TiledDisplay: skip transferring to host if all displays are on the rendering device

display/lava-extras/display/TiledDisplay.cc

@@ -5,6 +5,8 @@
 #include <lava/objects/Device.hh>
 #include <lava/objects/Image.hh>
 #include <lava/objects/MemoryChunk.hh>
+#include <thread>
+#include <iostream>
 
 namespace lava {
 
@@ -52,8 +54,13 @@ void TiledDisplay::submit(const SharedImage &image) {
         throw std::logic_error(
             "For now you should present to the original device, too");
 
+    // if true, we can skip transferring to other devices
+    bool all_master = std::all_of(begin(mConfig.tiles), end(mConfig.tiles),
+        [&](TileConfig const& cfg) { return cfg.device == master_idx; });
+
     // Step 1 - GPUs-Side: Work with mHostImages[gpu_side]
 
+    if (!all_master) {
         { // Step 1.1 transfer master image to Host
             auto cmd = mDevices[master_idx]->graphicsQueue().beginCommandBuffer();
             mHostImages[master_idx][gpu_side]->copyFrom(image, cmd);
@@ -70,6 +77,8 @@ void TiledDisplay::submit(const SharedImage &image) {
                     vk::ImageLayout::eTransferSrcOptimal);
             }
         }
+    }
+
     // Step 1.3 blit and present host images
     for (size_t i = 0; i < mConfig.tiles.size(); i++) {
         auto const &tile = mConfig.tiles[i];
@@ -110,6 +119,9 @@ void TiledDisplay::submit(const SharedImage &image) {
                                         vk::ImageLayout::ePresentSrcKHR, cmd);
         }
     }
+
+    auto cpu_side = (gpu_side + 1) % 2;
+    if (!all_master) {
         { // Step 1.4 change (non-master) hostImages layout back to general
             for (size_t i = 0; i < mDevices.size(); i++) {
                 if (i == master_idx)
@@ -123,7 +135,6 @@ void TiledDisplay::submit(const SharedImage &image) {
                 mDevices[i]->graphicsQueue().catchUp(1);
 
         // Step 2: Exchange Data between hostImages
-    auto cpu_side = (gpu_side + 1) % 2;
         {
             auto master_map =
                 mHostImages[master_idx][cpu_side]->memoryChunk()->map();
@@ -136,7 +147,7 @@ void TiledDisplay::submit(const SharedImage &image) {
                        mHostImages[master_idx][0]->levelBytes(0));
             }
         }
-
+    }
     gpu_side = cpu_side;
 }