Rearrange most of the internals of MakeContextCurrent. Put all of the code to
bind the new context up front. If that is successful, unbind the old context.
This saves a lot of code and removes some locking crazyiness.
This patch has been tested for indirect rendering with glxinfo, glxgears,
manywin, and wincopy.
do not import arrays for generic arb attribs if the array is not enabled (same as for generic attribs for nv vp is already done). Since the requested stride is 16, otherwise the code would end up doing lots of unnecessary import work (in doom3, trans_4_GLfloat_4f_raw caused by that was by far the single most time-consuming function in the r200 driver, not importing the disabled arrays caused the cpu time spent in the driver to drop from 45% to 30%, though real-world gain was pretty minimal as it's not really cpu bound here in the first place).
Remove code in the Python scripts that keeps the indentation of the
generated code consistent for all cases. Instead, pass the generated code
through the indent command. Unix tools for the win!
Divide categories into four groups as they are processed from the XML. Add
an iterator to iterate over the categories in order, starting with "core"
versions, then ARB extensions, then numbered non-ARB extensions, and finally
unnumbered extensions.
Use the new iterator in a couple places to ensure that output that is
grouped by catgory is generated in a consistent order.
More changes to the scripts are coming. The generated files will be
committed one time after all the changes are in. Too bad we're not using
GIT, or this would be easy. :(
The new method client_supported_for_indirect is used to determine whether or
not the client-side library supports the function for indirect rendering.
The may or may not have associated protocol that needs to be generated
(e.g., glVertexPointer does not, but glVertex3fv does).
i915: Attempt to schedule buffer swap on target vertical blank when possible.
This has some advantages over the traditional way of first waiting for the
target vertical blank and then emitting the buffer swap, e.g.
* glXSwapBuffers returns immediately, only the next time the driver needs the
hardware lock will it block until the target vertical blank. This should
allow applications that don't intermix rendering and other processing to
start processing for the next frame right away.
* It's less likely to produce tearing.