COMPMID-345 - In-place computation for Activation Layer

Change-Id: I25ebfccc3d3e758cc8164e0b33805c0bb303891a
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/78226
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Pablo Tello <pablo.tello@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

4 files changed, 53 insertions, 20 deletions

diff --git a/arm_compute/core/CL/kernels/CLActivationLayerKernel.h b/arm_compute/core/CL/kernels/CLActivationLayerKernel.h
index 490e70544b..df22574de8 100644
--- a/arm_compute/core/CL/kernels/CLActivationLayerKernel.h
+++ b/arm_compute/core/CL/kernels/CLActivationLayerKernel.h
@@ -24,23 +24,45 @@
 #ifndef __ARM_COMPUTE_CLACTIVATIONLAYERKERNEL_H__
 #define __ARM_COMPUTE_CLACTIVATIONLAYERKERNEL_H__
 
-#include "arm_compute/core/CL/ICLSimple3DKernel.h"
+#include "arm_compute/core/CL/ICLKernel.h"
 
 namespace arm_compute
 {
 class ICLTensor;
 
 /** Interface for the activation layer kernel. */
-class CLActivationLayerKernel : public ICLSimple3DKernel
+class CLActivationLayerKernel : public ICLKernel
 {
 public:
+    /** Default constructor */
+    CLActivationLayerKernel();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLActivationLayerKernel(const CLActivationLayerKernel &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLActivationLayerKernel &operator=(const CLActivationLayerKernel &) = delete;
+    /** Allow instances of this class to be moved */
+    CLActivationLayerKernel(CLActivationLayerKernel &&) = default;
+    /** Allow instances of this class to be moved */
+    CLActivationLayerKernel &operator=(CLActivationLayerKernel &&) = default;
+    /** Default destructor */
+    ~CLActivationLayerKernel() = default;
     /** Set the input and output tensor.
      *
-     * @param[in]  input    Source tensor. Data types supported: F16, F32, U16, S16.
-     * @param[out] output   Destination tensor. Data type should match the input data type.
-     * @param[in]  act_info Activation layer information.
+     * @note If the output tensor is a nullptr, the activation function will be performed in-place
+     *
+     * @param[in, out] input    Source tensor. In case of @p output tensor = nullptr, this tensor will store the result
+     *                          of the activation function. Data types supported: F16/F32.
+     * @param[out]     output   Destination tensor. Data type should match the input data type.
+     * @param[in]      act_info Activation layer information.
      */
-    void configure(const ICLTensor *input, ICLTensor *output, ActivationLayerInfo act_info);
+    void configure(ICLTensor *input, ICLTensor *output, ActivationLayerInfo act_info);
+
+    // Inherited methods overridden:
+    void run(const Window &window, cl::CommandQueue &queue) override;
+
+private:
+    ICLTensor *_input;
+    ICLTensor *_output;
 };
 }
 #endif /*__ARM_COMPUTE_CLACTIVATIONLAYERKERNEL_H__ */
diff --git a/arm_compute/core/NEON/kernels/NEActivationLayerKernel.h b/arm_compute/core/NEON/kernels/NEActivationLayerKernel.h
index 97f92d6a1e..539bca587a 100644
--- a/arm_compute/core/NEON/kernels/NEActivationLayerKernel.h
+++ b/arm_compute/core/NEON/kernels/NEActivationLayerKernel.h
@@ -25,14 +25,14 @@
 #define __ARM_COMPUTE_NEACTIVATIONLAYERKERNEL_H__
 
 #include "arm_compute/core/FixedPoint.h"
-#include "arm_compute/core/NEON/INESimpleKernel.h"
+#include "arm_compute/core/NEON/INEKernel.h"
 
 namespace arm_compute
 {
 class ITensor;
 
 /** Interface for the activation layer kernel. */
-class NEActivationLayerKernel : public INESimpleKernel
+class NEActivationLayerKernel : public INEKernel
 {
 public:
     /** Constructor */
@@ -47,11 +47,14 @@ public:
     NEActivationLayerKernel &operator=(NEActivationLayerKernel &&) = default;
     /** Set the input and output tensor.
      *
-     * @param[in]  input           Source tensor. Data types supported: QS8/F32.
-     * @param[out] output          Destination tensor. Data type supported: same as @p input
-     * @param[in]  activation_info Activation layer information.
+     * @note If the output tensor is a nullptr, the activation function will be performed in-place
+     *
+     * @param[in, out] input           Source tensor. In case of @p output tensor = nullptr, this tensor will store the result
+     *                                 of the activation function. Data types supported: QS8/F32.
+     * @param[out]     output          Destination tensor. Data type supported: same as @p input
+     * @param[in]      activation_info Activation layer information.
      */
-    void configure(const ITensor *input, ITensor *output, ActivationLayerInfo activation_info);
+    void configure(ITensor *input, ITensor *output, ActivationLayerInfo activation_info);
 
     // Inherited methods overridden:
     void run(const Window &window) override;
@@ -77,6 +80,8 @@ private:
     typename std::enable_if<std::is_same<T, qint8_t>::value, void>::type activation(const Window &window);
 
 private:
+    ITensor                      *_input;
+    ITensor                      *_output;
     ActivationFunctionExecutorPtr _func;
     ActivationLayerInfo           _act_info;
 };
diff --git a/arm_compute/runtime/CL/functions/CLActivationLayer.h b/arm_compute/runtime/CL/functions/CLActivationLayer.h
index 6468c996a2..3028afb25b 100644
--- a/arm_compute/runtime/CL/functions/CLActivationLayer.h
+++ b/arm_compute/runtime/CL/functions/CLActivationLayer.h
@@ -41,11 +41,14 @@ class CLActivationLayer : public ICLSimpleFunction
 public:
     /** Set the input and output tensor.
      *
-     * @param[in]  input    Source tensor. Data types supported: F16, F32, U16, S16.
-     * @param[out] output   Destination tensor. Data type should match the input data type.
-     * @param[in]  act_info Activation layer parameters.
+     * @note If the output tensor is a nullptr, the activation function will be performed in-place
+     *
+     * @param[in, out] input    Source tensor. In case of @p output tensor = nullptr, this tensor will store the result
+     *                          of the activation function. Data types supported: F16/F32.
+     * @param[out]     output   Destination tensor. Data type should match the input data type.
+     * @param[in]      act_info Activation layer parameters.
      */
-    void configure(const ICLTensor *input, ICLTensor *output, ActivationLayerInfo act_info);
+    void configure(ICLTensor *input, ICLTensor *output, ActivationLayerInfo act_info);
 };
 }
 #endif /* __ARM_COMPUTE_CLACTIVATIONLAYER_H__ */
diff --git a/arm_compute/runtime/NEON/functions/NEActivationLayer.h b/arm_compute/runtime/NEON/functions/NEActivationLayer.h
index 35366e16fb..b1a211553d 100644
--- a/arm_compute/runtime/NEON/functions/NEActivationLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEActivationLayer.h
@@ -41,11 +41,14 @@ class NEActivationLayer : public INESimpleFunction
 public:
     /** Set the input and output tensor.
      *
-     * @param[in]  input           Source tensor. Data type supported: QS8/F32.
-     * @param[out] output          Destination tensor. Data type supported: same as @p input
-     * @param[in]  activation_info Activation layer parameters.
+     * @note If the output tensor is a nullptr, the activation function will be performed in-place
+     *
+     * @param[in, out] input           Source tensor. In case of @p output tensor = nullptr, this tensor will store the result
+     *                                 of the activation function. Data types supported: QS8/F32.
+     * @param[out]     output          Destination tensor. Data type supported: same as @p input
+     * @param[in]      activation_info Activation layer parameters.
      */
-    void configure(const ITensor *input, ITensor *output, ActivationLayerInfo activation_info);
+    void configure(ITensor *input, ITensor *output, ActivationLayerInfo activation_info);
 };
 }
 #endif /* __ARM_COMPUTE_NEACTIVATIONLAYER_H__ */