Missed Optimization: Fold chained GEP with constant base into single GEP

[Leo0506]

define ptr @src(i32 %arg0, ptr %arg1) {
  %v0 = getelementptr inbounds nuw i8, ptr %arg1, i64 8148
  %v1 = icmp sgt i32 %arg0, 3
  %v2 = select i1 %v1, i64 55104, i64 21304
  %v3 = getelementptr i8, ptr %v0, i64 %v2
  ret ptr %v3
}

define ptr @tgt(i32 %arg0, ptr %arg1) {
  %v1 = icmp sgt i32 %arg0, 3
  %v2 = select i1 %v1, i64 63252, i64 29452
  %v3 = getelementptr i8, ptr %arg1, i64 %v2
  ret ptr %v3
}

alive2: https://alive2.llvm.org/ce/z/YvBj6A
godbolt: https://godbolt.org/z/hYW8Mvefj
Pattern found in: https://github.com/dtcxzyw/llvm-opt-benchmark/blob/main/bench/ffmpeg/optimized/msmpeg4enc.ll