<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/150302>150302</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            [MLIR][TOSA] Folding cast to bool gives wrong value
        </td>
    </tr>

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

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

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

<pre>
    I'm fixing what I believe to be a bug in StableHLO with our type convert folder, and I believe it may exist in other dialects as well?

Looking at https://www.mlplatform.org/tosa/tosa_spec.html#_cast:

```
if (is_same<out_t,bool_t>()) {
    out = (in != 0) ? true : false;
}
```

However with a lack of special casing bools Convert / Cast-like ops these are folded incorrectly:

```
$ cat /tmp/t.mlir 
func.func @const_f32_bool() -> tensor<i1> {
  %cst = arith.constant dense<10> : tensor<i32>
  %0 = tosa.cast %cst : (tensor<i32>) -> tensor<i1>
  func.return %0 : tensor<i1>
}
$ mlir-opt /tmp/t.mlir --canonicalize
module {
  func.func @const_f32_bool() -> tensor<i1> {
 %0 = "tosa.const"() <{values = dense<false> : tensor<i1>}> : () -> tensor<i1>
    return %0 : tensor<i1>
  }
}
```

Input was `10 != 0` so probably should be folded to `true`(?) Not sure if there are other cast-like ops in upstream dialects that may have this issue as well?
</pre>
<img width="1" height="1" alt="" src="http://email.email.llvm.org/o/eJykVMFu4zYU_Jrny4MNirTs-KCDYq-7QdMu0O3doKgniw0lCuSTvenXF5SVTTaboocChgTBnOGb4XB0jPbcExWQ30N-WOiRWx-KXxzp6PtfF5Wvn4sHkNsOG_vN9me8tprxAStyli6E7LEi1FiNZ7Q9fmVdOfr8-AWvllv0Y0B-HgiN7y8UGBvvagog96j7-g2NZez0M9I3GznxeG4pYG21I8MRdcQrOQfqCKIEUT56_5SG0Ywt8xBBlSCPII_X63XVucFpbnzoVj6cQR7ZRz2_TnEgs2q5cyDVyejICTpxwkbMP1HaBkHe2XiKuiNQez_yiUHuK-_diUF9AnkHcgdyh7C9B1EiIvqREdRhQvYIMksfYlqjjshhJARVYqNdJFAJBdvDu41BlJ_9lS4UbgZqdNo8oW8wDW61Q6NjUp4mibifbQV5xL2OvHT2idAPEbmlSKgD3Syv0fbGh0CG3fOHkkGu0eiJirshPVedswFBlM3Ym1V6IKyF8X3kU6PkKY1w8wGXoD4hUx99ALW3Wfp8MQZkbuLNGR0st6uJQfeMNfXJiX0mpvWqfEOhZHJ5xosJnc5vlY7slbJMZr8HfTzPxDUpCcRj6F94y5_Xzcci15gcWPrhZ1eWS6N731ujnf2bQJSdr0dH30X_L8u-KwYpb6ITAUg5Q0HtYXt_0W6kOK17MXKO1nsvJ1nbw8sf_z7AHOT_NAhxtuij_D70w8h41RFhIzLxehE2AqPHIfhKV-4ZY-tHV6f2mBPKPiHSPUlc8i7ddrnD3z1jHAOhbVKqwy3Vt4IwP2Te9jgOkQPp7rU6OPVV6pZWp7pqbUQb40hvOmVRF6reqZ1eUJFtc6WE2GZy0RaVUtum2oi1yfJNI2ijKVNql9ekiOo1LWwhhczFVqpsl98puWpImuyOZKOVVGZXw1pQp61bOXeZ6mgxbV5kuVBCLpyuyMWpfKVMwQJV3tpKpjYORYItq_EcYS2cjRxfidiym3r7t8eHPyA_QH7_55evJeQHPHpXp5KYLktqaO8dnu2FIl6D7884ZWcxBlf8WJ9ny-1YrYzvQB7TTvNrOQT_FxkGeZzmjyCPs4RLIf8JAAD__yPS5oI">