* Takes a string like "64K" and "128M" and returns the corresponding size in * bytes, expanding the multiple prefixes to the actual size, like "655

* @brief Perform a scalar gather operation. * @param data The data buffer to gather from. * @param indices The indices used to gather data. * @para

* @brief Perform a scalar scatter operation. * @param data The buffer to scatter data into. * @param indices The indices used to scatter data. * @p

* @brief Runs the benchmark loop for precompiled CUDA C++ kernels using * the high-level @b runtime API. It counts TOPS (Tensor Operations Per *

* @brief Runs the benchmark loop for precompiled CUDA C++ kernels using * the low-level @b driver API. It counts TOPS (Tensor Operations Per * Se

* `yyjson` does not only allows to override the allocator, but also passes a context object down: * * void *(* malloc)(void *ctx, size_t size)