transform vector by transpose of matrix (assuming orthonormal)

check for OpenGL errors

unload the D3D10 DLLs

This function returns the best GPU (with maximum GFLOPS)

/////////////////////////////////////////////////////////////////////////// Helper functions /////////////////////////////////////////////////////////

Beginning of GPU Architecture definitions

Allocate device-preferred unified memory and return buffer with numpy view.

Create numpy view of cuda.core Buffer via DLPack.

Verify that the NVVM result code is success, or terminate otherwise.

/////////////////////////////////////////////////////////////////////////// Create VBO ///////////////////////////////////////////////////////////////

General initialization call to pick the best CUDA Device

signed random float [-1, 1]

CUDA Utility Helper Functions

Determine if two pixel colors are within tolerance

Extension checking utility

Create NvMediaImage surface based on the input attributes. * Returns NVMEDIA_STATUS_OK on success */

* |A| = max { |A|*ones(m,1) } */

Custom ReLU operation using cuda.core. This function provides a TensorFlow-native interface to custom CUDA kernels compiled with cuda.co

/////////////////////////////////////////////////////////////////////////// /////////////////////////////////////////////////////////////////////////

/////////////////////////////////////////////////////////////////////////// /////////////////////////////////////////////////////////////////////////