Speedup copysign for half and bfloat16 types (#47413)

Summary:
This also avoids internal compiler error exceptions on aarch64 platforms and transitively fixes #47395

Pull Request resolved: #47413

Reviewed By: walterddr

Differential Revision: D24745921

Pulled By: malfet

fbshipit-source-id: 790e5b91d9116670c882d838b3862d5b47178d68

aten/src/ATen/native/cpu/BinaryOpsKernel.cpp

@@ -803,10 +803,27 @@ void heaviside_kernel(TensorIterator& iter) {
   });
 }
 
+template<typename T>
+T copysign(T a, T b) {
+  return std::copysign(a, b);
+}
+
+// Implement copysign for half precision floats using bit ops
+// Sign is the most significant bit for both half and bfloat16 types
+template<>
+c10::Half copysign(c10::Half a, c10::Half b) {
+  return c10::Half((a.x&0x7fff) | (b.x&0x8000), c10::Half::from_bits());
+}
+
+template<>
+c10::BFloat16 copysign(c10::BFloat16 a, c10::BFloat16 b) {
+   return c10::BFloat16((a.x&0x7fff) | (b.x&0x8000), c10::BFloat16::from_bits());
+}
+
 void copysign_kernel(TensorIterator& iter) {
   AT_DISPATCH_FLOATING_TYPES_AND2(kBFloat16, kHalf, iter.common_dtype(), "copysign_cpu", [&]() {
     cpu_kernel(iter, [](scalar_t a, scalar_t b) -> scalar_t {
-        return std::copysign(a, b);
+        return copysign(a, b);
     });
   });
 }