COMPMID-3097 Fuse activation with fully connected layer CL

Change-Id: I447030e69b9e565f2f81529a41af8c5e7ece7ecf
Signed-off-by: Giorgio Arena <giorgio.arena@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2702
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

5 files changed, 122 insertions, 109 deletions

diff --git a/arm_compute/core/PixelValue.h b/arm_compute/core/PixelValue.h
index c5f6608163..31bc55098a 100644
--- a/arm_compute/core/PixelValue.h
+++ b/arm_compute/core/PixelValue.h
@@ -41,11 +41,11 @@ public:
     }
     /** Initialize the union with a pixel value of chosen datatype
      *
-     * @param[in] v        int value.
+     * @param[in] v        value.
      * @param[in] datatype DataType that @p v have to be stored
      * @param[in] qinfo    (Optional) QuantizationInfo to apply in case of quantized data types to @p v
      */
-    PixelValue(int64_t v, DataType datatype, QuantizationInfo qinfo = QuantizationInfo())
+    PixelValue(double v, DataType datatype, QuantizationInfo qinfo = QuantizationInfo())
         : PixelValue()
     {
         switch(datatype)
@@ -57,13 +57,13 @@ public:
                 value.s8 = static_cast<int8_t>(v);
                 break;
             case DataType::QASYMM8:
-                value.u8 = quantize_qasymm8(static_cast<uint8_t>(v), qinfo);
+                value.u8 = quantize_qasymm8(static_cast<float>(v), qinfo);
                 break;
             case DataType::QASYMM8_SIGNED:
-                value.s8 = quantize_qasymm8_signed(static_cast<int8_t>(v), qinfo);
+                value.s8 = quantize_qasymm8_signed(static_cast<float>(v), qinfo);
                 break;
             case DataType::QSYMM8:
-                value.s8 = quantize_qsymm8(static_cast<int8_t>(v), qinfo);
+                value.s8 = quantize_qsymm8(static_cast<float>(v), qinfo);
                 break;
             case DataType::U16:
                 value.u16 = static_cast<uint16_t>(v);
@@ -72,10 +72,10 @@ public:
                 value.s16 = static_cast<int16_t>(v);
                 break;
             case DataType::QASYMM16:
-                value.u16 = quantize_qasymm16(static_cast<uint16_t>(v), qinfo);
+                value.u16 = quantize_qasymm16(static_cast<float>(v), qinfo);
                 break;
             case DataType::QSYMM16:
-                value.s16 = quantize_qsymm16(static_cast<int16_t>(v), qinfo);
+                value.s16 = quantize_qsymm16(static_cast<float>(v), qinfo);
                 break;
             case DataType::U32:
                 value.u32 = static_cast<uint32_t>(v);
@@ -96,10 +96,8 @@ public:
                 value.f32 = static_cast<float>(v);
                 break;
             case DataType::F64:
-                value.f64 = static_cast<double>(v);
-                break;
             default:
-                value.s64 = v;
+                value.f64 = v;
                 break;
         }
     }
diff --git a/arm_compute/core/Types.h b/arm_compute/core/Types.h
index 2030b171c6..cf689d757c 100644
--- a/arm_compute/core/Types.h
+++ b/arm_compute/core/Types.h
@@ -799,39 +799,6 @@ private:
     DimensionRoundingType _round_type;
 };
 
-/** Fully connected layer info */
-struct FullyConnectedLayerInfo
-{
-    DataLayout weights_trained_layout{ DataLayout::NCHW }; /**<  Layout that the weights have been trained with. */
-    bool       transpose_weights{ true };                  /**<  Transpose weights if true. */
-    bool       are_weights_reshaped{ false };              /**<  Reshape the weights tensor if false. */
-    bool       retain_internal_weights{ false };           /**<  Retain internal reshaped weights. */
-    bool       fp_mixed_precision{ false };                /**<  Use wider accumulators (32 bit instead of 16 for FP16) to improve accuracy. */
-
-    /** Sets the weights trained data layout
-     *
-     * @param[in] layout Data layout that the weights were trained with
-     *
-     * @return Updated object
-     */
-    FullyConnectedLayerInfo &set_weights_trained_layout(DataLayout layout)
-    {
-        weights_trained_layout = layout;
-        return *this;
-    }
-    /** Sets the transpose weights flag
-     *
-     * @param[in] should_transpose_weights Boolean flag indicating if weights should be transposed
-     *
-     * @return Updated object
-     */
-    FullyConnectedLayerInfo &set_transpose_weights(bool should_transpose_weights)
-    {
-        transpose_weights = should_transpose_weights;
-        return *this;
-    }
-};
-
 /** PriorBox layer info */
 class PriorBoxLayerInfo final
 {
@@ -1674,6 +1641,40 @@ private:
     bool               _enabled = { false };
 };
 
+/** Fully connected layer info */
+struct FullyConnectedLayerInfo
+{
+    DataLayout          weights_trained_layout{ DataLayout::NCHW }; /**<  Layout that the weights have been trained with. */
+    bool                transpose_weights{ true };                  /**<  Transpose weights if true. */
+    bool                are_weights_reshaped{ false };              /**<  Reshape the weights tensor if false. */
+    bool                retain_internal_weights{ false };           /**<  Retain internal reshaped weights. */
+    bool                fp_mixed_precision{ false };                /**<  Use wider accumulators (32 bit instead of 16 for FP16) to improve accuracy. */
+    ActivationLayerInfo activation_info{};                          /**<  Fused activation to apply after the matrix multiplication. */
+
+    /** Sets the weights trained data layout
+     *
+     * @param[in] layout Data layout that the weights were trained with
+     *
+     * @return Updated object
+     */
+    FullyConnectedLayerInfo &set_weights_trained_layout(DataLayout layout)
+    {
+        weights_trained_layout = layout;
+        return *this;
+    }
+    /** Sets the transpose weights flag
+     *
+     * @param[in] should_transpose_weights Boolean flag indicating if weights should be transposed
+     *
+     * @return Updated object
+     */
+    FullyConnectedLayerInfo &set_transpose_weights(bool should_transpose_weights)
+    {
+        transpose_weights = should_transpose_weights;
+        return *this;
+    }
+};
+
 /** Normalization Layer Information class */
 class NormalizationLayerInfo
 {
@@ -1944,16 +1945,16 @@ enum class GEMMLowpOutputStageType
 /** GEMMLowp output stage info */
 struct GEMMLowpOutputStageInfo
 {
-    GEMMLowpOutputStageType type{ GEMMLowpOutputStageType::NONE }; /**< GEMMLowp output stage type */
-    int32_t                 gemmlowp_offset{ 0 };                  /**< GEMMLowp output stage offset used for quantizing to QASYMM8 */
-    int32_t                 gemmlowp_multiplier{ 0 };              /**< GEMMLowp output stage multiplier used for quantizing to QASYMM8 */
-    int32_t                 gemmlowp_shift{ 0 };                   /**< GEMMLowp output stage shift used for quantizing to uint8 */
-    int32_t                 gemmlowp_min_bound{ 0 };               /**< GEMMLowp min value used to saturate down the output result before converting back to QASYMM8 */
-    int32_t                 gemmlowp_max_bound{ 0 };               /**< GEMMLowp max value used to saturate down the output result before converting back to QASYMM8 */
-    std::vector<int32_t>    gemmlowp_multipliers{};                /**< GEMMLowp output stage multiplier used for quantizing to QASYMM8 */
-    std::vector<int32_t>    gemmlowp_shifts{};                     /**< GEMMLowp output stage multiplier used for quantizing to QASYMM8 */
-    bool                    is_quantized_per_channel{ false };     /**< GEMMLowp quantized per-channel flag */
-    DataType                output_data_type{ DataType::UNKNOWN }; /**< Output tensor data type to use if the output is not initialized */
+    GEMMLowpOutputStageType type{ GEMMLowpOutputStageType::NONE };                        /**< GEMMLowp output stage type */
+    int32_t                 gemmlowp_offset{ 0 };                                         /**< GEMMLowp output stage offset used for quantizing to QASYMM8 */
+    int32_t                 gemmlowp_multiplier{ 0 };                                     /**< GEMMLowp output stage multiplier used for quantizing to QASYMM8 */
+    int32_t                 gemmlowp_shift{ 0 };                                          /**< GEMMLowp output stage shift used for quantizing to uint8 */
+    int32_t                 gemmlowp_min_bound{ std::numeric_limits<int32_t>::lowest() }; /**< GEMMLowp min value used to saturate down the output result before converting back to QASYMM8 */
+    int32_t                 gemmlowp_max_bound{ std::numeric_limits<int32_t>::max() };    /**< GEMMLowp max value used to saturate down the output result before converting back to QASYMM8 */
+    std::vector<int32_t>    gemmlowp_multipliers{};                                       /**< GEMMLowp output stage multiplier used for quantizing to QASYMM8 */
+    std::vector<int32_t>    gemmlowp_shifts{};                                            /**< GEMMLowp output stage multiplier used for quantizing to QASYMM8 */
+    bool                    is_quantized_per_channel{ false };                            /**< GEMMLowp quantized per-channel flag */
+    DataType                output_data_type{ DataType::UNKNOWN };                        /**< Output tensor data type to use if the output is not initialized */
 };
 
 /** GEMM LHS (Left Hand Side) matrix information */
diff --git a/arm_compute/graph/nodes/FullyConnectedLayerNode.h b/arm_compute/graph/nodes/FullyConnectedLayerNode.h
index 41a7bc9946..10c310dda2 100644
--- a/arm_compute/graph/nodes/FullyConnectedLayerNode.h
+++ b/arm_compute/graph/nodes/FullyConnectedLayerNode.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2019 ARM Limited.
+ * Copyright (c) 2018-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -43,6 +43,11 @@ public:
     FullyConnectedLayerNode(unsigned int            num_outputs,
                             QuantizationInfo        out_quant_info = QuantizationInfo(),
                             FullyConnectedLayerInfo fc_info        = FullyConnectedLayerInfo());
+    /** Sets fused activation
+     *
+     * @param[in] fused_activation Fused activation to set
+     */
+    void set_fused_activation(ActivationLayerInfo fused_activation);
     /** Computes weights descriptor
      *
      * @warning Works for inputs with 1D batch space
@@ -83,6 +88,8 @@ public:
     TensorDescriptor configure_output(size_t idx) const override;
     void accept(INodeVisitor &v) override;
 
+    static constexpr NodeType node_type = NodeType::FullyConnectedLayer;
+
 private:
     unsigned int            _num_outputs;
     QuantizationInfo        _out_quant_info;
diff --git a/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h b/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h
index f453879fd8..564135eed8 100644
--- a/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h
+++ b/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2019 ARM Limited.
+ * Copyright (c) 2017-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -68,24 +68,25 @@ public:
      * @param[in]  result_offset   Offset to be added to each element of the input matrix
      * @param[in]  result_mult_int Value to be multiplied to each element of the input matrix when once the result_offset has been add
      * @param[in]  result_shift    Number of bits to shift right the result before converting back to QASYMM8
-     * @param[in]  min             (Optional) Min value used to saturate down the output result before converting back to QASYMM8
+     * @param[in]  min             (Optional) Min value used to saturate down the output result before converting back to QASYMM8. Defaults to the minimum possible 32-bit signed integer.
      * @param[in]  max             (Optional) Max value used to saturate up the output result before converting back to QASYMM8,
-     *                             Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+     *                             Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      */
-    void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, int result_offset, int result_mult_int, int result_shift, int min = 0, int max = 0);
+    void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, int result_offset, int result_mult_int, int result_shift, int min = std::numeric_limits<int32_t>::lowest(),
+                   int max = std::numeric_limits<int32_t>::max());
     /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpQuantizeDownInt32ToUint8Scale
      *
      * @param[in] input  Input tensor. It is the output of @ref CLGEMMLowpMatrixMultiplyCore function. Data type supported: S32
      * @param[in] bias   Biases tensor. Only shared biases supported and it can be a nullptr if the addition of biases is not required.
      *                   Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
      * @param[in] output Output tensor. Data type supported: Data type supported: QASYMM8
-     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QASYMM8
+     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QASYMM8. Defaults to the minimum possible 32-bit signed integer.
      * @param[in] max    (Optional) Max value used to saturate up the output result before converting back to QASYMM8,
-     *                   Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+     *                   Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0);
+    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = std::numeric_limits<int32_t>::lowest(), int max = std::numeric_limits<int32_t>::max());
 };
 
 /** Basic function to execute CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint on OpenCL.
@@ -128,25 +129,25 @@ public:
      * @param[in]  result_fixedpoint_multiplier Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add
      * @param[in]  result_shift                 Number of bits to shift right the result after the fixed point multiplication
      * @param[in]  result_offset_after_shift    Offset to be applied to result before converting it back to QASYMM8
-     * @param[in]  min                          (Optional) Min value used to saturate down the output result before converting back to QASYMM8
+     * @param[in]  min                          (Optional) Min value used to saturate down the output result before converting back to QASYMM8. Defaults to the minimum possible 32-bit signed integer.
      * @param[in]  max                          (Optional) Max value used to saturate up the output result before converting back to QASYMM8,
-     *                                          Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+     *                                          Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      */
     void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, int result_fixedpoint_multiplier, int result_shift, int result_offset_after_shift,
-                   int min = 0, int max = 0);
+                   int min = std::numeric_limits<int32_t>::lowest(), int max = std::numeric_limits<int32_t>::max());
     /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint
      *
      * @param[in] input  Input tensor. It is the output of @ref CLGEMMLowpMatrixMultiplyCore function. Data type supported: S32
      * @param[in] bias   Biases tensor. Only shared biases supported and it can be a nullptr if the addition of biases is not required.
      *                   Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
      * @param[in] output Output tensor. Data type supported: Data type supported: QASYMM8
-     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QASYMM8
+     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QASYMM8. Defaults to the minimum possible 32-bit signed integer.
      * @param[in] max    (Optional) Max value used to saturate up the output result before converting back to QASYMM8,
-     *                            Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+     *                            Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0);
+    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = std::numeric_limits<int32_t>::lowest(), int max = std::numeric_limits<int32_t>::max());
 };
 
 /** Basic function to execute CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPoint on OpenCL.
@@ -189,25 +190,25 @@ public:
      * @param[in]  result_fixedpoint_multiplier Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add
      * @param[in]  result_shift                 Number of bits to shift right the result after the fixed point multiplication
      * @param[in]  result_offset_after_shift    Offset to be applied to result before converting it back to QASYMM8_SIGNED
-     * @param[in]  min                          (Optional) Min value used to saturate down the output result before converting back to QASYMM8_SIGNED. Defaults to 0
+     * @param[in]  min                          (Optional) Min value used to saturate down the output result before converting back to QASYMM8_SIGNED. Defaults to the minimum possible 32-bit signed integer.
      * @param[in]  max                          (Optional) Max value used to saturate up the output result before converting back to QASYMM8_SIGNED. Defaults to 0
-     *                                          Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+     *                                          Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      */
     void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, int result_fixedpoint_multiplier, int result_shift, int result_offset_after_shift,
-                   int min = 0, int max = 0);
+                   int min = std::numeric_limits<int32_t>::lowest(), int max = std::numeric_limits<int32_t>::max());
     /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPoint
      *
      * @param[in] input  Input tensor. It is the output of @ref CLGEMMLowpMatrixMultiplyCore function. Data type supported: S32
      * @param[in] bias   Biases tensor. Only shared biases supported and it can be a nullptr if the addition of biases is not required.
      *                   Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
      * @param[in] output Output tensor. Data type supported: Data type supported: QASYMM8_SIGNED
-     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QASYMM8_SIGNED. Defaults to 0
+     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QASYMM8_SIGNED. Defaults to the minimum possible 32-bit signed integer.
      * @param[in] max    (Optional) Max value used to saturate up the output result before converting back to QASYMM8_SIGNED. Defaults to 0
-     *                            Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+     *                            Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0);
+    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = std::numeric_limits<int32_t>::lowest(), int max = std::numeric_limits<int32_t>::max());
 };
 
 /** Basic function to execute CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloat on OpenCL.
@@ -230,24 +231,25 @@ public:
      * @param[out] output     Output tensor. Data type supported: Data type supported: QASYMM8
      * @param[in]  multiplier Float multiplier to be multiplied to each element of the input matrix
      * @param[in]  offset     Offset to be applied to result before converting it back to QASYMM8
-     * @param[in]  min        (Optional) Min value used to saturate down the output result before converting back to QASYMM8
+     * @param[in]  min        (Optional) Min value used to saturate down the output result before converting back to QASYMM8. Defaults to the minimum possible 32-bit signed integer.
      * @param[in]  max        (Optional) Max value used to saturate up the output result before converting back to QASYMM8,
-     *                        Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+     *                        Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      */
-    void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, float multiplier, int offset, int min = 0, int max = 0);
+    void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, float multiplier, int offset, int min = std::numeric_limits<int32_t>::lowest(),
+                   int max = std::numeric_limits<int32_t>::max());
     /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint
      *
      * @param[in] input  Input tensor. It is the output of @ref CLGEMMLowpMatrixMultiplyCore function. Data type supported: S32
      * @param[in] bias   Biases tensor. Only shared biases supported and it can be a nullptr if the addition of biases is not required.
      *                   Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
      * @param[in] output Output tensor. Data type supported: Data type supported: QASYMM8
-     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QASYMM8
+     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QASYMM8. Defaults to the minimum possible 32-bit signed integer.
      * @param[in] max    (Optional) Max value used to saturate up the output result before converting back to QASYMM8,
-     *                   Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+     *                   Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0);
+    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = std::numeric_limits<int32_t>::lowest(), int max = std::numeric_limits<int32_t>::max());
 };
 /** Basic function to execute CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPoint on OpenCL.
  *
@@ -288,24 +290,25 @@ public:
      * @param[out] output                       Output tensor. Data type supported: Data type supported: QSYMM16
      * @param[in]  result_fixedpoint_multiplier Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add
      * @param[in]  result_shift                 Number of bits to shift right the result after the fixed point multiplication
-     * @param[in]  min                          (Optional) Min value used to saturate down the output result before converting back to QSYMM16. Defaults to 0.
+     * @param[in]  min                          (Optional) Min value used to saturate down the output result before converting back to QSYMM16. Defaults to the minimum possible 32-bit signed integer.
      * @param[in]  max                          (Optional) Max value used to saturate up the output result before converting back to QSYMM16.
-     *                                          Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to 0.
+     *                                          Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      */
-    void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, int result_fixedpoint_multiplier, int result_shift, int min = 0, int max = 0);
+    void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, int result_fixedpoint_multiplier, int result_shift, int min = std::numeric_limits<int32_t>::lowest(),
+                   int max = std::numeric_limits<int32_t>::max());
     /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPoint
      *
      * @param[in] input  Input tensor info. It is the output of @ref CLGEMMLowpMatrixMultiplyCore function. Data type supported: S32
      * @param[in] bias   Biases tensor info. Only shared biases supported and it can be a nullptr if the addition of biases is not required.
      *                   Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
      * @param[in] output Output tensor info. Data type supported: Data type supported: QSYMM16
-     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QSYMM16. Defaults to 0.
+     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QSYMM16. Defaults to the minimum possible 32-bit signed integer.
      * @param[in] max    (Optional) Max value used to saturate up the output result before converting back to QSYMM16,
-     *                            Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to 0.
+     *                            Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0);
+    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = std::numeric_limits<int32_t>::lowest(), int max = std::numeric_limits<int32_t>::max());
 };
 } // namespace arm_compute
 #endif /*ARM_COMPUTE_CLGEMMLOWPOUTPUTSTAGE_H */
\ No newline at end of file
diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h
index ca2cbbc268..283b052917 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h
@@ -68,24 +68,25 @@ public:
      * @param[in]  result_offset   Offset to be added to each element of the input matrix
      * @param[in]  result_mult_int Value to be multiplied to each element of the input matrix when once the result_offset has been add
      * @param[in]  result_shift    Number of bits to shift right the result before converting back to QASYMM8
-     * @param[in]  min             (Optional) Min value used to saturate down the output result before converting back to QASYMM8
+     * @param[in]  min             (Optional) Min value used to saturate down the output result before converting back to QASYMM8. Defaults to the minimum possible 32-bit signed integer.
      * @param[in]  max             (Optional) Max value used to saturate up the output result before converting back to QASYMM8,
-     *                             Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+     *                             Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      */
-    void configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_offset, int result_mult_int, int result_shift, int min = 0, int max = 0);
+    void configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_offset, int result_mult_int, int result_shift, int min = std::numeric_limits<int32_t>::lowest(),
+                   int max = std::numeric_limits<int32_t>::max());
     /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpQuantizeDownInt32ToUint8Scale
      *
      * @param[in] input  Input tensor. It is the output of @ref NEGEMMLowpMatrixMultiplyCore function. Data type supported: S32
      * @param[in] bias   Biases tensor. Only shared biases supported and it can be a nullptr if the addition of biases is not required.
      *                   Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
      * @param[in] output Output tensor. Data type supported: Data type supported: QASYMM8
-     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QASYMM8
+     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QASYMM8. Defaults to the minimum possible 32-bit signed integer.
      * @param[in] max    (Optional) Max value used to saturate up the output result before converting back to QASYMM8,
-     *                   Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+     *                   Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0);
+    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = std::numeric_limits<int32_t>::lowest(), int max = std::numeric_limits<int32_t>::max());
 };
 
 /** Basic function to execute NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint on NEON.
@@ -128,24 +129,25 @@ public:
      * @param[in]  result_fixedpoint_multiplier Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add
      * @param[in]  result_shift                 Number of bits to shift right the result after the fixed point multiplication
      * @param[in]  result_offset_after_shift    Offset to be applied to result before converting it back to QASYMM8
-     * @param[in]  min                          (Optional) Min value used to saturate down the output result before converting back to QASYMM8
+     * @param[in]  min                          (Optional) Min value used to saturate down the output result before converting back to QASYMM8. Defaults to the minimum possible 32-bit signed integer.
      * @param[in]  max                          (Optional) Max value used to saturate up the output result before converting back to QASYMM8,
-     *                                          Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+     *                                          Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      */
-    void configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_fixedpoint_multiplier, int result_shift, int result_offset_after_shift, int min = 0, int max = 0);
+    void configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_fixedpoint_multiplier, int result_shift, int result_offset_after_shift,
+                   int min = std::numeric_limits<int32_t>::lowest(), int max = std::numeric_limits<int32_t>::max());
     /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint
      *
      * @param[in] input  Input tensor. It is the output of @ref NEGEMMLowpMatrixMultiplyCore function. Data type supported: S32
      * @param[in] bias   Biases tensor. Only shared biases supported and it can be a nullptr if the addition of biases is not required.
      *                   Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
      * @param[in] output Output tensor. Data type supported: Data type supported: QASYMM8
-     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QASYMM8
+     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QASYMM8. Defaults to the minimum possible 32-bit signed integer.
      * @param[in] max    (Optional) Max value used to saturate up the output result before converting back to QASYMM8,
-     *                            Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+     *                            Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0);
+    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = std::numeric_limits<int32_t>::lowest(), int max = std::numeric_limits<int32_t>::max());
 };
 /** Basic function to execute NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPoint on NEON.
  *
@@ -187,24 +189,25 @@ public:
      * @param[in]  result_fixedpoint_multiplier Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add
      * @param[in]  result_shift                 Number of bits to shift right the result after the fixed point multiplication
      * @param[in]  result_offset_after_shift    Offset to be applied to result before converting it back to QASYMM8_SIGNED
-     * @param[in]  min                          (Optional) Min value used to saturate down the output result before converting back to QASYMM8_SIGNED
+     * @param[in]  min                          (Optional) Min value used to saturate down the output result before converting back to QASYMM8_SIGNED. Defaults to the minimum possible 32-bit signed integer.
      * @param[in]  max                          (Optional) Max value used to saturate up the output result before converting back to QASYMM8_SIGNED,
-     *                                          Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+     *                                          Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      */
-    void configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_fixedpoint_multiplier, int result_shift, int result_offset_after_shift, int min = 0, int max = 0);
+    void configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_fixedpoint_multiplier, int result_shift, int result_offset_after_shift,
+                   int min = std::numeric_limits<int32_t>::lowest(), int max = std::numeric_limits<int32_t>::max());
     /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPoint
      *
      * @param[in] input  Input tensor. It is the output of @ref NEGEMMLowpMatrixMultiplyCore function. Data type supported: S32
      * @param[in] bias   Biases tensor. Only shared biases supported and it can be a nullptr if the addition of biases is not required.
      *                   Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
      * @param[in] output Output tensor. Data type supported: Data type supported: QASYMM8_SIGNED
-     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QASYMM8_SIGNED
+     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QASYMM8_SIGNED. Defaults to the minimum possible 32-bit signed integer.
      * @param[in] max    (Optional) Max value used to saturate up the output result before converting back to QASYMM8_SIGNED,
-     *                            Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+     *                            Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0);
+    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = std::numeric_limits<int32_t>::lowest(), int max = std::numeric_limits<int32_t>::max());
 };
 /** Basic function to execute NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPoint on NEON.
  *
@@ -245,24 +248,25 @@ public:
      * @param[out] output                       Output tensor. Data type supported: Data type supported: QSYMM16
      * @param[in]  result_fixedpoint_multiplier Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add
      * @param[in]  result_shift                 Number of bits to shift right the result after the fixed point multiplication
-     * @param[in]  min                          (Optional) Min value used to saturate down the output result before converting back to QSYMM16. Defaults to 0.
+     * @param[in]  min                          (Optional) Min value used to saturate down the output result before converting back to QSYMM16. Defaults to the minimum possible 32-bit signed integer.
      * @param[in]  max                          (Optional) Max value used to saturate up the output result before converting back to QSYMM16.
-     *                                          Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to 0.
+     *                                          Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      */
-    void configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_fixedpoint_multiplier, int result_shift, int min = 0, int max = 0);
+    void configure(const ITensor *input, const ITensor *bias, ITensor *output, int result_fixedpoint_multiplier, int result_shift, int min = std::numeric_limits<int32_t>::lowest(),
+                   int max = std::numeric_limits<int32_t>::max());
     /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint
      *
      * @param[in] input  Input tensor info. It is the output of @ref NEGEMMLowpMatrixMultiplyCore function. Data type supported: S32
      * @param[in] bias   Biases tensor info. Only shared biases supported and it can be a nullptr if the addition of biases is not required.
      *                   Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
      * @param[in] output Output tensor info. Data type supported: Data type supported: QSYMM16
-     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QSYMM16. Defaults to 0.
+     * @param[in] min    (Optional) Min value used to saturate down the output result before converting back to QSYMM16. Defaults to the minimum possible 32-bit signed integer.
      * @param[in] max    (Optional) Max value used to saturate up the output result before converting back to QSYMM16,
-     *                            Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to 0.
+     *                            Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to the maximum possible 32-bit signed integer.
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0);
+    static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = std::numeric_limits<int32_t>::lowest(), int max = std::numeric_limits<int32_t>::max());
 };
 
 /** Basic function to execute GEMMLowpQuantizeDown kernels on NEON.