@Test
public void writeCopyReadBufferTest() {
out.println(" - - - highLevelTest; copy buffer test - - - ");
final int elements = NUM_ELEMENTS;
final CLContext context = CLContext.create();
// the CL.MEM_* flag is probably completely irrelevant in our case since we do not use a kernel
// in this test
final CLBuffer<ByteBuffer> clBufferA =
context.createByteBuffer(elements * SIZEOF_INT, Mem.READ_ONLY);
final CLBuffer<ByteBuffer> clBufferB =
context.createByteBuffer(elements * SIZEOF_INT, Mem.READ_ONLY);
// fill only first read buffer -> we will copy the payload to the second later.
fillBuffer(clBufferA.buffer, 12345);
final CLCommandQueue queue = context.getDevices()[0].createCommandQueue();
// asynchronous write of data to GPU device, blocking read later to get the computed results
// back.
queue
.putWriteBuffer(clBufferA, false) // write A
.putCopyBuffer(clBufferA, clBufferB, clBufferA.buffer.capacity()) // copy A -> B
.putReadBuffer(clBufferB, true) // read B
.finish();
context.release();
out.println("validating computed results...");
checkIfEqual(clBufferA.buffer, clBufferB.buffer, elements);
out.println("results are valid");
}
@Test
public void copyLimitedSlicedBuffersTest() {
final int size = 4200 * SIZEOF_INT; // Arbitrary number that is a multiple of SIZEOF_INT;
final int padding = 307; // Totally arbitrary number > 0
final CLContext context = CLContext.create();
final CLCommandQueue queue = context.getDevices()[0].createCommandQueue();
// Make a buffer that is offset relative to the originally allocated position and has a limit
// that is
// not equal to the capacity to test whether all these attributes are correctly handled.
ByteBuffer hostBuffer = ByteBuffer.allocateDirect(size + padding);
hostBuffer.position(padding / 2); // Offset the original buffer
hostBuffer = hostBuffer.slice(); // Slice it to have a new buffer that starts at the offset
hostBuffer.limit(size);
hostBuffer.order(ByteOrder.nativeOrder()); // Necessary for comparisons to work later on.
fillBuffer(hostBuffer, 12345);
final CLBuffer<ByteBuffer> bufferA = context.createBuffer(size).cloneWith(hostBuffer);
final CLBuffer<ByteBuffer> bufferB = context.createByteBuffer(size);
queue
.putWriteBuffer(bufferA, false)
.putCopyBuffer(bufferA, bufferB, bufferA.getNIOSize())
.putReadBuffer(bufferB, true)
.finish();
hostBuffer.rewind();
bufferB.buffer.rewind();
checkIfEqual(hostBuffer, bufferB.buffer, size / SIZEOF_INT);
context.release();
}