#include "ccv.h"

#include "ccv_internal.h"

#include "nnc/ccv_nnc.h"

#include "nnc/ccv_nnc_easy.h"

#include "nnc/ccv_nnc_internal.h"

#ifdef USE_OPENMP

#include <omp.h>

#endif

#ifdef USE_DISPATCH

#include <dispatch/dispatch.h>

#endif

static int _ccv_nnc_swish_forw(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(input_size == 1);

  const ccv_nnc_tensor_t* a = inputs[0];

  assert(CCV_IS_TENSOR_CONTIGUOUS(a));

  assert(output_size == 1);

  ccv_nnc_tensor_t* b = outputs[0];

  assert(CCV_IS_TENSOR_CONTIGUOUS(b));

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

  int i;

  for (i = 0; i < CCV_NNC_MAX_DIM_ALLOC && a->info.dim[i] > 0; 
i++9
)

  {

    assert(a->info.dim[i] == b->info.dim[i]);

  }

  float* ap = a->data.f32;

  float* bp = b->data.f32;

  for (i = 0; i < count; 
i++2.41k
)

    bp[i] = ap[i] / (1. + exp(-ap[i]));

  return CCV_NNC_EXEC_SUCCESS;

}

static int _ccv_nnc_swish_back(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(input_size == 3);

  const ccv_nnc_tensor_t* g = inputs[0]; // gradient

  assert(CCV_IS_TENSOR_CONTIGUOUS(g));

  const ccv_nnc_tensor_t* a = inputs[1];

  assert(CCV_IS_TENSOR_CONTIGUOUS(a));

  assert(output_size == 1);

  ccv_nnc_tensor_t* h = outputs[0];

  assert(CCV_IS_TENSOR_CONTIGUOUS(h));

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

  int i;

  for (i = 0; i < CCV_NNC_MAX_DIM_ALLOC && g->info.dim[i] > 0; 
i++5
)

  {

    assert(a->info.dim[i] == g->info.dim[i]);

    assert(g->info.dim[i] == h->info.dim[i]);

  }

  float* ap = a->data.f32;

  float* gp = g->data.f32;

  float* hp = h->data.f32;

  /**

   * e^x*(x+e^x+1)/(e^x+1)^2

   * = x*e^x/(e^x+1)^2+e^x/(e^x+1) = x*e^x/(e^x+1)^2+y

   *

   * e^x/(e^x+1)^2 = ((e^x+1)^2-e^2x-1)/(2*(e^x+1)^2)

   * = 1/2*(1-e^2x/(e^x+1)^2-1/(e^x+1)^2)

   * = 1/2*(1-y^2-1/(e^x+1)^2)

   *

   * y = e^x/(e^x+1) = 1 - 1/(e^x+1)

   * 1/(e^x+1) = 1-y

   *

   * 1/2*(1-y^2-1/(e^x+1)^2) = 1/2*(1-y^2-(1-y)^2)

   * = 1/2*(1-y^2-1+2y-y^2)

   * = y-y^2

   *

   * x*e^x/(e^x+1)^2+y = x*(y-y^2)+y

   */

  for (i = 0; i < count; 
i++2.01k
)

  {

    const float x = ap[i];

    const float y = 1. / (1. + exp(-x));

    const float y2 = y * y;

    hp[i] = gp[i] * (x * (y - y2) + y);

  }

  return CCV_NNC_EXEC_SUCCESS;

}

REGISTER_COMMAND_BACKEND(CCV_NNC_SWISH_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_swish_forw;

}

REGISTER_COMMAND_BACKEND(CCV_NNC_SWISH_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_swish_back;

}