[InstCombine] Missed optimization: fold chained selects with exclusive condition but one identical arm

[XChy]

Alive2 proof: https://alive2.llvm.org/ce/z/ZkjFRM

Motivating example

define ptr @src(i32 %level, ptr %level.addr) local_unnamed_addr #0 {
entry:
  %c1 = icmp eq i32 %level, 0
  %c2 = icmp eq i32 %level, 1
  %s1 = select i1 %c1, ptr %level.addr, ptr null
  %s2 = select i1 %c2, ptr null, ptr %s1
  ret ptr %s2
}

can be folded to:

define ptr @tgt(i32 %level, ptr %level.addr) local_unnamed_addr #0 {
entry:
  %cond = icmp eq i32 %level, 0
  %s1 = select i1 %cond, ptr %level.addr, ptr null
  ret ptr %s1
}

Real-world motivation

This snippet of IR is derived from openssl/crypto/shlib/cmp_util.c@OSSL_CMP_print_to_bio (after O3 pipeline).
The example above is a reduced version. If you're interested in the original suboptimal IR and optimal IR, see also: https://godbolt.org/z/qrh916jj8

Let me know if you can confirm that it's an optimization opportunity, thanks.

[marcauberer]

I would like to work on this

[marcauberer]

This can be further generalized to: https://alive2.llvm.org/ce/z/pnK3jW

[sftlbcn]

duplicate of #82350?

[XChy]

duplicate of #82350?

Oh, it's a duplicate. Close it now.