COMPMID-1937: Adds support for DequantizationLayer for NEON/CL.

Change-Id: I4b73edd176a277294e0e42e642460bc61210778a
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/744
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Giuseppe Rossini <giuseppe.rossini@arm.com>

4 files changed, 41 insertions, 100 deletions

diff --git a/arm_compute/core/CL/kernels/CLDequantizationLayerKernel.h b/arm_compute/core/CL/kernels/CLDequantizationLayerKernel.h
index 25fd3378cb..3dfb19b306 100644
--- a/arm_compute/core/CL/kernels/CLDequantizationLayerKernel.h
+++ b/arm_compute/core/CL/kernels/CLDequantizationLayerKernel.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2018 ARM Limited.
+ * Copyright (c) 2017-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -30,11 +30,7 @@ namespace arm_compute
 {
 class ICLTensor;
 
-/** Interface for the dequantization layer kernel.
- *
- * @note The implementation supports only 3D input tensors.
- *
- */
+/** Interface for the dequantization layer kernel. */
 class CLDequantizationLayerKernel : public ICLKernel
 {
 public:
@@ -52,22 +48,18 @@ public:
     ~CLDequantizationLayerKernel() = default;
     /** Set the input, output, min and max.
      *
-     * @param[in]  input   Source tensor. Data types supported: U8.
-     * @param[out] output  Destination tensor. Data types supported: F32.
-     * @param[in]  min_max Pointer to the tensor with shape [2, batches] which stores the minimum and maximum value for each 3D input tensor.
-     *                     The dimensions over the second must match the batched dimensions of the input tensor. Data type supported: F32.
+     * @param[in]  input  Source tensor. Data types supported: QASYMM8.
+     * @param[out] output Destination tensor. Data types supported: F16/F32.
      */
-    void configure(const ICLTensor *input, ICLTensor *output, const ICLTensor *min_max);
+    void configure(const ICLTensor *input, ICLTensor *output);
     /** Static function to check if given info will lead to a valid configuration of @ref CLDequantizationLayerKernel
      *
-     * @param[in] input   Input tensor info. Data types supported: U8.
-     * @param[in] output  Output tensor info. Data types supported: F32.
-     * @param[in] min_max Info for the tensor with shape [2, batches] which stores the minimum and maximum value for each 3D input tensor.
-     *                    The dimensions over the second must match the batched dimensions of the input tensor. Data type supported: F32.
+     * @param[in] input  Input tensor info. Data types supported: QASYMM8.
+     * @param[in] output Output tensor info. Data types supported: F16/F32.
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *min_max);
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output);
 
     // Inherited methods overridden:
     void run(const Window &window, cl::CommandQueue &queue) override;
@@ -75,7 +67,6 @@ public:
 private:
     const ICLTensor *_input;
     ICLTensor       *_output;
-    const ICLTensor *_min_max;
 };
 } // namespace arm_compute
 #endif /*__ARM_COMPUTE_CLDEQUANTIZATIONLAYERKERNEL_H__ */
diff --git a/arm_compute/core/NEON/kernels/NEDequantizationLayerKernel.h b/arm_compute/core/NEON/kernels/NEDequantizationLayerKernel.h
index f48e76f340..7d215f5f7b 100644
--- a/arm_compute/core/NEON/kernels/NEDequantizationLayerKernel.h
+++ b/arm_compute/core/NEON/kernels/NEDequantizationLayerKernel.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2018 ARM Limited.
+ * Copyright (c) 2017-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -30,11 +30,7 @@ namespace arm_compute
 {
 class ITensor;
 
-/** Interface for the dequantization layer kernel.
- *
- * @note The implementation supports only 3D input tensors
- *
- */
+/** Interface for the dequantization layer kernel. */
 class NEDequantizationLayerKernel : public INEKernel
 {
 public:
@@ -54,24 +50,20 @@ public:
     NEDequantizationLayerKernel &operator=(NEDequantizationLayerKernel &&) = default;
     /** Default destructor */
     ~NEDequantizationLayerKernel() = default;
-    /** Set input, output, min and max.
+    /** Set input, output tensors.
      *
-     * @param[in]  input   Source tensor with at least 3 dimensions. The dimensions over the third will be interpreted as batches. Data type supported: U8.
-     * @param[out] output  Destination tensor with the same dimensions of input. Data type supported: F32.
-     * @param[in]  min_max Pointer to the tensor with shape [2, batches] which stores the minimum and maximum value for each 3D input tensor.
-     *                     The dimensions over the second must match the batched dimensions of the input tensor. Data type supported: F32
+     * @param[in]  input  Source tensor. Data type supported: QASYMM8.
+     * @param[out] output Destination tensor with the same dimensions of input. Data type supported: F16/F32.
      */
-    void configure(const ITensor *input, ITensor *output, const ITensor *min_max);
+    void configure(const ITensor *input, ITensor *output);
     /** Static function to check if given info will lead to a valid configuration of @ref NEDequantizationLayerKernel
      *
-     * @param[in] input   Input tensor info. Data types supported: U8.
-     * @param[in] output  Output tensor info. Data types supported: F32.
-     * @param[in] min_max Info for the tensor with shape [2, batches] which stores the minimum and maximum value for each 3D input tensor.
-     *                    The dimensions over the second must match the batched dimensions of the input tensor. Data type supported: F32.
+     * @param[in] input  Input tensor info. Data types supported: QASYMM8.
+     * @param[in] output Output tensor info. Data types supported: F16/F32.
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *min_max);
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output);
 
     // Inherited methods overridden:
     void run(const Window &window, const ThreadInfo &info) override;
@@ -79,7 +71,6 @@ public:
 private:
     const ITensor *_input;
     ITensor       *_output;
-    const ITensor *_min_max;
 };
 } // namespace arm_compute
 #endif /*__ARM_COMPUTE_NEDEQUANTIZATIONLAYERKERNEL_H__ */
diff --git a/arm_compute/runtime/CL/functions/CLDequantizationLayer.h b/arm_compute/runtime/CL/functions/CLDequantizationLayer.h
index efd28fc819..cf7c5761e4 100644
--- a/arm_compute/runtime/CL/functions/CLDequantizationLayer.h
+++ b/arm_compute/runtime/CL/functions/CLDequantizationLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2018 ARM Limited.
+ * Copyright (c) 2017-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -24,55 +24,33 @@
 #ifndef __ARM_COMPUTE_CLDEQUANTIZATIONLAYER_H__
 #define __ARM_COMPUTE_CLDEQUANTIZATIONLAYER_H__
 
-#include "arm_compute/runtime/IFunction.h"
-
-#include "arm_compute/core/CL/kernels/CLDequantizationLayerKernel.h"
-#include "arm_compute/runtime/Tensor.h"
+#include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
 #include "arm_compute/core/Types.h"
 
 namespace arm_compute
 {
+// Forward declarations
 class ICLTensor;
 
-/** Basic function to simulate a dequantization layer. This function calls the following CL kernels:
- *
- * -# @ref CLDequantizationLayerKernel
- *
- */
-class CLDequantizationLayer : public IFunction
+/** Basic function to run @ref CLDequantizationLayerKernel that dequantizes an input tensor */
+class CLDequantizationLayer : public ICLSimpleFunction
 {
 public:
-    /** Default constructor */
-    CLDequantizationLayer();
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    CLDequantizationLayer(const CLDequantizationLayer &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    CLDequantizationLayer &operator=(const CLDequantizationLayer &) = delete;
     /** Set the input and output tensors.
      *
-     * @param[in]  input   Source tensor with at least 3 dimensions. The dimensions over the third will be interpreted as batches. Data types supported: U8.
-     * @param[out] output  Destination tensor with the same dimensions of input. Data type supported: F32.
-     * @param[in]  min_max Pointer to the tensor with shape [2, batches] which stores the minimum and maximum value for each 3D input tensor.
-     *                     The dimensions over the second must match the batched dimensions of the input tensor. Data type supported: F32.
+     * @param[in]  input  Source tensor with at least 3 dimensions. The dimensions over the third will be interpreted as batches. Data types supported: QASYMM8.
+     * @param[out] output Destination tensor with the same dimensions of input. Data type supported: F16/F32.
      */
-    void configure(const ICLTensor *input, ICLTensor *output, const ICLTensor *min_max);
+    void configure(const ICLTensor *input, ICLTensor *output);
     /** Static function to check if given info will lead to a valid configuration of @ref CLDequantizationLayer
      *
-     * @param[in] input   Input tensor info. Data types supported: U8.
-     * @param[in] output  Output tensor info. Data type supported: F32.
-     * @param[in] min_max Info for the tensor with shape [2, batches] which stores the minimum and maximum value for each 3D input tensor.
-     *                    The dimensions over the second must match the batched dimensions of the input tensor. Data type supported: F32.
+     * @param[in] input  Input tensor info. Data types supported: QASYMM8.
+     * @param[in] output Output tensor info. Data type supported: F16/F32.
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *min_max);
-
-    // Inherited methods overridden:
-    void run() override;
-
-private:
-    CLDequantizationLayerKernel _dequantize_kernel;
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output);
 };
-}
+} // namespace arm_compute
 #endif /* __ARM_COMPUTE_CLDEQUANTIZATIONLAYER_H__ */
diff --git a/arm_compute/runtime/NEON/functions/NEDequantizationLayer.h b/arm_compute/runtime/NEON/functions/NEDequantizationLayer.h
index 90c454ef3e..b7c5bac844 100644
--- a/arm_compute/runtime/NEON/functions/NEDequantizationLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEDequantizationLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2018 ARM Limited.
+ * Copyright (c) 2017-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -24,52 +24,33 @@
 #ifndef __ARM_COMPUTE_NEDEQUANTIZATIONLAYER_H__
 #define __ARM_COMPUTE_NEDEQUANTIZATIONLAYER_H__
 
-#include "arm_compute/runtime/IFunction.h"
-
-#include "arm_compute/core/NEON/kernels/NEDequantizationLayerKernel.h"
+#include "arm_compute/runtime/NEON/INESimpleFunctionNoBorder.h"
 
 #include "arm_compute/core/Types.h"
 
 namespace arm_compute
 {
+// Forward declarations
 class ITensor;
 
-/** Basic function to simulate a dequantization layer. This function calls the following NEON kernels:
- *
- * @note The implementation supports only 3D input tensors
- *
- * -# @ref NEDequantizationLayerKernel
- *
- */
-class NEDequantizationLayer : public IFunction
+/** Basic function to run @ref NEDequantizationLayerKernel that dequantizes an input tensor */
+class NEDequantizationLayer : public INESimpleFunctionNoBorder
 {
 public:
-    /** Default constructor */
-    NEDequantizationLayer();
     /** Configure the kernel.
      *
-     * @param[in]  input   Source tensor with at least 3 dimensions. The dimensions over the third will be interpreted as batches. Data types supported: U8.
-     * @param[out] output  Destination tensor with the same dimensions of input. Data type supported: F32.
-     * @param[in]  min_max Pointer to the tensor with shape [2, batches] which stores the minimum and maximum value for each 3D input tensor.
-     *                     The dimensions over the second must match the batched dimensions of the input tensor. Data type supported: F32
+     * @param[in]  input  Source tensor. Data types supported: QASYMM8.
+     * @param[out] output Destination tensor with the same dimensions of input. Data type supported: F16/F32.
      */
-    void configure(const ITensor *input, ITensor *output, const ITensor *min_max);
+    void configure(const ITensor *input, ITensor *output);
     /** Static function to check if given info will lead to a valid configuration of @ref NEDequantizationLayer
      *
-     * @param[in] input   Input tensor info. Data types supported: U8.
-     * @param[in] output  Output tensor info. Data type supported: F32.
-     * @param[in] min_max Info for the tensor with shape [2, batches] which stores the minimum and maximum value for each 3D input tensor.
-     *                    The dimensions over the second must match the batched dimensions of the input tensor. Data type supported: F32.
+     * @param[in] input  Input tensor info. Data types supported: QASYMM8.
+     * @param[in] output Output tensor info. Data type supported: F16/F32.
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const ITensorInfo *min_max);
-
-    // Inherited methods overridden:
-    void run() override;
-
-private:
-    NEDequantizationLayerKernel _dequantize_kernel;
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output);
 };
-}
+} // namespace arm_compute
 #endif /* __ARM_COMPUTE_NEDEQUANTIZATIONLAYER_H__ */