diff --git a/CMakeLists.txt b/CMakeLists.txt
index 9dbcfaf0..94a039d5 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -4,11 +4,15 @@ project(sfizz VERSION 1.0.0 LANGUAGES CXX)
 # Set the highest possible standard
 set(CMAKE_CXX_STANDARD 14)
 
+# Enable LTO
+set(CMAKE_POLICY_DEFAULT_CMP0069 NEW) # To override the policy in abseil and benchmark
+set(CMAKE_INTERPROCEDURAL_OPTIMIZATION TRUE)
+
 if (CMAKE_CXX_COMPILER_ID MATCHES "Clang" AND NOT ANDROID)
     add_compile_options(-stdlib=libc++)
     # Presumably need the above for linking too, maybe other options missing as well
     add_link_options(-stdlib=libc++)   # New command on CMake master, not in 3.12 release
-else()
+else() 
     add_link_options(-lstdc++fs)
 endif()
 
diff --git a/sfizz/CMakeLists.txt b/sfizz/CMakeLists.txt
index b040cc2a..aa12a72f 100644
--- a/sfizz/CMakeLists.txt
+++ b/sfizz/CMakeLists.txt
@@ -37,19 +37,22 @@ endif()
 set(SFIZZ_SOURCES ${SFIZZ_SOURCES} ${SFIZZ_SIMD_SOURCES})
 
 add_library(sfizz_parser STATIC)
-target_sources(sfizz_parser PUBLIC Parser.cpp Opcode.cpp)
+target_sources(sfizz_parser PRIVATE Parser.cpp Opcode.cpp)
 target_include_directories(sfizz_parser PUBLIC .)
 if(UNIX)
     target_link_libraries(sfizz_parser PUBLIC stdc++fs)
+    # target_compile_options(sfizz_parser PUBLIC -fno-rtti -fno-exceptions)
 endif(UNIX)
 target_link_libraries(sfizz_parser PRIVATE absl::strings)
 
 add_library(sfizz STATIC ${SFIZZ_SOURCES})
 target_link_libraries(sfizz PRIVATE sfizz_parser)
 target_include_directories(sfizz PUBLIC .)
+find_package(Threads REQUIRED)
+target_link_libraries(sfizz PRIVATE Threads::Threads)
 if(UNIX)
     target_link_libraries(sfizz PUBLIC stdc++fs atomic)
-    target_compile_options(sfizz PRIVATE -fno-rtti -fno-exceptions)
+    # target_compile_options(sfizz PUBLIC -fno-rtti -fno-exceptions)
 endif(UNIX)
 target_link_libraries(sfizz PUBLIC absl::strings)
 target_link_libraries(sfizz PRIVATE sndfile absl::flat_hash_map)
diff --git a/sfizz/FilePool.cpp b/sfizz/FilePool.cpp
index 9eaa35a2..0e1c31a2 100644
--- a/sfizz/FilePool.cpp
+++ b/sfizz/FilePool.cpp
@@ -45,7 +45,7 @@ std::unique_ptr<AudioBuffer<T>> readFromFile(SndfileHandle& sndFile, int numFram
         sndFile.readf(tempReadBuffer->channelWriter(0), numFrames);
         ::readInterleaved<float>(tempReadBuffer->getSpan(0), returnedBuffer->getSpan(0), returnedBuffer->getSpan(1));
     }
-    return std::move(returnedBuffer);
+    return returnedBuffer;
 }
 
 absl::optional<sfz::FilePool::FileInformation> sfz::FilePool::getFileInformation(const std::string& filename, uint32_t offset) noexcept
diff --git a/sfizz/FilePool.h b/sfizz/FilePool.h
index dad2e0f7..fb30c353 100644
--- a/sfizz/FilePool.h
+++ b/sfizz/FilePool.h
@@ -38,11 +38,7 @@
 namespace sfz {
 class FilePool {
 public:
-    FilePool()
-        : fileLoadingThread(std::thread(&FilePool::loadingThread, this))
-        , garbageCollectionThread(std::thread(&FilePool::garbageThread, this))
-    {
-    }
+    FilePool() { }
 
     ~FilePool()
     {
@@ -75,11 +71,11 @@ private:
     moodycamel::BlockingReaderWriterQueue<FileLoadingInformation> loadingQueue { config::numVoices };
     void loadingThread() noexcept;
     void garbageThread() noexcept;
-    std::thread fileLoadingThread;
-    std::thread garbageCollectionThread;
+    bool quitThread { false };
+    std::thread fileLoadingThread { &FilePool::loadingThread, this };
+    std::thread garbageCollectionThread { &FilePool::garbageThread, this };
     std::vector<std::shared_ptr<AudioBuffer<float>>> fileHandles;
     std::mutex fileHandleMutex;
-    bool quitThread { false };
     absl::flat_hash_map<absl::string_view, std::shared_ptr<AudioBuffer<float>>> preloadedData;
     LEAK_DETECTOR(FilePool);
 };
diff --git a/sfizz/Synth.cpp b/sfizz/Synth.cpp
index 4f9875bc..a8f5a1ec 100644
--- a/sfizz/Synth.cpp
+++ b/sfizz/Synth.cpp
@@ -356,12 +356,14 @@ void sfz::Synth::noteOn(int delay, int channel, int noteNumber, uint8_t velocity
     }
 }
 
-void sfz::Synth::noteOff(int delay, int channel, int noteNumber, uint8_t velocity) noexcept
 {
     ASSERT(noteNumber < 128);
     ASSERT(noteNumber >= 0);
 
-    auto replacedVelocity = velocity == 0 ? midiState.getNoteVelocity(noteNumber) : velocity;
+    // FIXME: Some keyboards (e.g. Casio PX5S) can send a real note-off velocity. In this case, do we have a 
+    // way in sfz to specify that a release trigger should NOT use the note-on velocity?
+    // auto replacedVelocity = (velocity == 0 ? sfz::getNoteVelocity(noteNumber) : velocity);
+    auto replacedVelocity = midiState.getNoteVelocity(noteNumber);
     auto randValue = randNoteDistribution(Random::randomGenerator);
     for (auto& voice : voices)
         voice->registerNoteOff(delay, channel, noteNumber, replacedVelocity);
diff --git a/sfizz/Voice.cpp b/sfizz/Voice.cpp
index 8beee826..444e46b0 100644
--- a/sfizz/Voice.cpp
+++ b/sfizz/Voice.cpp
@@ -365,8 +365,8 @@ void sfz::Voice::fillWithData(AudioSpan<float> buffer) noexcept
     ::add<int>(sourcePosition, indices);
 
     //FIXME : all this casting is driving me crazy
-    const auto sampleEnd = static_cast<int>(region->trueSampleEnd()) - 1;
-    if (region->shouldLoop() && static_cast<size_t>(sampleEnd) <= source.getNumFrames()) {
+    const auto sampleEnd = min(static_cast<int>(region->trueSampleEnd()), static_cast<int>(source.getNumFrames())) - 1;
+    if (region->shouldLoop() && region->loopRange.getEnd() <= source.getNumFrames()) {
         const auto offset = sampleEnd - static_cast<int>(region->loopRange.getStart());
         for (auto* index = indices.begin(); index < indices.end(); ++index) {
             if (*index > sampleEnd) {
diff --git a/tests/SIMDHelpersT.cpp b/tests/SIMDHelpersT.cpp
index 4fd8075f..138df243 100644
--- a/tests/SIMDHelpersT.cpp
+++ b/tests/SIMDHelpersT.cpp
@@ -528,7 +528,7 @@ TEST_CASE("[Helpers] Linear Ramp (SIMD)")
     std::array<float, 6> output;
     std::array<float, 6> expected { v, v + v, v + v + v, v + v + v + v, v + v + v + v + v, v + v + v + v + v + v };
     linearRamp<float, true>(absl::MakeSpan(output), start, v);
-    REQUIRE(output == expected);
+    REQUIRE(approxEqual<float>(output, expected));
 }
 
 TEST_CASE("[Helpers] Linear Ramp (SIMD vs scalar)")