COMPMID-2727: Add support for split sizes in NESplit

Signed-off-by: Kurtis Charnock <kurtis.charnock@arm.com>
Change-Id: I14cb2711f3a02bd5f50976cb78fe5865e2062891
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/c/VisualCompute/ComputeLibrary/+/214133
Tested-by: bsgcomp <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/2433
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

3 files changed, 77 insertions, 129 deletions

diff --git a/arm_compute/runtime/NEON/functions/NESplit.h b/arm_compute/runtime/NEON/functions/NESplit.h
index e4d62048e6..69aef793d5 100644
--- a/arm_compute/runtime/NEON/functions/NESplit.h
+++ b/arm_compute/runtime/NEON/functions/NESplit.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2019 ARM Limited.
+ * Copyright (c) 2018-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -24,8 +24,10 @@
 #ifndef ARM_COMPUTE_NESPLIT_H
 #define ARM_COMPUTE_NESPLIT_H
 
+#include "arm_compute/core/ITensor.h"
 #include "arm_compute/core/Types.h"
 
+#include "arm_compute/runtime/CPP/functions/CPPSplit.h"
 #include "arm_compute/runtime/IFunction.h"
 #include "arm_compute/runtime/NEON/functions/NESlice.h"
 
@@ -34,43 +36,12 @@
 
 namespace arm_compute
 {
-// Forward declarations
-class ITensor;
-
 /** Basic function to split a tensor along a given axis */
-class NESplit : public IFunction
+class NESplit : public CPPSplit<NESlice>
 {
 public:
-    /** Default constructor */
-    NESplit();
-    /** Initialise the kernel's input and outputs.
-     *
-     * @param[in]  input   The input tensor. Data types supported: All
-     * @param[out] outputs A vector containing the output tensors. Data types supported: Same as @p input.
-     *                     The output tensors should match the input tensor dimensions for all shape dimensions apart
-     *                     from the split dimension.
-     * @param[in]  axis    Axis on which to split the input.
-     */
-    void configure(const ITensor *input, const std::vector<ITensor *> &outputs, unsigned int axis);
-    /** Static function to check if given info will lead to a valid configuration of @ref NESplit
-     *
-     * @param[in] input   The input tensor info. Data types supported: All
-     * @param[in] outputs A vector containing the output tensors' info. Data types supported: Same as @p input.
-     *                    The output tensors should match the input tensor dimensions for all shape dimensions apart
-     *                    from the split dimension
-     * @param[in] axis    Axis on which to split the input.
-     *
-     * @return a status
-     */
-    static Status validate(const ITensorInfo *input, const std::vector<ITensorInfo *> &outputs, unsigned int axis);
-
     // Inherited methods overridden:
     void run() override;
-
-private:
-    std::vector<ITensor *> _outputs_vector;
-    std::vector<NESlice>   _slice_functions;
-    unsigned int           _num_outputs;
 };
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_NESPLIT_H */
diff --git a/src/runtime/NEON/functions/NESplit.cpp b/src/runtime/NEON/functions/NESplit.cpp
index 0373ab6f88..d512fe5c09 100644
--- a/src/runtime/NEON/functions/NESplit.cpp
+++ b/src/runtime/NEON/functions/NESplit.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2019 ARM Limited.
+ * Copyright (c) 2018-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -34,100 +34,6 @@
 
 namespace arm_compute
 {
-NESplit::NESplit()
-    : _outputs_vector(), _slice_functions(), _num_outputs(0)
-{
-}
-
-void NESplit::configure(const ITensor *input, const std::vector<ITensor *> &outputs, unsigned int axis)
-{
-    // Create Slice functions
-    _num_outputs = outputs.size();
-    _slice_functions.resize(_num_outputs);
-
-    // Get output shape
-    const TensorShape output_shape = arm_compute::misc::shape_calculator::compute_split_shape(input->info(), axis, _num_outputs);
-
-    // Extract output tensor info
-    std::vector<ITensorInfo *> outputs_info;
-    for(auto &output : outputs)
-    {
-        ARM_COMPUTE_ERROR_ON_NULLPTR(output);
-        outputs_info.emplace_back(output->info());
-    }
-
-    // Validate
-    ARM_COMPUTE_ERROR_THROW_ON(NESplit::validate(input->info(), outputs_info, axis));
-
-    const size_t axis_split_step = output_shape[axis];
-    unsigned int axis_offset     = 0;
-
-    // Start/End coordinates
-    Coordinates start_coords;
-    Coordinates end_coords;
-    for(unsigned int d = 0; d < output_shape.num_dimensions(); ++d)
-    {
-        end_coords.set(d, -1);
-    }
-
-    for(unsigned int i = 0; i < _num_outputs; i++)
-    {
-        // Update coordinate on axis
-        start_coords.set(axis, axis_offset);
-        end_coords.set(axis, axis_offset + axis_split_step);
-
-        // Configure slice function
-        _slice_functions[i].configure(input, outputs[i], start_coords, end_coords);
-
-        // Set valid region from shape
-        outputs[i]->info()->set_valid_region(ValidRegion(Coordinates(), output_shape));
-
-        // Update axis offset
-        axis_offset += axis_split_step;
-    }
-}
-
-Status NESplit::validate(const ITensorInfo *input, const std::vector<ITensorInfo *> &outputs, unsigned int axis)
-{
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input);
-    ARM_COMPUTE_RETURN_ERROR_ON(axis >= input->num_dimensions());
-    ARM_COMPUTE_RETURN_ERROR_ON(outputs.size() < 2);
-
-    // Get output shape
-    const TensorShape output_shape = arm_compute::misc::shape_calculator::compute_split_shape(input, axis, outputs.size());
-    ARM_COMPUTE_RETURN_ERROR_ON(output_shape.total_size() == 0);
-
-    const size_t axis_split_step = output_shape[axis];
-    unsigned int axis_offset     = 0;
-
-    // Start/End coordinates
-    Coordinates start_coords;
-    Coordinates end_coords;
-    for(unsigned int d = 0; d < output_shape.num_dimensions(); ++d)
-    {
-        end_coords.set(d, -1);
-    }
-
-    // Validate output tensors
-    for(const auto &output : outputs)
-    {
-        ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output);
-
-        // Output auto inizialitation if not yet initialized
-        TensorInfo tmp_output_info = *output->clone();
-        auto_init_if_empty(tmp_output_info, input->clone()->set_is_resizable(true).set_tensor_shape(output_shape));
-
-        // Update coordinate on axis
-        start_coords.set(axis, axis_offset);
-        end_coords.set(axis, axis_offset + axis_split_step);
-
-        ARM_COMPUTE_RETURN_ON_ERROR(NESlice::validate(input, output, start_coords, end_coords));
-        axis_offset += axis_split_step;
-    }
-
-    return Status{};
-}
-
 void NESplit::run()
 {
     for(unsigned i = 0; i < _num_outputs; ++i)
diff --git a/tests/validation/NEON/Split.cpp b/tests/validation/NEON/Split.cpp
index 91b3b9a047..5ad19a69ac 100644
--- a/tests/validation/NEON/Split.cpp
+++ b/tests/validation/NEON/Split.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2019 ARM Limited.
+ * Copyright (c) 2018-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -65,6 +65,29 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(
     const Status status = NESplit::validate(&input_info.clone()->set_is_resizable(false), outputs_info_ptr, axis);
     ARM_COMPUTE_EXPECT(bool(status) == expected, framework::LogLevel::ERRORS);
 }
+
+DATA_TEST_CASE(ValidateSplitShapes, framework::DatasetMode::ALL, zip(zip(zip(
+        framework::dataset::make("InputInfo", { TensorInfo(TensorShape(27U, 3U, 16U, 2U), 1, DataType::F32),
+                                                TensorInfo(TensorShape(27U, 3U, 16U, 2U), 1, DataType::F32)
+        }),
+        framework::dataset::make("Axis", { 2, 2 })),
+        framework::dataset::make("Splits", { std::vector<TensorInfo>{TensorInfo(TensorShape(27U, 3U, 4U,  2U), 1, DataType::F32),
+                                                                     TensorInfo(TensorShape(27U, 3U, 4U,  2U), 1, DataType::F32),
+                                                                     TensorInfo(TensorShape(27U, 3U, 8U,  2U), 1, DataType::F32)},
+                                             std::vector<TensorInfo>{TensorInfo(TensorShape(27U, 3U, 3U,  2U), 1, DataType::F32),
+                                                                     TensorInfo(TensorShape(27U, 3U, 13U, 2U), 1, DataType::F32)} })),
+        framework::dataset::make("Expected", { true, true })),
+        input_info, axis, splits, expected)
+{
+    std::vector<ITensorInfo*> outputs_info_ptr;
+
+    for(auto &split : splits)
+    {
+        outputs_info_ptr.emplace_back(const_cast<TensorInfo*>(&split));
+    }
+    const Status status = NESplit::validate(&input_info.clone()->set_is_resizable(false), outputs_info_ptr, axis);
+    ARM_COMPUTE_EXPECT(bool(status) == expected, framework::LogLevel::ERRORS);
+}
 // clang-format on
 // *INDENT-ON*
 
@@ -95,9 +118,45 @@ DATA_TEST_CASE(Configuration,
     }
 }
 
+DATA_TEST_CASE(ConfigurationSplitShapes,
+               framework::DatasetMode::ALL,
+               combine(datasets::SmallSplitShapesDataset(), framework::dataset::make("DataType", { DataType::F16, DataType::F32 })),
+               shape, axis, split_shapes, data_type)
+{
+    // Create tensors
+    Tensor              src = create_tensor<Tensor>(shape, data_type);
+    std::vector<Tensor> dsts;
+
+    for(const auto &split_shape : split_shapes)
+    {
+        Tensor dst = create_tensor<Tensor>(split_shape, data_type);
+        dsts.push_back(std::move(dst));
+    }
+
+    std::vector<ITensor *> dsts_ptrs;
+    for(auto &dst : dsts)
+    {
+        dsts_ptrs.emplace_back(&dst);
+    }
+
+    // Create and Configure function
+    NESplit split;
+    split.configure(&src, dsts_ptrs, axis);
+
+    // Validate valid regions
+    for(auto &dst : dsts)
+    {
+        const ValidRegion valid_region = shape_to_valid_region(dst.info()->tensor_shape());
+        validate(dst.info()->valid_region(), valid_region);
+    }
+}
+
 template <typename T>
 using NESplitFixture = SplitFixture<Tensor, ITensor, Accessor, NESplit, T>;
 
+template <typename T>
+using NESplitShapesFixture = SplitShapesFixture<Tensor, ITensor, Accessor, NESplit, T>;
+
 TEST_SUITE(Float)
 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
 TEST_SUITE(FP16)
@@ -151,6 +210,18 @@ FIXTURE_DATA_TEST_CASE(RunLarge,
         validate(Accessor(_target[i]), _reference[i]);
     }
 }
+
+FIXTURE_DATA_TEST_CASE(RunSmallSplitShapes,
+                       NESplitShapesFixture<float>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(datasets::SmallSplitShapesDataset(), framework::dataset::make("DataType", DataType::F32)))
+{
+    // Validate outputs
+    for(unsigned int i = 0; i < _target.size(); ++i)
+    {
+        validate(Accessor(_target[i]), _reference[i]);
+    }
+}
 TEST_SUITE_END() // FP32
 TEST_SUITE_END() // Float