Update docs for semaphore-aware Sequence evalAsync overload

Author: cone-forest

docs/overview/advanced-examples.rst

@@ -162,6 +162,9 @@ Async/Await Example
 
 A simple example of asynchronous submission can be found below.
 
+You can also use async submissions with Vulkan semaphores to synchronize
+Kompute-generated submits with user-managed queue submits.
+
 First we are able to create the manager as we normally would.
 
 .. code-block:: cpp
@@ -233,15 +236,25 @@ The parameter provided is the maximum amount of time to wait in nanoseconds. Whe
 
     auto sq = mgr.sequence();
 
-    // Run Async Kompute operation on the parameters provided
-    sq->evalAsync<kp::OpAlgoDispatch>(algo);
+    // Optional: pass submit-level synchronization primitives so this submit
+    // waits/signals alongside user-managed queue work
+    std::vector<vk::Semaphore> waitSemaphores = { externalWaitSemaphore };
+    std::vector<vk::PipelineStageFlags> waitDstStageMasks = {
+        vk::PipelineStageFlagBits::eComputeShader
+    };
+    std::vector<vk::Semaphore> signalSemaphores = { externalSignalSemaphore };
+    auto opAlgo = std::make_shared<kp::OpAlgoDispatch>(algo);
+    sq->evalAsync(opAlgo, waitSemaphores, waitDstStageMasks, signalSemaphores);
 
     // Here we can do other work
 
-    // When we're ready we can wait 
+    // When we're ready we can wait
     // The default wait time is UINT64_MAX
     sq->evalAwait();
 
+``evalAwait()`` must be called before invoking ``evalAsync()`` again on the
+same ``Sequence``.
+
 
 Finally, below you can see that we can also run syncrhonous commands without having to change anything.
 

docs/overview/custom-operations.rst

@@ -33,7 +33,7 @@ Below you
    * - preEval()
      - When the Sequence is Evaluated this preEval is called across all operations before dispatching the batch of recorded commands to the GPU. This is useful for example if you need to copy data from local to host memory.
    * - postEval()
-     - After the sequence is Evaluated this postEval is called across all operations. When running asynchronously the postEval is called when you call `evalAwait()`, which is why it's important to always run evalAwait() to ensure the process doesn't go into inconsistent state.
+     - After the sequence is Evaluated this postEval is called across all operations. In asynchronous flows postEval is called when you run `evalAwait()`, and `evalAwait()` must be called before triggering `evalAsync()` again on the same sequence to avoid inconsistent state.
 
 
 Simple Operation Extending OpAlgoBase

src/include/kompute/Sequence.hpp

@@ -154,18 +154,22 @@ class Sequence : public std::enable_shared_from_this<Sequence>
 
     /**
      * Eval Async sends all the recorded and stored operations in the vector of
-     * operations into the gpu as a submit job without a barrier. EvalAwait()
-     * must ALWAYS be called after to ensure the sequence is terminated
-     * correctly.
+     * operations into the gpu as a submit job without a barrier.
+     *
+     * evalAwait() must be called before invoking evalAsync() again on this same
+     * Sequence to complete the previous async run and reset internal state.
      *
      * @return Boolean stating whether execution was successful.
      */
     std::shared_ptr<Sequence> evalAsync();
     /**
      * Eval Async sends all recorded operations as a submit job and allows
      * submit-level GPU synchronization by providing wait and signal semaphores.
-     * EvalAwait() must ALWAYS be called after to ensure the sequence is
-     * terminated correctly.
+     *
+     * This overload is useful for synchronizing Kompute submissions with
+     * user-managed queue submissions without forcing CPU-side synchronization.
+     * evalAwait() must be called before invoking evalAsync() again on this same
+     * Sequence to complete the previous async run and reset internal state.
      *
      * @param waitSemaphores Semaphores that must be signaled before this submit
      * starts executing.
@@ -182,18 +186,22 @@ class Sequence : public std::enable_shared_from_this<Sequence>
       const std::vector<vk::Semaphore>& signalSemaphores);
     /**
      * Clears currnet operations to record provided one in the vector of
-     * operations into the gpu as a submit job without a barrier. EvalAwait()
-     * must ALWAYS be called after to ensure the sequence is terminated
-     * correctly.
+     * operations into the gpu as a submit job without a barrier.
+     *
+     * evalAwait() must be called before invoking evalAsync() again on this same
+     * Sequence to complete the previous async run and reset internal state.
      *
      * @return Boolean stating whether execution was successful.
      */
     std::shared_ptr<Sequence> evalAsync(std::shared_ptr<OpBase> op);
     /**
      * Clears current operations, records the provided one and submits with
      * optional wait/signal semaphores for submit-level GPU synchronization.
-     * EvalAwait() must ALWAYS be called after to ensure the sequence is
-     * terminated correctly.
+     *
+     * This overload is useful for synchronizing Kompute submissions with
+     * user-managed queue submissions without forcing CPU-side synchronization.
+     * evalAwait() must be called before invoking evalAsync() again on this same
+     * Sequence to complete the previous async run and reset internal state.
      *
      * @param op Operation to record prior to submit.
      * @param waitSemaphores Semaphores that must be signaled before this submit