COMPMID-2051 Refactor shape_calculator::calculate_concatenate_shape

Change-Id: Ibf316718d11fa975d75f226925747b21c4efd127
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-on: https://review.mlplatform.org/c/974
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, 108 insertions, 57 deletions

diff --git a/src/runtime/CL/functions/CLConcatenateLayer.cpp b/src/runtime/CL/functions/CLConcatenateLayer.cpp
index b9b3c5bb80..b8224d2cce 100644
--- a/src/runtime/CL/functions/CLConcatenateLayer.cpp
+++ b/src/runtime/CL/functions/CLConcatenateLayer.cpp
@@ -56,15 +56,7 @@ void CLConcatenateLayer::configure(const std::vector<ICLTensor *> &inputs_vector
         ARM_COMPUTE_ERROR_ON_NULLPTR(t);
         return t->info();
     });
-    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);
-    }
+    TensorShape 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());
@@ -143,19 +135,6 @@ Status CLConcatenateLayer::validate(const std::vector<ITensorInfo *> &inputs_vec
     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();
-    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);
-    }
-    auto_init_if_empty(tmp_output_info, output_shape, 1, inputs_vector[0]->data_type());
-
     unsigned int offset = 0;
     switch(axis)
     {
@@ -166,19 +145,19 @@ Status CLConcatenateLayer::validate(const std::vector<ITensorInfo *> &inputs_vec
                 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));
+                    ARM_COMPUTE_RETURN_ON_ERROR(CLWidthConcatenate2TensorsKernel::validate(inputs_vector[0], inputs_vector[1], output));
                     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));
+                    ARM_COMPUTE_RETURN_ON_ERROR(CLWidthConcatenate4TensorsKernel::validate(inputs_vector[0], inputs_vector[1], inputs_vector[2], inputs_vector[3], output));
                     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));
+                        ARM_COMPUTE_RETURN_ON_ERROR(CLWidthConcatenateLayerKernel::validate(input, offset, output));
                         offset += input->dimension(axis);
                     }
                     break;
@@ -189,7 +168,7 @@ Status CLConcatenateLayer::validate(const std::vector<ITensorInfo *> &inputs_vec
         {
             for(const auto &input : inputs_vector)
             {
-                ARM_COMPUTE_RETURN_ON_ERROR(CLHeightConcatenateLayerKernel::validate(input, offset, &tmp_output_info));
+                ARM_COMPUTE_RETURN_ON_ERROR(CLHeightConcatenateLayerKernel::validate(input, offset, output));
                 offset += input->dimension(axis);
             }
             break;
@@ -198,7 +177,7 @@ Status CLConcatenateLayer::validate(const std::vector<ITensorInfo *> &inputs_vec
         {
             for(const auto &input : inputs_vector)
             {
-                ARM_COMPUTE_RETURN_ON_ERROR(CLDepthConcatenateLayerKernel::validate(input, offset, &tmp_output_info));
+                ARM_COMPUTE_RETURN_ON_ERROR(CLDepthConcatenateLayerKernel::validate(input, offset, output));
                 offset += input->dimension(axis);
             }
             break;
@@ -207,6 +186,12 @@ Status CLConcatenateLayer::validate(const std::vector<ITensorInfo *> &inputs_vec
             ARM_COMPUTE_ERROR("Axis not supported");
     }
 
+    if(output->total_size() != 0)
+    {
+        TensorShape output_shape = arm_compute::misc::shape_calculator::calculate_concatenate_shape(inputs_vector, axis);
+        ARM_COMPUTE_RETURN_ERROR_ON(output_shape.total_size() != output->tensor_shape().total_size());
+    }
+
     return Status{};
 }
 
diff --git a/src/runtime/CL/functions/CLDepthConcatenateLayer.cpp b/src/runtime/CL/functions/CLDepthConcatenateLayer.cpp
index e46647a20c..4a5f845631 100644
--- a/src/runtime/CL/functions/CLDepthConcatenateLayer.cpp
+++ b/src/runtime/CL/functions/CLDepthConcatenateLayer.cpp
@@ -56,7 +56,7 @@ void CLDepthConcatenateLayer::configure(const std::vector<ICLTensor *> &inputs_v
     _concat_kernels_vector  = arm_compute::support::cpp14::make_unique<CLDepthConcatenateLayerKernel[]>(_num_inputs);
     _border_handlers_vector = arm_compute::support::cpp14::make_unique<CLFillBorderKernel[]>(_num_inputs);
 
-    TensorShape output_shape = arm_compute::misc::shape_calculator::calculate_depth_concatenate_shape(inputs_vector_info);
+    TensorShape output_shape = arm_compute::misc::shape_calculator::calculate_concatenate_shape(inputs_vector_info, Window::DimZ);
 
     // Output auto inizialitation if not yet initialized
     auto_init_if_empty(*output->info(), output_shape, 1, inputs_vector[0]->info()->data_type());
@@ -82,7 +82,7 @@ Status CLDepthConcatenateLayer::validate(const std::vector<ITensorInfo *> &input
 
     // Output auto inizialitation if not yet initialized
     TensorInfo  tmp_output_info = *output->clone();
-    TensorShape output_shape    = arm_compute::misc::shape_calculator::calculate_depth_concatenate_shape(inputs_vector);
+    TensorShape output_shape    = arm_compute::misc::shape_calculator::calculate_concatenate_shape(inputs_vector, Window::DimZ);
     auto_init_if_empty(tmp_output_info, output_shape, 1, inputs_vector[0]->data_type());
 
     unsigned int depth_offset = 0;
diff --git a/src/runtime/GLES_COMPUTE/functions/GCConcatenateLayer.cpp b/src/runtime/GLES_COMPUTE/functions/GCConcatenateLayer.cpp
new file mode 100644
index 0000000000..506f648171
--- /dev/null
+++ b/src/runtime/GLES_COMPUTE/functions/GCConcatenateLayer.cpp
@@ -0,0 +1,81 @@
+/*
+ * Copyright (c) 2019 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/GLES_COMPUTE/functions/GCConcatenateLayer.h"
+
+#include "arm_compute/core/Error.h"
+#include "arm_compute/core/GLES_COMPUTE/IGCTensor.h"
+#include "arm_compute/core/PixelValue.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+#include "arm_compute/runtime/GLES_COMPUTE/GCScheduler.h"
+#include "support/ToolchainSupport.h"
+
+namespace arm_compute
+{
+GCConcatenateLayer::GCConcatenateLayer()
+    : _concat_kernels(),
+      _num_inputs(0),
+      _axis(Window::DimZ)
+{
+}
+
+void GCConcatenateLayer::configure(std::vector<IGCTensor *> inputs_vector, IGCTensor *output, size_t axis)
+{
+    ARM_COMPUTE_ERROR_ON(inputs_vector.size() < 2);
+
+    _num_inputs = inputs_vector.size();
+    _axis       = axis;
+
+    TensorShape 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());
+
+    unsigned int offset = 0;
+    switch(axis)
+    {
+        case Window::DimZ:
+        {
+            for(unsigned int i = 0; i < _num_inputs; ++i)
+            {
+                auto kernel = support::cpp14::make_unique<GCDepthConcatenateLayerKernel>();
+                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("Axis not supported");
+    }
+}
+
+void GCConcatenateLayer::run()
+{
+    for(auto &kernel : _concat_kernels)
+    {
+        GCScheduler::get().dispatch(*kernel, true);
+    }
+}
+} // namespace arm_compute
diff --git a/src/runtime/NEON/functions/NEConcatenateLayer.cpp b/src/runtime/NEON/functions/NEConcatenateLayer.cpp
index e02c0c2c7a..b8cfa2b8f2 100644
--- a/src/runtime/NEON/functions/NEConcatenateLayer.cpp
+++ b/src/runtime/NEON/functions/NEConcatenateLayer.cpp
@@ -56,15 +56,7 @@ void NEConcatenateLayer::configure(const std::vector<ITensor *> &inputs_vector,
         ARM_COMPUTE_ERROR_ON_NULLPTR(inputs_vector.at(i));
         inputs_vector_info.emplace_back(inputs_vector.at(i)->info());
     }
-    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);
-    }
+    TensorShape 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());
@@ -109,19 +101,6 @@ Status NEConcatenateLayer::validate(const std::vector<ITensorInfo *> &inputs_vec
     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output);
     ARM_COMPUTE_RETURN_ERROR_ON(inputs_vector.size() < 2);
 
-    // Output auto inizialitation if not yet initialized
-    TensorInfo  tmp_output_info = *output->clone();
-    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);
-    }
-    auto_init_if_empty(tmp_output_info, output_shape, 1, inputs_vector[0]->data_type());
-
     unsigned int offset = 0;
     for(const auto &input : inputs_vector)
     {
@@ -130,17 +109,17 @@ Status NEConcatenateLayer::validate(const std::vector<ITensorInfo *> &inputs_vec
         {
             case Window::DimX:
             {
-                ARM_COMPUTE_RETURN_ON_ERROR(NEWidthConcatenateLayerKernel::validate(input, offset, &tmp_output_info));
+                ARM_COMPUTE_RETURN_ON_ERROR(NEWidthConcatenateLayerKernel::validate(input, offset, output));
                 break;
             }
             case Window::DimY:
             {
-                ARM_COMPUTE_RETURN_ON_ERROR(NEHeightConcatenateLayerKernel::validate(input, offset, &tmp_output_info));
+                ARM_COMPUTE_RETURN_ON_ERROR(NEHeightConcatenateLayerKernel::validate(input, offset, output));
                 break;
             }
             case Window::DimZ:
             {
-                ARM_COMPUTE_RETURN_ON_ERROR(NEDepthConcatenateLayerKernel::validate(input, offset, &tmp_output_info));
+                ARM_COMPUTE_RETURN_ON_ERROR(NEDepthConcatenateLayerKernel::validate(input, offset, output));
                 break;
             }
             default:
@@ -149,6 +128,12 @@ Status NEConcatenateLayer::validate(const std::vector<ITensorInfo *> &inputs_vec
         offset += input->dimension(axis);
     }
 
+    if(output->total_size() != 0)
+    {
+        TensorShape output_shape = arm_compute::misc::shape_calculator::calculate_concatenate_shape(inputs_vector, axis);
+        ARM_COMPUTE_RETURN_ERROR_ON(output_shape.total_size() != output->tensor_shape().total_size());
+    }
+
     return Status{};
 }
 
diff --git a/src/runtime/NEON/functions/NEDepthConcatenateLayer.cpp b/src/runtime/NEON/functions/NEDepthConcatenateLayer.cpp
index 49db855f21..b814bffa96 100644
--- a/src/runtime/NEON/functions/NEDepthConcatenateLayer.cpp
+++ b/src/runtime/NEON/functions/NEDepthConcatenateLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2018 ARM Limited.
+ * Copyright (c) 2017-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -54,7 +54,7 @@ void NEDepthConcatenateLayer::configure(const std::vector<ITensor *> &inputs_vec
     {
         inputs_vector_info.emplace_back(inputs_vector.at(i)->info());
     }
-    TensorShape output_shape = arm_compute::misc::shape_calculator::calculate_depth_concatenate_shape(inputs_vector_info);
+    TensorShape output_shape = arm_compute::misc::shape_calculator::calculate_concatenate_shape(inputs_vector_info, Window::DimZ);
 
     // Output auto inizialitation if not yet initialized
     auto_init_if_empty(*output->info(), output_shape, 1, inputs_vector[0]->info()->data_type());
@@ -80,7 +80,7 @@ Status NEDepthConcatenateLayer::validate(const std::vector<ITensorInfo *> &input
 
     // Output auto inizialitation if not yet initialized
     TensorInfo  tmp_output_info = *output->clone();
-    TensorShape output_shape    = arm_compute::misc::shape_calculator::calculate_depth_concatenate_shape(inputs_vector);
+    TensorShape output_shape    = arm_compute::misc::shape_calculator::calculate_concatenate_shape(inputs_vector, Window::DimZ);
     auto_init_if_empty(tmp_output_info, output_shape, 1, inputs_vector[0]->data_type());
 
     unsigned int depth_offset = 0;