COMPMID-1734: Implement CLSelect

Change-Id: I49b2e8b4200c9ed654736d9451e4ab9c073b4b10

12 files changed, 1155 insertions, 0 deletions

diff --git a/arm_compute/core/CL/CLKernels.h b/arm_compute/core/CL/CLKernels.h
index cc76231345..1457d7b3ac 100644
--- a/arm_compute/core/CL/CLKernels.h
+++ b/arm_compute/core/CL/CLKernels.h
@@ -116,6 +116,7 @@
 #include "arm_compute/core/CL/kernels/CLReshapeLayerKernel.h"
 #include "arm_compute/core/CL/kernels/CLScaleKernel.h"
 #include "arm_compute/core/CL/kernels/CLScharr3x3Kernel.h"
+#include "arm_compute/core/CL/kernels/CLSelectKernel.h"
 #include "arm_compute/core/CL/kernels/CLSobel3x3Kernel.h"
 #include "arm_compute/core/CL/kernels/CLSobel5x5Kernel.h"
 #include "arm_compute/core/CL/kernels/CLSobel7x7Kernel.h"
diff --git a/arm_compute/core/CL/kernels/CLSelectKernel.h b/arm_compute/core/CL/kernels/CLSelectKernel.h
new file mode 100644
index 0000000000..031166c37c
--- /dev/null
+++ b/arm_compute/core/CL/kernels/CLSelectKernel.h
@@ -0,0 +1,85 @@
+/*
+ * Copyright (c) 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_CLSELECTKERNEL_H__
+#define __ARM_COMPUTE_CLSELECTKERNEL_H__
+
+#include "arm_compute/core/CL/ICLKernel.h"
+#include "arm_compute/core/Types.h"
+
+namespace arm_compute
+{
+// Forward declarations
+class ICLTensor;
+
+/** OpenCL interface for executing the select kernel
+ *
+ * Select is computed by:
+ * @f[ output(i) = condition(i) ? x(i) : y(i) @f]
+ **/
+class CLSelectKernel : public ICLKernel
+{
+public:
+    /** Default constructor */
+    CLSelectKernel();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLSelectKernel(const CLSelectKernel &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLSelectKernel &operator=(const CLSelectKernel &) = delete;
+    /** Allow instances of this class to be moved */
+    CLSelectKernel(CLSelectKernel &&) = default;
+    /** Allow instances of this class to be moved */
+    CLSelectKernel &operator=(CLSelectKernel &&) = default;
+    /** Default destructor */
+    ~CLSelectKernel() = default;
+    /** Initialise the kernel's inputs and output.
+     *
+     * @param[in]  c      Condition input tensor. Data types supported: U8.
+     * @param[in]  x      First input tensor. Data types supported: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32.
+     * @param[out] y      Second input tensor. Data types supported: Same as @p x
+     * @param[in]  output Output tensor. Data types supported: Same as @p x.
+     */
+    void configure(const ICLTensor *c, const ICLTensor *x, const ICLTensor *y, ICLTensor *output);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLSelectKernel
+     *
+     * @param[in] c      Condition input tensor. Data types supported: U8.
+     * @param[in] x      First input tensor. Data types supported: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32.
+     * @param[in] y      Second input tensor. Data types supported: Same as @p x
+     * @param[in] output Output tensor. Data types supported: Same as @p x.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *c, const ITensorInfo *x, const ITensorInfo *y, const ITensorInfo *output);
+
+    // Inherited methods overridden:
+    void run(const Window &window, cl::CommandQueue &queue) override;
+
+private:
+    const ICLTensor *_c;             /**< Condition tensor */
+    const ICLTensor *_x;             /**< Source tensor 1 */
+    const ICLTensor *_y;             /**< Source tensor 2 */
+    ICLTensor       *_output;        /**< Destination tensor */
+    bool             _has_same_rank; /**< Flag that indicates if condition tensor and other inputs have the same rank */
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_CLWHEREKERNEL_H__ */
diff --git a/arm_compute/runtime/CL/CLFunctions.h b/arm_compute/runtime/CL/CLFunctions.h
index 2a8294f28b..b4c5c2eb80 100644
--- a/arm_compute/runtime/CL/CLFunctions.h
+++ b/arm_compute/runtime/CL/CLFunctions.h
@@ -117,6 +117,7 @@
 #include "arm_compute/runtime/CL/functions/CLReshapeLayer.h"
 #include "arm_compute/runtime/CL/functions/CLScale.h"
 #include "arm_compute/runtime/CL/functions/CLScharr3x3.h"
+#include "arm_compute/runtime/CL/functions/CLSelect.h"
 #include "arm_compute/runtime/CL/functions/CLSlice.h"
 #include "arm_compute/runtime/CL/functions/CLSobel3x3.h"
 #include "arm_compute/runtime/CL/functions/CLSobel5x5.h"
diff --git a/arm_compute/runtime/CL/functions/CLSelect.h b/arm_compute/runtime/CL/functions/CLSelect.h
new file mode 100644
index 0000000000..dab62a7ec1
--- /dev/null
+++ b/arm_compute/runtime/CL/functions/CLSelect.h
@@ -0,0 +1,60 @@
+/*
+ * Copyright (c) 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_CLSELECT_H__
+#define __ARM_COMPUTE_CLSELECT_H__
+
+#include "arm_compute/runtime/CL/ICLSimpleFunction.h"
+
+#include "arm_compute/core/Types.h"
+
+namespace arm_compute
+{
+// Forward declarations
+class ICLTensor;
+
+/** Basic function to run @ref CLSelect */
+class CLSelect : public ICLSimpleFunction
+{
+public:
+    /** Initialise the kernel's inputs and output.
+     *
+     * @param[in]  c      Condition input tensor. Data types supported: U8.
+     * @param[in]  x      First input tensor. Data types supported: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32.
+     * @param[in]  y      Second input tensor. Data types supported: Same as @p x
+     * @param[out] output Output tensor. Data types supported: Same as @p x.
+     */
+    void configure(const ICLTensor *c, const ICLTensor *x, const ICLTensor *y, ICLTensor *output);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLSelect
+     *
+     * @param[in] c      Condition input tensor. Data types supported: U8.
+     * @param[in] x      First input tensor. Data types supported: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32.
+     * @param[in] y      Second input tensor. Data types supported: Same as @p x
+     * @param[in] output Output tensor. Data types supported: Same as @p x.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *c, const ITensorInfo *x, const ITensorInfo *y, const ITensorInfo *output);
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_CLSELECT_H__ */
diff --git a/src/core/CL/CLKernelLibrary.cpp b/src/core/CL/CLKernelLibrary.cpp
index 3a002e808d..12944061a9 100644
--- a/src/core/CL/CLKernelLibrary.cpp
+++ b/src/core/CL/CLKernelLibrary.cpp
@@ -386,6 +386,9 @@ const std::map<std::string, std::string> CLKernelLibrary::_kernel_program_map =
     { "scale_bilinear_quantized_nchw", "scale_quantized.cl" },
     { "scale_bilinear_quantized_nhwc", "scale_quantized.cl" },
     { "scharr3x3", "scharr_filter.cl" },
+    { "select_same_rank", "select.cl" },
+    { "select_different_rank_2", "select.cl" },
+    { "select_different_rank_n", "select.cl" },
     { "sobel3x3", "sobel_filter.cl" },
     { "sobel_separable5x1", "sobel_filter.cl" },
     { "sobel_separable1x5", "sobel_filter.cl" },
@@ -814,6 +817,10 @@ const std::map<std::string, std::string> CLKernelLibrary::_program_source_map =
 #include "./cl_kernels/scharr_filter.clembed"
     },
     {
+        "select.cl",
+#include "./cl_kernels/select.clembed"
+    },
+    {
         "sobel_filter.cl",
 #include "./cl_kernels/sobel_filter.clembed"
     },
diff --git a/src/core/CL/cl_kernels/select.cl b/src/core/CL/cl_kernels/select.cl
new file mode 100644
index 0000000000..d783ae212e
--- /dev/null
+++ b/src/core/CL/cl_kernels/select.cl
@@ -0,0 +1,216 @@
+/*
+ * Copyright (c) 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 "helpers.h"
+
+#if defined(DATA_TYPE) && defined(SELECT_DATA_TYPE) && defined(VEC_SIZE)
+/** This function perform a select operation between two tensors when condition tensor has the same rank.
+ *
+ * @attention The data_type need to be passed at compile time using -DDATA_TYPE: e.g. -DDATA_TYPE=uchar
+ * @attention The select operation data_type need to be passed at compile time using -DSELECT_DATA_TYPE: e.g. -DSELECT_DATA_TYPE=uchar
+ * @attention Vector size should be given as a preprocessor argument using -DVEC_SIZE=size. e.g. -DVEC_SIZE=16
+ *
+ * @param[in]  c_ptr                             Pointer to the source tensor. Supported data types: U8
+ * @param[in]  c_stride_x                        Stride of the source tensor in X dimension (in bytes)
+ * @param[in]  c_step_x                          c_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  c_stride_y                        Stride of the source tensor in Y dimension (in bytes)
+ * @param[in]  c_step_y                          c_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  c_stride_z                        Stride of the source tensor in Z dimension (in bytes)
+ * @param[in]  c_step_z                          c_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  c_offset_first_element_in_bytes   The offset of the first element in the source tensor
+ * @param[in]  x_ptr                             Pointer to the source tensor. Supported data types: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32
+ * @param[in]  x_stride_x                        Stride of the source tensor in X dimension (in bytes)
+ * @param[in]  x_step_x                          x_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  x_stride_y                        Stride of the source tensor in Y dimension (in bytes)
+ * @param[in]  x_step_y                          x_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  x_stride_z                        Stride of the source tensor in Z dimension (in bytes)
+ * @param[in]  x_step_z                          x_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  x_offset_first_element_in_bytes   The offset of the first element in the source tensor
+ * @param[in]  y_ptr                             Pointer to the source tensor. Supported data types: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32
+ * @param[in]  y_stride_x                        Stride of the source tensor in X dimension (in bytes)
+ * @param[in]  y_step_x                          y_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  y_stride_y                        Stride of the source tensor in Y dimension (in bytes)
+ * @param[in]  y_step_y                          y_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  y_stride_z                        Stride of the source tensor in Z dimension (in bytes)
+ * @param[in]  y_step_z                          y_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  y_offset_first_element_in_bytes   The offset of the first element in the source tensor
+ * @param[out] out_ptr                           Pointer to the destination tensor. Supported data types: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32
+ * @param[in]  out_stride_x                      Stride of the destination tensor in X dimension (in bytes)
+ * @param[in]  out_step_x                        out_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  out_stride_y                      Stride of the destination tensor in Y dimension (in bytes)
+ * @param[in]  out_step_y                        out_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  out_stride_z                      Stride of the source tensor in Z dimension (in bytes)
+ * @param[in]  out_step_z                        out_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  out_offset_first_element_in_bytes The offset of the first element in the destination tensor
+ */
+__kernel void select_same_rank(
+    TENSOR3D_DECLARATION(c),
+    TENSOR3D_DECLARATION(x),
+    TENSOR3D_DECLARATION(y),
+    TENSOR3D_DECLARATION(out))
+{
+    // Get pixels pointer
+    Tensor3D c_t   = CONVERT_TO_TENSOR3D_STRUCT(c);
+    Tensor3D x_t   = CONVERT_TO_TENSOR3D_STRUCT(x);
+    Tensor3D y_t   = CONVERT_TO_TENSOR3D_STRUCT(y);
+    Tensor3D out_t = CONVERT_TO_TENSOR3D_STRUCT(out);
+
+    // Load values
+    VEC_DATA_TYPE(SELECT_DATA_TYPE, VEC_SIZE)
+    in_c = CONVERT((VLOAD(VEC_SIZE)(0, (__global uchar *)c_t.ptr)), VEC_DATA_TYPE(SELECT_DATA_TYPE, VEC_SIZE));
+    VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)
+    in_x = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)x_t.ptr);
+    VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)
+    in_y = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)y_t.ptr);
+
+    // Calculate and store result
+    VSTORE(VEC_SIZE)
+    (select(in_y, in_x, in_c > (SELECT_DATA_TYPE)0), 0, (__global DATA_TYPE *)out_t.ptr);
+}
+
+/** This function perform a select operation between two tensors when condition tensor has a different rank.
+ *
+ * @attention The data_type need to be passed at compile time using -DDATA_TYPE: e.g. -DDATA_TYPE=uchar
+ * @attention The select operation data_type need to be passed at compile time using -DSELECT_DATA_TYPE: e.g. -DSELECT_DATA_TYPE=uchar
+ * @attention Vector size should be given as a preprocessor argument using -DVEC_SIZE=size. e.g. -DVEC_SIZE=16
+ *
+ * @param[in]  c_ptr                             Pointer to the source tensor. Supported data types: U8
+ * @param[in]  c_stride_x                        Stride of the source tensor in X dimension (in bytes)
+ * @param[in]  c_step_x                          c_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  c_offset_first_element_in_bytes   The offset of the first element in the source tensor
+ * @param[in]  x_ptr                             Pointer to the source tensor. Supported data types: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32
+ * @param[in]  x_stride_x                        Stride of the source tensor in X dimension (in bytes)
+ * @param[in]  x_step_x                          x_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  x_stride_y                        Stride of the source tensor in Y dimension (in bytes)
+ * @param[in]  x_step_y                          x_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  x_stride_z                        Stride of the source tensor in Z dimension (in bytes)
+ * @param[in]  x_step_z                          x_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  x_offset_first_element_in_bytes   The offset of the first element in the source tensor
+ * @param[in]  y_ptr                             Pointer to the source tensor. Supported data types: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32
+ * @param[in]  y_stride_x                        Stride of the source tensor in X dimension (in bytes)
+ * @param[in]  y_step_x                          y_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  y_stride_y                        Stride of the source tensor in Y dimension (in bytes)
+ * @param[in]  y_step_y                          y_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  y_stride_z                        Stride of the source tensor in Z dimension (in bytes)
+ * @param[in]  y_step_z                          y_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  y_offset_first_element_in_bytes   The offset of the first element in the source tensor
+ * @param[out] out_ptr                           Pointer to the destination tensor. Supported data types: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32
+ * @param[in]  out_stride_x                      Stride of the destination tensor in X dimension (in bytes)
+ * @param[in]  out_step_x                        out_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  out_stride_y                      Stride of the destination tensor in Y dimension (in bytes)
+ * @param[in]  out_step_y                        out_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  out_stride_z                      Stride of the source tensor in Z dimension (in bytes)
+ * @param[in]  out_step_z                        out_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  out_offset_first_element_in_bytes The offset of the first element in the destination tensor
+ */
+__kernel void select_different_rank_2(
+    VECTOR_DECLARATION(c),
+    TENSOR3D_DECLARATION(x),
+    TENSOR3D_DECLARATION(y),
+    TENSOR3D_DECLARATION(out))
+{
+    const int c_idx = get_global_id(1);
+
+    // Get pixels pointer
+    Vector   c_t   = CONVERT_TO_VECTOR_STRUCT_NO_STEP(c);
+    Tensor3D x_t   = CONVERT_TO_TENSOR3D_STRUCT(x);
+    Tensor3D y_t   = CONVERT_TO_TENSOR3D_STRUCT(y);
+    Tensor3D out_t = CONVERT_TO_TENSOR3D_STRUCT(out);
+
+    // Load values
+    VEC_DATA_TYPE(SELECT_DATA_TYPE, VEC_SIZE)
+    in_c = *((__global uchar *)(c_t.ptr + c_idx * c_t.stride_x));
+    VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)
+    in_x = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)x_t.ptr);
+    VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)
+    in_y = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)y_t.ptr);
+
+    // Calculate and store result
+    VSTORE(VEC_SIZE)
+    (select(in_y, in_x, in_c > (SELECT_DATA_TYPE)0), 0, (__global DATA_TYPE *)out_t.ptr);
+}
+#endif /* defined(DATA_TYPE) && defined(SELECT_DATA_TYPE) && defined(VEC_SIZE) */
+
+#if defined(DATA_TYPE) && defined(SELECT_DATA_TYPE) && defined(VEC_SIZE) && defined(DEPTH_SIZE)
+/** This function perform a select operation between two tensors when condition tensor has a different rank.
+ *
+ * @attention The data_type need to be passed at compile time using -DDATA_TYPE: e.g. -DDATA_TYPE=uchar
+ * @attention The select operation data_type need to be passed at compile time using -DSELECT_DATA_TYPE: e.g. -DSELECT_DATA_TYPE=uchar
+ * @attention Vector size should be given as a preprocessor argument using -DVEC_SIZE=size. e.g. -DVEC_SIZE=16
+ *
+ * @param[in]  c_ptr                             Pointer to the source tensor. Supported data types: U8
+ * @param[in]  c_stride_x                        Stride of the source tensor in X dimension (in bytes)
+ * @param[in]  c_step_x                          c_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  c_offset_first_element_in_bytes   The offset of the first element in the source tensor
+ * @param[in]  x_ptr                             Pointer to the source tensor. Supported data types: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32
+ * @param[in]  x_stride_x                        Stride of the source tensor in X dimension (in bytes)
+ * @param[in]  x_step_x                          x_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  x_stride_y                        Stride of the source tensor in Y dimension (in bytes)
+ * @param[in]  x_step_y                          x_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  x_stride_z                        Stride of the source tensor in Z dimension (in bytes)
+ * @param[in]  x_step_z                          x_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  x_offset_first_element_in_bytes   The offset of the first element in the source tensor
+ * @param[in]  y_ptr                             Pointer to the source tensor. Supported data types: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32
+ * @param[in]  y_stride_x                        Stride of the source tensor in X dimension (in bytes)
+ * @param[in]  y_step_x                          y_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  y_stride_y                        Stride of the source tensor in Y dimension (in bytes)
+ * @param[in]  y_step_y                          y_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  y_stride_z                        Stride of the source tensor in Z dimension (in bytes)
+ * @param[in]  y_step_z                          y_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  y_offset_first_element_in_bytes   The offset of the first element in the source tensor
+ * @param[out] out_ptr                           Pointer to the destination tensor. Supported data types: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32
+ * @param[in]  out_stride_x                      Stride of the destination tensor in X dimension (in bytes)
+ * @param[in]  out_step_x                        out_stride_x * number of elements along X processed per workitem(in bytes)
+ * @param[in]  out_stride_y                      Stride of the destination tensor in Y dimension (in bytes)
+ * @param[in]  out_step_y                        out_stride_y * number of elements along Y processed per workitem(in bytes)
+ * @param[in]  out_stride_z                      Stride of the source tensor in Z dimension (in bytes)
+ * @param[in]  out_step_z                        out_stride_z * number of elements along Z processed per workitem(in bytes)
+ * @param[in]  out_offset_first_element_in_bytes The offset of the first element in the destination tensor
+ */
+__kernel void select_different_rank_n(
+    VECTOR_DECLARATION(c),
+    TENSOR3D_DECLARATION(x),
+    TENSOR3D_DECLARATION(y),
+    TENSOR3D_DECLARATION(out))
+{
+    const int c_idx = get_global_id(2) / DEPTH_SIZE;
+
+    // Get pixels pointer
+    Vector   c_t   = CONVERT_TO_VECTOR_STRUCT_NO_STEP(c);
+    Tensor3D x_t   = CONVERT_TO_TENSOR3D_STRUCT(x);
+    Tensor3D y_t   = CONVERT_TO_TENSOR3D_STRUCT(y);
+    Tensor3D out_t = CONVERT_TO_TENSOR3D_STRUCT(out);
+
+    // Load values
+    VEC_DATA_TYPE(SELECT_DATA_TYPE, VEC_SIZE)
+    in_c = *((__global uchar *)(c_t.ptr + c_idx * c_t.stride_x));
+    VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)
+    in_x = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)x_t.ptr);
+    VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)
+    in_y = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)y_t.ptr);
+
+    // Calculate and store result
+    VSTORE(VEC_SIZE)
+    (select(in_y, in_x, in_c > (SELECT_DATA_TYPE)0), 0, (__global DATA_TYPE *)out_t.ptr);
+}
+#endif /* defined(DATA_TYPE) && defined(SELECT_DATA_TYPE) && defined(VEC_SIZE) && defined(DEPTH_SIZE) */
\ No newline at end of file
diff --git a/src/core/CL/kernels/CLSelectKernel.cpp b/src/core/CL/kernels/CLSelectKernel.cpp
new file mode 100644
index 0000000000..c9e5da0670
--- /dev/null
+++ b/src/core/CL/kernels/CLSelectKernel.cpp
@@ -0,0 +1,203 @@
+/*
+ * Copyright (c) 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/CL/kernels/CLSelectKernel.h"
+
+#include "arm_compute/core/CL/CLHelpers.h"
+#include "arm_compute/core/CL/CLKernelLibrary.h"
+#include "arm_compute/core/CL/CLValidate.h"
+#include "arm_compute/core/CL/ICLTensor.h"
+#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/core/Utils.h"
+#include "arm_compute/core/Window.h"
+
+#include "support/ToolchainSupport.h"
+
+namespace arm_compute
+{
+namespace
+{
+Status validate_arguments(const ITensorInfo *c, const ITensorInfo *x, const ITensorInfo *y, const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(x);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(x,
+                                                         1,
+                                                         DataType::U8, DataType::S8, DataType::QASYMM8,
+                                                         DataType::U16, DataType::S16,
+                                                         DataType::U32, DataType::S32,
+                                                         DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(x, y);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(x, y);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(c, 1, DataType::U8);
+
+    const bool is_same_rank = (c->tensor_shape().num_dimensions() == x->tensor_shape().num_dimensions());
+    ARM_COMPUTE_RETURN_ERROR_ON(is_same_rank && (x->tensor_shape() != c->tensor_shape()));
+    ARM_COMPUTE_RETURN_ERROR_ON(!is_same_rank && ((c->tensor_shape().num_dimensions() > 1) || (c->tensor_shape().x() != x->tensor_shape()[x->tensor_shape().num_dimensions() - 1])));
+
+    if(output != nullptr && output->total_size() != 0)
+    {
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(x, output);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(x, output);
+    }
+
+    return Status{};
+}
+
+std::pair<Status, Window> validate_and_configure_window(ITensorInfo *c, ITensorInfo *x, ITensorInfo *y, ITensorInfo *output)
+{
+    if(output != nullptr)
+    {
+        // Output tensor auto initialization if not yet initialized
+        auto_init_if_empty(*output, *x->clone());
+    }
+
+    const bool is_same_rank = (c->tensor_shape().num_dimensions() == x->tensor_shape().num_dimensions());
+
+    const unsigned int num_elems_processed_per_iteration = 16 / x->element_size();
+
+    // Configure kernel window
+    Window                 win = calculate_max_window(*x, Steps(num_elems_processed_per_iteration));
+    AccessWindowHorizontal x_access(x, 0, num_elems_processed_per_iteration);
+    AccessWindowHorizontal y_access(y, 0, num_elems_processed_per_iteration);
+    bool                   window_changed = update_window_and_padding(win, x_access, y_access);
+
+    // Update window for condition
+    if(is_same_rank)
+    {
+        AccessWindowHorizontal c_access(c, 0, num_elems_processed_per_iteration);
+        window_changed = window_changed || update_window_and_padding(win, c_access);
+    }
+
+    // Update window for output
+    if(output != nullptr)
+    {
+        AccessWindowHorizontal output_access(output, 0, num_elems_processed_per_iteration);
+        window_changed = window_changed || update_window_and_padding(win, output_access);
+        output_access.set_valid_region(win, x->valid_region());
+    }
+
+    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
+    return std::make_pair(err, win);
+}
+} // namespace
+
+CLSelectKernel::CLSelectKernel()
+    : _c(nullptr), _x(nullptr), _y(nullptr), _output(nullptr), _has_same_rank(false)
+{
+}
+void CLSelectKernel::configure(const ICLTensor *c, const ICLTensor *x, const ICLTensor *y, ICLTensor *output)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(c, x, y, output);
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(c->info(), x->info(), y->info(), output->info()));
+
+    _c             = c;
+    _x             = x;
+    _y             = y;
+    _output        = output;
+    _has_same_rank = (c->info()->tensor_shape().num_dimensions() == x->info()->tensor_shape().num_dimensions());
+
+    const unsigned int num_elems_processed_per_iteration = 16 / x->info()->element_size();
+
+    // Set build options
+    CLBuildOptions build_opts;
+    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(x->info()->data_type()));
+    build_opts.add_option("-DSELECT_DATA_TYPE=" + get_cl_select_type_from_data_type(x->info()->data_type()));
+    build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration));
+
+    // Create kernel
+    std::string kernel_name = "select";
+    if(_has_same_rank)
+    {
+        kernel_name += "_same_rank";
+    }
+    else
+    {
+        const bool is_input_rank_greater_than_two = x->info()->tensor_shape().num_dimensions() > 2;
+        if(is_input_rank_greater_than_two)
+        {
+            const size_t width      = x->info()->tensor_shape().x();
+            const size_t height     = x->info()->tensor_shape().y();
+            const size_t outer_size = x->info()->tensor_shape()[x->info()->tensor_shape().num_dimensions() - 1];
+            const size_t depth_size = x->info()->tensor_shape().total_size() / (width * height * outer_size);
+            build_opts.add_option("-DDEPTH_SIZE=" + support::cpp11::to_string(depth_size));
+        }
+        kernel_name += "_different_rank";
+        kernel_name += is_input_rank_greater_than_two ? "_n" : "_2";
+    }
+    _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel(kernel_name, build_opts.options()));
+
+    // Configure kernel window
+    auto win_config = validate_and_configure_window(c->info(), x->info(), y->info(), output->info());
+    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
+    ICLKernel::configure_internal(win_config.second);
+
+    _config_id = "select_";
+    _config_id += string_from_data_type(x->info()->data_type());
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(x->info()->dimension(0));
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(x->info()->dimension(1));
+    _config_id += "_";
+    _config_id += support::cpp11::to_string(x->info()->dimension(2));
+}
+
+Status CLSelectKernel::validate(const ITensorInfo *c, const ITensorInfo *x, const ITensorInfo *y, const ITensorInfo *output)
+{
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(c, x, y, output));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(c->clone().get(), x->clone().get(), y->clone().get(), output->clone().get()).first);
+    return Status{};
+}
+
+void CLSelectKernel::run(const arm_compute::Window &window, cl::CommandQueue &queue)
+{
+    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
+    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
+
+    Window collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
+    Window slice     = collapsed.first_slice_window_3D();
+
+    if(!_has_same_rank)
+    {
+        Window vector_slice = window.first_slice_window_1D();
+        vector_slice.set(Window::DimX, Window::Dimension(0, 0, 0));
+        unsigned int idx = 0;
+        add_1D_tensor_argument(idx, _c, vector_slice);
+    }
+
+    do
+    {
+        unsigned int idx = _has_same_rank ? 0 : num_arguments_per_1D_tensor();
+        if(_has_same_rank)
+        {
+            add_3D_tensor_argument(idx, _c, slice);
+        }
+        add_3D_tensor_argument(idx, _x, slice);
+        add_3D_tensor_argument(idx, _y, slice);
+        add_3D_tensor_argument(idx, _output, slice);
+
+        enqueue(queue, *this, slice, lws_hint());
+    }
+    while(collapsed.slide_window_slice_3D(slice));
+}
+} // namespace arm_compute
diff --git a/src/runtime/CL/functions/CLSelect.cpp b/src/runtime/CL/functions/CLSelect.cpp
new file mode 100644
index 0000000000..90c368e9b8
--- /dev/null
+++ b/src/runtime/CL/functions/CLSelect.cpp
@@ -0,0 +1,45 @@
+/*
+ * Copyright (c) 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/CL/functions/CLSelect.h"
+
+#include "arm_compute/core/CL/kernels/CLSelectKernel.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/CL/CLScheduler.h"
+
+using namespace arm_compute;
+
+namespace arm_compute
+{
+void CLSelect::configure(const ICLTensor *c, const ICLTensor *x, const ICLTensor *y, ICLTensor *output)
+{
+    auto k = arm_compute::support::cpp14::make_unique<CLSelectKernel>();
+    k->configure(c, x, y, output);
+    _kernel = std::move(k);
+}
+
+Status CLSelect::validate(const ITensorInfo *c, const ITensorInfo *x, const ITensorInfo *y, const ITensorInfo *output)
+{
+    return CLSelectKernel::validate(c, x, y, output);
+}
+} // namespace arm_compute
diff --git a/tests/validation/CL/Select.cpp b/tests/validation/CL/Select.cpp
new file mode 100644
index 0000000000..e3159caa93
--- /dev/null
+++ b/tests/validation/CL/Select.cpp
@@ -0,0 +1,262 @@
+/*
+ * Copyright (c) 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/Types.h"
+#include "arm_compute/runtime/CL/CLTensor.h"
+#include "arm_compute/runtime/CL/CLTensorAllocator.h"
+#include "arm_compute/runtime/CL/functions/CLSelect.h"
+#include "tests/CL/CLAccessor.h"
+#include "tests/PaddingCalculator.h"
+#include "tests/datasets/ShapeDatasets.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Macros.h"
+#include "tests/framework/datasets/Datasets.h"
+#include "tests/validation/Validation.h"
+#include "tests/validation/fixtures/SelectFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace
+{
+auto configuration_dataset = combine(framework::dataset::concat(datasets::SmallShapes(), datasets::LargeShapes()),
+                                     framework::dataset::make("has_same_rank", { false, true }));
+auto run_small_dataset = combine(datasets::SmallShapes(), framework::dataset::make("has_same_rank", { false, true }));
+auto run_large_dataset = combine(datasets::LargeShapes(), framework::dataset::make("has_same_rank", { false, true }));
+
+} // namespace
+TEST_SUITE(CL)
+TEST_SUITE(Select)
+
+// *INDENT-OFF*
+// clang-format off
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip(zip(
+        framework::dataset::make("CInfo", { TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S8), // Invalid condition datatype
+                                            TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8), // Invalid output datatype
+                                            TensorInfo(TensorShape(13U), 1, DataType::U8),          // Invalid c shape
+                                            TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8), // Mismatching shapes
+                                            TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
+                                            TensorInfo(TensorShape(2U), 1, DataType::U8),
+        }),
+        framework::dataset::make("XInfo",{ TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
+                                           TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
+                                           TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
+                                           TensorInfo(TensorShape(32U, 10U, 2U), 1, DataType::F32),
+                                           TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+                                           TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+        })),
+        framework::dataset::make("YInfo",{ TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
+                                           TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
+                                           TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
+                                           TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+                                           TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+                                           TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+        })),
+        framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
+                                                TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S8),
+                                                TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
+                                                TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+                                                TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+                                                TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+        })),
+        framework::dataset::make("Expected", { false, false, false, false, true, true})),
+        c_info, x_info, y_info, output_info, expected)
+{
+    Status s = CLSelect::validate(&c_info.clone()->set_is_resizable(false),
+                                  &x_info.clone()->set_is_resizable(false),
+                                  &y_info.clone()->set_is_resizable(false),
+                                  &output_info.clone()->set_is_resizable(false));
+    ARM_COMPUTE_EXPECT(bool(s) == expected, framework::LogLevel::ERRORS);
+}
+// clang-format on
+// *INDENT-ON*
+
+template <typename T>
+using CLSelectFixture = SelectValidationFixture<CLTensor, CLAccessor, CLSelect, T>;
+
+TEST_SUITE(Float)
+TEST_SUITE(F16)
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, configuration_dataset,
+               shape, same_rank)
+{
+    const DataType dt = DataType::F16;
+
+    // Create tensors
+    CLTensor ref_c = create_tensor<CLTensor>(detail::select_condition_shape(shape, same_rank), DataType::U8);
+    CLTensor ref_x = create_tensor<CLTensor>(shape, dt);
+    CLTensor ref_y = create_tensor<CLTensor>(shape, dt);
+    CLTensor dst   = create_tensor<CLTensor>(shape, dt);
+
+    // Create and Configure function
+    CLSelect select;
+    select.configure(&ref_c, &ref_x, &ref_y, &dst);
+
+    // Validate valid region
+    const ValidRegion valid_region = shape_to_valid_region(shape);
+    validate(dst.info()->valid_region(), valid_region);
+
+    // Validate padding
+    const int         step    = 16 / arm_compute::data_size_from_type(dt);
+    const PaddingSize padding = PaddingCalculator(shape.x(), step).required_padding();
+    if(same_rank)
+    {
+        validate(ref_c.info()->padding(), padding);
+    }
+    validate(ref_x.info()->padding(), padding);
+    validate(ref_y.info()->padding(), padding);
+    validate(dst.info()->padding(), padding);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmall,
+                       CLSelectFixture<half>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(run_small_dataset, framework::dataset::make("DataType", DataType::F16)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge,
+                       CLSelectFixture<half>,
+                       framework::DatasetMode::NIGHTLY,
+                       combine(run_large_dataset, framework::dataset::make("DataType", DataType::F16)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // F16
+
+TEST_SUITE(FP32)
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, configuration_dataset,
+               shape, same_rank)
+{
+    const DataType dt = DataType::F32;
+
+    // Create tensors
+    CLTensor ref_c = create_tensor<CLTensor>(detail::select_condition_shape(shape, same_rank), DataType::U8);
+    CLTensor ref_x = create_tensor<CLTensor>(shape, dt);
+    CLTensor ref_y = create_tensor<CLTensor>(shape, dt);
+    CLTensor dst   = create_tensor<CLTensor>(shape, dt);
+
+    // Create and Configure function
+    CLSelect select;
+    select.configure(&ref_c, &ref_x, &ref_y, &dst);
+
+    // Validate valid region
+    const ValidRegion valid_region = shape_to_valid_region(shape);
+    validate(dst.info()->valid_region(), valid_region);
+
+    // Validate padding
+    const int         step    = 16 / arm_compute::data_size_from_type(dt);
+    const PaddingSize padding = PaddingCalculator(shape.x(), step).required_padding();
+    if(same_rank)
+    {
+        validate(ref_c.info()->padding(), padding);
+    }
+    validate(ref_x.info()->padding(), padding);
+    validate(ref_y.info()->padding(), padding);
+    validate(dst.info()->padding(), padding);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmall,
+                       CLSelectFixture<float>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(run_small_dataset, framework::dataset::make("DataType", DataType::F32)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge,
+                       CLSelectFixture<float>,
+                       framework::DatasetMode::NIGHTLY,
+                       combine(run_large_dataset, framework::dataset::make("DataType", DataType::F32)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // F32
+TEST_SUITE_END() // Float
+
+TEST_SUITE(Quantized)
+TEST_SUITE(QASYMM8)
+DATA_TEST_CASE(Configuration, framework::DatasetMode::ALL, configuration_dataset,
+               shape, same_rank)
+{
+    const DataType dt = DataType::QASYMM8;
+
+    // Create tensors
+    CLTensor ref_c = create_tensor<CLTensor>(detail::select_condition_shape(shape, same_rank), DataType::U8);
+    CLTensor ref_x = create_tensor<CLTensor>(shape, dt);
+    CLTensor ref_y = create_tensor<CLTensor>(shape, dt);
+    CLTensor dst   = create_tensor<CLTensor>(shape, dt);
+
+    // Create and Configure function
+    CLSelect select;
+    select.configure(&ref_c, &ref_x, &ref_y, &dst);
+
+    // Validate valid region
+    const ValidRegion valid_region = shape_to_valid_region(shape);
+    validate(dst.info()->valid_region(), valid_region);
+
+    // Validate padding
+    const int         step    = 16 / arm_compute::data_size_from_type(dt);
+    const PaddingSize padding = PaddingCalculator(shape.x(), step).required_padding();
+    if(same_rank)
+    {
+        validate(ref_c.info()->padding(), padding);
+    }
+    validate(ref_x.info()->padding(), padding);
+    validate(ref_y.info()->padding(), padding);
+    validate(dst.info()->padding(), padding);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmall,
+                       CLSelectFixture<uint8_t>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(run_small_dataset, framework::dataset::make("DataType", DataType::QASYMM8)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLarge,
+                       CLSelectFixture<uint8_t>,
+                       framework::DatasetMode::NIGHTLY,
+                       combine(run_large_dataset, framework::dataset::make("DataType", DataType::QASYMM8)))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference);
+}
+TEST_SUITE_END() // QASYMM8
+TEST_SUITE_END() // Quantized
+
+TEST_SUITE_END() // Select
+TEST_SUITE_END() // CL
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
\ No newline at end of file
diff --git a/tests/validation/fixtures/SelectFixture.h b/tests/validation/fixtures/SelectFixture.h
new file mode 100644
index 0000000000..2ddc97a6cf
--- /dev/null
+++ b/tests/validation/fixtures/SelectFixture.h
@@ -0,0 +1,148 @@
+/*
+ * Copyright (c) 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_TEST_SELECT_FIXTURE
+#define ARM_COMPUTE_TEST_SELECT_FIXTURE
+
+#include "arm_compute/core/TensorShape.h"
+#include "arm_compute/core/Types.h"
+#include "tests/AssetsLibrary.h"
+#include "tests/Globals.h"
+#include "tests/IAccessor.h"
+#include "tests/framework/Asserts.h"
+#include "tests/framework/Fixture.h"
+#include "tests/validation/reference/Select.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace detail
+{
+/** Get the expected shape of the condition tensor
+ *
+ * @param shape               Shape of the other input tensor to select
+ * @param has_same_same_rank  Boolean that flags if the condition has the same rank with the other tensors
+ *
+ * @return the expected condition shape
+ */
+inline TensorShape select_condition_shape(TensorShape shape, bool has_same_same_rank)
+{
+    TensorShape condition_shape = shape;
+    if(!has_same_same_rank)
+    {
+        condition_shape = TensorShape(shape[shape.num_dimensions() - 1]);
+    }
+    return condition_shape;
+}
+} // namespace detail
+
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class SelectValidationFixture : public framework::Fixture
+{
+public:
+    template <typename...>
+    void setup(TensorShape shape, bool has_same_same_rank, DataType data_type)
+    {
+        TensorShape condition_shape = detail::select_condition_shape(shape, has_same_same_rank);
+
+        _target    = compute_target(shape, condition_shape, data_type);
+        _reference = compute_reference(shape, condition_shape, data_type);
+    }
+
+protected:
+    template <typename U>
+    void fill(U &&tensor, int i)
+    {
+        library->fill_tensor_uniform(tensor, i);
+    }
+    template <typename U>
+    void fill_bool(U &&tensor, int i)
+    {
+        ARM_COMPUTE_ERROR_ON(tensor.data_type() != DataType::U8);
+        library->fill_tensor_uniform(tensor, i, static_cast<uint8_t>(0), static_cast<uint8_t>(1));
+    }
+
+    TensorType compute_target(const TensorShape &shape, const TensorShape &condition_shape, DataType data_type)
+    {
+        // Create tensors
+        TensorType c_t   = create_tensor<TensorType>(condition_shape, DataType::U8);
+        TensorType x_t   = create_tensor<TensorType>(shape, data_type);
+        TensorType y_t   = create_tensor<TensorType>(shape, data_type);
+        TensorType dst_t = create_tensor<TensorType>(shape, data_type);
+
+        // Create and configure function
+        FunctionType select;
+        select.configure(&c_t, &x_t, &y_t, &dst_t);
+
+        ARM_COMPUTE_EXPECT(c_t.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(x_t.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(y_t.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(dst_t.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+        // Allocate tensors
+        c_t.allocator()->allocate();
+        x_t.allocator()->allocate();
+        y_t.allocator()->allocate();
+        dst_t.allocator()->allocate();
+
+        ARM_COMPUTE_EXPECT(!c_t.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(!x_t.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(!y_t.info()->is_resizable(), framework::LogLevel::ERRORS);
+        ARM_COMPUTE_EXPECT(!dst_t.info()->is_resizable(), framework::LogLevel::ERRORS);
+
+        // Fill tensors
+        fill_bool(AccessorType(c_t), 0);
+        fill(AccessorType(x_t), 1);
+        fill(AccessorType(y_t), 2);
+
+        // Compute function
+        select.run();
+
+        return dst_t;
+    }
+
+    SimpleTensor<T> compute_reference(const TensorShape &shape, const TensorShape &condition_shape, DataType data_type)
+    {
+        // Create reference
+        SimpleTensor<uint8_t> ref_c{ condition_shape, DataType::U8 };
+        SimpleTensor<T>       ref_x{ shape, data_type };
+        SimpleTensor<T>       ref_y{ shape, data_type };
+
+        // Fill reference
+        fill_bool(ref_c, 0);
+        fill(ref_x, 1);
+        fill(ref_y, 2);
+
+        return reference::select<T>(ref_c, ref_x, ref_y);
+    }
+
+    TensorType      _target{};
+    SimpleTensor<T> _reference{};
+};
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_TEST_SELECT_FIXTURE */
diff --git a/tests/validation/reference/Select.cpp b/tests/validation/reference/Select.cpp
new file mode 100644
index 0000000000..91b6b66c42
--- /dev/null
+++ b/tests/validation/reference/Select.cpp
@@ -0,0 +1,84 @@
+/*
+ * Copyright (c) 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 "Select.h"
+
+#include "arm_compute/core/Types.h"
+#include "tests/validation/Helpers.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> select(const SimpleTensor<uint8_t> &c, const SimpleTensor<T> &x, const SimpleTensor<T> &y)
+{
+    // Check if condition has the same rank as c
+    const bool is_same_rank = (c.shape().num_dimensions() == x.shape().num_dimensions());
+
+    // Check shapes
+    ARM_COMPUTE_ERROR_ON(x.shape() != y.shape());
+    ARM_COMPUTE_ERROR_ON(is_same_rank && (x.shape() != c.shape()));
+    ARM_COMPUTE_ERROR_ON(!is_same_rank && (c.shape().num_dimensions() > 1) && (c.shape().x() != x.shape()[x.shape().num_dimensions() - 1]));
+
+    // Create reference
+    SimpleTensor<T> dst{ x.shape(), x.data_type(), 1 };
+
+    // Run select core
+    if(is_same_rank)
+    {
+        for(int i = 0; i < x.num_elements(); ++i)
+        {
+            dst[i] = c[i] > 0 ? x[i] : y[i];
+        }
+    }
+    else
+    {
+        T *output_ptr = dst.data();
+
+        const int outer_size = c.num_elements();
+        const int inner_size = x.num_elements() / outer_size;
+        size_t    offset     = 0;
+
+        for(int i = 0; i < outer_size; ++i)
+        {
+            const T *input_ptr = c[i] > 0 ? x.data() : y.data();
+            memcpy(output_ptr + offset, input_ptr + offset, inner_size * sizeof(T));
+            offset += inner_size;
+        }
+    }
+
+    return dst;
+}
+
+template SimpleTensor<uint8_t> select(const SimpleTensor<uint8_t> &c, const SimpleTensor<uint8_t> &x, const SimpleTensor<uint8_t> &y);
+template SimpleTensor<half> select(const SimpleTensor<uint8_t> &c, const SimpleTensor<half> &x, const SimpleTensor<half> &y);
+template SimpleTensor<float> select(const SimpleTensor<uint8_t> &c, const SimpleTensor<float> &x, const SimpleTensor<float> &y);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/reference/Select.h b/tests/validation/reference/Select.h
new file mode 100644
index 0000000000..394f36748d
--- /dev/null
+++ b/tests/validation/reference/Select.h
@@ -0,0 +1,43 @@
+/*
+ * Copyright (c) 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_TEST_SELECT_H__
+#define __ARM_COMPUTE_TEST_SELECT_H__
+
+#include "tests/SimpleTensor.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+namespace reference
+{
+template <typename T>
+SimpleTensor<T> select(const SimpleTensor<uint8_t> &c, const SimpleTensor<T> &x, const SimpleTensor<T> &y);
+} // namespace reference
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_TEST_SELECT_H__ */