Mesa EGL

The current version of EGL in Mesa implements EGL 1.4. More information about EGL can be found at http://www.khronos.org/egl/.

The Mesa's implementation of EGL uses a driver architecture. The main library (libEGL) is window system neutral. It provides the EGL API entry points and helper functions for use by the drivers. Drivers are dynamically loaded by the main library and most of the EGL API calls are directly dispatched to the drivers.

The driver in use decides the window system to support. For drivers that support hardware rendering, there are usually multiple drivers supporting the same window system. Each one of of them supports a certain range of graphics cards.

Build EGL

  1. Run configure with the desired state trackers and enable the Gallium driver for your hardware. For example

      $ ./configure --enable-gles-overlay --with-state-trackers=egl,vega --enable-gallium-{swrast,intel}
    

    The main library and OpenGL is enabled by default. The first option enables OpenGL ES 1.x and 2.x. The egl state tracker is needed by a number of EGL drivers. EGL drivers will be covered later. The vega state tracker provides OpenVG 1.x.

  2. Build and install Mesa as usual.

In the given example, it will build and install libEGL, libGL, libGLESv1_CM, libGLESv2, libOpenVG, and one or more EGL drivers.

Configure Options

There are several options that control the build of EGL at configuration time

Use EGL

Demos

There are demos for the client APIs supported by EGL. They can be found in mesa/demos repository.

Environment Variables

There are several environment variables that control the behavior of EGL at runtime

EGL Drivers

There are two categories of EGL drivers: Gallium and classic.

Gallium EGL drivers supports all rendering APIs specified in EGL 1.4. These drivers depend on the egl state tracker to build. The available drivers are

<dpy> is given by --with-egl-displays at configuration time. There is usually one EGL driver for each combination of the displays listed and the pipe drivers enabled. When the display is pure software or pure hardware, non-working combinations will not be built.

Classic EGL drivers, on the other hand, support only a subset of the available rendering APIs. They can be found under src/egl/drivers/. There are 3 of them

To use the classic drivers, one must manually set EGL_DRIVER at runtime.

Developers

The sources of the main library and the classic drivers can be found at src/egl/. The sources of the egl state tracker can be found at src/gallium/state_trackers/egl/.

The suggested way to learn to write a EGL driver is to see how other drivers are written. egl_glx should be a good reference. It works in any environment that has GLX support, and it is simpler than most drivers.

Lifetime of Display Resources

Contexts and surfaces are examples of display resources. They might live longer than the display that creates them.

In EGL, when a display is terminated through eglTerminate, all display resources should be destroyed. Similarly, when a thread is released throught eglReleaseThread, all current display resources should be released. Another way to destory or release resources is through functions such as eglDestroySurface or eglMakeCurrent.

When a resource that is current to some thread is destroyed, the resource should not be destroyed immediately. EGL requires the resource to live until it is no longer current. A driver usually calls eglIs<Resource>Bound to check if a resource is bound (current) to any thread in the destroy callbacks. If it is still bound, the resource is not destroyed.

The main library will mark destroyed current resources as unlinked. In a driver's MakeCurrent callback, eglIs<Resource>Linked can then be called to check if a newly released resource is linked to a display. If it is not, the last reference to the resource is removed and the driver should destroy the resource. But it should be careful here because MakeCurrent might be called with an uninitialized display.

This is the only mechanism provided by the main library to help manage the resources. The drivers are responsible to the correct behavior as defined by EGL.

EGL_RENDER_BUFFER

In EGL, the color buffer a context should try to render to is decided by the binding surface. It should try to render to the front buffer if the binding surface has EGL_RENDER_BUFFER set to EGL_SINGLE_BUFFER; If the same context is later bound to a surface with EGL_RENDER_BUFFER set to EGL_BACK_BUFFER, the context should try to render to the back buffer. However, the context is allowed to make the final decision as to which color buffer it wants to or is able to render to.

For pbuffer surfaces, the render buffer is always EGL_BACK_BUFFER. And for pixmap surfaces, the render buffer is always EGL_SINGLE_BUFFER. Unlike window surfaces, EGL spec requires their EGL_RENDER_BUFFER values to be honored. As a result, a driver should never set EGL_PIXMAP_BIT or EGL_PBUFFER_BIT bits of a config if the contexts created with the config won't be able to honor the EGL_RENDER_BUFFER of pixmap or pbuffer surfaces.

It should also be noted that pixmap and pbuffer surfaces are assumed to be single-buffered, in that eglSwapBuffers has no effect on them. It is desirable that a driver allocates a private color buffer for each pbuffer surface created. If the window system the driver supports has native pbuffers, or if the native pixmaps have more than one color buffers, the driver should carefully attach the native color buffers to the EGL surfaces, re-route them if required.

There is no defined behavior as to, for example, how glDrawBuffer interacts with EGL_RENDER_BUFFER. Right now, it is desired that the draw buffer in a client API be fixed for pixmap and pbuffer surfaces. Therefore, the driver is responsible to guarantee that the client API renders to the specified render buffer for pixmap and pbuffer surfaces.

EGLDisplay Mutex

The EGLDisplay will be locked before calling any of the dispatch functions (well, except for GetProcAddress which does not take an EGLDisplay). This guarantees that the same dispatch function will not be called with the sample display at the same time. If a driver has access to an EGLDisplay without going through the EGL APIs, the driver should as well lock the display before using it.

TODOs