[clang][Interp] Fix zero-initializing of floating types

[tbaeder]

Summary

After the discussion in https://reviews.llvm.org/D141472, this fixes zero-initializing members of floating type

[aaron.ballman]

LGTM with a suggestion.

clang/lib/AST/Interp/ByteCodeExprGen.cpp

956–960

Desperately needs reformatting, but can do this in a one-liner that's actually a bit more clear.

clang/lib/AST/Interp/ByteCodeExprGen.cpp

bool ByteCodeExprGen<Emitter>::visitZeroInitializer(PrimType T, const Expr *E) {
   case PT_Sint64:
     return this->emitZeroSint64(E);
   case PT_Uint64:
     return this->emitZeroUint64(E);
   case PT_Ptr:
     return this->emitNullPtr(E);
   case PT_FnPtr:
     return this->emitNullFnPtr(E);
-  case PT_Float:
-    assert(false);
+  case PT_Float: {
+    return this->emitConstFloat(
+        APFloat::getZero(Ctx.getFloatSemantics(E->getType())), E);
+  }
   }
   llvm_unreachable("unknown primitive type");
 }
 
 template <class Emitter>
 bool ByteCodeExprGen<Emitter>::dereference(
     const Expr *LV, DerefKind AK, llvm::function_ref<bool(PrimType)> Direct,
     llvm::function_ref<bool(PrimType)> Indirect) {
   if (std::optional<PrimType> T = classify(LV->getType())) {

clang/test/AST/Interp/floats.cpp

namespace compound {
 
   constexpr float f2() {
     float f = __FLT_MAX__;
     f += 1.0;
     return f;
   }
   static_assert(f2() == __FLT_MAX__, "");
 }
+
+namespace ZeroInit {
+  template<typename FloatT>
+  struct A {
+    int a;
+    FloatT f;
+  };
+
+  constexpr A<float> a{12};
+  static_assert(a.f == 0.0f);
+
+  constexpr A<double> b{12};
+  static_assert(a.f == 0.0);
+};