COMPMID-1816: Use parallel reduction on 0 axis in CL ARG_MIN/ARG_MAX

Introducing new CLArgMinMax kernel

Change-Id: I0b8254207cc3859d19ceef9b6429cf5c1c586db0
Signed-off-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/2202
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>

2 files changed, 22 insertions, 15 deletions

diff --git a/arm_compute/runtime/CL/functions/CLArgMinMaxLayer.h b/arm_compute/runtime/CL/functions/CLArgMinMaxLayer.h
index 1b465a4866..21cded0417 100644
--- a/arm_compute/runtime/CL/functions/CLArgMinMaxLayer.h
+++ b/arm_compute/runtime/CL/functions/CLArgMinMaxLayer.h
@@ -21,10 +21,13 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef __ARM_COMPUTE_CLARGMINMAXLAYER_H__
-#define __ARM_COMPUTE_CLARGMINMAXLAYER_H__
+#ifndef ARM_COMPUTE_CLARGMINMAXLAYER_H
+#define ARM_COMPUTE_CLARGMINMAXLAYER_H
 
+#include "arm_compute/core/CL/kernels/CLArgMinMaxLayerKernel.h"
+#include "arm_compute/core/CL/kernels/CLReshapeLayerKernel.h"
 #include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/IFunction.h"
 #include "arm_compute/runtime/IMemoryManager.h"
 #include "arm_compute/runtime/MemoryGroup.h"
@@ -33,7 +36,6 @@ namespace arm_compute
 {
 class ITensorInfo;
 class ICLTensor;
-class CLReductionOperation;
 
 /** Function to calculate the index of the minimum or maximum values in a
  *  tensor based on an axis.
@@ -53,19 +55,18 @@ public:
     CLArgMinMaxLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
     /** Set the input and output tensors.
      *
-     * @param[in]  input  Input source tensor, this could be written if @ref CLReductionOperation
-     *                    manipulates its border for better performance. Data types supported: F16/F32.
+     * @param[in]  input  Input source tensor. Data types supported: F16/F32.
      * @param[in]  axis   Axis to find max/min index.
      * @param[out] output Output source tensor. Data types supported: U32/S32.
-     * @param[in]  op     Operation to perform: min or max
+     * @param[in]  op     Reduction operation to perform. Operations supported: ARG_IDX_MAX, ARG_IDX_MIN
      */
-    void configure(ICLTensor *input, int axis, ICLTensor *output, const ReductionOperation &op);
+    void configure(const ICLTensor *input, int axis, ICLTensor *output, const ReductionOperation &op);
     /** Static function to check if given info will lead to a valid configuration of @ref CLArgMinMaxLayer
      *
      * @param[in] input  Input source tensor info. Data types supported: F16/F32.
      * @param[in] axis   Axis to find max/min index.
      * @param[in] output Output source tensor info. Data types supported: U32/S32.
-     * @param[in] op     Operation to perform: min or max
+     * @param[in] op     Reduction operation to perform. Operations supported: ARG_IDX_MAX, ARG_IDX_MIN
      *
      * @return a status
      */
@@ -75,7 +76,13 @@ public:
     void run() override;
 
 private:
-    std::unique_ptr<CLReductionOperation> _reduction_function;
+    MemoryGroup                         _memory_group;
+    std::vector<CLTensor>               _results_vector;
+    CLTensor                            _not_reshaped_output;
+    std::vector<CLArgMinMaxLayerKernel> _reduction_kernels_vector;
+    CLReshapeLayerKernel                _reshape_kernel;
+    unsigned int                        _num_of_stages;
+    unsigned int                        _reduction_axis;
 };
 } // namespace arm_compute
-#endif /* __ARM_COMPUTE_CLARGMINMAXLAYER_H__ */
+#endif /* ARM_COMPUTE_CLARGMINMAXLAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLReductionOperation.h b/arm_compute/runtime/CL/functions/CLReductionOperation.h
index 405e1177fd..9e0bf03ffe 100644
--- a/arm_compute/runtime/CL/functions/CLReductionOperation.h
+++ b/arm_compute/runtime/CL/functions/CLReductionOperation.h
@@ -21,8 +21,8 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef __ARM_COMPUTE_CLREDUCTIONOPERATION_H__
-#define __ARM_COMPUTE_CLREDUCTIONOPERATION_H__
+#ifndef ARM_COMPUTE_CLREDUCTIONOPERATION_H
+#define ARM_COMPUTE_CLREDUCTIONOPERATION_H
 
 #include "arm_compute/core/CL/kernels/CLFillBorderKernel.h"
 #include "arm_compute/core/CL/kernels/CLReductionOperationKernel.h"
@@ -57,7 +57,7 @@ public:
      * @param[in]  input     Source tensor. Data types supported: QASYMM8/F16/F32.
      * @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, 1, 2, 3
-     * @param[in]  op        Reduction operation to perform.
+     * @param[in]  op        Reduction operation to perform. Operations supported: MEAN_SUM, PROD, SUM_SQUARE, SUM, MIN, MAX
      * @param[in]  keep_dims (Optional) Whether to keep the reduced dimension after the operation. Defaults to true.
      */
     void configure(ICLTensor *input, ICLTensor *output, unsigned int axis, ReductionOperation op, bool keep_dims = true);
@@ -67,7 +67,7 @@ public:
      * @param[in] input     Source tensor info. Data types supported: QASYMM8/F16/F32.
      * @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, 1, 2, 3
-     * @param[in] op        Reduction operation to perform.
+     * @param[in] op        Reduction operation to perform. Operations supported: MEAN_SUM, PROD, SUM_SQUARE, SUM, MIN, MAX
      * @param[in] keep_dims (Optional) Whether to keep the reduced dimension after the operation. Defaults to true.
      *
      * @return a status
@@ -92,4 +92,4 @@ private:
     bool                                    _is_reshape_required;
 };
 } // namespace arm_compute
-#endif /*__ARM_COMPUTE_CLREDUCTIONOPERATION_H__ */
+#endif /* ARM_COMPUTE_CLREDUCTIONOPERATION_H */
\ No newline at end of file