Adding NVIDIA 'Threaded optimization' workaround/fix at initialization on Windows for javaws/applets.

It has been observed that for some combinations, eg:
     - Windows 7 64bit (other variants may apply too)
     - NVIDIA 8600M GT
     - 260.99

the NVIDIA setting of 'Threaded optimization' := 'auto' (default) causes the JVM to simply crash
in case of javaws and [jnlp] applets.

'Threaded Optimization' := 'off' works reliable

'Threaded Optimization' := 'on'  never works with javaws and applets on the above configuration

A user could workaround this by setting 'Threaded Optimization' := 'off',
however, this would disable many users on the spot,
since you cannot ask the average user for such a task, if she only wants to see a web page.

This patch 'fixes' the 'auto' mode by running the eager GL profile initialization
within a block of single CPU affinity:

  SetProcessAffinityMask(pid, 1);
  try {
    initProfilesForDeviceImpl(device);
  } finally {
    SetProcessAffinityMask(pid, sysValue);
  }

Hopefully we can remove this hack with a driver fix.
However this workaround is as little invasive as possible.

1 files changed, 5 insertions, 0 deletions

diff --git a/make/stub_includes/win32/wingdi.h b/make/stub_includes/win32/wingdi.h
index 7862b5533..b169621f5 100644
--- a/make/stub_includes/win32/wingdi.h
+++ b/make/stub_includes/win32/wingdi.h
@@ -194,6 +194,11 @@ WINUSERAPI int  WINAPI ReleaseDC(HWND hWnd, HDC hDC);
 WINUSERAPI BOOL WINAPI DestroyWindow(HWND hWnd);
 WINUSERAPI DWORD WINAPI GetObjectType(HGDIOBJ h);
 
+WINUSERAPI HANDLE WINAPI GetCurrentProcess(void);
+WINUSERAPI BOOL WINAPI GetProcessAffinityMask(HANDLE hProcess,PDWORD_PTR lpProcessAffinityMask,PDWORD_PTR lpSystemAffinityMask);
+WINUSERAPI BOOL WINAPI SetProcessAffinityMask(HANDLE hProcess,DWORD_PTR dwProcessAffinityMask);
+
+
 // Routines for changing gamma ramp of display device
 WINGDIAPI BOOL        WINAPI GetDeviceGammaRamp(HDC,LPVOID);
 WINGDIAPI BOOL        WINAPI SetDeviceGammaRamp(HDC,LPVOID);