kernels/test/op_select_copy_test.cpp - platform/external/executorch - Git at Google

 /*
  * Copyright (c) Meta Platforms, Inc. and affiliates.
  * All rights reserved.
  *
  * This source code is licensed under the BSD-style license found in the
  * LICENSE file in the root directory of this source tree.
  */

 #include <executorch/kernels/test/FunctionHeaderWrapper.h> // Declares the operator
 #include <executorch/kernels/test/TestUtil.h>
 #include <executorch/kernels/test/supported_features.h>
 #include <executorch/runtime/core/exec_aten/exec_aten.h>
 #include <executorch/runtime/core/exec_aten/testing_util/tensor_factory.h>
 #include <executorch/runtime/core/exec_aten/testing_util/tensor_util.h>
 #include <executorch/runtime/core/exec_aten/util/scalar_type_util.h>

 #include <gtest/gtest.h>
 #include <sys/types.h>

 using namespace ::testing;
 using exec_aten::ArrayRef;
 using exec_aten::ScalarType;
 using exec_aten::Tensor;
 using torch::executor::testing::TensorFactory;

 class OpSelectCopyIntOutTest : public OperatorTest {
  protected:
   Tensor& op_select_copy_int_out(
       const Tensor& self,
       int64_t dim,
       int64_t index,
       Tensor& out) {
     return torch::executor::aten::select_copy_outf(
         context_, self, dim, index, out);
   }

   template <class CTYPE, exec_aten::ScalarType DTYPE>
   void test_dtype() {
     TensorFactory<DTYPE> tf;

     // Based on the split defintion, if we split any dim()=3 and size(1)=2
     // tensor along first dim to two tensors [ret_0, ret_1], the ret_0 and ret_1
     // shall be equal to x[:, 0, :] and x[:, 1, :] e.g. x[i, 0, j] = ret_0[i, j]
     // and x[i, 1, j] = ret_1[i, j] for any i in [-x.size(0), x.size(0)) and j
     // in
     // [-x.size(2), x.size(2))
     // Therefore we design the following tensor x for test easily: it is a
     // tensor formed by stacking tensors ones(3, 4) and zeros(3,4) along the
     // first dim. So if we select the tensor along the first dim by the above
     // rules, the ret_0 should be ones(3, 4) and ret_1 should be zeros(3, 4)

     // clang-format off
     Tensor x = tf.make(
         {3, 2, 4},
         {
           // all ones below are from x,
           // and all zeros are from y.
           // [0, :, :]
           1, 1, 1, 1, // [0, 0, :]
           0, 0, 0, 0, // [0, 1, :]

           // [1, :, :]
           1, 1, 1, 1, // [1, 0, :]
           0, 0, 0, 0, // [1, 1, :]

           // [2, :, :]
           1, 1, 1, 1, // [2, 0, :]
           0, 0, 0, 0, // [2, 1, :]
         });
     // clang-format on

     // Expected values for out_0 and ret_0 after the test are all ones(3, 4)
     // based on the above rules. So here we set the default value of out_0 as
     // zeros(3, 4) on purpose, to eliminate the influence to the final result
     // from initial value. Same for out_1 and ret_1.

     Tensor out_0 = tf.zeros({3, 4});
     Tensor out_1 = tf.ones({3, 4});
     Tensor ret_0 = op_select_copy_int_out(x, /*dim=*/1, /*index=*/0, out_0);
     Tensor ret_1 = op_select_copy_int_out(x, /*dim=*/1, /*index=*/1, out_1);

     EXPECT_TENSOR_EQ(ret_0, out_0);
     EXPECT_TENSOR_EQ(ret_1, out_1);

     EXPECT_TENSOR_EQ(ret_0, tf.ones({3, 4}));
     EXPECT_TENSOR_EQ(ret_1, tf.zeros({3, 4}));
   }

   // Run the test by selecting Tensor x on given dim and all available indexes
   // on that dimension
   void run_test_cases(
       const Tensor& x,
       ssize_t dim,
       const std::vector<Tensor>& expected) {
     // Generated out tensor sharing same size and dtype with expected tensor
     TensorFactory<ScalarType::Double> tf;

     const std::vector<int32_t> out_size(
         expected[0].sizes().begin(), expected[0].sizes().end());
     Tensor out = tf.ones(out_size);

     for (ssize_t idx = 0; idx < x.size(dim); idx++) {
       // Should always return the provided out Tensor.
       // The ret shall meet the expectation.
       Tensor ret = op_select_copy_int_out(x, dim, idx, out);
       EXPECT_TENSOR_EQ(out, ret);
       EXPECT_TENSOR_EQ(out, expected[idx]);

       ret = op_select_copy_int_out(x, dim, /*index=*/idx - x.size(dim), out);
       EXPECT_TENSOR_EQ(out, ret);

       EXPECT_TENSOR_EQ(out, expected[idx]);
     }
   }
 };

 TEST_F(OpSelectCopyIntOutTest, SelectFrontDimAllIndexes) {
   TensorFactory<ScalarType::Double> tf;

   // clang-format off
   Tensor x = tf.make(
       {2, 3, 4},
       {
           // [0, :, :]
           1.,   2.,   3.,   4., // [0, 0, :]
           5.,   6.,   7.,   8., // [0, 1, :]
           9.,  10.,  11.,  12., // [0, 2, :]

           // [1, :, :]
          -1.,  -2.,  -3.,  -4., // [1, 0, :]
          -5.,  -6.,  -7.,  -8., // [1, 1, :]
          -9., -10., -11., -12., // [1, 2, :]
       });
   // clang-format on

   // Try to select the tensor from the input front (0th dimension)
   // The size of output tensor should follow these rules:
   // - output.size(i) shall equal input.size(i) if i < dim,
   // - output.size(i) shall equal input.size(i+1) if i >= dim
   const std::vector<int32_t> out_size = {3, 4};

   Tensor out = tf.zeros(out_size);

   // clang-format off
   std::vector<Tensor> expected_rets = {
     // Expected result when choosing from the 0th dimension and 0th index
     // The result should equal x[0，:, :]
     tf.make(
       out_size,
       {
         1.,   2.,   3.,   4., // [0, :]
         5.,   6.,   7.,   8., // [1, :]
         9.,  10.,  11.,  12., // [2, :]
       }),

     // Expected result when choosing from the 0th dimension and 1st index
     // The result should euqal x[1, :, :]
     tf.make(
       out_size,
       {
         -1.,  -2.,  -3.,  -4., // [0, :]
         -5.,  -6.,  -7.,  -8., // [1, :]
         -9., -10., -11., -12., // [2, :]
       })
   };
   // clang-format on

   run_test_cases(x, /*dim=*/0, expected_rets);
 }

 TEST_F(OpSelectCopyIntOutTest, SelectMiddleDimAllIndexes) {
   TensorFactory<ScalarType::Double> tf;

   // clang-format off
   Tensor x = tf.make(
       {2, 3, 4},
       {
           // [0, :, :]
           1.,   2.,   3.,   4., // [0, 0, :]
           5.,   6.,   7.,   8., // [0, 1, :]
           9.,  10.,  11.,  12., // [0, 2, :]

           // [1, :, :]
          -1.,  -2.,  -3.,  -4., // [1, 0, :]
          -5.,  -6.,  -7.,  -8., // [1, 1, :]
          -9., -10., -11., -12., // [1, 2, :]
       });
   // clang-format on

   // Try to select the tensor from the input front (0th dimension)
   // The size of output tensor should follow these rules:
   // - output.size(i) shall equal input.size(i) if i < dim,
   // - output.size(i) shall equal input.size(i+1) if i >= dim
   const std::vector<int32_t> out_size = {2, 4};

   Tensor out = tf.zeros(out_size);

   // clang-format off
   std::vector<Tensor> expected_rets = {
     // Expected result when choosing from the 1st dimension and 0th index
     // The result should equal x[:，0, :]
     tf.make(
       out_size,
       {
          1.,   2.,   3.,   4., // [0, :]
         -1.,  -2.,  -3.,  -4., // [1, :]
       }),
     // Expected result when choosing from the 1st dimension and 1st index
     // The result should equal x[:, 1, :]
     tf.make(
       out_size,
       {
          5.,   6.,   7.,   8., // [0, :]
         -5.,  -6.,  -7.,  -8., // [1, :]
       }),
     // Expected result when choosing from the 1st dimension and 2th index
     // The result should equal x[:，2, :]
     tf.make(
       out_size,
       {
          9.,  10.,  11.,  12., // [0, :]
         -9., -10., -11., -12., // [1, :]
       })
   };
   // clang-format on

   run_test_cases(x, /*dim=*/1, expected_rets);
 }

 TEST_F(OpSelectCopyIntOutTest, SelectEndDimAllIndexes) {
   TensorFactory<ScalarType::Double> tf;

   // clang-format off
   Tensor x = tf.make(
     {2, 3, 4},
     {
       // [0, :, :]
       1.,   2.,   3.,   4., // [0, 0, :]
       5.,   6.,   7.,   8., // [0, 1, :]
       9.,  10.,  11.,  12., // [0, 2, :]

       // [1, :, :]
       -1.,  -2.,  -3.,  -4., // [1, 0, :]
       -5.,  -6.,  -7.,  -8., // [1, 1, :]
       -9., -10., -11., -12., // [1, 2, :]
     });
   // clang-format on

   // Try to select the tensor from the input front (0th dimension)
   // The size of output tensor should follow these rules:
   // - output.size(i) shall equal input.size(i) if i < dim,
   // - output.size(i) shall equal input.size(i+1) if i >= dim
   const std::vector<int32_t> out_size = {2, 3};

   Tensor out = tf.zeros(out_size);

   // clang-format off
   std::vector<Tensor> expected_rets = {
     // Expected result when choosing from the 2nd dimension and 0th index
     // The result should equal x[:，:, 0] (a.k.a 0th column of x data layout)
     tf.make(
       out_size,
       {
          1.,  5.,  9.,  // [0, :]
         -1., -5., -9.,  // [1, :]
       }),
     // Expected result when choosing from the 2nd dimension and 1st index
     // The result should equal x[:，:, 1] (a.k.a 1st column of x data layout)
     tf.make(
       out_size,
       {
          2.,  6.,  10.,  // [0, :]
         -2., -6., -10.,  // [1, :]
       }),
     // Expected result when choosing from the 2nd dimension and 2nd index
     // The result should equal x[:，:, 2] (a.k.a 2nd column of x data layout)
     tf.make(
       out_size,
       {
          3.,  7.,  11.,  // [0, :]
         -3., -7., -11.,  // [1, :]
       }),
     // Expected result when choosing from the 2nd dimension and 3rd index
     // The result should equal x[:，:, 3] (a.k.a 3rd column of x data layout)
     tf.make(
       out_size,
       {
          4.,  8.,  12.,  // [0, :]
         -4., -8., -12.,  // [1, :]
       })
   };
   // clang-format on

   run_test_cases(x, /*dim=*/2, expected_rets);
 }

 /// A generic smoke test that works for any dtype that supports ones() and
 /// zeros().
 TEST_F(OpSelectCopyIntOutTest, AllDtypesSupported) {
 #define TEST_ENTRY(ctype, dtype) test_dtype<ctype, ScalarType::dtype>();
   ET_FORALL_REAL_TYPES_AND(Bool, 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()
   // for those types.
 }

 //////////////////////////////////////////////////////////////////////////////
 // The following tests focus on empty-size tensor and empty tensor.
 // Here we first define the term:
 // empty-size tensor: size is [] but do have data (e.g.tensor(5))
 // empty tensor: size is not [] and the size of at least one
 // dim is zero, and does not have data in it (e.g ones(1,0,2,3))

 // In this test we are gonnna find if our select function support vector tensor
 // input and empty-size tensor output. Such combination is quite normal in real
 // world (e.g. select(torch.range(10), 0, 5, out) == tensor(5))
 TEST_F(OpSelectCopyIntOutTest, VectorInputSupported) {
   TensorFactory<ScalarType::Int> tf;

   Tensor x = tf.make({10}, {0, 1, 2, 3, 4, 5, 6, 7, 8, 9});

   // Make an empty-size out tensor and demonstrate that it has data.
   Tensor out = tf.make({}, {0});
   EXPECT_EQ(out.numel(), 1);

   // pass the empty-size tensor to the function,
   Tensor expect = tf.make({}, {5});
   op_select_copy_int_out(x, /*dim=*/0, /*index=*/5, out);
   EXPECT_TENSOR_EQ(out, expect);
 }

 // This test focuses on the support for empty tensor (dim() > 0) input and empty
 // tensor output
 TEST_F(OpSelectCopyIntOutTest, EmptyTensorNonZeroNDimsInputSupported) {
   TensorFactory<ScalarType::Int> tf;

   // Using empty tensors as input.
   Tensor x = tf.make({3, 0, 10, 3}, {});
   EXPECT_EQ(x.numel(), 0);

   // Output whose shape is appropriate for selecting along dim(2)
   Tensor out = tf.make({3, 0, 3}, {});
   EXPECT_EQ(out.numel(), 0);

   Tensor ret = op_select_copy_int_out(x, /*dim=*/2, /*index=*/3, out);
   EXPECT_EQ(ret.numel(), 0);
   // Success if it doesn't assert on the weird-shaped empty input and the
   // ret is still a empty array
 }

 // Apply select on dim() == 0 empty tensor input and empty tensor output
 TEST_F(OpSelectCopyIntOutTest, EmptyTensorZeroNDimsInputDies) {
   TensorFactory<ScalarType::Int> tf;

   // Using empty tensors as input.
   Tensor x = tf.make({0}, {});
   EXPECT_EQ(x.numel(), 0);

   // Output whose shape is appropriate for selecting along dim(0)
   Tensor out = tf.make({}, {0});
   EXPECT_EQ(out.numel(), 1);

   // Expected failure when slicing on the dimension with length 0 since no space
   // on the dimension could be sliced. (out of bound error)
   ET_EXPECT_KERNEL_FAILURE(
       context_, op_select_copy_int_out(x, /*dim=*/0, /*index=*/0, out));
 }
 ///////////////////////////////////////////////////////////////////////

 TEST_F(OpSelectCopyIntOutTest, DimOutOfBoundDies) {
   TensorFactory<ScalarType::Int> tf;

   Tensor x = tf.ones({1, 1, 1});
   Tensor out = tf.zeros({1, 1});

   // Some invalid dim values.
   const std::vector<int32_t> invalid_dims = {3, 4, 5, -4, -5, -6};
   for (ssize_t dim : invalid_dims) {
     ET_EXPECT_KERNEL_FAILURE(
         context_, op_select_copy_int_out(x, dim, /*index=*/0, out));
   }
 }

 TEST_F(OpSelectCopyIntOutTest, MismatchedDtypesDies) {
   TensorFactory<ScalarType::Int> tf_int;
   TensorFactory<ScalarType::Float> tf_float;
   Tensor x = tf_int.zeros({1, 2, 2});

   // Size is compatible to the output, but a mismatched dtype.
   Tensor out = tf_float.ones({2, 2});

   ET_EXPECT_KERNEL_FAILURE(
       context_, op_select_copy_int_out(x, /*dim=*/0, /*index=*/0, out));
 }

 TEST_F(OpSelectCopyIntOutTest, OutMatchNumelLackDimAtEndDies) {
   if (torch::executor::testing::SupportedFeatures::get()->is_aten) {
     GTEST_SKIP() << "ATen kernel can handle out with mismatched dimensions";
   }
   TensorFactory<ScalarType::Int> tf;
   Tensor x = tf.zeros({1, 2, 2, 1});

   // Out shares the same dtype and numel as the expected output, but a
   // mixmatched size (out.dim() should always one lower than x.dim())
   Tensor out = tf.ones({2, 2});

   ET_EXPECT_KERNEL_FAILURE(
       context_, op_select_copy_int_out(x, /*dim=*/0, /*index=*/0, out));
 }

 TEST_F(OpSelectCopyIntOutTest, OutMatchNumelExtraDimAtFrontDies) {
   if (torch::executor::testing::SupportedFeatures::get()->is_aten) {
     GTEST_SKIP() << "ATen kernel can handle out with mismatched dimensions";
   }
   TensorFactory<ScalarType::Int> tf;
   Tensor x = tf.zeros({2, 2});

   // Out shares the same dtype and numel as the expected output, but a
   // mixmatched size (out.dim() should always one lower than x.dim())
   Tensor out = tf.ones({1, 2});

   ET_EXPECT_KERNEL_FAILURE(
       context_, op_select_copy_int_out(x, /*dim=*/0, /*index=*/0, out));
 }

 TEST_F(OpSelectCopyIntOutTest, OutSizeMismatchDimDies) {
   if (torch::executor::testing::SupportedFeatures::get()->is_aten) {
     GTEST_SKIP() << "ATen kernel can handle out with mismatched dimensions";
   }
   TensorFactory<ScalarType::Int> tf;

   Tensor x = tf.zeros({2, 4, 7, 5});

   // Should be {2, 4, 5} to match the x when calling select() with dim 2.
   Tensor out = tf.zeros({2, 4, 7});

   ET_EXPECT_KERNEL_FAILURE(
       context_, op_select_copy_int_out(x, /*dim=*/2, /*index=*/3, out));
 }

 /* %python
 import torch
 torch.manual_seed(0)
 x = torch.rand(2, 3, 4)
 res = torch.select(x, 1, 2)
 op = "op_select_copy_int_out"
 opt_extra_params = "1, 2,"
 dtype = "ScalarType::Float"
 check = "EXPECT_TENSOR_EQ" */

 TEST_F(OpSelectCopyIntOutTest, DynamicShapeUpperBoundSameAsExpected) {
   /* %python
   out_args = "{2, 4}, torch::executor::TensorShapeDynamism::DYNAMIC_BOUND"
   %rewrite(unary_op) */

   TensorFactory<ScalarType::Float> tf;

   Tensor x = tf.make(
       {2, 3, 4},
       {0.49625658988952637,  0.7682217955589294,  0.08847743272781372,
        0.13203048706054688,  0.30742281675338745, 0.6340786814689636,
        0.4900934100151062,   0.8964447379112244,  0.455627977848053,
        0.6323062777519226,   0.3488934636116028,  0.40171730518341064,
        0.022325754165649414, 0.16885894536972046, 0.2938884496688843,
        0.518521785736084,    0.6976675987243652,  0.800011396408081,
        0.16102945804595947,  0.28226858377456665, 0.6816085577011108,
        0.9151939749717712,   0.39709991216659546, 0.8741558790206909});
   Tensor expected = tf.make(
       {2, 4},
       {0.455627977848053,
        0.6323062777519226,
        0.3488934636116028,
        0.40171730518341064,
        0.6816085577011108,
        0.9151939749717712,
        0.39709991216659546,
        0.8741558790206909});

   Tensor out =
       tf.zeros({2, 4}, torch::executor::TensorShapeDynamism::DYNAMIC_BOUND);
   op_select_copy_int_out(x, 1, 2, out);
   EXPECT_TENSOR_EQ(out, expected);
 }

 TEST_F(OpSelectCopyIntOutTest, DynamicShapeUpperBoundLargerThanExpected) {
   if (!torch::executor::testing::SupportedFeatures::get()->output_resize) {
     GTEST_SKIP() << "Dynamic shape not supported";
   }
   /* %python
   out_args = "{5, 5}, torch::executor::TensorShapeDynamism::DYNAMIC_BOUND"
   %rewrite(unary_op) */

   TensorFactory<ScalarType::Float> tf;

   Tensor x = tf.make(
       {2, 3, 4},
       {0.49625658988952637,  0.7682217955589294,  0.08847743272781372,
        0.13203048706054688,  0.30742281675338745, 0.6340786814689636,
        0.4900934100151062,   0.8964447379112244,  0.455627977848053,
        0.6323062777519226,   0.3488934636116028,  0.40171730518341064,
        0.022325754165649414, 0.16885894536972046, 0.2938884496688843,
        0.518521785736084,    0.6976675987243652,  0.800011396408081,
        0.16102945804595947,  0.28226858377456665, 0.6816085577011108,
        0.9151939749717712,   0.39709991216659546, 0.8741558790206909});
   Tensor expected = tf.make(
       {2, 4},
       {0.455627977848053,
        0.6323062777519226,
        0.3488934636116028,
        0.40171730518341064,
        0.6816085577011108,
        0.9151939749717712,
        0.39709991216659546,
        0.8741558790206909});

   Tensor out =
       tf.zeros({5, 5}, torch::executor::TensorShapeDynamism::DYNAMIC_BOUND);
   op_select_copy_int_out(x, 1, 2, out);
   EXPECT_TENSOR_EQ(out, expected);
 }

 TEST_F(OpSelectCopyIntOutTest, DynamicShapeUnbound) {
   if (!torch::executor::testing::SupportedFeatures::get()->output_resize) {
     GTEST_SKIP() << "Dynamic shape not supported";
   }
   /* %python
   out_args = "{1, 1}, torch::executor::TensorShapeDynamism::DYNAMIC_UNBOUND"
   %rewrite(unary_op) */

   TensorFactory<ScalarType::Float> tf;

   Tensor x = tf.make(
       {2, 3, 4},
       {0.49625658988952637,  0.7682217955589294,  0.08847743272781372,
        0.13203048706054688,  0.30742281675338745, 0.6340786814689636,
        0.4900934100151062,   0.8964447379112244,  0.455627977848053,
        0.6323062777519226,   0.3488934636116028,  0.40171730518341064,
        0.022325754165649414, 0.16885894536972046, 0.2938884496688843,
        0.518521785736084,    0.6976675987243652,  0.800011396408081,
        0.16102945804595947,  0.28226858377456665, 0.6816085577011108,
        0.9151939749717712,   0.39709991216659546, 0.8741558790206909});
   Tensor expected = tf.make(
       {2, 4},
       {0.455627977848053,
        0.6323062777519226,
        0.3488934636116028,
        0.40171730518341064,
        0.6816085577011108,
        0.9151939749717712,
        0.39709991216659546,
        0.8741558790206909});

   Tensor out =
       tf.zeros({1, 1}, torch::executor::TensorShapeDynamism::DYNAMIC_UNBOUND);
   op_select_copy_int_out(x, 1, 2, out);
   EXPECT_TENSOR_EQ(out, expected);
 }
	/*
	* Copyright (c) Meta Platforms, Inc. and affiliates.
	* All rights reserved.
	*
	* This source code is licensed under the BSD-style license found in the
	* LICENSE file in the root directory of this source tree.
	*/

	#include <executorch/kernels/test/FunctionHeaderWrapper.h> // Declares the operator
	#include <executorch/kernels/test/TestUtil.h>
	#include <executorch/kernels/test/supported_features.h>
	#include <executorch/runtime/core/exec_aten/exec_aten.h>
	#include <executorch/runtime/core/exec_aten/testing_util/tensor_factory.h>
	#include <executorch/runtime/core/exec_aten/testing_util/tensor_util.h>
	#include <executorch/runtime/core/exec_aten/util/scalar_type_util.h>

	#include <gtest/gtest.h>
	#include <sys/types.h>

	using namespace ::testing;
	using exec_aten::ArrayRef;
	using exec_aten::ScalarType;
	using exec_aten::Tensor;
	using torch::executor::testing::TensorFactory;

	class OpSelectCopyIntOutTest : public OperatorTest {
	protected:
	Tensor& op_select_copy_int_out(
	const Tensor& self,
	int64_t dim,
	int64_t index,
	Tensor& out) {
	return torch::executor::aten::select_copy_outf(
	context_, self, dim, index, out);
	}

	template <class CTYPE, exec_aten::ScalarType DTYPE>
	void test_dtype() {
	TensorFactory<DTYPE> tf;

	// Based on the split defintion, if we split any dim()=3 and size(1)=2
	// tensor along first dim to two tensors [ret_0, ret_1], the ret_0 and ret_1
	// shall be equal to x[:, 0, :] and x[:, 1, :] e.g. x[i, 0, j] = ret_0[i, j]
	// and x[i, 1, j] = ret_1[i, j] for any i in [-x.size(0), x.size(0)) and j
	// in
	// [-x.size(2), x.size(2))
	// Therefore we design the following tensor x for test easily: it is a
	// tensor formed by stacking tensors ones(3, 4) and zeros(3,4) along the
	// first dim. So if we select the tensor along the first dim by the above
	// rules, the ret_0 should be ones(3, 4) and ret_1 should be zeros(3, 4)

	// clang-format off
	Tensor x = tf.make(
	{3, 2, 4},
	{
	// all ones below are from x,
	// and all zeros are from y.
	// [0, :, :]
	1, 1, 1, 1, // [0, 0, :]
	0, 0, 0, 0, // [0, 1, :]

	// [1, :, :]
	1, 1, 1, 1, // [1, 0, :]
	0, 0, 0, 0, // [1, 1, :]

	// [2, :, :]
	1, 1, 1, 1, // [2, 0, :]
	0, 0, 0, 0, // [2, 1, :]
	});
	// clang-format on

	// Expected values for out_0 and ret_0 after the test are all ones(3, 4)
	// based on the above rules. So here we set the default value of out_0 as
	// zeros(3, 4) on purpose, to eliminate the influence to the final result
	// from initial value. Same for out_1 and ret_1.

	Tensor out_0 = tf.zeros({3, 4});
	Tensor out_1 = tf.ones({3, 4});
	Tensor ret_0 = op_select_copy_int_out(x, /dim=/1, /index=/0, out_0);
	Tensor ret_1 = op_select_copy_int_out(x, /dim=/1, /index=/1, out_1);

	EXPECT_TENSOR_EQ(ret_0, out_0);
	EXPECT_TENSOR_EQ(ret_1, out_1);

	EXPECT_TENSOR_EQ(ret_0, tf.ones({3, 4}));
	EXPECT_TENSOR_EQ(ret_1, tf.zeros({3, 4}));
	}

	// Run the test by selecting Tensor x on given dim and all available indexes
	// on that dimension
	void run_test_cases(
	const Tensor& x,
	ssize_t dim,
	const std::vector<Tensor>& expected) {
	// Generated out tensor sharing same size and dtype with expected tensor
	TensorFactory<ScalarType::Double> tf;

	const std::vector<int32_t> out_size(
	expected[0].sizes().begin(), expected[0].sizes().end());
	Tensor out = tf.ones(out_size);

	for (ssize_t idx = 0; idx < x.size(dim); idx++) {
	// Should always return the provided out Tensor.
	// The ret shall meet the expectation.
	Tensor ret = op_select_copy_int_out(x, dim, idx, out);
	EXPECT_TENSOR_EQ(out, ret);
	EXPECT_TENSOR_EQ(out, expected[idx]);

	ret = op_select_copy_int_out(x, dim, /index=/idx - x.size(dim), out);
	EXPECT_TENSOR_EQ(out, ret);

	EXPECT_TENSOR_EQ(out, expected[idx]);
	}
	}
	};

	TEST_F(OpSelectCopyIntOutTest, SelectFrontDimAllIndexes) {
	TensorFactory<ScalarType::Double> tf;

	// clang-format off
	Tensor x = tf.make(
	{2, 3, 4},
	{
	// [0, :, :]
	1., 2., 3., 4., // [0, 0, :]
	5., 6., 7., 8., // [0, 1, :]
	9., 10., 11., 12., // [0, 2, :]

	// [1, :, :]
	-1., -2., -3., -4., // [1, 0, :]
	-5., -6., -7., -8., // [1, 1, :]
	-9., -10., -11., -12., // [1, 2, :]
	});
	// clang-format on

	// Try to select the tensor from the input front (0th dimension)
	// The size of output tensor should follow these rules:
	// - output.size(i) shall equal input.size(i) if i < dim,
	// - output.size(i) shall equal input.size(i+1) if i >= dim
	const std::vector<int32_t> out_size = {3, 4};

	Tensor out = tf.zeros(out_size);

	// clang-format off
	std::vector<Tensor> expected_rets = {
	// Expected result when choosing from the 0th dimension and 0th index
	// The result should equal x[0，:, :]
	tf.make(
	out_size,
	{
	1., 2., 3., 4., // [0, :]
	5., 6., 7., 8., // [1, :]
	9., 10., 11., 12., // [2, :]
	}),

	// Expected result when choosing from the 0th dimension and 1st index
	// The result should euqal x[1, :, :]
	tf.make(
	out_size,
	{
	-1., -2., -3., -4., // [0, :]
	-5., -6., -7., -8., // [1, :]
	-9., -10., -11., -12., // [2, :]
	})
	};
	// clang-format on

	run_test_cases(x, /dim=/0, expected_rets);
	}

	TEST_F(OpSelectCopyIntOutTest, SelectMiddleDimAllIndexes) {
	TensorFactory<ScalarType::Double> tf;

	// clang-format off
	Tensor x = tf.make(
	{2, 3, 4},
	{
	// [0, :, :]
	1., 2., 3., 4., // [0, 0, :]
	5., 6., 7., 8., // [0, 1, :]
	9., 10., 11., 12., // [0, 2, :]

	// [1, :, :]
	-1., -2., -3., -4., // [1, 0, :]
	-5., -6., -7., -8., // [1, 1, :]
	-9., -10., -11., -12., // [1, 2, :]
	});
	// clang-format on

	// Try to select the tensor from the input front (0th dimension)
	// The size of output tensor should follow these rules:
	// - output.size(i) shall equal input.size(i) if i < dim,
	// - output.size(i) shall equal input.size(i+1) if i >= dim
	const std::vector<int32_t> out_size = {2, 4};

	Tensor out = tf.zeros(out_size);

	// clang-format off
	std::vector<Tensor> expected_rets = {
	// Expected result when choosing from the 1st dimension and 0th index
	// The result should equal x[:，0, :]
	tf.make(
	out_size,
	{
	1., 2., 3., 4., // [0, :]
	-1., -2., -3., -4., // [1, :]
	}),
	// Expected result when choosing from the 1st dimension and 1st index
	// The result should equal x[:, 1, :]
	tf.make(
	out_size,
	{
	5., 6., 7., 8., // [0, :]
	-5., -6., -7., -8., // [1, :]
	}),
	// Expected result when choosing from the 1st dimension and 2th index
	// The result should equal x[:，2, :]
	tf.make(
	out_size,
	{
	9., 10., 11., 12., // [0, :]
	-9., -10., -11., -12., // [1, :]
	})
	};
	// clang-format on

	run_test_cases(x, /dim=/1, expected_rets);
	}

	TEST_F(OpSelectCopyIntOutTest, SelectEndDimAllIndexes) {
	TensorFactory<ScalarType::Double> tf;

	// clang-format off
	Tensor x = tf.make(
	{2, 3, 4},
	{
	// [0, :, :]
	1., 2., 3., 4., // [0, 0, :]
	5., 6., 7., 8., // [0, 1, :]
	9., 10., 11., 12., // [0, 2, :]

	// [1, :, :]
	-1., -2., -3., -4., // [1, 0, :]
	-5., -6., -7., -8., // [1, 1, :]
	-9., -10., -11., -12., // [1, 2, :]
	});
	// clang-format on

	// Try to select the tensor from the input front (0th dimension)
	// The size of output tensor should follow these rules:
	// - output.size(i) shall equal input.size(i) if i < dim,
	// - output.size(i) shall equal input.size(i+1) if i >= dim
	const std::vector<int32_t> out_size = {2, 3};

	Tensor out = tf.zeros(out_size);

	// clang-format off
	std::vector<Tensor> expected_rets = {
	// Expected result when choosing from the 2nd dimension and 0th index
	// The result should equal x[:，:, 0] (a.k.a 0th column of x data layout)
	tf.make(
	out_size,
	{
	1., 5., 9., // [0, :]
	-1., -5., -9., // [1, :]
	}),
	// Expected result when choosing from the 2nd dimension and 1st index
	// The result should equal x[:，:, 1] (a.k.a 1st column of x data layout)
	tf.make(
	out_size,
	{
	2., 6., 10., // [0, :]
	-2., -6., -10., // [1, :]
	}),
	// Expected result when choosing from the 2nd dimension and 2nd index
	// The result should equal x[:，:, 2] (a.k.a 2nd column of x data layout)
	tf.make(
	out_size,
	{
	3., 7., 11., // [0, :]
	-3., -7., -11., // [1, :]
	}),
	// Expected result when choosing from the 2nd dimension and 3rd index
	// The result should equal x[:，:, 3] (a.k.a 3rd column of x data layout)
	tf.make(
	out_size,
	{
	4., 8., 12., // [0, :]
	-4., -8., -12., // [1, :]
	})
	};
	// clang-format on

	run_test_cases(x, /dim=/2, expected_rets);
	}

	/// A generic smoke test that works for any dtype that supports ones() and
	/// zeros().
	TEST_F(OpSelectCopyIntOutTest, AllDtypesSupported) {
	#define TEST_ENTRY(ctype, dtype) test_dtype<ctype, ScalarType::dtype>();
	ET_FORALL_REAL_TYPES_AND(Bool, 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()
	// for those types.
	}

	//////////////////////////////////////////////////////////////////////////////
	// The following tests focus on empty-size tensor and empty tensor.
	// Here we first define the term:
	// empty-size tensor: size is [] but do have data (e.g.tensor(5))
	// empty tensor: size is not [] and the size of at least one
	// dim is zero, and does not have data in it (e.g ones(1,0,2,3))

	// In this test we are gonnna find if our select function support vector tensor
	// input and empty-size tensor output. Such combination is quite normal in real
	// world (e.g. select(torch.range(10), 0, 5, out) == tensor(5))
	TEST_F(OpSelectCopyIntOutTest, VectorInputSupported) {
	TensorFactory<ScalarType::Int> tf;

	Tensor x = tf.make({10}, {0, 1, 2, 3, 4, 5, 6, 7, 8, 9});

	// Make an empty-size out tensor and demonstrate that it has data.
	Tensor out = tf.make({}, {0});
	EXPECT_EQ(out.numel(), 1);

	// pass the empty-size tensor to the function,
	Tensor expect = tf.make({}, {5});
	op_select_copy_int_out(x, /dim=/0, /index=/5, out);
	EXPECT_TENSOR_EQ(out, expect);
	}

	// This test focuses on the support for empty tensor (dim() > 0) input and empty
	// tensor output
	TEST_F(OpSelectCopyIntOutTest, EmptyTensorNonZeroNDimsInputSupported) {
	TensorFactory<ScalarType::Int> tf;

	// Using empty tensors as input.
	Tensor x = tf.make({3, 0, 10, 3}, {});
	EXPECT_EQ(x.numel(), 0);

	// Output whose shape is appropriate for selecting along dim(2)
	Tensor out = tf.make({3, 0, 3}, {});
	EXPECT_EQ(out.numel(), 0);

	Tensor ret = op_select_copy_int_out(x, /dim=/2, /index=/3, out);
	EXPECT_EQ(ret.numel(), 0);
	// Success if it doesn't assert on the weird-shaped empty input and the
	// ret is still a empty array
	}

	// Apply select on dim() == 0 empty tensor input and empty tensor output
	TEST_F(OpSelectCopyIntOutTest, EmptyTensorZeroNDimsInputDies) {
	TensorFactory<ScalarType::Int> tf;

	// Using empty tensors as input.
	Tensor x = tf.make({0}, {});
	EXPECT_EQ(x.numel(), 0);

	// Output whose shape is appropriate for selecting along dim(0)
	Tensor out = tf.make({}, {0});
	EXPECT_EQ(out.numel(), 1);

	// Expected failure when slicing on the dimension with length 0 since no space
	// on the dimension could be sliced. (out of bound error)
	ET_EXPECT_KERNEL_FAILURE(
	context_, op_select_copy_int_out(x, /dim=/0, /index=/0, out));
	}
	///////////////////////////////////////////////////////////////////////

	TEST_F(OpSelectCopyIntOutTest, DimOutOfBoundDies) {
	TensorFactory<ScalarType::Int> tf;

	Tensor x = tf.ones({1, 1, 1});
	Tensor out = tf.zeros({1, 1});

	// Some invalid dim values.
	const std::vector<int32_t> invalid_dims = {3, 4, 5, -4, -5, -6};
	for (ssize_t dim : invalid_dims) {
	ET_EXPECT_KERNEL_FAILURE(
	context_, op_select_copy_int_out(x, dim, /index=/0, out));
	}
	}

	TEST_F(OpSelectCopyIntOutTest, MismatchedDtypesDies) {
	TensorFactory<ScalarType::Int> tf_int;
	TensorFactory<ScalarType::Float> tf_float;
	Tensor x = tf_int.zeros({1, 2, 2});

	// Size is compatible to the output, but a mismatched dtype.
	Tensor out = tf_float.ones({2, 2});

	ET_EXPECT_KERNEL_FAILURE(
	context_, op_select_copy_int_out(x, /dim=/0, /index=/0, out));
	}

	TEST_F(OpSelectCopyIntOutTest, OutMatchNumelLackDimAtEndDies) {
	if (torch::executor::testing::SupportedFeatures::get()->is_aten) {
	GTEST_SKIP() << "ATen kernel can handle out with mismatched dimensions";
	}
	TensorFactory<ScalarType::Int> tf;
	Tensor x = tf.zeros({1, 2, 2, 1});

	// Out shares the same dtype and numel as the expected output, but a
	// mixmatched size (out.dim() should always one lower than x.dim())
	Tensor out = tf.ones({2, 2});

	ET_EXPECT_KERNEL_FAILURE(
	context_, op_select_copy_int_out(x, /dim=/0, /index=/0, out));
	}

	TEST_F(OpSelectCopyIntOutTest, OutMatchNumelExtraDimAtFrontDies) {
	if (torch::executor::testing::SupportedFeatures::get()->is_aten) {
	GTEST_SKIP() << "ATen kernel can handle out with mismatched dimensions";
	}
	TensorFactory<ScalarType::Int> tf;
	Tensor x = tf.zeros({2, 2});

	// Out shares the same dtype and numel as the expected output, but a
	// mixmatched size (out.dim() should always one lower than x.dim())
	Tensor out = tf.ones({1, 2});

	ET_EXPECT_KERNEL_FAILURE(
	context_, op_select_copy_int_out(x, /dim=/0, /index=/0, out));
	}

	TEST_F(OpSelectCopyIntOutTest, OutSizeMismatchDimDies) {
	if (torch::executor::testing::SupportedFeatures::get()->is_aten) {
	GTEST_SKIP() << "ATen kernel can handle out with mismatched dimensions";
	}
	TensorFactory<ScalarType::Int> tf;

	Tensor x = tf.zeros({2, 4, 7, 5});

	// Should be {2, 4, 5} to match the x when calling select() with dim 2.
	Tensor out = tf.zeros({2, 4, 7});

	ET_EXPECT_KERNEL_FAILURE(
	context_, op_select_copy_int_out(x, /dim=/2, /index=/3, out));
	}

	/* %python
	import torch
	torch.manual_seed(0)
	x = torch.rand(2, 3, 4)
	res = torch.select(x, 1, 2)
	op = "op_select_copy_int_out"
	opt_extra_params = "1, 2,"
	dtype = "ScalarType::Float"
	check = "EXPECT_TENSOR_EQ" */

	TEST_F(OpSelectCopyIntOutTest, DynamicShapeUpperBoundSameAsExpected) {
	/* %python
	out_args = "{2, 4}, torch::executor::TensorShapeDynamism::DYNAMIC_BOUND"
	%rewrite(unary_op) */

	TensorFactory<ScalarType::Float> tf;

	Tensor x = tf.make(
	{2, 3, 4},
	{0.49625658988952637, 0.7682217955589294, 0.08847743272781372,
	0.13203048706054688, 0.30742281675338745, 0.6340786814689636,
	0.4900934100151062, 0.8964447379112244, 0.455627977848053,
	0.6323062777519226, 0.3488934636116028, 0.40171730518341064,
	0.022325754165649414, 0.16885894536972046, 0.2938884496688843,
	0.518521785736084, 0.6976675987243652, 0.800011396408081,
	0.16102945804595947, 0.28226858377456665, 0.6816085577011108,
	0.9151939749717712, 0.39709991216659546, 0.8741558790206909});
	Tensor expected = tf.make(
	{2, 4},
	{0.455627977848053,
	0.6323062777519226,
	0.3488934636116028,
	0.40171730518341064,
	0.6816085577011108,
	0.9151939749717712,
	0.39709991216659546,
	0.8741558790206909});

	Tensor out =
	tf.zeros({2, 4}, torch::executor::TensorShapeDynamism::DYNAMIC_BOUND);
	op_select_copy_int_out(x, 1, 2, out);
	EXPECT_TENSOR_EQ(out, expected);
	}

	TEST_F(OpSelectCopyIntOutTest, DynamicShapeUpperBoundLargerThanExpected) {
	if (!torch::executor::testing::SupportedFeatures::get()->output_resize) {
	GTEST_SKIP() << "Dynamic shape not supported";
	}
	/* %python
	out_args = "{5, 5}, torch::executor::TensorShapeDynamism::DYNAMIC_BOUND"
	%rewrite(unary_op) */

	TensorFactory<ScalarType::Float> tf;

	Tensor x = tf.make(
	{2, 3, 4},
	{0.49625658988952637, 0.7682217955589294, 0.08847743272781372,
	0.13203048706054688, 0.30742281675338745, 0.6340786814689636,
	0.4900934100151062, 0.8964447379112244, 0.455627977848053,
	0.6323062777519226, 0.3488934636116028, 0.40171730518341064,
	0.022325754165649414, 0.16885894536972046, 0.2938884496688843,
	0.518521785736084, 0.6976675987243652, 0.800011396408081,
	0.16102945804595947, 0.28226858377456665, 0.6816085577011108,
	0.9151939749717712, 0.39709991216659546, 0.8741558790206909});
	Tensor expected = tf.make(
	{2, 4},
	{0.455627977848053,
	0.6323062777519226,
	0.3488934636116028,
	0.40171730518341064,
	0.6816085577011108,
	0.9151939749717712,
	0.39709991216659546,
	0.8741558790206909});

	Tensor out =
	tf.zeros({5, 5}, torch::executor::TensorShapeDynamism::DYNAMIC_BOUND);
	op_select_copy_int_out(x, 1, 2, out);
	EXPECT_TENSOR_EQ(out, expected);
	}

	TEST_F(OpSelectCopyIntOutTest, DynamicShapeUnbound) {
	if (!torch::executor::testing::SupportedFeatures::get()->output_resize) {
	GTEST_SKIP() << "Dynamic shape not supported";
	}
	/* %python
	out_args = "{1, 1}, torch::executor::TensorShapeDynamism::DYNAMIC_UNBOUND"
	%rewrite(unary_op) */

	TensorFactory<ScalarType::Float> tf;

	Tensor x = tf.make(
	{2, 3, 4},
	{0.49625658988952637, 0.7682217955589294, 0.08847743272781372,
	0.13203048706054688, 0.30742281675338745, 0.6340786814689636,
	0.4900934100151062, 0.8964447379112244, 0.455627977848053,
	0.6323062777519226, 0.3488934636116028, 0.40171730518341064,
	0.022325754165649414, 0.16885894536972046, 0.2938884496688843,
	0.518521785736084, 0.6976675987243652, 0.800011396408081,
	0.16102945804595947, 0.28226858377456665, 0.6816085577011108,
	0.9151939749717712, 0.39709991216659546, 0.8741558790206909});
	Tensor expected = tf.make(
	{2, 4},
	{0.455627977848053,
	0.6323062777519226,
	0.3488934636116028,
	0.40171730518341064,
	0.6816085577011108,
	0.9151939749717712,
	0.39709991216659546,
	0.8741558790206909});

	Tensor out =
	tf.zeros({1, 1}, torch::executor::TensorShapeDynamism::DYNAMIC_UNBOUND);
	op_select_copy_int_out(x, 1, 2, out);
	EXPECT_TENSOR_EQ(out, expected);
	}