COMPMID-3532: Align data type support between doxygen and implementation - CPP

The patch also removes some unused NEON kernels.

Change-Id: I4a7622f31c88ee038b21874614a981764a03122a
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3509
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Sheri Zhang <sheri.zhang@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

17 files changed, 21 insertions, 999 deletions

diff --git a/Android.bp b/Android.bp
index d19db113d5..b0cacea7b1 100644
--- a/Android.bp
+++ b/Android.bp
@@ -276,7 +276,6 @@ cc_library_static {
         "src/core/NEON/kernels/NEFastCornersKernel.cpp",
         "src/core/NEON/kernels/NEFillArrayKernel.cpp",
         "src/core/NEON/kernels/NEFillBorderKernel.cpp",
-        "src/core/NEON/kernels/NEFillInnerBorderKernel.cpp",
         "src/core/NEON/kernels/NEFlattenLayerKernel.cpp",
         "src/core/NEON/kernels/NEFloorKernel.cpp",
         "src/core/NEON/kernels/NEFuseBatchNormalizationKernel.cpp",
@@ -289,10 +288,8 @@ cc_library_static {
         "src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel.cpp",
         "src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.cpp",
         "src/core/NEON/kernels/NEGEMMLowpReductionKernel.cpp",
-        "src/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.cpp",
         "src/core/NEON/kernels/NEGEMMMatrixAdditionKernel.cpp",
         "src/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.cpp",
-        "src/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.cpp",
         "src/core/NEON/kernels/NEGEMMTranspose1xWKernel.cpp",
         "src/core/NEON/kernels/NEGatherKernel.cpp",
         "src/core/NEON/kernels/NEGaussian3x3Kernel.cpp",
diff --git a/arm_compute/core/CPP/kernels/CPPBoxWithNonMaximaSuppressionLimitKernel.h b/arm_compute/core/CPP/kernels/CPPBoxWithNonMaximaSuppressionLimitKernel.h
index 3fa83a6d6d..ac797bdf77 100644
--- a/arm_compute/core/CPP/kernels/CPPBoxWithNonMaximaSuppressionLimitKernel.h
+++ b/arm_compute/core/CPP/kernels/CPPBoxWithNonMaximaSuppressionLimitKernel.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2019 ARM Limited.
+ * Copyright (c) 2018-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -24,8 +24,6 @@
 #ifndef ARM_COMPUTE_CPPBOXWITHNONMAXIMASUPPRESSIONLIMITKERNEL_H
 #define ARM_COMPUTE_CPPBOXWITHNONMAXIMASUPPRESSIONLIMITKERNEL_H
 
-#include "arm_compute/core/IArray.h"
-#include "arm_compute/core/IHOG.h"
 #include "arm_compute/core/NEON/INEKernel.h"
 #include "arm_compute/core/Types.h"
 
@@ -62,7 +60,7 @@ public:
      * @param[out] classes          The classes output tensor of size [N]. Data types supported: Same as @p scores_in
      * @param[out] batch_splits_out (Optional) The batch splits output tensor [batch_size]. Data types supported: Same as @p scores_in
      * @param[out] keeps            (Optional) The keeps output tensor of size [N]. Data types supported: Same as@p scores_in
-     * @param[out] keeps_size       (Optional) Number of filtered indices per class tensor of size [num_classes]. Data types supported: Same as @p scores_in
+     * @param[out] keeps_size       (Optional) Number of filtered indices per class tensor of size [num_classes]. Data types supported: U32
      * @param[in]  info             (Optional) BoxNMSLimitInfo information.
      */
     void configure(const ITensor *scores_in, const ITensor *boxes_in, const ITensor *batch_splits_in, ITensor *scores_out, ITensor *boxes_out, ITensor *classes,
diff --git a/arm_compute/core/CPP/kernels/CPPPermuteKernel.h b/arm_compute/core/CPP/kernels/CPPPermuteKernel.h
index e75152f4ea..6a60849bf3 100644
--- a/arm_compute/core/CPP/kernels/CPPPermuteKernel.h
+++ b/arm_compute/core/CPP/kernels/CPPPermuteKernel.h
@@ -56,15 +56,15 @@ public:
 
     /** Set the input and output of the kernel.
      *
-     * @param[in]  input  The input tensor to permute. Data types supported: U8/S8/QASYMM8/QASYMM8_SIGNED/U16/S16/F16/U32/S32/F32
-     * @param[out] output The output tensor. Data types supported: Same as @p input
+     * @param[in]  input  The input tensor to permute. Data types supported: All.
+     * @param[out] output The output tensor. Data types supported: same as @p input
      * @param[in]  perm   Permutation vector
      */
     void configure(const ITensor *input, ITensor *output, const PermutationVector &perm);
     /** Static function to check if given info will lead to a valid configuration of @ref CPPPermuteKernel
      *
-     * @param[in] input  The input tensor to permute. Data types supported: U8/S8/QASYMM8/QASYMM8_SIGNED/U16/S16/F16/U32/S32/F32
-     * @param[in] output The output tensor. Data types supported: Same as @p input
+     * @param[in] input  The input tensor to permute. Data types supported: All.
+     * @param[in] output The output tensor. Data types supported: same as @p input
      * @param[in] perm   Permutation vector
      *
      * @return a status
diff --git a/arm_compute/core/CPP/kernels/CPPTopKVKernel.h b/arm_compute/core/CPP/kernels/CPPTopKVKernel.h
index 4b9bfdd3c9..bba183ce3a 100644
--- a/arm_compute/core/CPP/kernels/CPPTopKVKernel.h
+++ b/arm_compute/core/CPP/kernels/CPPTopKVKernel.h
@@ -54,7 +54,7 @@ public:
     /** Set the input and output of the kernel.
      *
      * @param[in]  predictions A batch_size x classes tensor. Data types supported: F16/S32/F32/QASYMM8/QASYMM8_SIGNED
-     * @param[in]  targets     A batch_size 1D tensor of class ids. Data types supported: S32
+     * @param[in]  targets     A batch_size 1D tensor of class ids. Data types supported: U32
      * @param[out] output      Computed precision at @p k as a bool 1D tensor. Data types supported: U8
      * @param[in]  k           Number of top elements to look at for computing precision.
      */
@@ -63,7 +63,7 @@ public:
     /** Static function to check if given info will lead to a valid configuration of @ref CPPTopKVKernel
      *
      * @param[in] predictions A batch_size x classes tensor info. Data types supported: F16/S32/F32/QASYMM8/QASYMM8_SIGNED
-     * @param[in] targets     A batch_size 1D tensor info of class ids. Data types supported: S32
+     * @param[in] targets     A batch_size 1D tensor info of class ids. Data types supported: U32
      * @param[in] output      Computed precision at @p k as a bool 1D tensor info. Data types supported: U8
      * @param[in] k           Number of top elements to look at for computing precision.
      *
diff --git a/arm_compute/core/CPP/kernels/CPPUpsampleKernel.h b/arm_compute/core/CPP/kernels/CPPUpsampleKernel.h
index 9fbc9b697c..eb23d1a185 100644
--- a/arm_compute/core/CPP/kernels/CPPUpsampleKernel.h
+++ b/arm_compute/core/CPP/kernels/CPPUpsampleKernel.h
@@ -55,8 +55,8 @@ public:
 
     /** Set the input and output of the kernel.
      *
-     * @param[in]  input  The input tensor to upsample. Data types supported: F32/F16/QASYMM8/QASYMM8_SIGNED
-     * @param[out] output The output tensor. Data types supported: Same as @p input
+     * @param[in]  input  The input tensor to upsample. Data types supported: All.
+     * @param[out] output The output tensor. Data types supported: same as @p input.
      * @param[in]  info   Padding info.
      */
     void configure(const ITensor *input, ITensor *output, const PadStrideInfo &info);
diff --git a/arm_compute/core/NEON/NEKernels.h b/arm_compute/core/NEON/NEKernels.h
index 1c87b11030..1deb64a8de 100644
--- a/arm_compute/core/NEON/NEKernels.h
+++ b/arm_compute/core/NEON/NEKernels.h
@@ -70,7 +70,6 @@
 #include "arm_compute/core/NEON/kernels/NEFastCornersKernel.h"
 #include "arm_compute/core/NEON/kernels/NEFillArrayKernel.h"
 #include "arm_compute/core/NEON/kernels/NEFillBorderKernel.h"
-#include "arm_compute/core/NEON/kernels/NEFillInnerBorderKernel.h"
 #include "arm_compute/core/NEON/kernels/NEFlattenLayerKernel.h"
 #include "arm_compute/core/NEON/kernels/NEFloorKernel.h"
 #include "arm_compute/core/NEON/kernels/NEFuseBatchNormalizationKernel.h"
@@ -84,10 +83,8 @@
 #include "arm_compute/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel.h"
 #include "arm_compute/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.h"
 #include "arm_compute/core/NEON/kernels/NEGEMMLowpReductionKernel.h"
-#include "arm_compute/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.h"
 #include "arm_compute/core/NEON/kernels/NEGEMMMatrixAdditionKernel.h"
 #include "arm_compute/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.h"
-#include "arm_compute/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.h"
 #include "arm_compute/core/NEON/kernels/NEGEMMTranspose1xWKernel.h"
 #include "arm_compute/core/NEON/kernels/NEGatherKernel.h"
 #include "arm_compute/core/NEON/kernels/NEGaussian3x3Kernel.h"
diff --git a/arm_compute/core/NEON/kernels/NEFillInnerBorderKernel.h b/arm_compute/core/NEON/kernels/NEFillInnerBorderKernel.h
deleted file mode 100644
index 9c1059e606..0000000000
--- a/arm_compute/core/NEON/kernels/NEFillInnerBorderKernel.h
+++ /dev/null
@@ -1,79 +0,0 @@
-/*
- * Copyright (c) 2016-2019 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.
- */
-#ifndef ARM_COMPUTE_NEFILLINNERBORDERKERNEL_H
-#define ARM_COMPUTE_NEFILLINNERBORDERKERNEL_H
-
-#include "arm_compute/core/NEON/INEKernel.h"
-#include "arm_compute/core/PixelValue.h"
-#include "arm_compute/core/Types.h"
-
-namespace arm_compute
-{
-class ITensor;
-
-/** Interface for the kernel to fill the interior borders */
-class NEFillInnerBorderKernel : public INEKernel
-{
-public:
-    const char *name() const override
-    {
-        return "NEFillInnerBorderKernel";
-    }
-    /** Default constructor */
-    NEFillInnerBorderKernel();
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    NEFillInnerBorderKernel(const NEFillInnerBorderKernel &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    NEFillInnerBorderKernel &operator=(const NEFillInnerBorderKernel &) = delete;
-    /** Allow instances of this class to be moved */
-    NEFillInnerBorderKernel(NEFillInnerBorderKernel &&) = default;
-    /** Allow instances of this class to be moved */
-    NEFillInnerBorderKernel &operator=(NEFillInnerBorderKernel &&) = default;
-    /** Default destructor */
-    ~NEFillInnerBorderKernel() = default;
-
-    /** Initialise the function.
-     *
-     * @note This kernel fills the borders within the XY-planes.
-     *
-     * @param[in,out] input                 Tensor to process. Data types supported: U8/S16/S32/F32.
-     * @param[in]     border_size           Size of the border to fill in elements.
-     * @param[in]     constant_border_value (Optional) Constant value to use for borders if border_mode is set to CONSTANT.
-     *
-     */
-    void configure(ITensor *input, BorderSize border_size, const PixelValue &constant_border_value = PixelValue());
-
-    // Inherited methods overridden:
-    void run(const Window &window, const ThreadInfo &info) override;
-
-private:
-    template <typename T>
-    void fill_value_single_channel(const Window &window);
-
-    ITensor   *_tensor;
-    BorderSize _border_size;
-    PixelValue _constant_border_value;
-};
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_NEFILLINNERBORDERKERNEL_H */
diff --git a/arm_compute/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.h b/arm_compute/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.h
deleted file mode 100644
index a3ba57e4ab..0000000000
--- a/arm_compute/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.h
+++ /dev/null
@@ -1,75 +0,0 @@
-/*
- * Copyright (c) 2017-2019 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.
- */
-#ifndef ARM_COMPUTE_NEGEMMMATRIXACCUMULATEBIASESKERNEL_H
-#define ARM_COMPUTE_NEGEMMMATRIXACCUMULATEBIASESKERNEL_H
-
-#include "arm_compute/core/NEON/INEKernel.h"
-
-namespace arm_compute
-{
-class ITensor;
-/** NEON kernel to add a bias to each row of the input tensor */
-class NEGEMMMatrixAccumulateBiasesKernel : public INEKernel
-{
-public:
-    const char *name() const override
-    {
-        return "NEGEMMMatrixAccumulateBiasesKernel";
-    }
-    /** Default constructor */
-    NEGEMMMatrixAccumulateBiasesKernel();
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    NEGEMMMatrixAccumulateBiasesKernel(const NEGEMMMatrixAccumulateBiasesKernel &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    NEGEMMMatrixAccumulateBiasesKernel &operator=(const NEGEMMMatrixAccumulateBiasesKernel &) = delete;
-    /** Allow instances of this class to be moved */
-    NEGEMMMatrixAccumulateBiasesKernel(NEGEMMMatrixAccumulateBiasesKernel &&) = default;
-    /** Allow instances of this class to be moved */
-    NEGEMMMatrixAccumulateBiasesKernel &operator=(NEGEMMMatrixAccumulateBiasesKernel &&) = default;
-    /** Default destructor */
-    ~NEGEMMMatrixAccumulateBiasesKernel() = default;
-    /** Set the accumulate buffer and the biases of the kernel.
-     *
-     * @param[in, out] accum  The accumulate tensor to convert. Data type supported: F32
-     * @param[in]      biases The shared biases tensor to append. It must be 1D Tensor. Data type supported: Same as @p input
-     */
-    void configure(ITensor *accum, const ITensor *biases);
-    /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMMatrixAccumulateBiasesKernel
-     *
-     * @param[in] accum  The accumulate tensor to convert. Data type supported: F32
-     * @param[in] biases The shared biases tensor to append. It must be 1D Tensor. Data type supported: Same as @p input
-     *
-     * @return a status
-     */
-    static Status validate(const ITensorInfo *accum, const ITensorInfo *biases);
-
-    // Inherited methods overridden:
-    void run(const Window &window, const ThreadInfo &info) override;
-
-private:
-    ITensor       *_accum;
-    const ITensor *_biases;
-};
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_NEGEMMMATRIXACCUMULATEBIASESKERNEL_H */
diff --git a/arm_compute/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.h b/arm_compute/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.h
deleted file mode 100644
index f5635dd58c..0000000000
--- a/arm_compute/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.h
+++ /dev/null
@@ -1,99 +0,0 @@
-/*
- * Copyright (c) 2016-2019 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.
- */
-#ifndef ARM_COMPUTE_NEGEMMMATRIXVECTORMULTIPLYKERNEL_H_
-#define ARM_COMPUTE_NEGEMMMATRIXVECTORMULTIPLYKERNEL_H_
-
-#include "arm_compute/core/NEON/INESimpleKernel.h"
-
-namespace arm_compute
-{
-class ITensor;
-
-/** Interface for the GEMM matrix vector multiply kernel. **/
-class NEGEMMMatrixVectorMultiplyKernel : public INESimpleKernel
-{
-public:
-    const char *name() const override
-    {
-        return "NEGEMMMatrixVectorMultiplyKernel";
-    }
-    /** Default constructor */
-    NEGEMMMatrixVectorMultiplyKernel();
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    NEGEMMMatrixVectorMultiplyKernel(const NEGEMMMatrixVectorMultiplyKernel &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    NEGEMMMatrixVectorMultiplyKernel &operator=(const NEGEMMMatrixVectorMultiplyKernel &) = delete;
-    /** Allow instances of this class to be moved */
-    NEGEMMMatrixVectorMultiplyKernel(NEGEMMMatrixVectorMultiplyKernel &&) = default;
-    /** Allow instances of this class to be moved */
-    NEGEMMMatrixVectorMultiplyKernel &operator=(NEGEMMMatrixVectorMultiplyKernel &&) = default;
-    /** Initialise the kernel's input and output.
-     *
-     * @param[in]  input0 First Input tensor. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32
-     * @param[in]  input1 Second Input tensor. Data types supported: same as @p input.
-     * @param[out] output Output tensor which stores the interleaved matrix. Data type supported: same as @p input, S32 for QASYMM8/QASYMM8_SIGNED input.
-     */
-    void configure(const ITensor *input0, const ITensor *input1, ITensor *output);
-
-    /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMMatrixVectorMultiplyKernel
-     *
-     * @param[in] input0 First Input tensor. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32
-     * @param[in] input1 Second Input tensor. Data types supported: same as @p input.
-     * @param[in] output Output tensor which stores the interleaved matrix. Data type supported: same as @p input, S32 for QASYMM8/QASYMM8_SIGNED input.
-     *
-     * @return a status
-     */
-    static Status validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output);
-
-    // Inherited methods overridden:
-    void run(const Window &window, const ThreadInfo &info) override;
-    BorderSize border_size() const override;
-
-private:
-    /** Template function to run the matrix vector multiplication
-     *
-     * @tparam I0 Input 0 type
-     * @tparam I1 Input 1 type
-     * @tparam O  Output type
-     *
-     * @param[in] window_in  Input region. (Must be a valid region of the window returned by window()).
-     * @param[in] window_w   Weights region. (Must be a valid region of the window returned by window()).
-     * @param[in] window_out Output region.(Must be a valid region of the window returned by window()).
-     */
-    template <typename I0, typename I1, typename O>
-    void matrix_vector_multiply(const Window &window_in, const Window &window_w, const Window &window_out);
-    /** Common signature for all the specialised matrix vector multiplication functions */
-    using GEMMMatrixVectorMultiplyFunctionPtr = void (NEGEMMMatrixVectorMultiplyKernel::*)(const Window &window_in,
-                                                                                           const Window &window_w,
-                                                                                           const Window &window_out);
-
-private:
-    GEMMMatrixVectorMultiplyFunctionPtr _func;
-    const ITensor                      *_input0;
-    const ITensor                      *_input1;
-    ITensor                            *_output;
-    BorderSize                          _border_size;
-};
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_NEGEMMMATRIXVECTORMULTIPLYKERNEL_H_*/
diff --git a/arm_compute/runtime/CPP/functions/CPPBoxWithNonMaximaSuppressionLimit.h b/arm_compute/runtime/CPP/functions/CPPBoxWithNonMaximaSuppressionLimit.h
index b6f55b5bf8..b1ea84dcda 100644
--- a/arm_compute/runtime/CPP/functions/CPPBoxWithNonMaximaSuppressionLimit.h
+++ b/arm_compute/runtime/CPP/functions/CPPBoxWithNonMaximaSuppressionLimit.h
@@ -58,7 +58,7 @@ public:
      * @param[out] classes          The classes output tensor of size [N]. Data types supported: Same as @p scores_in
      * @param[out] batch_splits_out (Optional) The batch splits output tensor. Data types supported: Same as @p scores_in
      * @param[out] keeps            (Optional) The keeps output tensor of size [N]. Data types supported: Same as @p scores_in
-     * @param[in]  keeps_size       (Optional) Number of filtered indices per class tensor of size [num_classes]. Data types supported: Same as @p scores_in
+     * @param[in]  keeps_size       (Optional) Number of filtered indices per class tensor of size [num_classes]. Data types supported: U32.
      * @param[in]  info             (Optional) BoxNMSLimitInfo information.
      */
     void configure(const ITensor *scores_in, const ITensor *boxes_in, const ITensor *batch_splits_in, ITensor *scores_out, ITensor *boxes_out, ITensor *classes,
@@ -76,7 +76,7 @@ public:
      * @param[in] classes          The classes output tensor of size [N]. Data types supported: Same as @p scores_in
      * @param[in] batch_splits_out (Optional) The batch splits output tensor. Data types supported: Same as @p scores_in
      * @param[in] keeps            (Optional) The keeps output tensor of size [N]. Data types supported: Same as @p scores_in
-     * @param[in] keeps_size       (Optional) Number of filtered indices per class tensor of size [num_classes]. Data types supported: Same as @p scores_in
+     * @param[in] keeps_size       (Optional) Number of filtered indices per class tensor of size [num_classes]. Data types supported: U32.
      * @param[in] info             (Optional) BoxNMSLimitInfo information.
      *
      * @return a status
diff --git a/arm_compute/runtime/CPP/functions/CPPSplit.h b/arm_compute/runtime/CPP/functions/CPPSplit.h
index 6adcbc3323..611cc41157 100644
--- a/arm_compute/runtime/CPP/functions/CPPSplit.h
+++ b/arm_compute/runtime/CPP/functions/CPPSplit.h
@@ -47,8 +47,8 @@ public:
     }
     /** Static function to check if given info will lead to a valid configuration of @ref CPPSplit
      *
-     * @param[in] input   The input tensor info. Data types supported: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32.
-     * @param[in] outputs A vector containing the output tensors' info. Data types supported: Same as @p input.
+     * @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.
diff --git a/arm_compute/runtime/CPP/functions/CPPTopKV.h b/arm_compute/runtime/CPP/functions/CPPTopKV.h
index c94e277312..fb93bdf764 100644
--- a/arm_compute/runtime/CPP/functions/CPPTopKV.h
+++ b/arm_compute/runtime/CPP/functions/CPPTopKV.h
@@ -39,7 +39,7 @@ public:
     /** Set the input and output of the kernel.
      *
      * @param[in]  predictions A batch_size x classes tensor. Data types supported: F16/S32/F32/QASYMM8/QASYMM8_SIGNED
-     * @param[in]  targets     A batch_size 1D tensor of class ids. Data types supported: S32
+     * @param[in]  targets     A batch_size 1D tensor of class ids. Data types supported: U32
      * @param[out] output      Computed precision at @p k as a bool 1D tensor. Data types supported: U8
      * @param[in]  k           Number of top elements to look at for computing precision.
      */
@@ -48,7 +48,7 @@ public:
     /** Static function to check if given info will lead to a valid configuration of @ref CPPTopKVKernel
      *
      * @param[in] predictions A batch_size x classes tensor info. Data types supported: F16/S32/F32/QASYMM8/QASYMM8_SIGNED
-     * @param[in] targets     A batch_size 1D tensor info of class ids. Data types supported: S32
+     * @param[in] targets     A batch_size 1D tensor info of class ids. Data types supported: U32
      * @param[in] output      Computed precision at @p k as a bool 1D tensor info. Data types supported: U8
      * @param[in] k           Number of top elements to look at for computing precision.
      *
diff --git a/arm_compute/runtime/CPP/functions/CPPUpsample.h b/arm_compute/runtime/CPP/functions/CPPUpsample.h
index 3e1852bc30..080e86b676 100644
--- a/arm_compute/runtime/CPP/functions/CPPUpsample.h
+++ b/arm_compute/runtime/CPP/functions/CPPUpsample.h
@@ -38,8 +38,8 @@ class CPPUpsample : public ICPPSimpleFunction
 public:
     /** Configure the upsample CPP kernel
      *
-     * @param[in]  input  The input tensor to upsample. Data types supported: F32/F16/QASYMM8/QASYMM8_SIGNED
-     * @param[out] output The output tensor. Data types supported: Same as @p input
+     * @param[in]  input  The input tensor to upsample. Data types supported: All.
+     * @param[out] output The output tensor. Data types supported: same as @p input
      * @param[in]  info   Padding information
      */
     void configure(const ITensor *input, ITensor *output, const PadStrideInfo &info);
diff --git a/docs/00_introduction.dox b/docs/00_introduction.dox
index 3668fafbc8..85146cb15d 100644
--- a/docs/00_introduction.dox
+++ b/docs/00_introduction.dox
@@ -343,7 +343,7 @@ v20.02 Public major release
      - @ref NEElementwiseMin
      - @ref NEElementwiseSquaredDiff
      - @ref NEFullyConnectedLayer
-     - @ref NEGEMMMatrixVectorMultiplyKernel
+     - NEGEMMMatrixVectorMultiplyKernel
      - @ref NEPixelWiseMultiplication
      - @ref NEPoolingLayer
      - @ref NEPReluLayer
@@ -915,7 +915,7 @@ v17.12 Public major release
  - New NEON kernels / functions
     - arm_compute::NEGEMMLowpAArch64A53Kernel / arm_compute::NEGEMMLowpAArch64Kernel / arm_compute::NEGEMMLowpAArch64V8P4Kernel / arm_compute::NEGEMMInterleavedBlockedKernel / arm_compute::NEGEMMLowpAssemblyMatrixMultiplyCore
     - arm_compute::NEHGEMMAArch64FP16Kernel
-    - @ref NEDepthwiseConvolutionLayer3x3Kernel / NEDepthwiseIm2ColKernel / @ref NEGEMMMatrixVectorMultiplyKernel / NEDepthwiseVectorToTensorKernel / @ref NEDepthwiseConvolutionLayer
+    - @ref NEDepthwiseConvolutionLayer3x3Kernel / NEDepthwiseIm2ColKernel / NEGEMMMatrixVectorMultiplyKernel / NEDepthwiseVectorToTensorKernel / @ref NEDepthwiseConvolutionLayer
     - @ref NEGEMMLowpOffsetContributionKernel / @ref NEGEMMLowpMatrixAReductionKernel / @ref NEGEMMLowpMatrixBReductionKernel / @ref NEGEMMLowpMatrixMultiplyCore
     - @ref NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel / @ref NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint
     - NEWinogradLayer / NEWinogradLayerKernel
@@ -1037,7 +1037,7 @@ v17.03.1 First Major public release of the sources
    - @ref NETransposeKernel / @ref NETranspose
    - @ref NELogits1DMaxKernel, NELogits1DShiftExpSumKernel, NELogits1DNormKernel / @ref NESoftmaxLayer
    - @ref NEIm2ColKernel, @ref NECol2ImKernel, NEConvolutionLayerWeightsReshapeKernel / @ref NEConvolutionLayer
-   - @ref NEGEMMMatrixAccumulateBiasesKernel / @ref NEFullyConnectedLayer
+   - NEGEMMMatrixAccumulateBiasesKernel / @ref NEFullyConnectedLayer
    - @ref NEGEMMLowpMatrixMultiplyKernel / NEGEMMLowp
 
 v17.03 Sources preview
diff --git a/src/core/NEON/kernels/NEFillInnerBorderKernel.cpp b/src/core/NEON/kernels/NEFillInnerBorderKernel.cpp
deleted file mode 100644
index 50060b2376..0000000000
--- a/src/core/NEON/kernels/NEFillInnerBorderKernel.cpp
+++ /dev/null
@@ -1,138 +0,0 @@
-/*
- * Copyright (c) 2016-2019 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/core/NEON/kernels/NEFillInnerBorderKernel.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/Validate.h"
-#include "arm_compute/core/Window.h"
-
-#include <algorithm>
-#include <cstddef>
-#include <cstdint>
-
-using namespace arm_compute;
-
-namespace arm_compute
-{
-class Coordinates;
-} // namespace arm_compute
-
-NEFillInnerBorderKernel::NEFillInnerBorderKernel()
-    : _tensor(nullptr), _border_size(0), _constant_border_value(static_cast<float>(0.f))
-{
-}
-
-void NEFillInnerBorderKernel::configure(ITensor *input, BorderSize border_size, const PixelValue &constant_border_value)
-{
-    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8, DataType::S16, DataType::S32, DataType::F32);
-
-    _tensor                = input;
-    _border_size           = border_size;
-    _constant_border_value = constant_border_value;
-
-    Window win;
-    win.set(Window::DimX, Window::Dimension(0, 1, 1));
-    win.set(Window::DimY, Window::Dimension(0, 1, 1));
-    win.use_tensor_dimensions(_tensor->info()->tensor_shape(), Window::DimZ);
-    INEKernel::configure(win);
-}
-
-void NEFillInnerBorderKernel::run(const Window &window, const ThreadInfo &info)
-{
-    ARM_COMPUTE_UNUSED(info);
-    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
-    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
-
-    // If there is no border: early exit
-    if(_border_size.empty())
-    {
-        return;
-    }
-
-    switch(_tensor->info()->data_type())
-    {
-        case DataType::U8:
-            fill_value_single_channel<uint8_t>(window);
-            break;
-        case DataType::S16:
-            fill_value_single_channel<int16_t>(window);
-            break;
-        case DataType::S32:
-            fill_value_single_channel<int32_t>(window);
-            break;
-        case DataType::F32:
-            static_assert(sizeof(float) == 4, "Float must be 32 bit");
-            fill_value_single_channel<float>(window);
-            break;
-        default:
-            ARM_COMPUTE_ERROR("Not handled");
-            break;
-    }
-}
-
-template <typename T>
-void NEFillInnerBorderKernel::fill_value_single_channel(const Window &window)
-{
-    const size_t stride = _tensor->info()->strides_in_bytes()[1];
-    const size_t width  = _tensor->info()->dimension(0);
-    const size_t height = _tensor->info()->dimension(1);
-
-    T constant_border_value;
-    _constant_border_value.get(constant_border_value);
-
-    // Left and right border
-    // All X values are set at once
-    Window vertical(window);
-    vertical.set(Window::DimY, Window::Dimension(0, height, 1));
-
-    Iterator vertical_it(_tensor, vertical);
-
-    execute_window_loop(vertical, [&](const Coordinates &)
-    {
-        std::fill_n(reinterpret_cast<T *>(vertical_it.ptr()), _border_size.left, constant_border_value);
-        std::fill_n(reinterpret_cast<T *>(vertical_it.ptr()) + width - _border_size.right, _border_size.right, constant_border_value);
-    },
-    vertical_it);
-
-    // Top and bottom border
-    // All values are set at once
-    Iterator horizontal_it(_tensor, window);
-
-    execute_window_loop(window, [&](const Coordinates &)
-    {
-        for(size_t i = 0; i < _border_size.top; ++i)
-        {
-            std::fill_n(reinterpret_cast<T *>(horizontal_it.ptr() + i * stride), width, constant_border_value);
-        }
-
-        for(size_t i = 0; i < _border_size.bottom; ++i)
-        {
-            std::fill_n(reinterpret_cast<T *>(horizontal_it.ptr() + (height - i - 1) * stride), width, constant_border_value);
-        }
-    },
-    horizontal_it);
-}
diff --git a/src/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.cpp b/src/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.cpp
deleted file mode 100644
index 5ac2323896..0000000000
--- a/src/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.cpp
+++ /dev/null
@@ -1,169 +0,0 @@
-/*
- * Copyright (c) 2017-2019 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/core/NEON/kernels/NEGEMMMatrixAccumulateBiasesKernel.h"
-
-#include "arm_compute/core/AccessWindowStatic.h"
-#include "arm_compute/core/CPP/Validate.h"
-#include "arm_compute/core/Error.h"
-#include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/ITensor.h"
-#include "arm_compute/core/NEON/NEFixedPoint.h"
-#include "arm_compute/core/Types.h"
-#include "arm_compute/core/Utils.h"
-#include "arm_compute/core/Validate.h"
-#include "arm_compute/core/Window.h"
-
-#include <arm_neon.h>
-#include <cstddef>
-#include <cstdint>
-
-using namespace arm_compute;
-
-namespace
-{
-inline Status validate_arguments(const ITensorInfo *accum, const ITensorInfo *biases)
-{
-    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(accum);
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(accum, 1, DataType::F16, DataType::F32);
-    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(biases, accum);
-    ARM_COMPUTE_RETURN_ERROR_ON(biases->num_dimensions() > 1);
-    ARM_COMPUTE_RETURN_ERROR_ON(biases->dimension(0) != accum->dimension(0));
-
-    return Status{};
-}
-
-inline std::pair<Status, Window> validate_and_configure_window(ITensorInfo *accum, ITensorInfo *biases)
-{
-    constexpr unsigned int num_elems_processed_per_iteration = 16;
-
-    // Configure kernel window
-    Window win = calculate_max_window(*accum, Steps(num_elems_processed_per_iteration));
-
-    bool window_changed = update_window_and_padding(win,
-                                                    AccessWindowHorizontal(accum, 0, num_elems_processed_per_iteration),
-                                                    AccessWindowStatic(biases, 0, 0, ceil_to_multiple(biases->dimension(0), num_elems_processed_per_iteration), biases->tensor_shape().y()));
-
-    AccessWindowHorizontal output_access(accum, 0, num_elems_processed_per_iteration);
-
-    // Set the valid region for the accum tensor
-    Coordinates coord;
-    coord.set_num_dimensions(accum->num_dimensions());
-    output_access.set_valid_region(win, ValidRegion(coord, accum->tensor_shape()));
-
-    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
-    return std::make_pair(err, win);
-}
-} // namespace
-
-NEGEMMMatrixAccumulateBiasesKernel::NEGEMMMatrixAccumulateBiasesKernel()
-    : _accum(nullptr), _biases(nullptr)
-{
-}
-
-void NEGEMMMatrixAccumulateBiasesKernel::configure(ITensor *accum, const ITensor *biases)
-{
-    ARM_COMPUTE_ERROR_ON_NULLPTR(accum, biases);
-
-    // Perform validate step
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(accum->info(), biases->info()));
-
-    _biases = biases;
-    _accum  = accum;
-
-    // Configure kernel window
-    auto win_config = validate_and_configure_window(accum->info(), biases->info());
-    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
-    INEKernel::configure(win_config.second);
-}
-
-Status NEGEMMMatrixAccumulateBiasesKernel::validate(const ITensorInfo *accum, const ITensorInfo *biases)
-{
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(accum, biases));
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(accum->clone().get(), biases->clone().get()).first);
-
-    return Status{};
-}
-
-void NEGEMMMatrixAccumulateBiasesKernel::run(const Window &window, const ThreadInfo &info)
-{
-    ARM_COMPUTE_UNUSED(info);
-    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
-    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
-
-    Window win_biases;
-    win_biases.set(Window::DimX, Window::Dimension(window.x().start(), window.x().end(), window.x().step()));
-    win_biases.set(Window::DimY, Window::Dimension(0, 1, 1));
-
-    Iterator in0_out(_accum, window);
-    Iterator in1(_biases, win_biases);
-
-    switch(_accum->info()->data_type())
-    {
-        case DataType::F32:
-        {
-            execute_window_loop(window, [&](const Coordinates &)
-            {
-                const float32x4x4_t accum  = vld4q_f32(reinterpret_cast<const float *>(in0_out.ptr()));
-                const float32x4x4_t biases = vld4q_f32(reinterpret_cast<const float *>(in1.ptr()));
-                const float32x4x4_t res =
-                {
-                    {
-                        vaddq_f32(accum.val[0], biases.val[0]),
-                        vaddq_f32(accum.val[1], biases.val[1]),
-                        vaddq_f32(accum.val[2], biases.val[2]),
-                        vaddq_f32(accum.val[3], biases.val[3])
-                    }
-                };
-
-                vst4q_f32(reinterpret_cast<float *>(in0_out.ptr()), res);
-            },
-            in0_out, in1);
-            break;
-        }
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-        case DataType::F16:
-        {
-            execute_window_loop(window, [&](const Coordinates &)
-            {
-                const float16x8x2_t accum  = vld2q_f16(reinterpret_cast<const float16_t *>(in0_out.ptr()));
-                const float16x8x2_t biases = vld2q_f16(reinterpret_cast<const float16_t *>(in1.ptr()));
-                const float16x8x2_t res =
-                {
-                    {
-                        vaddq_f16(accum.val[0], biases.val[0]),
-                        vaddq_f16(accum.val[1], biases.val[1])
-                    }
-                };
-
-                vst2q_f16(reinterpret_cast<float16_t *>(in0_out.ptr()), res);
-            },
-            in0_out, in1);
-            break;
-        }
-#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
-        default:
-            ARM_COMPUTE_ERROR("Data type not supported");
-            break;
-    }
-}
diff --git a/src/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.cpp b/src/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.cpp
deleted file mode 100644
index cf8411c55f..0000000000
--- a/src/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.cpp
+++ /dev/null
@@ -1,410 +0,0 @@
-/*
- * Copyright (c) 2016-2019 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/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.h"
-
-#include "arm_compute/core/AccessWindowStatic.h"
-#include "arm_compute/core/CPP/Validate.h"
-#include "arm_compute/core/Error.h"
-#include "arm_compute/core/Helpers.h"
-#include "arm_compute/core/ITensor.h"
-#include "arm_compute/core/NEON/INEKernel.h"
-#include "arm_compute/core/Types.h"
-#include "arm_compute/core/Validate.h"
-#include "arm_compute/core/Window.h"
-
-#include <arm_neon.h>
-#include <cstddef>
-#include <cstdint>
-#include <tuple>
-
-namespace arm_compute
-{
-namespace
-{
-Status validate_arguments(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output)
-{
-    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input0);
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input0, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::F16, DataType::F32);
-    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input0, input1);
-    if(is_data_type_quantized_asymmetric(input0->data_type()))
-    {
-        ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::S32);
-    }
-    else
-    {
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input0, output);
-    }
-
-    ARM_COMPUTE_RETURN_ERROR_ON(input0->num_dimensions() == input1->num_dimensions());
-    ARM_COMPUTE_RETURN_ERROR_ON(input0->dimension(2) != input1->dimension(1));
-    ARM_COMPUTE_RETURN_ERROR_ON(input0->dimension(DataLayoutDimension::HEIGHT) != output->dimension(DataLayoutDimension::HEIGHT));
-    ARM_COMPUTE_RETURN_ERROR_ON(input1->dimension(DataLayoutDimension::WIDTH) != output->dimension(DataLayoutDimension::WIDTH));
-
-    return Status{};
-}
-
-std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input0, ITensorInfo *input1, ITensorInfo *output)
-{
-    const unsigned int num_elems_read_per_iteration = 16 / input0->element_size();
-
-    Window win = calculate_max_window(*input0, Steps(num_elems_read_per_iteration));
-
-    AccessWindowHorizontal input0_access(input0, 0, num_elems_read_per_iteration);
-    AccessWindowHorizontal input1_access(input1, 0, num_elems_read_per_iteration);
-    AccessWindowStatic     output_access(output, 0, 0, output->dimension(0), output->dimension(1));
-
-    bool window_changed = update_window_and_padding(win, input0_access, input1_access, output_access);
-
-    output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape()));
-
-    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
-    return std::make_pair(err, win);
-}
-} // namespace
-
-template <typename I0, typename I1, typename O>
-void NEGEMMMatrixVectorMultiplyKernel::matrix_vector_multiply(const Window &window_in, const Window &window_w, const Window &window_out)
-{
-    ARM_COMPUTE_ERROR("Unsupported data types");
-    ARM_COMPUTE_UNUSED(window_in);
-    ARM_COMPUTE_UNUSED(window_w);
-    ARM_COMPUTE_UNUSED(window_out);
-}
-
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-template <>
-void NEGEMMMatrixVectorMultiplyKernel::matrix_vector_multiply<half, half, half>(const Window &window_in,
-                                                                                const Window &window_w,
-                                                                                const Window &window_out)
-{
-    Iterator in(_input0, window_in);
-    Iterator in2(_input1, window_w);
-    Iterator out(_output, window_out);
-
-    const int input_w          = _input0->info()->dimension(0);
-    const int input_h          = _input0->info()->dimension(1);
-    const int input_stride_x   = _input0->info()->strides_in_bytes().x();
-    const int weights_stride_x = _input1->info()->strides_in_bytes().x();
-    const int weights_stride_y = _input1->info()->strides_in_bytes().y();
-    const int output_stride_x  = _output->info()->strides_in_bytes().x();
-
-    execute_window_loop(window_in, [&](const Coordinates & id)
-    {
-        // Get pointers
-        const uint8_t *const input_ptr   = in.ptr();
-        const uint8_t *const weights_ptr = in2.ptr() + id.z() * weights_stride_y;
-        auto                 output_ptr  = reinterpret_cast<__fp16 *>(out.ptr() + (id.y() + id.z() * input_h) * output_stride_x);
-
-        float16x8_t row_dot = vdupq_n_f16(0.f);
-        for(int i = 0; i < input_w; i += 8)
-        {
-            const auto input   = vld1q_f16(reinterpret_cast<const __fp16 *>(input_ptr + i * input_stride_x));
-            const auto weights = vld1q_f16(reinterpret_cast<const __fp16 *>(weights_ptr + i * weights_stride_x));
-            row_dot            = vaddq_f16(row_dot, vmulq_f16(input, weights));
-        }
-
-        auto temp = vadd_f16(vget_high_f16(row_dot), vget_low_f16(row_dot));
-        temp      = vpadd_f16(temp, temp);
-        temp      = vpadd_f16(temp, temp);
-
-        *output_ptr = vget_lane_f16(temp, 0);
-    },
-    in, in2, out);
-}
-#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
-
-template <>
-void NEGEMMMatrixVectorMultiplyKernel::matrix_vector_multiply<float, float, float>(const Window &window_in,
-                                                                                   const Window &window_w,
-                                                                                   const Window &window_out)
-{
-    Iterator in(_input0, window_in);
-    Iterator in2(_input1, window_w);
-    Iterator out(_output, window_out);
-
-    const int input_w          = _input0->info()->dimension(0);
-    const int input_h          = _input0->info()->dimension(1);
-    const int input_stride_x   = _input0->info()->strides_in_bytes().x();
-    const int weights_stride_x = _input1->info()->strides_in_bytes().x();
-    const int weights_stride_y = _input1->info()->strides_in_bytes().y();
-    const int output_stride_x  = _output->info()->strides_in_bytes().x();
-
-    execute_window_loop(window_in, [&](const Coordinates & id)
-    {
-        // Get pointers
-        const uint8_t *const input_ptr   = in.ptr();
-        const uint8_t *const weights_ptr = in2.ptr() + id.z() * weights_stride_y;
-        auto                 output_ptr  = reinterpret_cast<float *>(out.ptr() + (id.y() + id.z() * input_h) * output_stride_x);
-
-        float32x4_t row_dot = vdupq_n_f32(0.f);
-        for(int i = 0; i < input_w; i += 4)
-        {
-            const auto input   = vld1q_f32(reinterpret_cast<const float *>(input_ptr + i * input_stride_x));
-            const auto weights = vld1q_f32(reinterpret_cast<const float *>(weights_ptr + i * weights_stride_x));
-            row_dot            = vaddq_f32(row_dot, vmulq_f32(input, weights));
-        }
-
-        auto temp = vadd_f32(vget_high_f32(row_dot), vget_low_f32(row_dot));
-        temp      = vpadd_f32(temp, temp);
-
-        *output_ptr = vget_lane_f32(temp, 0);
-    },
-    in, in2, out);
-}
-
-template <>
-void NEGEMMMatrixVectorMultiplyKernel::matrix_vector_multiply<uint8_t, uint8_t, int32_t>(const Window &window_in,
-                                                                                         const Window &window_w,
-                                                                                         const Window &window_out)
-{
-    Iterator in(_input0, window_in);
-    Iterator in2(_input1, window_w);
-    Iterator out(_output, window_out);
-
-    const int input_offset   = -_input0->info()->quantization_info().uniform().offset;
-    const int weights_offset = -_input1->info()->quantization_info().uniform().offset;
-
-    const int input_w          = _input0->info()->dimension(0);
-    const int input_h          = _input0->info()->dimension(1);
-    const int input_stride_x   = _input0->info()->strides_in_bytes().x();
-    const int weights_stride_x = _input1->info()->strides_in_bytes().x();
-    const int weights_stride_y = _input1->info()->strides_in_bytes().y();
-    const int output_stride_x  = _output->info()->strides_in_bytes().x();
-    const int read_step        = 16 / _input0->info()->element_size();
-
-    const int32x4_t v_input_offset   = vdupq_n_s32(input_offset);
-    const int32x4_t v_weights_offset = vdupq_n_s32(weights_offset);
-
-    execute_window_loop(window_in, [&](const Coordinates & id)
-    {
-        // Get pointers
-        const uint8_t *const input_ptr   = in.ptr();
-        const uint8_t *const weights_ptr = in2.ptr() + id.z() * weights_stride_y;
-        auto                 output_ptr  = reinterpret_cast<int32_t *>(out.ptr() + (id.y() + id.z() * input_h) * output_stride_x);
-
-        int32x4_t row_dot = vdupq_n_s32(0);
-        for(int i = 0; i < input_w; i += read_step)
-        {
-            // Read values
-            const auto input   = vld1q_u8(reinterpret_cast<const uint8_t *>(input_ptr + i * input_stride_x));
-            const auto weights = vld1q_u8(reinterpret_cast<const uint8_t *>(weights_ptr + i * weights_stride_x));
-
-            // Add offsets
-            const int32x4x4_t input_s32 =
-            {
-                {
-                    vaddw_s16(v_input_offset, vreinterpret_s16_u16(vget_low_u16(vmovl_u8(vget_low_u8(input))))),
-                    vaddw_s16(v_input_offset, vreinterpret_s16_u16(vget_high_u16(vmovl_u8(vget_low_u8(input))))),
-                    vaddw_s16(v_input_offset, vreinterpret_s16_u16(vget_low_u16(vmovl_u8(vget_high_u8(input))))),
-                    vaddw_s16(v_input_offset, vreinterpret_s16_u16(vget_high_u16(vmovl_u8(vget_high_u8(input)))))
-                }
-            };
-            const int32x4x4_t weights_s32 =
-            {
-                {
-                    vaddw_s16(v_weights_offset, vreinterpret_s16_u16(vget_low_u16(vmovl_u8(vget_low_u8(weights))))),
-                    vaddw_s16(v_weights_offset, vreinterpret_s16_u16(vget_high_u16(vmovl_u8(vget_low_u8(weights))))),
-                    vaddw_s16(v_weights_offset, vreinterpret_s16_u16(vget_low_u16(vmovl_u8(vget_high_u8(weights))))),
-                    vaddw_s16(v_weights_offset, vreinterpret_s16_u16(vget_high_u16(vmovl_u8(vget_high_u8(weights)))))
-                }
-            };
-
-            // Dot
-            row_dot = vaddq_s32(row_dot, vmulq_s32(input_s32.val[0], weights_s32.val[0]));
-            row_dot = vaddq_s32(row_dot, vmulq_s32(input_s32.val[1], weights_s32.val[1]));
-            row_dot = vaddq_s32(row_dot, vmulq_s32(input_s32.val[2], weights_s32.val[2]));
-            row_dot = vaddq_s32(row_dot, vmulq_s32(input_s32.val[3], weights_s32.val[3]));
-        }
-
-        // Reduction
-        auto temp = vadd_s32(vget_high_s32(row_dot), vget_low_s32(row_dot));
-        temp      = vpadd_s32(temp, temp);
-
-        *output_ptr = vget_lane_s32(temp, 0);
-    },
-    in, in2, out);
-}
-
-template <>
-void NEGEMMMatrixVectorMultiplyKernel::matrix_vector_multiply<int8_t, int8_t, int32_t>(const Window &window_in,
-                                                                                       const Window &window_w,
-                                                                                       const Window &window_out)
-{
-    Iterator in(_input0, window_in);
-    Iterator in2(_input1, window_w);
-    Iterator out(_output, window_out);
-
-    const int input_offset   = -_input0->info()->quantization_info().uniform().offset;
-    const int weights_offset = -_input1->info()->quantization_info().uniform().offset;
-
-    const int input_w          = _input0->info()->dimension(0);
-    const int input_h          = _input0->info()->dimension(1);
-    const int input_stride_x   = _input0->info()->strides_in_bytes().x();
-    const int weights_stride_x = _input1->info()->strides_in_bytes().x();
-    const int weights_stride_y = _input1->info()->strides_in_bytes().y();
-    const int output_stride_x  = _output->info()->strides_in_bytes().x();
-    const int read_step        = 16 / _input0->info()->element_size();
-
-    const int32x4_t v_input_offset   = vdupq_n_s32(input_offset);
-    const int32x4_t v_weights_offset = vdupq_n_s32(weights_offset);
-
-    execute_window_loop(window_in, [&](const Coordinates & id)
-    {
-        // Get pointers
-        const uint8_t *const input_ptr   = in.ptr();
-        const uint8_t *const weights_ptr = in2.ptr() + id.z() * weights_stride_y;
-        auto                 output_ptr  = reinterpret_cast<int32_t *>(out.ptr() + (id.y() + id.z() * input_h) * output_stride_x);
-
-        int32x4_t row_dot = vdupq_n_s32(0);
-        for(int i = 0; i < input_w; i += read_step)
-        {
-            // Read values
-            const auto input   = vld1q_s8(reinterpret_cast<const int8_t *>(input_ptr + i * input_stride_x));
-            const auto weights = vld1q_s8(reinterpret_cast<const int8_t *>(weights_ptr + i * weights_stride_x));
-
-            // Add offsets
-            const int32x4x4_t input_s32 =
-            {
-                {
-                    vaddw_s16(v_input_offset, vget_low_s16(vmovl_s8(vget_low_s8(input)))),
-                    vaddw_s16(v_input_offset, vget_high_s16(vmovl_s8(vget_low_s8(input)))),
-                    vaddw_s16(v_input_offset, vget_low_s16(vmovl_s8(vget_high_s8(input)))),
-                    vaddw_s16(v_input_offset, vget_high_s16(vmovl_s8(vget_high_s8(input))))
-                }
-            };
-            const int32x4x4_t weights_s32 =
-            {
-                {
-                    vaddw_s16(v_weights_offset, vget_low_s16(vmovl_s8(vget_low_s8(weights)))),
-                    vaddw_s16(v_weights_offset, vget_high_s16(vmovl_s8(vget_low_s8(weights)))),
-                    vaddw_s16(v_weights_offset, vget_low_s16(vmovl_s8(vget_high_s8(weights)))),
-                    vaddw_s16(v_weights_offset, vget_high_s16(vmovl_s8(vget_high_s8(weights))))
-                }
-            };
-
-            // Dot
-            row_dot = vaddq_s32(row_dot, vmulq_s32(input_s32.val[0], weights_s32.val[0]));
-            row_dot = vaddq_s32(row_dot, vmulq_s32(input_s32.val[1], weights_s32.val[1]));
-            row_dot = vaddq_s32(row_dot, vmulq_s32(input_s32.val[2], weights_s32.val[2]));
-            row_dot = vaddq_s32(row_dot, vmulq_s32(input_s32.val[3], weights_s32.val[3]));
-        }
-
-        // Reduction
-        auto temp = vadd_s32(vget_high_s32(row_dot), vget_low_s32(row_dot));
-        temp      = vpadd_s32(temp, temp);
-
-        *output_ptr = vget_lane_s32(temp, 0);
-    },
-    in, in2, out);
-}
-
-NEGEMMMatrixVectorMultiplyKernel::NEGEMMMatrixVectorMultiplyKernel()
-    : _func(nullptr), _input0(nullptr), _input1(nullptr), _output(nullptr), _border_size(0)
-{
-}
-
-BorderSize NEGEMMMatrixVectorMultiplyKernel::border_size() const
-{
-    return _border_size;
-}
-
-void NEGEMMMatrixVectorMultiplyKernel::configure(const ITensor *input0, const ITensor *input1, ITensor *output)
-{
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input0, input1, output);
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input0->info(), input1->info(), output->info()));
-
-    _input0 = input0;
-    _input1 = input1;
-    _output = output;
-
-    // Set appropriate function to run
-    switch(input0->info()->data_type())
-    {
-        case DataType::QASYMM8:
-            _func = &NEGEMMMatrixVectorMultiplyKernel::matrix_vector_multiply<uint8_t, uint8_t, int32_t>;
-            break;
-        case DataType::QASYMM8_SIGNED:
-            _func = &NEGEMMMatrixVectorMultiplyKernel::matrix_vector_multiply<int8_t, int8_t, int32_t>;
-            break;
-#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-        case DataType::F16:
-            _func = &NEGEMMMatrixVectorMultiplyKernel::matrix_vector_multiply<half, half, half>;
-            break;
-#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
-        case DataType::F32:
-            _func = &NEGEMMMatrixVectorMultiplyKernel::matrix_vector_multiply<float, float, float>;
-            break;
-        default:
-            ARM_COMPUTE_ERROR("Unsupported data type");
-    }
-
-    // Configure kernel window
-    const unsigned int num_elems_read_per_iteration = 16 / _input0->info()->element_size();
-
-    const unsigned int border_x = ceil_to_multiple(input0->info()->dimension(0), num_elems_read_per_iteration) - input0->info()->dimension(0);
-    _border_size                = BorderSize(0, border_x);
-
-    auto win_config = validate_and_configure_window(input0->info(), input1->info(), output->info());
-    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
-    INEKernel::configure(win_config.second);
-}
-
-Status NEGEMMMatrixVectorMultiplyKernel::validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output)
-{
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input0, input1, output);
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input0, input1, output));
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input0->clone().get(), input1->clone().get(), output->clone().get()).first);
-    return Status{};
-}
-
-void NEGEMMMatrixVectorMultiplyKernel::run(const Window &window, const ThreadInfo &info)
-{
-    ARM_COMPUTE_UNUSED(info);
-    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
-    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
-    ARM_COMPUTE_ERROR_ON(_func == nullptr);
-
-    Window window_slice = window.first_slice_window_3D();
-
-    Window window_in(window);
-    Window window_weights(window_slice);
-    Window window_out(window);
-
-    // Setup input0 slice
-    window_in.set(Window::DimX, Window::Dimension(0, _input0->info()->dimension(0), _input0->info()->dimension(0)));
-    window_in.set(Window::DimY, Window::Dimension(0, _input0->info()->dimension(1), 1));
-    window_in.set(Window::DimZ, Window::Dimension(0, _input0->info()->dimension(2), 1));
-
-    // Setup input1 and output slice. Their dimensions are increased in the kernel.
-    window_weights.set(Window::DimX, Window::Dimension(0, 0, 0));
-    window_weights.set(Window::DimY, Window::Dimension(0, 0, 0));
-    window_weights.set(Window::DimZ, Window::Dimension(0, 0, 0));
-
-    window_out.set(Window::DimX, Window::Dimension(0, 0, 0));
-    window_out.set(Window::DimY, Window::Dimension(0, 0, 0));
-    window_out.set(Window::DimZ, Window::Dimension(0, 0, 0));
-
-    (this->*_func)(window_in, window_weights, window_out);
-}
-} // namespace arm_compute