Missed Optimization: Fold boolean-indexed array GEP into select of base vs. element pointer

[Leo0506]

define ptr @src(i32 %arg1, ptr %arg0) {
  %v0 = add i32 %arg1, -1
  %v1 = icmp ne i32 %v0, 0
  %v2 = zext i1 %v1 to i64
  %v3 = getelementptr [2 x ptr], ptr %arg0, i64 0, i64 %v2
  ret ptr %v3
}

define ptr @tgt(i32 %arg1, ptr %arg0) {
  %v0 = icmp eq i32 %arg1, 1
  %v1 = getelementptr [2 x ptr], ptr %arg0, i64 0, i64 1
  %v2 = select i1 %v0, ptr %arg0, ptr %v1
  ret ptr %v2
}

alive2: https://alive2.llvm.org/ce/z/7Z-Z-Q
godbolt: https://godbolt.org/z/9asvWEPsP
Pattern found in: https://github.com/dtcxzyw/llvm-opt-benchmark/blob/main/bench/linux/optimized/gen8_ppgtt.ll

[Leo0506]

Sorry, this is a misread on my side. This isn’t a missed optimization. Closing as not a missed optimization, apologies for the noise.