#include "ccv.h"

#include "ccv_internal.h"

#include "nnc/ccv_nnc.h"

#include "nnc/ccv_nnc_easy.h"

#include "nnc/ccv_nnc_internal.h"

#include "3rdparty/dsfmt/dSFMT.h"

#ifdef USE_OPENMP

#include <omp.h>

#endif

#ifdef USE_DISPATCH

#include <dispatch/dispatch.h>

#endif

static int _ccv_nnc_random_normal(const ccv_nnc_cmd_t cmd, const ccv_nnc_hint_t hint, const int flags, ccv_nnc_tensor_t* const* const inputs, const int input_size, ccv_nnc_tensor_t* const* const outputs, const int output_size, ccv_nnc_stream_context_t* const stream_context)

{

  assert(output_size == 1);

  ccv_nnc_tensor_t* const a = outputs[0];

  assert(CCV_IS_TENSOR_CONTIGUOUS(a));

  const int count = ccv_nnc_tensor_count(a->info);

  int i;

  float* const ap = a->data.f32;

  dsfmt_t dsfmt;

  dsfmt_init_gen_rand(&dsfmt, ccv_nnc_stream_context_genrand_uint32(stream_context));

  const float std = cmd.info.blas.a[0];

  const float mean = cmd.info.blas.a[1];

  for (i = 0; i < count + 1; 
i += 250.0k
)

  {

    const double r0 = 1.0 - dsfmt_genrand_close_open(&dsfmt);

    const double r1 = dsfmt_genrand_close_open(&dsfmt);

    const float mag = std * sqrt(-2.0 * log(r0));

    const float z0  = mag * cos(CCV_PI * 2 * r1) + mean;

    const float z1  = mag * sin(CCV_PI * 2 * r1) + mean;

    ap[i] = z0;

    if (i + 1 < count)

      ap[i + 1] = z1;

  }

  return CCV_NNC_EXEC_SUCCESS;

}

REGISTER_COMMAND_BACKEND(CCV_NNC_RANDOM_NORMAL_FORWARD, CCV_NNC_BACKEND_CPU_REF)(ccv_nnc_cmd_backend_registry_t* const registry)

{

  registry->tensor_formats = CCV_TENSOR_FORMAT_NHWC | CCV_TENSOR_FORMAT_NCHW | CCV_TENSOR_FORMAT_CHWN;

  registry->tensor_datatypes = CCV_32F;

  registry->tensor_memory = CCV_TENSOR_CPU_MEMORY;

  registry->algorithms = 1;

  registry->exec = _ccv_nnc_random_normal;

}

REGISTER_COMMAND_BACKEND(CCV_NNC_RANDOM_NORMAL_BACKWARD, CCV_NNC_BACKEND_CPU_REF)(ccv_nnc_cmd_backend_registry_t* const registry)

{

  registry->tensor_formats = CCV_TENSOR_FORMAT_NHWC | CCV_TENSOR_FORMAT_NCHW | CCV_TENSOR_FORMAT_CHWN;

  registry->tensor_datatypes = CCV_32F;

  registry->tensor_memory = CCV_TENSOR_CPU_MEMORY;

  registry->algorithms = 1;

  registry->exec = _ccv_nnc_random_normal;

}