diff --git a/src/CodeGen_LLVM.cpp b/src/CodeGen_LLVM.cpp
index e0f7635e462c..74dda7e5fd34 100644
--- a/src/CodeGen_LLVM.cpp
+++ b/src/CodeGen_LLVM.cpp
@@ -1513,19 +1513,20 @@ void CodeGen_LLVM::visit(const Reinterpret *op) {
 
     } else {
         if (src.is_scalar() && dst.is_vector()) {
-            // If we're converting from a scalar to a vector, first produce
-            // a temporary vector of the required size by splatting
-            // the source scalar. Note that the scalar size may be different.
-            value = create_broadcast(value, (dst.bits() * dst.lanes()) / src.bits());
+            // If the source type is a scalar, we promote it to an
+            // equivalent vector of width one before doing the
+            // bitcast, because llvm's bitcast operator doesn't
+            // want to convert between scalars and vectors.
+            value = create_broadcast(value, 1);
         }
         if (src.is_vector() && dst.is_scalar()) {
-            // If we're converting from a vector to a scalar, first change
-            // element count/type, and then extract the first lane.
-            Type tmp = dst.with_lanes((src.bits() * src.lanes()) / dst.bits());
-            llvm_dst = llvm_type_of(tmp);
+            // Similarly, if we're converting from a vector to a
+            // scalar, convert to a vector of width 1 first, and
+            // then extract the first lane.
+            llvm_dst = get_vector_type(llvm_dst, 1);
         }
         value = builder->CreateBitCast(value, llvm_dst);
-        if (src.is_vector() && dst.is_scalar() && value->getType()->isVectorTy()) {
+        if (src.is_vector() && dst.is_scalar()) {
             value = builder->CreateExtractElement(value, (uint64_t)0);
         }
     }
diff --git a/src/Deinterleave.cpp b/src/Deinterleave.cpp
index 5d46d60bf09e..e368d851d615 100644
--- a/src/Deinterleave.cpp
+++ b/src/Deinterleave.cpp
@@ -322,6 +322,15 @@ class Deinterleaver : public IRGraphMutator {
         }
     }
 
+    Expr visit(const Reinterpret *op) override {
+        if (op->type.is_scalar()) {
+            return op;
+        } else {
+            Type t = op->type.with_lanes(new_lanes);
+            return Reinterpret::make(t, mutate(op->value));
+        }
+    }
+
     Expr visit(const Call *op) override {
         Type t = op->type.with_lanes(new_lanes);
 
diff --git a/src/IR.cpp b/src/IR.cpp
index ffded0cbade1..740234b8e31f 100644
--- a/src/IR.cpp
+++ b/src/IR.cpp
@@ -23,14 +23,8 @@ Expr Cast::make(Type t, Expr v) {
 Expr Reinterpret::make(Type t, Expr v) {
     user_assert(v.defined()) << "reinterpret of undefined Expr\n";
     int from_bits = v.type().bits() * v.type().lanes();
-    int from_vector = v.type().is_vector();
     int to_bits = t.bits() * t.lanes();
-    int to_vector = t.is_vector();
-    user_assert((from_bits == to_bits) ||
-                (from_vector && !to_vector && from_bits > to_bits &&
-                 from_bits % to_bits == 0) ||
-                (to_vector && !from_vector && to_bits > from_bits &&
-                 to_bits % from_bits == 0))
+    user_assert(from_bits == to_bits)
         << "Reinterpret cast from type " << v.type()
         << " which has " << from_bits
         << " bits, to type " << t
diff --git a/src/VectorizeLoops.cpp b/src/VectorizeLoops.cpp
index e2960739ed86..7dcd79d24664 100644
--- a/src/VectorizeLoops.cpp
+++ b/src/VectorizeLoops.cpp
@@ -527,6 +527,16 @@ class VectorSubs : public IRMutator {
         }
     }
 
+    Expr visit(const Reinterpret *op) override {
+        Expr value = mutate(op->value);
+        if (value.same_as(op->value)) {
+            return op;
+        } else {
+            Type t = op->type.with_lanes(value.type().lanes());
+            return Reinterpret::make(t, value);
+        }
+    }
+
     string get_widened_var_name(const string &name) {
         return name + ".widened." + vectorized_vars.back().name;
     }