Support Half/BFloat16 in tril (#7893)

Partial fix for #7748.

Author: swolchok

kernels/portable/cpu/op_tril.cpp

@@ -158,7 +158,7 @@ Tensor& tril_out(
   clear_out(out);
 
   ScalarType out_type = out.scalar_type();
-  ET_SWITCH_REAL_TYPES_AND(Bool, out_type, ctx, __func__, CTYPE, [&]() {
+  ET_SWITCH_REALHBBF16_TYPES(out_type, ctx, __func__, CTYPE, [&]() {
     tril_kernel<CTYPE>(ctx, self, diagonal, out);
   });
 

kernels/test/op_tril_test.cpp

@@ -727,7 +727,7 @@ class OpTrilTest : public OperatorTest {
     test_tril_out_multi_unequal_dim<ScalarType::DTYPE>(); \
   }
 
-ET_FORALL_REAL_TYPES_AND(Bool, GENERATE_GENERIC_TEST)
+ET_FORALL_REALHBBF16_TYPES(GENERATE_GENERIC_TEST)
 
 // Create generic tests for real dtypes. Tensors have diverse values.
 #define GENERATE_REAL_TEST(_, DTYPE)                          \
@@ -738,7 +738,7 @@ ET_FORALL_REAL_TYPES_AND(Bool, GENERATE_GENERIC_TEST)
     test_tril_out_randint_multi_unequal<ScalarType::DTYPE>(); \
   }
 
-ET_FORALL_REAL_TYPES(GENERATE_REAL_TEST)
+ET_FORALL_REALHBBF16_TYPES(GENERATE_REAL_TEST)
 
 TEST_F(OpTrilTest, InvalidInputShapesDies) {
   TensorFactory<ScalarType::Int> tf;