Mentions légales du service

Skip to content
Snippets Groups Projects
Commit 2d1487c4 authored by hhakim's avatar hhakim
Browse files

Update to gpu_mod@bd7eced3 and add functions to MatDense<FPP,GPU2>: add, abs, normL1 with tests.

parent dfa70815
Branches
Tags
No related merge requests found
Hide whitespace changes
Inline Side-by-side
misc/test/src/C++/test_matdense_gpu_mod.cpp.in
+ 182
− 1
View file @ 2d1487c4
......@@ -27,12 +27,14 @@ void test_mul_gpu_dense()
auto err_diff = cpu_mat1_mat2_ref;
err_diff -= cpu_mat1_mat2_test;
cout << "err mul.: " << err_diff.norm()/cpu_mat1_mat2_ref.norm() << endl;
cout << "OK" << endl;
cout << "Mul. GPUDense*CPUDense in CPUDense" << endl;
cpu_mat1_mat2_test = *cpu_mat2;
gpu_mat1.multiply(cpu_mat1_mat2_test);
err_diff = cpu_mat1_mat2_ref;
err_diff -= cpu_mat1_mat2_test;
cout << "err mul.: " << err_diff.norm()/cpu_mat1_mat2_ref.norm() << endl;
cout << "OK" << endl;
cout << "Mul. GPUDense*CPUSparse in CPUDense" << endl;
Faust::MatSparse<double,Cpu> cpu_mat2_sparse(*cpu_mat2);
// cout << cpu_mat2_sparse.to_string(false, true) << endl;
......@@ -45,6 +47,7 @@ void test_mul_gpu_dense()
// cout << cpu_mat1_mat2_ref.to_string(false, true) << endl;
// cout << cpu_mat1_mat2_test.to_string(false, true) << endl;
cout << "err mul.: " << err_diff.norm()/cpu_mat1_mat2_ref.norm() << endl;
cout << "OK" << endl;
cout << "Mul. GPUDense*CPUSparse in GPUDense" << endl;
MatDense<double, GPU2> gpu_mat1_mat2_test(nrows, ncols2);
gpu_mat1.multiply(cpu_mat2_sparse, gpu_mat1_mat2_test);
......@@ -52,6 +55,22 @@ void test_mul_gpu_dense()
err_diff = gpu_mat1_mat2_test_to_cpu;
err_diff -= cpu_mat1_mat2_ref;
cout << "err mul.: " << err_diff.norm()/cpu_mat1_mat2_ref.norm() << endl;
cout << "OK" << endl;
cout << "MatDense*cpu_vec" << endl;
Faust::Vect<double, Cpu>* vec = Faust::Vect<double,Cpu>::rand(cpu_mat1->getNbCol());
Faust::Vect<double, Cpu> vec_copy(*vec);
cout << vec->norm() << endl;
cout << vec_copy.norm() << endl;
gpu_mat1.multiply(*vec);
vec_copy = cpu_mat1->multiply(vec_copy);
// cout << vec->size() << endl;
// cout << vec_copy.size() << endl;
// cout << vec->norm() << endl;
// cout << vec_copy.norm() << endl;
auto err_vec_diff = vec_copy;
err_vec_diff -= *vec;
assert(err_vec_diff.norm() / vec_copy.norm() < 1e-3);
cout << "OK" << endl;
}
void test_resize()
......@@ -61,6 +80,7 @@ void test_resize()
auto cpu_mat1 = Faust::MatDense<double,Cpu>::randMat(nrows,ncols);
MatDense<double,GPU2> gpu_mat1(nrows, ncols, cpu_mat1->getData());
gpu_mat1.resize(45, 12);
cout << "OK" << endl;
}
void test_setones()
......@@ -76,6 +96,7 @@ void test_setones()
auto err_diff = *cpu_mat1;
err_diff -= gpu_mat1_to_cpu;
cout << "err setones.: " << err_diff.norm()/cpu_mat1->norm() << endl;
cout << "OK" << endl;
}
void test_setzeros()
......@@ -88,6 +109,7 @@ void test_setzeros()
auto gpu_mat1_to_cpu = gpu_mat1.tocpu();
// cout << gpu_mat1_to_cpu.to_string(false, true) << endl;
assert(gpu_mat1_to_cpu.norm() == 0.);
cout << "OK" << endl;
}
void test_seteyes()
......@@ -103,7 +125,7 @@ void test_seteyes()
auto err_diff = *cpu_mat1;
err_diff -= gpu_mat1_to_cpu;
cout << "err seteyes.: " << err_diff.norm()/cpu_mat1->norm() << endl;
cout << "OK" << endl;
}
void test_clone()
......@@ -120,6 +142,157 @@ void test_clone()
// cout << gpu_mat1_to_cpu.to_string(false, true) << endl;
// cout << gpu_mat1_clone_to_cpu.to_string(false, true) << endl;
cout << "err clone: " << err_diff.norm()/cpu_mat1->norm() << endl;
cout << "OK" << endl;
}
void test_transpose()
{
cout << "Test transpose" << endl;
faust_unsigned_int nrows = 11, ncols = 9;
auto cpu_mat1 = Faust::MatDense<double,Cpu>::randMat(nrows,ncols);
MatDense<double,GPU2> gpu_mat1(nrows, ncols, cpu_mat1->getData());
// cout << cpu_mat1->to_string(false, true) << endl;
cpu_mat1->transpose();
gpu_mat1.transpose();
auto gpu_mat1_to_cpu = gpu_mat1.tocpu();
auto err_diff = *cpu_mat1;
err_diff -= gpu_mat1_to_cpu;
cout << "err transpose: " << err_diff.norm()/cpu_mat1->norm() << endl;
// cout << gpu_mat1_to_cpu.to_string(false, true) << endl;
assert(nrows == cpu_mat1->getNbCol() && nrows == gpu_mat1_to_cpu.getNbCol() && ncols == gpu_mat1_to_cpu.getNbRow() && ncols == cpu_mat1->getNbRow());
cout << "OK" << endl;
}
void test_adjoint()
{
cout << "Test adjoint" << endl;
faust_unsigned_int nrows = 11, ncols = 9;
auto cpu_mat1 = Faust::MatDense<double,Cpu>::randMat(nrows,ncols);
MatDense<double,GPU2> gpu_mat1(nrows, ncols, cpu_mat1->getData());
// cout << cpu_mat1->to_string(false, true) << endl;
cpu_mat1->adjoint();
gpu_mat1.adjoint();
auto gpu_mat1_to_cpu = gpu_mat1.tocpu();
auto err_diff = *cpu_mat1;
err_diff -= gpu_mat1_to_cpu;
cout << "err adjoint: " << err_diff.norm()/cpu_mat1->norm() << endl;
// cout << gpu_mat1_to_cpu.to_string(false, true) << endl;
assert(nrows == cpu_mat1->getNbCol() && nrows == gpu_mat1_to_cpu.getNbCol() && ncols == gpu_mat1_to_cpu.getNbRow() && ncols == cpu_mat1->getNbRow());
cout << "OK" << endl;
}
void test_conjugate()
{
cout << "Test conjugate" << endl;
faust_unsigned_int nrows = 11, ncols = 9;
auto cpu_mat1 = Faust::MatDense<double,Cpu>::randMat(nrows,ncols);
MatDense<double,GPU2> gpu_mat1(nrows, ncols, cpu_mat1->getData());
// cout << cpu_mat1->to_string(false, true) << endl;
cpu_mat1->conjugate();
gpu_mat1.conjugate();
auto gpu_mat1_to_cpu = gpu_mat1.tocpu();
auto err_diff = *cpu_mat1;
err_diff -= gpu_mat1_to_cpu;
cout << "err conjugate: " << err_diff.norm()/cpu_mat1->norm() << endl;
// cout << gpu_mat1_to_cpu.to_string(false, true) << endl;
assert(ncols == cpu_mat1->getNbCol() && ncols == gpu_mat1_to_cpu.getNbCol() && nrows == gpu_mat1_to_cpu.getNbRow() && nrows == cpu_mat1->getNbRow());
cout << "OK" << endl;
}
void test_norm()
{
cout << "test spectral norm" << endl;
faust_unsigned_int nrows = 12, ncols = 9;
int flag;
auto cpu_mat1 = Faust::MatDense<double,Cpu>::randMat(nrows,ncols);
MatDense<double,GPU2> gpu_mat1(nrows, ncols, cpu_mat1->getData());
auto gpu_norm = gpu_mat1.spectralNorm(1000, (float)1e-6);
auto cpu_norm = cpu_mat1->spectralNorm(1000, 1e-6, flag);
// cout << gpu_norm << endl;
// cout << cpu_norm << endl;
assert(abs(cpu_norm-gpu_norm) < 1e-6);
cout << "OK" << endl;
cout << "test frob norm" << endl;
gpu_norm = gpu_mat1.norm();
cpu_norm = cpu_mat1->norm();
assert(abs(cpu_norm-gpu_norm) < 1e-6);
cout << "OK" << endl;
cout << "test l1 norm" << endl;
gpu_norm = gpu_mat1.normL1();
cpu_norm = cpu_mat1->normL1();
// cout << "gpu: " << gpu_norm << " cpu: " << cpu_norm << endl;
assert(abs(cpu_norm-gpu_norm) < 1e-6);
cout << "OK" << endl;
}
void test_normalize()
{
cout << "test normalize" << endl;
faust_unsigned_int nrows = 11, ncols = 9;
auto cpu_mat1 = Faust::MatDense<double,Cpu>::randMat(nrows,ncols);
MatDense<double,GPU2> gpu_mat1(nrows, ncols, cpu_mat1->getData());
cpu_mat1->normalize();
gpu_mat1.normalize();
auto gpu_mat1_to_cpu = gpu_mat1.tocpu();
auto err_diff = *cpu_mat1;
err_diff -= gpu_mat1_to_cpu;
cout << "err normalize: " << err_diff.norm()/cpu_mat1->norm() << endl;
assert(abs(gpu_mat1.norm()-1) < 1e-6);
cout << "OK" << endl;
}
void test_scalar_mul()
{
cout << "test scalar mul" << endl;
faust_unsigned_int nrows = 11, ncols = 9;
auto lambda = 12.;
auto cpu_mat1 = Faust::MatDense<double,Cpu>::randMat(nrows,ncols);
MatDense<double,GPU2> gpu_mat1(nrows, ncols, cpu_mat1->getData());
cpu_mat1->scalarMultiply(lambda);
gpu_mat1.scalarMultiply(lambda);
auto gpu_mat1_to_cpu = gpu_mat1.tocpu();
auto err_diff = *cpu_mat1;
err_diff -= gpu_mat1_to_cpu;
cout << "err scalar mul: " << err_diff.norm()/cpu_mat1->norm() << endl;
cout << "OK" << endl;
}
void test_abs()
{
cout << "test abs" << endl;
faust_unsigned_int nrows = 11, ncols = 9;
auto lambda = 12.;
auto cpu_mat1 = Faust::MatDense<double,Cpu>::randMat(nrows,ncols);
cpu_mat1->scalarMultiply(-1);
cout << cpu_mat1->to_string(false, true) << endl;
MatDense<double,GPU2> gpu_mat1(nrows, ncols, cpu_mat1->getData());
cpu_mat1->abs();
gpu_mat1.abs();
auto gpu_mat1_to_cpu = gpu_mat1.tocpu();
auto err_diff = *cpu_mat1;
cout << gpu_mat1_to_cpu.to_string(false, true) << endl;
cout << cpu_mat1->to_string(false, true) << endl;
err_diff -= gpu_mat1_to_cpu;
cout << "err scalar mul: " << err_diff.norm()/cpu_mat1->norm() << endl;
cout << "OK" << endl;
}
void test_add()
{
cout << "test add" << endl;
faust_unsigned_int nrows = 11, ncols = 9;
auto lambda = 12.;
auto cpu_mat1 = Faust::MatDense<double,Cpu>::randMat(nrows,ncols);
auto cpu_mat2 = Faust::MatDense<double,Cpu>::randMat(nrows,ncols);
MatDense<double,GPU2> gpu_mat1(nrows, ncols, cpu_mat1->getData());
cpu_mat1->add(*cpu_mat2);
gpu_mat1.add(*cpu_mat2);
auto gpu_mat1_to_cpu = gpu_mat1.tocpu();
auto err_diff = *cpu_mat1;
err_diff -= gpu_mat1_to_cpu;
auto err = err_diff.norm()/cpu_mat1->norm();
assert(err < 1e-6);
cout << "OK" << endl;
}
int main(int argc, char** argv)
......@@ -131,5 +304,13 @@ int main(int argc, char** argv)
test_setzeros();
test_seteyes();
test_clone();
test_transpose();
test_conjugate();
test_adjoint();
test_norm();
test_normalize();
test_scalar_mul();
test_abs();
test_add();
return EXIT_SUCCESS;
}
src/faust_linear_operator/GPU2/faust_MatDense_gpu.h
+ 4
− 0
View file @ 2d1487c4
......@@ -16,6 +16,8 @@ namespace Faust
public:
MatDense(const faust_unsigned_int nbRow, const faust_unsigned_int nbCol, const FPP* data = nullptr, const bool no_alloc=false);
void add(MatDense<FPP,Cpu> const& A);
void add(MatDense<FPP,GPU2> const& A);
// vec = this * vec
Vect<FPP, Cpu> multiply(const Vect<FPP, Cpu> &vec);
void multiply(MatDense<FPP, GPU2> &other, const char op_this='N');
......@@ -32,8 +34,10 @@ namespace Faust
void transpose();
void conjugate();
void adjoint();
void abs();
Real<FPP> spectralNorm(const faust_unsigned_int nbr_iter_max, const float threshold);
Real<FPP> norm();
Real<FPP> normL1();
void normalize();
MatDense<FPP, GPU2>* clone();
MatDense<FPP, Cpu> tocpu();
......
src/faust_linear_operator/GPU2/faust_MatDense_gpu.hpp.in
+ 28
− 0
View file @ 2d1487c4
......@@ -190,3 +190,31 @@ void Faust::MatDense<@FAUST_SCALAR_FOR_GM@,GPU2>::normalize()
auto dsm_funcs = ((gm_DenseMatFunc_@GM_SCALAR@*)this->dsm_funcs);
dsm_funcs->normalize(gpu_mat);
}
template<>
void Faust::MatDense<@FAUST_SCALAR_FOR_GM@,GPU2>::abs()
{
auto dsm_funcs = ((gm_DenseMatFunc_@GM_SCALAR@*)this->dsm_funcs);
dsm_funcs->abs(gpu_mat);
}
template<>
Real<@FAUST_SCALAR_FOR_GM@> Faust::MatDense<@FAUST_SCALAR_FOR_GM@,GPU2>::normL1()
{
auto dsm_funcs = ((gm_DenseMatFunc_@GM_SCALAR@*)this->dsm_funcs);
return dsm_funcs->norm_l1(gpu_mat);
}
template<>
void Faust::MatDense<@FAUST_SCALAR_FOR_GM@, GPU2>::add(MatDense<@FAUST_SCALAR_FOR_GM@,GPU2> const& A)
{
auto dsm_funcs = ((gm_DenseMatFunc_@GM_SCALAR@*)this->dsm_funcs);
dsm_funcs->add_gpu_dsm(this->gpu_mat, A.gpu_mat);
}
template<>
void Faust::MatDense<@FAUST_SCALAR_FOR_GM@, GPU2>::add(MatDense<@FAUST_SCALAR_FOR_GM@,Cpu> const& A)
{
auto dsm_funcs = ((gm_DenseMatFunc_@GM_SCALAR@*)this->dsm_funcs);
dsm_funcs->add_cpu_dsm(gpu_mat, A.getData(), A.getNbRow(), A.getNbCol());
}
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment