Advanced (sarek.Spoc_core.Advanced)

8a. Unified Memory (CUDA Managed Memory)

module Unified : sig ... end

Unified memory allows automatic page migration between CPU and GPU. Note: Requires backend support - currently a placeholder API.

8b. Pinned/Mapped Memory

module Pinned : sig ... end

Pinned memory enables faster transfers and zero-copy access. Note: Requires backend support - currently a placeholder API.

8c. Async Kernel Launch with Futures

module type FUTURE = sig ... end

Future representing a pending async operation

module Future : sig ... end

val run_async : 
  Kernel.t ->
  Kernel.args ->
  _block:Spoc_framework.Framework_sig.dims ->
  _grid:Spoc_framework.Framework_sig.dims ->
  Device.t ->
  unit Future.t

Run a kernel asynchronously and return a future

8d. Device Memory Pools

module Pool : sig ... end

Memory pool for efficient allocation/deallocation

8e. Graph-based Execution (CUDA Graphs)

module Graph : sig ... end

Computation graph for capturing and replaying operations. Note: Full implementation requires deep backend integration.

Convenience Re-exports

val has_advanced_features : Device.t -> bool

Check if any advanced features are available on device