CNTK/Tests/UnitTests/MathTests/MatrixBlasTests.cpp

348 строки
14 KiB
C++

//
// Copyright (c) Microsoft. All rights reserved.
// Licensed under the MIT license. See LICENSE.md file in the project root for full license information.
//
#include "stdafx.h"
#include <math.h>
#ifdef _WIN32
#include <crtdefs.h>
#endif
#include "../../../Source/Math/Matrix.h"
#include "../../../Source/Math/CPUMatrix.h"
using namespace Microsoft::MSR::CNTK;
namespace Microsoft { namespace MSR { namespace CNTK { namespace Test {
const size_t c_value_1 = 1;
const size_t c_value_2 = 2;
const size_t c_value_3 = 3;
BOOST_AUTO_TEST_SUITE(CPUMatrixSuite)
BOOST_FIXTURE_TEST_CASE(MatrixMultiplyTest, RandomSeedFixture)
{
// Part 1: Multiply with identity matrix
SingleMatrix matrixA = SingleMatrix::RandomGaussian(64, 23, c_deviceIdZero, 0, 2, IncrementCounter());
SingleMatrix matrixB = SingleMatrix::Eye(23, c_deviceIdZero);
SingleMatrix matrixC = matrixA * matrixB;
foreach_coord (i, j, matrixA)
{
BOOST_CHECK_EQUAL(matrixA(i, j), matrixC(i, j));
}
SingleMatrix matrix0(3, 3, c_deviceIdZero);
matrix0(0, 0) = 1;
matrix0(0, 1) = 2;
matrix0(0, 2) = 3;
matrix0(1, 0) = 4;
matrix0(1, 1) = 5;
matrix0(1, 2) = 6;
matrix0(2, 0) = 7;
matrix0(2, 1) = 8;
matrix0(2, 2) = 9;
// Part 2: Compare with Octave on toy example
SingleMatrix matrix1(3, 4, c_deviceIdZero);
matrix1(0, 0) = 8;
matrix1(0, 1) = 9;
matrix1(0, 2) = 3;
matrix1(0, 3) = 45;
matrix1(1, 0) = 3;
matrix1(1, 1) = 4;
matrix1(1, 2) = 56;
matrix1(1, 3) = 1;
matrix1(2, 0) = -3;
matrix1(2, 1) = 5;
matrix1(2, 2) = 2;
matrix1(2, 3) = 6;
SingleMatrix matrix2 = matrix0 * matrix1;
BOOST_CHECK_EQUAL(5, matrix2(0, 0));
BOOST_CHECK_EQUAL(32, matrix2(0, 1));
BOOST_CHECK_EQUAL(121, matrix2(0, 2));
BOOST_CHECK_EQUAL(65, matrix2(0, 3));
BOOST_CHECK_EQUAL(29, matrix2(1, 0));
BOOST_CHECK_EQUAL(86, matrix2(1, 1));
BOOST_CHECK_EQUAL(304, matrix2(1, 2));
BOOST_CHECK_EQUAL(221, matrix2(1, 3));
BOOST_CHECK_EQUAL(53, matrix2(2, 0));
BOOST_CHECK_EQUAL(140, matrix2(2, 1));
BOOST_CHECK_EQUAL(487, matrix2(2, 2));
BOOST_CHECK_EQUAL(377, matrix2(2, 3));
}
BOOST_FIXTURE_TEST_CASE(MatrixMultiplyAndPlusAndMinus, RandomSeedFixture)
{
// Part 1: Multiply with identity matrix
SingleMatrix matrixA = SingleMatrix::RandomGaussian(64, 23, c_deviceIdZero, 0, 2, IncrementCounter());
SingleMatrix matrixB = SingleMatrix::Eye(23, c_deviceIdZero);
SingleMatrix matrixB1 = SingleMatrix::RandomUniform(64, 23, c_deviceIdZero, -95.23f, 43.5f, IncrementCounter());
SingleMatrix matrixB2 = SingleMatrix::RandomUniform(64, 23, c_deviceIdZero, 23.23f, 143.5f, IncrementCounter());
SingleMatrix C = ((matrixA * matrixB) + matrixB1) - matrixB2;
foreach_coord (i, j, matrixA)
{
BOOST_CHECK_EQUAL(matrixA(i, j) + matrixB1(i, j) - matrixB2(i, j), C(i, j));
}
// Part 3: compare with CPUMatrix results
// TODO: Split into separate test case WI# 82
CPUMatrix<float> cpuMatrix1 = CPUMatrix<float>::RandomUniform(429, 1024, -1, 1, IncrementCounter());
CPUMatrix<float> cpuMatrix2 = CPUMatrix<float>::RandomUniform(429, 1024, -2, 2, IncrementCounter());
CPUMatrix<float> cpuMatrix3 = CPUMatrix<float>::RandomUniform(1024, 1024, -3, 1, IncrementCounter());
CPUMatrix<float> copyCpuMatrix3(cpuMatrix3);
CPUMatrix<float>::MultiplyAndAdd(cpuMatrix1, true, cpuMatrix2, false, cpuMatrix3);
SingleMatrix singleMatrix1(429, 1024, cpuMatrix1.Buffer(), matrixFlagNormal);
SingleMatrix singleMatrix2(429, 1024, cpuMatrix2.Buffer(), matrixFlagNormal);
SingleMatrix singleMatrix3(1024, 1024, copyCpuMatrix3.Buffer(), matrixFlagNormal);
SingleMatrix::MultiplyAndAdd(singleMatrix1, true, singleMatrix2, false, singleMatrix3);
foreach_coord (i, j, singleMatrix3)
{
BOOST_CHECK(fabs(singleMatrix3(i, j) - cpuMatrix3(i, j)) < c_epsilonFloat5E4);
}
// TODO: Split into separate test case WI# 82
SingleMatrix singleMatrix4 = SingleMatrix::Ones(8, 9, c_deviceIdZero);
SingleMatrix singleMatrix5 = SingleMatrix::Ones(8, 1, c_deviceIdZero);
singleMatrix5(4, 0) = -5.5;
SingleMatrix::ScaleAndAdd(1, singleMatrix5, singleMatrix4);
foreach_coord (i, j, singleMatrix4)
{
if (i != 4)
{
BOOST_CHECK_EQUAL(2, singleMatrix4(i, j));
}
else
{
BOOST_CHECK_EQUAL(-4.5, singleMatrix4(i, j));
}
}
}
BOOST_FIXTURE_TEST_CASE(MatrixColumnwiseScaleAndWeightedAdd, RandomSeedFixture)
{
size_t m = 256;
size_t n = 64;
for(int deviceId : {-1, 0})
{
SingleMatrix singleMatrixA(deviceId);
singleMatrixA.AssignTruncateBottomOf(SingleMatrix::RandomUniform(m, n, deviceId, -200, 1, IncrementCounter()), 0);
const SingleMatrix singleMatrixB = SingleMatrix::RandomUniform(n, 1, deviceId, 0, 1, IncrementCounter());
SingleMatrix singleMatrixAcsc = singleMatrixA.DeepClone();
singleMatrixAcsc.SwitchToMatrixType(SPARSE, matrixFormatSparseCSC, true);
SingleMatrix singleMatrixCexpected = SingleMatrix::RandomUniform(m, n, deviceId, 0, 1, IncrementCounter());
SingleMatrix singleMatrixCdense = singleMatrixCexpected.DeepClone();
SingleMatrix singleMatrixCsparse = singleMatrixCexpected.DeepClone();
SingleMatrix singleMatrixBdiag(n, n, deviceId);
singleMatrixBdiag.SetValue(0);
singleMatrixBdiag.SetDiagonalValue(singleMatrixB);
for(float beta : {0.0f, 1.0f})
{
SingleMatrix::MultiplyAndWeightedAdd(1, singleMatrixA, false, singleMatrixBdiag, false, beta, singleMatrixCexpected);
SingleMatrix::ColumnwiseScaleAndWeightedAdd(1, singleMatrixA, singleMatrixB, beta, singleMatrixCdense);
SingleMatrix::ColumnwiseScaleAndWeightedAdd(1, singleMatrixAcsc, singleMatrixB, beta, singleMatrixCsparse);
BOOST_CHECK(singleMatrixCexpected.IsEqualTo(singleMatrixCdense, c_epsilonFloatE4));
BOOST_CHECK(singleMatrixCexpected.IsEqualTo(singleMatrixCsparse, c_epsilonFloatE4));
}
}
}
BOOST_FIXTURE_TEST_CASE(MatrixScaleAndAdd, RandomSeedFixture)
{
std::mt19937 rng(0);
const int seed = rng();
const SingleMatrix singleMatrixA = SingleMatrix::RandomUniform(1024, 512, c_deviceIdZero , - 12.34f, 55.2312f, seed + 0);
const SingleMatrix singleMatrixB = SingleMatrix::RandomUniform(1024, 512, c_deviceIdZero, -12.34f, 55.2312f, seed + 1);
SingleMatrix singleMatrixC(singleMatrixB.DeepClone());
const float alpha = 0.34213f;
SingleMatrix::ScaleAndAdd(alpha, singleMatrixA, singleMatrixC);
foreach_coord (i, j, singleMatrixC)
{
BOOST_CHECK(fabsf(singleMatrixC(i, j) - (alpha * singleMatrixA(i, j) + singleMatrixB(i, j))) < c_epsilonFloatE5);
}
// Test 2
// TODO: Split into separate test case WI# 82
const SingleMatrix singleMatrixA1 = SingleMatrix::RandomUniform(1024, 512, c_deviceIdZero, -12.34f, 55.2312f, seed + 2);
const SingleMatrix singleMatrixB1 = SingleMatrix::RandomUniform(1024, 512, c_deviceIdZero, -12.34f, 55.2312f, seed + 3);
SingleMatrix singleMatrixC1(singleMatrixB1.DeepClone()); // C1==B1
const float beta = -1.4654f;
SingleMatrix::ScaleAndAdd(alpha, singleMatrixA1, beta, singleMatrixC1); // C1=alpha*A1+beta*C1
foreach_coord (i, j, singleMatrixC1)
{
BOOST_CHECK(fabsf(singleMatrixC1(i, j) - (alpha * singleMatrixA1(i, j) + beta * singleMatrixB1(i, j))) < c_epsilonFloatE5);
}
// Test 3 - columnwise
// TODO: Split into separate test case WI# 82
const SingleMatrix singleMatrixA2 = SingleMatrix::RandomUniform(1024, 1, c_deviceIdZero, -12.34f, 55.2312f, seed + 4);
const SingleMatrix singleMatrixB2 = SingleMatrix::RandomUniform(1024, 512, c_deviceIdZero, -12.34f, 55.2312f, seed + 5); // Column
SingleMatrix singleMatrixC2(singleMatrixB2.DeepClone()); // C2==B2
const float betaOne = 1;
SingleMatrix::ScaleAndAdd(alpha, singleMatrixA2, betaOne, singleMatrixC2); // C2=alpha*A1+beta*C1
foreach_coord (i, j, singleMatrixC2)
{
float x = singleMatrixC2(i, j);
float y = (alpha * singleMatrixA2(i, 0) + betaOne * singleMatrixB2(i, j));
BOOST_CHECK(fabsf(x - y) < c_epsilonFloatE5);
}
}
BOOST_FIXTURE_TEST_CASE(MatrixScaleAndAdd_double, RandomSeedFixture)
{
std::mt19937 rng(0);
const int seed = rng();
DoubleMatrix matrixA = DoubleMatrix::RandomUniform(1024, 512, c_deviceIdZero, -12.34, 55.2312, seed + 0);
DoubleMatrix matrixB = DoubleMatrix::RandomUniform(1024, 512, c_deviceIdZero, -12.34, 55.2312, seed + 1);
DoubleMatrix matrixC(matrixB.DeepClone());
const float alpha = 0.34213f;
DoubleMatrix::ScaleAndAdd(alpha, matrixA, matrixC);
foreach_coord (i, j, matrixC)
{
BOOST_CHECK(fabsf(static_cast<float>(matrixC(i, j) - (alpha * matrixA(i, j) + matrixB(i, j)))) < c_epsilonDoubleE11);
}
// Test 2
// TODO: Split into separate test case WI# 82
DoubleMatrix matrixA1 = DoubleMatrix::RandomUniform(1024, 512, c_deviceIdZero, -12.34f, 55.2312f, seed + 2);
DoubleMatrix matrixB1 = DoubleMatrix::RandomUniform(1024, 512, c_deviceIdZero, -12.34f, 55.2312f, seed + 3);
DoubleMatrix matrixC1(matrixB1.DeepClone()); // C1==B1
const float beta = -1.4654f;
DoubleMatrix::ScaleAndAdd(alpha, matrixA1, beta, matrixC1); // C1=alpha*A1+beta*C1
foreach_coord (i, j, matrixC1)
{
BOOST_CHECK(fabsf(static_cast<float>(matrixC1(i, j) - (alpha * matrixA1(i, j) + beta * matrixB1(i, j)))) < c_epsilonDoubleE11);
}
// Test 3 - columnwise
// TODO: Split into separate test case WI# 82
DoubleMatrix matrixA2 = DoubleMatrix::RandomUniform(1024, 1, c_deviceIdZero, -12.34, 55.2312, seed + 4);
DoubleMatrix matrixB2 = DoubleMatrix::RandomUniform(1024, 512, c_deviceIdZero, -12.34, 55.2312, seed + 5); // Column
DoubleMatrix matrixC2(matrixB2.DeepClone()); // C2==B2
const float betaOne = 1;
DoubleMatrix::ScaleAndAdd(alpha, matrixA2, betaOne, matrixC2); // C2=alpha*A1+beta*C1
foreach_coord (i, j, matrixC2)
{
float x = static_cast<float>(matrixC2(i, j));
float y = static_cast<float>((alpha * matrixA2(i, 0)) + (betaOne * matrixB2(i, j)));
BOOST_CHECK(fabsf(x - y) < c_epsilonDoubleE11);
}
}
BOOST_FIXTURE_TEST_CASE(MatrixNorms, RandomSeedFixture)
{
SingleMatrix matrix0(c_value_2, c_value_3, c_deviceIdZero);
matrix0(0, 0) = 1;
matrix0(0, 1) = 2;
matrix0(0, 2) = 3;
matrix0(1, 0) = 4;
matrix0(1, 1) = 5;
matrix0(1, 2) = 6;
SingleMatrix matrix3(c_deviceIdZero);
matrix0.VectorNorm1(matrix3, true);
SingleMatrix matrix2(c_value_1, c_value_3, c_deviceIdZero);
matrix2(0, 0) = 5;
matrix2(0, 1) = 7;
matrix2(0, 2) = 9;
BOOST_CHECK(matrix3.IsEqualTo(matrix2));
matrix0.VectorNorm1(matrix3, false);
matrix2.Resize(2, 1);
matrix2(0, 0) = 6;
matrix2(1, 0) = 15;
BOOST_CHECK(matrix3.IsEqualTo(matrix2));
matrix0.VectorNorm2(matrix3, true);
matrix2.Resize(1, 3);
matrix2(0, 0) = 4.1231f;
matrix2(0, 1) = 5.3852f;
matrix2(0, 2) = 6.7082f;
BOOST_CHECK(matrix3.IsEqualTo(matrix2, c_epsilonFloat5E4));
matrix0.VectorNorm2(matrix3, false);
matrix2.Resize(2, 1);
matrix2(0, 0) = 3.7417f;
matrix2(1, 0) = 8.7750f;
BOOST_CHECK(matrix3.IsEqualTo(matrix2, c_epsilonFloat5E4));
SingleMatrix matrix00(c_value_2, c_value_3, c_deviceIdZero);
matrix00(0, 0) = 1;
matrix00(0, 1) = 2;
matrix00(0, 2) = 3;
matrix00(1, 0) = 4;
matrix00(1, 1) = 5;
matrix00(1, 2) = 6;
SingleMatrix matrix1(c_deviceIdZero);
matrix00.VectorMax(matrix1, matrix3, true);
matrix2.Resize(1, 3);
matrix2(0, 0) = 4;
matrix2(0, 1) = 5;
matrix2(0, 2) = 6;
BOOST_CHECK(matrix3.IsEqualTo(matrix2, c_epsilonFloatE4));
matrix00.VectorMax(matrix1, matrix3, false);
matrix2.Resize(2, 1);
matrix2(0, 0) = 3;
matrix2(1, 0) = 6;
BOOST_CHECK(matrix3.IsEqualTo(matrix2, c_epsilonFloatE4));
matrix0.VectorNormInf(matrix3, true);
matrix2.Resize(1, 3);
matrix2(0, 0) = 4;
matrix2(0, 1) = 5;
matrix2(0, 2) = 6;
BOOST_CHECK(matrix3.IsEqualTo(matrix2, c_epsilonFloatE4));
matrix0.VectorNormInf(matrix3, false);
matrix2.Resize(2, 1);
matrix2(0, 0) = 3;
matrix2(1, 0) = 6;
BOOST_CHECK(matrix3.IsEqualTo(matrix2));
matrix00(0, 0) = 1;
matrix00(0, 1) = 2;
matrix00(0, 2) = 3;
matrix00(1, 0) = 4;
matrix00(1, 1) = 5;
matrix00(1, 2) = 6;
BOOST_CHECK_EQUAL(6, matrix00.MatrixNormInf());
BOOST_CHECK(abs(matrix0.FrobeniusNorm() - 9.5394f) < c_epsilonFloatE4);
BOOST_CHECK(abs(matrix0.MatrixNormInf() - 6) < c_epsilonFloatE4);
BOOST_CHECK_EQUAL(21, matrix00.MatrixNorm1());
Matrix<float> matrixA = Matrix<float>::Eye(4096, c_deviceIdZero);
BOOST_CHECK_EQUAL(4096, matrixA.MatrixNorm0());
Matrix<float> matrixB = Matrix<float>::Eye(5, c_deviceIdZero);
BOOST_CHECK_EQUAL(5, matrixB.MatrixNorm0());
}
BOOST_FIXTURE_TEST_CASE(MatrixInnerProductOfMatrices, RandomSeedFixture)
{
SingleMatrix vector1(c_value_2, c_value_3, c_deviceIdZero);
vector1(0, 0) = 1;
vector1(0, 1) = 2;
vector1(0, 2) = 3;
vector1(1, 0) = 4;
vector1(1, 1) = 5;
vector1(1, 2) = 6;
SingleMatrix vector2(c_value_2, c_value_3, c_deviceIdZero);
vector2(0, 0) = 7;
vector2(0, 1) = 8;
vector2(0, 2) = 9;
vector2(1, 0) = 10;
vector2(1, 1) = 11;
vector2(1, 2) = 12;
const float ip = SingleMatrix::InnerProductOfMatrices(vector1, vector2);
BOOST_CHECK_EQUAL(217, ip);
}
BOOST_AUTO_TEST_SUITE_END()
}
} } }