Add MatGeneric<FPP,GPU2> abstract parent class for MatSparse/Dense<FPP,GPU2>...

Add MatGeneric<FPP,GPU2> abstract parent class for MatSparse/Dense<FPP,GPU2> and implement getNbRow/Col() in child classes.

gpu_mod @ f73ff026

-Subproject commit e4f1060e5e2973fb2ac4b59520c158c03ec60116
+Subproject commit f73ff0267aac9210fa6d235b550d27db036872cd

src/faust_linear_operator/CPU/faust_gpu_mod.h

@@ -61,7 +61,7 @@ namespace Faust
 
 		void insert(const Faust::MatGeneric<FPP,Cpu>* M, int32_t id);
 
-		/** TODO: this function should be deleted because it definitely slower than the version 2 */
+		/** TODO: this function should be deleted because it is definitely slower than the version 2 */
 		FPP power_iteration(int32_t max_iter, Real<FPP> threshold, int& flag);
 
 		/** This version differs from power_iteration in the way it computes F'Fv in this order F'(Fv)

src/faust_linear_operator/CPU/faust_gpu_mod.hpp

@@ -18,6 +18,7 @@ bool Faust::FaustGPU<FPP>::are_cpu_mat_all_known(const std::vector<MatGeneric<FP
 
 namespace Faust
 {
+	//TODO: these functions have to move elsewhere more appropriate or simply disappear
 	template<>
 		void set_one<double>(double* scal)
 		{

src/faust_linear_operator/CPU/faust_gpu_mod_gen.hpp.in

@@ -200,7 +200,7 @@ namespace Faust
 	Faust::RefManager FaustGPU<@FAUST_SCALAR_FOR_GM@>::ref_man([](void *fact)
 		{
 			auto gp_funcs = GPUModHandler::get_singleton()->gp_funcs();
-			//normally cpu_gpu_map must contains a the key fac if ref_man knew it (see ctor)
+			//normally cpu_gpu_map must contain the fact key if ref_man knew it (see ctor)
 			gp_funcs->free_mat(Faust::FaustGPU<@FAUST_SCALAR_FOR_GM@>::cpu_gpu_map[fact]);
 			Faust::FaustGPU<@FAUST_SCALAR_FOR_GM@>::cpu_gpu_map.erase(fact);
 		});

src/faust_linear_operator/GPU2/faust_MatDense_gpu.h

@@ -2,18 +2,17 @@
 #define __FAUST_MATDENSE_GPU2__
 #ifdef USE_GPU_MOD
 #include "faust_MatDense.h"
+#include "faust_MatGeneric_gpu.h"
 #include "faust_gpu_mod_utils.h"
 namespace Faust
 {
-
-	template<typename FPP,FDevice DEVICE>
-    class MatDense;
-
 	template <typename FPP>
 	void gemm(const MatDense<FPP, GPU2> &A, const MatDense<FPP, GPU2> &B, MatDense<FPP, GPU2> &C, const FPP& alpha, const FPP& beta, const char opA, const char opB);
 
+
+
 	template<typename FPP>
-		class MatDense<FPP, GPU2> : MatDense<FPP, Cpu>
+		class MatDense<FPP, GPU2> : public MatGeneric<FPP,GPU2>
 		{
 			friend void gemm<>(const MatDense<FPP, GPU2> &A, const MatDense<FPP, GPU2> &B, MatDense<FPP, GPU2> &C, const FPP& alpha, const FPP& beta, const char opA, const char opB);
 
@@ -86,11 +85,14 @@ namespace Faust
 				MatDense<FPP, Cpu> tocpu(const void* stream=nullptr) const;
 				void Display() const;
 				std::string to_string(const bool transpose=false, const bool displaying_small_mat_elts=false) const;
+				int32_t getNbRow() const;
+				int32_t getNbCol() const;
 			private:
 				gm_DenseMat_t gpu_mat;
 		};
 
-};
+
+}
 #include "faust_MatDense_gpu_double.hpp"
 #endif
 #endif

src/faust_linear_operator/GPU2/faust_MatDense_gpu.hpp.in

 //TODO: move to CPP
 namespace Faust
 {
+	    template<>
+		int32_t Faust::MatDense<@FAUST_SCALAR_FOR_GM@, GPU2>::getNbRow() const
+		{
+		    int32_t nrows;
+		    auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
+		    dsm_funcs->info(gpu_mat, &nrows, nullptr);
+		    return nrows;
+		}
+
+	    template<>
+		int32_t Faust::MatDense<@FAUST_SCALAR_FOR_GM@, GPU2>::getNbCol() const
+		{
+		    int32_t ncols;
+		    auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
+		    dsm_funcs->info(gpu_mat, nullptr, &ncols);
+		    return ncols;
+		}
+
 	template<>
 		Faust::MatDense<@FAUST_SCALAR_FOR_GM@,GPU2>::MatDense(
 				const faust_unsigned_int nbRow,
@@ -8,10 +26,8 @@ namespace Faust
 				const @FAUST_SCALAR_FOR_GM@* data/*=nullptr*/,
 				const bool no_alloc/*= false*/,
 				const int32_t dev_id/*=-1*/,
-				const void* stream/*=nullptr*/)
+				const void* stream/*=nullptr*/) : MatGeneric<@FAUST_SCALAR_FOR_GM@,GPU2>()
 		{
-			this->dim1 = nbRow;
-			this->dim2 = nbCol;
 			auto gp_funcs = GPUModHandler::get_singleton()->gp_funcs();
 			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
 			// create the matrix in the device dev_id memory
@@ -45,18 +61,17 @@ namespace Faust
 			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
 			dsm_funcs->mul_gpu_dsm_ext(this->gpu_mat, other.gpu_mat, other.gpu_mat, OP_NOTRANSP, OP_NOTRANSP);
 			//TODO: update dims (in transpose/adjoint case)
-			other.dim1 = dim1;
 		}
 
 	template<>
 		void Faust::MatDense<@FAUST_SCALAR_FOR_GM@,GPU2>::multiply(MatDense<@FAUST_SCALAR_FOR_GM@, Cpu> &other, const char op_this)
 		{
 			// other = this * other
+			//TODO: update dims (in transpose/adjoint case)
 			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
-			MatDense<@FAUST_SCALAR_FOR_GM@, GPU2> other_gpu_mat(other.dim1, other.dim2,other.getData());
+			MatDense<@FAUST_SCALAR_FOR_GM@, GPU2> other_gpu_mat(other.getNbRow(), other.getNbCol(), other.getData());
+			other.resize(this->getNbRow(), other.getNbCol());
 			dsm_funcs->mul_gpu_dsm_tocpu_ext(this->gpu_mat, other_gpu_mat.gpu_mat, other.getData(), OP_NOTRANSP, OP_NOTRANSP);
-			//TODO: update dims (in transpose/adjoint case)
-			other.dim1 = dim1;
 		}
 
 	template<>
@@ -64,7 +79,7 @@ namespace Faust
 		{
 			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
 			MatDense<@FAUST_SCALAR_FOR_GM@, GPU2> gpu_vec(vec.size(), 1, vec.getData());
-			Faust::Vect<@FAUST_SCALAR_FOR_GM@, Cpu> out_v(dim2);
+			Faust::Vect<@FAUST_SCALAR_FOR_GM@, Cpu> out_v(getNbCol());
 			dsm_funcs->mul_gpu_dsm_tocpu_ext(this->gpu_mat, gpu_vec.gpu_mat, out_v.getData(), OP_NOTRANSP, OP_NOTRANSP);
 			return out_v;
 		}
@@ -113,7 +128,7 @@ namespace Faust
 	template<>
 		Faust::MatDense<@FAUST_SCALAR_FOR_GM@, Cpu> Faust::MatDense<@FAUST_SCALAR_FOR_GM@,GPU2>::tocpu(const void* stream/*=nullptr*/) const
 		{
-			MatDense<@FAUST_SCALAR_FOR_GM@, Cpu> cpu_mat(dim1, dim2);
+			MatDense<@FAUST_SCALAR_FOR_GM@, Cpu> cpu_mat(getNbRow(), getNbCol());
 			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
 			dsm_funcs->set_stream(gpu_mat, stream);
 			dsm_funcs->tocpu(gpu_mat, cpu_mat.getData());
@@ -125,12 +140,12 @@ namespace Faust
 		void Faust::MatDense<@FAUST_SCALAR_FOR_GM@,GPU2>::resize(const faust_unsigned_int nbRow, const faust_unsigned_int nbCol)
 		{
 			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
-			dsm_funcs->resize(gpu_mat, nbRow, nbCol);
-			this->dim1 = nbRow;
-			this->dim2 = nbCol;
+			dsm_funcs->resize(gpu_mat, getNbRow(), getNbCol());
+#ifndef NDEBUG
 			int32_t new_nrows, new_ncols;
 			dsm_funcs->info(gpu_mat, &new_nrows, &new_ncols);
-			assert(nbRow == new_nrows && new_ncols == nbCol);
+			assert(getNbRow() == new_nrows && new_ncols == getNbCol());
+#endif
 		}
 
 	template<>
@@ -145,7 +160,7 @@ namespace Faust
 		{
 			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
 			dsm_funcs->setzeros(gpu_mat);
-			this->isZeros = true;
+			this->is_zeros = true;
 		}
 
 	template<>
@@ -177,7 +192,7 @@ namespace Faust
 			// use a stream (nullptr means default)
 			dsm_funcs->set_stream(this->gpu_mat, stream);
 			auto gpu_mat = dsm_funcs->clone(this->gpu_mat);
-			auto clone_mat = new Faust::MatDense<@FAUST_SCALAR_FOR_GM@,GPU2>(dim1, dim2, nullptr, /*no_alloc*/true, dev_id);
+			auto clone_mat = new Faust::MatDense<@FAUST_SCALAR_FOR_GM@,GPU2>(getNbRow(), getNbCol(), nullptr, /*no_alloc*/true, dev_id);
 			clone_mat->gpu_mat = gpu_mat;
 			// return to previous device
 			gp_funcs->set_dev(cur_dev_id);
@@ -191,9 +206,6 @@ namespace Faust
 		{
 			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
 			dsm_funcs->transpose(this->gpu_mat);
-			auto tmp = dim1;
-			dim1 = dim2;
-			dim2 = tmp;
 		}
 
 	template<>
@@ -201,9 +213,6 @@ namespace Faust
 		{
 			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
 			dsm_funcs->adjoint(this->gpu_mat);
-			auto tmp = dim1;
-			dim1 = dim2;
-			dim2 = tmp;
 		}
 
 	template<>
@@ -332,12 +341,11 @@ namespace Faust
 			// other = this * other
 			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
 			dsm_funcs->mul_gpu_dsm_ext(this->gpu_mat, A.gpu_mat, this->gpu_mat, OP_NOTRANSP, OP_NOTRANSP);
-			Faust::MatDense<@FAUST_SCALAR_FOR_GM@,GPU2> new_this(dim1, A.dim2, nullptr);
+			Faust::MatDense<@FAUST_SCALAR_FOR_GM@,GPU2> new_this(getNbRow(), A.getNbCol(), nullptr);
 			dsm_funcs->mul_gpu_dsm_ext(this->gpu_mat, A.gpu_mat, new_this.gpu_mat, OP_NOTRANSP, OP_NOTRANSP);
 			auto tmp = this->gpu_mat;
 			this->gpu_mat = new_this.gpu_mat;
 			new_this.gpu_mat = tmp;
-			this->dim2 = A.dim2;
 		}
 
 	template<>
@@ -346,12 +354,11 @@ namespace Faust
 			// this = this * other
 			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
 			MatDense<@FAUST_SCALAR_FOR_GM@,GPU2> gpuA(A.getNbRow(), A.getNbCol(), A.getData());
-			Faust::MatDense<@FAUST_SCALAR_FOR_GM@,GPU2> new_this(dim1, A.dim2, nullptr);
+			Faust::MatDense<@FAUST_SCALAR_FOR_GM@,GPU2> new_this(getNbRow(), A.getNbCol(), nullptr);
 			dsm_funcs->mul_gpu_dsm_ext(this->gpu_mat, gpuA.gpu_mat, new_this.gpu_mat, OP_NOTRANSP, OP_NOTRANSP);
 			auto tmp = this->gpu_mat;
 			this->gpu_mat = new_this.gpu_mat;
 			new_this.gpu_mat = tmp;
-			this->dim2 = A.dim2;
 		}
 
 	template<>
@@ -371,8 +378,9 @@ namespace Faust
 		{
 			auto spm_funcs = GPUModHandler::get_singleton()->spm_funcs(@FAUST_SCALAR_FOR_GM@(0));
 			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
-			if(isZeros)
+			if(is_zeros)
 			{
+			    	resize(S.getNbRow(), getNbCol());
 				setZeros();
 			}
 			else
@@ -383,7 +391,6 @@ namespace Faust
 				dsm_funcs->free(this->gpu_mat);
 				spm_funcs->free(gpu_S);
 				this->gpu_mat = gpu_out;
-				this->dim1 = S.getNbRow();
 			}
 		}
 
@@ -392,9 +399,7 @@ namespace Faust
 		{
 
 			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs(@FAUST_SCALAR_FOR_GM@(0));
-			dim1 = A.getNbRow();
-			dim2 = A.getNbCol();
-			auto gpu_mat = dsm_funcs->togpu(dim1, dim2, const_cast<@FAUST_SCALAR_FOR_GM@*>(A.getData()));
+			auto gpu_mat = dsm_funcs->togpu(getNbRow(), getNbCol(), const_cast<@FAUST_SCALAR_FOR_GM@*>(A.getData()));
 			dsm_funcs->free(this->gpu_mat);
 			this->gpu_mat = gpu_mat;
 		}
@@ -413,8 +418,6 @@ namespace Faust
 			auto gpu_mat = dsm_funcs->clone(A.gpu_mat);
 			dsm_funcs->free(this->gpu_mat);
 			this->gpu_mat = gpu_mat;
-			this->dim1 = A.dim1;
-			this->dim2 = A.dim2;
 		}
 
 	template <>
@@ -456,4 +459,6 @@ namespace Faust
 		{
 		    std::cout << this->to_string();
 		}
+
+
 };

src/faust_linear_operator/GPU2/faust_MatGeneric_gpu.h

+#ifndef __FAUST_MATGENERIC_GPU__
+#define __FAUST_MATGENERIC_GPU__
+namespace Faust
+{
+	//TODO: this class is temporary, ideally MatSparse<FPP,GPU2> and MatDense<FPP,GPU2> should extend the MatGeneric<FPP, Device> class
+	//TODO: keep this class until MatSparse<FPP,GPU2> and MatDense<FPP,GPU2> fully implement the MatGeneric<FPP, Device> methods
+	// The interest of this class is mostly to make Transform capable of storing generic matrix
+	template<typename FPP>
+		class MatGeneric<FPP, GPU2>
+		{
+			protected:
+				bool is_identity;
+				bool is_zeros;
+			public:
+				virtual int32_t getNbRow() const=0;
+				virtual int32_t getNbCol() const=0;
+				MatGeneric();
+				virtual ~MatGeneric();
+		};
+
+}
+#include "faust_MatGeneric_gpu.hpp"
+#endif

src/faust_linear_operator/GPU2/faust_MatGeneric_gpu.hpp

+namespace Faust
+{
+	template<typename FPP>
+	MatGeneric<FPP,GPU2>::~MatGeneric()
+	{
+	}
+
+	template<typename FPP>
+	MatGeneric<FPP,GPU2>::MatGeneric() : is_zeros(false), is_identity(false)
+	{
+	}
+}

src/faust_linear_operator/GPU2/faust_MatSparse_gpu.h

@@ -3,15 +3,12 @@
 #ifdef USE_GPU_MOD
 #include "faust_gpu_mod_utils.h"
 #include "faust_constant.h"
+#include "faust_MatGeneric_gpu.h"
 namespace Faust
 {
 
-	template<typename FPP,FDevice DEVICE>
-    class MatSparse;
-
-
 	template<typename FPP>
-		class MatSparse<FPP, GPU2>
+		class MatSparse<FPP, GPU2> : public MatGeneric<FPP,GPU2>
 		{
 
 			public:
@@ -66,8 +63,8 @@ namespace Faust
 				void setZeros();
 				MatSparse<FPP, GPU2>* clone(const int32_t dev_id=-1, const void* stream=nullptr);
 				void move(const int32_t dev_id=-1, const void* stream=nullptr);
-				int32_t getNbRow();
-				int32_t getNbCol();
+				int32_t getNbRow() const;
+				int32_t getNbCol() const;
 				int32_t getNonZeros();
 				int32_t getDevice() const;
 				void Display() const;
@@ -75,8 +72,6 @@ namespace Faust
 				~MatSparse();
 
 			private:
-				int32_t nbRow;
-				int32_t nbCol;
 				gm_SparseMat_t gpu_mat;
 		};
 

src/faust_linear_operator/GPU2/faust_MatSparse_gpu.hpp.in

 //TODO: move to cpp.in
 namespace Faust
 {
+	template<>
+		int32_t Faust::MatSparse<@FAUST_SCALAR_FOR_GM@, GPU2>::getNbRow() const
+		{
+			auto spm_funcs = GPUModHandler::get_singleton()->spm_funcs((@FAUST_SCALAR_FOR_GM@)(0));
+			int32_t nbRow;
+			spm_funcs->info(gpu_mat, &nbRow, nullptr, nullptr); //TODO gpu_mod: add get_nnz, get_nrows, get_ncols
+			return nbRow;
+		}
+
+	template<>
+		int32_t Faust::MatSparse<@FAUST_SCALAR_FOR_GM@, GPU2>::getNbCol() const
+		{
+			auto spm_funcs = GPUModHandler::get_singleton()->spm_funcs((@FAUST_SCALAR_FOR_GM@)(0));
+			int32_t nbCol;
+			spm_funcs->info(gpu_mat, nullptr, &nbCol, nullptr); //TODO gpu_mod: add get_nnz, get_nrows, get_ncols
+			return nbCol;
+		}
 
 
 	template<>
@@ -12,7 +29,7 @@ namespace Faust
 			const int32_t* rowptr, /*= nullptr,*/
 			const int32_t* colinds, /*= nullptr,*/
 			const int32_t dev_id,/*=-1,*/
-			const void* stream)/*=nullptr)*/ : nbRow(nbRow), nbCol(nbCol)
+			const void* stream)/*=nullptr)*/
 	{
 		if(values != nullptr)
 		{
@@ -24,7 +41,6 @@ namespace Faust
 			gp_funcs->set_dev(dev_id);
 			gpu_mat = spm_funcs->togpu_stream(nbRow, nbCol, nnz, rowptr, colinds, values, stream);
 			gp_funcs->set_dev(cur_dev_id);
-			spm_funcs->info(gpu_mat, &this->nbRow, &this->nbCol, /*nnz*/ nullptr); //normally useless (bcause dims are already initialized), but init from GPU again just in case of inconsistency that would show up in tests
 		}
 		else gpu_mat = nullptr;
 	}
@@ -62,8 +78,6 @@ namespace Faust
 	{
 			auto spm_funcs = GPUModHandler::get_singleton()->spm_funcs((@FAUST_SCALAR_FOR_GM@)(0));
 			spm_funcs->copy(mat.gpu_mat, gpu_mat);
-			nbRow = mat.nbRow;
-			nbCol = mat.nbCol;
 	}
 
 	template<>
@@ -114,8 +128,6 @@ namespace Faust
 		auto spm_funcs = GPUModHandler::get_singleton()->spm_funcs((@FAUST_SCALAR_FOR_GM@)(0));
 		spm_funcs->free(this->gpu_mat);
 		this->gpu_mat = gmat.gpu_mat;
-		this->nbRow = gmat.nbRow;
-		this->nbCol = gmat.nbCol;
 		gmat.gpu_mat = nullptr; // to avoid freeing the new gpu_mat when leaving this scope
 	}
 
@@ -171,8 +183,6 @@ namespace Faust
 	template<>
 		void Faust::MatSparse<@FAUST_SCALAR_FOR_GM@, GPU2>::resize(int32_t nnz, int32_t nrows, int32_t ncols)
 		{
-			nbRow = nrows;
-			nbCol = ncols;
 			auto spm_funcs = GPUModHandler::get_singleton()->spm_funcs((@FAUST_SCALAR_FOR_GM@)(0));
 			spm_funcs->resize(gpu_mat, nnz, nrows, ncols);
 		}
@@ -204,7 +214,7 @@ namespace Faust
 			// use a stream (nullptr means default)
 			spm_funcs->set_stream(this->gpu_mat, stream);
 			auto gpu_mat = spm_funcs->clone(this->gpu_mat);
-			auto clone_mat = new Faust::MatSparse<@FAUST_SCALAR_FOR_GM@,GPU2>(nbRow, nbCol, /*nnz*/ 0, /*values*/ nullptr, /*rowptr */ nullptr, /*colinds*/ nullptr, dev_id, stream);
+			auto clone_mat = new Faust::MatSparse<@FAUST_SCALAR_FOR_GM@,GPU2>(getNbRow(), getNbCol(), /*nnz*/ 0, /*values*/ nullptr, /*rowptr */ nullptr, /*colinds*/ nullptr, dev_id, stream);
 			clone_mat->gpu_mat = gpu_mat;
 			// return to previous device
 			gp_funcs->set_dev(cur_dev_id);
@@ -223,17 +233,6 @@ namespace Faust
 			spm_funcs->set_stream(gpu_mat, nullptr);
 		}
 
-	template<>
-		int32_t Faust::MatSparse<@FAUST_SCALAR_FOR_GM@, GPU2>::getNbRow()
-		{
-			return nbRow;
-		}
-
-	template<>
-		int32_t Faust::MatSparse<@FAUST_SCALAR_FOR_GM@, GPU2>::getNbCol()
-		{
-			return nbCol;
-		}
 
 	template<>
 		int32_t Faust::MatSparse<@FAUST_SCALAR_FOR_GM@, GPU2>::getNonZeros()

src/faust_linear_operator/faust_MatGeneric.h

@@ -216,6 +216,7 @@ namespace Faust
 			faust_unsigned_int dim2;
 			bool is_ortho;
 			bool is_identity;
+			//TODO: add is_zeros