[Mesa-users] Mesa OpenGL functions dispatching to noop version of glapi_mapi_tmp.h

[Michael Saunders]

I am employing two Mesa OpenGL contexts in a Qt5 app that are part of the
same context share group, https://doc.qt.io/qt-5/qopenglcontextgroup.html,
in two different threads. One thread renders to an offscreen OpenGL
framebuffer object. After each frame is complete the thread detaches the
texture from the framebuffer object and passes it, along with an OpenGL
frame sync, to the main thread for compositing with the on-screen
framebuffer. This allows the application GUI to remain responsive while
heavy-duty rendering is being performed. This works well with NVidia OpenGL
drivers installed but not with Mesa's swrast or Mesa LLVM drivers that I
built from source. Stepping into the Mesa source code I can see the thread
that renders to the offscreen framebuffer object steps into OpenGL
functions defined in glapi_mapi_tmp.h where every function stub is
implemented as "noop<GLfunction>" where <GLfunction> is the name of the
OpenGL function with the "gl" prefix removed. The functions are complete
no-ops.

Is this evidence that I didn't compile Mesa correctly for concurrency or
something else? The Mesa configure log indicated that I do have GLX_TLS
enabled.

Thanks,
Michael
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-users/attachments/20190401/638a9244/attachment.html>