Mentions légales du service

Skip to content
Snippets Groups Projects
Commit 0c034cbe authored by hhakim's avatar hhakim
Browse files

Implement the TransformHelperPoly Faust-vector polynomial specialized multiplication on GPU.

parent cd61c0db
Branches
Tags
No related merge requests found
Hide whitespace changes
Inline Side-by-side
src/faust_linear_operator/CPU/faust_TransformHelperPoly.h
+ 7
− 3
View file @ 0c034cbe
......@@ -17,7 +17,7 @@ namespace Faust
};
template<typename FPP>
TransformHelper<FPP, Cpu>* basisChebyshev(MatSparse<FPP,Cpu>* L, int32_t K, MatSparse<FPP, Cpu>* T0=nullptr, BasisLaziness lazy_instantiation=INSTANTIATE_ONCE_AND_FOR_ALL);
TransformHelper<FPP, Cpu>* basisChebyshev(MatSparse<FPP,Cpu>* L, int32_t K, MatSparse<FPP, Cpu>* T0=nullptr, bool on_gpu=false, BasisLaziness lazy_instantiation=INSTANTIATE_ONCE_AND_FOR_ALL);
template<typename FPP>
void poly(int d, uint K, int n, const FPP* basisX, const FPP* coeffs, FPP* out);
......@@ -38,13 +38,15 @@ namespace Faust
std::vector<bool> is_fact_created;
BasisLaziness laziness;
bool T0_is_arbitrary;
bool mul_and_combi_lin_on_gpu;
// ctor is private on purpose (callees must call basisChebyshev() to instantiate an object
TransformHelperPoly(uint K,
MatSparse<FPP, Cpu> *L,
MatSparse<FPP, Cpu> *rR=nullptr,
MatSparse<FPP, Cpu> *T0=nullptr,
BasisLaziness laziness=INSTANTIATE_COMPUTE_AND_FREE);
BasisLaziness laziness=INSTANTIATE_COMPUTE_AND_FREE,
bool on_gpu=false);
TransformHelperPoly(uint K, const TransformHelperPoly<FPP>& src);
......@@ -109,6 +111,8 @@ namespace Faust
Vect<FPP,Cpu> multiply(const Vect<FPP,Cpu> &x, const bool transpose=false, const bool conjugate=false);
Vect<FPP,Cpu> multiply(const FPP* x, const bool transpose=false, const bool conjugate=false);
void multiply(const FPP* x, FPP* y, const bool transpose=false, const bool conjugate=false);
void multiply_cpu(const FPP* x, FPP* y, const bool transpose=false, const bool conjugate=false);
void multiply_gpu(const FPP* x, FPP* y, const bool transpose=false, const bool conjugate=false);
MatDense<FPP, Cpu> multiply(const MatDense<FPP,Cpu> &X, const bool transpose=false, const bool conjugate=false);
void multiply(const FPP* X, int n, FPP* out, const bool transpose=false, const bool conjugate=false);
TransformHelper<FPP, Cpu>* next(uint K);
......@@ -148,7 +152,7 @@ namespace Faust
void create_rR(const MatSparse<FPP,Cpu>* L);
~TransformHelperPoly();
friend TransformHelper<FPP,Cpu>* basisChebyshev<>(MatSparse<FPP,Cpu>* L, int32_t K, MatSparse<FPP, Cpu>* T0, BasisLaziness lazy_instantiation);
friend TransformHelper<FPP,Cpu>* basisChebyshev<>(MatSparse<FPP,Cpu>* L, int32_t K, MatSparse<FPP, Cpu>* T0, bool on_gpu, BasisLaziness lazy_instantiation);
};
......
src/faust_linear_operator/CPU/faust_TransformHelperPoly.hpp
+ 65
− 8
View file @ 0c034cbe
......@@ -9,7 +9,8 @@ namespace Faust
MatSparse<FPP, Cpu>* L,
MatSparse<FPP, Cpu>* rR /* = nullptr*/,
MatSparse<FPP,Cpu> *T0 /*= nullptr*/,
BasisLaziness laziness /*= INSTANTIATE_COMPUTE_AND_FREE */) : TransformHelper<FPP,Cpu>()
BasisLaziness laziness /*= INSTANTIATE_COMPUTE_AND_FREE */,
bool on_gpu /*= false*/) : TransformHelper<FPP,Cpu>()
{
// assuming L is symmetric
this->L = L;
......@@ -39,6 +40,7 @@ namespace Faust
if(laziness == NOT_LAZY)
this->basisChebyshev_all();
this->mul_and_combi_lin_on_gpu = on_gpu;
}
template<typename FPP>
......@@ -120,7 +122,7 @@ namespace Faust
}
template<typename FPP>
void TransformHelperPoly<FPP>::multiply(const FPP* x, FPP* y, const bool transpose/*=false*/, const bool conjugate/*=false*/)
void TransformHelperPoly<FPP>::multiply_cpu(const FPP* x, FPP* y, const bool transpose/*=false*/, const bool conjugate/*=false*/)
{
/**
* Recurrence relation (k=1 to K):
......@@ -142,7 +144,7 @@ namespace Faust
if(K == 0)
return;
Eigen::Map<Eigen::Matrix<FPP, Eigen::Dynamic, 1>> x_vec(const_cast<FPP*>(x), d);
Eigen::Map<Eigen::Matrix<FPP, Eigen::Dynamic, 1>>v2(const_cast<FPP*>(y+d), d);
Eigen::Map<Eigen::Matrix<FPP, Eigen::Dynamic, 1>> v2(const_cast<FPP*>(y+d), d);
v2 = L->mat*x_vec;
if(K == 1) // not necessary but clearer
return;
......@@ -152,6 +154,57 @@ namespace Faust
Eigen::Map<Eigen::Matrix<FPP, Eigen::Dynamic, 1>>v2_(const_cast<FPP*>(y+d*(i-2)), d);
Eigen::Map<Eigen::Matrix<FPP, Eigen::Dynamic, 1>>v1_(const_cast<FPP*>(y+d*(i-3)), d);
new_v2_ = L->mat*v2_*2-v1_;
}
}
template<typename FPP>
void TransformHelperPoly<FPP>::multiply_gpu(const FPP* x, FPP* y, const bool transpose/*=false*/, const bool conjugate/*=false*/)
{
#ifdef USE_GPU_MOD
int d = L->getNbRow();
uint K = this->size()-1;
Vect<FPP, GPU2> gpu_v1(d, x);
Vect<FPP, GPU2> gpu_v2(gpu_v1);
Vect<FPP, GPU2> gpu_new_v2(d);
const MatSparse<FPP, GPU2> gpu_L(*this->L);
MatSparse<FPP, GPU2> gpu_twoL(gpu_L);
gpu_twoL *= 2;
memcpy(y, x, sizeof(FPP)*d);
if(K == 0)
return;
Eigen::Map<Eigen::Matrix<FPP, Eigen::Dynamic, 1>> v2(const_cast<FPP*>(y+d), d);
// gpu_v2 == x
gpu_v2.multiplyLeft(gpu_L);
gpu_v2.tocpu(y+d); // v2 = L->mat*x_vec;
if(K == 1) // not necessary but clearer
return;
for(int i=3;i<=K+1;i++)
{
Eigen::Map<Eigen::Matrix<FPP, Eigen::Dynamic, 1>>new_v2_cpu(const_cast<FPP*>(y+d*(i-1)), d);
gpu_new_v2 = gpu_v2;
gpu_new_v2.multiplyLeft(const_cast<const MatSparse<FPP,GPU2>&>(gpu_twoL));
gpu_new_v2 -= gpu_v1; // new_v2_ = L->mat*v2_*2-v1_;
// prepare next it
gpu_v1 = gpu_v2;
gpu_v2 = gpu_new_v2;
gpu_new_v2.tocpu(new_v2_cpu.data());
}
#else
throw std::runtime_error("USE_GPU_MOD option must be enabled at compiling time to use this function (TransformHelperPoly<FPP>::multiply_gpu).");
#endif
}
template<typename FPP>
void TransformHelperPoly<FPP>::multiply(const FPP* x, FPP* y, const bool transpose/*=false*/, const bool conjugate/*=false*/)
{
if(this->mul_and_combi_lin_on_gpu)
{
multiply_gpu(x, y, transpose, conjugate);
}
else
{
multiply_cpu(x, y, transpose, conjugate);
}
}
......@@ -254,7 +307,7 @@ namespace Faust
int d = L->getNbRow();
uint K = this->size()-1;
auto scale = (K+1)*d;
#pragma omp parallel for
#pragma omp parallel for
for(int i=0;i<n;i++)
{
// Vect<FPP,Cpu> x(d, X.getData()+i*d);
......@@ -278,7 +331,7 @@ namespace Faust
if(std::find(std::begin(col_ids), std::end(col_ids), id) == std::end(col_ids))
col_ids.push_back(id);
}
#pragma omp parallel for
#pragma omp parallel for
for(auto i=col_ids.begin(); i < col_ids.end();i++)
{
auto id = *i;
......@@ -511,7 +564,7 @@ namespace Faust
}
template<typename FPP>
TransformHelper<FPP,Cpu>* basisChebyshev(MatSparse<FPP,Cpu>* L, int32_t K, MatSparse<FPP, Cpu>* T0/*=nullptr*/, BasisLaziness lazy_instantiation/*=INSTANTIATE_COMPUTE_AND_FREE*/)
TransformHelper<FPP,Cpu>* basisChebyshev(MatSparse<FPP,Cpu>* L, int32_t K, MatSparse<FPP, Cpu>* T0/*=nullptr*/, bool on_gpu/*=false*/, BasisLaziness lazy_instantiation/*=INSTANTIATE_COMPUTE_AND_FREE*/)
{
// assuming L is symmetric
auto basisP = new TransformHelperPoly<FPP>(
......@@ -519,7 +572,8 @@ namespace Faust
new MatSparse<FPP, Cpu>(*L),
nullptr /* rR initialized by the ctor */,
T0,
lazy_instantiation);
lazy_instantiation,
on_gpu);
return basisP;
}
......@@ -803,7 +857,10 @@ namespace Faust
str <<", density "<< density << ", nnz_sum "<<this->get_total_nnz() << ", " << this->size() << " factor(s): "<< std::endl;
for (int i=0 ; i<this->size() ; i++)
{
str << "- FACTOR " << i;
str << "- ";
if(this->mul_and_combi_lin_on_gpu)
str << "GPU ";
str << "FACTOR " << i;
density = (double) this->get_fact_nnz(i) / this->get_fact_nb_rows(i) / this->get_fact_nb_cols(i);
str << Faust::MatGeneric<FPP,Cpu>::to_string(this->get_fact_nb_rows(i), this->get_fact_nb_cols(i), this->is_transposed, density, this->get_fact_nnz(i), /* is_identity */ i == this->size()-1, Sparse);
}
......
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment