COMPMID-791: Generic Depthwise Convolution Layer NEON QASYMM8

Change-Id: I33cf54e68f6c097ac58b6f16c3f9a720978f09cd
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/117289
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>

1 files changed, 25 insertions, 4 deletions

diff --git a/arm_compute/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.h b/arm_compute/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.h
index 5ea83901f4..95fe916a3c 100644
--- a/arm_compute/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.h
+++ b/arm_compute/core/NEON/kernels/NEGEMMMatrixVectorMultiplyKernel.h
@@ -49,7 +49,7 @@ public:
     NEGEMMMatrixVectorMultiplyKernel &operator=(NEGEMMMatrixVectorMultiplyKernel &&) = default;
     /** Initialise the kernel's input and output.
      *
-     * @param[in]  input0 First Input tensor. Data types supported: F16/F32
+     * @param[in]  input0 First Input tensor. Data types supported: QASYMM8/F32
      * @param[in]  input1 Second Input tensor. Data types supported: same as @p input.
      * @param[out] output Output tensor which stores the interleaved matrix. Data type supported: same as @p input.
      */
@@ -57,11 +57,32 @@ public:
 
     // Inherited methods overridden:
     void run(const Window &window, const ThreadInfo &info) override;
+    BorderSize border_size() const override;
 
 private:
-    const ITensor *_input0;
-    const ITensor *_input1;
-    ITensor       *_output;
+    /** Template function to run the matrix vector multiplication
+     *
+     * @tparam I0 Input 0 type
+     * @tparam I1 Input 1 type
+     * @tparam O  Output type
+     *
+     * @param[in] window_in  Input region. (Must be a valid region of the window returned by window()).
+     * @param[in] window_w   Weights region. (Must be a valid region of the window returned by window()).
+     * @param[in] window_out Output region.(Must be a valid region of the window returned by window()).
+     */
+    template <typename I0, typename I1, typename O>
+    void matrix_vector_multiply(const Window &window_in, const Window &window_w, const Window &window_out);
+    /** Common signature for all the specialised matrix vector multiplication functions */
+    using GEMMMatrixVectorMultiplyFunctionPtr = void (NEGEMMMatrixVectorMultiplyKernel::*)(const Window &window_in,
+                                                                                           const Window &window_w,
+                                                                                           const Window &window_out);
+
+private:
+    GEMMMatrixVectorMultiplyFunctionPtr _func;
+    const ITensor                      *_input0;
+    const ITensor                      *_input1;
+    ITensor                            *_output;
+    BorderSize                          _border_size;
 };
 } // namespace arm_compute
 #endif /*__ARM_COMPUTE_NEGEMMMATRIXVECTORMULTIPLYKERNEL_H_*/