* @brief Executes operators on a CUDA-enabled device (non-JIT path) * * Optionally takes a stream for asynchronous execution. * This executo

* Parameters needed to execute a recursive filter. */

* @brief Executes operators on a CUDA-enabled device using JIT compilation * * This executor uses NVRTC (NVIDIA Runtime Compilation) to JIT-com

Common cache parameters that every cache entry needs

Pure NVRTC implementation Cache both module and function to prevent resource leaks CUmodule must remain loaded for CUfunction to be valid

Enum to define how a capability query is propagated/aggregated

* Test covariance parameters for equality. Unlike the hash, all parameters must * match. */

* Crude hash on cpv to get a reasonably good delta for collisions. This doesn't * need to be perfect, but fast enough to not slow down lookups, and d