COMPMID-699 - Add NEON functions for im2col and col2im

Change-Id: Ie4a5cbd42f412d28de92d787e955e7d977918371
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/110737
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: BSG Visual Compute Jenkins server to access repositories on http://mpd-gerrit.cambridge.arm.com <bsgcomp@arm.com>

11 files changed, 185 insertions, 6 deletions

diff --git a/arm_compute/core/CL/kernels/CLCol2ImKernel.h b/arm_compute/core/CL/kernels/CLCol2ImKernel.h
index 807748cfde..bd86da1b5e 100644
--- a/arm_compute/core/CL/kernels/CLCol2ImKernel.h
+++ b/arm_compute/core/CL/kernels/CLCol2ImKernel.h
@@ -66,7 +66,7 @@ public:
 
     /** Set the input and output of the kernel.
      *
-     * @param[in]  input          The input tensor to convert. Data types supported: QS8/QS16/F16/F32
+     * @param[in]  input          The input tensor to convert. Data types supported: QS8/QS16/QASYMM8/F16/F32
      * @param[out] output         The output tensor. 3 lower dimensions represent a single output [width, height, OFM],
      *                            while the rest represent batch of outputs. Data types supported: Same as @p input
      * @param[in]  convolved_dims Output convolved dimensions.
diff --git a/arm_compute/core/NEON/kernels/NECol2ImKernel.h b/arm_compute/core/NEON/kernels/NECol2ImKernel.h
index d537d49c58..93ec37a610 100644
--- a/arm_compute/core/NEON/kernels/NECol2ImKernel.h
+++ b/arm_compute/core/NEON/kernels/NECol2ImKernel.h
@@ -66,7 +66,7 @@ public:
 
     /** Set the input and output of the kernel.
      *
-     * @param[in]  input          The input tensor to convert. Data types supported: U8/S8/QS8/U16/S16/QS16/F16/U32/S32/F32
+     * @param[in]  input          The input tensor to convert. Data types supported: U8/S8/QS8/QASYMM8/U16/S16/QS16/F16/U32/S32/F32
      * @param[out] output         The output tensor. 3 lower dimensions represent a single output [width, height, OFM],
      *                            while the rest represent batch of outputs. Data types supported: Same as @p input
      * @param[in]  convolved_dims Output convolved dimensions.
diff --git a/arm_compute/core/NEON/kernels/NEIm2ColKernel.h b/arm_compute/core/NEON/kernels/NEIm2ColKernel.h
index 1a0735ea8f..176799e824 100644
--- a/arm_compute/core/NEON/kernels/NEIm2ColKernel.h
+++ b/arm_compute/core/NEON/kernels/NEIm2ColKernel.h
@@ -73,7 +73,8 @@ public:
     /** Set the input and output of the kernel.
      *
      * @param[in]  input       The input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM],
-     *                         while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QS8/QS16/F16/F32
+     *                         while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QS8/QS16/QASYMM8/F16/F32
+     *                         Note: QASYMM8 works only for has_bias = false
      * @param[out] output      The output tensor. Data types supported: Same as @p input
      * @param[in]  kernel_dims The kernel dimensions (width and height).
      * @param[in]  conv_info   Contains padding and stride information described in @ref PadStrideInfo.
diff --git a/arm_compute/runtime/NEON/NEFunctions.h b/arm_compute/runtime/NEON/NEFunctions.h
index 4b36202fd9..5baaa50d40 100644
--- a/arm_compute/runtime/NEON/NEFunctions.h
+++ b/arm_compute/runtime/NEON/NEFunctions.h
@@ -39,6 +39,7 @@
 #include "arm_compute/runtime/NEON/functions/NECannyEdge.h"
 #include "arm_compute/runtime/NEON/functions/NEChannelCombine.h"
 #include "arm_compute/runtime/NEON/functions/NEChannelExtract.h"
+#include "arm_compute/runtime/NEON/functions/NECol2Im.h"
 #include "arm_compute/runtime/NEON/functions/NEColorConvert.h"
 #include "arm_compute/runtime/NEON/functions/NEConvolution.h"
 #include "arm_compute/runtime/NEON/functions/NEConvolutionLayer.h"
@@ -74,6 +75,7 @@
 #include "arm_compute/runtime/NEON/functions/NEHOGMultiDetection.h"
 #include "arm_compute/runtime/NEON/functions/NEHarrisCorners.h"
 #include "arm_compute/runtime/NEON/functions/NEHistogram.h"
+#include "arm_compute/runtime/NEON/functions/NEIm2Col.h"
 #include "arm_compute/runtime/NEON/functions/NEIntegralImage.h"
 #include "arm_compute/runtime/NEON/functions/NEL2Normalize.h"
 #include "arm_compute/runtime/NEON/functions/NELaplacianPyramid.h"
diff --git a/arm_compute/runtime/NEON/functions/NECol2Im.h b/arm_compute/runtime/NEON/functions/NECol2Im.h
new file mode 100644
index 0000000000..7794d8673f
--- /dev/null
+++ b/arm_compute/runtime/NEON/functions/NECol2Im.h
@@ -0,0 +1,49 @@
+/*
+ * Copyright (c) 2017 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_NECOL2IM_H__
+#define __ARM_COMPUTE_NECOL2IM_H__
+
+#include "arm_compute/runtime/NEON/INESimpleFunction.h"
+
+#include "arm_compute/core/Types.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** Basic function to run @ref NECol2ImKernel */
+class NECol2Im : public INESimpleFunction
+{
+public:
+    /** Configure the col2im NEON kernel
+     *
+     * @param[in]  input          The input tensor to convert. Data types supported: QS8/QS16/QASYMM8/F16/F32
+     * @param[out] output         The output tensor. 3 lower dimensions represent a single output [width, height, OFM],
+     *                            while the rest represent batch of outputs. Data types supported: Same as @p input
+     * @param[in]  convolved_dims Output convolved dimensions.
+     */
+    void configure(const ITensor *input, ITensor *output, std::pair<unsigned int, unsigned int> convolved_dims);
+};
+}
+#endif /* __ARM_COMPUTE_NECOL2IM_H__ */
diff --git a/arm_compute/runtime/NEON/functions/NEIm2Col.h b/arm_compute/runtime/NEON/functions/NEIm2Col.h
new file mode 100644
index 0000000000..913eafabea
--- /dev/null
+++ b/arm_compute/runtime/NEON/functions/NEIm2Col.h
@@ -0,0 +1,53 @@
+/*
+ * Copyright (c) 2017 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_NEIM2COL_H__
+#define __ARM_COMPUTE_NEIM2COL_H__
+
+#include "arm_compute/runtime/NEON/INESimpleFunction.h"
+
+#include "arm_compute/core/Size2D.h"
+#include "arm_compute/core/Types.h"
+
+namespace arm_compute
+{
+class ITensor;
+
+/** Basic function to run @ref NEIm2ColKernel */
+class NEIm2Col : public INESimpleFunction
+{
+public:
+    /** Configure the im2col NEON kernel
+     *
+     * @param[in]  input       The input tensor to convert. 3 lower dimensions represent a single input [width, height, IFM],
+     *                         while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QS8/QS16/QASYMM8/F16/F32
+     *                         Note: QASYMM8 works only for has_bias = false
+     * @param[out] output      The output tensor. Data types supported: Same as @p input
+     * @param[in]  kernel_dims The kernel dimensions (width and height).
+     * @param[in]  conv_info   Contains padding and stride information described in @ref PadStrideInfo.
+     * @param[in]  has_bias    In case biases are provided expands the matrix with 1.
+     */
+    void configure(const ITensor *input, ITensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias);
+};
+}
+#endif /* __ARM_COMPUTE_NEIM2COL_H__ */
diff --git a/src/core/CL/kernels/CLCol2ImKernel.cpp b/src/core/CL/kernels/CLCol2ImKernel.cpp
index f288752ce6..4521b435b3 100644
--- a/src/core/CL/kernels/CLCol2ImKernel.cpp
+++ b/src/core/CL/kernels/CLCol2ImKernel.cpp
@@ -43,7 +43,7 @@ CLCol2ImKernel::CLCol2ImKernel()
 
 void CLCol2ImKernel::configure(const ICLTensor *input, ICLTensor *output, std::pair<unsigned int, unsigned int> convolved_dims)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QS16, DataType::F16, DataType::F32);
+    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QS16, DataType::QASYMM8, DataType::F16, DataType::F32);
     ARM_COMPUTE_ERROR_ON_NULLPTR(output);
 
     TensorShape output_shape = input->info()->tensor_shape();
diff --git a/src/core/NEON/kernels/NECol2ImKernel.cpp b/src/core/NEON/kernels/NECol2ImKernel.cpp
index 460d37e85d..68fc50ced6 100644
--- a/src/core/NEON/kernels/NECol2ImKernel.cpp
+++ b/src/core/NEON/kernels/NECol2ImKernel.cpp
@@ -69,7 +69,8 @@ NECol2ImKernel::NECol2ImKernel()
 
 void NECol2ImKernel::configure(const ITensor *input, ITensor *output, std::pair<unsigned int, unsigned int> convolved_dims)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::S8, DataType::QS8, DataType::U16, DataType::S16, DataType::QS16, DataType::U32, DataType::S32, DataType::F16,
+    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::S8, DataType::QS8, DataType::QASYMM8, DataType::U16, DataType::S16, DataType::QS16, DataType::U32, DataType::S32,
+                                                  DataType::F16,
                                                   DataType::F32);
     ARM_COMPUTE_ERROR_ON_NULLPTR(output);
 
diff --git a/src/core/NEON/kernels/NEIm2ColKernel.cpp b/src/core/NEON/kernels/NEIm2ColKernel.cpp
index b03d5b2d03..090e554834 100644
--- a/src/core/NEON/kernels/NEIm2ColKernel.cpp
+++ b/src/core/NEON/kernels/NEIm2ColKernel.cpp
@@ -278,9 +278,10 @@ NEIm2ColKernel::NEIm2ColKernel()
 
 void NEIm2ColKernel::configure(const ITensor *input, ITensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias)
 {
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F16, DataType::F32, DataType::QS8, DataType::QS16);
+    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F16, DataType::F32, DataType::QS8, DataType::QS16, DataType::QASYMM8);
     ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
     ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT_POSITION(input, output);
+    ARM_COMPUTE_ERROR_ON(input->info()->data_type() == DataType::QASYMM8 && has_bias);
 
     _input          = input;
     _output         = output;
diff --git a/src/runtime/NEON/functions/NECol2Im.cpp b/src/runtime/NEON/functions/NECol2Im.cpp
new file mode 100644
index 0000000000..2dc4ebd49a
--- /dev/null
+++ b/src/runtime/NEON/functions/NECol2Im.cpp
@@ -0,0 +1,36 @@
+/*
+ * Copyright (c) 2017 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/NECol2Im.h"
+
+#include "arm_compute/core/NEON/kernels/NECol2ImKernel.h"
+#include "support/ToolchainSupport.h"
+
+using namespace arm_compute;
+
+void NECol2Im::configure(const ITensor *input, ITensor *output, std::pair<unsigned int, unsigned int> convolved_dims)
+{
+    auto k = arm_compute::support::cpp14::make_unique<NECol2ImKernel>();
+    k->configure(input, output, convolved_dims);
+    _kernel = std::move(k);
+}
diff --git a/src/runtime/NEON/functions/NEIm2Col.cpp b/src/runtime/NEON/functions/NEIm2Col.cpp
new file mode 100644
index 0000000000..4ed591523b
--- /dev/null
+++ b/src/runtime/NEON/functions/NEIm2Col.cpp
@@ -0,0 +1,36 @@
+/*
+ * Copyright (c) 2017 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/NEIm2Col.h"
+
+#include "arm_compute/core/NEON/kernels/NEIm2ColKernel.h"
+#include "support/ToolchainSupport.h"
+
+using namespace arm_compute;
+
+void NEIm2Col::configure(const ITensor *input, ITensor *output, const Size2D &kernel_dims, const PadStrideInfo &conv_info, bool has_bias)
+{
+    auto k = arm_compute::support::cpp14::make_unique<NEIm2ColKernel>();
+    k->configure(input, output, kernel_dims, conv_info, has_bias);
+    _kernel = std::move(k);
+}