWKMrvnPq3BigHgZziScGvlXAAAA__9zAHu7">