Port NEGEMMConv2d to memory injecting interface
Resolves: COMPMID-4506, COMPMID-4570
Change-Id: I6d37a06da141f1fcfcaa8525322a319cb0234791
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5824
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
index dc9783f..ff88876 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
@@ -24,32 +24,15 @@
#ifndef ARM_COMPUTE_NEGEMMLOWPMATRIXMULTIPLYCORE_H
#define ARM_COMPUTE_NEGEMMLOWPMATRIXMULTIPLYCORE_H
-#include "NEActivationLayer.h"
-#include "arm_compute/core/ITensorPack.h"
#include "arm_compute/runtime/IFunction.h"
#include "arm_compute/runtime/IMemoryManager.h"
#include "arm_compute/runtime/IWeightsManager.h"
-#include "arm_compute/runtime/MemoryGroup.h"
-#include "arm_compute/runtime/Tensor.h"
#include <memory>
namespace arm_compute
{
class ITensor;
-class NEConvertQuantizedSignednessKernel;
-class NEGEMMInterleave4x4Kernel;
-class NEGEMMLowpMatrixMultiplyKernel;
-class NEGEMMLowpOffsetContributionKernel;
-class NEGEMMLowpOffsetContributionOutputStageKernel;
-class NEGEMMLowpMatrixAReductionKernel;
-class NEGEMMLowpMatrixBReductionKernel;
-class NEGEMMTranspose1xWKernel;
-namespace cpu
-{
-class CpuGemmAssemblyDispatch;
-}
-
/** Basic function to execute GEMMLowpMatrixMultiplyCore. This function calls the following kernels if the DOT product instruction is not available:
*
* -# @ref NEGEMMInterleave4x4Kernel
@@ -119,14 +102,7 @@
void configure(const ITensor *a, const ITensor *b, const ITensor *c, ITensor *output, const GEMMInfo &gemm_info = GEMMInfo());
/** Static function to check if given info will lead to a valid configuration of @ref NEGEMMLowpMatrixMultiplyCore
*
- * @note The @p output type is S32 if @p gemm_info.type == GEMMLowpOutputStageType::NONE. It is QASYMM8/QASYMM8_SIGNED otherwise
- *
- * @param[in] a First input tensor info (Matrix A). Data type supported: QASYMM8/QASYMM8_SIGNED.
- * @param[in] b Second input tensor info (Matrix B). Data type supported: QASYMM8/QASYMM8_SIGNED/QSYMM8/QSYMM8_PER_CHANNEL.
- * @param[in] c Third input tensor info (Matrix C). It can be a nullptr. Data type supported: S32
- * @param[in] output Output tensor info. Data type supported: Data type supported: S32/QASYMM8/QASYMM8_SIGNED
- * @param[in] gemm_info (Optional) Specifies if the matrix A and/or matrix B have been reshaped and
- * if the reshape of matrix B should be executed only for the first run
+ * Similar to @ref NEGEMMLowpMatrixMultiplyCore::configure()
*
* @return a status
*/
@@ -137,41 +113,8 @@
void prepare() override;
private:
- MemoryGroup _memory_group;
- IWeightsManager *_weights_manager;
- std::unique_ptr<cpu::CpuGemmAssemblyDispatch> _asm_glue;
- std::unique_ptr<NEGEMMLowpMatrixMultiplyKernel> _mm_kernel;
- std::unique_ptr<NEGEMMInterleave4x4Kernel> _mtx_a_reshape_kernel;
- std::unique_ptr<NEGEMMTranspose1xWKernel> _mtx_b_reshape_kernel;
- std::unique_ptr<NEGEMMLowpMatrixAReductionKernel> _mtx_a_reduction_kernel;
- std::unique_ptr<NEGEMMLowpMatrixBReductionKernel> _mtx_b_reduction_kernel;
- std::unique_ptr<NEGEMMLowpOffsetContributionKernel> _offset_contribution_kernel;
- std::unique_ptr<NEGEMMLowpOffsetContributionOutputStageKernel> _offset_contribution_output_stage_kernel;
- NEActivationLayer _activation_func;
- std::unique_ptr<NEConvertQuantizedSignednessKernel> _convert_to_signed_asymm;
- std::unique_ptr<NEConvertQuantizedSignednessKernel> _convert_from_signed_asymm;
-
- Tensor _vector_sum_col;
- Tensor _vector_sum_row;
- Tensor _tmp_a;
- Tensor _tmp_b;
- Tensor _mm_result_s32;
- Tensor _signed_a;
- Tensor _signed_output;
- const ITensor *_original_b;
- int32_t _a_offset;
- int32_t _b_offset;
-
- bool _run_vector_matrix_multiplication;
- bool _assembly_path;
- bool _fused_assembly_path;
- bool _reshape_b_only_on_first_run;
- bool _is_prepared;
- bool _fuse_output_stage;
- bool _run_activation;
- bool _flip_signedness;
-
- ITensorPack _asm_glue_tensors{};
+ struct Impl;
+ std::unique_ptr<struct Impl> _impl;
};
} // namespace arm_compute
#endif /*ARM_COMPUTE_NEGEMMLOWPMATRIXMULTIPLYCORE_H */