Merge pull request #4984 from Sonicadvance1/shm_guaranteed_or_your_money_back

SHMStats: Add a 16-byte alignment guarantee

FEXCore/include/FEXCore/Utils/SHMStats.h

@@ -39,10 +39,12 @@ enum class AppType : uint8_t {
   WIN_WOW64,
 };
 
+// Only append new members to the end of {ThreadStatsHeader, ThreadStats} to allow old tools time to support new information.
+// FEX isn't guaranteeing /not/ breaking compatibility with versions, but trying to not cause too much churn.
 struct ThreadStatsHeader {
   uint8_t Version;
   AppType app_type;
-  uint8_t _pad[2];
+  uint16_t ThreadStatsSize;
   char fex_version[48];
   std::atomic<uint32_t> Head;
   std::atomic<uint32_t> Size;
@@ -63,6 +65,9 @@ struct ThreadStats {
   uint64_t AccumulatedFloatFallbackCount;
 };
 
+// Ensure 16-byte alignment to take advantage of ARM single-copy atomicity.
+static_assert(sizeof(ThreadStats) % 16 == 0, "Needs to be 16-byte aligned!");
+
 template<typename T, size_t FlatOffset = 0>
 class AccumulationBlock final {
 public:

Source/Common/SHMStats.cpp

@@ -13,10 +13,11 @@ void StatAllocBase::SaveHeader(FEXCore::SHMStats::AppType AppType) {
   Head = reinterpret_cast<FEXCore::SHMStats::ThreadStatsHeader*>(Base);
   Head->Size.store(CurrentSize, std::memory_order_relaxed);
   Head->Version = FEXCore::SHMStats::STATS_VERSION;
+  Head->app_type = AppType;
+  Head->ThreadStatsSize = sizeof(FEXCore::SHMStats::ThreadStats);
 
   std::string_view GitString = GIT_DESCRIBE_STRING;
   strncpy(Head->fex_version, GitString.data(), std::min(GitString.size(), sizeof(Head->fex_version)));
-  Head->app_type = AppType;
 
   Stats = reinterpret_cast<FEXCore::SHMStats::ThreadStats*>(reinterpret_cast<uint64_t>(Base) + sizeof(FEXCore::SHMStats::ThreadStatsHeader));