aboutsummaryrefslogtreecommitdiff
path: root/arm_compute/runtime/NEON/functions/NEElementwiseOperations.h
diff options
context:
space:
mode:
Diffstat (limited to 'arm_compute/runtime/NEON/functions/NEElementwiseOperations.h')
-rw-r--r--arm_compute/runtime/NEON/functions/NEElementwiseOperations.h322
1 files changed, 280 insertions, 42 deletions
diff --git a/arm_compute/runtime/NEON/functions/NEElementwiseOperations.h b/arm_compute/runtime/NEON/functions/NEElementwiseOperations.h
index cac105cdb9..ebf2277d1f 100644
--- a/arm_compute/runtime/NEON/functions/NEElementwiseOperations.h
+++ b/arm_compute/runtime/NEON/functions/NEElementwiseOperations.h
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2018-2020 ARM Limited.
+ * Copyright (c) 2018-2021, 2023 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -25,29 +25,59 @@
#define ARM_COMPUTE_NEELEMENTWISEOPERATIONS_H
#include "arm_compute/core/Types.h"
-#include "arm_compute/runtime/NEON/INESimpleFunction.h"
+#include "arm_compute/function_info/ActivationLayerInfo.h"
+#include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/NEON/INEOperator.h"
namespace arm_compute
{
class ITensor;
-/** Basic function to run @ref NEArithmeticOperationKernel for max
+/** Basic function to run @ref cpu::kernels::CpuArithmeticKernel for max
*
* @note The tensor data type for the inputs must be QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
* @note The function performs a max operation between two tensors.
*/
-class NEElementwiseMax : public INESimpleFunction
+class NEElementwiseMax : public IFunction
{
public:
+ /** Default Constructor */
+ NEElementwiseMax();
+ /** Default Destructor */
+ ~NEElementwiseMax();
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEElementwiseMax(const NEElementwiseMax &) = delete;
+ /** Default move constructor */
+ NEElementwiseMax(NEElementwiseMax &&);
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEElementwiseMax &operator=(const NEElementwiseMax &) = delete;
+ /** Default move assignment operator */
+ NEElementwiseMax &operator=(NEElementwiseMax &&);
/** Initialise the kernel's inputs, output and conversion policy.
*
+ * Valid data layouts:
+ * - All
+ *
+ * Valid data type configurations:
+ * |src0 |src1 |dst |
+ * |:--------------|:--------------|:--------------|
+ * |QASYMM8 |QASYMM8 |QASYMM8 |
+ * |QASYMM8_SIGNED |QASYMM8_SIGNED |QASYMM8_SIGNED |
+ * |S32 |S32 |S32 |
+ * |S16 |S16 |S16 |
+ * |F16 |F16 |F16 |
+ * |F32 |F32 |F32 |
+ *
* @param[in, out] input1 First tensor input. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
* @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1.
* @param[out] output Output tensor. Data types supported: Same as @p input1.
* @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported.
*/
- void configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
- /** Static function to check if given info will lead to a valid configuration of @ref NEArithmeticOperationKernel for max
+ void configure(ITensor *input1,
+ ITensor *input2,
+ ITensor *output,
+ const ActivationLayerInfo &act_info = ActivationLayerInfo());
+ /** Static function to check if given info will lead to a valid configuration of @ref cpu::kernels::CpuArithmeticKernel for max
*
* @param[in] input1 First tensor input info. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
* @param[in] input2 Second tensor input info. Data types supported: Same as @p input1.
@@ -56,26 +86,64 @@ public:
*
* @return a status
*/
- static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
+ static Status validate(const ITensorInfo *input1,
+ const ITensorInfo *input2,
+ const ITensorInfo *output,
+ const ActivationLayerInfo &act_info = ActivationLayerInfo());
+
+ // Inherited methods overridden:
+ void run() override;
+
+private:
+ struct Impl;
+ std::unique_ptr<Impl> _impl;
};
-/** Basic function to run @ref NEArithmeticOperationKernel for min
+/** Basic function to run @ref cpu::kernels::CpuArithmeticKernel for min
*
* @note The tensor data type for the inputs must be QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
* @note The function performs a min operation between two tensors.
*/
-class NEElementwiseMin : public INESimpleFunction
+class NEElementwiseMin : public IFunction
{
public:
+ /** Default Constructor */
+ NEElementwiseMin();
+ /** Default Destructor */
+ ~NEElementwiseMin();
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEElementwiseMin(const NEElementwiseMin &) = delete;
+ /** Default move constructor */
+ NEElementwiseMin(NEElementwiseMin &&);
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEElementwiseMin &operator=(const NEElementwiseMin &) = delete;
+ /** Default move assignment operator */
+ NEElementwiseMin &operator=(NEElementwiseMin &&);
/** Initialise the kernel's inputs, output and conversion policy.
*
+ * Valid data layouts:
+ * - All
+ *
+ * Valid data type configurations:
+ * |src0 |src1 |dst |
+ * |:--------------|:--------------|:--------------|
+ * |QASYMM8 |QASYMM8 |QASYMM8 |
+ * |QASYMM8_SIGNED |QASYMM8_SIGNED |QASYMM8_SIGNED |
+ * |S32 |S32 |S32 |
+ * |S16 |S16 |S16 |
+ * |F16 |F16 |F16 |
+ * |F32 |F32 |F32 |
+ *
* @param[in, out] input1 First tensor input. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
* @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1.
* @param[out] output Output tensor. Data types supported: Same as @p input1.
* @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported.
*/
- void configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
- /** Static function to check if given info will lead to a valid configuration of @ref NEArithmeticOperationKernel for min
+ void configure(ITensor *input1,
+ ITensor *input2,
+ ITensor *output,
+ const ActivationLayerInfo &act_info = ActivationLayerInfo());
+ /** Static function to check if given info will lead to a valid configuration of @ref cpu::kernels::CpuArithmeticKernel for min
*
* @param[in] input1 First tensor input info. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
* @param[in] input2 Second tensor input info. Data types supported: Same as @p input1.
@@ -84,26 +152,64 @@ public:
*
* @return a status
*/
- static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
+ static Status validate(const ITensorInfo *input1,
+ const ITensorInfo *input2,
+ const ITensorInfo *output,
+ const ActivationLayerInfo &act_info = ActivationLayerInfo());
+
+ // Inherited methods overridden:
+ void run() override;
+
+private:
+ struct Impl;
+ std::unique_ptr<Impl> _impl;
};
-/** Basic function to run @ref NEArithmeticOperationKernel for squared difference
+/** Basic function to run @ref cpu::kernels::CpuArithmeticKernel for squared difference
*
* @note The tensor data type for the inputs must be QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
* @note The function performs a squared different operation between two tensors (i.e., out[i] = (in1[i] - in2[i])^2
*/
-class NEElementwiseSquaredDiff : public INESimpleFunction
+class NEElementwiseSquaredDiff : public IFunction
{
public:
+ /** Default Constructor */
+ NEElementwiseSquaredDiff();
+ /** Default Destructor */
+ ~NEElementwiseSquaredDiff();
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEElementwiseSquaredDiff(const NEElementwiseSquaredDiff &) = delete;
+ /** Default move constructor */
+ NEElementwiseSquaredDiff(NEElementwiseSquaredDiff &&);
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEElementwiseSquaredDiff &operator=(const NEElementwiseSquaredDiff &) = delete;
+ /** Default move assignment operator */
+ NEElementwiseSquaredDiff &operator=(NEElementwiseSquaredDiff &&);
/** Initialise the kernel's inputs, output and conversion policy.
*
+ * Valid data layouts:
+ * - All
+ *
+ * Valid data type configurations:
+ * |src0 |src1 |dst |
+ * |:--------------|:--------------|:--------------|
+ * |QASYMM8 |QASYMM8 |QASYMM8 |
+ * |QASYMM8_SIGNED |QASYMM8_SIGNED |QASYMM8_SIGNED |
+ * |S32 |S32 |S32 |
+ * |S16 |S16 |S16 |
+ * |F16 |F16 |F16 |
+ * |F32 |F32 |F32 |
+ *
* @param[in, out] input1 First tensor input. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
* @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1.
* @param[out] output Output tensor. Data types supported: Same as @p input1.
* @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported.
*/
- void configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
- /** Static function to check if given info will lead to a valid configuration of @ref NEArithmeticOperationKernel for squared difference
+ void configure(ITensor *input1,
+ ITensor *input2,
+ ITensor *output,
+ const ActivationLayerInfo &act_info = ActivationLayerInfo());
+ /** Static function to check if given info will lead to a valid configuration of @ref cpu::kernels::CpuArithmeticKernel for squared difference
*
* @param[in] input1 First tensor input info. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
* @param[in] input2 Second tensor input info. Data types supported: Same as @p input1.
@@ -112,26 +218,60 @@ public:
*
* @return a status
*/
- static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
+ static Status validate(const ITensorInfo *input1,
+ const ITensorInfo *input2,
+ const ITensorInfo *output,
+ const ActivationLayerInfo &act_info = ActivationLayerInfo());
+
+ // Inherited methods overridden:
+ void run() override;
+
+private:
+ struct Impl;
+ std::unique_ptr<Impl> _impl;
};
-/** Basic function to run @ref NEArithmeticOperationKernel for division
+/** Basic function to run @ref cpu::kernels::CpuArithmeticKernel for division
*
* @note The tensor data type for the inputs must be F16/F32.
* @note The function performs a squared different operation between two tensors (i.e., out[i] = in1[i] / in2[i])
*/
-class NEElementwiseDivision : public INESimpleFunction
+class NEElementwiseDivision : public IFunction
{
public:
+ /** Default Constructor */
+ NEElementwiseDivision();
+ /** Default Destructor */
+ ~NEElementwiseDivision();
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEElementwiseDivision(const NEElementwiseDivision &) = delete;
+ /** Default move constructor */
+ NEElementwiseDivision(NEElementwiseDivision &&);
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEElementwiseDivision &operator=(const NEElementwiseDivision &) = delete;
+ /** Default move assignment operator */
+ NEElementwiseDivision &operator=(NEElementwiseDivision &&);
/** Initialise the kernel's inputs, output and conversion policy.
*
+ * Valid data layouts:
+ * - All
+ *
+ * Valid data type configurations:
+ * |src0 |src1 |dst |
+ * |:--------------|:--------------|:--------------|
+ * |F16 |F16 |F16 |
+ * |F32 |F32 |F32 |
+ *
* @param[in, out] input1 First tensor input. Data types supported: F16/F32.
* @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1.
* @param[out] output Output tensor. Data types supported: Same as @p input1.
* @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported.
*/
- void configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
- /** Static function to check if given info will lead to a valid configuration of @ref NEArithmeticOperationKernel for division
+ void configure(ITensor *input1,
+ ITensor *input2,
+ ITensor *output,
+ const ActivationLayerInfo &act_info = ActivationLayerInfo());
+ /** Static function to check if given info will lead to a valid configuration of @ref cpu::kernels::CpuArithmeticKernel for division
*
* @param[in] input1 First tensor input info. Data types supported: F16/F32.
* @param[in] input2 Second tensor input info. Data types supported: Same as @p input1.
@@ -140,27 +280,61 @@ public:
*
* @return a status
*/
- static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
+ static Status validate(const ITensorInfo *input1,
+ const ITensorInfo *input2,
+ const ITensorInfo *output,
+ const ActivationLayerInfo &act_info = ActivationLayerInfo());
+
+ // Inherited methods overridden:
+ void run() override;
+
+private:
+ struct Impl;
+ std::unique_ptr<Impl> _impl;
};
-/** Basic function to run @ref NEArithmeticOperationKernel for power
+/** Basic function to run @ref cpu::kernels::CpuArithmeticKernel for power
*
* @note The tensor data type for the inputs must be F16/F32.
* @note The function performs a elementwise power of in1 to in2 (i.e., out[i] = in1[i] ^ in2[i])
* @note For an exponent that is a float, this function will only work with a positive base.
*/
-class NEElementwisePower : public INESimpleFunction
+class NEElementwisePower : public IFunction
{
public:
+ /** Default Constructor */
+ NEElementwisePower();
+ /** Default Destructor */
+ ~NEElementwisePower();
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEElementwisePower(const NEElementwisePower &) = delete;
+ /** Default move constructor */
+ NEElementwisePower(NEElementwisePower &&);
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEElementwisePower &operator=(const NEElementwisePower &) = delete;
+ /** Default move assignment operator */
+ NEElementwisePower &operator=(NEElementwisePower &&);
/** Initialise the kernel's inputs, output and conversion policy.
*
+ * Valid data layouts:
+ * - All
+ *
+ * Valid data type configurations:
+ * |src0 |src1 |dst |
+ * |:--------------|:--------------|:--------------|
+ * |F16 |F16 |F16 |
+ * |F32 |F32 |F32 |
+ *
* @param[in, out] input1 First tensor input. Data types supported: F16/F32.
* @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1.
* @param[out] output Output tensor. Data types supported: Same as @p input1.
* @param[in] act_info (Optional) Activation layer information in case of a fused activation. Currently not supported.
*/
- void configure(ITensor *input1, ITensor *input2, ITensor *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
- /** Static function to check if given info will lead to a valid configuration of @ref NEArithmeticOperationKernel for power
+ void configure(ITensor *input1,
+ ITensor *input2,
+ ITensor *output,
+ const ActivationLayerInfo &act_info = ActivationLayerInfo());
+ /** Static function to check if given info will lead to a valid configuration of @ref cpu::kernels::CpuArithmeticKernel for power
*
* @param[in] input1 First tensor input info. Data types supported: F16/F32.
* @param[in] input2 Second tensor input info. Data types supported: Same as @p input1.
@@ -169,62 +343,125 @@ public:
*
* @return a status
*/
- static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info = ActivationLayerInfo());
+ static Status validate(const ITensorInfo *input1,
+ const ITensorInfo *input2,
+ const ITensorInfo *output,
+ const ActivationLayerInfo &act_info = ActivationLayerInfo());
+
+ // Inherited methods overridden:
+ void run() override;
+
+private:
+ struct Impl;
+ std::unique_ptr<Impl> _impl;
};
-/** Basic function to run @ref NEComparisonOperationKernel.
+/** Basic function to run @ref cpu::kernels::CpuComparisonKernel.
*
- * @note The tensor data type for the inputs must be QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
+ * @note The tensor data type for the inputs must be U8/QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
* @note The function performs a comparison operation between two tensors.
*/
-class NEElementwiseComparison : public INESimpleFunction
+class NEElementwiseComparison : public IFunction
{
public:
+ /** Default Constructor */
+ NEElementwiseComparison();
+ /** Default Destructor */
+ ~NEElementwiseComparison();
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEElementwiseComparison(const NEElementwiseComparison &) = delete;
+ /** Default move constructor */
+ NEElementwiseComparison(NEElementwiseComparison &&);
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEElementwiseComparison &operator=(const NEElementwiseComparison &) = delete;
+ /** Default move assignment operator */
+ NEElementwiseComparison &operator=(NEElementwiseComparison &&);
/** Initialise the kernel's inputs, output and conversion policy.
*
- * @param[in, out] input1 First tensor input. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
+ * Valid data layouts:
+ * - All
+ *
+ * Valid data type configurations:
+ * |src0 |src1 |dst |
+ * |:--------------|:--------------|:-----|
+ * |QASYMM8 |QASYMM8 |U8 |
+ * |QASYMM8_SIGNED |QASYMM8_SIGNED |U8 |
+ * |S32 |S32 |U8 |
+ * |U8 |U8 |U8 |
+ * |S16 |S16 |U8 |
+ * |F16 |F16 |U8 |
+ * |F32 |F32 |U8 |
+ *
+ * @param[in, out] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
* @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1.
- * @param[out] output Output tensor. Data types supported: U16/U32.
+ * @param[out] output Output tensor. Data types supported: U8.
* @param[in] op Comparison Operation to be performed.
*/
void configure(ITensor *input1, ITensor *input2, ITensor *output, ComparisonOperation op);
- /** Static function to check if given info will lead to a valid configuration of @ref NEComparisonOperationKernel
+ /** Static function to check if given info will lead to a valid configuration of @ref cpu::kernels::CpuComparisonKernel
*
- * @param[in] input1 First tensor input info. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
+ * @param[in] input1 First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
* @param[in] input2 Second tensor input info. Data types supported: Same as @p input1.
- * @param[in] output Output tensor info. Data types supported: U16/U32.
+ * @param[in] output Output tensor info. Data types supported: U8.
* @param[in] op Comparison Operation to be performed.
*
* @return a status
*/
- static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ComparisonOperation op);
+ static Status
+ validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, ComparisonOperation op);
+
+ // Inherited methods overridden:
+ void run() override;
+
+private:
+ struct Impl;
+ std::unique_ptr<Impl> _impl;
};
-/** Basic function to run @ref NEComparisonOperationKernel
+/** Basic function to run @ref cpu::kernels::CpuComparisonKernel
*
- * @note The tensor data type for the inputs must be QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
+ * @note The tensor data type for the inputs must be U8/QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
* @note The function performs a comparison operation between two tensors.
*/
template <ComparisonOperation op>
-class NEElementwiseComparisonStatic : public INESimpleFunction
+class NEElementwiseComparisonStatic : public IFunction
{
public:
+ /** Default Constructor */
+ NEElementwiseComparisonStatic();
+ /** Default Destructor */
+ ~NEElementwiseComparisonStatic();
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEElementwiseComparisonStatic(const NEElementwiseComparisonStatic &) = delete;
+ /** Default move constructor */
+ NEElementwiseComparisonStatic(NEElementwiseComparisonStatic &&);
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ NEElementwiseComparisonStatic &operator=(const NEElementwiseComparisonStatic &) = delete;
+ /** Default move assignment operator */
+ NEElementwiseComparisonStatic &operator=(NEElementwiseComparisonStatic &&);
/** Initialise the kernel's inputs, output and conversion policy.
*
- * @param[in, out] input1 First tensor input. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
+ * @param[in, out] input1 First tensor input. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
* @param[in, out] input2 Second tensor input. Data types supported: Same as @p input1.
* @param[out] output Output tensor. Data types supported: U16/U32.
*/
void configure(ITensor *input1, ITensor *input2, ITensor *output);
- /** Static function to check if given info will lead to a valid configuration of @ref NEComparisonOperationKernel
+ /** Static function to check if given info will lead to a valid configuration of @ref cpu::kernels::CpuComparisonKernel
*
- * @param[in] input1 First tensor input info. Data types supported: QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
+ * @param[in] input1 First tensor input info. Data types supported: U8/QASYMM8/QASYMM8_SIGNED/S16/F16/S32/F32.
* @param[in] input2 Second tensor input info. Data types supported: Same as @p input1.
* @param[in] output Output tensor info. Data types supported: U16/U32.
*
* @return a status
*/
static Status validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output);
+
+ // Inherited methods overridden:
+ void run() override;
+
+private:
+ struct Impl;
+ std::unique_ptr<Impl> _impl;
};
/** Basic function to run equal comparison. */
@@ -239,5 +476,6 @@ using NEGreaterEqual = NEElementwiseComparisonStatic<ComparisonOperation::Greate
using NELess = NEElementwiseComparisonStatic<ComparisonOperation::Less>;
/** Basic function to run less-equal comparison. */
using NELessEqual = NEElementwiseComparisonStatic<ComparisonOperation::LessEqual>;
+
} // namespace arm_compute
#endif /* ARM_COMPUTE_NEELEMENTWISEOPERATIONS_H */