MPSQUARK / BAVCL

Hardware-accelerated Vector Compute Library for .NET Containing Quality of life improvements and functionality intended for data science, graphical processing and GPGPU.
Other
10 stars 2 forks source link

added kernel caching inside the GPU class #2

Closed NullandKale closed 3 years ago

NullandKale commented 3 years ago

I moved all the kernels into into the GPU class, and precompiled them after we get the accelerator. This moves around 400ms of latency away from the first kernel operation and into constructing the GPU. Screenshot 2021-05-11 233847 Screenshot 2021-05-11 233958

I also changed a few memory allocations to reduce garbage generation.