Work around a MinGW thread_local bug

MinGW-w64 generates bad code when accessing extern thread_local objects.
Wrapper functions are used to ensure it only accesses them from the same place
they're defined. This unfortunately adds a bit of overhead for what should be a
relatively simple thing.

These functions are inlined for non-MinGW targets, avoiding the overhead on
non-affected targets.

4 files changed, 31 insertions, 15 deletions

diff --git a/router/al.cpp b/router/al.cpp
index 5e273236..db18206e 100644
--- a/router/al.cpp
+++ b/router/al.cpp
@@ -11,31 +11,31 @@ std::atomic<DriverIface*> CurrentCtxDriver{nullptr};
 
 #define DECL_THUNK1(R,n,T1) AL_API R AL_APIENTRY n(T1 a)                      \
 {                                                                             \
-    DriverIface *iface = ThreadCtxDriver;                                     \
+    DriverIface *iface = GetThreadDriver();                                   \
     if(!iface) iface = CurrentCtxDriver.load(std::memory_order_acquire);      \
     return iface->n(a);                                                       \
 }
 #define DECL_THUNK2(R,n,T1,T2) AL_API R AL_APIENTRY n(T1 a, T2 b) \
 {                                                                             \
-    DriverIface *iface = ThreadCtxDriver;                                     \
+    DriverIface *iface = GetThreadDriver();                                   \
     if(!iface) iface = CurrentCtxDriver.load(std::memory_order_acquire);      \
     return iface->n(a, b);                                                    \
 }
 #define DECL_THUNK3(R,n,T1,T2,T3) AL_API R AL_APIENTRY n(T1 a, T2 b, T3 c) \
 {                                                                             \
-    DriverIface *iface = ThreadCtxDriver;                                     \
+    DriverIface *iface = GetThreadDriver();                                   \
     if(!iface) iface = CurrentCtxDriver.load(std::memory_order_acquire);      \
     return iface->n(a, b, c);                                                 \
 }
 #define DECL_THUNK4(R,n,T1,T2,T3,T4) AL_API R AL_APIENTRY n(T1 a, T2 b, T3 c, T4 d) \
 {                                                                             \
-    DriverIface *iface = ThreadCtxDriver;                                     \
+    DriverIface *iface = GetThreadDriver();                                   \
     if(!iface) iface = CurrentCtxDriver.load(std::memory_order_acquire);      \
     return iface->n(a, b, c, d);                                              \
 }
 #define DECL_THUNK5(R,n,T1,T2,T3,T4,T5) AL_API R AL_APIENTRY n(T1 a, T2 b, T3 c, T4 d, T5 e) \
 {                                                                             \
-    DriverIface *iface = ThreadCtxDriver;                                     \
+    DriverIface *iface = GetThreadDriver();                                   \
     if(!iface) iface = CurrentCtxDriver.load(std::memory_order_acquire);      \
     return iface->n(a, b, c, d, e);                                           \
 }
@@ -46,7 +46,7 @@ std::atomic<DriverIface*> CurrentCtxDriver{nullptr};
  */
 AL_API ALenum AL_APIENTRY alGetError(void)
 {
-    DriverIface *iface = ThreadCtxDriver;
+    DriverIface *iface = GetThreadDriver();
     if(!iface) iface = CurrentCtxDriver.load(std::memory_order_acquire);
     return iface ? iface->alGetError() : AL_NO_ERROR;
 }
diff --git a/router/alc.cpp b/router/alc.cpp
index d43c78ce..5532c047 100644
--- a/router/alc.cpp
+++ b/router/alc.cpp
@@ -469,21 +469,21 @@ ALC_API ALCboolean ALC_APIENTRY alcMakeContextCurrent(ALCcontext *context)
      */
     if(idx < 0)
     {
-        DriverIface *oldiface = ThreadCtxDriver;
+        DriverIface *oldiface = GetThreadDriver();
         if(oldiface) oldiface->alcSetThreadContext(nullptr);
         oldiface = CurrentCtxDriver.exchange(nullptr);
         if(oldiface) oldiface->alcMakeContextCurrent(nullptr);
     }
     else
     {
-        DriverIface *oldiface = ThreadCtxDriver;
+        DriverIface *oldiface = GetThreadDriver();
         if(oldiface && oldiface != &DriverList[idx])
             oldiface->alcSetThreadContext(nullptr);
         oldiface = CurrentCtxDriver.exchange(&DriverList[idx]);
         if(oldiface && oldiface != &DriverList[idx])
             oldiface->alcMakeContextCurrent(nullptr);
     }
-    ThreadCtxDriver = nullptr;
+    SetThreadDriver(nullptr);
 
     return ALC_TRUE;
 }
@@ -526,7 +526,7 @@ ALC_API void ALC_APIENTRY alcDestroyContext(ALCcontext *context)
 
 ALC_API ALCcontext* ALC_APIENTRY alcGetCurrentContext(void)
 {
-    DriverIface *iface = ThreadCtxDriver;
+    DriverIface *iface = GetThreadDriver();
     if(!iface) iface = CurrentCtxDriver.load();
     return iface ? iface->alcGetCurrentContext() : nullptr;
 }
@@ -932,10 +932,10 @@ ALC_API ALCboolean ALC_APIENTRY alcSetThreadContext(ALCcontext *context)
 
     if(!context)
     {
-        DriverIface *oldiface = ThreadCtxDriver;
+        DriverIface *oldiface = GetThreadDriver();
         if(oldiface && !oldiface->alcSetThreadContext(nullptr))
             return ALC_FALSE;
-        ThreadCtxDriver = nullptr;
+        SetThreadDriver(nullptr);
         return ALC_TRUE;
     }
 
@@ -944,10 +944,10 @@ ALC_API ALCboolean ALC_APIENTRY alcSetThreadContext(ALCcontext *context)
     {
         if(DriverList[idx].alcSetThreadContext(context))
         {
-            DriverIface *oldiface = ThreadCtxDriver;
+            DriverIface *oldiface = GetThreadDriver();
             if(oldiface != &DriverList[idx])
             {
-                ThreadCtxDriver = &DriverList[idx];
+                SetThreadDriver(&DriverList[idx]);
                 if(oldiface) oldiface->alcSetThreadContext(nullptr);
             }
             return ALC_TRUE;
@@ -960,7 +960,7 @@ ALC_API ALCboolean ALC_APIENTRY alcSetThreadContext(ALCcontext *context)
 
 ALC_API ALCcontext* ALC_APIENTRY alcGetThreadContext(void)
 {
-    DriverIface *iface = ThreadCtxDriver;
+    DriverIface *iface = GetThreadDriver();
     if(iface) return iface->alcGetThreadContext();
     return nullptr;
 }
diff --git a/router/router.cpp b/router/router.cpp
index 731db063..7eb0bddc 100644
--- a/router/router.cpp
+++ b/router/router.cpp
@@ -24,6 +24,11 @@ thread_local DriverIface *ThreadCtxDriver;
 enum LogLevel LogLevel = LogLevel_Error;
 FILE *LogFile;
 
+#ifdef __MINGW32__
+DriverIface *GetThreadDriver() noexcept { return ThreadCtxDriver; }
+void SetThreadDriver(DriverIface *driver) noexcept { ThreadCtxDriver = driver; }
+#endif
+
 static void LoadDriverList(void);
 
 
diff --git a/router/router.h b/router/router.h
index 253494de..f9ebd4a4 100644
--- a/router/router.h
+++ b/router/router.h
@@ -172,6 +172,17 @@ extern std::vector<DriverIface> DriverList;
 extern thread_local DriverIface *ThreadCtxDriver;
 extern std::atomic<DriverIface*> CurrentCtxDriver;
 
+/* HACK: MinGW generates bad code when accessing an extern thread_local object.
+ * Add a wrapper function for it that only accesses it where it's defined.
+ */
+#ifdef __MINGW32__
+DriverIface *GetThreadDriver() noexcept;
+void SetThreadDriver(DriverIface *driver) noexcept;
+#else
+inline DriverIface *GetThreadDriver() noexcept { return ThreadCtxDriver; }
+inline void SetThreadDriver(DriverIface *driver) noexcept { ThreadCtxDriver = driver; }
+#endif
+
 
 class PtrIntMap {
     void **mKeys{nullptr};