COMPMID-2062 Rework CL ConcatenateLayer

Change-Id: I5a60cd0e822b8912132a6785057921bbf6ef8f7f
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-on: https://review.mlplatform.org/c/951
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>

5 files changed, 166 insertions, 122 deletions

diff --git a/arm_compute/runtime/CL/functions/CLConcatenateLayer.h b/arm_compute/runtime/CL/functions/CLConcatenateLayer.h
index 25a0153f51..5cf09c8ee0 100644
--- a/arm_compute/runtime/CL/functions/CLConcatenateLayer.h
+++ b/arm_compute/runtime/CL/functions/CLConcatenateLayer.h
@@ -77,13 +77,9 @@ public:
     void run() override;
 
 private:
-    void configure_h_concatenate(std::vector<ICLTensor *> inputs_vector, ICLTensor *output);
-    static Status validate_h_concatenate(const std::vector<ITensorInfo *> &inputs_vector, const ITensorInfo *output);
-
-    std::unique_ptr<IFunction>                        _concat_function;
-    std::unique_ptr<CLHeightConcatenateLayerKernel[]> _hconcat_kernels;
-    unsigned int                                      _num_inputs;
-    unsigned int                                      _axis;
+    std::vector<std::unique_ptr<ICLKernel>> _concat_kernels;
+    unsigned int                            _num_inputs;
+    unsigned int                            _axis;
 };
 }
 #endif /* __ARM_COMPUTE_CLCONCATENATELAYER_H__ */
diff --git a/arm_compute/runtime/CL/functions/CLDepthConcatenateLayer.h b/arm_compute/runtime/CL/functions/CLDepthConcatenateLayer.h
index aef5d63654..33f751f6db 100644
--- a/arm_compute/runtime/CL/functions/CLDepthConcatenateLayer.h
+++ b/arm_compute/runtime/CL/functions/CLDepthConcatenateLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2018 ARM Limited.
+ * Copyright (c) 2017-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -41,6 +41,8 @@ class ICLTensor;
 
 /** Basic function to execute concatenate tensors along z axis. This function calls the following kernels:
  *
+ * @deprecated This function is deprecated and will be removed in release 19.08
+ *
  * -# @ref CLFillBorderKernel (executed if input's lowest two dimensions are smaller than respective output's dimensions)
  * -# @ref CLDepthConcatenateLayerKernel
  *
diff --git a/arm_compute/runtime/CL/functions/CLWidthConcatenateLayer.h b/arm_compute/runtime/CL/functions/CLWidthConcatenateLayer.h
index 55b65dadc4..a87ec3094c 100644
--- a/arm_compute/runtime/CL/functions/CLWidthConcatenateLayer.h
+++ b/arm_compute/runtime/CL/functions/CLWidthConcatenateLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018 ARM Limited.
+ * Copyright (c) 2018-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -42,6 +42,8 @@ class ICLTensor;
 
 /** Basic function to execute concatenate tensors along x axis. This function calls the following kernel:
  *
+ * @deprecated This function is deprecated and will be removed in release 19.08
+ *
  * -# @ref CLWidthConcatenateLayerKernel
  * -# @ref CLWidthConcatenate2TensorsKernel (if there are exactly 2 input tensors)
  * -# @ref CLWidthConcatenate4TensorsKernel (if there are exactly 4 input tensors)
diff --git a/src/runtime/CL/functions/CLConcatenateLayer.cpp b/src/runtime/CL/functions/CLConcatenateLayer.cpp
index 13164fdff9..7edea3efac 100644
--- a/src/runtime/CL/functions/CLConcatenateLayer.cpp
+++ b/src/runtime/CL/functions/CLConcatenateLayer.cpp
@@ -38,39 +38,16 @@
 namespace arm_compute
 {
 CLConcatenateLayer::CLConcatenateLayer()
-    : _concat_function(nullptr),
-      _hconcat_kernels(),
+    : _concat_kernels(),
       _num_inputs(0),
       _axis(Window::DimX)
 {
 }
 
-Status CLConcatenateLayer::validate_h_concatenate(const std::vector<ITensorInfo *> &inputs_vector, const ITensorInfo *output) // NOLINT
-{
-    const unsigned int num_inputs = inputs_vector.size();
-
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output);
-    ARM_COMPUTE_RETURN_ERROR_ON(num_inputs < 2);
-
-    // Output auto inizialitation if not yet initialized
-    TensorInfo        tmp_output_info = *output->clone();
-    const TensorShape output_shape    = arm_compute::misc::shape_calculator::calculate_concatenate_shape(inputs_vector, Window::DimY);
-    auto_init_if_empty(tmp_output_info, output_shape, 1, inputs_vector[0]->data_type());
-
-    unsigned int height_offset = 0;
-    // Validate generic case of WidthConcatenate kernel
-    for(const auto &input : inputs_vector)
-    {
-        ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input);
-        ARM_COMPUTE_RETURN_ON_ERROR(CLHeightConcatenateLayerKernel::validate(input, height_offset, &tmp_output_info));
-        height_offset += input->dimension(Window::DimY);
-    }
-
-    return Status{};
-}
-
-void CLConcatenateLayer::configure_h_concatenate(std::vector<ICLTensor *> inputs_vector, ICLTensor *output) // NOLINT
+void CLConcatenateLayer::configure(const std::vector<ICLTensor *> &inputs_vector, ICLTensor *output, DataLayoutDimension axis)
 {
+    ARM_COMPUTE_ERROR_ON(output == nullptr);
+    _axis       = get_data_layout_dimension_index(output->info()->data_layout(), axis);
     _num_inputs = inputs_vector.size();
 
     std::vector<ITensorInfo *> inputs_vector_info(inputs_vector.size());
@@ -79,103 +56,166 @@ void CLConcatenateLayer::configure_h_concatenate(std::vector<ICLTensor *> inputs
         ARM_COMPUTE_ERROR_ON_NULLPTR(t);
         return t->info();
     });
-
-    const TensorShape output_shape = arm_compute::misc::shape_calculator::calculate_concatenate_shape(inputs_vector, Window::DimY);
+    TensorShape output_shape{};
+    if(_axis == Window::DimZ)
+    {
+        output_shape = arm_compute::misc::shape_calculator::calculate_depth_concatenate_shape(inputs_vector);
+    }
+    else
+    {
+        output_shape = arm_compute::misc::shape_calculator::calculate_concatenate_shape(inputs_vector, _axis);
+    }
 
     // Output auto inizialitation if not yet initialized
     auto_init_if_empty(*output->info(), output_shape, 1, inputs_vector[0]->info()->data_type());
+    ARM_COMPUTE_ERROR_THROW_ON(CLConcatenateLayer::validate(inputs_vector_info, output->info(), axis));
 
-    ARM_COMPUTE_ERROR_THROW_ON(CLConcatenateLayer::validate_h_concatenate(inputs_vector_info, output->info()));
-
-    // Configure generic case WidthConcatenate kernels
-    _hconcat_kernels = arm_compute::support::cpp14::make_unique<CLHeightConcatenateLayerKernel[]>(_num_inputs);
-
-    unsigned int height_offset = 0;
-    unsigned int i             = 0;
-    std::transform(inputs_vector.begin(), inputs_vector.end(), inputs_vector.begin(), [&](ICLTensor * t)
-    {
-        auto &kernel = _hconcat_kernels[i++];
-        kernel.configure(t, height_offset, output);
-        height_offset += t->info()->dimension(Window::DimY);
-        return t;
-    });
-}
-
-void CLConcatenateLayer::configure(const std::vector<ICLTensor *> &inputs_vector, ICLTensor *output, DataLayoutDimension axis)
-{
-    ARM_COMPUTE_ERROR_ON(output == nullptr);
-    _axis = get_data_layout_dimension_index(output->info()->data_layout(), axis);
+    unsigned int offset = 0;
     switch(_axis)
     {
-        case 0:
+        case Window::DimX:
         {
-            auto func = support::cpp14::make_unique<CLWidthConcatenateLayer>();
-            func->configure(inputs_vector, output);
-            _concat_function = std::move(func);
+            switch(_num_inputs)
+            {
+                case 2:
+                {
+                    // Configure WidthConcatenate2Tensors kernel
+                    auto kernel = support::cpp14::make_unique<CLWidthConcatenate2TensorsKernel>();
+                    kernel->configure(inputs_vector.at(0), inputs_vector.at(1), output);
+                    _concat_kernels.emplace_back(std::move(kernel));
+                    break;
+                }
+                case 4:
+                {
+                    // Configure WidthConcatenate4Tensors kernel
+                    auto kernel = support::cpp14::make_unique<CLWidthConcatenate4TensorsKernel>();
+                    kernel->configure(inputs_vector.at(0), inputs_vector.at(1), inputs_vector.at(2), inputs_vector.at(3), output);
+                    _concat_kernels.emplace_back(std::move(kernel));
+                    break;
+                }
+                default:
+                {
+                    // Configure generic case WidthConcatenate kernels
+                    for(unsigned int i = 0; i < _num_inputs; ++i)
+                    {
+                        auto kernel = support::cpp14::make_unique<CLWidthConcatenateLayerKernel>();
+                        kernel->configure(inputs_vector.at(i), offset, output);
+                        offset += inputs_vector.at(i)->info()->dimension(_axis);
+                        _concat_kernels.emplace_back(std::move(kernel));
+                    }
+                    break;
+                }
+            }
             break;
         }
-        case 1:
+        case Window::DimY:
         {
-            configure_h_concatenate(inputs_vector, output);
+            for(unsigned int i = 0; i < _num_inputs; ++i)
+            {
+                auto kernel = support::cpp14::make_unique<CLHeightConcatenateLayerKernel>();
+                kernel->configure(inputs_vector.at(i), offset, output);
+                offset += inputs_vector.at(i)->info()->dimension(_axis);
+                _concat_kernels.emplace_back(std::move(kernel));
+            }
             break;
         }
-        case 2:
+        case Window::DimZ:
         {
-            auto func = support::cpp14::make_unique<CLDepthConcatenateLayer>();
-            func->configure(inputs_vector, output);
-            _concat_function = std::move(func);
+            for(unsigned int i = 0; i < _num_inputs; ++i)
+            {
+                auto kernel = support::cpp14::make_unique<CLDepthConcatenateLayerKernel>();
+                kernel->configure(inputs_vector.at(i), offset, output);
+                offset += inputs_vector.at(i)->info()->dimension(_axis);
+                _concat_kernels.emplace_back(std::move(kernel));
+            }
             break;
         }
         default:
-            ARM_COMPUTE_ERROR("Concatenation is supported across width, height and depth only!");
+            ARM_COMPUTE_ERROR("Axis not supported");
     }
 }
 
 Status CLConcatenateLayer::validate(const std::vector<ITensorInfo *> &inputs_vector, const ITensorInfo *output, DataLayoutDimension axis)
 {
     ARM_COMPUTE_RETURN_ERROR_ON(output == nullptr);
+    const unsigned int num_inputs = inputs_vector.size();
 
-    switch(get_data_layout_dimension_index(output->data_layout(), axis))
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output);
+    ARM_COMPUTE_RETURN_ERROR_ON(num_inputs < 2);
+    const unsigned int _axis = get_data_layout_dimension_index(inputs_vector[0]->data_layout(), axis);
+
+    // Output auto inizialitation if not yet initialized
+    TensorInfo  tmp_output_info = *output->clone();
+    TensorShape output_shape{};
+    if(_axis == Window::DimZ)
     {
-        case 0:
-            ARM_COMPUTE_RETURN_ON_ERROR(CLWidthConcatenateLayer::validate(inputs_vector, output));
-            break;
-        case 1:
-            ARM_COMPUTE_RETURN_ON_ERROR(CLConcatenateLayer::validate_h_concatenate(inputs_vector, output));
-            break;
-        case 2:
-            ARM_COMPUTE_RETURN_ON_ERROR(CLDepthConcatenateLayer::validate(inputs_vector, output));
-            break;
-        default:
-            ARM_COMPUTE_RETURN_ERROR_MSG("Concatenation is supported across width and depth only!");
+        output_shape = arm_compute::misc::shape_calculator::calculate_depth_concatenate_shape(inputs_vector);
     }
-    return Status{};
-}
+    else
+    {
+        output_shape = arm_compute::misc::shape_calculator::calculate_concatenate_shape(inputs_vector, _axis);
+    }
+    auto_init_if_empty(tmp_output_info, output_shape, 1, inputs_vector[0]->data_type());
 
-void CLConcatenateLayer::run()
-{
+    unsigned int offset = 0;
     switch(_axis)
     {
-        case 0:
-        case 2:
+        case Window::DimX:
         {
-            ARM_COMPUTE_ERROR_ON(_concat_function == nullptr);
-            _concat_function->run();
+            switch(num_inputs)
+            {
+                case 2:
+                    // Validate WidthConcatenate2Tensors kernels if there are 2 inputs
+                    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(inputs_vector[0], inputs_vector[1]);
+                    ARM_COMPUTE_RETURN_ON_ERROR(CLWidthConcatenate2TensorsKernel::validate(inputs_vector[0], inputs_vector[1], &tmp_output_info));
+                    break;
+                case 4:
+                    // Validate WidthConcatenate4Tensors kernels if there are 4 inputs
+                    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(inputs_vector[0], inputs_vector[1], inputs_vector[2], inputs_vector[3]);
+                    ARM_COMPUTE_RETURN_ON_ERROR(CLWidthConcatenate4TensorsKernel::validate(inputs_vector[0], inputs_vector[1], inputs_vector[2], inputs_vector[3], &tmp_output_info));
+                    break;
+                default:
+                    // Validate generic case of WidthConcatenate kernel
+                    for(const auto &input : inputs_vector)
+                    {
+                        ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input);
+                        ARM_COMPUTE_RETURN_ON_ERROR(CLWidthConcatenateLayerKernel::validate(input, offset, &tmp_output_info));
+                        offset += input->dimension(_axis);
+                    }
+                    break;
+            }
             break;
         }
-        case 1:
+        case Window::DimY:
         {
-            for(unsigned int i = 0; i < _num_inputs; ++i)
+            for(const auto &input : inputs_vector)
             {
-                CLScheduler::get().enqueue(_hconcat_kernels[i], true);
+                ARM_COMPUTE_RETURN_ON_ERROR(CLHeightConcatenateLayerKernel::validate(input, offset, &tmp_output_info));
+                offset += input->dimension(_axis);
             }
             break;
         }
-        default:
+        case Window::DimZ:
         {
-            ARM_COMPUTE_ERROR("Axis not supported");
+            for(const auto &input : inputs_vector)
+            {
+                ARM_COMPUTE_RETURN_ON_ERROR(CLDepthConcatenateLayerKernel::validate(input, offset, &tmp_output_info));
+                offset += input->dimension(_axis);
+            }
             break;
         }
+        default:
+            ARM_COMPUTE_ERROR("Axis not supported");
+    }
+
+    return Status{};
+}
+
+void CLConcatenateLayer::run()
+{
+    for(auto &kernel : _concat_kernels)
+    {
+        CLScheduler::get().enqueue(*kernel, true);
     }
 }
 } // namespace arm_compute
diff --git a/tests/validation/CL/DepthConcatenateLayer.cpp b/tests/validation/CL/DepthConcatenateLayer.cpp
index 01477f9fc3..f4a693ca7d 100644
--- a/tests/validation/CL/DepthConcatenateLayer.cpp
+++ b/tests/validation/CL/DepthConcatenateLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2018 ARM Limited.
+ * Copyright (c) 2017-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -24,14 +24,14 @@
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/CLTensorAllocator.h"
-#include "arm_compute/runtime/CL/functions/CLDepthConcatenateLayer.h"
+#include "arm_compute/runtime/CL/functions/CLConcatenateLayer.h"
 #include "tests/CL/CLAccessor.h"
 #include "tests/datasets/ShapeDatasets.h"
 #include "tests/framework/Asserts.h"
 #include "tests/framework/Macros.h"
 #include "tests/framework/datasets/Datasets.h"
 #include "tests/validation/Validation.h"
-#include "tests/validation/fixtures/DepthConcatenateLayerFixture.h"
+#include "tests/validation/fixtures/ConcatenateLayerFixture.h"
 
 namespace arm_compute
 {
@@ -73,8 +73,7 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(
         inputs_vector_info_raw.emplace_back(&input);
     }
 
-    bool is_valid = bool(CLDepthConcatenateLayer::validate(inputs_vector_info_raw,
-                                                           &output_info.clone()->set_is_resizable(false)));
+    bool is_valid = bool(CLConcatenateLayer::validate(inputs_vector_info_raw, &output_info.clone()->set_is_resizable(false), DataLayoutDimension::CHANNEL));
     ARM_COMPUTE_EXPECT(is_valid == expected, framework::LogLevel::ERRORS);
 }
 // clang-format on
@@ -83,37 +82,38 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(
 TEST_CASE(Configuration, framework::DatasetMode::ALL)
 {
     // Create tensors
-    CLTensor src1 = create_tensor<CLTensor>(TensorShape(32U, 32U, 128U), DataType::F32, 1);
+    CLTensor src1 = create_tensor<CLTensor>(TensorShape(128U, 32U, 32U), DataType::F32, 1);
     CLTensor src2 = create_tensor<CLTensor>(TensorShape(32U, 32U, 32U), DataType::F32, 1);
+    CLTensor src3 = create_tensor<CLTensor>(TensorShape(16U, 32U, 32U), DataType::F32, 1);
     CLTensor dst;
 
     ARM_COMPUTE_EXPECT(src1.info()->is_resizable(), framework::LogLevel::ERRORS);
     ARM_COMPUTE_EXPECT(src2.info()->is_resizable(), framework::LogLevel::ERRORS);
+    ARM_COMPUTE_EXPECT(src3.info()->is_resizable(), framework::LogLevel::ERRORS);
     ARM_COMPUTE_EXPECT(dst.info()->is_resizable(), framework::LogLevel::ERRORS);
 
     // Create and configure function
-    CLDepthConcatenateLayer concat_layer;
+    CLConcatenateLayer concat_layer;
 
-    concat_layer.configure({ &src1, &src2 }, &dst);
-
-    // Validate valid region
-    const ValidRegion valid_region = shape_to_valid_region(TensorShape(32U, 32U, 160U));
-    validate(dst.info()->valid_region(), valid_region);
+    concat_layer.configure({ &src1, &src2, &src3 }, &dst, DataLayoutDimension::CHANNEL);
 }
-
 template <typename T>
-using CLDepthConcatenateLayerFixture = DepthConcatenateLayerValidationFixture<CLTensor, ICLTensor, CLAccessor, CLDepthConcatenateLayer, T>;
+using CLDepthConcatenateLayerFixture = ConcatenateLayerValidationFixture<CLTensor, ICLTensor, CLAccessor, CLConcatenateLayer, T>;
 
 TEST_SUITE(Float)
 TEST_SUITE(FP16)
-FIXTURE_DATA_TEST_CASE(RunSmall, CLDepthConcatenateLayerFixture<half>, framework::DatasetMode::PRECOMMIT, combine(datasets::Small2DShapes(), framework::dataset::make("DataType",
-                                                                                                                  DataType::F16)))
+FIXTURE_DATA_TEST_CASE(RunSmall, CLDepthConcatenateLayerFixture<half>, framework::DatasetMode::PRECOMMIT, combine(combine(concat(datasets::Small3DShapes(), datasets::Tiny4DShapes()),
+                                                                                                                  framework::dataset::make("DataType",
+                                                                                                                          DataType::F16)),
+                                                                                                                  framework::dataset::make("Axis", 2)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, CLDepthConcatenateLayerFixture<half>, framework::DatasetMode::NIGHTLY, combine(datasets::Large2DShapes(), framework::dataset::make("DataType",
-                                                                                                                DataType::F16)))
+FIXTURE_DATA_TEST_CASE(RunLarge, CLDepthConcatenateLayerFixture<half>, framework::DatasetMode::NIGHTLY, combine(combine(concat(datasets::Large3DShapes(), datasets::Small4DShapes()),
+                                                                                                                        framework::dataset::make("DataType",
+                                                                                                                                DataType::F16)),
+                                                                                                                framework::dataset::make("Axis", 2)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference);
@@ -121,14 +121,17 @@ FIXTURE_DATA_TEST_CASE(RunLarge, CLDepthConcatenateLayerFixture<half>, framework
 TEST_SUITE_END()
 
 TEST_SUITE(FP32)
-FIXTURE_DATA_TEST_CASE(RunSmall, CLDepthConcatenateLayerFixture<float>, framework::DatasetMode::PRECOMMIT, combine(datasets::Small2DShapes(), framework::dataset::make("DataType",
-                                                                                                                   DataType::F32)))
+FIXTURE_DATA_TEST_CASE(RunSmall, CLDepthConcatenateLayerFixture<float>, framework::DatasetMode::PRECOMMIT, combine(combine(concat(datasets::Small3DShapes(), datasets::Tiny4DShapes()),
+                                                                                                                   framework::dataset::make("DataType",
+                                                                                                                           DataType::F32)),
+                                                                                                                   framework::dataset::make("Axis", 2)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, CLDepthConcatenateLayerFixture<float>, framework::DatasetMode::NIGHTLY, combine(datasets::DepthConcatenateLayerShapes(), framework::dataset::make("DataType",
-                                                                                                                 DataType::F32)))
+FIXTURE_DATA_TEST_CASE(RunLarge, CLDepthConcatenateLayerFixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::ConcatenateLayerShapes(), framework::dataset::make("DataType",
+                                                                                                                 DataType::F32)),
+                                                                                                                 framework::dataset::make("Axis", 2)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference);
@@ -136,17 +139,19 @@ FIXTURE_DATA_TEST_CASE(RunLarge, CLDepthConcatenateLayerFixture<float>, framewor
 TEST_SUITE_END()
 TEST_SUITE_END()
 
-
 TEST_SUITE(Quantized)
 TEST_SUITE(QASYMM8)
-FIXTURE_DATA_TEST_CASE(RunSmall, CLDepthConcatenateLayerFixture<uint8_t>, framework::DatasetMode::PRECOMMIT, combine(datasets::Small2DShapes(), framework::dataset::make("DataType",
-                                                                                                                     DataType::QASYMM8)))
+FIXTURE_DATA_TEST_CASE(RunSmall, CLDepthConcatenateLayerFixture<uint8_t>, framework::DatasetMode::PRECOMMIT, combine(combine(concat(datasets::Small3DShapes(), datasets::Tiny4DShapes()),
+                                                                                                                     framework::dataset::make("DataType",
+                                                                                                                             DataType::QASYMM8)),
+                                                                                                                     framework::dataset::make("Axis", 2)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, CLDepthConcatenateLayerFixture<uint8_t>, framework::DatasetMode::NIGHTLY, combine(datasets::DepthConcatenateLayerShapes(), framework::dataset::make("DataType",
-                                                                                                                   DataType::QASYMM8)))
+FIXTURE_DATA_TEST_CASE(RunLarge, CLDepthConcatenateLayerFixture<uint8_t>, framework::DatasetMode::NIGHTLY, combine(combine(datasets::ConcatenateLayerShapes(), framework::dataset::make("DataType",
+                                                                                                                   DataType::QASYMM8)),
+                                                                                                                   framework::dataset::make("Axis", 2)))
 {
     // Validate output
     validate(CLAccessor(_target), _reference);
@@ -154,7 +159,6 @@ FIXTURE_DATA_TEST_CASE(RunLarge, CLDepthConcatenateLayerFixture<uint8_t>, framew
 TEST_SUITE_END()
 TEST_SUITE_END()
 
-
 TEST_SUITE_END()
 TEST_SUITE_END()
 } // namespace validation