<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/112878>112878</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            [mlir] Inconsistent results for linalg.transpose
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
            mlir
      </td>
    </tr>

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

    <tr>
      <th>Reporter</th>
      <td>
          wangyongj1a
      </td>
    </tr>
</table>

<pre>
    I have the following MLIR program:
test.mlir:
```
module {

 func.func @func1() -> f32 {
    %true = arith.constant true
 %c-4534883404942705875_i64 = arith.constant -4534883404942705875 : i64
 %c-30978575762189800_i64 = arith.constant -30978575762189800 : i64
 %c-169620577492_i64 = arith.constant -169620577492 : i64
 %c-63283138746_i64 = arith.constant -63283138746 : i64
    %from_elements = tensor.from_elements %c-4534883404942705875_i64, %c-30978575762189800_i64, %c-169620577492_i64, %c-63283138746_i64  : tensor<2x2xi64>
    %cast = tensor.cast %from_elements : tensor<2x2xi64> to tensor<?x?xi64>
    %4 = scf.if %true -> (tensor<?x?xi64>) {
      scf.yield %cast : tensor<?x?xi64>
    } else {
      scf.yield %cast : tensor<?x?xi64>
    }
    %transposed = linalg.transpose ins(%4 : tensor<?x?xi64>) outs(%cast : tensor<?x?xi64>) permutation = [1, 0]
 %collapsed_8 = tensor.collapse_shape %cast [[0, 1]] : tensor<?x?xi64> into tensor<?xi64>
    %collapsed_10 = tensor.collapse_shape %transposed [[0, 1]] : tensor<?x?xi64> into tensor<?xi64>
    %idx0 = index.constant 0
 %idx1 = index.constant 1
    %idx2 = index.constant 2
    %idx3 = index.constant 3
    %extracted0 = tensor.extract %collapsed_10[%idx0] : tensor<?xi64>
    vector.print %extracted0 : i64
    %extracted1 = tensor.extract %collapsed_10[%idx1] : tensor<?xi64>
    vector.print %extracted1 : i64
    %extracted2 = tensor.extract %collapsed_10[%idx2] : tensor<?xi64>
    vector.print %extracted2 : i64
    %extracted3 = tensor.extract %collapsed_10[%idx3] : tensor<?xi64>
    vector.print %extracted3 : i64

    %111 = arith.sitofp %extracted2 :i64 to f32
 return %111 : f32
  }
}
```
When I ran ```/data/tmp/v1018/llvm-project/build/bin/mlir-opt --one-shot-bufferize="dialect-filter=tensor,linalg,bufferization" --convert-index-to-llvm --convert-linalg-to-loops --convert-vector-to-llvm --expand-strided-metadata --convert-scf-to-cf --lower-affine --one-shot-bufferize="dialect-filter=tensor" --func-bufferize --one-shot-bufferize="dialect-filter=arith" --convert-arith-to-llvm --finalize-memref-to-llvm --convert-func-to-llvm --reconcile-unrealized-casts test.mlir | /data/tmp/v1018/llvm-project/build/bin/mlir-cpu-runner -e func1 --shared-libs=/data/tmp/v1018/llvm-project/build/lib/libmlir_runner_utils.so,/data/tmp/v1018/llvm-project/build/lib/libmlir_c_runner_utils.so``` on the program, I got the result of:
```
-4534883404942705875
-169620577492
-30978575762189800
-63283138746
-3.097857e+16
```
However, when I ran ```/data/tmp/v1018/llvm-project/build/bin/mlir-opt --linalg-generalize-named-ops --one-shot-bufferize="dialect-filter=tensor,linalg,bufferization" --convert-index-to-llvm --convert-linalg-to-loops --convert-vector-to-llvm --expand-strided-metadata --convert-scf-to-cf --lower-affine --one-shot-bufferize="dialect-filter=tensor" --func-bufferize --one-shot-bufferize="dialect-filter=arith" --convert-arith-to-llvm --finalize-memref-to-llvm --convert-func-to-llvm --reconcile-unrealized-casts test.mlir | /data/tmp/v1018/llvm-project/build/bin/mlir-cpu-runner -e func1 --shared-libs=/data/tmp/v1018/llvm-project/build/lib/libmlir_runner_utils.so,/data/tmp/v1018/llvm-project/build/lib/libmlir_c_runner_utils.so``` on the program, I got the result of:
```
-4534883404942705875
-169620577492
-169620577492
-63283138746
-1.696206e+11
```
The above two results seem to be inconsistent. I'm not sure if there is any bug in my program or if the wrong usage of the above passes caused these results.
My git version is 18ac0178ad7f839bdb3376b7a0c6b69a24520a06.
</pre>
<img width="1px" height="1px" alt="" src="http://email.email.llvm.org/o/eJzsWF9vqzoS_zTOywiEbf4-5KFtbrSV9r6sVtrHysCQ-ArsyDZpej79ykATktDu9vSetyO1JJk_P_9mmLEHhLVypxDXJHkkyWYlerfXZv0q1O5Nq91fVKxKXb-tn2Evjghuj9DottWvUu3gz38-_wsORu-M6Ah_INGGRA8OrQu7VpqzhKTR9Df87HTdtwgke5zUwxWaXlWhvwCJI_9JCcsJKyAg_A9oOLt4AAAQljjTIxC-AWGk24eVVtYJ5cDLJ0PCkiqIEx7nOY-juIhZFiV5lrzINF5yXbIFwh9ApvEckkdFlidZkqWM5kUeRR8C3lkuwdG0SFmUZFlcsA-R5kZLIClnOac8z-L0Q4yZzQ3EmNLG6O4FW-xQOTsgOFRWm_BG8UlaCXv6JEdn7W3IZ8VtGAPPkQXhT-zETt6a_3HFuxLWzemOv-_jWUQCpy9iwrcn_7-wyJhSWzWhbM71N1QnYflHAKy4rlsYAN4ktvWM-MP_QSDbALYW_z64m2YSyh60xXqIspVKtLvwLAWp7NCPQxY-xmcF6N5Npv-TDSvggKbrnXBSq2FhkjxSXwsRSTaz6tZtKw4W65f86jZP4he7Fwe8JMDvZY-Rh6EeJtl8xgKkui2ApQo7M6DR5xTmmfwFRGR9GglIVePp0tvRJVuyPtElE3oLxJas2K0VX7LiV1Z4ckZUDuur3EzS2_T5rIxxLCbkNuYjVk6b8GCkcndr3W9hZzX9ChX6PSr0cyrsK1TY96jcng03VPhXqPDvUeHXVK4IUUpnJ5SVTjeH-zj8AeC0P_0nZ4OuN-oC8DDTXfa0y5fr0eM_e1TwDEYouGjYthZOELZ13YGw7ZFGNCds27bHLjgY_RdWjrBt2cu29p9SEbb1802gDw6CQCsM7F67oOybBo38gYRvCGO1FC1WLmhk69AQvplyyJ7GvZWwp3ePYfcjjEEQVFod0bhg6LbA6cDTmMlH30Gh9cHONONdmLng6SBUHVhnZI110KETPtCZi60ab181EAStfkUTiKaRCn8mKk_eT20Xh6-gDFVwnYFBNAun8ZHLHxh02BlsFnIzLH8RG6y0qmSLQa8MDr514M8HC-cZFUjmx46frYDq0AemVwoNBDhMsBSCwO6FwTpoZWmHcL-E3spyvPoFXkbwl97J1oZWE_b0LbjqDvC9C0CrYbh_n-bZEzzDTrtBaND2rQPdfDTSL82Ck2Y-502iu8Fwks8mv3fLcDRFwh5purj0P_QrHtG3Fbz-gu6e-m2HCs1Yfkp0WAdj7_3u_d-9_7v3v9L7C6L7tqfhYJUObU8XV_33HkGU-ojgXvVE0oJF7Py8UPrnFT-qSutQuRCeCcs6UNqB7Q2CbHxs_osFod6g7HcgFXRv7zkAbSYjeDVa7aC3YoegR9G48EFYixYq0ftR3-3RvmfLhiPJP99gJx0c0Vj_eCMt0FxUEc1yUWdNzouyLjnP0jITUZWWaSFYnLBIRGm4qte8LnghVrimGStYnOUJXe3XSUSTPIloE9MkKxqRZFiXvGkYK5oqRlzJNYtY7CuDRjzjNOSUYRpVSZanRcq5IHGEnZBt6Msm1Ga3ktb2uKaU5Vm-akWJrR3eBzE2vMdhfiJdmfVQZmW_sySOWmmdvSA46drhHdLgkGzgeZb-8-1ptLl7sFz1pl3vnTtYX2BsS9h2J92-L8NKd1Nt35f4wNgStp1IH9fsvwEAAP__wb2vIA">