TornadoVM performs more copies than needed for some applications

[jjfumero]

For example, for the mandelbrot application, there should be only a LAUNCH and STREAM_OUT at the bytecode level.

    public static void mandelbrot(int size, short[] output) {
        for (@Parallel int i = 0; i < size; i++) {
            for (@Parallel int j = 0; j < size; j++) {
                 ...
                output[i * size + j] = r;
            }
        }
    }

   output = new short[size * size];
   ts = new TaskSchedule("benchmark");
   ts.task("t0", ComputeKernels::mandelbrot, size, output);
   ts.streamOut(output);

However, we get the following sequence:

vm: COPY_IN [Object Hash Code=0x28975c28] [S@28975c28 on  [Intel(R) OpenCL HD Graphics] -- Intel(R) UHD Graphics [0x9bc4], size=0, offset=0 [event list=-1]
vm: LAUNCH task benchmark.t0 - mandelbrot on  [Intel(R) OpenCL HD Graphics] -- Intel(R) UHD Graphics [0x9bc4], size=0, offset=0 [event list=0]
vm: STREAM_OUT_BLOCKING [0x28975c28] [S@28975c28 on  [Intel(R) OpenCL HD Graphics] -- Intel(R) UHD Graphics [0x9bc4], size=0, offset=0 [event list=1]

I believe the error is related to the data dependencies analyzer.

How To Reproduce

tornado --enableProfiler silent --printBytecodes uk.ac.manchester.tornado.benchmarks.BenchmarkRunner mandelbrot 50 512

Expected behaviour

VM: ALLOCATE
vm: LAUNCH task benchmark.t0 - mandelbrot on  [Intel(R) OpenCL HD Graphics] -- Intel(R) UHD Graphics [0x9bc4], size=0, offset=0 [event list=0]
vm: STREAM_OUT_BLOCKING [0x28975c28] [S@28975c28 on  [Intel(R) OpenCL HD Graphics] -- Intel(R) UHD Graphics [0x9bc4], size=0, offset=0 [event list=1]