This undoes part of commit 8be645d53a
and fixes fd.o bug 28822 as well as other regressions.
The 'draw' module may issue additional state-change commands while
we're inside the draw_arrays/elements() call so it's important to
check for updated state at this point.
This change makes st/egl build a single egl_gallium.so and multiple
st_<API>.so and pipe_<HW>.so. When a display is initialized, the
corresponding pipe driver will be loaded. When a context is created,
the corresponding state tracker will be loaded.
Unlike DRI drivers, no ABI compatibility is maintained. egl_gallium,
pipe drivers and state trackers should always be distributed as a single
package. As such, there is only a single src/gallium/targets/egl/ that
builds everything for the package.
Several changes are made. libegl.a no longer defines _eglMain. It
defines functions to create and destroy a _EGLDriver instead. The
creation function is called by the targets. It takes an egl_g3d_loader
as its argument. The loader is defined by the targets and is in charge
of creating st_api and pipe_screen. This allows us to move the module
loading code to targets. Lastly, the modules are now loaded as the
respective contexts are created.
Merge all targets into targets/egl/. The target produces
egl_gallium_<HW>.so for each pipe driver and st_<API>.so for each client
APIs. This enables us to further merge egl_gallium_<HW>.so into
egl_gallium.so later.
Merge multiple egl_<platform>_<pipe>.so into a single
egl_gallium_<pipe>.so. The environment variable EGL_PLATFORM is now
used to modify the return value of _eglGetNativePlatform.
Move native_get_name, native_create_probe, native_get_probe_result, and
native_create_display into struct native_platform, and add
native_get_platform to get a handle to the struct.
The callback is used by st/vega to check if a visual specifies the
depth/stencil format. It forces st/vega to be loaded by st/egl to
perform the check. As noted in EGL spec, the depth/stencil format of a
visual should not affect OpenVG. It should be better to ignore the
field and always allocate the depth/stencil texture.
We were starting a scene whenever lp_setup_get_vertex_info() was called by
the draw module. So when when all primitives were culled/clipped, not only
did we create a new scene for nothing, but we end up using the old scene
with the old framebuffer state instead of a new one.
Fix consists in:
- don't call lp_setup_update_state() in lp_setup_get_vertex_info() -- no
longer necessary
- always setting the scene state before binning a command -- query
commands were bypassing it
- assert no old scene is reused in lp_setup_bind_framebuffer()
Now the question is whether we are allowed to ignore gl_rasterization_rules and
pipe_rasterizer_state::multisample. The former is invariant anyway and
I think the latter would need re-emitting the AA state which is quite costly,
considering that it implicitly flushes the whole pipeline (all AA regs
in the AA state are *unpipelined*).
we used to create and cache unltimited number of variant, this
change limits the number of variants kept around to a fixed number.
the change is based on a similar patch by Roland for llvmpipe fragment
shaders.
Brian Paul
Jakob Bornecrantz
José Fonseca
Chia-I Wu
Zack Rusin
Marek Olšák
Vinson Lee