<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/78144>78144</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            some wierd thing about wrapping tensor.
        </td>
    </tr>

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

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

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

<pre>
    Hi there~ I found a bug(i think) in the performance of wrapping “`arith.constant dense<0.0> :  tensor<512x512xf32>`” in  a function like 
    `func.func @create_tensor0() -> tensor<512x512xf32> {`
    `    %cst = arith.constant dense<0.0> :  tensor<512x512xf32>`
    `    return %cst : tensor<512x512xf32>`
  }`
 , and to call it as %output in linalg.matmul's output arg.
Warp or not give me the different result, which confuse me a lot.
See detail in discourse [page](https://discourse.llvm.org/t/why-wrapping-the-creation-of-arith-constant-in-a-function-will-introduce-difference/76143) please.
    I tried in main branch，still act like the same.
    In main branch, the compile cmd shoud delete the "allow-return-allocs" part.
</pre>
<img width="1px" height="1px" alt="" src="http://email.email.llvm.org/o/eJyklE-PszYQxj_NcBmBjIFADhx2k4363ir10GPl2AO4r7GR_2z6XvrZK5PNplupUqUewLJm_Juxn0cjQtCzJRqhe4XuXIgUF-fHnz2FcJmcN-6diqtTP8afNMaFPP2J33ByySoUeE0z8CEHtP0O_Ija5iTcyE_Or8JKQjfhzYtt03ZGeOMwMDie4MCE13GppLMhChtRkQ0EzYlVDJo3hOYFMZINzkNz6mr-R_6mhkPzBgf2AJ1zRRQ4JSujdhaN_k4I7AzsBRERDiyHqvxDaJn0JCL9dgcz4ENuuswF_6UWQv-a6_2duC-8kyEiNGf8fxf5J9hTTN4--S__4TD05-cO-AmFVRgdSmEM6ogiZJ5LcUsxP5jRVpi5WkVckwHeB_yICT9Xd8yvwm_oPFoXcdbvhCvtyio9TeTJRvQUkom52m3RckHp7JTCnijQuPgB-oUIFUWhTa6sdJAu-UAI3esmZoLuDHxYYtwCNC_AL8Avn0mVMe9r5fwM_BKBX27Lj_LhpTIuVO5yamdLN5W7DOVDhlLbUpQPW5Q3bUypbfROJUnl4xaSgF_6Q9022QebIRGoekryDaPXpHLjq9AWr15YucDbBV5PMJxC1MagkPHuuvw8QaxfAF8P8tOeJN26aUMoV4VhcUmhIkPxTgDOhTHuVt6NUOaNDMA5bsLHqlBjo47NURQ01j3rmro_dkOxjN3A64YfqCfB2XCd1NTUvG_lUTHZirYu9MgZb1ldd3XNu26oWMv6YejkoWu4oraHltEqtPl89UKHkGjsh7ptCyOuZMI-Jji3dMM9CJznqeHHfKa8pjlAy4wOMTwpUUdDY3Ar4U2TV_u0mFFcXYrP0XB3eVUkb8avdph1XNK1km4FfsnUj6XcvPudZDbG3kvIWuZe_woAAP__AzaD2A">