COMPMID-881: RSH new arm_gemm interface.

Change-Id: I1e2a1a77097d8017c274af3f97eba6964f80f5fa
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/122592
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>

4 files changed, 25 insertions, 26 deletions

diff --git a/arm_compute/runtime/NEON/functions/NEGEMM.h b/arm_compute/runtime/NEON/functions/NEGEMM.h
index f2b6ef77bd..5279995be4 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMM.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMM.h
@@ -25,7 +25,6 @@
 #define __ARM_COMPUTE_NEGEMM_H__
 
 #include "arm_compute/core/NEON/kernels/NEFillBorderKernel.h"
-#include "arm_compute/core/NEON/kernels/NEGEMMAssemblyBaseKernel.h"
 #include "arm_compute/core/NEON/kernels/NEGEMMInterleave4x4Kernel.h"
 #include "arm_compute/core/NEON/kernels/NEGEMMMatrixAdditionKernel.h"
 #include "arm_compute/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.h"
@@ -35,6 +34,8 @@
 #include "arm_compute/runtime/MemoryGroup.h"
 #include "arm_compute/runtime/Tensor.h"
 
+#include "arm_compute/runtime/NEON/AssemblyHelper.h"
+
 #include <memory>
 
 namespace arm_compute
@@ -73,19 +74,19 @@ public:
     void run() override;
 
 private:
-    MemoryGroup                               _memory_group;
-    NEGEMMInterleave4x4Kernel                 _interleave_kernel;
-    NEGEMMTranspose1xWKernel                  _transpose_kernel;
-    NEGEMMMatrixMultiplyKernel                _mm_kernel;
-    std::unique_ptr<NEGEMMAssemblyBaseKernel> _mm_optimised_kernel;
-    NEGEMMMatrixAdditionKernel                _ma_kernel;
-    Tensor                                    _tmp_a;
-    Tensor                                    _tmp_b;
-    Tensor                                    _workspace;
-    bool                                      _run_vector_matrix_multiplication;
-    bool                                      _run_addition;
-    bool                                      _is_first_run;
-    bool                                      _reshape_b_only_on_first_run;
+    MemoryGroup                _memory_group;
+    NEGEMMInterleave4x4Kernel  _interleave_kernel;
+    NEGEMMTranspose1xWKernel   _transpose_kernel;
+    NEGEMMMatrixMultiplyKernel _mm_kernel;
+    AssemblyKernelGlueF32      _asm_glue;
+    NEGEMMMatrixAdditionKernel _ma_kernel;
+    Tensor                     _tmp_a;
+    Tensor                     _tmp_b;
+    Tensor                     _workspace;
+    bool                       _run_vector_matrix_multiplication;
+    bool                       _run_addition;
+    bool                       _is_first_run;
+    bool                       _reshape_b_only_on_first_run;
 };
 }
 #endif /*__ARM_COMPUTE_NEGEMM_H__ */
diff --git a/arm_compute/runtime/NEON/functions/NEGEMMConvolutionLayer.h b/arm_compute/runtime/NEON/functions/NEGEMMConvolutionLayer.h
index ac5f4caa78..4ae8ee1fb3 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMMConvolutionLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMMConvolutionLayer.h
@@ -36,6 +36,7 @@
 #include "arm_compute/core/NEON/kernels/NEWeightsReshapeKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/MemoryGroup.h"
+#include "arm_compute/runtime/NEON/AssemblyHelper.h"
 #include "arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h"
 #include "arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h"
 #include "arm_compute/runtime/Tensor.h"
@@ -149,22 +150,14 @@ private:
      * @param[in]  reshape_info   (Optional) GEMM reshape info. If is_interleaved_transposed = true, this object must contain the information to understand how the matrix A and matrix B have been reshaped
      */
     void configure_mm(const ITensor *input, const ITensor *weights, ITensor *output, bool is_interleaved, const GEMMReshapeInfo &reshape_info = GEMMReshapeInfo());
-    /** Prepare the appropriate assembly optimized kernel
-     *
-     * @param[in] ci CPU information
-     * @param[in] M  M parameter of matrix multiplication
-     * @param[in] N  N parameter of matrix multiplication
-     * @param[in] K  K parameter of matrix multiplication
-     */
-    void configure_asm_mm(const struct CPUInfo &ci, int M, int N, int K);
 
 private:
+    AssemblyKernelGlueF32                               _asm_glue;
     MemoryGroup                                         _memory_group;
     NEIm2ColKernel                                      _input_im2col_kernel;
     NEGEMMInterleave4x4Kernel                           _input_interleave_kernel;
     NEConvolutionLayerReshapeWeights                    _reshape_weights;
     NEGEMMMatrixMultiplyKernel                          _mm_kernel;
-    std::unique_ptr<NEGEMMAssemblyBaseKernel>           _mm_optimised_kernel;
     NEGEMMLowpMatrixMultiplyCore                        _mm_gemmlowp;
     NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint _gemmlowp_output_stage;
     NECol2ImKernel                                      _output_col2im_kernel;
diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpAssemblyMatrixMultiplyCore.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpAssemblyMatrixMultiplyCore.h
index 3d213a7668..f09c94e726 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMMLowpAssemblyMatrixMultiplyCore.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpAssemblyMatrixMultiplyCore.h
@@ -1,6 +1,6 @@
 
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -29,6 +29,7 @@
 #include "arm_compute/runtime/IFunction.h"
 #include "arm_compute/runtime/IMemoryManager.h"
 #include "arm_compute/runtime/MemoryGroup.h"
+#include "arm_compute/runtime/NEON/AssemblyHelper.h"
 #include "arm_compute/runtime/Tensor.h"
 
 #include <memory>
@@ -58,6 +59,8 @@ public:
 
 private:
     MemoryGroup                _memory_group;
+    AssemblyKernelGlueU8U32    _asm_glue_unsigned;
+    AssemblyKernelGlueS8S32    _asm_glue_signed;
     std::unique_ptr<INEKernel> _mm_kernel;
     std::unique_ptr<INEKernel> _mtx_a_reshape_kernel;
     std::unique_ptr<INEKernel> _mtx_b_reshape_kernel;
diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
index eddb3a26b7..95776f829a 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -30,6 +30,7 @@
 #include "arm_compute/runtime/IFunction.h"
 #include "arm_compute/runtime/IMemoryManager.h"
 #include "arm_compute/runtime/MemoryGroup.h"
+#include "arm_compute/runtime/NEON/AssemblyHelper.h"
 #include "arm_compute/runtime/Tensor.h"
 
 #include <memory>
@@ -48,7 +49,6 @@ class ITensor;
  * otherwise if the DOT product instruction is available:
  *
  *  -# @ref NEGEMMInterleaveBlockedKernel
- *  -# @ref NEGEMMLowpAArch64V8P4Kernel
  *  -# @ref NEGEMMLowpOffsetContributionKernel
  *
 */
@@ -90,6 +90,8 @@ public:
 
 private:
     MemoryGroup                        _memory_group;
+    AssemblyKernelGlueU8U32            _asm_glue_unsigned;
+    AssemblyKernelGlueS8S32            _asm_glue_signed;
     std::unique_ptr<INEKernel>         _mm_kernel;
     std::unique_ptr<INEKernel>         _mtx_a_reshape_kernel;
     std::unique_ptr<INEKernel>         _mtx_b_reshape_kernel;