diff --git a/filament/src/Scene.cpp b/filament/src/Scene.cpp
index 541af9a74e..a9337bc3f3 100644
--- a/filament/src/Scene.cpp
+++ b/filament/src/Scene.cpp
@@ -28,6 +28,7 @@
 #include <utils/compiler.h>
 #include <utils/EntityManager.h>
 #include <utils/Range.h>
+#include <utils/Systrace.h>
 #include <utils/Zip2Iterator.h>
 
 #include <algorithm>
@@ -50,6 +51,8 @@ void FScene::prepare(const mat4f& worldOriginTransform) {
     // TODO: can we skip this in most cases? Since we rely on indices staying the same,
     //       we could only skip, if nothing changed in the RCM.
 
+    SYSTRACE_CALL();
+
     FEngine& engine = mEngine;
     EntityManager& em = engine.getEntityManager();
     FRenderableManager& rcm = engine.getRenderableManager();
diff --git a/filament/src/ShadowMap.cpp b/filament/src/ShadowMap.cpp
index 36e1f9f666..36594dd7af 100644
--- a/filament/src/ShadowMap.cpp
+++ b/filament/src/ShadowMap.cpp
@@ -27,6 +27,7 @@
 #include <backend/DriverEnums.h>
 
 #include <utils/debug.h>
+#include <utils/Systrace.h>
 
 #include <limits>
 
@@ -1061,6 +1062,8 @@ float ShadowMap::texelSizeWorldSpace(const mat4f& Wp, const mat4f& MbMtF) const
 template<typename Casters, typename Receivers>
 void ShadowMap::visitScene(const FScene& scene, uint32_t visibleLayers,
         Casters casters, Receivers receivers) noexcept {
+    SYSTRACE_CALL();
+
     using State = FRenderableManager::Visibility;
     FScene::RenderableSoa const& UTILS_RESTRICT soa = scene.getRenderableData();
     float3 const* const UTILS_RESTRICT worldAABBCenter = soa.data<FScene::WORLD_AABB_CENTER>();
diff --git a/filament/src/View.cpp b/filament/src/View.cpp
index 473193a4b0..016c306ea5 100644
--- a/filament/src/View.cpp
+++ b/filament/src/View.cpp
@@ -743,6 +743,7 @@ void FView::prepareVisibleRenderables(JobSystem& js,
 
 void FView::cullRenderables(JobSystem& js,
         FScene::RenderableSoa& renderableData, Frustum const& frustum, size_t bit) noexcept {
+    SYSTRACE_CALL();
 
     float3 const* worldAABBCenter = renderableData.data<FScene::WORLD_AABB_CENTER>();
     float3 const* worldAABBExtent = renderableData.data<FScene::WORLD_AABB_EXTENT>();
diff --git a/libs/utils/include/utils/JobSystem.h b/libs/utils/include/utils/JobSystem.h
index 31382bad6f..241d69922c 100644
--- a/libs/utils/include/utils/JobSystem.h
+++ b/libs/utils/include/utils/JobSystem.h
@@ -470,6 +470,7 @@ struct ParallelForJobData {
     }
 
     void parallelWithJobs(JobSystem& js, JobSystem::Job* parent) noexcept {
+        assert(parent);
 
         // We first split about the number of threads we have, and only then we split the rest
         // in a single thread (but execute the final cut in new jobs, see parallel() below),
diff --git a/libs/utils/src/JobSystem.cpp b/libs/utils/src/JobSystem.cpp
index fc2b8cfe48..46c719b6b2 100644
--- a/libs/utils/src/JobSystem.cpp
+++ b/libs/utils/src/JobSystem.cpp
@@ -253,8 +253,13 @@ void JobSystem::wake() noexcept {
     lock.lock();
     // this empty critical section is needed -- it guarantees that notifiy_all() happens
     // after the condition variables are set.
-    lock.unlock();
+
+    // We signal the condition inside the lock (which is not required), because this seems to
+    // yield to better scheduling on Android. When we signal outside the critical section,
+    // it looks like this thread gives its time slice to the waking thread and just sits there
+    // being runnable, but not running.
     mWaiterCondition.notify_all();
+    lock.unlock();
 }
 
 inline JobSystem::ThreadState& JobSystem::getState() noexcept {