aboutsummaryrefslogtreecommitdiff
path: root/arm_compute/runtime
diff options
context:
space:
mode:
Diffstat (limited to 'arm_compute/runtime')
-rw-r--r--arm_compute/runtime/CL/ICLOperator.h2
-rw-r--r--arm_compute/runtime/CL/functions/CLConcatenateLayer.h69
-rw-r--r--arm_compute/runtime/CL/functions/CLLSTMLayer.h9
3 files changed, 65 insertions, 15 deletions
diff --git a/arm_compute/runtime/CL/ICLOperator.h b/arm_compute/runtime/CL/ICLOperator.h
index 7ad0c8ce0e..c9fdd864de 100644
--- a/arm_compute/runtime/CL/ICLOperator.h
+++ b/arm_compute/runtime/CL/ICLOperator.h
@@ -54,7 +54,7 @@ public:
ICLOperator &operator=(ICLOperator &&) = default;
// Inherited methods overridden:
- void run(InputTensorMap inputs, OutputTensorMap outputs, OperatorTensorMap workspace) override final;
+ void run(InputTensorMap inputs, OutputTensorMap outputs, OperatorTensorMap workspace) override;
void prepare(OperatorTensorMap constants) override final;
protected:
diff --git a/arm_compute/runtime/CL/functions/CLConcatenateLayer.h b/arm_compute/runtime/CL/functions/CLConcatenateLayer.h
index c5869345c7..4e8a95be43 100644
--- a/arm_compute/runtime/CL/functions/CLConcatenateLayer.h
+++ b/arm_compute/runtime/CL/functions/CLConcatenateLayer.h
@@ -24,6 +24,7 @@
#ifndef ARM_COMPUTE_CLCONCATENATELAYER_H
#define ARM_COMPUTE_CLCONCATENATELAYER_H
+#include "arm_compute/runtime/CL/ICLOperator.h"
#include "arm_compute/runtime/IFunction.h"
#include "arm_compute/core/CL/ICLKernel.h"
@@ -51,16 +52,25 @@ class CLConcatenateLayer : public IFunction
public:
/** Default constructor */
CLConcatenateLayer();
+ /** Destructor */
+ ~CLConcatenateLayer();
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ CLConcatenateLayer(const CLConcatenateLayer &) = delete;
+ /** Default move constructor */
+ CLConcatenateLayer(CLConcatenateLayer &&);
+ /** Prevent instances of this class from being copied (As this class contains pointers) */
+ CLConcatenateLayer &operator=(const CLConcatenateLayer &) = delete;
+ /** Default move assignment operator */
+ CLConcatenateLayer &operator=(CLConcatenateLayer &&);
/** Initialise the kernel's inputs vector and output.
*
* @note Input and output tensor dimensions preconditions defer depending on the concatenation axis.
* @note Preconditions can be found respectively at @ref CLWidthConcatenateLayerKernel, @ref CLHeightConcatenateLayerKernel and @ref CLDepthConcatenateLayerKernel.
*
- * @param[in,out] inputs_vector The vectors containing all the tensors to concatenate. Data types supported: All.
+ * @param[in,out] inputs_vector The vectors containing all the tensors to concatenate. Data types supported: All
* @param[out] output Output tensor. Data types supported: Same as @p input.
* @param[in] axis Concatenation axis. Supported underlying concatenation axis are 0, 1, 2 and 3.
*/
- void configure(std::vector<ICLTensor *> &inputs_vector, ICLTensor *output, size_t axis);
void configure(std::vector<const ICLTensor *> &inputs_vector, ICLTensor *output, size_t axis);
/** Initialise the kernel's inputs vector and output.
*
@@ -68,11 +78,10 @@ public:
* @note Preconditions can be found respectively at @ref CLWidthConcatenateLayerKernel, @ref CLHeightConcatenateLayerKernel and @ref CLDepthConcatenateLayerKernel.
*
* @param[in] compile_context The compile context to be used.
- * @param[in,out] inputs_vector The vectors containing all the tensors to concatenate. Data types supported: All.
+ * @param[in,out] inputs_vector The vectors containing all the tensors to concatenate. Data types supported: All
* @param[out] output Output tensor. Data types supported: Same as @p input.
* @param[in] axis Concatenation axis. Supported underlying concatenation axis are 0, 1, 2 and 3.
*/
- void configure(const CLCompileContext &compile_context, std::vector<ICLTensor *> &inputs_vector, ICLTensor *output, size_t axis);
void configure(const CLCompileContext &compile_context, std::vector<const ICLTensor *> &inputs_vector, ICLTensor *output, size_t axis);
/** Static function to check if given info will lead to a valid configuration of @ref CLConcatenateLayer
*
@@ -85,22 +94,64 @@ public:
*
* @return a status
*/
- static Status validate(const std::vector<ITensorInfo *> &inputs_vector, const ITensorInfo *output, size_t axis);
static Status validate(const std::vector<const ITensorInfo *> &inputs_vector, const ITensorInfo *output, size_t axis);
// Inherited methods overridden:
void run() override;
private:
- template <typename TensorType>
- void configure_internal(const CLCompileContext &compile_context, std::vector<TensorType *> &&inputs_vector, ICLTensor *output, size_t axis);
+ struct Impl;
+ std::unique_ptr<Impl> _impl;
+};
+
+namespace experimental
+{
+/** Basic function to execute concatenate tensors along a given axis. This function calls the following kernels:
+ *
+ * -# @ref CLWidthConcatenateLayerKernel (if underlying concatenation axis is 0).
+ * -# @ref CLHeightConcatenateLayerKernel (if underlying concatenation axis is 1).
+ * -# @ref CLDepthConcatenateLayerKernel (if underlying concatenation axis is 2).
+ * -# @ref CLBatchConcatenateLayerKernel (if underlying concatenation axis is 3).
+ */
+class CLConcatenateLayer : public ICLOperator
+{
+public:
+ /** Default constructor */
+ CLConcatenateLayer();
+ /** Initialise the kernel's inputs vector and output.
+ *
+ * @note Input and output tensor dimensions preconditions defer depending on the concatenation axis.
+ * @note Preconditions can be found respectively at @ref CLWidthConcatenateLayerKernel, @ref CLHeightConcatenateLayerKernel and @ref CLDepthConcatenateLayerKernel.
+ *
+ *
+ * @param[in] compile_context The compile context to be used.
+ * @param[in,out] inputs_vector The vectors containing all the tensors to concatenate. Data types supported: All
+ * @param[out] output Output tensor. Data types supported: Same as @p input.
+ * @param[in] axis Concatenation axis. Supported underlying concatenation axis are 0, 1, 2 and 3.
+ */
+ void configure(const CLCompileContext &compile_context, const std::vector<ITensorInfo *> &inputs_vector, ITensorInfo *output, size_t axis);
+ /** Static function to check if given info will lead to a valid configuration of @ref NEConcatenateLayer
+ *
+ * @note Input and output tensor dimensions preconditions defer depending on the concatenation axis.
+ * @note Preconditions can be found respectively at @ref CLWidthConcatenateLayerKernel, @ref CLHeightConcatenateLayerKernel and @ref CLDepthConcatenateLayerKernel.
+ *
+ * @param[in] inputs_vector The vectors containing all the tensors info to concatenate. Data types supported: All
+ * @param[in] output Output tensor info. Data types supported: Same as @p input.
+ * @param[in] axis Concatenation axis. Supported underlying concatenation axis are 0, 1, 2 and 3.
+ *
+ * @return a status
+ */
+ static Status validate(const std::vector<const ITensorInfo *> &inputs_vector, const ITensorInfo *output, size_t axis);
- template <typename TensorInfoType>
- static Status validate_internal(const std::vector<TensorInfoType *> &inputs_vector, const ITensorInfo *output, size_t axis);
+ // Inherited methods overridden:
+ MemoryRequirements workspace() const override;
+ void run(InputTensorMap inputs, OutputTensorMap outputs, OperatorTensorMap workspace) override;
+private:
std::vector<std::unique_ptr<ICLKernel>> _concat_kernels;
unsigned int _num_inputs;
unsigned int _axis;
};
+} // namespace experimental
} // namespace arm_compute
#endif /* ARM_COMPUTE_CLCONCATENATELAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLLSTMLayer.h b/arm_compute/runtime/CL/functions/CLLSTMLayer.h
index 2e44eed6f6..e5733cd784 100644
--- a/arm_compute/runtime/CL/functions/CLLSTMLayer.h
+++ b/arm_compute/runtime/CL/functions/CLLSTMLayer.h
@@ -30,7 +30,6 @@
#include "arm_compute/core/CL/kernels/CLElementwiseOperationKernel.h"
#include "arm_compute/core/CL/kernels/CLMemsetKernel.h"
#include "arm_compute/core/CL/kernels/CLPixelWiseMultiplicationKernel.h"
-#include "arm_compute/core/CL/kernels/CLWidthConcatenate2TensorsKernel.h"
#include "arm_compute/core/Types.h"
#include "arm_compute/runtime/CL/CLTensor.h"
#include "arm_compute/runtime/CL/functions/CLActivationLayer.h"
@@ -232,10 +231,10 @@ private:
CLCopyKernel _copy_cell_state;
CLCopyKernel _copy_output;
CLConcatenateLayer _concat_scratch_buffer;
- CLWidthConcatenate2TensorsKernel _concat_inputs_forget_gate;
- CLWidthConcatenate2TensorsKernel _concat_weights_forget_gate;
- CLWidthConcatenate2TensorsKernel _concat_weights_input_gate;
- CLWidthConcatenate2TensorsKernel _concat_weights_output;
+ CLConcatenateLayer _concat_inputs_forget_gate;
+ CLConcatenateLayer _concat_weights_forget_gate;
+ CLConcatenateLayer _concat_weights_input_gate;
+ CLConcatenateLayer _concat_weights_output;
CLMemsetKernel _ones_memset_kernel;
CLMeanStdDevNormalizationLayer _mean_std_norm_input_gate;
CLPixelWiseMultiplicationKernel _pixelwise_mul_input_gate_coeff;
generated by cgit v1.2.1 (git 2.23.0) at 2024-05-15 07:25:39 +0000