No need to jump through hoops to track __DRIdrivers and avoid dlopening the
same .so more than twice, dlopen() does this internally. Besides, we
were already bypassing this and dlopening drivers for each screen,
whether or not they were already dlopened.
The DriverAPI is internal to the DRI drivers and GetDrawableMSC
obsoletes GetMSC. Also, since the DRI driver interface has not yet
been released, just drop the getMSC function from the DRI interface
instead using the ABI preserving version mechanism.
Finally, using void pointer privates in the DRI interface is not allowed,
always pass the actual types around (__DRIdrawable in this case) to
enhance type safety and readability of the code.
Instead of generating the glapi sources for the xserver and commiting
them to the xserver tree, we can keep them in the mesa tree and change
the xserver build to use the files from the mesa tree.
This makes the xserver glx build more robust as it reduces the chances
for mismatches of the glX API used in the xserver vs. in mesa.
If the server didn't send a value, assume it's not supported.
A more generic solution might be better for this kind of problem, but an
attempt for this failed (see https://bugs.freedesktop.org/show_bug.cgi?id=9264)
and this allows compiz to work with drivers that support
GL_EXT_framebuffer_object.
Consolidate support for synchronizing to and retrieving vblank counters. Also
fix the core vblank code to return monotonic MSC counters, which are required
by some GLX extensions. Adding support for multiple pipes to a low level
driver is fairly easy, the Intel 965 driver provides simple example code (see
intel_buffers.c:intelWindowMoved()).
The new code bumps the media stream counter extension version to 2 and adds a
new getDrawableMSC callback. This callback takes a drawablePrivate pointer,
which is used to calculate the MSC value seen by clients based on the actual
vblank counter(s) returned from the kernel. The new drawable private fields
are as follows:
- vblSeq - used for tracking vblank counts for buffer swapping
- vblFlags - flags (e.g. current pipe), updated by low level driver
- msc_base - MSC counter from the last time the current pipe changed
- vblank_base - kernel DRM vblank counter from the last time the pipe changed
Using the above variables, the core vblank code (in vblank.c) can calculate a
monotonic MSC value. The low level DRI drivers are responsible for updating
the current pipe (by setting VBLANK_FLAG_SECONDARY for example in vblFlags)
along with msc_base and vblank_base whenever the pipe associated with a given
drawable changes (again, see intelWindowMoved for an example of this).
Drivers should fill in the GetDrawableMSC DriverAPIRec field to point to
driDrawableGetMSC32 and add code for pipe switching as outlined above to fully
support the new scheme.
The old version just set GLX_DRAWABLE_TYPE to GLX_WINDOW_BIT for
configs received through glXGetVisualConfigs and to
GLX_WINDOW_BIT | GLX_PIXMAP_BIT for configs received as FBConfigs.
The X server needs to send this info, but keep the workaround for now.
Fixes#12835.
The old implementation fetches fbconfigs or glx visuals once and assumes the list
describes both fbconfigs and glx visuals. This patch splits it up and fetches
visuals and fbconfigs in two steps and keep the two lists separate. A server
could have no glx visuals or no glx fbconfigs and the old code wouldn't know the
difference.
This new API lets the loader examine DRI level extensions provided by the
driver in a forward compatible manner.
Much of the churn in the DRI interface is adding support for new
extensions or removing old, unused extensions. This new extension
mechanism lets the loader query the extensions provided by the driver
and implement the extensions it knows about. Deprecating extensions
is done by not exporting that extension in the list, which doesn't
require keeping old function pointers around to preserve ABI.
The entrypoints take a mix of __DRIscreen * and void * (screen private)
arguments (similarly for contexts and drawables). This patch does away
with passing the private void pointer and always only passes the fully
typed __DRIscreen pointer and always as the first argument.
This makes the interface more consistent and increases type safety, and
catches a bug where we would pass a screen private to DRIdrawable::getSBC.
The screenConfigs field of __DRIscreen points back to the containing
__GLXscreenConfigs struct. This is a serious abstraction violation; it
assumes that the loader is libGL and that there *is* a __GLXscreenConfigs
type in the loader.
Using the containerOf macro, we can get from the __DRIscreen pointer to
the containing __GLXscreenConfigs struct, at a place in the stack
where the above is a valid assumption. Besides, the __DRI* structs shouldn't
hold state other than the private pointer.
__glXinitialize() can't be called with the GLX lock held. Just
pass in the __GLXscreenConfigs pointer so we don't have to look it
up in __GLXdisplayPrivate.
As for createDrawable and destroyDrawable, these functions immediately
upon entry to driCreateNewContext and immediately before exit from
driDestroyContext. Instead of passing function pointers back and forth
just obtain the drm_context_t prior to calling DRIscreen::createNewContext
and pass it as a parameter.
This change also lets us keep the DRI context XID in the libGL loader only.
All the DRI driver did was call the createDrawable callback immediately
upon entry to DRIscreen::createNewDrawable to get the drm_drawable_t.
We can just call that before calling into the DRI driver and pass the
returned drm_drawable_t as an argument to the DRI entry point.
Likewise for destroyDrawable.
Also, DRIdrawablePrivate::draw isn't used anywhere, and since the
driver no longer needs the XID of the drawable we can now drop that.
Many DRI entry points took a __DRInativeDisplay pointer and a screen
index as arguments. The only use for the native display pointer was to
pass it back to the loader when looking up the __DRIscreen for the given
screen index.
Instead, let's just pass in the __DRIscreen pointer directly, which
let's drop the __DRInativeDisplay type and the getScreen function.
The assumption is now that the loader will be able to retrieve context
from the __DRIscreen pointer when necessary.
This fixes a regression from commit f81b1dbe374fe446f6ef676e70a72952ffb47d4e:
Since then, driDestroyDisplay gets called from __glXFreeDisplayPrivate. It
dlcloses the handles associated with the display but fails to remove their
references from the Drivers list, so subsequent calls to OpenDriver return a
stale handle and an invalid createNewScreenFunc pointer. The attempt to call
the latter results in a segfault when running amoeba, e.g.
With this, tools like ximagesrc in gstreamer correctly see updates from GL
rendering. Support requires that the Xdamage library be current (but will be
disabled if not present) plus a new X Server with support for the new
XDamagePost request. libGL now has a new interface version, and also links
against libXdamage and libXfixes to support it, but backwards compatibility
is retained.
Currently, all drivers report damage at SwapBuffers time through common code --
front buffer rendering doesn't result in damage being reported. Also, the
damage is against the root window, as our drivers don't yet render to backing
store when they should (composited environments).
Fixes a GLX protocol problem when binding an indirect rendering context
after a direct rendering context. In this case, the oldContetTag sent to
the server should be None, but the currectContextTag stored in the direct
rendering context (i.e., -1) was sent instead.
I always build with -DGLX_USE_TLS, so I never hit these paths. glapi.h is
required in some places because _glapi_Dispatch is declared there, but
_glapi_tls_Dispatch is declared in glthread.h.
glDeleteTextures and glDeleteTexturesEXT were erroneously listed as
aliases of each other. For anything /except/ GLX protocol they are
aliases. This set of changes allows functions that are functionally
identical but have different GLX protocol to be listed as aliases.
When building with GLX_INDIRECT_RENDERING set, different static
functions are used. These functions determine whether the current
context is direct rendering or not. If the context is direct
rendering, the aliased function (e.g., glDeleteTextures in the case of
glDeleteTexturesEXT) is called. If the context is not direct
rendering, the correct GLX protocol is sent.
For a deeper explanation of what is changed, please see:
http://dri.freedesktop.org/wiki/PartiallyAliasedFunctions
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.
There were two sets of bugs in the vertex program (ARB and NV)
protocol. First, several of the ARB functions were missing the
'doubles_in_order="true"' annotation. Second, after the ARB decided
that glVertexAttrib*ARB functions must not alias fixed-function state
for GLSL, Nvidia re-assigned GLX protocol opcodes for
glVertexAttrib*NV (circa Septeber 2004). For some reason gl_API.xml
was never updated to reflect this, and the updated version of the
GL_NV_vertex_program spec never made into the registry.
Nvidia no longer supports this extension, and they no longer export its
entry points from their libGL. There's no reason for us to keep dragging it
around either.
The functions for GL_EXT_paletted_texture that do not share GLX
protocol with GL_ARB_imaging are supposed to alias the similar
functions from GL_SGI_color_table. They didn't. This patch corrects
this problem and enables GLX protocol for both extensions.
Since this removes 3 entries from the dispatch table, this change
creates a lot of changes in the generated files.
tells the scripts to assign an available offset to the function. The
important changes are in src/mesa/glapi/gl_XML.py and
src/mesa/glapi/*.xml.
Since the DRI drivers only depend on functions required by the ABI
(e.g., GL 1.2 + ARB_multitexture) having fixed offsets, all functions
not in the ABI use "assign" mode. This has caused the offset of
basically every function outside the ABI to change. I have verified
that a libGL with this patch works with a DRI driver without the patch.
Futher, several function were removed from the dispatch tables
altogether. These are the functions for the following extensions:
GL_SGIS_texture_filter4
GL_SGIS_texture4D
GL_SGIS_detail_texture
GL_SGIS_sharpen_texture
GL_SGIX_sprite
GL_SGIX_instruments
GL_SGIX_framezoom
GL_SGIX_tag_sample_buffer
GL_SGIX_reference_plane
GL_SGIX_flush_raster
GL_SGIX_list_priority
GL_SGIX_fragment_lighting
GL_PGI_misc_hints
GL_EXT_index_material
GL_EXT_index_func
GL_3DFX_tbuffer
This removes 50 functions from the dispatch table.
Some applications end up dlopening libGL without RTLD_GLOBAL, so the libGL
symbols referenced by the driver can't be unresolved when libGL dlopens it.
This attempts to make the libGL symbols visible to the driver by dlopening
libGL (again) with RTLD_GLOBAL before dlopening the driver and dlclosing
the obtained handle afterwards.
__glXCalcualteUsableExtensions. By having both callers of
__glXProcessServerString initialize the bit-fields passed in, we guarantee
that the correct amount of data is initialized:
This fixes bugzilla #7353.
But later, in __glXPopArrayState(), we were using that field to set
the active_texture_unit. The value was garbage and could lead to
things blowing up.
See bug 6863.
big, ugly, error prone switch-statement is replaced with a compact table.
I also added numerous comments, including a comment explaining how the
format parameter is validated.
Explicitly pass GL_FLOAT as the type in the cases where that is the only
possible value (e.g., everywhere except the call to glColorPointer).
Validate that stride is >= 0.
Tested with all modes (including the two error modes) of
progs/tests/interleave.c.
Bug: #5001, #5058
Reviewed by: Brian Paul
should be defined. It was flawed on some platforms (e.g., Darwin & mingw).
Instead, rely on the build system to define it on the compiler command line.
This also reverts ajax's hand-edit to indirect_size.c. I'll fix that on the
X.org side of things later today.
the driver. The loader now takes care of this for the driver.
Remove _glapi_DispatchTSD and give _glapi_Dispatch its semantic (i.e.,
having a NULL value means that the application is multithreaded and
_glapi_get_dispatch must be called).
Gut all of the dispatch override code. This removes _glapi_RealDispatch,
_glapi_tls_RealDispatch, _glapi_begin_dispatch_override,
_glapi_end_dispatch_override, and _glapi_get_override_dispatch.
Remove _glapi_get_proc_address, _glapi_get_proc_name, _glapi_get_version,
and _glapi_check_table from the loader / driver interface.
Reviewed by: Brian Paul
Make sure that all the values used in the clean-up code are initialized at
some point. It is still unclear to me as to why GCC does not complain. I
suspect the problem may be due to the depth of if-statement nesting.
direct-rendering is possible on a given screen. It is not possible to
determine this by just looking at driDisplay.private in the display
structure. However, it is possible to determine this by looking at
driScreen.private in the screen structure. This field is only non-NULL if a
DRI driver has been successfully loaded and bootstrapped. Once we get to
that point, we know that direct-rendering is at least theoretically
possible.
This fixes bugzilla #2875.
DRI drivers using the "old" interface relied on certain fields being at
certain offsets. Removing the structures would break that. Now that the
old interface is no longer supported, the dead data can finally die.
linux-dri, but it disables support for direct-rendering in it. This config
mainly exists so that I can verify that changes to libGL haven't broken
builds on non-DRI platforms.
_glapi_add_entrypoint has been replaced by a new routine called
_glapi_add_dispatch. This new routine dynamically assignes dispatch offsets
to functions added. This allows IHVs to add support for extension functions
that do not have assigned dispatch offsets.
It also means that a driver has no idea what offset will be assigned to a
function. The vast majority of the changes in this commit account for that.
An additional table, driDispatchRemapTable, is added. Functions not in the
Linux OpenGL ABI (i.e., anything not in GL 1.2 + ARB_multitexture) has a
fixed offset in this new table. The entry in this table specifies the
offset in of the function in the real dispatch table.
The internal interface was also bumped from version 20050725 to 20050727.
This has been tested with various programs in progs/demos on:
radeon (Radeon Mobility M6)
r128 (Rage 128 Pro)
mga (G400)
that are currently obtained via glXGetProcAddress and all of the XF86DRI
functions are replaced with a funciton table. This table will be passed to
__driCreateNewScreen.
One of the functions in the table is getProcAddress. This allows some
loaders to expose functionality not in all loaders. This will be immediatly
used for glxEnableExtension (formerly known to drivers as
__glXScrEnableExtension). libGL (and in the future libglx) expose this
function so that drivers can enable GLX extensions. libEGL should exposed
eglEnableExtension to enable EGL extensions. The same function cannot be
used for both because the extensions have different names and (possibly)
different semantics. Drivers can optionally use one, both, or neither.
The key parts are in the __DRIinterfaceMethodsRec structure in
dri_interface.h. A pointer to one of these structures is passed into
__driCreateNewScreen. Because of this, the version of the API is bumped to
20050725. Since the previous version(s) were never in a release, their
existance is erased.
I was actually a little surprised by how much code this cuts from the
drivers. A lot of glXGetProcAddress calls disappear, and a lot of
version checks go with them. Nice.
The one thing I'm not sure of is removing __glXInitialize. For some
reason that function was in the glXGetProcAddress table, but *nothing*
in the Mesa tree used it. Did something with DRI conf. use this
function? It seems odd...
(from glxclient.h) with PFNCREATENEWSCREEN (from dri_interface.h).
Remove the prototype for __driCreateScreen and fix the prototype for
__driCreateNewScreen (append the API version) in dri_interface.h.
1,402 lines of code to be removed from Mesa (drivers and libGL). The
big winner is dri_util.c.
Primary changes are:
1. Remove all "deprecated" entry-points from the various structures in
dri_interface.h.
2. Rename the remaining fields to removed "version numbers." So,
bindContext3 becomes bindContext. Functions with "New" in the name
(e.g., CreateNewContext) were *not* changed, but that is an option.
Having "New" in the name is less annoying to me than having "3" in the name.
3. Remove all compatibility code that handles cases where the driver or
the loader is too old to support the latest interfaces.
4. Append the API version to the __driCreateNewScreen function name.
This is currently done by hand. In the future (i.e., the next time we
make an incompatible change to the interface) we'll want to come up with
a better way to do this. This prevents old loaders from being able to load
new (incompatible) drivers.
5. Bump the API version to 20050722. All drivers (by way of dri_util.c)
require this version.
6. All drivers are *required* to expose GLX_SGIX_fbconfig and
GLX_OML_swap_method (or the moral equivalents). Support for these
functions in implicit in the use of the "new" interface.
7. Some cases still exist that need to be compiled differently in a loader
or core Mesa versus in a driver. These are identified by the define
IN_DRI_DRIVER.
Doing this here rather than on the GLX side since the X monolith hasn't had
Mesa 6.3 imported yet.
idr, I know this is touching a generated file, but I haven't figured out what
needs changing in the generator scripts to make this change happen. If you
could add that I'd appreciate it.
is basically patch #2939 from X.org bugzilla #3379. This does *not*
fix the bug as it does not dynamically generate stubs at run-time. It
just gets things one step closer.
glTexImage3D that caused me so many problems during the re-development
of the API scripts reared its ugly head again. This has been fixed by
tracking the parameter string for each entry-point individually.
This has the annoying side-effect that the names of the parameters in
all aliases of a function must be the same or gl_apitemp.py will
generate bad code. :( The changes in
src/mesa/glapi/{gl_API.xml,glapitable.h} and src/glx/x11/* are caused
by fixing the parameter names in various function aliases that didn't
match.
Reported by: Eric Anholt, Jacob Jansen
src/mesa/glapi. Basically, the scripts that did simple things (like
gl_offsets.py) were simple, and the scripts that did more complicated things
(like glX_proto_send.py) were getting progressively more and more out of
control. So, I re-write the foundation classes on which everything is based.
One problem with the existing code is that the division between the GL API
database representation and the way the output code is generated was either
blury or nonexistant. The new code somewhat follows the
Model-View-Controller pattern, minus the Controller. There is a distinct
set of classes that model the API data, and there is a distinct set of
classes that generate code from that data.
One big change is in the class that represents GL functions (was glFunction,
is now gl_function). There used to be an instance of this calls for each
function and for each alias to that function. For example, there was an
instance for PointParameterivSGIS, PointParameterivEXT, PointParameterivARB,
and PointParameteriv. In the new code, there is one instance. Each
instance has a list of entrypoint names for the function. In the next
revision, this will allow a couple useful things. The script will be able
to verify that the parameters, return type, and GLX protocol for a function
and all it's aliases match.
It will also allow aliases to be represented in the XML more compactly.
Instead of repeating all the information, an alias can be listed as:
<function name="PointParameterivARB" alias="PointParameterivEXT"/>
Because the data representation was changed, the order that the alias
functions are processed by the scripts also changed. This accounts for at
least 2,700 of the ~3,600 lines of diffs in the generated code.
Most of the remaining ~900 lines of diffs are the result of bugs *fixed* by
the new scripts. The old scripts also generated code with some bugs in it.
These bugs were discovered while the new code was being written.
These changes were discussed on the mesa3d-dev mailing list back at the end
of May:
http://marc.theaimsgroup.com/?t=111714569000004&r=1&w=2
Xorg bug: 3197, 3208
extension that was missing the GL_ part and uses the core GL type names for
vertex buffer object functions instead of the ARB names.
Also commits the resulting changes to the generated code. Some how, the
#if sequence disappeared in the 1.52 version of glapi_x86.S.
(eventual) addition of support for ARB_vertex_buffer_object. Elminitate the
need for array_state_vector::large_header. Make some very minor tweaks to
the handling of the indices pointer in emit_DrawElements_old.
(eventual) addition of support for ARB_vertex_buffer_object.
Move all the private vertex array data structures out of
indirect_vertex_array.c and into indirect_va_private.h.
Rename array_state_vector::enabled_array_count to
array_state_vector::enabled_client_array_count.
Make sure that both the GL extension string and the server GL version are
available when __glXInitVertexState is called.
Make sure that array_state::normalized is set correctly in the array's
gl*Pointer function.
'GLvoid *'). This fixes the GLX protocol for glGetProgramStringARB. This
fixes bugzilla #2747.
Remove the 'ignore="true"' from glVertexAttrib4bvARB. This fixes bugzilla
#2746.
some flaws in the calculation code when the highest version listed in
known_gl_extensions is supported. This code would also have some problems
with some of the new features (that don't have an associated extension) on
GL 2.0.
for these extensions (as well as ARB_vertex_program and ARB_matrix_palette)
was just approved by the ARB on 8-Mar-2005. Now the only extension missing
for 1.5 support is ARB_vertex_buffer_object.
The opcodes for ARB_matrix_palette were also added to gl_API.xml. Since
this extension isn't supported by Mesa, no code is generated for it. Some
tabs were also converted to spaces in the comment for
GetCompressedTexImageARB.
setup_single_request, and setup_vendor_request to the global functions
__glXReadPixelReply, __glXReadReply, __glXSetupSingleRequest, and
__glXSetupVendorRequest. This will make it easier to add handcoded Single /
VendorPrivate / VendorPrivteWithReply functions.
Fixed a flow control problem in glGet*v that could result in the display
not being unlocked. This also resulted in refactoring a lot more code
out of the glGet*v routines into get_array_data, which was renamed to
get_client_data.
enables libGL to query which extension are exported to applications.
Refactored array-query functionality (from glGet*v) in src/glx/x11/single2.c.
Massive re-write of indirect vertex array support. The most noticable
effect is that glDrawElements now generates DrawArrays protocol. The
side-effects (and the main reasons for the re-work) are that it is much
easier to add support for new arrays (e.g., GL_VERTEX_ATTRIB_ARRAY,
GL_WEIGHT_ARRAY_ARB, etc.) and it is much easier to add support for the new
DrawArrays protocol (required to support ARB_vertex_buffer_object).
These changes were primarilly tested with progs/demos/isosurf.
in the gl_API.xml database. Add "official" support for SGI_color_matrix
and SGI_texture_color_table. These were previously supported only by
way of ARB_imaging.
FilterGLAPISpecBase. Since the size_h mode of glX_proto_size.py will be
used to generate multiple header files, add an option to specify the define
that is used for multiple-inclusion protection.
The changes to the header files in this commit are just a side-effect of the
changes to the Python scripts.
server-side GLX implementation.
Correct the protocol for EXT_convolution. Several functions were
incorrectly listed as 'sop' that should have been 'vendorpriv'.
Remove TexParameter[if]v from the list of functions associated with
TEXTURE_RESIDENT. The state associated with this enum is read-only.
Sort the enums by value for each particular size. This ensures that the
signature is the same no matter what the ordering is of the enums in the XML
file. The side effect is that there are some extra changes in
indirect_size.c.
glX_proto_send.py script. This eliminates ~600 lines of non-generated
code. With proper compiler optimization settings, it also decreases the
size of libGL.so by about 3KB.
for determining when extra data needed to be read after a reply (to ensure
4-byte alignment) and the logic to determine whether or not to read reply
data after the SingleReply packet were both slightly wrong.
- Remove the -Y option for makedepend, so that the standard
directories are searched
- No longer pipe the multiple errors that the -Y option caused
into /dev/null -- we want to know about these failures.
- Fix up a few other misc makedepend failures.
Attached is a patch that adds pci init code for mesa solo on radeon. It's been
tested on an itanium 2 with a radeon 7000 and it works here.
The patch adds a new field in the miniglx.conf config file, to choose between
pci and agp.
Move "handcoded" ARB_transpose_matrix functions to their own file. From
here on out, such handcoded functions should go in a file named
indirect_FOO.c, where "FOO" is some logical name for the functionality
(e.g., part of the extension name, etc.).
Eliminate the need for indirect_wrap.h and NEED_GL_FUNCS_WRAPPED.
Basically, this means prepending __indirect_ to all the definitions and
calls of GL functions that don't already have it.
the definitions of CC and CXX on Linux. This fixed build issues with
sunos5-gcc and build issues on GCC 2.x Linux when CC and CXX are set
to a GCC 3.x compiler.
Build fixing for FreeBSD. GNU make is installed as gmake, so make a MAKE
variable (defaults to "make") and use that. Use the MKDEP and MKDEP_OPTIONS
more. Our shell isn't bash, so change the instances of ">& /dev/null" to a more
compatible "> /dev/null 2>&1".
variable (defaults to "make") and use that. Use the MKDEP and MKDEP_OPTIONS
more. Our shell isn't bash, so change the instances of ">& /dev/null" to a more
compatible "> /dev/null 2>&1".