Exhibits a weird behavior in Emgu.CV:
When targeting .NET 4.8/OpenCL and running on the GPU, CvInvoke.Pow(umat, 2, destination) gives results different from CvInvoke.Multiply(umat, umat, destination).
The different results are probably due to the GPU rounding behavior, however this difference does not appear:
- When targeting .NET 4.8 and running on the CPU
- When targeting .NET 4.8 but using the NVidia GPU
- When targeting .NET 6 (be it on the CPU or on Intel GPU or on NVidia GPU)
The values with .NET 6 are the same as .NET 4 on CPU.
.NET 4 with Intel GPU is the only outlier.
To sum it up, the weird behavior only appears when using .NET 4.8 and the Intel GPU. In all other cases (CPU vs GPU, NVidia vs Intel or .NET 6 vs .NET 4.8), the results are the same.
In order to witness the difference, build the example project with net48 target, then run it once without arguments (will generate the GPU results) and another time passing it --noopencl.
Compare the two generated files: they will be different.
PS: I witness all of this on a machine with an Intel UHD Graphics 3.0 card (driver version: 30.0.101.1003)