Dead code produced in non optimized builds

[ghost]

For

bool b()  {
        union U { bool b; int i; }
        U u;
        return u.b;
}

LDC 1.32.1 produces

define zeroext i1 @_D7example1bFZb() #0 !dbg !4 {
  %u = alloca %example.b.U, align 4               ; [#uses = 2, size/byte = 4]
  %1 = bitcast %example.b.U* %u to i8*, !dbg !7   ; [#uses = 1] [debug line = app/example.d:3:9]
  call void @llvm.memset.p0i8.i64(i8* align 1 %1, i8 0, i64 4, i1 false), !dbg !7 ; [debug line = app/example.d:3:9]
  %2 = getelementptr inbounds %example.b.U, %example.b.U* %u, i32 0, i32 0 ; [#uses = 2, type = i8*]
  %3 = load i8, i8* %2, align 1, !dbg !8          ; [#uses = 1] [debug line = app/example.d:4:9]
  %4 = trunc i8 %3 to i1, !dbg !8                 ; [#uses = 0] [debug line = app/example.d:4:9]
  %5 = load i8, i8* %2, align 1, !dbg !8          ; [#uses = 1] [debug line = app/example.d:4:9]
  %6 = trunc i8 %5 to i1, !dbg !8                 ; [#uses = 1] [debug line = app/example.d:4:9]
  ret i1 %6, !dbg !8                              ; [debug line = app/example.d:4:9]
}

%3 and %4 should not be emitted, that's dead code.

[JohanEngelen]

Non-optimized code is by definition sub-optimal. The double load is probably happening to simplify compiler implementation. I doubt whether it is worth spending time to eliminate it, as it will quickly be eliminated in during optimization.

[ghost]

Obviously DCE handles that very well but those two lines are a symptom of an internal bug.

[dnadlinger]

Is it a bug, though, or just slightly suboptimal codegen in unoptimised mode? If you do suspect it's a symptom of an actual defect, I suggest you look through the internals (or -vv) output to see what's actually going on.