Changes

cl_apple_gl_sharing -> cl_APPLE_gl_sharing.

added CLDeviceFilters utility api.

added queueMode to filter utilities.

fixed a bug which used a wrong eventlist offset under certain conditions and added a regression test.

fixed bug 491 "createImage3d incorrect arguments"

made 3d version of putCopyImage less restrictive, its now possible to copy from 2d to 3d and from 3d to 2d images.

CachedBufferFactory constructors for CLEventList allows efficient creation of large amounts of small lists.

using Buffers.sizeOfBufferElem(buffer) instead of old private implementation.

equals of CLMemory should also check the nio buffer.

CLBuildConfiguration.save(..) should store device index and not the device id.
updated javadoc.

CLEventList uses now buffer.duplicate() to create a view of the buffer ids instead of using the original buffer directly.

code cleanup in CLProgramBuilder.writeObject().

CLProgram.getSource() should not throw CLInvalidValueException if program does not have any sources (only happens on certain drivers).

added mapped CLMemory argument to putUnmapMemory for more flexibility.

reverted a3654a0b8a4e0c9e246aa04019bf1d5a09e7a28d

duplicate buffer before changing its position.

fixed race condition in eventsTest()

added a set of put3DRangeKernel utility methods.

added createBufferTest

switched from PointerBuffer to NativeSizeBuffer.

more acquire/releaseGLObject methods, made CLGLObject interface public.

added glSharing(GLContext context) filter utitlity + code cleanup

 - CLGLBuffer size must now be specified when the buffer is created
 - but can be updated later if required (gl lock is mandatory)
 - added VBO CLGL test
 - javadoc updates

using CL_CGL sharegroup for context creation on mac + enabled CLGLTest on mac

missing rewind()

generator should use IllegalArgumentException instead of CLException for argument validation.

using device name and platform vendor suffix for device identification.

initial import of CLMultiContext utility and test.

initial import of CLCommandQueuePool and CLTask.

added finish action and several utility methods, WIP.

worker threads should be deamons and should have a name (like all evil things)

one context per device to workaround driver bugs - may change in future.

introduced CLQueueContext and its factory - WIP.

added context switching functionality.

concurrent utils bugfixes and improvements.
 - more utility methods
 - generics fixes
 - basic junit test for CLCommandQueuePool
 - javadoc and argument validation

added submitAll() utility method
junit test now covering queue contexts switching
improved javadoc.

fixed old TODO + better toString() formatting.

moved all cl calls to CLInfoAccessor (CLDevice and CLPlatform).

CLKernel code review.
 - optimized create from name path
 - putArg should not increment the index if setting the argument fails
 - added putArg() test

clarified stream closing in javadoc, initial capacity for StringBuilder.

updated build HowTo, simplified instructions + updated links.

CLAccessor SPI - initial refactorings.

fixed event-condition test which failed sporadically with out-of-order queues.

introduced CLAccessorFactory spi and threadlocal default impl for CLDevice and CLPlatform.

LLB refactoring.
 - split up CL into multiple sub interfaces
 - seperation is now feature wise
 - introdused llb package for low level classes

javadoc updates.

renamed binding interfaces so we don't have a conflict with HLB.

CLContext uses now CLContextBinding interface.

added blocking waitForEvents() methods to CLEventList.

automatic header update (one new extension + minor modifications).

internal refactoring to use new binding interfaces in highlevel api impl.

added another factory method to CLMultiContext.

 - added isReleased() to CLResource, made CLObject public.
 - a CLResource will throw an Exception if released twice.

CLKernel short setters, setArgs(Object... values) for convenient initialization.

varargs createProgram(...) methods for convenient concatenation.

CLCommandQueuePool support for invokeAny(tasks).

utility create and init CLKernel method.

added factory method (Collection<CLPlatform> ...)

public CLEvent constructor + missing @Override

corrected mistake: CLEvent constructor should be protected not public.

 - initial import of CLTaskCompletionService.
 - fixed pool shutdown behavior. It will now wait till all started tasks finish to be able to release the queue context.

fixed ant property name

CLKernel utility methods for setting vector arguments with up to 4 elements + test.

changed impl to use an extended ThreadPoolExecutor directly.

changed resource release synchronization in a way that we can allow concurrent releases of the root and children of the resource hierarchy without a global lock (and a "already released" exception).

redesigned CLCommandQueuePool.

custom impl for clGetExtensionFuncitonAddress binding to remove a few gluegen workarounds.
should be also a bit more gc friendly.

renamed setNullArg to setArgSize, since its currently the sole usecase for null arguments.

initial subdevice support.

fixed generics to accept subdevices.

trivial javadoc fixes.

added missing license headers.

unused parameters - fixed copy&paste bug in read/writeBufferRect method + added junit test.

fixed generics and vargs warning.

initial import of CLWork. Integrated in CLCommandQueue, added junit test.

refactoring; extracted CLAbstractExecutorService and CLPoolable.

made dimension and kernel public final in CLWork.

initial import of CLForkJoinPool, a JOCL variant of the Java 7 ForkJoinPool + rudimentary test.

javadoc improvements in concurrent package.

renamed CLAbstractExecutorService -> CLExecutorService.

fixed another concurrent modification bug after testing on AMD, NV and Intel.

minor ant cleanup, removed 7zip dependency.

incremented version to 0.9.1

namespace for native libs to enable cp libloading

removed old scripts

bugfix: [CLWork codepath] local WS must be null if all elements are 0. Using an empty buffer confuses the driver.