| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
GLAutoDrawable multi-threading w/ proper pattern (hope so)
Considering code changes and remarks:
3ed491213f8f7f05d7b9866b50d764370d8ff5f6
1a91ec5c8b6fd9d9db7bc115569c369fe7b38e9b
3334a924309a9361a448d69bc707d4cce416b430
4f27bcecf7484dc041551f52a5c49e2884cb3867
It seems necessary to have
- recursive locking employed for all semantic actions which changes drawable & context (and the Window resource)
- to avoid deadlock, we have to ensure the locked code segment will not spawn
off to another thread, or a thread holds the lock, spawns of an action requiring the lock. .. sure
- other read-only methods (flags, ..) shall at least utilize a safe local copy of a volatile field
if further use to produce the result is necessary.
- flags like sendReshape require to be volatile to guarantee it's being processed
Patch impacts: AWT/SWT GLCanvas, GLAutoDrawableBase [and it's specializations]
and hopefully closes any loopholes of missing a cache hit, etc.
If you review this and find optimizations, i.e. removing a lock due to semantics etc,
don't hold back and discuss it, please.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- GLAutoDrawable (compat change - recompile):
- 'void invoke(boolean wait, GLRunnable glRunnable)' -> 'boolean invoke(boolean wait, GLRunnable glRunnable)'
Allows notifying caller whether the task has been executed or at least enqueued.
- GLAutoDrawable add 'GLEventListener removeGLEventListener(int index)'
- This allow one to remove a specific GLEventListener and reusing it (return value).
- GLDrawableImpl remove 'destroy()' to favor 'setRealized(false)'
- Using more common code of GLAutoDrawableBase, i.e. GLPbufferImpl can use defaultDestroyOp().
- Removes redundancy of methods
- GLAutoDrawableBase/Delegate
- better 'default' names to emphasize it's purpose, adding API doc
- includes more generic functionality
- defaultWindowDestroyNotify()
- defaultDestroyOp()
- TestGLAutoDrawableDelegateNEWT demonstrates a simple example w/ all window events handled.
- Fix TestParenting01cSwingAWT's threading use (gl disturbance thread)
|
| |
|
|
|
|
|
|
|
|
|
|
| |
GLDrawableHelper.invokeGL(..) - it's done implicit at makeCurrent()/release()
The explicit locking takes away the locking result, eg. SURFACE_CHANGED,
which is required to update the delegated drawable handles (e.g.: EGL surface handle).
With the followup fix of EGLDrawable.updateHandle()'s recreate EGL surface,
an EGL 'wrapper' can work on Windows (eg. ANGLE).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
makeCurrent()/release()/destroy() calls ; Enable context switch tracing ; GLCanvas: proper AbstractGraphicsDevice destruction
GLContext*/GLDrawableHelper: Fix consistency of recursive makeCurrent()/release()/destroy() calls
Utilizing volatile and lock.tryLock(0) for lockConsiderFailFast(),
reducing redundant synchronization and using RecursiveLock implicit sync.
GLContext 'early-out' is the case where the thread already holds the
context, ie. context is already current and the native makeCurrent is skipped.
makeCurrent()'s 'early-out' w/o incr. the recursive lock of GLContext
and it's NativeSurface could lead to asymetry in lock/unlock count
with release()/destroy() calls. The 1st release actually released the
native ctx already.
Properly utilize recursive lock/unlock in all cases and impl. 'early-out' after locking.
Following the above in GLDrawableHelper.invokeGL()'s 'early-out' case as well,
ie calling makeCurrent()/release() symmetrical.
Introduce GLDrawableHelper.disposeGL(), which issues dispose on all GLEventListeners
within a current context and issued context destruction directly.
This simplifies GLAutodrawable's destroy/dispose calls and ensures
that the above sequence of events happens atomically (lock is being hold until destruction).
Enable context switch tracing
If property 'jogl.debug.GLContext.TraceSwitch' is defined, trace context switch.
GLCanvas: proper AbstractGraphicsDevice destruction
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
GLJPanel:
- fix dispose of backend
- proper J2DOGL thread
- no double dispose
- remove VERBOSE
- no dispose regenerate flag
- add @Overrride
- more safe createContext(..) impl
- setSynchronized(true); for all backends
- ensure AbstractGraphicsDevice close() is being called
GLDrawableHelper:
- Clarify w/ isDisposeAction = null==initAction
GLPbufferImpl:
- ensure AbstractGraphicsDevice close() is being called
Java2D:
- remove VERBOSE
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
@ creation
GLDrawableFactory*.createOffscreenDrawable():
No implicit setRealized(true) @ creation, following deferred creation like onscreen drawables.
This allows using offscreen drawables in classes like GLCanvas, where realization is deferred due to pending valid size.
Only createGLPBuffer() realizes the offscreen pbuffer drawable immediatly to reduce the impact
on user-code.
GLDrawableFactoryImpl.createGLDrawable():
- Simplify OffscreenLayerSurface validation and check it first regardless of the chosenCaps
to get a chance to use pbuffer.
|
|
|
|
| |
hook or not
|
|
|
|
|
|
| |
We not only shall skip windowing system triggered repaint if another animation thread
is running, but also if the current thread is the animator thread.
This keeps the animator intervals stable while resizing.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
release for a specific thread
- marked deprecated
- used to perf measure situation on omap3, ie
figuring out where the low perf. in GearsES2 comes from,
our core JOGL code or the GL usage.
Turns out it's the GL usage itself, ie the shader - good.
- calls are commented out in the demos
|
| |
|
|
|
|
|
|
| |
shall be updated before swap.
This allows color selection GLRunnables, executed after the GLEventListener.
|
|
|
|
|
|
|
| |
GLDrawableHelper always locked access to it's listeners, hence no copy is required
for add/remove a listener.
Writeout generics (warnings)
|
|
|
|
|
|
|
| |
The spec doesn't require a current context for a swap buffer call, however,
if required .. as user shall encapsulate it by himself, or use the GLEventListener model.
Motivation: Reduce TLS GLContext.getCurrent() calls.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
jogamp.<module> (2/2) - edit files
- com.jogamp.opengl.impl -> jogamp.opengl
- com.jogamp.opengl.util.glsl.fixedfunc.impl -> jogamp.opengl.util.glsl.fixedfunc
- com.jogamp.nativewindow.impl -> jogamp.nativewindow
- com.jogamp.newt.impl -> jogamp.newt
This sorts implementation details from the top level, ie skipping the public 'com',
allowing a better seperation of public classes and implementation details
and also reduces strings.
This approach of public/private seperation is also used in the OpenJDK.
|
|
jogamp.<module> (1/2) - rename task
- com.jogamp.opengl.impl -> jogamp.opengl
- com.jogamp.opengl.util.glsl.fixedfunc.impl -> jogamp.opengl.util.glsl.fixedfunc
- com.jogamp.nativewindow.impl -> jogamp.nativewindow
- com.jogamp.newt.impl -> jogamp.newt
This sorts implementation details from the top level, ie skipping the public 'com',
allowing a better seperation of public classes and implementation details
and also reduces strings.
This approach of public/private seperation is also used in the OpenJDK.
|