<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/143219>143219</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            [flang][perf] Adding noalias for function arguments affects function specialization
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
            flang
      </td>
    </tr>

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

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

<pre>
    CPU2017/548 slowed down after https://github.com/llvm/llvm-project/commit/a0d699a8e686cba99690cf28463d14526c5bfbc8, so this change was partially reverted by https://github.com/llvm/llvm-project/commit/d16ecad968244ca42ce09cb3199dea1ec994b211

I could reproduce it on neoverse-v1 with `-Ofast -flto -mcpu=neoverse-v1 -fuse-ld=lld -mmlir -force-no-alias=false/true`.

It looks like the `digits_2` function does not pass the function specialization threshold on the Latency bonus.

Without noalias:
```
FnSpecialization: Specialization bonus {Latency = 2424 (71%)}
```

With noalias:
```
FnSpecialization: Specialization bonus {Latency = 1098 (34%)}
FnSpecialization: No possible specializations found for _QMbrute_forcePdigits_2
```

The current threshold is 40%.

It looks like the big portion of the latency bounds comes from the multiple `mod(row, 3)` expressions, which are not CSEd without noalias - so function specialization computes a lot of bonus for them when `row` is a constant.  With noalias, there is a single `srem` instruction that computes `mod(row, 3)` at the beginning of the function.

I will investigate more to see if the function specialization misses to account for some bonuses.
</pre>
<img width="1" height="1" alt="" src="http://email.email.llvm.org/o/eJysVd2O4zYPfRrlhnAgyz-xL3KRmXwBPqA_W2yLXg5kibLVkSVDlJPOPn1hJ5md2XYKFC0QIBZF8fAcUqIksr1H3LPqgQlx_iKfBxktE4JVx42c0xDi_m7ddEG_7B8__SJ4vmPiVJUNkAsX1KDDxYM0CSMMKU3EigMTJyZOvU3D3G1VGJk4OXe-_2VTDL-hSkycVBhHu3xIruu2lQ3WTa062bZ1y5URTVkXOi8rUauqM51qmHgECpAGS6AG6XuEiySYZExWOvcCEc8YE2roXv5FOjqvUUnd1o0oSyVLoZC3qivyttUoc1RtW3Yizxk_MH74P6gwOw0Rpxj0rBBsguDBYzhjJMzOOVxsGoDVPPvRSEqQGZcCZKOaZlYc3zpmZibMnGbF0TkN2Tg6GyEzISrMfMiks5JYcTTSETJxSnFGVvPtLZUELoRnAmefEdKAC6a2vU30JFjNwcxeJRs86IAEPiSYJNHq-bpFEyornf0i12UaItIQnIZ1gfCdTOjVC3TBz3QD_tWmIcwJfLgluBhZzW8_fjj5z-_CsuIA7y3XeMB2D3cAVhxBlKIEJppdzkTFRMt2x28i39D_e-ict80CXZRvof8q2g8BpkBkO4ffqEdgwuw1mBDh6afvuzgnfFqL-em1LH-i8_OAoOYY0ac36luCkjNRfVjqzvYwhbgSCmY1uddSzV4TqDAigYlhXHfH2SU7ubVJxqCZaGK4LFesWNjWHPD3KSLRwmMxXwarBpAR18Z5_Pw_vfb1m7pDtlzPjzpJhXGaExJIcCEtOV51X8RJA45wGdAvySxp1HxhLEEFT0n6tAV4V2bxuJyJePUi6_srEYo4roc9pTirWwvL9BX9I7YyXWXE3npvfX8X8U7nLjxcrHNg_Rkp2V4mhDFEhBSAEMGav71NoyVCWpylUmH2aWVPYcSrGEjbjd4Xui1aucF9vitbUe1E0WyGPVaV0brsil3V1Hld511TNbIwJZe7tlLlxu4FFxWv-Y7zcpc3W73jpqxMWWMhiqI1rOQ4Suu2y8u3DbHfWKIZ93lZiLzdONmho9tEME76_jYO4n59Kru5J1ZyZynR1xDJJreOkeuJ6siqhwmjYdURDlovSt7bY-H6qoyM_TyiTwTSGFSJPhJtM0e3_8ev-cqMmDjdyJ334o8AAAD__z0tS-Y">