win32: Use thread local storage to get us a dispatch table per thread.

src/dispatch_common.h

@@ -94,3 +94,10 @@ bool epoxy_extension_in_string(const char *extension_list, const char *ext);
 
 extern void UNWRAPPED_PROTO(epoxy_glBegin_unwrapped)(GLenum primtype);
 extern void UNWRAPPED_PROTO(epoxy_glEnd_unwrapped)(void);
+
+#if USING_DISPATCH_TABLE
+void gl_init_dispatch_table(void);
+void wgl_init_dispatch_table(void);
+extern uint32_t gl_tls_index, gl_tls_size;
+extern uint32_t wgl_tls_index, wgl_tls_size;
+#endif

src/dispatch_wgl.c

@@ -58,3 +58,62 @@ epoxy_has_wgl_extension(HDC hdc, const char *ext)
 
     return epoxy_extension_in_string(getext(hdc), ext);
 }
+
+static void
+reset_dispatch_table(void)
+{
+    gl_init_dispatch_table();
+    wgl_init_dispatch_table();
+}
+
+/**
+ * This global symbol is apparently looked up by Windows when loading
+ * a DLL, but it doesn't declare the prototype.
+ */
+BOOL WINAPI
+DllMain(HINSTANCE dll, DWORD reason, LPVOID reserved);
+
+BOOL WINAPI
+DllMain(HINSTANCE dll, DWORD reason, LPVOID reserved)
+{
+    void *data;
+
+    switch (reason) {
+    case DLL_PROCESS_ATTACH:
+        gl_tls_index = TlsAlloc();
+        if (gl_tls_index == TLS_OUT_OF_INDEXES)
+            return FALSE;
+        wgl_tls_index = TlsAlloc();
+        if (wgl_tls_index == TLS_OUT_OF_INDEXES)
+            return FALSE;
+
+        /* FALLTHROUGH */
+
+    case DLL_THREAD_ATTACH:
+        data = LocalAlloc(LPTR, gl_tls_size);
+        TlsSetValue(gl_tls_index, data);
+
+        data = LocalAlloc(LPTR, wgl_tls_size);
+        TlsSetValue(wgl_tls_index, data);
+
+        reset_dispatch_table();
+        break;
+
+    case DLL_THREAD_DETACH:
+    case DLL_PROCESS_DETACH:
+        data = TlsGetValue(gl_tls_index);
+        LocalFree(data);
+
+        data = TlsGetValue(wgl_tls_index);
+        LocalFree(data);
+        break;
+
+        if (reason == DLL_PROCESS_DETACH) {
+            TlsFree(gl_tls_index);
+            TlsFree(wgl_tls_index);
+        }
+        break;
+    }
+
+    return TRUE;
+}

src/gen_dispatch.py

@@ -794,11 +794,22 @@ class Generator(object):
         self.outln('};')
         self.outln('')
 
+        self.outln('uint32_t {0}_tls_index;'.format(self.target))
+        self.outln('uint32_t {0}_tls_size = sizeof(struct dispatch_table);'.format(self.target))
+        self.outln('')
+
         self.outln('static inline struct dispatch_table *')
         self.outln('get_dispatch_table(void)')
         self.outln('{')
-        self.outln('    /* XXX: Make this thread-local and swapped on makecurrent on win32. */')
-        self.outln('	return &resolver_table;')
+        self.outln('	return TlsGetValue({0}_tls_index);'.format(self.target))
+        self.outln('}')
+        self.outln('')
+
+        self.outln('void')
+        self.outln('{0}_init_dispatch_table(void)'.format(self.target))
+        self.outln('{')
+        self.outln('    struct dispatch_table *dispatch_table = get_dispatch_table();')
+        self.outln('    memcpy(dispatch_table, &resolver_table, sizeof(resolver_table));')
         self.outln('}')
         self.outln('')