COMPMID-1574 Implement ReduceMean in OpenCL
Change-Id: Id331199f569f52a37280a9ada5bf84694580b93c
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/152843
Tested-by: bsgcomp <bsgcomp@arm.com>
Reviewed-by: Michele DiGiorgio <michele.digiorgio@arm.com>
diff --git a/arm_compute/core/CL/kernels/CLReductionOperationKernel.h b/arm_compute/core/CL/kernels/CLReductionOperationKernel.h
index 60e2f08..ef24fd5 100644
--- a/arm_compute/core/CL/kernels/CLReductionOperationKernel.h
+++ b/arm_compute/core/CL/kernels/CLReductionOperationKernel.h
@@ -50,25 +50,27 @@
/** Set the input and output tensors.
*
- * @param[in] input Source tensor. Data types supported: F16/F32. Data layouts supported: NCHW.
+ * @param[in] input Source tensor. Data types supported: QASYMM8/F16/F32. Data layouts supported: NCHW.
* @param[out] output Destination tensor. Data types and data layouts supported: Same as @p input.
* Output will have the same number of dimensions as input.
- * @param[in] axis Axis along which to reduce. Supported reduction axis : 0
+ * @param[in] axis Axis along which to reduce. Supported reduction axis : 0,1,2,3
* @param[in] op Reduction operation to perform.
+ * @param[in] width (Optional) In case of x-axis we also need to provide the width of the input image.
*/
- void configure(const ICLTensor *input, ICLTensor *output, unsigned int axis, ReductionOperation op);
+ void configure(const ICLTensor *input, ICLTensor *output, unsigned int axis, ReductionOperation op, unsigned int width = 0);
/** Static function to check if given info will lead to a valid configuration of @ref CLReductionOperationKernel.
*
- * @param[in] input Source tensor info. Data types supported: F16/F32. Data layouts supported: NCHW.
+ * @param[in] input Source tensor info. Data types supported: QASYMM8/F16/F32. Data layouts supported: NCHW.
* @param[in] output Destination tensor info. Data types and data layouts supported: Same as @p input.
* Output will have the same number of dimensions as input.
- * @param[in] axis Axis along which to reduce. Supported reduction axis : 0
+ * @param[in] axis Axis along which to reduce. Supported reduction axis : 0,1,2,3
* @param[in] op Reduction operation to perform.
+ * @param[in] width (Optional) In case of x-axis we also need to provide the width of the input image.
*
* @return a status
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *output, unsigned int axis, ReductionOperation op);
+ static Status validate(const ITensorInfo *input, const ITensorInfo *output, unsigned int axis, ReductionOperation op, unsigned int width = 0);
// Inherited methods overridden:
void run(const Window &window, cl::CommandQueue &queue) override;
diff --git a/arm_compute/core/Types.h b/arm_compute/core/Types.h
index 9b81ecc..c0350bc 100644
--- a/arm_compute/core/Types.h
+++ b/arm_compute/core/Types.h
@@ -522,6 +522,7 @@
{
SUM_SQUARE, /**< Sum of squares */
SUM, /**< Sum */
+ MEAN_SUM, /**< Mean of sum */
};
/** The normalization type used for the normalization layer */
diff --git a/arm_compute/runtime/CL/CLFunctions.h b/arm_compute/runtime/CL/CLFunctions.h
index d87a828..6a614f7 100644
--- a/arm_compute/runtime/CL/CLFunctions.h
+++ b/arm_compute/runtime/CL/CLFunctions.h
@@ -104,6 +104,7 @@
#include "arm_compute/runtime/CL/functions/CLRNNLayer.h"
#include "arm_compute/runtime/CL/functions/CLROIAlignLayer.h"
#include "arm_compute/runtime/CL/functions/CLROIPoolingLayer.h"
+#include "arm_compute/runtime/CL/functions/CLReduceMean.h"
#include "arm_compute/runtime/CL/functions/CLReductionOperation.h"
#include "arm_compute/runtime/CL/functions/CLRemap.h"
#include "arm_compute/runtime/CL/functions/CLReorgLayer.h"
diff --git a/arm_compute/runtime/CL/functions/CLReduceMean.h b/arm_compute/runtime/CL/functions/CLReduceMean.h
new file mode 100644
index 0000000..5a919e5
--- /dev/null
+++ b/arm_compute/runtime/CL/functions/CLReduceMean.h
@@ -0,0 +1,78 @@
+/*
+ * 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_CL_REDUCE_MEAN_H__
+#define __ARM_COMPUTE_CL_REDUCE_MEAN_H__
+
+#include "arm_compute/runtime/CL/ICLSimpleFunction.h"
+#include "arm_compute/runtime/CL/functions/CLArithmeticDivision.h"
+#include "arm_compute/runtime/CL/functions/CLReductionOperation.h"
+#include "arm_compute/runtime/CL/functions/CLReshapeLayer.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+
+namespace arm_compute
+{
+// Forward Declarations
+class ICLTensor;
+
+/** Basic function to perform reduce operation */
+class CLReduceMean : public IFunction
+{
+public:
+ /** Default constructor */
+ CLReduceMean(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+ /** Configure kernel
+ *
+ * @note Supported tensor rank: up to 4
+ *
+ * @param[in] input Source tensor. Data type supported: QASYMM8/F16/F32
+ * @param[in] reduction_axis Reduction axis vector.
+ * @param[in] keep_dims If positive, retains reduced dimensions with length 1.
+ * @param[out] output Destination tensor. Data type supported: Same as @p input
+ */
+ void configure(ICLTensor *input, const Coordinates &reduction_axis, bool keep_dims, ICLTensor *output);
+
+ /** Static function to check if given info will lead to a valid configuration of @ref CLReduceMean
+ *
+ * @param[in] input Source tensor. Data type supported: QASYMM8/F16/F32
+ * @param[in] reduction_axis Reduction axis vector.
+ * @param[in] keep_dims If positive, retains reduced dimensions with length 1.
+ * @param[in] output Destination tensor. Data type supported: Same as @p input
+ *
+ * @return A status
+ */
+ static Status validate(const ITensorInfo *input, const Coordinates &reduction_axis, bool keep_dims, const ITensorInfo *output);
+
+ // Inherited methods overridden:
+ void run() override;
+
+private:
+ CLMemoryGroup _memory_group;
+ std::unique_ptr<CLReductionOperation[]> _reduction_kernels{ nullptr };
+ std::unique_ptr<CLTensor[]> _reduced_outs{ nullptr };
+ CLReshapeLayer _reshape;
+ unsigned int _reduction_ops;
+ bool _keep_dims;
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_CL_REDUCE_MEAN_H__ */
diff --git a/arm_compute/runtime/CL/functions/CLReductionOperation.h b/arm_compute/runtime/CL/functions/CLReductionOperation.h
index d862aff..4208178 100644
--- a/arm_compute/runtime/CL/functions/CLReductionOperation.h
+++ b/arm_compute/runtime/CL/functions/CLReductionOperation.h
@@ -53,7 +53,7 @@
/** Set the input and output tensors.
*
- * @param[in] input Source tensor. Data types supported: F16/F32. Data layouts supported: NCHW.
+ * @param[in] input Source tensor. Data types supported: QASYMM8/F16/F32. Data layouts supported: NCHW.
* @param[out] output Destination tensor. Data types and data layouts supported: Same as @p input.
* @param[in] axis Axis along which to reduce. Supported reduction axis : 0
* @param[in] op Reduction operation to perform.
@@ -62,7 +62,7 @@
/** Static function to check if given info will lead to a valid configuration of @ref CLReductionOperation.
*
- * @param[in] input Source tensor info. Data types supported: F16/F32. Data layouts supported: NCHW.
+ * @param[in] input Source tensor info. Data types supported: QASYMM8/F16/F32. Data layouts supported: NCHW.
* @param[in] output Destination tensor info. Data types and data layouts supported: Same as @p input.
* @param[in] axis Axis along which to reduce. Supported reduction axis : 0
* @param[in] op Reduction operation to perform.
@@ -80,6 +80,8 @@
std::unique_ptr<CLReductionOperationKernel[]> _reduction_kernels_vector{ nullptr };
std::unique_ptr<CLFillBorderKernel[]> _border_handlers_vector{ nullptr };
unsigned int _num_of_stages;
+ unsigned int _reduction_axis;
+ bool _is_quantized;
};
}
#endif /*__ARM_COMPUTE_CLREDUCTIONOPERATION_H__ */