<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/82414>82414</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            [InstCombine] Missed optimization : fold `X > C2 ? X + C1 : C2 + C1` to `max(X, C2) + C1`
        </td>
    </tr>

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

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

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

<pre>
    Alive2 proof: https://alive2.llvm.org/ce/z/ERjNs4

### Motivating example 

```llvm
define i32 @src(i32 %x) {
  %add = add nsw i32 %x, 16
  %cmp = icmp sgt i32 %x, 1008
  %s = select i1 %cmp, i32 %add, i32 1024
  ret i32 %s
}
```

can be folded to:

```llvm
define i32 @tgt(i32 %x) {
  %smax = call i32 @llvm.smax.i32(i32 %x, i32 1008)
  %add = add nuw nsw i32 %smax, 16
 ret i32 %add
}
```

LLVM does well when `C1` or `C2` is not constant, but when both are constants, LLVM missed it. Though this example doesn't show better codegen, I think it's a better canonicalization.

### Real-world motivation

This snippet of IR is derived from [protobuf/generated_message_tctable_lite.cc](https://github.com/dtcxzyw/llvm-opt-benchmark/blob/main/bench/protobuf/original/generated_message_tctable_lite.cc.ll@_ZN6google8protobuf13RepeatedFieldIiE14GrowNoAnnotateEii) after O3 pipeline (original IR is from [llvm-opt-benchmark](https://github.com/dtcxzyw/llvm-opt-benchmark)).
Original IR is too big to attach here, email me to get it please.

**Let me know if you can confirm that it's an optimization opportunity, thanks.**
</pre>
<img width="1px" height="1px" alt="" src="http://email.email.llvm.org/o/eJykVV1r4zoQ_TXKy9Bgy47jPOShTepLoe1CWS7lvhTZmtjayhojTZq2v_4i52OTZe-ylwWTWJozI53jo5EKwbQOcSlmN2K2nqgtd-SXz6vuY1KT_lheW_OGEgZPtBHZNXTMQxDZtZCVkJUao1Nr3_op-VbIqkEhq08hq9unb48hF8laJNeHX5ntH3ggNm-KjWsB31U_WIQLYJHsn1h3P6VxYxyCySSIPAm-EbIcB3L2LuQCxPxmD4Q4pbQGka0h_ruwg-_IFaTFGbDphxFo4kto-RKZJOUZNozIgBYbBpMe0iPwkKS0Po7SRObHVI-nsuHAb77-geg5-0Y5qBE2ZDVqYIpi_x9xuOVfihN69T5SaZS1x6TxC8bI1GTyIv1IKCmFXPyXxtvduc6xzoXUZxJEkX5DhPv7vx9AEwbYobWw69CBKJJVKooEyI_vMr6bAI4YGnKBleO4bL3lfUJN3IHyeIqGGB5L9yYE1GB4Cl872rYdcGfCyY1xZSfknCF0tIMamdFDQxpbdLHIXcS7VzAs5DyAOkGUI2caZc2nYkNu-vMD8ITKXu3IWw394SyQO4d-jbsJzgwDMtAG7p4iU43evKGGjacexOxm8MRUbzdCVi069IpRv_QYgmrxhRtWtcUXaxinTSNmayHLy-PbGu629bShXshKc_P--bETsopmuKKBr2p0Tdcr_ypkVVuqhax6ZVwcxYiQ1dkOyJvWOGV_ZzNTa0WevPzzWLRErcXyWCfNnnDAmFoZtPrO3Kb5X552j3TtHLFivDUmelptotxfMhjMgDbaX8jyuIWDXEeZfsLnz8RYCLk4fNkvl0syEdSmBSZQzKrpoEOP0THYK2Ohxxhq43lgGCyqgD9YJD73yBH56mgHZgMftI3GijbeGN8Dd4pPznNAA5v-4DegYSDPW2f4I67KnXKvYbovO9HLTC-yhZrgMp0nZV4kZVFMumWxSIpSzZomSYsizZTS-Xyx2CiVpwuVp3piljKReSJlIhM5z-V0Vs8XZa3zrMFZPctqkScjwdNVMDEhbHFZyjzNJ1bVaMN4x0jpcAdjUEgZrxy_HCWut22IjcgEDt-rsGE7Xk53LvCK-to4FLM1POzP7wX1eD3FrhmbwzOI7BZWEkRWwTMIeQOrdETEuXEUuwdTBI_tqnyOeq3k2DCPgMnW2-UvbDL24INHBk_fsGEhq5FdELIa2f8bAAD__9spOzM">