[llvm] [mlir] Split the llvm::ThreadPool into an abstract base class and an impleme… (PR #82094)

[Alexandre Ganea via llvm-commits]

================
@@ -49,16 +46,31 @@ class ThreadPoolTaskGroup;
 /// available threads are used up by tasks waiting for a task that has no thread
 /// left to run on (this includes waiting on the returned future). It should be
 /// generally safe to wait() for a group as long as groups do not form a cycle.
-class ThreadPool {
+class ThreadPoolInterface {
+  // The actual method to enqueue a task to be defined by the concrete
+  // implementation.
+  virtual void asyncEnqueue(std::function<void()> Task,
+                            ThreadPoolTaskGroup *Group) = 0;
+
 public:
-  /// Construct a pool using the hardware strategy \p S for mapping hardware
-  /// execution resources (threads, cores, CPUs)
-  /// Defaults to using the maximum execution resources in the system, but
-  /// accounting for the affinity mask.
-  ThreadPool(ThreadPoolStrategy S = hardware_concurrency());
+  // Destroying the pool will drain the pending tasks and wait. The current
----------------
aganea wrote:

Same here?

https://github.com/llvm/llvm-project/pull/82094