#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_softmax_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, const ccv_nnc_stream_context_t* const stream_context)
{
	assert(input_size == 1);
	const ccv_nnc_tensor_t* a = inputs[0];
	assert(!CCV_IS_TENSOR_VIEW(a));
	assert(output_size == 1);
	ccv_nnc_tensor_t* b = outputs[0];
	assert(!CCV_IS_TENSOR_VIEW(a));
	int i, count = ccv_nnc_tensor_count(a->info);
	for (i = 0; i < CCV_NNC_MAX_DIM_ALLOC && a->info.dim[i] > 0; i++)
	{
		assert(a->info.dim[i] == b->info.dim[i]);
	}
	float* ap = a->data.f32;
	float* bp = b->data.f32;
	double maxval = ap[0];
	for (i = 1; i < count; i++)
		if (ap[i] > maxval)
			maxval = ap[i];
	double sumval = 0;
	for (i = 0; i < count; i++)
		sumval += (bp[i] = expf(ap[i] - maxval));
	sumval = 1.0 / sumval;
	for (i = 0; i < count; i++)
		bp[i] *= sumval;
	return CCV_NNC_EXEC_SUCCESS;
}

static int _ccv_nnc_softmax_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, const ccv_nnc_stream_context_t* const stream_context)
{
	assert(0 && "This should never be called.");
	return CCV_NNC_EXEC_INVALID;
}

REGISTER_COMMAND_BACKEND(CCV_NNC_SOFTMAX_FORWARD, CCV_NNC_BACKEND_CPU_REF)(ccv_nnc_cmd_backend_registry_t* const registry)
{
	registry->tensor_formats = CCV_TENSOR_FORMAT_NHWC;
	registry->tensor_datatypes = CCV_32F;
	registry->tensor_memory = CCV_TENSOR_CPU_MEMORY;
	registry->algorithms = 1;
	registry->exec = _ccv_nnc_softmax_forw;
}

REGISTER_COMMAND_BACKEND(CCV_NNC_SOFTMAX_BACKWARD, CCV_NNC_BACKEND_CPU_REF)(ccv_nnc_cmd_backend_registry_t* const registry)
{
	registry->tensor_formats = CCV_TENSOR_FORMAT_NHWC;
	registry->tensor_datatypes = CCV_32F;
	registry->tensor_memory = CCV_TENSOR_CPU_MEMORY;
	registry->algorithms = 1;
	registry->exec = _ccv_nnc_softmax_back;
}