/* NiuTrans.Tensor - an open-source tensor library
* Copyright (C) 2017, Natural Language Processing Lab, Northestern University.
* All rights reserved.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
*   http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/

/*
* $Created by: XIAO Tong (email: xiaotong@mail.neu.edu.cn) 2018-04-24
*/

#include "../../XTensor.h"
#include "../../XName.h"
#include "Multiply.h"
#include "Multiply.cuh"
#include "MultiplyDim.h"

namespace nts { // namespace nts(NiuTrans.Tensor)

/*
element-wise product of two tensors

c(i) = a(i)*b(i) + \alpha * c(i)
where i is the index of the item

>> a - tensor a
>> b - tensor b
>> c - result tensor
>> alpha - the coefficient
>> leadingDim - the dimension along which we perform broadcasting
*/
void _Multiply(const XTensor * a, const XTensor * b, XTensor * c, DTYPE alpha, int leadingDim)
{
	int leadingDimRDI = a->order - leadingDim - 1;
    CheckNTErrors((a->unitNum <= c->unitNum && b->unitNum <= c->unitNum),
                  "Unmatched tensors in multiplication!");
    CheckNTErrors((a->order == b->order && a->order == c->order), 
                  "Unmatched tensors!");

#ifdef USE_CUDA
    if (a->devID >= 0 || b->devID >= 0 || c->devID >= 0) {
        _CudaMultiply(a, b, c, alpha, leadingDim);
        return;
    }
#endif

    int stride = 1;
    int blockSizeA = 1;
    int blockSizeB = 1;
    int blockSizeC = 1;
    int blockNum = 1;
    int dimensionSizeA = a->dimSizeRDI[leadingDimRDI];
    int dimensionSizeB = b->dimSizeRDI[leadingDimRDI];
    int dimensionSizeC = c->dimSizeRDI[leadingDimRDI];

    for (int i = 0; i < a->order; i++) {
        if (i != leadingDimRDI) {
            CheckNTErrors((a->dimSizeRDI[i] == b->dimSizeRDI[i] &&
                           a->dimSizeRDI[i] == c->dimSizeRDI[i]),
                          "Unmatched tensors!");
        }
        if (i < leadingDimRDI)
            stride *= a->dimSizeRDI[i];
    }

    blockSizeA = stride * dimensionSizeA;
    blockSizeB = stride * dimensionSizeB;
    blockSizeC = stride * dimensionSizeC;
    blockNum = a->unitNum / blockSizeA;

    if (!a->isSparse && !b->isSparse) {
        if (a->dataType == DEFAULT_DTYPE && b->dataType == DEFAULT_DTYPE) {
            if (a->unitNum == c->unitNum && b->unitNum == c->unitNum) {
                int size = a->unitNum;
                DTYPE * ap = (DTYPE*)a->data;
                DTYPE * bp = (DTYPE*)b->data;
                DTYPE * cp = (DTYPE*)c->data;
                if (alpha == 0) {
                    for (int i = 0; i < size; i++)
                        cp[i] = ap[i] * bp[i];
                }
                else {
                    for (int i = 0; i < size; i++)
                        cp[i] = ap[i] * bp[i] + alpha * cp[i];
                }
            }
            else {
                for (int k = 0; k < blockNum; k++) {

                    for (int ci = 0, ai = 0, bi = 0; ci < dimensionSizeC; ci++, ai++, bi++) {
                        if (ai >= dimensionSizeA)
                            ai = 0;
                        if (bi >= dimensionSizeB)
                            bi = 0;
                        DTYPE * ap = (DTYPE*)a->data + k * blockSizeA + ai * stride;
                        DTYPE * bp = (DTYPE*)b->data + k * blockSizeB + bi * stride;
                        DTYPE * cp = (DTYPE*)c->data + k * blockSizeC + ci * stride;
                        for (int j = 0; j < stride; j++)
                            cp[j] = ap[j] * bp[j] + cp[j] * alpha;
                    }
                }
            }
        }
        else {
            // TODO!!
            ShowNTErrors("TODO!");
        }
    }
    else {
        // TODO!!
        ShowNTErrors("TODO!");
    }
}

/*
element-wise product of two tensors (do it on site)
keep the result in the input tensor a and return nothing

a(i) = a(i)*b(i) + \alpha * a(i)
where i is the index of the item

>> a - tensor a (where keep the result)
>> b - tensor b
>> alpha - the coefficient
>> leadingDim - the dimension along which we perform broadcasting
*/
void _MultiplyMe(XTensor * a, const XTensor * b, DTYPE alpha, int leadingDim)
{
    _Multiply(a, b, a, alpha, leadingDim);
}

/* 
return a dimension if the multiplication is performed as MultiplyDim (in more details in MultiplyDim.h)
>> a - a tensor
>> b - another tensor for multiplication
*/
int GetMultiplyDimIndex(const XTensor &a, const XTensor &b)
{
    if(a.order < b.order)
        return -1;
    if(XTensor::IsSameShaped(&a, &b))
        return -1;

    int hitCount = 0;
    int hitDim = -1;
    for(int i = 0; i < b.order; i++){
        if(b.dimSize[b.order - 1 - i] == 1)
            continue;
        else if(b.dimSize[b.order - 1 - i] == a.dimSize[a.order - 1 - i]){
            hitCount++;
            hitDim = a.order - b.order + i;
        }
    }

    if(hitCount == 1)
        return hitDim;
    else
        return -1;
}

/*
element-wise product of two tensors (return a XTensor structure)
make a new tensor c to keep the result and return it

c(i) = a(i)*b(i)
where i is the index of the item

>> a - tensor a
>> b - tensor b
>> leadingDim - the dimension along which we perform broadcasting
<< return - the product of the tensors
*/
XTensor Multiply(const XTensor &a, const XTensor &b, DTYPE alpha, int leadingDim)
{

    XTensor c(&a);
    c.SetTMPFlag();
    
    int n = GetMultiplyDimIndex(a, b);

    if(n == -1){
        CheckNTErrors(a.dimSize[leadingDim] == b.dimSize[leadingDim], "TODO!");
    
        /* call _Multiply function */
        _Multiply(&a, &b, &c, 0, leadingDim);
    
        /* tensor connections */
        XLink::MakeLink(&a, &b, &c, MATH_MULTIPLY);
        XLink::AddParamToHead(&c, alpha);
        XLink::AddParamToHeadInt(&c, leadingDim);
    }
    else if(n >= 0 && n < a.order){
        /* call _MultiplyDim function */
        _MultiplyDim(&a, &b, &c, n, alpha);

        /* tensor connections */
        XLink::MakeLink(&a, &b, &c, MATH_MULTIPLYDIM);
        XLink::AddParamToHeadInt(&c, n);
        XLink::AddParamToHead(&c, alpha);
    }
    else{
        ShowNTErrors("Something is wrong!");
    }

    return c;
}

} // namespace nts(NiuTrans.Tensor)