win32: Delay using dispatch tables until the second MakeCurrent.

Fixes #11.

4 files changed, 55 insertions, 45 deletions

diff --git a/README.md b/README.md
index 0beb63c..5d10539 100644
--- a/README.md
+++ b/README.md
@@ -120,12 +120,6 @@ epoxy (in a way that might change the device or pixel format), then
 epoxy needs to be notified of the change using the
 ```epoxy_handle_external_wglMakeCurrent()``` function.
 
-The win32 dispatch layer is currently slower than it should be in the
-single-context (or multi-context, but same device and pixel format)
-case.  We need to switch to using the linux-like function pointer
-stubs, and detect when transitioning to multi-device/format and hook
-in the per-thread dispatch table at that point.
-
-The win32 wglMakeCurrent () variants are slower than they should be,
+The win32 wglMakeCurrent() variants are slower than they should be,
 because they should be caching the resolved dispatch tables instead of
 resetting an entire thread-local dispatch table every time.
diff --git a/src/dispatch_common.h b/src/dispatch_common.h
index aa8d04e..5974aa0 100644
--- a/src/dispatch_common.h
+++ b/src/dispatch_common.h
@@ -76,10 +76,23 @@
 #define WRAPPER_VISIBILITY static GLAPIENTRY
 #define WRAPPER(x) x ## _wrapped
 
-#if USING_DISPATCH_TABLE
-#define GEN_GLOBAL_REWRITE_PTR(name, args, passthrough)
-#define GEN_GLOBAL_REWRITE_PTR_RET(ret, name, args, passthrough)
+#define GEN_GLOBAL_REWRITE_PTR(name, args, passthrough)          \
+    static EPOXYCALL void                                        \
+    name##_global_rewrite_ptr args                               \
+    {                                                            \
+        name = (void *)name##_resolver();                        \
+        name passthrough;                                        \
+    }
 
+#define GEN_GLOBAL_REWRITE_PTR_RET(ret, name, args, passthrough) \
+    static EPOXYCALL ret                                         \
+    name##_global_rewrite_ptr args                               \
+    {                                                            \
+        name = (void *)name##_resolver();                        \
+        return name passthrough;                                 \
+    }
+
+#if USING_DISPATCH_TABLE
 #define GEN_DISPATCH_TABLE_REWRITE_PTR(name, args, passthrough)            \
     static EPOXYCALL void                                                  \
     name##_dispatch_table_rewrite_ptr args                                 \
@@ -115,21 +128,6 @@
     }
 
 #else
-#define GEN_GLOBAL_REWRITE_PTR(name, args, passthrough)          \
-    static EPOXYCALL void                                        \
-    name##_global_rewrite_ptr args                               \
-    {                                                            \
-        name = (void *)name##_resolver();                        \
-        name passthrough;                                        \
-    }
-
-#define GEN_GLOBAL_REWRITE_PTR_RET(ret, name, args, passthrough) \
-    static EPOXYCALL ret                                         \
-    name##_global_rewrite_ptr args                               \
-    {                                                            \
-        name = (void *)name##_resolver();                        \
-        return name passthrough;                                 \
-    }
 #define GEN_DISPATCH_TABLE_REWRITE_PTR(name, args, passthrough)
 #define GEN_DISPATCH_TABLE_REWRITE_PTR_RET(ret, name, args, passthrough)
 #define GEN_DISPATCH_TABLE_THUNK(name, args, passthrough)
@@ -176,9 +174,12 @@ extern void UNWRAPPED_PROTO(glEnd_unwrapped)(void);
 
 #if USING_DISPATCH_TABLE
 void gl_init_dispatch_table(void);
+void gl_switch_to_dispatch_table(void);
 void wgl_init_dispatch_table(void);
+void wgl_switch_to_dispatch_table(void);
 extern uint32_t gl_tls_index, gl_tls_size;
 extern uint32_t wgl_tls_index, wgl_tls_size;
+
 #define wglMakeCurrent_unwrapped epoxy_wglMakeCurrent_unwrapped
 #define wglMakeContextCurrentARB_unwrapped epoxy_wglMakeContextCurrentARB_unwrapped
 #define wglMakeContextCurrentEXT_unwrapped epoxy_wglMakeContextCurrentEXT_unwrapped
@@ -187,4 +188,4 @@ extern BOOL UNWRAPPED_PROTO(wglMakeCurrent_unwrapped)(HDC hdc, HGLRC hglrc);
 extern BOOL UNWRAPPED_PROTO(wglMakeContextCurrentARB_unwrapped)(HDC hDrawDC, HDC hReadDC, HGLRC hglrc);
 extern BOOL UNWRAPPED_PROTO(wglMakeContextCurrentEXT_unwrapped)(HDC hDrawDC, HDC hReadDC, HGLRC hglrc);
 extern BOOL UNWRAPPED_PROTO(wglMakeAssociatedContextCurrentAMD_unwrapped)(HGLRC hglrc);
-#endif
+#endif /* _WIN32_ */
diff --git a/src/dispatch_wgl.c b/src/dispatch_wgl.c
index 1156419..5b3a8ef 100644
--- a/src/dispatch_wgl.c
+++ b/src/dispatch_wgl.c
@@ -27,6 +27,9 @@
 
 #include "dispatch_common.h"
 
+static bool first_context_current = false;
+static bool already_switched_to_dispatch_table = false;
+
 /**
  * If we can determine the WGL extension support from the current
  * context, then return that, otherwise give the answer that will just
@@ -63,16 +66,27 @@ epoxy_has_wgl_extension(HDC hdc, const char *ext)
  * Does the work necessary to update the win32 per-thread dispatch
  * tables when wglMakeCurrent() is called.
  *
- * Right now, we just reset everything to "do wglGetProcAddress()
- * again".  This could be improved in the future to track a resolved
- * dispatch table per context and reuse it when the context is made
- * current again.
+ * Right now, we use global function pointers until the second
+ * MakeCurrent occurs, at which point we switch to dispatch tables.
+ * This could be improved in the future to track a resolved dispatch
+ * table per context and reuse it when the context is made current
+ * again.
  */
 PUBLIC void
 epoxy_handle_external_wglMakeCurrent(void)
 {
-    gl_init_dispatch_table();
-    wgl_init_dispatch_table();
+    if (!first_context_current) {
+        first_context_current = true;
+    } else {
+        if (!already_switched_to_dispatch_table) {
+            already_switched_to_dispatch_table = true;
+            gl_switch_to_dispatch_table();
+            wgl_switch_to_dispatch_table();
+        }
+
+        gl_init_dispatch_table();
+        wgl_init_dispatch_table();
+    }
 }
 
 /**
@@ -96,6 +110,8 @@ DllMain(HINSTANCE dll, DWORD reason, LPVOID reserved)
         if (wgl_tls_index == TLS_OUT_OF_INDEXES)
             return FALSE;
 
+        first_context_current = false;
+
         /* FALLTHROUGH */
 
     case DLL_THREAD_ATTACH:
@@ -105,7 +121,6 @@ DllMain(HINSTANCE dll, DWORD reason, LPVOID reserved)
         data = LocalAlloc(LPTR, wgl_tls_size);
         TlsSetValue(wgl_tls_index, data);
 
-        epoxy_handle_external_wglMakeCurrent();
         break;
 
     case DLL_THREAD_DETACH:
diff --git a/src/gen_dispatch.py b/src/gen_dispatch.py
index 5af23d6..072a0fe 100755
--- a/src/gen_dispatch.py
+++ b/src/gen_dispatch.py
@@ -611,18 +611,12 @@ class Generator(object):
                                                                        func.args_decl,
                                                                        func.args_list))
 
-    def write_linux_function_pointer(self, func):
+    def write_function_pointer(self, func):
         self.outln('{0}{1} epoxy_{2} = epoxy_{2}_global_rewrite_ptr;'.format(func.public,
                                                                              func.ptr_type,
                                                                              func.wrapped_name))
         self.outln('')
 
-    def write_win32_function_pointer(self, func):
-        self.outln('{0}{1} epoxy_{2} = epoxy_{2}_dispatch_table_thunk;'.format(func.public,
-                                                                               func.ptr_type,
-                                                                               func.wrapped_name))
-        self.outln('')
-
     def write_provider_enums(self):
         # Writes the enum declaration for the list of providers
         # supported by gl_provider_resolver()
@@ -728,6 +722,7 @@ class Generator(object):
         self.outln(' */')
         self.outln('')
         self.outln('#include <stdlib.h>')
+        self.outln('#include <string.h>')
         self.outln('#include <stdio.h>')
         self.outln('')
         self.outln('#include "dispatch_common.h"')
@@ -788,16 +783,21 @@ class Generator(object):
         self.outln('}')
         self.outln('')
 
-        for func in self.sorted_functions:
-            self.write_win32_function_pointer(func)
-
-        self.outln('#else /* !USING_DISPATCH_TABLE */')
+        self.outln('void')
+        self.outln('{0}_switch_to_dispatch_table(void)'.format(self.target))
+        self.outln('{')
 
         for func in self.sorted_functions:
-            self.write_linux_function_pointer(func)
+            self.outln('    epoxy_{0} = epoxy_{0}_dispatch_table_thunk;'.format(func.wrapped_name))
+
+        self.outln('}')
+        self.outln('')
 
         self.outln('#endif /* !USING_DISPATCH_TABLE */')
 
+        for func in self.sorted_functions:
+            self.write_function_pointer(func)
+
 argparser = argparse.ArgumentParser(description='Generate GL dispatch wrappers.')
 argparser.add_argument('files', metavar='file.xml', nargs='+', help='GL API XML files to be parsed')
 argparser.add_argument('--dir', metavar='dir', required=True, help='Destination directory')