1 | initial version |
OpenCL path
Actually macro CV_OCL_RUN
returns from function if the call succeeded. There is also check for _matD.isUMat()
- if it is not then OpenCL branch will be skipped. If you compiled with -DWITH_OPENCL=OFF
then these functions will be stripped completely.
CPU path
Then _cv::hal::gemm32/64f_ is called which tries Lapack implementation first (or user-defined implementation via HAL replacement mechanism), then uses basic CPU implementation
So, if you build with -DWITH_LAPACK=ON
cmake option and have MKL or some other Lapack backend available, then gemm
should fall into that part.