Compositor: Full frame Ellipse Mask node

Adds full frame implementation to this node operation.
No functional changes.
3x times faster than tiled fallback.

Reviewed By: jbakker

Differential Revision: https://developer.blender.org/D11635

source/blender/compositor/operations/COM_EllipseMaskOperation.cc

@@ -20,6 +20,8 @@
 #include "BLI_math.h"
 #include "DNA_node_types.h"
 
+#include <functional>
+
 namespace blender::compositor {
 
 EllipseMaskOperation::EllipseMaskOperation()
@@ -114,6 +116,77 @@ void EllipseMaskOperation::executePixelSampled(float output[4],
   }
 }
 
+void EllipseMaskOperation::update_memory_buffer_partial(MemoryBuffer *output,
+                                                        const rcti &area,
+                                                        Span<MemoryBuffer *> inputs)
+{
+  MaskFunc mask_func;
+  switch (m_maskType) {
+    case CMP_NODE_MASKTYPE_ADD:
+      mask_func = [](const bool is_inside, const float *mask, const float *value) {
+        return is_inside ? MAX2(mask[0], value[0]) : mask[0];
+      };
+      break;
+    case CMP_NODE_MASKTYPE_SUBTRACT:
+      mask_func = [](const bool is_inside, const float *mask, const float *value) {
+        return is_inside ? CLAMPIS(mask[0] - value[0], 0, 1) : mask[0];
+      };
+      break;
+    case CMP_NODE_MASKTYPE_MULTIPLY:
+      mask_func = [](const bool is_inside, const float *mask, const float *value) {
+        return is_inside ? mask[0] * value[0] : 0;
+      };
+      break;
+    case CMP_NODE_MASKTYPE_NOT:
+      mask_func = [](const bool is_inside, const float *mask, const float *value) {
+        if (is_inside) {
+          return mask[0] > 0.0f ? 0.0f : value[0];
+        }
+        return mask[0];
+      };
+      break;
+  }
+  apply_mask(output, area, inputs, mask_func);
+}
+
+void EllipseMaskOperation::apply_mask(MemoryBuffer *output,
+                                      const rcti &area,
+                                      Span<MemoryBuffer *> inputs,
+                                      MaskFunc mask_func)
+{
+  const MemoryBuffer *input_mask = inputs[0];
+  const MemoryBuffer *input_value = inputs[1];
+  const float op_w = this->getWidth();
+  const float op_h = this->getHeight();
+  const float half_w = this->m_data->width / 2.0f;
+  const float half_h = this->m_data->height / 2.0f;
+  const float tx = half_w * half_w;
+  const float ty = half_h * half_h;
+  for (const int y : YRange(area)) {
+    const float op_ry = y / op_h;
+    const float dy = (op_ry - this->m_data->y) / m_aspectRatio;
+    float *out = output->get_elem(area.xmin, y);
+    const float *mask = input_mask->get_elem(area.xmin, y);
+    const float *value = input_value->get_elem(area.xmin, y);
+    for (const int x : XRange(area)) {
+      const float op_rx = x / op_w;
+      const float dx = op_rx - this->m_data->x;
+      const float rx = this->m_data->x + (m_cosine * dx + m_sine * dy);
+      const float ry = this->m_data->y + (-m_sine * dx + m_cosine * dy);
+      float sx = rx - this->m_data->x;
+      sx *= sx;
+      float sy = ry - this->m_data->y;
+      sy *= sy;
+      const bool inside = ((sx / tx) + (sy / ty)) < 1.0f;
+      out[0] = mask_func(inside, mask, value);
+
+      mask += input_mask->elem_stride;
+      value += input_value->elem_stride;
+      out += output->elem_stride;
+    }
+  }
+}
+
 void EllipseMaskOperation::deinitExecution()
 {
   this->m_inputMask = nullptr;

source/blender/compositor/operations/COM_EllipseMaskOperation.h

@@ -18,12 +18,14 @@
 
 #pragma once
 
-#include "COM_NodeOperation.h"
+#include "COM_MultiThreadedOperation.h"
 
 namespace blender::compositor {
 
-class EllipseMaskOperation : public NodeOperation {
+class EllipseMaskOperation : public MultiThreadedOperation {
  private:
+  using MaskFunc = std::function<float(bool is_inside, const float *mask, const float *value)>;
+
   /**
    * Cached reference to the inputProgram
    */
@@ -64,6 +66,16 @@ class EllipseMaskOperation : public NodeOperation {
   {
     this->m_maskType = maskType;
   }
+
+  void update_memory_buffer_partial(MemoryBuffer *output,
+                                    const rcti &area,
+                                    Span<MemoryBuffer *> inputs) override;
+
+ private:
+  void apply_mask(MemoryBuffer *output,
+                  const rcti &area,
+                  Span<MemoryBuffer *> inputs,
+                  MaskFunc mask_func);
 };
 
 }  // namespace blender::compositor