Missed Optimization: simplify smin difference to signed less-than

[Leo0506]

define i64 @src(i64 %arg0, ptr %arg1) {
  %v0 = load i64, ptr %arg1, align 8
  %v1 = tail call i64 @llvm.smin.i64(i64 %arg0, i64 %v0)
  %v2 = add nuw nsw i64 %arg0, 1
  %v3 = tail call i64 @llvm.smin.i64(i64 %v2, i64 %v0)
  %v4 = sub i64 %v3, %v1
  ret i64 %v4
}

define i64 @tgt(i64 %arg0, ptr %arg1) {
  %v0 = load i64, ptr %arg1, align 8
  %v1 = icmp slt i64 %arg0, %v0
  %v3 = select i1 %v1, i64 1, i64 0
  ret i64 %v3
}

alive2: https://alive2.llvm.org/ce/z/LAm4bL
godbolt: https://godbolt.org/z/Y4r9499jv

Pattern found in: https://github.com/dtcxzyw/llvm-opt-benchmark/blob/main/bench/cpython/optimized/arraymodule.ll

[benwu25]

Hi, unless someone else wants to tackle it or this is not something that should be implemented, could I be assigned to this issue? I will try for a few days and request to be unassigned if I struggle to make progress.