FastCV extension 3rd Post

[adsha-quic]

Adding FastCV extensions for merge, split, gemm and arithm APIs add, subtract

Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

• I agree to contribute to the project under Apache 2 License.
• To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
• The PR is proposed to the proper branch
• There is a reference to the original bug report and related work
• There is accuracy test, performance test and test data in opencv_extra repository, if applicable
  Patch to opencv_extra has the same branch name.
• The feature is well documented and sample code can be built with the project CMake

[asmorkalov]

OpenCV HAL has GEMM options:

/**
The function performs generalized matrix multiplication similar to the gemm functions in BLAS level 3:
\f$D = \alpha*AB+\beta*C\f$

@param src1 pointer to input \f$M\times N\f$ matrix \f$A\f$ or \f$A^T\f$ stored in row major order.
@param src1_step number of bytes between two consequent rows of matrix \f$A\f$ or \f$A^T\f$.
@param src2 pointer to input \f$N\times K\f$ matrix \f$B\f$ or \f$B^T\f$ stored in row major order.
@param src2_step number of bytes between two consequent rows of matrix \f$B\f$ or \f$B^T\f$.
@param alpha \f$\alpha\f$ multiplier before \f$AB\f$
@param src3 pointer to input \f$M\times K\f$ matrix \f$C\f$ or \f$C^T\f$ stored in row major order.
@param src3_step number of bytes between two consequent rows of matrix \f$C\f$ or \f$C^T\f$.
@param beta \f$\beta\f$ multiplier before \f$C\f$
@param dst pointer to input \f$M\times K\f$ matrix \f$D\f$ stored in row major order.
@param dst_step number of bytes between two consequent rows of matrix \f$D\f$.
@param m number of rows in matrix \f$A\f$ or \f$A^T\f$, equals to number of rows in matrix \f$D\f$
@param n number of columns in matrix \f$A\f$ or \f$A^T\f$
@param k number of columns in matrix \f$B\f$ or \f$B^T\f$, equals to number of columns in matrix \f$D\f$
@param flags algorithm options (combination of CV_HAL_GEMM_1_T, ...).
 */

//! @addtogroup core_hal_interface_matrix_multiplication Matrix multiplication
//! @{
inline int hal_ni_gemm32f(const float* src1, size_t src1_step, const float* src2, size_t src2_step,
                          float alpha, const float* src3, size_t src3_step, float beta, float* dst, size_t dst_step,
                          int m, int n, int k, int flags) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_gemm64f(const double* src1, size_t src1_step, const double* src2, size_t src2_step,
                          double alpha, const double* src3, size_t src3_step, double beta, double* dst, size_t dst_step,
                          int m, int n, int k, int flags) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_gemm32fc(const float* src1, size_t src1_step, const float* src2, size_t src2_step,
                          float alpha, const float* src3, size_t src3_step, float beta, float* dst, size_t dst_step,
                          int m, int n, int k, int flags) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_gemm64fc(const double* src1, size_t src1_step, const double* src2, size_t src2_step,
                          double alpha, const double* src3, size_t src3_step, double beta, double* dst, size_t dst_step,
                          int m, int n, int k, int flags) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
//! @}

I propose to implement rather then add extension.

[asmorkalov]

The same question on HAL.

[adsha-quic]

We are trying optimizations to achieve consistent perf across various targets. So for now we added it in extension