+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Allocate a chunk of memory in the device.
+ *
+ * @param dev_id
+ * Device ID requiring allocated memory.
+ * @param size
+ * Number of bytes to allocate.
+ * Requesting 0 will do nothing.
+ * @param align
+ * If 0, the return is a pointer that is suitably aligned
+ * for any kind of variable (in the same manner as malloc()).
+ * Otherwise, the return is a pointer that is a multiple of *align*.
+ * In this case, it must obviously be a power of two.
+ *
+ * @return
+ * A pointer to the allocated memory, otherwise NULL and rte_errno is set:
+ * - ENODEV if invalid dev_id
+ * - EINVAL if align is not a power of two
+ * - ENOTSUP if operation not supported by the driver
+ * - E2BIG if size is higher than limit
+ * - ENOMEM if out of space
+ * - EPERM if driver error
+ */
+__rte_experimental
+void *rte_gpu_mem_alloc(int16_t dev_id, size_t size, unsigned int align)
+__rte_alloc_size(2);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Deallocate a chunk of memory allocated with rte_gpu_mem_alloc().
+ *
+ * @param dev_id
+ * Reference device ID.
+ * @param ptr
+ * Pointer to the memory area to be deallocated.
+ * NULL is a no-op accepted value.
+ *
+ * @return
+ * 0 on success, -rte_errno otherwise:
+ * - ENODEV if invalid dev_id
+ * - ENOTSUP if operation not supported by the driver
+ * - EPERM if driver error
+ */
+__rte_experimental
+int rte_gpu_mem_free(int16_t dev_id, void *ptr);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Register a chunk of memory on the CPU usable by the device.
+ *
+ * @param dev_id
+ * Device ID requiring allocated memory.
+ * @param size
+ * Number of bytes to allocate.
+ * Requesting 0 will do nothing.
+ * @param ptr
+ * Pointer to the memory area to be registered.
+ * NULL is a no-op accepted value.
+
+ * @return
+ * A pointer to the allocated memory, otherwise NULL and rte_errno is set:
+ * - ENODEV if invalid dev_id
+ * - EINVAL if reserved flags
+ * - ENOTSUP if operation not supported by the driver
+ * - E2BIG if size is higher than limit
+ * - ENOMEM if out of space
+ * - EPERM if driver error
+ */
+__rte_experimental
+int rte_gpu_mem_register(int16_t dev_id, size_t size, void *ptr);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Deregister a chunk of memory previously registered with rte_gpu_mem_register()
+ *
+ * @param dev_id
+ * Reference device ID.
+ * @param ptr
+ * Pointer to the memory area to be unregistered.
+ * NULL is a no-op accepted value.
+ *
+ * @return
+ * 0 on success, -rte_errno otherwise:
+ * - ENODEV if invalid dev_id
+ * - ENOTSUP if operation not supported by the driver
+ * - EPERM if driver error
+ */
+__rte_experimental
+int rte_gpu_mem_unregister(int16_t dev_id, void *ptr);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Enforce a GPU write memory barrier.
+ *
+ * @param dev_id
+ * Reference device ID.
+ *
+ * @return
+ * 0 on success, -rte_errno otherwise:
+ * - ENODEV if invalid dev_id
+ * - ENOTSUP if operation not supported by the driver
+ * - EPERM if driver error
+ */
+__rte_experimental
+int rte_gpu_wmb(int16_t dev_id);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Create a communication flag that can be shared
+ * between CPU threads and device workload to exchange some status info
+ * (e.g. work is done, processing can start, etc..).
+ *
+ * @param dev_id
+ * Reference device ID.
+ * @param devflag
+ * Pointer to the memory area of the devflag structure.
+ * @param mtype
+ * Type of memory to allocate the communication flag.
+ *
+ * @return
+ * 0 on success, -rte_errno otherwise:
+ * - ENODEV if invalid dev_id
+ * - EINVAL if invalid inputs
+ * - ENOTSUP if operation not supported by the driver
+ * - ENOMEM if out of space
+ * - EPERM if driver error
+ */
+__rte_experimental
+int rte_gpu_comm_create_flag(uint16_t dev_id,
+ struct rte_gpu_comm_flag *devflag,
+ enum rte_gpu_comm_flag_type mtype);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Deallocate a communication flag.
+ *
+ * @param devflag
+ * Pointer to the memory area of the devflag structure.
+ *
+ * @return
+ * 0 on success, -rte_errno otherwise:
+ * - ENODEV if invalid dev_id
+ * - EINVAL if NULL devflag
+ * - ENOTSUP if operation not supported by the driver
+ * - EPERM if driver error
+ */
+__rte_experimental
+int rte_gpu_comm_destroy_flag(struct rte_gpu_comm_flag *devflag);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Set the value of a communication flag as the input value.
+ * Flag memory area is treated as volatile.
+ * The flag must have been allocated with RTE_GPU_COMM_FLAG_CPU.
+ *
+ * @param devflag
+ * Pointer to the memory area of the devflag structure.
+ * @param val
+ * Value to set in the flag.
+ *
+ * @return
+ * 0 on success, -rte_errno otherwise:
+ * - EINVAL if invalid input params
+ */
+__rte_experimental
+int rte_gpu_comm_set_flag(struct rte_gpu_comm_flag *devflag,
+ uint32_t val);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Get the value of the communication flag.
+ * Flag memory area is treated as volatile.
+ * The flag must have been allocated with RTE_GPU_COMM_FLAG_CPU.
+ *
+ * @param devflag
+ * Pointer to the memory area of the devflag structure.
+ * @param val
+ * Flag output value.
+ *
+ * @return
+ * 0 on success, -rte_errno otherwise:
+ * - EINVAL if invalid input params
+ */
+__rte_experimental
+int rte_gpu_comm_get_flag_value(struct rte_gpu_comm_flag *devflag,
+ uint32_t *val);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Create a communication list that can be used to share packets
+ * between CPU and device.
+ * Each element of the list contains:
+ * - a packet list of RTE_GPU_COMM_LIST_PKTS_MAX elements
+ * - number of packets in the list
+ * - a status flag to communicate if the packet list is FREE,
+ * READY to be processed, DONE with processing.
+ *
+ * The list is allocated in CPU-visible memory.
+ * At creation time, every list is in FREE state.
+ *
+ * @param dev_id
+ * Reference device ID.
+ * @param num_comm_items
+ * Number of items in the communication list.
+ *
+ * @return
+ * A pointer to the allocated list, otherwise NULL and rte_errno is set:
+ * - EINVAL if invalid input params
+ */
+__rte_experimental
+struct rte_gpu_comm_list *rte_gpu_comm_create_list(uint16_t dev_id,
+ uint32_t num_comm_items);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Destroy a communication list.
+ *
+ * @param comm_list
+ * Communication list to be destroyed.
+ * @param num_comm_items
+ * Number of items in the communication list.
+ *
+ * @return
+ * 0 on success, -rte_errno otherwise:
+ * - EINVAL if invalid input params
+ */
+__rte_experimental
+int rte_gpu_comm_destroy_list(struct rte_gpu_comm_list *comm_list,
+ uint32_t num_comm_items);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Populate the packets list of the communication item
+ * with info from a list of mbufs.
+ * Status flag of that packet list is set to READY.
+ *
+ * @param comm_list_item
+ * Communication list item to fill.
+ * @param mbufs
+ * List of mbufs.
+ * @param num_mbufs
+ * Number of mbufs.
+ *
+ * @return
+ * 0 on success, -rte_errno otherwise:
+ * - EINVAL if invalid input params
+ * - ENOTSUP if mbufs are chained (multiple segments)
+ */
+__rte_experimental
+int rte_gpu_comm_populate_list_pkts(struct rte_gpu_comm_list *comm_list_item,
+ struct rte_mbuf **mbufs, uint32_t num_mbufs);
+
+/**
+ * @warning
+ * @b EXPERIMENTAL: this API may change without prior notice.
+ *
+ * Reset a communication list item to the original state.
+ * The status flag set to FREE and mbufs are returned to the pool.
+ *
+ * @param comm_list_item
+ * Communication list item to reset.
+ *
+ * @return
+ * 0 on success, -rte_errno otherwise:
+ * - EINVAL if invalid input params
+ */
+__rte_experimental
+int rte_gpu_comm_cleanup_list(struct rte_gpu_comm_list *comm_list_item);
+