Provide optimised version of QOpenGLFramebufferObject::blitFramebuffer

Profiling shows that the call to glGetIntegerv() in this function that
queries for the currently bound framebuffer is very expensive. It's
the top hit on the CPU when profiling a Qt3D application using a
Scene3D Qt Quick 2 item. The reason it is so expensive is that this
call forces the OpenGL driver to propagate the OpenGL state through
all of the queued commands in order to answer the query. It may
also induce a pipeline stall depending upon the driver implementation.

As this function gets called on the hot path every frame whenever using
a Scene3D item and may also be called in plain Qt Quick when using
ShaderEffect items; the layer property of QQuickItem; or when updating
the Qt Quick glyph cache texture on Core profile contexts, it is very
much worthy of optimization.

This commit adds an overload of the blitFramebuffer() call that allows
the caller to provide a policy that can either:

* keep the existing behavior of restoring the previous framebuffer
  binding,
* restore the default framebuffer, or
* don't restore anything and let the caller be responsible for it.

This will allow consumers such as Qt Quick and Qt 3D to use the
optimised code path. The existing overloads of blitFramebuffer()
retain the current behavior by calling with the policy to restore the
previous framebuffer binding.

Upon making this change, the cost of blitFramebuffer() is massively
reduced. A follow up commit will optimize this further.

Change-Id: I417abb7da916ae5088f6817e4eff8ea02c8c5803
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

0 files changed, 0 insertions, 0 deletions