Class ClMatMul

src/gpu/cl/operators/ClMatMul.h:43–98

Basic operator to execute BatchMatMul on OpenCL. This operator calls the following OpenCL kernels: * * -# @ref kernels::ClMatMulNativeKernel */

 *  -# @ref kernels::ClMatMulNativeKernel
 */
class ClMatMul : public IClOperator
{
public:
    /** Constructor */
    ClMatMul();
    /** Default destructor */
    ~ClMatMul() = default;
    /** Initialise the kernel's inputs and output
     *
     * Valid data layouts:
     * - All
     *
     * Valid data type configurations:
     * |lhs            |rhs            |dst            |
     * |:--------------|:--------------|:--------------|
     * |F32            |F32            |F32            |
     * |F16            |F16            |F16            |
     * |QASYMM8_SIGNED |QASYMM8_SIGNED |QASYMM8_SIGNED |
     * |QASYMM8        |QASYMM8        |QASYMM8        |
     *
     * @note BatchMatMul: Batched Matrix Multiply - [A * B], Multiplies all slices (slice is an element of a batch) of Tensors A and B
     *                    and stores the result in the dst tensor of the same batch size.
     *                    Batch here is number of slices from A and B multiplied at a time, do not confuse with the batch dimension 'N' of NHWC/NCHW
     *                    For NHWC for example: the batch is the higher dimensions H * N, and in general it is H * all higher dimensions.
     * @note All tensors must have the same data type.
     *
     * @param[in]  compile_context The compile context to be used.
     * @param[in]  lhs             Left-hand side tensor info. Data types supported: F16/F32/QASYMM8_SIGNED/QASYMM8.
     * @param[in]  rhs             Right-hand side tensor info. Data types supported: same as @p lhs.
     * @param[out] dst             Output tensor to store the result of the batched matrix multiplication. Data types supported: same as @p lhs.
     * @param[in]  matmul_info     Contains MatMul operation information described in @ref MatMulInfo.
     * @param[in]  act_info        Class containing information about fused activation function.
     */
    void configure(const CLCompileContext    &compile_context,
                   ITensorInfo               *lhs,
                   ITensorInfo               *rhs,
                   ITensorInfo               *dst,
                   const MatMulInfo          &matmul_info,
                   const ActivationLayerInfo &act_info = ActivationLayerInfo());
    /** Static function to check if given info will lead to a valid configuration
     *
     * Similar to @ref ClMatMul::configure()
     *
     * @return a status
     */
    static Status validate(const ITensorInfo         *lhs,
                           const ITensorInfo         *rhs,
                           const ITensorInfo         *dst,
                           const MatMulInfo          &matmul_info,
                           const ActivationLayerInfo &act_info = ActivationLayerInfo());
    // Inherited methods overridden:
    void run(ITensorPack &tensors) override;

private:
    std::unique_ptr<opencl::IClKernel> _matmul_kernel{nullptr};
};
} // namespace opencl
} // namespace arm_compute