Compositor: Full frame Levels node

Adds full frame implementation to this node operations.
No functional changes.

Reviewed By: jbakker

Differential Revision: https://developer.blender.org/D11749

2 files changed, 29 insertions, 1 deletions

diff --git a/source/blender/compositor/intern/COM_ExecutionSystem.cc b/source/blender/compositor/intern/COM_ExecutionSystem.cc
index cd2738fc2c7..a449d86deb9 100644
--- a/source/blender/compositor/intern/COM_ExecutionSystem.cc
+++ b/source/blender/compositor/intern/COM_ExecutionSystem.cc
@@ -82,6 +82,7 @@ ExecutionSystem::ExecutionSystem(RenderData *rd,
       BLI_assert_msg(0, "Non implemented execution model");
       break;
   }
+  num_work_threads_ = WorkScheduler::get_num_cpu_threads();
 }
 
 ExecutionSystem::~ExecutionSystem()
@@ -130,7 +131,7 @@ void ExecutionSystem::execute_work(const rcti &work_rect,
 
   /* Split work vertically to maximize continuous memory. */
   const int work_height = BLI_rcti_size_y(&work_rect);
-  const int num_sub_works = MIN2(WorkScheduler::get_num_cpu_threads(), work_height);
+  const int num_sub_works = MIN2(num_work_threads_, work_height);
   const int split_height = num_sub_works == 0 ? 0 : work_height / num_sub_works;
   int remaining_height = work_height - split_height * num_sub_works;
 
diff --git a/source/blender/compositor/intern/COM_ExecutionSystem.h b/source/blender/compositor/intern/COM_ExecutionSystem.h
index 38c3432a8ec..ad6f9fd4bdb 100644
--- a/source/blender/compositor/intern/COM_ExecutionSystem.h
+++ b/source/blender/compositor/intern/COM_ExecutionSystem.h
@@ -31,6 +31,7 @@ class ExecutionGroup;
 #include "DNA_node_types.h"
 
 #include "BLI_vector.hh"
+#include "atomic_ops.h"
 
 namespace blender::compositor {
 
@@ -150,6 +151,11 @@ class ExecutionSystem {
    */
   ExecutionModel *execution_model_;
 
+  /**
+   * Number of cpu threads available for work execution.
+   */
+  int num_work_threads_;
+
   ThreadMutex work_mutex_;
   ThreadCondition work_finished_cond_;
 
@@ -201,6 +207,27 @@ class ExecutionSystem {
 
   void execute_work(const rcti &work_rect, std::function<void(const rcti &split_rect)> work_func);
 
+  /**
+   * Multi-threadedly execute given work function passing work_rect splits as argument. On
+   * finished, caller thread will call reduce_func for each thread result.
+   */
+  template<typename TResult>
+  void execute_work(const rcti &work_rect,
+                    std::function<TResult(const rcti &split_rect)> work_func,
+                    TResult &join,
+                    std::function<void(TResult &join, const TResult &chunk)> reduce_func)
+  {
+    Array<TResult> chunks(num_work_threads_);
+    int num_started = 0;
+    execute_work(work_rect, [&](const rcti &split_rect) {
+      const int current = atomic_fetch_and_add_int32(&num_started, 1);
+      chunks[current] = work_func(split_rect);
+    });
+    for (const int i : IndexRange(num_started)) {
+      reduce_func(join, chunks[i]);
+    }
+  }
+
   bool is_breaked() const;
 
  private: