Reorder added

Adds Reorder kernel exposing blocking reorders from arm_gemm

Resolves ONCPUML-1232

Change-Id: I42bf4166311fe1771565134d3ed7039fc8e30230
Signed-off-by: David Svantesson <david.svantesson@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9500
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

8 files changed, 542 insertions, 0 deletions

diff --git a/tests/BUILD.bazel b/tests/BUILD.bazel
index 8122652991..606f403f38 100644
--- a/tests/BUILD.bazel
+++ b/tests/BUILD.bazel
@@ -106,6 +106,13 @@ cc_binary(
                 "ARM_COMPUTE_ENABLE_BF16",
                 ],
                 "//conditions:default": [],
+              }) +
+        select({
+                "//:sve_validation_flag": [
+                "ENABLE_SVE",
+                "ARM_COMPUTE_ENABLE_SVE",
+                ],
+                "//conditions:default": [],
               })
 )
 
diff --git a/tests/CMakeLists.txt b/tests/CMakeLists.txt
index 163ab82e91..c4b12e770a 100644
--- a/tests/CMakeLists.txt
+++ b/tests/CMakeLists.txt
@@ -107,6 +107,7 @@ target_sources(
           validation/reference/DepthwiseConvolutionLayer.cpp
           validation/reference/FullyConnectedLayer.cpp
           validation/reference/ConvolutionLayer.cpp
+          validation/reference/Reorder.cpp
           framework/Framework.cpp
           framework/Utils.cpp
           framework/Exceptions.cpp
diff --git a/tests/datasets/ReorderLayerDataset.h b/tests/datasets/ReorderLayerDataset.h
new file mode 100644
index 0000000000..8e1a8422b2
--- /dev/null
+++ b/tests/datasets/ReorderLayerDataset.h
@@ -0,0 +1,158 @@
+/*
+ * Copyright (c) 2023 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ACL_TESTS_DATASETS_REORDERLAYERDATASET
+#define ACL_TESTS_DATASETS_REORDERLAYERDATASET
+
+#include "arm_compute/core/TensorShape.h"
+#include "arm_compute/core/Types.h"
+#include "utils/TypePrinter.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace datasets
+{
+/** [ReorderLayer datasets] **/
+class ReorderLayerDataset
+{
+public:
+    using type = std::tuple<TensorShape, TensorShape, WeightFormat, WeightFormat>;
+
+    struct iterator
+    {
+        iterator(std::vector<TensorShape>::const_iterator  in_it,
+                 std::vector<TensorShape>::const_iterator  out_it,
+                 std::vector<WeightFormat>::const_iterator _wf_in_it,
+                 std::vector<WeightFormat>::const_iterator _wf_out_it)
+            : _in_it{ std::move(in_it) },
+              _out_it{ std::move(out_it) },
+              _wf_in_it{ std::move(_wf_in_it) },
+              _wf_out_it{ std::move(_wf_out_it) }
+        {
+        }
+
+        std::string description() const
+        {
+            std::stringstream description;
+            description << "In=" << *_in_it << ":";
+            description << "Out=" << *_out_it << ":";
+            description << "Wf_In=" << *_wf_in_it << ":";
+            description << "Wf_Out=" << *_wf_out_it;
+            return description.str();
+        }
+
+        ReorderLayerDataset::type operator*() const
+        {
+            return std::make_tuple(*_in_it, *_out_it, *_wf_in_it, *_wf_out_it);
+        }
+
+        iterator &operator++()
+        {
+            ++_in_it;
+            ++_out_it;
+            ++_wf_in_it;
+            ++_wf_out_it;
+
+            return *this;
+        }
+
+    private:
+        std::vector<TensorShape>::const_iterator  _in_it;
+        std::vector<TensorShape>::const_iterator  _out_it;
+        std::vector<WeightFormat>::const_iterator _wf_in_it;
+        std::vector<WeightFormat>::const_iterator _wf_out_it;
+    };
+
+    iterator begin() const
+    {
+        return iterator(_in_shapes.begin(), _out_shapes.begin(), _in_wfs.begin(), _out_wfs.begin());
+    }
+
+    int size() const
+    {
+        return std::min(_in_shapes.size(), std::min(_out_shapes.size(), std::min(_in_wfs.size(), _out_wfs.size())));
+    }
+
+    void add_config(TensorShape in, TensorShape out, WeightFormat in_wf, WeightFormat out_wf)
+    {
+        _in_shapes.emplace_back(std::move(in));
+        _out_shapes.emplace_back(std::move(out));
+        _in_wfs.emplace_back(std::move(in_wf));
+        _out_wfs.emplace_back(std::move(out_wf));
+    }
+
+    // protected:
+    ReorderLayerDataset()                       = default;
+    ReorderLayerDataset(ReorderLayerDataset &&) = default;
+
+    private:
+    std::vector<TensorShape>  _in_shapes{};
+    std::vector<TensorShape>  _out_shapes{};
+    std::vector<WeightFormat> _in_wfs{};
+    std::vector<WeightFormat> _out_wfs{};
+};
+
+/** [ReorderLayer datasets] **/
+
+class ReorderLayerDatasetBlock4 final : public ReorderLayerDataset
+{
+    public:
+    ReorderLayerDatasetBlock4()
+    {
+        add_config(TensorShape(10U, 9U), TensorShape(10U, 12U), WeightFormat::OHWI, WeightFormat::OHWIo4);
+        add_config(TensorShape(16U, 16U), TensorShape(16U, 16U), WeightFormat::OHWI, WeightFormat::OHWIo4);
+        add_config(TensorShape(10U, 511U), TensorShape(10U, 512U), WeightFormat::OHWI, WeightFormat::OHWIo4);
+        add_config(TensorShape(234U, 301U), TensorShape(234U, 304U), WeightFormat::OHWI, WeightFormat::OHWIo4);
+        add_config(TensorShape(1024U, 1024U), TensorShape(1024U, 1024U), WeightFormat::OHWI, WeightFormat::OHWIo4);
+        add_config(TensorShape(10U, 9U, 1U, 1U), TensorShape(10U, 12U, 1U, 1U), WeightFormat::OHWI, WeightFormat::OHWIo4);
+        add_config(TensorShape(16U, 16U, 1U, 1U), TensorShape(16U, 16U, 1U, 1U), WeightFormat::OHWI, WeightFormat::OHWIo4);
+        add_config(TensorShape(10U, 511U, 1U, 1U), TensorShape(10U, 512U, 1U, 1U), WeightFormat::OHWI, WeightFormat::OHWIo4);
+        add_config(TensorShape(234U, 301U, 1U, 1U), TensorShape(234U, 304U, 1U, 1U), WeightFormat::OHWI, WeightFormat::OHWIo4);
+        add_config(TensorShape(1024U, 1024U, 1U, 1U), TensorShape(1024U, 1024U, 1U, 1U), WeightFormat::OHWI, WeightFormat::OHWIo4);
+    }
+};
+
+class ReorderLayerDatasetBlock8 final : public ReorderLayerDataset
+{
+    public:
+    ReorderLayerDatasetBlock8()
+    {
+        add_config(TensorShape(10U, 9U), TensorShape(10U, 16U), WeightFormat::OHWI, WeightFormat::OHWIo8);
+        add_config(TensorShape(16U, 16U), TensorShape(16U, 16U), WeightFormat::OHWI, WeightFormat::OHWIo8);
+        add_config(TensorShape(10U, 511U), TensorShape(10U, 512U), WeightFormat::OHWI, WeightFormat::OHWIo8);
+        add_config(TensorShape(234U, 301U), TensorShape(234U, 304U), WeightFormat::OHWI, WeightFormat::OHWIo8);
+        add_config(TensorShape(1024U, 1024U), TensorShape(1024U, 1024U), WeightFormat::OHWI, WeightFormat::OHWIo8);
+        add_config(TensorShape(10U, 9U, 1U, 1U), TensorShape(10U, 16U, 1U, 1U), WeightFormat::OHWI, WeightFormat::OHWIo8);
+        add_config(TensorShape(16U, 16U, 1U, 1U), TensorShape(16U, 16U, 1U, 1U), WeightFormat::OHWI, WeightFormat::OHWIo8);
+        add_config(TensorShape(10U, 511U, 1U, 1U), TensorShape(10U, 512U, 1U, 1U), WeightFormat::OHWI, WeightFormat::OHWIo8);
+        add_config(TensorShape(234U, 301U, 1U, 1U), TensorShape(234U, 304U, 1U, 1U), WeightFormat::OHWI, WeightFormat::OHWIo8);
+        add_config(TensorShape(1024U, 1024U, 1U, 1U), TensorShape(1024U, 1024U, 1U, 1U), WeightFormat::OHWI, WeightFormat::OHWIo8);
+    }
+};
+
+} // namespace datasets
+} // namespace test
+} // namespace arm_compute
+#endif /* ACL_TESTS_DATASETS_REORDERLAYERDATASET */
diff --git a/tests/validation/CMakeLists.txt b/tests/validation/CMakeLists.txt
index eef062b5f9..448e96c4f9 100644
--- a/tests/validation/CMakeLists.txt
+++ b/tests/validation/CMakeLists.txt
@@ -138,6 +138,7 @@ if(ENABLE_NEON)
             NEON/FullyConnectedLayer.cpp
             NEON/ConvolutionLayer.cpp
             NEON/StridedSlice.cpp
+            NEON/ReorderLayer.cpp
             NEON/UNIT/DynamicTensor.cpp
             NEON/UNIT/TensorAllocator.cpp
             NEON/UNIT/MemoryManager.cpp
diff --git a/tests/validation/NEON/ReorderLayer.cpp b/tests/validation/NEON/ReorderLayer.cpp
new file mode 100644
index 0000000000..89d860fa4f
--- /dev/null
+++ b/tests/validation/NEON/ReorderLayer.cpp
@@ -0,0 +1,66 @@
+/*
+ * Copyright (c) 2023 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/runtime/NEON/functions/NEReorderLayer.h"
+#include "arm_compute/runtime/Tensor.h"
+#include "tests/NEON/Accessor.h"
+#include "tests/datasets/ReorderLayerDataset.h"
+#include "tests/datasets/ShapeDatasets.h"
+#include "tests/framework/Macros.h"
+#include "tests/framework/datasets/Datasets.h"
+#include "tests/validation/Validation.h"
+#include "tests/validation/fixtures/ReorderFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+TEST_SUITE(NEON)
+TEST_SUITE(ReorderLayer)
+
+template <typename T>
+using NEReorderLayerAlias = ReorderValidationFixture<Tensor, Accessor, NEReorderLayer, T>;
+
+TEST_SUITE(FP32)
+#if defined(ARM_COMPUTE_ENABLE_SVE)
+FIXTURE_DATA_TEST_CASE(RunBlock8, NEReorderLayerAlias<float>, framework::DatasetMode::ALL, combine(datasets::ReorderLayerDatasetBlock8(), framework::dataset::make("DataType", DataType::F32)))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+#endif // ARM_COMPUTE_ENABLE_SVE
+FIXTURE_DATA_TEST_CASE(RunBlock4, NEReorderLayerAlias<float>, framework::DatasetMode::ALL, combine(datasets::ReorderLayerDatasetBlock4(), framework::dataset::make("DataType", DataType::F32)))
+{
+    // Validate output
+    validate(Accessor(_target), _reference);
+}
+
+TEST_SUITE_END() // FP32
+
+TEST_SUITE_END() // ReorderLayer
+TEST_SUITE_END() // NEON
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/fixtures/ReorderFixture.h b/tests/validation/fixtures/ReorderFixture.h
new file mode 100644
index 0000000000..bb3e3730df
--- /dev/null
+++ b/tests/validation/fixtures/ReorderFixture.h
@@ -0,0 +1,109 @@
+/*
+ * Copyright (c) 2023 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ACL_TESTS_VALIDATION_FIXTURES_REORDERFIXTURE
+#define ACL_TESTS_VALIDATION_FIXTURES_REORDERFIXTURE
+
+#include "arm_compute/core/TensorShape.h"
+#include "arm_compute/core/Types.h"
+#include "tests/AssetsLibrary.h"
+#include "tests/Globals.h"
+#include "tests/IAccessor.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Fixture.h"
+#include "tests/validation/reference/Reorder.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+/** [ReorderLayer fixture] **/
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class ReorderValidationFixture : public framework::Fixture
+{
+public:
+    template <typename...>
+    void setup(TensorShape input_shape, TensorShape output_shape, WeightFormat input_wf, WeightFormat output_wf, DataType data_type)
+    {
+        _target    = compute_target(input_shape, output_shape, input_wf, output_wf, data_type);
+        _reference = compute_reference(input_shape, output_shape, output_wf, data_type);
+    }
+
+    protected:
+    template <typename U>
+    void fill(U &&tensor)
+    {
+        library->fill_tensor_uniform(tensor, 0);
+    }
+
+    TensorType compute_target(const TensorShape &input_shape, const TensorShape &output_shape, WeightFormat input_wf, WeightFormat output_wf, DataType data_type)
+    {
+        // Create tensors
+        TensorType src = create_tensor<TensorType>(input_shape, data_type);
+        TensorType dst = create_tensor<TensorType>(output_shape, data_type);
+
+        // Create and configure function
+        FunctionType reorder;
+
+        reorder.configure(&src, &dst, input_wf, output_wf);
+
+        ARM_COMPUTE_ASSERT(src.info()->is_resizable());
+        ARM_COMPUTE_ASSERT(dst.info()->is_resizable());
+
+        // Allocate tensors
+        src.allocator()->allocate();
+        dst.allocator()->allocate();
+
+        ARM_COMPUTE_ASSERT(!src.info()->is_resizable());
+        ARM_COMPUTE_ASSERT(!dst.info()->is_resizable());
+
+        // Fill tensors
+        fill(AccessorType(src));
+
+        // Compute function
+        reorder.run();
+
+        return dst;
+    }
+
+    SimpleTensor<T> compute_reference(const TensorShape &input_shape, const TensorShape &output_shape, WeightFormat output_wf, DataType data_type)
+    {
+        // Create reference
+        SimpleTensor<T> src{ input_shape, data_type };
+
+        // Fill reference
+        fill(src);
+
+        return reference::reorder_layer<T>(src, output_shape, output_wf);
+    }
+
+    TensorType      _target{};
+    SimpleTensor<T> _reference{};
+};
+/** [ReorderLayer fixture] **/
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* ACL_TESTS_VALIDATION_FIXTURES_REORDERFIXTURE */
diff --git a/tests/validation/reference/Reorder.cpp b/tests/validation/reference/Reorder.cpp
new file mode 100644
index 0000000000..8abb372596
--- /dev/null
+++ b/tests/validation/reference/Reorder.cpp
@@ -0,0 +1,156 @@
+/*
+ * Copyright (c) 2023 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "Reorder.h"
+#include "src/core/NEON/kernels/arm_gemm/utils.hpp"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+
+/*
+ * Generic transform.
+ *
+ * Assuming the untransposed case, this works by first reading <BlockBy>
+ * consecutive values from the first input row.  This same number of values
+ * are then read from the next <IntBy-1> rows.  Now return to the first
+ * input row and repeat.
+ *
+ * Need to cope with the work requested in either dimension not actually
+ * being a multiple of the block sizes.
+ */
+template <unsigned int tIntBy, unsigned int BlockBy, bool Transposed, size_t TOutSize, size_t TInSize, typename d_type, arm_gemm::VLType vlt>
+struct Transform_ref
+{
+    template <typename TOut, typename TIn>
+    static void Transform(TOut &out, const TIn in, const int stride,
+                          const int y0, const int ymax, const int x0, const int xmax)
+    {
+        // NOTE: This code is disabled to avoid the call to get_vector_length(), so templated transforms will not be
+        // correct for SVE.  This is not an issue as we have specializations for all SVE cases.
+        // For SVE cases we multiply the interleave factor by the vector length.
+        // const unsigned int IntBy = tIntBy * (vlt == VLType::SVE ? get_vector_length<TOut>() / BlockBy : 1);
+        const unsigned int IntBy     = tIntBy;
+        int                out_index = 0;
+
+        const int n_whole_y_blocks = (ymax - y0) / IntBy;
+        const int y_remainders     = (ymax - y0) % IntBy;
+        const int n_y_blocks       = n_whole_y_blocks + (y_remainders ? 1 : 0);
+
+        const int n_whole_x_blocks = (xmax - x0) / BlockBy;
+        const int x_remainders     = (xmax - x0) % BlockBy;
+        const int n_x_blocks       = n_whole_x_blocks + (x_remainders ? 1 : 0);
+
+        // "Y" loop: advance down the rows of the source IntBy rows at a time.
+        // Set up fill_rows to show the number rows to copy from, and blank_rows
+        // for the number of blank rows to add.
+        for(int y_block = 0; y_block < n_y_blocks; y_block++)
+        {
+            const int fill_rows  = (y_block < n_whole_y_blocks) ? IntBy : y_remainders;
+            const int blank_rows = IntBy - fill_rows;
+
+            const int y_base = y0 + (y_block * IntBy);
+
+            // So now advance along this block of rows, BlockBy columns at a time.
+            for(int x_block = 0; x_block < n_x_blocks; x_block++)
+            {
+                const int fill_cols  = (x_block < n_whole_x_blocks) ? BlockBy : x_remainders;
+                const int blank_cols = BlockBy - fill_cols;
+
+                const int x_base = x0 + (x_block * BlockBy);
+
+                for(int row = 0; row < fill_rows; row++)
+                {
+                    for(int col = 0; col < fill_cols; col++)
+                    {
+                        // In-range copy.  If it's transposed, we reverse the sense of rows and columns here.
+                        if(Transposed)
+                        {
+                            out[out_index] = in[(x_base + col) * stride + y_base + row];
+                            out_index++;
+                        }
+                        else
+                        {
+                            out[out_index] = in[(y_base + row) * stride + x_base + col];
+                            out_index++;
+                        }
+                    }
+                    // "col" tail - row is in range but column is out of range.
+                    for(int col = 0; col < blank_cols; col++)
+                    {
+                        out[out_index] = 0;
+                        out_index++;
+                    }
+                }
+                // "row" tail - row is out of range so fill with zeros always.
+                const d_type zeroval = 0;
+                const int    pads    = blank_rows * (fill_cols + blank_cols);
+
+                for(int i = 0; i < pads; i++)
+                {
+                    out[out_index] = zeroval;
+                }
+
+                out_index += pads;
+            }
+        }
+    }
+};
+
+template <typename T>
+SimpleTensor<T> reorder_layer(const SimpleTensor<T> &src, const TensorShape &output_shape, WeightFormat output_wf)
+{
+    SimpleTensor<T> dst{ output_shape, src.data_type() };
+    const int       cols = src.shape()[0];
+    const int       rows = src.shape()[1];
+
+    switch(output_wf)
+    {
+        case WeightFormat::OHWIo4:
+        {
+            Transform_ref<4, 1, true, sizeof(float), sizeof(float), float, arm_gemm::VLType::None>::Transform<SimpleTensor<T> &, SimpleTensor<T>>(dst, src, rows, 0, rows, 0, cols);
+            break;
+        }
+        case WeightFormat::OHWIo8:
+        {
+            Transform_ref<8, 1, true, sizeof(float), sizeof(float), float, arm_gemm::VLType::None>::Transform<SimpleTensor<T> &, SimpleTensor<T>>(dst, src, rows, 0, rows, 0, cols);
+            break;
+        }
+        default:
+            break;
+    }
+
+    return dst;
+}
+
+template SimpleTensor<float> reorder_layer(const SimpleTensor<float> &src, const TensorShape &output_shape, WeightFormat output_wf);
+
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/reference/Reorder.h b/tests/validation/reference/Reorder.h
new file mode 100644
index 0000000000..94ee5078f8
--- /dev/null
+++ b/tests/validation/reference/Reorder.h
@@ -0,0 +1,44 @@
+/*
+ * Copyright (c) 2023 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ACL_TESTS_VALIDATION_REFERENCE_REORDER
+#define ACL_TESTS_VALIDATION_REFERENCE_REORDER
+
+#include "tests/SimpleTensor.h"
+#include "tests/Types.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> reorder_layer(const SimpleTensor<T> &src, const TensorShape &output_shape, WeightFormat output_wf);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* ACL_TESTS_VALIDATION_REFERENCE_REORDER */