* \brief Allocate from a pool * * This is an alternate spelling for cudaMallocFromPoolAsync * made available through operator overloading. * * \s

* \brief \hl Allocates page-locked memory on the host * * Allocates \p size bytes of host memory that is page-locked and accessible * to the device

* \ingroup CUDART_GRAPH * \brief Get the currently running device graph id. * * Get the currently running device graph id. * \return Returns t

convenience function to avoid source breakage in c++ code

* \ingroup CUDART_EXECUTION * \brief Programmatic grid dependency synchronization * * This device function will block the thread until all direc

* \ingroup CUDART_EXECUTION * \brief Programmatic dependency trigger * * This device function ensures the programmatic launch completion edges /

only returns [Some] for exact division

Formula and coefficients from <https://en.wikipedia.org/wiki/Error_function#Numerical_approximations> (Abramowitz and Stegun), Max error `3e-7`.

TODO variant that extracts the entire subgraph up to a given set of values, keeping the exact same inputs? Extract a small subgraph consisting of all

Returns the first available cuda device if any.

* Type cast to \p float operator. */