Revert "Push rocm to slow path (#46216)" (#46728)

Summary:
This reverts commit bc1ce58.

Fixes #{issue number}

Pull Request resolved: #46728

Reviewed By: cpuhrsch

Differential Revision: D24482783

Pulled By: izdeby

fbshipit-source-id: 619b710a8e790b9878e7317f672b4947e7b88145

aten/src/ATen/native/ForeachUtils.h

@@ -58,10 +58,6 @@ void check_foreach_api_restrictions(TensorList tensors, ArrayRef<double> scalars
 // - Resulting tensor must have the same dtype as the input one
 bool can_use_fast_route(TensorList tensors, Scalar scalar) {
   TORCH_CHECK(tensors.size() > 0, "Tensor list must have at least one tensor.");
-
-#ifdef __HIP_PLATFORM_HCC__
-  return false;
-#else
   auto expected_device = tensors[0].device();
 
   for (auto t : tensors) {
@@ -98,13 +94,9 @@ bool can_use_fast_route(TensorList tensors, Scalar scalar) {
   }
 
   return true;
-#endif
 }
 
 bool can_use_fast_route(TensorList tensors) {
-#ifdef __HIP_PLATFORM_HCC__
-  return false;
-#else
   TORCH_CHECK(tensors.size() > 0, "Tensor list must have at least one tensor.");
   auto expected_device = tensors[0].device();
 
@@ -123,13 +115,9 @@ bool can_use_fast_route(TensorList tensors) {
   }
 
   return true;
-#endif
 }
 
 bool can_use_fast_route(TensorList tensors1, TensorList tensors2) {
-#ifdef __HIP_PLATFORM_HCC__
-  return false;
-#else
   auto expected_device = tensors1[0].device();
 
   for (int64_t i = 0; i < tensors1.size(); i++) {
@@ -161,13 +149,9 @@ bool can_use_fast_route(TensorList tensors1, TensorList tensors2) {
   }
 
   return true;
-#endif
 }
 
 bool can_use_fast_route(TensorList tensors1, TensorList tensors2, TensorList tensors3) {
-#ifdef __HIP_PLATFORM_HCC__
-  return false;
-#else
   auto expected_device = tensors1[0].device();
 
   for (int64_t i = 0; i < tensors1.size(); i++) {
@@ -208,7 +192,6 @@ bool can_use_fast_route(TensorList tensors1, TensorList tensors2, TensorList ten
   }
 
   return true;
-#endif
 }
 
 bool can_use_fast_route(TensorList tensors1, TensorList tensors2, TensorList tensors3, ArrayRef<double> scalars) {
@@ -221,11 +204,7 @@ bool can_use_fast_route(TensorList tensors, ArrayRef<double> scalars) {
   TORCH_CHECK(scalars.size() > 0, "Scalars list must have at least one value.");
   TORCH_CHECK(tensors.size() == scalars.size(), "Tensor list must have same number of elements as scalar list.");
 
-#ifdef __HIP_PLATFORM_HCC__
-  return false;
-#else
   return can_use_fast_route(tensors);
-#endif
 }
 
 }

test/test_foreach.py

@@ -160,6 +160,7 @@ def test_exp(self, device, dtype):
     #
     # Pointwise ops
     #
+    @skipCUDAIfRocm
     @dtypes(*torch.testing.get_all_dtypes(include_bfloat16=False, include_bool=False, include_complex=False))
     def test_addcmul(self, device, dtype):
         if self.device_type == 'cpu':
@@ -261,7 +262,6 @@ def test_max_min_inf_nan(self, device, dtype):
     #
     # Ops with scalar
     #
-    @skipCUDAIfRocm
     @dtypes(*torch.testing.get_all_dtypes())
     def test_int_scalar(self, device, dtype):
         for N in N_values:
@@ -303,7 +303,6 @@ def test_int_scalar(self, device, dtype):
     # We need to update codegen to correctly handle function overloads with float[] and int[].
     # As optimizers work with float tensors, the result will always be torch.float32 for now.
     # Current schema is using 'float[]' as scalar list type.
-    @skipCUDAIfRocm
     @dtypes(*torch.testing.get_all_dtypes())
     def test_int_scalarlist(self, device, dtype):
         for N in N_values:
@@ -452,7 +451,6 @@ def test_float_scalarlist(self, device, dtype):
                 else:
                     self.assertEqual(tensors, expected)
 
-    @skipCUDAIfRocm
     @dtypes(*torch.testing.get_all_dtypes())
     def test_complex_scalar(self, device, dtype):
         for N in N_values:
@@ -516,7 +514,6 @@ def test_complex_scalarlist(self, device, dtype):
                 with self.assertRaisesRegex(TypeError, "argument 'scalars' must be tuple of floats"):
                     foreach_bin_op_(tensors, scalars)
 
-    @skipCUDAIfRocm
     @dtypes(*torch.testing.get_all_dtypes())
     def test_bool_scalar(self, device, dtype):
         for N in N_values: