COMPMID-540 Replace NEDeconvolutionLayerUpsampleKernel with NEScaleKernel

Change-Id: Ic29557cca24447ef40fa2cfca84f208b4d43f8de
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/122180
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele DiGiorgio <michele.digiorgio@arm.com>
Reviewed-by: Pablo Tello <pablo.tello@arm.com>

8 files changed, 277 insertions, 32 deletions

diff --git a/arm_compute/core/CPP/CPPKernels.h b/arm_compute/core/CPP/CPPKernels.h
index f55f41b0eb..a0c5707a79 100644
--- a/arm_compute/core/CPP/CPPKernels.h
+++ b/arm_compute/core/CPP/CPPKernels.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -29,5 +29,6 @@
 #include "arm_compute/core/CPP/kernels/CPPDetectionWindowNonMaximaSuppressionKernel.h"
 #include "arm_compute/core/CPP/kernels/CPPPermuteKernel.h"
 #include "arm_compute/core/CPP/kernels/CPPSortEuclideanDistanceKernel.h"
+#include "arm_compute/core/CPP/kernels/CPPUpsampleKernel.h"
 
 #endif /* __ARM_COMPUTE_CPPKERNELS_H__ */
diff --git a/arm_compute/core/CPP/kernels/CPPUpsampleKernel.h b/arm_compute/core/CPP/kernels/CPPUpsampleKernel.h
new file mode 100644
index 0000000000..e814c76c7d
--- /dev/null
+++ b/arm_compute/core/CPP/kernels/CPPUpsampleKernel.h
@@ -0,0 +1,77 @@
+/*
+ * Copyright (c) 2017-2018 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 __ARM_COMPUTE_CPPUPSAMPLEKERNEL_H__
+#define __ARM_COMPUTE_CPPUPSAMPLEKERNEL_H__
+
+#include "arm_compute/core/CPP/ICPPKernel.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** CPP kernel to perform tensor upsample.
+ *
+ */
+class CPPUpsampleKernel : public ICPPKernel
+{
+public:
+    const char *name() const override
+    {
+        return "CPPUpsampleKernel";
+    }
+    /** Default constructor */
+    CPPUpsampleKernel();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CPPUpsampleKernel(const CPPUpsampleKernel &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CPPUpsampleKernel &operator=(const CPPUpsampleKernel &) = delete;
+    /** Allow instances of this class to be moved */
+    CPPUpsampleKernel(CPPUpsampleKernel &&) = default;
+    /** Allow instances of this class to be moved */
+    CPPUpsampleKernel &operator=(CPPUpsampleKernel &&) = default;
+    /** Default destructor */
+    ~CPPUpsampleKernel() = default;
+
+    /** Set the input and output of the kernel.
+     *
+     * @param[in]  input              The input tensor to upsample. Data types supported: F32
+     * @param[out] output             The output tensor. Data types supported: Same as @p input
+     * @param[in]  info               Padding info.
+     * @param[in]  inner_border_right The number of zeros added to right edge of the input.
+     * @param[in]  inner_border_top   The number of zeros added to top edge of the input.
+     */
+    void configure(const ITensor *input, ITensor *output, const PadStrideInfo &info, unsigned int inner_border_right, unsigned int inner_border_top);
+
+    // Inherited methods overridden:
+    void run(const Window &window, const ThreadInfo &info) override;
+    bool is_parallelisable() const override;
+
+private:
+    const ITensor *_input;
+    ITensor       *_output;
+    PadStrideInfo  _info;
+    std::pair<unsigned int, unsigned int> _inner_border;
+};
+} // namespace arm_compute
+#endif /*__ARM_COMPUTE_CPPUPSAMPLEKERNEL_H__ */
diff --git a/arm_compute/runtime/CPP/CPPFunctions.h b/arm_compute/runtime/CPP/CPPFunctions.h
index 1f01ffac82..4b7e1afbb5 100644
--- a/arm_compute/runtime/CPP/CPPFunctions.h
+++ b/arm_compute/runtime/CPP/CPPFunctions.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -26,5 +26,6 @@
 
 /* Header regrouping all the CPP functions */
 #include "arm_compute/runtime/CPP/functions/CPPPermute.h"
+#include "arm_compute/runtime/CPP/functions/CPPUpsample.h"
 
 #endif /* __ARM_COMPUTE_CPPFUNCTIONS_H__ */
diff --git a/arm_compute/runtime/CPP/functions/CPPUpsample.h b/arm_compute/runtime/CPP/functions/CPPUpsample.h
new file mode 100644
index 0000000000..06df866349
--- /dev/null
+++ b/arm_compute/runtime/CPP/functions/CPPUpsample.h
@@ -0,0 +1,50 @@
+/*
+ * Copyright (c) 2017-2018 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 __ARM_COMPUTE_CPPUPSAMPLE_H__
+#define __ARM_COMPUTE_CPPUPSAMPLE_H__
+
+#include "arm_compute/runtime/CPP/ICPPSimpleFunction.h"
+
+#include "arm_compute/core/Types.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** Basic function to run @ref CPPUpsample */
+class CPPUpsample : public ICPPSimpleFunction
+{
+public:
+    /** Configure the upsample CPP kernel
+     *
+     * @param[in]  input              The input tensor to upsample. Data types supported: F32
+     * @param[out] output             The output tensor. Data types supported: Same as @p input
+     * @param[in]  info               Padding information
+     * @param[in]  inner_border_right The number of zeros added to right edge of the input.
+     * @param[in]  inner_border_top   The number of zeros added to top edge of the input.
+     */
+    void configure(const ITensor *input, ITensor *output, const PadStrideInfo &info, unsigned int inner_border_right, unsigned int inner_border_top);
+};
+}
+#endif /* __ARM_COMPUTE_CPPUPSAMPLE_H__ */
diff --git a/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h b/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h
index 7566131488..c148c3ff69 100644
--- a/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h
@@ -24,6 +24,7 @@
 #ifndef __ARM_COMPUTE_NEDECONVOLUTIONLAYER_H__
 #define __ARM_COMPUTE_NEDECONVOLUTIONLAYER_H__
 
+#include "arm_compute/runtime/CPP/functions/CPPUpsample.h"
 #include "arm_compute/runtime/NEON/functions/NEConvolutionLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h"
 
@@ -97,6 +98,7 @@ public:
 private:
     MemoryGroup        _memory_group;
     NEConvolutionLayer _conv_f;
+    CPPUpsample        _upsample_f;
     Tensor             _scaled_output;
     ITensor           *_input;
     PadStrideInfo      _info;
diff --git a/src/core/CPP/kernels/CPPUpsampleKernel.cpp b/src/core/CPP/kernels/CPPUpsampleKernel.cpp
new file mode 100644
index 0000000000..d77d9c118f
--- /dev/null
+++ b/src/core/CPP/kernels/CPPUpsampleKernel.cpp
@@ -0,0 +1,101 @@
+/*
+ * Copyright (c) 2017-2018 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/core/CPP/kernels/CPPUpsampleKernel.h"
+
+#include "arm_compute/core/Error.h"
+#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/ITensor.h"
+#include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/core/Validate.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+
+#include <cstddef>
+#include <cstdint>
+
+using namespace arm_compute;
+
+CPPUpsampleKernel::CPPUpsampleKernel()
+    : _input(nullptr), _output(nullptr), _info(), _inner_border()
+{
+}
+
+bool CPPUpsampleKernel::is_parallelisable() const
+{
+    return false;
+}
+
+void CPPUpsampleKernel::configure(const ITensor *input, ITensor *output, const PadStrideInfo &info, unsigned int inner_border_right, unsigned int inner_border_top)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+
+    _input        = input;
+    _output       = output;
+    _info         = info;
+    _inner_border = std::make_pair(inner_border_right, inner_border_top);
+
+    // Configure kernel window
+    Window win = calculate_max_window(*input->info(), Steps());
+
+    // The CPPUpsampleKernel doesn't need padding so update_window_and_padding() can be skipped
+    Coordinates coord;
+    coord.set_num_dimensions(output->info()->num_dimensions());
+    output->info()->set_valid_region(ValidRegion(coord, output->info()->tensor_shape()));
+
+    ICPPKernel::configure(win);
+}
+
+void CPPUpsampleKernel::run(const Window &window, const ThreadInfo &info)
+{
+    ARM_COMPUTE_UNUSED(info);
+    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
+    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICPPKernel::window(), window);
+
+    // Initialize _scaled_output buffer
+    const int width_scaled  = _output->info()->dimension(0);
+    const int height_scaled = _output->info()->dimension(1);
+    const int stride_x      = _info.stride().first;
+    const int stride_y      = _info.stride().second;
+    const int start_x       = _info.pad().first;
+    const int start_y       = _inner_border.second + _info.pad().second;
+    const int end_y         = height_scaled - _info.pad().second;
+    const int end_x         = width_scaled - _inner_border.first - _info.pad().first;
+
+    std::fill_n(_output->buffer(), _output->info()->total_size(), 0);
+
+    // Create window
+    Window window_out(window);
+    window_out.set(Window::DimX, Window::Dimension(start_x, end_x, stride_x));
+    window_out.set(Window::DimY, Window::Dimension(start_y, end_y, stride_y));
+
+    // Create iterators
+    Iterator in(_input, window);
+    Iterator out(_output, window_out);
+
+    execute_window_loop(window, [&](const Coordinates & id)
+    {
+        *(reinterpret_cast<float *>(out.ptr())) = *(reinterpret_cast<const float *>(in.ptr()));
+    },
+    in, out);
+}
diff --git a/src/runtime/CPP/functions/CPPUpsample.cpp b/src/runtime/CPP/functions/CPPUpsample.cpp
new file mode 100644
index 0000000000..619b7e1f43
--- /dev/null
+++ b/src/runtime/CPP/functions/CPPUpsample.cpp
@@ -0,0 +1,36 @@
+/*
+ * Copyright (c) 2017-2018 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/CPP/functions/CPPUpsample.h"
+
+#include "arm_compute/core/CPP/kernels/CPPUpsampleKernel.h"
+#include "support/ToolchainSupport.h"
+
+using namespace arm_compute;
+
+void CPPUpsample::configure(const ITensor *input, ITensor *output, const PadStrideInfo &info, unsigned int inner_border_right, unsigned int inner_border_top)
+{
+    auto k = arm_compute::support::cpp14::make_unique<CPPUpsampleKernel>();
+    k->configure(input, output, info, inner_border_right, inner_border_top);
+    _kernel = std::move(k);
+}
\ No newline at end of file
diff --git a/src/runtime/NEON/functions/NEDeconvolutionLayer.cpp b/src/runtime/NEON/functions/NEDeconvolutionLayer.cpp
index c1ba5dd36e..693d7a4f70 100644
--- a/src/runtime/NEON/functions/NEDeconvolutionLayer.cpp
+++ b/src/runtime/NEON/functions/NEDeconvolutionLayer.cpp
@@ -34,6 +34,7 @@ using namespace arm_compute::misc::shape_calculator;
 NEDeconvolutionLayer::NEDeconvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager) // NOLINT
     : _memory_group(std::move(memory_manager)),
       _conv_f(),
+      _upsample_f(),
       _scaled_output(),
       _input(nullptr),
       _info(),
@@ -79,44 +80,20 @@ void NEDeconvolutionLayer::configure(ITensor *input, const ITensor *weights, con
 
     // Allocate auxiliary tensors
     _scaled_output.allocator()->allocate();
+
+    // configure upsample function
+    _upsample_f.configure(input, &_scaled_output, info, inner_border_right, inner_border_top);
 }
 
 void NEDeconvolutionLayer::run()
 {
     _memory_group.acquire();
 
-    // Initialize _scaled_output buffer
-    const int width_in      = _input->info()->dimension(0);
-    const int height_in     = _input->info()->dimension(1);
-    const int width_scaled  = _scaled_output.info()->dimension(0);
-    const int height_scaled = _scaled_output.info()->dimension(1);
-    const int num_2d_slices = _input->info()->tensor_shape().total_size() / (width_in * height_in);
-    const int stride_x      = _info.stride().first;
-    const int stride_y      = _info.stride().second;
-
-    std::fill_n(_scaled_output.buffer(), _scaled_output.info()->total_size(), 0);
-
-    // scaled_output is the input for the forward convolution. We copy the input elements to scaled_output
-    // and insert rows and columns with zeroes depending on the stride values.
-    for(int slice = 0; slice < num_2d_slices; ++slice)
-    {
-        const int start_x = _info.pad().first;
-        const int start_y = _inner_border.second + _info.pad().second;
-        const int end_y   = height_scaled - _info.pad().second;
-        const int end_x   = width_scaled - _inner_border.first - _info.pad().first;
-
-        for(int yi = start_y, in_y = 0; yi < end_y; yi += stride_y, in_y++)
-        {
-            for(int xi = start_x, in_x = 0; xi < end_x; xi += stride_x, in_x++)
-            {
-                const auto in = *(reinterpret_cast<float *>(_input->buffer() + _input->info()->offset_element_in_bytes(Coordinates(in_x, in_y, slice))));
-                *(reinterpret_cast<float *>(_scaled_output.buffer() + _scaled_output.info()->offset_element_in_bytes(Coordinates(xi, yi, slice)))) = in;
-            }
-        }
-    }
+    // Run upsample kernel
+    _upsample_f.run();
 
     // Run convolution layer
     _conv_f.run();
 
     _memory_group.release();
-}
+}
\ No newline at end of file