COMPMID-1386: Add support for converting weights for CL.
Change-Id: I62e3ead903366baeeb1488f233a9b8b0c388c9de
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/140403
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
diff --git a/arm_compute/runtime/CL/functions/CLConvertFullyConnectedWeights.h b/arm_compute/runtime/CL/functions/CLConvertFullyConnectedWeights.h
index 77e9d15..9bfade4 100644
--- a/arm_compute/runtime/CL/functions/CLConvertFullyConnectedWeights.h
+++ b/arm_compute/runtime/CL/functions/CLConvertFullyConnectedWeights.h
@@ -39,7 +39,7 @@
*
* @param[in] input Source weights tensor to convert. Must be 2 dimensional. Data types supported: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32.
* @param[out] output The converted weights tensor. Shape and Data Type: Same as @p input.
- * @param[in] original_input_shape Shape of the original input tensor (the one entering fully connected layer). Must be in NCHW format.
+ * @param[in] original_input_shape Shape of the original input tensor (the one entering fully connected layer).
* @param[in] data_layout The data layout the weights have been trained in.
*/
void configure(const ICLTensor *input, ICLTensor *output, const TensorShape &original_input_shape, DataLayout data_layout);
@@ -47,7 +47,7 @@
*
* @param[in] input Source weights tensor info to convert. Must be 2 dimensional. Data types supported: U8/S8/QASYMM8/U16/S16/U32/S32/F16/F32.
* @param[in] output The converted weights tensor info. Shape and Data Type: Same as @p input.
- * @param[in] original_input_shape Shape of the original input tensor (the one entering fully connected layer). Must be in NCHW format.
+ * @param[in] original_input_shape Shape of the original input tensor (the one entering fully connected layer).
* @param[in] data_layout The data layout the weights have been trained in.
*/
static Status validate(const ITensorInfo *input, const ITensorInfo *output, const TensorShape &original_input_shape, DataLayout data_layout);
diff --git a/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h b/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h
index 3357868..6b8d7a9 100644
--- a/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h
+++ b/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h
@@ -31,6 +31,7 @@
#include "arm_compute/core/CL/kernels/CLTransposeKernel.h"
#include "arm_compute/runtime/CL/CLMemoryGroup.h"
#include "arm_compute/runtime/CL/CLTensor.h"
+#include "arm_compute/runtime/CL/functions/CLConvertFullyConnectedWeights.h"
#include "arm_compute/runtime/CL/functions/CLGEMM.h"
#include "arm_compute/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.h"
#include "arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h"
@@ -86,32 +87,26 @@
CLFullyConnectedLayer &operator=(CLFullyConnectedLayer &&) = default;
/** Set the input and output tensors.
*
- * @param[in] input Source tensor. Data type supported: QASYMM8/F16/F32.
- * @param[in] weights Weights tensor. The weights must be 2 dimensional. Data type supported: Same as @p input
- * @param[in] biases Bias tensor. It can be nullptr. Data type supported:Same as @p input.
- * @param[out] output Destination tensor. Data type supported: Same as @p input.
- * @param[in] transpose_weights (Optional) Transpose weights if true. Defaults to true.
- * @param[in] are_weights_reshaped (Optional) Reshape the weights tensor if false. Defaults to false.
- * @param[in] retain_internal_weights (Optional) Retain internal reshaped weights. Defaults to false.
- * Used for reconfiguration purposes.
+ * @param[in] input Source tensor. Data type supported: QASYMM8/F16/F32.
+ * @param[in] weights Weights tensor. The weights must be 2 dimensional. Data type supported: Same as @p input
+ * @param[in] biases Bias tensor. It can be nullptr. Data type supported:Same as @p input.
+ * @param[out] output Destination tensor. Data type supported: Same as @p input.
+ * @param[in] fc_info (Optional) Fully connected layer additional info
*/
- void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output, bool transpose_weights = true, bool are_weights_reshaped = false,
- bool retain_internal_weights = false);
+ void configure(const ICLTensor *input, const ICLTensor *weights, const ICLTensor *biases, ICLTensor *output,
+ FullyConnectedLayerInfo fc_info = FullyConnectedLayerInfo());
/** Static function to check if given info will lead to a valid configuration of @ref CLFullyConnectedLayer
*
- * @param[in] input Source tensor. Data type supported: QASYMM8/F16/F32.
- * @param[in] weights Weights tensor. The weights must be 2 dimensional. Data type supported: Same as @p input
- * @param[in] biases Bias tensor. It can be nullptr. Data type supported:Same as @p input.
- * @param[in] output Destination tensor. Data type supported: Same as @p input.
- * @param[in] transpose_weights (Optional) Transpose weights if true. Defaults to true.
- * @param[in] are_weights_reshaped (Optional) Reshape the weights tensor if false. Defaults to false.
- * @param[in] retain_internal_weights (Optional) Retain internal reshaped weights. Defaults to false.
- * Used for reconfiguration purposes.
+ * @param[in] input Source tensor. Data type supported: QASYMM8/F16/F32.
+ * @param[in] weights Weights tensor. The weights must be 2 dimensional. Data type supported: Same as @p input
+ * @param[in] biases Bias tensor. It can be nullptr. Data type supported:Same as @p input.
+ * @param[in] output Destination tensor. Data type supported: Same as @p input.
+ * @param[in] fc_info (Optional) Fully connected layer additional info
*
* @return a status
*/
- static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output, bool transpose_weights = true, bool are_weights_reshaped = false,
- bool retain_internal_weights = false);
+ static Status validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *biases, const ITensorInfo *output,
+ FullyConnectedLayerInfo fc_info = FullyConnectedLayerInfo());
//Inherited methods override
void run() override;
@@ -124,6 +119,7 @@
CLMemoryGroup _memory_group;
CLIm2ColKernel _im2col_kernel;
+ CLConvertFullyConnectedWeights _convert_weights;
CLFullyConnectedLayerReshapeWeights _reshape_weights_kernel;
CLGEMM _mm_gemm;
CLGEMMLowpMatrixMultiplyCore _mm_gemmlowp;
@@ -131,11 +127,14 @@
CLGEMMMatrixAccumulateBiasesKernel _accumulate_biases_kernel;
CLTensor _im2col_output;
CLTensor _gemmlowp_output;
+ CLTensor _converted_weights_output;
CLTensor _reshape_weights_output;
+ bool _are_weights_converted;
bool _are_weights_reshaped;
bool _is_fc_after_conv;
bool _accumulate_biases;
bool _is_quantized;
+ bool _is_prepared;
const ICLTensor *_original_weights;
};
}