One of the problems with krhplatform.h (besides it being a terrible
idea in the first place) is that it's not present on many systems at
all, including apple, win32, and linux before late 2009.
So, unless we introduce the first ./configure dependency to the core
library build, we really do need to define these standard types to
their standard values. But if we allow khrplatform.h inclusion as
well, we'll get long-vs-int redefinition warnings on 32-bit. So, we
have to completely replicate the header, which is a thing we've done
for everything else except for eglplatform.h.
A testcase will be used to make sure that our types don't drift from
the system khrplatform.h types, if it even exists.
Fixes#17
For performance, I want to be able to make single-context (well,
single-pixel-format-and-device) apps be able to directly call GL
functions through function pointers. Bake that into the ABI now so I
can get a release out the door and fix this up later.
This also fixes the lack of __stdcall annotation on the
PFNWHATEVERPROC typedefs.
This lets the compiler generate faster function calls (call through
function pointer, instead of call into a linker-generated stub func
containing jump to function pointer).
For example, on desktop 2.1 GL on Apple, there's no glBindVertexArray,
but there is glBindVertexArrayAPPLE, and as far as a caller is
concerned, the APPLE variant should be able to stand in for the
core/ARB version. Similarly for trying to do FBOs on an old Mesa
implementation that didn't have ARB_fbo yet, but did have EXT_fbo.
In addition to the failing testcase, there were a couple of
regressions in piglit's attribs test: one from glBegin_unwrapped vs
glBegin confusion in the __asm__ directives we were generating, and
one where the function pointers apparently were just getting mixed up
at application runtime.
These tell the linker to generate GNU_IFUNC relocs that rewrite the
PLT entries in the user's address space to point to our resolved GL
function, so there's no extra function pointer. It also, as a bonus,
cuts 400k out of the library.
This requires a toolchain from 2010 or so. Unfortunately, it's going
to take a bit more investigation to find out what specific bits are
required.
Fixes#4
We now initialize our dispatch table with function pointers that go
and do the rewrite, then we never have to check for a NULL table entry
again. On my 64-bit build, epoxy_glClear() drops from 83 bytes to 14,
while the total library size only goes up by 5%.
This also paves the way for fixing our dispatch table management using
TLS on windows.
This is going to change for macos and win32, and this will be easier
than trying to spread that logic through the python code and into the
generated code.
I think we don't want to be checking that the context is actually of
the declared types -- if the extension is exposed in the extension
string, the entrypoints had better be there.
This totally replaces the getprocaddress and dlsym code, which was
basically just stubs up until now. The is_glx/is_egl() is also
dropped -- they weren't doing anything, and the only false answer they
could give is if the dlopen were to fail.
I was thinking at one point that part of the build was going to
require not including the #defines from the generated code, but would
want these prototypes. It turns out that's not the case (and if it
is, I'll just wrap the #defines in an ifdef).
Unfortunately, for GLX 1.4+ entrypoints (just glxGetProcAddress
currently) or extensions, if there isn't a context bound then we don't
have a dpy and screen available to provide useful debug messages. Oh
well.
Eric Anholt