/////////////////////////////////////////////////////////////////////////// Helper function, returning uniformly distributed random float in [low, hig

/////////////////////////////////////////////////////////////////////////// Common functions /////////////////////////////////////////////////////////

render Mandelbrot image using CUDA or CPU

/////////////////////////////////////////////////////////////////////////// render a simple 3D scene /////////////////////////////////////////////////

step the simulation

random signed vector

step the simulation

convenience accessors

Polynomial approximation of cumulative normal distribution function

/////////////////////////////////////////////////////////////////////////// Polynomial approximation of cumulative normal distribution function //////

////////////////////////////////////////////////////////////////////////// Polynomial approximation of cumulative normal distribution function ///////

Checks if the shader is compiled.

////////////////////////////////////////////////////////////////////////// \brief resize image \param[in] src image to downscale \param[in]

////////////////////////////////////////////////////////////////////////// \brief load 4-channel unsigned byte image and convert it to single channel

Allocates an array with random float entries.

Allocates an array with random float entries.

Allocates an array with random float entries.

************************************************************************** * Performs DCT of vector of 8 elements. * * \param FirstIn [IN] - P

************************************************************************** * Performs IDCT of vector of 8 elements. * * \param FirstIn [IN] -

////////////////////////////////////////////////////////////////////////// \brief host texture fetch read from arbitrary position within image using

////////////////////////////////////////////////////////////////////////// \brief upscale one component of a displacement field \param[in] src

////////////////////////////////////////////////////////////////////////// \brief save optical flow in format described on vision.middlebury.edu/flow

Probe the driver for a (long, critical) split it actually accepts. Escalates the alignment granularity from `min_partition_size` upward in

Get or compile the ReLU kernels for a given device. Parameters ---------- device : Device CUDA device object Returns

Launch the delay kernel on `long_stream`, wait `launch_gap_s` on the host, launch the critical kernel on `critical_stream`, and time both wit

Attach shader to a program

Backward pass: compute gradient. For y = x², the derivative is dy/dx = 2x Therefore: grad_x = grad_output * 2x Para

render scene depth to texture (this is to ensure that particle are correctly occluded in the low-resolution render buffer)

Benchmark a 1D kernel with given threads-per-block configuration. Uses CUDA events for accurate GPU timing. Returns timing statistics as

bind to the FrameBuffer Object

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

************************************************************************** * The routine clamps the input value to integer byte range [0, 255] * * \p

Global cleanup function Shared by both GL and non-GL code paths