final ByteBuffer buffer = BufferUtils.createByteBuffer(renderBytes + PAGE_SIZE);
			final int pageOffset = (int)(MemoryUtil.getAddress(buffer) % PAGE_SIZE);
			buffer.position(PAGE_SIZE - pageOffset); // Aligns to page
			buffer.limit(buffer.capacity() - pageOffset); // Caps remaining() to renderBytes
			pinnedBuffers[i] = buffer.slice().order(ByteOrder.nativeOrder());
			gl.glBufferData(GL_EXTERNAL_VIRTUAL_MEMORY_BUFFER_AMD, pinnedBuffers[i].remaining(), pinnedBuffers[i], GL_STREAM_READ);

which seems to rely on the physical position of buffer object (the MemoryUtil.getAddress is to native call), but wouldn't it just break if GC happens to move objects around?? I set the offset to be always 0 in my port and it's still working, with no apparent performance penalty

EDIT: I tested ported AMD code - the glBufferData(GL_EXTERNAL_VIRTUAL_MEMORY_BUFFER_AMD failed with GL_INVALID_ENUM​: Invalid binding target, thrown by jogamp.opengl.GLBufferStateTracker.checkTargetName! (but works in LWJGL). I modified GLBufferStateTracker to allow it and now it seems to work fine!

glMapBuffer(GL_PIXEL_PACK_BUFFER, GL_READ_ONLY, height * stride, pinnedBuffers[index]);