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

Make addmv data qutity small

Browse Source
This commit is contained in:
kradchen
2023-12-19 14:20:13 +08:00
parent 81078bd69f
commit f56b04f014
1 changed files with 34 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files

74
test/CudaMatrix_Test.cpp
View File

@@ -279,18 +279,17 @@ TEST_F(CudaMatrix_Test, MatrixAddScalar) {
TEST_F(CudaMatrix_Test, MatrixAddmm) { TEST_F(CudaMatrix_Test, MatrixAddmm) {
//real //real
{ {
auto A = -Aurora::ones(4096,23519,1); auto A = -Aurora::ones(100,100,1);
auto B = Aurora::ones(4096,23519,1)*8; auto B = Aurora::ones(100,100,1)*8;
auto C = B-1; auto C = B-1;
auto dA = A.toDeviceMatrix(); auto dA = A.toDeviceMatrix();
auto dB = B.toDeviceMatrix(); auto dB = B.toDeviceMatrix();
//Matrix& + Matrix& //Matrix& + Matrix&
auto dC = (dA+dB); auto dC = (dA+dB);
EXPECT_TRUE(!dC.isComplex()); EXPECT_TRUE(!dC.isComplex());
EXPECT_EQ(4096, dC.getDimSize(0)); EXPECT_EQ(100, dC.getDimSize(0));
EXPECT_EQ(23519, dC.getDimSize(1)); EXPECT_EQ(100, dC.getDimSize(1));
EXPECT_EQ(1, dC.getDimSize(2)); EXPECT_EQ(1, dC.getDimSize(2));
#pragma omp parallel for
for (size_t i = 0; i < C.getDataSize(); i++) for (size_t i = 0; i < C.getDataSize(); i++)
{ {
EXPECT_FLOAT_EQ(C[i],dC.getValue(i)); EXPECT_FLOAT_EQ(C[i],dC.getValue(i));
@@ -298,33 +297,31 @@ TEST_F(CudaMatrix_Test, MatrixAddmm) {
//Matrix&& + Matrix& //Matrix&& + Matrix&
dC = A.toDeviceMatrix()+dB; dC = A.toDeviceMatrix()+dB;
EXPECT_TRUE(!dC.isComplex()); EXPECT_TRUE(!dC.isComplex());
EXPECT_EQ(4096, dC.getDimSize(0)); EXPECT_EQ(100, dC.getDimSize(0));
EXPECT_EQ(23519, dC.getDimSize(1)); EXPECT_EQ(100, dC.getDimSize(1));
EXPECT_EQ(1, dC.getDimSize(2)); EXPECT_EQ(1, dC.getDimSize(2));
// #pragma omp parallel for for (size_t i = 0; i < C.getDataSize(); i++)
// for (size_t i = 0; i < C.getDataSize(); i++) {
// { EXPECT_FLOAT_EQ(C[i],dC.getValue(i));
// EXPECT_FLOAT_EQ(C[i],dC.getValue(i)); }
// }
//Matrix& + Matrix&& //Matrix& + Matrix&&
dC = dB+A.toDeviceMatrix(); dC = dB+A.toDeviceMatrix();
EXPECT_TRUE(!dC.isComplex()); EXPECT_TRUE(!dC.isComplex());
EXPECT_EQ(4096, dC.getDimSize(0)); EXPECT_EQ(100, dC.getDimSize(0));
EXPECT_EQ(23519, dC.getDimSize(1)); EXPECT_EQ(100, dC.getDimSize(1));
EXPECT_EQ(1, dC.getDimSize(2)); EXPECT_EQ(1, dC.getDimSize(2));
// #pragma omp parallel for for (size_t i = 0; i < C.getDataSize(); i++)
// for (size_t i = 0; i < C.getDataSize(); i++) {
// { EXPECT_FLOAT_EQ(C[i],dC.getValue(i));
// EXPECT_FLOAT_EQ(C[i],dC.getValue(i)); }
// }
} }
//complex //complex
{ {
float *dataA = Aurora::random(4096*23519*2); float *dataA = Aurora::random(100*100*2);
float *dataB = Aurora::random(4096*23519*2); float *dataB = Aurora::random(100*100*2);
auto A = Aurora::Matrix::fromRawData(dataA, 4096,23519,1,Aurora::Complex); auto A = Aurora::Matrix::fromRawData(dataA, 100,100,1,Aurora::Complex);
auto B = Aurora::Matrix::fromRawData(dataB, 4096,23519,1,Aurora::Complex); auto B = Aurora::Matrix::fromRawData(dataB, 100,100,1,Aurora::Complex);
auto C = B+A; auto C = B+A;
auto dA = A.toDeviceMatrix(); auto dA = A.toDeviceMatrix();
@@ -332,10 +329,9 @@ TEST_F(CudaMatrix_Test, MatrixAddmm) {
//Matrix& + Matrix&, //Matrix& + Matrix&,
auto dC = (dA+dB); auto dC = (dA+dB);
EXPECT_TRUE(dC.isComplex()); EXPECT_TRUE(dC.isComplex());
EXPECT_EQ(4096, dC.getDimSize(0)); EXPECT_EQ(100, dC.getDimSize(0));
EXPECT_EQ(23519, dC.getDimSize(1)); EXPECT_EQ(100, dC.getDimSize(1));
EXPECT_EQ(1, dC.getDimSize(2)); EXPECT_EQ(1, dC.getDimSize(2));
#pragma omp parallel for
for (size_t i = 0; i < C.getDataSize(); i++) for (size_t i = 0; i < C.getDataSize(); i++)
{ {
EXPECT_FLOAT_EQ(C[i],dC.getValue(i)); EXPECT_FLOAT_EQ(C[i],dC.getValue(i));
@@ -343,25 +339,23 @@ TEST_F(CudaMatrix_Test, MatrixAddmm) {
//Matrix&& + Matrix&, //Matrix&& + Matrix&,
dC = A.toDeviceMatrix()+dB; dC = A.toDeviceMatrix()+dB;
EXPECT_TRUE(dC.isComplex()); EXPECT_TRUE(dC.isComplex());
EXPECT_EQ(4096, dC.getDimSize(0)); EXPECT_EQ(100, dC.getDimSize(0));
EXPECT_EQ(23519, dC.getDimSize(1)); EXPECT_EQ(100, dC.getDimSize(1));
EXPECT_EQ(1, dC.getDimSize(2)); EXPECT_EQ(1, dC.getDimSize(2));
// #pragma omp parallel for for (size_t i = 0; i < C.getDataSize(); i++)
// for (size_t i = 0; i < C.getDataSize(); i++) {
// { EXPECT_FLOAT_EQ(C[i],dC.getValue(i));
// EXPECT_FLOAT_EQ(C[i],dC.getValue(i)); }
// }
//Matrix& + Matrix&&, //Matrix& + Matrix&&,
dC = dB+A.toDeviceMatrix(); dC = dB+A.toDeviceMatrix();
EXPECT_TRUE(dC.isComplex()); EXPECT_TRUE(dC.isComplex());
EXPECT_EQ(4096, dC.getDimSize(0)); EXPECT_EQ(100, dC.getDimSize(0));
EXPECT_EQ(23519, dC.getDimSize(1)); EXPECT_EQ(100, dC.getDimSize(1));
EXPECT_EQ(1, dC.getDimSize(2)); EXPECT_EQ(1, dC.getDimSize(2));
// #pragma omp parallel for for (size_t i = 0; i < C.getDataSize(); i++)
// for (size_t i = 0; i < C.getDataSize(); i++) {
// { EXPECT_FLOAT_EQ(C[i],dC.getValue(i));
// EXPECT_FLOAT_EQ(C[i],dC.getValue(i)); }
// }
} }
} }