code size optimizations

- prevent unrolling of some loops. e.g. one loop generated more than
  1 KB of code!

filament/src/VertexBuffer.cpp

@@ -134,6 +134,7 @@ FVertexBuffer::FVertexBuffer(FEngine& engine, const VertexBuffer::Builder& build
 
     auto const& declaredAttributes = mDeclaredAttributes;
     auto const& attributes = mAttributes;
+    #pragma nounroll
     for (size_t i = 0, n = attributeArray.size(); i < n; ++i) {
         if (declaredAttributes[i]) {
             attributeArray[i].offset = attributes[i].offset;

filament/src/driver/opengl/OpenGLDriver.cpp

@@ -401,6 +401,7 @@ void OpenGLDriver::unbindTexture(GLenum target, GLuint texture_id) noexcept {
 
 void OpenGLDriver::unbindSampler(GLuint sampler) noexcept {
     // unbind this sampler from all the units it might be bound to
+    #pragma nounroll    // clang generates >800B of code!!!
     for (GLuint unit = 0; unit < MAX_TEXTURE_UNITS; unit++) {
         if (state.textures.units[unit].sampler == sampler) {
             bindSampler(unit, 0);
@@ -1260,6 +1261,7 @@ void OpenGLDriver::destroyVertexBuffer(Driver::VertexBufferHandle vbh) {
         // bindings of bound buffers are reset to 0
         const size_t targetIndex = getIndexForBufferTarget(GL_ARRAY_BUFFER);
         auto& target = state.buffers.genericBinding[targetIndex];
+        #pragma nounroll
         for (GLuint b : eb->gl.buffers) {
             if (target == b) {
                 target = 0;
@@ -1326,6 +1328,8 @@ void OpenGLDriver::destroyUniformBuffer(Driver::UniformBufferHandle ubh) {
         // bindings of bound buffers are reset to 0
         const size_t targetIndex = getIndexForBufferTarget(GL_UNIFORM_BUFFER);
         auto& target = state.buffers.targets[targetIndex];
+
+        #pragma nounroll // clang generates >1 KiB of code!!
         for (auto& buffer : target.buffers) {
             if (buffer.name == ub->gl.ubo.id) {
                 buffer.name = 0;

libs/filabridge/include/filament/SamplerBindingMap.h

@@ -19,8 +19,8 @@
 
 #include <filament/EngineEnums.h>
 
+#include <tsl/robin_map.h>
 #include <vector>
-#include <unordered_map>
 
 namespace filament {
 
@@ -89,7 +89,7 @@ private:
         return ((uint32_t) blockIndex << 8) + localOffset;
     }
     std::vector<SamplerBindingInfo> mBindingList;
-    std::unordered_map<BindingKey, SamplerBindingInfo> mBindingMap;
+    tsl::robin_map<BindingKey, SamplerBindingInfo> mBindingMap;
     uint8_t mSamplerBlockOffsets[filament::BindingPoints::COUNT] = { UNKNOWN_OFFSET };
 };
 

libs/utils/src/Profiler.cpp

@@ -63,6 +63,7 @@ Profiler::Profiler() noexcept {
 }
 
 Profiler::~Profiler() noexcept {
+    #pragma nounroll
     for (int fd : mCountersFd) {
         if (fd >= 0) {
             close(fd);
@@ -72,6 +73,7 @@ Profiler::~Profiler() noexcept {
 
 uint32_t Profiler::resetEvents(uint32_t eventMask) noexcept {
     // close all counters
+    #pragma nounroll
     for (int& fd : mCountersFd) {
         if (fd >= 0) {
             close(fd);
@@ -133,7 +135,7 @@ uint32_t Profiler::resetEvents(uint32_t eventMask) noexcept {
                 mEnabledEvents |= EV_L1D_MISSES;
             }
         }
-    
+
         if (eventMask & EV_BPU_REFS) {
             pe.type = PERF_TYPE_HARDWARE;
             pe.config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS;
@@ -143,7 +145,7 @@ uint32_t Profiler::resetEvents(uint32_t eventMask) noexcept {
                 mEnabledEvents |= EV_BPU_REFS;
             }
         }
-    
+
         if (eventMask & EV_BPU_MISSES) {
             pe.type = PERF_TYPE_HARDWARE;
             pe.config = PERF_COUNT_HW_BRANCH_MISSES;
@@ -153,7 +155,7 @@ uint32_t Profiler::resetEvents(uint32_t eventMask) noexcept {
                 mEnabledEvents |= EV_BPU_MISSES;
             }
         }
-    
+
 #ifdef __ARM_ARCH
         if (eventMask & EV_L1I_REFS) {
             pe.type = PERF_TYPE_RAW;