<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/123639>123639</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            SLP Vectorizer miscompilation
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
            miscompilation,
            llvm:SLPVectorizer
      </td>
    </tr>

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

    <tr>
      <th>Reporter</th>
      <td>
          aleks-tmb
      </td>
    </tr>
</table>

<pre>
    During our local testing, we identified a miscompilation where the SLP Vectorizer appears to be the cause. Here is the reproducer we managed to prepare:
https://godbolt.org/z/9nePcq7Go

The issue can be demonstrated using the `lli` tool. Unfortunately, I cannot use alive2 due to unknown libcall errors, such as @llvm.vector.insert.v32i32.v16i32 not being recognized by alive2

```
$ bin/opt -passes=slp-vectorizer -mattr=+avx2 src.ll -S > tgt.ll
$ bin/lli src.ll
Result: -256
$ bin/lli tgt.ll
Result: -272
```

The pseudocode represented by the IR is
```
t = 0
for (i = 0; i < 16; i++) {
    s -= (t + i)
    t = i
}
```
```
s -= (1 + 0) + (2 + 1) + (3 + 2) + (4 + 3) + ... + (14 + 15) + (15 + 16)
s = -256
```
This indicates that the SLP Vectorizer pass produces incorrect IR.

The issue became reproducible after patch `[SLP] Match poison as instruction with the same opcode` made by @alexey-bataev 
https://github.com/llvm/llvm-project/commit/b8703369daf777706196ff914c0376c27adde3cf

I'll provide more details on how the patch affects the issue as soon as I receive them.
</pre>
<img width="1" height="1" alt="" src="http://email.email.llvm.org/o/eJyUVM1u4zYQfhr6MrAgDS0pOuhgr-s2wBYIkm3vFDW22KVIlaSUTZ6-IOU0TraXAgI0GM7PNz_fCO_VxRC1rDyw8rgRcxisa4Wm734bxm7T2f6lPc5OmQvY2YG2UmgI5IMyF4Zf4JlA9WSCOivqQcCovLTjpLQIyhp4HsgRhIHg6esD_EkyWKdeyYGYJhLOQ7DQrQZSzJ4y-C06KJ9UjiZn-1mSi3lGYcSF-ugyOZqEI8b3LN8PIUw-inhieLrYvrM6ZNZdGJ5eGZ4aQw_y7_pXy_Jo_m2I8f0cM5qYvKfRGh-cCNTD7GOpMTmrcq0Vq3II1uoM_jBn68JsRCD9Eku_jwGMDTB7AqHVQgj9TBHfbL4b-2xAq04KrYGcs85HJz_LAYQHtsu1XsZsSS3JlPHkQrZwVByzpagUR4ixO4p4HEl7MeqVeuherrnWaliVX798z3AHnTIMT3YKsJ2E9-QZP3o9bZf31m9HEYJj_MjwIJYfCN7JTGvYPgHjv0C4hEzrD-G0Vlcjlu8fyc86ML6HLZbVT3b_ut_a1fgZ6jqHydPcW2n7ddbkyYS1xjiB-0dQ_pNjAMaPEKWzdcDwTl0V_ABR_AJFlWSGh_Q1wOoDy_cAAB620ZjhXQCGyai5Pq1xVcxWHz-DvZHfQxQpRJ4y4CFqMAnFjYYnAW80uyTwN02WZW8vxfpUlDfWRbnqqhWoTyDf2n6D6tugPCjTKykCRe6I8F-kixsBV0pFe2mdIxng_jH7zI2OpBjfGag6TSDOIQUJcojsYOXh6esDK4_we1JNVnlr4nKrSKdZrhdAhSFh8TGeneKwI6lG0VMcNNvlQtMPetl2Igha4GdGqzDMXSbtmFZsefttJ2f_IhkYnqQdRxWF7q7OOa-aXpzruq7zqmiq87kpdjLndSWxFn1PXJ7Xcu8Z1lrHjiyqJxiti9cgCKU9WAODfU7I14rF-UwyrIdp7ZHw4O1a8X2kKKklnbIx2_Qt7xveiA21Rc1rftcUuNsMbVV2uzNvSsSm2WFdC2rKuhZdmeOu59hsVIs5lnmBeVHlJfKs7wouu1oWXVPyqq7ZLqdRKJ2l82HdZZPAtAXyijcbLTrSPt1zxI_HmCEy_MIQUw_5_unrw_tuxMfyuHFt6mw3X3y8UMoH_54oqKCp_bRTH3NsZqfb_z2-VIFneLoWsbT4TwAAAP__he8IEw">