Implement MatBSR<FPP, GPU2>::bsrgemm and update to gpu_mod@a1dd1ade.

gpu_mod @ a1dd1ade

-Subproject commit 0f2c75e43083e6bacc0fae9e1ce5ce23873b75ed
+Subproject commit a1dd1adebe85d108efe6576f35192fe39f81d885

src/faust_linear_operator/GPU2/faust_MatBSR_gpu.cpp.in

@@ -112,11 +112,24 @@ namespace Faust
 			char2gm_Op(opThis, gop_this);
 			auto bsr_funcs = GPUModHandler::get_singleton()->bsr_funcs((@FAUST_SCALAR_FOR_GM@)(0));
 			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs((@FAUST_SCALAR_FOR_GM@)(0));
-			auto gpu_dmat = bsr_funcs->mul_gpu_dsm_ext(gpu_mat, M.gpu_mat, nullptr, gop_this, OP_NOTRANSP);
+			auto gpu_dmat = bsr_funcs->mul_gpu_dsm_ext(gpu_mat, M.gpu_mat, nullptr, gop_this, OP_NOTRANSP, /*alpha*/ nullptr, /*beta*/ nullptr);
 			dsm_funcs->free(M.get_gpu_mat_ptr());
 			M.gpu_mat = gpu_dmat;
 		}
 
+	template<>
+		void MatBSR<@FAUST_SCALAR_FOR_GM@,GPU2>::bsrgemm(const MatBSR<@FAUST_SCALAR_FOR_GM@,GPU2>& A, const MatDense<@FAUST_SCALAR_FOR_GM@,GPU2>& B, MatDense<@FAUST_SCALAR_FOR_GM@,GPU2>& C, const @FAUST_SCALAR_FOR_GM@& alpha, const @FAUST_SCALAR_FOR_GM@& beta, const char opA/*='N'*/, const char opB /*= 'N'*/)
+		{
+			gm_Op gop_A;
+			gm_Op gop_B;
+			char2gm_Op(opA, gop_A);
+			char2gm_Op(opB, gop_B);
+			auto bsr_funcs = GPUModHandler::get_singleton()->bsr_funcs((@FAUST_SCALAR_FOR_GM@)(0));
+			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs((@FAUST_SCALAR_FOR_GM@)(0));
+			auto gpu_dmat = bsr_funcs->mul_gpu_dsm_ext(A.gpu_mat, B.gpu_mat, C.gpu_mat, gop_A, gop_B, (const @GM_SCALAR@*) reinterpret_cast<const @GM_REINTERPRET_CAST_SCALAR@*>(&alpha), (const @GM_SCALAR@*) reinterpret_cast<const @GM_REINTERPRET_CAST_SCALAR@*>(&beta));
+			dsm_funcs->free(C.get_gpu_mat_ptr());
+		}
+
 	template<>
 		void MatBSR<@FAUST_SCALAR_FOR_GM@,GPU2>::multiply(Vect<@FAUST_SCALAR_FOR_GM@,GPU2>& vec, char opThis/*='N'*/) const
 		{
@@ -125,7 +138,7 @@ namespace Faust
 			auto bsr_funcs = GPUModHandler::get_singleton()->bsr_funcs((@FAUST_SCALAR_FOR_GM@)(0));
 			auto dsm_funcs = GPUModHandler::get_singleton()->dsm_funcs((@FAUST_SCALAR_FOR_GM@)(0));
 			// the backend code is specialized for vector (the number of columns is tested to call another function, see gpu_mod)
-			auto gpu_dmat = bsr_funcs->mul_gpu_dsm_ext(gpu_mat, vec.gpu_mat, nullptr, gop_this, OP_NOTRANSP);
+			auto gpu_dmat = bsr_funcs->mul_gpu_dsm_ext(gpu_mat, vec.gpu_mat, nullptr, gop_this, OP_NOTRANSP, nullptr, nullptr);
 			dsm_funcs->free(vec.get_gpu_mat_ptr());
 			vec.gpu_mat = gpu_dmat;
 		}

src/faust_linear_operator/GPU2/faust_MatBSR_gpu.h

@@ -79,6 +79,7 @@ namespace Faust
 			/*********** own member functions **************/
 			void multiply(MatDense<FPP,GPU2>& M, char opThis='N') const;
 			void multiply(Vect<FPP,GPU2>& vec, char opThis='N') const;
+			static void bsrgemm(const MatBSR<FPP, GPU2>& A, const MatDense<FPP,GPU2>& B, MatDense<FPP,GPU2>& C, const FPP& alpha, const FPP& beta, const char opThis/*='N'*/, const char opB /*= 'N'*/);
 
 			void tocpu(int32_t* browptr, int32_t* bcolinds, FPP* bdata, int32_t* nrows=nullptr, int32_t* ncols=nullptr, int32_t *bnrows=nullptr, int32_t *bncol=nullptr, int32_t* bnnz=nullptr) const;
 			void tocpu(MatBSR<FPP, Cpu> &cpu_mat) const;