TMultiply.cpp

/* 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: Lin Ye (email: linye2015@outlook.com) 2018-06-15
*/

#include "TMultiply.h"

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

/* 
case 1: element-wise product of two tensors
c(i) = a(i)*b(i) + \alpha * c(i) 
In this case, (2, 1)  (2, 1) -> (2, 1), leadingDim=0, alpha=0.
*/
bool TestMultiply1()
{
	/* a source tensor of size (2, 1) */
	int sOrder1 = 2;
	int * sDimSize1 = new int[sOrder1];
	sDimSize1[0] = 2;
	sDimSize1[1] = 1;

	int sUnitNum1 = 1;
	for (int i = 0; i < sOrder1; i++)
		sUnitNum1 *= sDimSize1[i];

	/* a source tensor of size (2, 1) */
	int sOrder2 = 2;
	int * sDimSize2 = new int[sOrder2];
	sDimSize2[0] = 2;
	sDimSize2[1] = 1;

	int sUnitNum2 = 1;
	for (int i = 0; i < sOrder2; i++)
		sUnitNum2 *= sDimSize2[i];

	/* a target tensor of size (2, 1) */
	int tOrder = 2;
	int * tDimSize = new int[tOrder];
	tDimSize[0] = 2;
	tDimSize[1] = 1;

	int tUnitNum = 1;
	for (int i = 0; i < tOrder; i++)
		tUnitNum *= tDimSize[i];

	DTYPE sData1[2][1] = { {0.0F}, 
                           {1.0F} };
	DTYPE sData2[2][1] = { {2.0F},
                           {3.0F} };
	DTYPE answer[2][1] = { {0.0F},
                           {3.0F} };

	/* CPU test */
	bool cpuTest = true;

	/* create tensors */
	XTensor * s1 = NewTensor(sOrder1, sDimSize1);
	XTensor * s2 = NewTensor(sOrder2, sDimSize2);
	XTensor * t = NewTensor(tOrder, tDimSize);
	XTensor * tMe = NewTensor(tOrder, tDimSize);
    XTensor tUser;

	/* initialize variables */
	s1->SetData(sData1, sUnitNum1);
	tMe->SetData(sData1, sUnitNum1);
	s2->SetData(sData2, sUnitNum2);
	t->SetZeroAll();

	/* call Multiply function */
	_Multiply(s1, s2, t, 0, 0);
	_MultiplyMe(tMe, s2, 0, 0);
    tUser = Multiply(*s1, *s2, 0);

	/* check results */
	cpuTest = t->CheckData(answer, tUnitNum) 
        && tMe->CheckData(answer, tUnitNum) && tUser.CheckData(answer, tUnitNum);

#ifdef USE_CUDA
	/* GPU test */
	bool gpuTest = true;

	/* create tensor */
	XTensor * sGPU1 = NewTensor(sOrder1, sDimSize1, X_FLOAT, 1.0F, 0);
	XTensor * sGPU2 = NewTensor(sOrder2, sDimSize2, X_FLOAT, 1.0F, 0);
	XTensor * tGPU = NewTensor(tOrder, tDimSize, X_FLOAT, 1.0F, 0);
	XTensor * tMeGPU = NewTensor(tOrder, tDimSize, X_FLOAT, 1.0F, 0);
    XTensor tUserGPU;

	/* Initialize variables */
	sGPU1->SetData(sData1, sUnitNum1);
	tMeGPU->SetData(sData1, sUnitNum1);
	sGPU2->SetData(sData2, sUnitNum2);
	tGPU->SetZeroAll();

	/* call Multiply function */
	_Multiply(sGPU1, sGPU2, tGPU, 0, 0);
	_MultiplyMe(tMeGPU, sGPU2, 0, 0);
    tUserGPU = Multiply(*sGPU1, *sGPU2, 0);

	/* check results */
	gpuTest = tGPU->CheckData(answer, tUnitNum)
        && tMeGPU->CheckData(answer, tUnitNum) && tUserGPU.CheckData(answer, tUnitNum);

	/* destroy variables */
    delete s1;
    delete s2;
    delete t;
    delete tMe;
    delete sGPU1;
    delete sGPU2;
    delete tGPU;
    delete tMeGPU;
    delete[] sDimSize1;
    delete[] sDimSize2;
    delete[] tDimSize;

	return cpuTest && gpuTest;
#else
    /* destroy variables */
    delete s1;
    delete s2;
    delete t;
    delete tMe;
    delete[] sDimSize1;
    delete[] sDimSize2;
    delete[] tDimSize;

	return cpuTest;
#endif // USE_CUDA
}

/* 
case 2: element-wise product of two tensors
c(i) = a(i)*b(i) + \alpha * c(i)
In this case, (2, 2)  (2, 2) -> (2, 2), leadingDim=0, alpha=0.
*/
bool TestMultiply2()
{
	/* a source tensor of size (2, 2) */
	int sOrder1 = 2;
	int * sDimSize1 = new int[sOrder1];
	sDimSize1[0] = 2;
	sDimSize1[1] = 2;

	int sUnitNum1 = 1;
	for (int i = 0; i < sOrder1; i++)
		sUnitNum1 *= sDimSize1[i];

	/* a source tensor of size (2, 2) */
	int sOrder2 = 2;
	int * sDimSize2 = new int[sOrder2];
	sDimSize2[0] = 2;
	sDimSize2[1] = 2;

	int sUnitNum2 = 1;
	for (int i = 0; i < sOrder2; i++)
		sUnitNum2 *= sDimSize2[i];

	/* a target tensor of size (2, 2) */
	int tOrder = 2;
	int * tDimSize = new int[tOrder];
	tDimSize[0] = 2;
	tDimSize[1] = 2;

	int tUnitNum = 1;
	for (int i = 0; i < tOrder; i++)
		tUnitNum *= tDimSize[i];

	DTYPE sData1[2][2] = { {0.0F, 1.0F},
	                       {2.0F, 3.0F} };
	DTYPE sData2[2][2] = { {0.0F, 1.0F},
	                       {2.0F, 3.0F} };
	DTYPE answer[2][2] = { {0.0F, 1.0F},
	                       {4.0F, 9.0F} };

	/* CPU test */
	bool cpuTest = true;

	/* create tensors */
	XTensor * s1 = NewTensor(sOrder1, sDimSize1);
	XTensor * s2 = NewTensor(sOrder2, sDimSize2);
	XTensor * t = NewTensor(tOrder, tDimSize);
    XTensor * tMe = NewTensor(tOrder, tDimSize);
    XTensor tUser;

	/* initialize variables */
	s1->SetData(sData1, sUnitNum1);
	tMe->SetData(sData1, sUnitNum1);
	s2->SetData(sData2, sUnitNum2);
	t->SetZeroAll();

	/* call Multiply function */
	_Multiply(s1, s2, t, 0, 0);
	_MultiplyMe(tMe, s2, 0, 0);
    tUser = Multiply(*s1, *s2, 0);

	/* check results */
	cpuTest = t->CheckData(answer, tUnitNum) 
        && tMe->CheckData(answer, tUnitNum) && tUser.CheckData(answer, tUnitNum);

#ifdef USE_CUDA
	/* GPU test */
	bool gpuTest = true;

	/* create tensor */
	XTensor * sGPU1 = NewTensor(sOrder1, sDimSize1, X_FLOAT, 1.0F, 0);
	XTensor * sGPU2 = NewTensor(sOrder2, sDimSize2, X_FLOAT, 1.0F, 0);
	XTensor * tGPU = NewTensor(tOrder, tDimSize, X_FLOAT, 1.0F, 0);
    XTensor * tMeGPU = NewTensor(tOrder, tDimSize, X_FLOAT, 1.0F, 0);
    XTensor tUserGPU;

	/* Initialize variables */
	sGPU1->SetData(sData1, sUnitNum1);
	tMeGPU->SetData(sData1, sUnitNum1);
	sGPU2->SetData(sData2, sUnitNum2);
	tGPU->SetZeroAll();

	/* call Multiply function */
	_Multiply(sGPU1, sGPU2, tGPU, 0, 0);
	_MultiplyMe(tMeGPU, sGPU2, 0, 0);
    tUserGPU = Multiply(*sGPU1, *sGPU2, 0);

	/* check results */
	gpuTest = tGPU->CheckData(answer, tUnitNum) && 
              tMeGPU->CheckData(answer, tUnitNum) && 
              tUserGPU.CheckData(answer, tUnitNum);

	/* destroy variables */
    delete s1;
    delete s2;
    delete t;
    delete tMe;
    delete sGPU1;
    delete sGPU2;
    delete tGPU;
    delete tMeGPU;
    delete[] sDimSize1;
    delete[] sDimSize2;
    delete[] tDimSize;

	return cpuTest && gpuTest;
#else
    /* destroy variables */
    delete s1;
    delete s2;
    delete t;
    delete tMe;
    delete[] sDimSize1;
    delete[] sDimSize2;
    delete[] tDimSize;

	return cpuTest;
#endif // USE_CUDA
}

/* 
case 3: element-wise product of two tensors, c(i) = a(i)*b(i) + \alpha * c(i)
In this case, (2, 2)  (2, 2) -> (2, 2), leadingDim=1, alpha=0.
*/
bool TestMultiply3()
{
	/* a source tensor of size (2, 2) */
	int sOrder1 = 2;
	int * sDimSize1 = new int[sOrder1];
	sDimSize1[0] = 2;
	sDimSize1[1] = 2;

	int sUnitNum1 = 1;
	for (int i = 0; i < sOrder1; i++)
		sUnitNum1 *= sDimSize1[i];

	/* a source tensor of size (2, 2) */
	int sOrder2 = 2;
	int * sDimSize2 = new int[sOrder2];
	sDimSize2[0] = 2;
	sDimSize2[1] = 2;

	int sUnitNum2 = 1;
	for (int i = 0; i < sOrder2; i++)
		sUnitNum2 *= sDimSize2[i];

	/* a target tensor of size (2, 2) */
	int tOrder = 2;
	int * tDimSize = new int[tOrder];
	tDimSize[0] = 2;
	tDimSize[1] = 2;

	int tUnitNum = 1;
	for (int i = 0; i < tOrder; i++)
		tUnitNum *= tDimSize[i];

	DTYPE sData1[2][2] = { {0.0F, 1.0F},
	                       {2.0F, 3.0F} };
	DTYPE sData2[2][2] = { {0.0F, 1.0F},
	                       {2.0F, 3.0F} };
	DTYPE answer[2][2] = { {0.0F, 1.0F},
	                       {4.0F, 9.0F} };

	/* CPU test */
	bool cpuTest = true;

	/* create tensors */
	XTensor * s1 = NewTensor(sOrder1, sDimSize1);
	XTensor * s2 = NewTensor(sOrder2, sDimSize2);
	XTensor * t = NewTensor(tOrder, tDimSize);

	/* initialize variables */
	s1->SetData(sData1, sUnitNum1);
	s2->SetData(sData2, sUnitNum2);
	t->SetZeroAll();

	/* call MultiplyElementWise function */
	_Multiply(s1, s2, t, 0, 1);

	/* check results */
	cpuTest = t->CheckData(answer, tUnitNum);

#ifdef USE_CUDA
	/* GPU test */
	bool gpuTest = true;

	/* create tensor */
	XTensor * sGPU1 = NewTensor(sOrder1, sDimSize1, X_FLOAT, 1.0F, 0);
	XTensor * sGPU2 = NewTensor(sOrder2, sDimSize2, X_FLOAT, 1.0F, 0);
	XTensor * tGPU = NewTensor(tOrder, tDimSize, X_FLOAT, 1.0F, 0);

	/* Initialize variables */
	sGPU1->SetData(sData1, sUnitNum1);
	sGPU2->SetData(sData2, sUnitNum2);
	tGPU->SetZeroAll();

	/* call MultiplyElementWise function */
	_Multiply(sGPU1, sGPU2, tGPU, 0, 1);

	/* check results */
	gpuTest = tGPU->CheckData(answer, tUnitNum);

	/* destroy variables */
    delete s1;
    delete s2;
    delete t;
    delete sGPU1;
    delete sGPU2;
    delete tGPU;
    delete[] sDimSize1;
    delete[] sDimSize2;
    delete[] tDimSize;

	return cpuTest && gpuTest;
#else
    /* destroy variables */
    delete s1;
    delete s2;
    delete t;
    delete[] sDimSize1;
    delete[] sDimSize2;
    delete[] tDimSize;

	return cpuTest;
#endif // USE_CUDA
}

/* other cases */
/*
TODO!!
*/

/* test for Multiply Function */
bool TestMultiply()
{
	XPRINT(0, stdout, "[TEST Multiply] element-wise product of two tensors \n");
	bool returnFlag = true, caseFlag = true;

	/* case 1 test */
	caseFlag = TestMultiply1();

	if (!caseFlag) {
		returnFlag = false;
		XPRINT(0, stdout, ">> case 1 failed!\n");
	}
	else
		XPRINT(0, stdout, ">> case 1 passed!\n");

	/* case 2 test */
	caseFlag = TestMultiply2();

	if (!caseFlag) {
		returnFlag = false;
		XPRINT(0, stdout, ">> case 2 failed!\n");
	}
	else
		XPRINT(0, stdout, ">> case 2 passed!\n");

	/* case 3 test */
	caseFlag = TestMultiply3();

	if (!caseFlag) {
		returnFlag = false;
		XPRINT(0, stdout, ">> case 3 failed!\n");
	}
	else
		XPRINT(0, stdout, ">> case 3 passed!\n");

	/* other cases test */
	/*
	TODO!!
	*/

	if (returnFlag) {
		XPRINT(0, stdout, ">> All Passed!\n");
	}
	else
		XPRINT(0, stdout, ">> Failed!\n");

	XPRINT(0, stdout, "\n");

	return returnFlag;
}

} // namespace nts(NiuTrans.Tensor)