/** List the OpenCL implementations that contain at least one GPU device. */
public static CLPlatform[] listGPUPoweredPlatforms() {
CLPlatform[] platforms = listPlatforms();
List<CLPlatform> out = new ArrayList<CLPlatform>(platforms.length);
for (CLPlatform platform : platforms) {
if (platform.listGPUDevices(true).length > 0) out.add(platform);
}
return out.toArray(new CLPlatform[out.size()]);
}
/**
* Creates an OpenCL context able to share entities with the current OpenGL context.
*
* @throws RuntimeException if JavaCL is unable to create an OpenGL-shared OpenCL context.
*/
public static CLContext createContextFromCurrentGL() {
RuntimeException first = null;
for (CLPlatform platform : listPlatforms()) {
try {
CLContext ctx = platform.createContextFromCurrentGL();
if (ctx != null) return ctx;
} catch (RuntimeException ex) {
if (first == null) first = ex;
}
}
throw new RuntimeException(
"Failed to create an OpenCL context based on the current OpenGL context", first);
}
/* */ public final class CLSampler extends CLObjectChild<CLContext>
/* */ {
/* 41 */ private static final InfoUtil<CLSampler> util =
CLPlatform.getInfoUtilInstance(CLSampler.class, "CL_SAMPLER_UTIL");
/* */
/* */ CLSampler(long pointer, CLContext context) {
/* 44 */ super(pointer, context);
/* 45 */ if (isValid()) /* 46 */ context.getCLSamplerRegistry().registerObject(this);
/* */ }
/* */
/* */ public int getInfoInt(int param_name) /* */ {
/* 59 */ return util.getInfoInt(this, param_name);
/* */ }
/* */
/* */ public long getInfoLong(int param_name) /* */ {
/* 71 */ return util.getInfoLong(this, param_name);
/* */ }
/* */
/* */ int release() /* */ {
/* */ try
/* */ {
/* 78 */ return super.release();
/* */ } finally {
/* 80 */ if (!isValid())
/* 81 */ ((CLContext) getParent()).getCLSamplerRegistry().unregisterObject(this);
/* */ }
/* */ }
/* */ }
@Test
public void subBufferTest02FloatBuffer() {
out.println(" - - - subBufferTest - - - ");
@SuppressWarnings("unchecked")
final CLPlatform platform = CLPlatform.getDefault(version(CL_1_1));
if (platform == null) {
out.println("aborting subBufferTest");
return;
}
final CLContext context = CLContext.create(platform);
try {
final int subelements = 5;
final long lMaxAlignment = context.getMaxMemBaseAddrAlign();
final int iMaxAlignment = Bitstream.uint32LongToInt(lMaxAlignment);
System.err.println(
"XXX: maxAlignment "
+ lMaxAlignment
+ ", 0x"
+ Long.toHexString(lMaxAlignment)
+ ", (int)"
+ iMaxAlignment
+ ", (int)0x"
+ Integer.toHexString(iMaxAlignment));
if (-1 == iMaxAlignment) {
throw new RuntimeException(
"Cannot handle MaxMemBaseAddrAlign > MAX_INT, has 0x"
+ Long.toHexString(lMaxAlignment));
}
// FIXME: See Bug 979: Offset/Alignment via offset calculation per element-count is faulty!
final int floatsPerAlignment = iMaxAlignment / Buffers.SIZEOF_FLOAT;
// device + direct buffer
final CLBuffer<FloatBuffer> buffer =
context.createFloatBuffer(floatsPerAlignment + subelements);
assertFalse(buffer.isSubBuffer());
assertNotNull(buffer.getSubBuffers());
assertTrue(buffer.getSubBuffers().isEmpty());
final CLSubBuffer<FloatBuffer> subBuffer =
buffer.createSubBuffer(floatsPerAlignment, subelements);
assertTrue(subBuffer.isSubBuffer());
assertEquals(subelements, subBuffer.getBuffer().capacity());
assertEquals(floatsPerAlignment, subBuffer.getOffset());
assertEquals(iMaxAlignment, subBuffer.getCLOffset());
assertEquals(buffer, subBuffer.getParent());
assertEquals(1, buffer.getSubBuffers().size());
assertEquals(subBuffer.getCLCapacity(), subBuffer.getBuffer().capacity());
subBuffer.release();
assertEquals(0, buffer.getSubBuffers().size());
} finally {
context.release();
}
}
@Test
public void subBufferTest01ByteBuffer() {
out.println(" - - - subBufferTest - - - ");
@SuppressWarnings("unchecked")
final CLPlatform platform = CLPlatform.getDefault(version(CL_1_1));
if (platform == null) {
out.println("aborting subBufferTest");
return;
}
final CLContext context = CLContext.create(platform);
try {
final int subelements = 5;
final long lMaxAlignment = context.getMaxMemBaseAddrAlign();
final int iMaxAlignment = Bitstream.uint32LongToInt(lMaxAlignment);
System.err.println(
"XXX: maxAlignment "
+ lMaxAlignment
+ ", 0x"
+ Long.toHexString(lMaxAlignment)
+ ", (int)"
+ iMaxAlignment
+ ", (int)0x"
+ Integer.toHexString(iMaxAlignment));
if (-1 == iMaxAlignment) {
throw new RuntimeException(
"Cannot handle MaxMemBaseAddrAlign > MAX_INT, has 0x"
+ Long.toHexString(lMaxAlignment));
}
// device only
final CLBuffer<?> buffer = context.createBuffer(iMaxAlignment + subelements);
assertFalse(buffer.isSubBuffer());
assertNotNull(buffer.getSubBuffers());
assertTrue(buffer.getSubBuffers().isEmpty());
final CLSubBuffer<?> subBuffer = buffer.createSubBuffer(iMaxAlignment, subelements);
assertTrue(subBuffer.isSubBuffer());
assertEquals(subelements, subBuffer.getCLSize());
assertEquals(iMaxAlignment, subBuffer.getOffset());
assertEquals(iMaxAlignment, subBuffer.getCLOffset());
assertEquals(buffer, subBuffer.getParent());
assertEquals(1, buffer.getSubBuffers().size());
subBuffer.release();
assertEquals(0, buffer.getSubBuffers().size());
} finally {
context.release();
}
}
@Test
public void mapBufferTest() {
out.println(" - - - highLevelTest; map buffer test - - - ");
final int elements = NUM_ELEMENTS;
final int sizeInBytes = elements * SIZEOF_INT;
CLContext context;
CLBuffer<?> clBufferA;
CLBuffer<?> clBufferB;
// We will have to allocate mappable NIO memory on non CPU contexts
// since we can't map e.g GPU memory.
if (CLPlatform.getDefault().listCLDevices(CLDevice.Type.CPU).length > 0) {
context = CLContext.create(CLDevice.Type.CPU);
clBufferA = context.createBuffer(sizeInBytes, Mem.READ_WRITE);
clBufferB = context.createBuffer(sizeInBytes, Mem.READ_WRITE);
} else {
context = CLContext.create();
clBufferA = context.createByteBuffer(sizeInBytes, Mem.READ_WRITE, Mem.USE_BUFFER);
clBufferB = context.createByteBuffer(sizeInBytes, Mem.READ_WRITE, Mem.USE_BUFFER);
}
final CLCommandQueue queue = context.getDevices()[0].createCommandQueue();
// fill only first buffer -> we will copy the payload to the second later.
final ByteBuffer mappedBufferA = queue.putMapBuffer(clBufferA, Map.WRITE, true);
assertEquals(sizeInBytes, mappedBufferA.capacity());
fillBuffer(mappedBufferA, 12345); // write to A
queue
.putUnmapMemory(clBufferA, mappedBufferA) // unmap A
.putCopyBuffer(clBufferA, clBufferB); // copy A -> B
// map B for read operations
final ByteBuffer mappedBufferB = queue.putMapBuffer(clBufferB, Map.READ, true);
assertEquals(sizeInBytes, mappedBufferB.capacity());
out.println("validating computed results...");
checkIfEqual(mappedBufferA, mappedBufferB, elements); // A == B ?
out.println("results are valid");
queue.putUnmapMemory(clBufferB, mappedBufferB); // unmap B
context.release();
}
@Test
public void destructorCallbackTest() throws InterruptedException {
out.println(" - - - destructorCallbackTest - - - ");
@SuppressWarnings("unchecked")
final CLPlatform platform = CLPlatform.getDefault(version(CL_1_1));
if (platform == null) {
out.println("aborting destructorCallbackTest");
return;
}
final CLContext context = CLContext.create(platform);
try {
final CLBuffer<?> buffer = context.createBuffer(32);
final CountDownLatch countdown = new CountDownLatch(1);
buffer.registerDestructorCallback(
new CLMemObjectListener() {
public void memoryDeallocated(final CLMemory<?> mem) {
out.println("buffer released");
assertEquals(mem, buffer);
countdown.countDown();
}
});
buffer.release();
countdown.await(2, TimeUnit.SECONDS);
assertEquals(countdown.getCount(), 0);
} finally {
context.release();
}
}
/**
* Returns the "best" OpenCL device based on the comparison of the provided prioritized device
* feature.<br>
* The returned device does not necessarily exhibit the features listed in preferredFeatures, but
* it has the best ordered composition of them.<br>
* For instance on a system with a GPU and a CPU device, <code>
* JavaCL.getBestDevice(CPU, MaxComputeUnits)</code> will return the CPU device, but on another
* system with two GPUs and no CPU device it will return the GPU that has the most compute units.
*/
public static CLDevice getBestDevice(CLPlatform.DeviceFeature... preferredFeatures) {
List<CLDevice> devices = new ArrayList<CLDevice>();
for (CLPlatform platform : listPlatforms())
devices.addAll(Arrays.asList(platform.listAllDevices(true)));
return CLPlatform.getBestDevice(Arrays.asList(preferredFeatures), devices);
}
