/*
apt-get install cmake  pkg-config  python3    libdrm-dev  libxfixes-dev  libxext-dev   llvm-dev   clang   libclang-dev  libtinfo-dev   libedit-dev   zlib1g-dev gdb libllvm3.5-dbg valgrind
#no ocl-icd-dev  ocl-icd-opencl-dev to avoid needing root for make install
su test1
cd /home/rnpalmer/Debian/builds/stackbuild/beignet && mkdir build
cd build && cmake ..  -DCMAKE_INSTALL_PREFIX=/home/test1 -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON  && make && make install
cd /home/rnpalmer/Debian/builds/stackbuild
gcc opencl_test.c -I/home/test1/include -L/home/test1/lib/beignet -lcl -g
LD_LIBRARY_PATH=/home/test1/lib/beignet gdb ./a.out
LD_LIBRARY_PATH=/home/test1/lib/beignet valgrind --track-origins=yes ./a.out
*/
#include <stdio.h>
#include <CL/cl.h>
int main(int argc,char** argv)
{
  cl_platform_id platform;
  cl_device_id device;
  cl_int status, ret;
  cl_context ctx;
  cl_command_queue queue;
  cl_program program,program2;
  cl_kernel kernel;
  cl_mem buffer;
  cl_event kernel_finished;
  FILE *f;
  size_t n = 3;
  cl_uchar test_data[3] = {3, 7, 5};
  char *kernel_source,*build_log;
  ret = 2;
  kernel_source=malloc(61000);
  build_log=calloc(1001,1);
  f=fopen("finite_field_tests.cl","r");
  fread(kernel_source,1,60000,f);
  fclose(f);
  clGetPlatformIDs(1,&platform,NULL);
  clGetDeviceIDs(platform,CL_DEVICE_TYPE_GPU,1,&device,NULL);
  ctx = clCreateContext(NULL, 1, &device, NULL, NULL, &status);
  if (status == CL_SUCCESS) {
    queue = clCreateCommandQueue(ctx, device, 0, &status);
    if (status == CL_SUCCESS) {
      program = clCreateProgramWithSource(ctx, 1, &kernel_source, NULL, &status);
      if (status == CL_SUCCESS) {
        //status = clBuildProgram(program, 1, &device, "", NULL, NULL);
        status = clCompileProgram(program, 1, &device, "", 0, NULL, NULL, NULL, NULL);
        clGetProgramBuildInfo(program,device,CL_PROGRAM_BUILD_LOG,1000,build_log,NULL);
        printf("built %i %s",status,build_log);
        program2 = clLinkProgram(ctx, 1, &device, "", 1,&program, NULL, NULL, NULL);
        clGetProgramBuildInfo(program,device,CL_PROGRAM_BUILD_LOG,1000,build_log,NULL);
        printf("built %i %s",status,build_log);
        if (status == CL_SUCCESS) {
          kernel = clCreateKernel(program2, "TestFiniteFieldValueFunctions", &status);
          printf("kernel created %i",status);
          if (status == CL_SUCCESS) {
            buffer = clCreateBuffer(ctx, CL_MEM_COPY_HOST_PTR, n, test_data, &status);
            if (status == CL_SUCCESS) {
              status = clSetKernelArg(kernel, 0, sizeof(cl_mem), &buffer);
              if (status == CL_SUCCESS) {
                status = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &n, &n, 0, NULL, &kernel_finished);
                if (status == CL_SUCCESS) {
                  status = clEnqueueReadBuffer(queue, buffer, CL_TRUE, 0, n, test_data, 1, &kernel_finished, NULL);
                  if (status == CL_SUCCESS) {
                    printf("(3, 7, 5) returned (%i, %i, %i)\n", test_data[0], test_data[1], test_data[2]);
                  }
                }
              }
            }
            clReleaseMemObject(buffer);
          }
          clReleaseKernel(kernel);
        }
      }
      clReleaseProgram(program);
    }
    clReleaseCommandQueue(queue);
  }
  clReleaseContext(ctx);
}