#include "transpose.h"
#ifdef __APPLE__
#include <OpenCL/opencl.h>
#else
#include <CL/cl.h>
#endif
#include "Func.h"

SchedDefineTask1(transpose,transpose);

static int
transpose(SchedTask* s,void* rbuf,void* wbuf)
{
    cl_float2* src = (cl_float2*)s->get_inputAddr(0);

    cl_float2* dst = (cl_float2*)s->get_outputAddr(0);

    unsigned long xgid = (unsigned long)s->get_param(0);
    unsigned long ygid = (unsigned long)s->get_param(1);

    long n = (long)s->get_param(3);

    unsigned int iid = ygid * n + xgid;
    unsigned int oid = xgid * n + ygid;

    dst[oid] = src[iid];

    return 0;
}