Support Half/BFloat16 in split_copy (pytorch#7901)

Partial fix for pytorch#7748.

Author: swolchok

kernels/portable/cpu/op_split_copy.cpp

@@ -58,10 +58,10 @@ void split_copy_Tensor_out(
   ScalarType in_type = input.scalar_type();
   ScalarType out_type = out[0].scalar_type();
 
-  ET_SWITCH_REAL_TYPES_AND(
-      Bool, in_type, ctx, "split_copy.Tensor_out", CTYPE_IN, [&]() {
-        ET_SWITCH_REAL_TYPES_AND(
-            Bool, out_type, ctx, "split_copy.Tensor_out", CTYPE_OUT, [&]() {
+  ET_SWITCH_REALHBBF16_TYPES(
+      in_type, ctx, "split_copy.Tensor_out", CTYPE_IN, [&]() {
+        ET_SWITCH_REALHBBF16_TYPES(
+            out_type, ctx, "split_copy.Tensor_out", CTYPE_OUT, [&]() {
               const CTYPE_IN* input_data = input.const_data_ptr<CTYPE_IN>();
               for (size_t i = 0, e = out.size(); i < e; ++i) {
                 size_t out_step = out[i].size(dim) * trailing_dims;

kernels/test/op_split_copy_test.cpp

@@ -298,7 +298,7 @@ TEST_F(OpSplitCopyTensorOutTest, LargerSplitSizeDoesNothing) {
 
 TEST_F(OpSplitCopyTensorOutTest, AllDtypesSupported) {
 #define TEST_ENTRY(ctype, dtype) test_dtype<ScalarType::dtype>();
-  ET_FORALL_REAL_TYPES_AND(Bool, TEST_ENTRY);
+  ET_FORALL_REALHBBF16_TYPES(TEST_ENTRY);
 #undef TEST_ENTRY
   // TODO: Also add tests for half, complex, quantized, and other types. Easiest
   // way to do that would be to make TensorFactory support zeros() and ones()