<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/137147>137147</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            [mlir][vector] Update how OOB accesses are handled by vector.extract/vector.insert
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
            mlir:vectorops,
            mlir:vector
      </td>
    </tr>

    <tr>
      <th>Assignees</th>
      <td>
            banach-space
      </td>
    </tr>

    <tr>
      <th>Reporter</th>
      <td>
          banach-space
      </td>
    </tr>
</table>

<pre>
    At the moment, only `-1` is used to "model" poison values - i.e., accesses that produce poison:

```mlir
  %c_neg_1 = arith.constant -1 : index
  %0 = vector.insert %value_to_store, %dest[%c_neg_1] : vector<5xf32> into vector<4x5xf32>
  %1 = vector.extract %src[%c_neg_1, 0] : f32 from vector<4x5xf32>
```

>From the discussion so far, it seems that:
* We should treat any out-of-bounds (OOB) index — not just `-1` — as producing a poison value (see [@dcaballe’s supportive comment](https://github.com/llvm/llvm-project/issues/134516#issuecomment-2789687910)).
* We should allow using `%poison = ub.poison : index` directly as an index that generates poison.

This direction would bring the dialect more in line with LLVM semantics. From [LLVM LangRef](https://llvm.org/docs/LangRef.html#id184):

> If idx exceeds the length of val for a fixed-length vector, the result is a [poison value](https://llvm.org/docs/LangRef.html#poisonvalues)

**VOICES AGAINST / CONCERNS**

We should be mindful of [@kuhar’s comment](https://github.com/llvm/llvm-project/issues/134516#issuecomment-2781890324). While not opposed to the proposed direction:

> If others would prefer consistency with the LLVM instructions, I won't object.

... we should keep him in the loop and ensure any concerns he raises are addressed.

**PRIOR DISCUSSION**

Relevant context and discussion can be found in:
* https://github.com/llvm/llvm-project/issues/134516
* https://github.com/llvm/llvm-project/pull/135498
* https://github.com/llvm/llvm-project/pull/136579
 
Also, this issue is to track **NEXT STEPS** from this [comment](https://github.com/llvm/llvm-project/pull/136579#issuecomment-2821065107).
</pre>
<img width="1" height="1" alt="" src="http://email.email.llvm.org/o/eJy0Vl1vozgU_TXOy1UQGAjhIQ9p2qwizTajpjOzb5XBl-AZYyPbtOm_X9mQJu3uvky1EhLCvl8-5_hymbXiqBBXJL8hlFZMsbqd257VSCgl-e2MDa7VZnW9M6s0f12tHbgWodMdKkfoBrSSr0AW8TwhixiEhcEiB6eBUNppjpJQCr0WVit4ZnJAC3MQEUbemdU1WosWXMsc9EbzocbJmqRrEodnEY9PJ4Uh8RqA0Lx-Unh8SoCkt8CMcG1Ua2UdUw7mfnUNQnE8nc3jYPiMtdMmEsqicX45FPTk9JN12qCviNCco3UBmHMSkt-GiKM7STf5qUkpSe9AKKcvy9npvHFOm1ynxZMzrA55ranfZ6AbiM9pmpRCY3T3H4Hf4BjR2XpLTwkXth6sFVqB1dAw44MKBxaxGwGeEKVr-IFgWz1IDs4gc8DUK-jBzXUzr_SguAVCl_v9DaHliCOQO0qWMSkzUNrBz8G6C-mXPWYnEoU6AntHu49oEcGfO4t5zSomJZ5dSwt26HttnHhGqHUX5JXfErpsneutL51uCd0ehWuHKqp1R-hWyufza94b_RNrR-hWWDugJXSbpFmeLAhNw8oUdE6LZblYFmUSE1oSWkb_wIRJqV9gsP4QHmeaTwfxZA5V9PZ1FtkiBi4M1k6-egiYmkALqj6iQsMc2gmOaOTtsRV28vKUvYTMlfE5RzaZxNpBpw2CUCCFQngRroUvX77_CRY7ppyobQRBACS_CetfmDo-YPNvyHmUIm2OhG65rj0-k3HUuk56lHiyzDwmbxcvvYNdA4KfAE81IrehNInq6FrQjScWGm2AQSNOyOfTzqRbugnmBu0gne8MzJd5rYnfKnMMMLYSX208iZrQ9ff9bnN3gPUf69394REI3cJmf7-5e7g_jAaj8YXpCqETijeD9McZpflraJm50uX_LcZkWcYp9cBH8KMVEsMN032vp0bqQeyNHr_fFPORJe1aNHbSUW-wQQO-JQrrUNWvo3Z8qKAToawzQwhkPVE7eNGK0MKBrnzlk0ijKIKXN7R-IfbQis7rMQhB6x6Y4oDKDgZDF6m1qtEoC555JnxvZ36Lc-M7PY-u-fr6sNs_wO3usPl2OOz299ckPaDEZ9_Oa60cnlxIdNXjaqY8fY1vVyDUpbd9mqLfDNMPUoYgeVYuPx1kkRde2kDi9VpaPV4mYSEU6y-TF4Zh9S8YQbu_--sRDo93Xyelj3-Q4ELym89p-H1RHxW8pEm8yJO48Aqe8VXKy7RkM1wlRZZny2JRFrN2taw4TyqaIo8xL5dNlhcFK5s6YaxkSZLOxIrGNI8zmsVllqRFlGdVlVVxVjZJwReckyzGjgkZnTvELNSxStIiyYqZZBVKO000YVJI12Mj0r31Uw3dfNyYZh2zCuethqMlWSyFdfaSwwknw5wUPPNbkt9MzvktfOs5cwitfoH9_uYyzXjFt0xxiRyq1w__f0K37-aQ2WDk6jOiDcd_XtG_AwAA__-Snx1U">