sunwen/Aurora
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add +,-,*,/,neg,pow to CudaMatrix

Browse Source
This commit is contained in:
kradchen
2023-11-20 09:48:39 +08:00
parent aabe9d1fd6
commit 761aea2acc
3 changed files with 585 additions and 112 deletions
Download Patch File Download Diff File Expand all files Collapse all files

240
test/CudaMatrix_Test.cpp
View File

@@ -186,6 +186,246 @@ TEST_F(CudaMatrix_Test, MatrixMul) {
}
}
TEST_F(CudaMatrix_Test, MatrixSub) {
auto A = Aurora::zeros(1000,1,1);
auto B = Aurora::zeros(1000,1,1);
for (size_t i = 0; i < 1000; i++)
{
A[i] = -1;
B[i] = i;
}
printf("Test CudaMatrix operator-(const CudaMatrix &aMatrix) const \r\n");
//CudaMatrix operator+(const CudaMatrix &aMatrix) const
auto C = A-B;
auto dA = A.toDeviceMatrix();
auto dB = B.toDeviceMatrix();
auto dC = (dA-dB);
auto dhC = dC.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(C[i],dhC[i]);
}
printf("Test CudaMatrix operator-(float aScalar) const \r\n");
//CudaMatrix operator+(float aScalar) const
auto D = C-0.5;
auto dD = dC-0.5;
auto dhD = dD.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(D[i],dhD[i]);
}
printf("Test CudaMatrix operator-(float aScalar, const CudaMatrix &aMatrix) \r\n");
// CudaMatrix operator+(float aScalar, const CudaMatrix &aMatrix)
D = 0.5-C;
dD = 0.5 - dC;
dhD = dD.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(D[i],dhD[i]);
}
printf("Test CudaMatrix &operator-(float aScalar, CudaMatrix &&aMatrix) \r\n");
// CudaMatrix &operator+(float aScalar, CudaMatrix &&aMatrix)
{
auto dD2 = 0.5 - (dA-dB);
dhD = dD2.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(D[i],dhD[i]);
}
}
printf("Test CudaMatrix &operator-(CudaMatrix &&aMatrix, float aScalar) \r\n");
// CudaMatrix &operator+(CudaMatrix &&aMatrix, float aScalar)
{
auto E = C-0.5;
auto dE2 = (dA-dB)-0.5;
auto dhE = dE2.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(E[i],dhE[i]);
}
}
//CudaMatrix operator+(CudaMatrix &&aMatrix) const
printf("Test CudaMatrix operator-(CudaMatrix &&aMatrix) const \r\n");
{
auto D = A-C;
auto dD2 = dA-(dA-dB);
dhD = dD2.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(D[i],dhD[i]);
}
}
//CudaMatrix operator+(CudaMatrix &&aMatrix,CudaMatrix &aOther)
printf("Test CudaMatrix operator-(CudaMatrix &&aMatrix,CudaMatrix &aOther) \r\n");
{
auto D = C-A;
auto dD2 = (dA-dB)-dA;
dhD = dD2.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(D[i],dhD[i]);
}
}
}
TEST_F(CudaMatrix_Test, MatrixDiv) {
auto A = Aurora::zeros(1000,1,1);
auto B = Aurora::zeros(1000,1,1);
for (size_t i = 0; i < 1000; i++)
{
A[i] = -1;
B[i] = i;
}
printf("Test CudaMatrix operator/(const CudaMatrix &aMatrix) const \r\n");
//CudaMatrix operator+(const CudaMatrix &aMatrix) const
auto C = A/B;
auto dA = A.toDeviceMatrix();
auto dB = B.toDeviceMatrix();
auto dC = (dA/dB);
auto dhC = dC.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(C[i],dhC[i]);
}
printf("Test CudaMatrix operator/(float aScalar) const \r\n");
//CudaMatrix operator+(float aScalar) const
auto D = C/0.5;
auto dD = dC/0.5;
auto dhD = dD.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(D[i],dhD[i]);
}
printf("Test CudaMatrix operator/(float aScalar, const CudaMatrix &aMatrix) \r\n");
// CudaMatrix operator+(float aScalar, const CudaMatrix &aMatrix)
D = 0.5/C;
dD = 0.5 / dC;
dhD = dD.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(D[i],dhD[i]);
}
printf("Test CudaMatrix &operator/(float aScalar, CudaMatrix &&aMatrix) \r\n");
// CudaMatrix &operator+(float aScalar, CudaMatrix &&aMatrix)
{
auto dD2 = 0.5 / (dA/dB);
dhD = dD2.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(D[i],dhD[i]);
}
}
printf("Test CudaMatrix &operator/(CudaMatrix &&aMatrix, float aScalar) \r\n");
// CudaMatrix &operator+(CudaMatrix &&aMatrix, float aScalar)
{
auto E = C/0.5;
auto dE2 = (dA/dB)/0.5;
auto dhE = dE2.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(E[i],dhE[i]);
}
}
//CudaMatrix operator+(CudaMatrix &&aMatrix) const
printf("Test CudaMatrix operator/(CudaMatrix &&aMatrix) const \r\n");
{
auto D = A/C;
auto dD2 = dA/(dA/dB);
dhD = dD2.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(D[i],dhD[i]);
}
}
//CudaMatrix operator+(CudaMatrix &&aMatrix,CudaMatrix &aOther)
printf("Test CudaMatrix operator/(CudaMatrix &&aMatrix,CudaMatrix &aOther) \r\n");
{
auto D = C/A;
auto dD2 = (dA/dB)/dA;
dhD = dD2.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(D[i],dhD[i]);
}
}
}
TEST_F(CudaMatrix_Test, MatrixPow){
auto A = Aurora::zeros(1000,1,1);
auto B = Aurora::zeros(1000,1,1);
for (size_t i = 0; i < 1000; i++)
{
A[i] = -1+0.2*i;
}
auto dA= A.toDeviceMatrix();
{
auto R = A^0;
auto dR = dA^0;
auto dhR = dR.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(R[i],dhR[i]);
}
}
{
auto R = A^1;
auto dR = dA^1;
auto dhR = dR.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(R[i],dhR[i]);
}
}
{
auto R = A^2;
auto dR = dA^2;
auto dhR = dR.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(R[i],dhR[i]);
}
}
{
auto R = A^3;
auto dR = dA^3;
auto dhR = dR.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(R[i],dhR[i]);
}
}
{
auto R = A^5;
auto dR = dA^5;
auto dhR = dR.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(R[i],dhR[i]);
}
}
}
TEST_F(CudaMatrix_Test, MatrixNeg){
auto A = Aurora::zeros(1000,1,1);
auto B = Aurora::zeros(1000,1,1);
for (size_t i = 0; i < 1000; i++)
{
A[i] = -1+0.2*i;
}
auto dA= A.toDeviceMatrix();
{
auto R = -A;
auto dR = -dA;
auto dhR = dR.toHostMatrix();
for (size_t i = 0; i < 1000; i++)
{
ASSERT_FLOAT_EQ(R[i],dhR[i]);
}
}
}
TEST_F(CudaMatrix_Test, matrixfunction)
{
printf("Test CudaMatrix block(int aDim,int aBeginIndx, int aEndIndex) const\r\n");