blas_node.h File Reference

BLAS (Basic Linear Algebra Subprograms) library node definitions. More...

#include <optional>
#include "sdfg/codegen/code_snippet_factory.h"
#include "sdfg/data_flow/library_nodes/math/math_node.h"

Go to the source code of this file.

Classes
class	sdfg::math::blas::BLASNode
	Base class for BLAS operation nodes. More...
class	sdfg::math::blas::BLASLibDependency

Namespaces
namespace	sdfg
namespace	sdfg::math
namespace	sdfg::math::blas

Enumerations
enum	sdfg::math::blas::BLAS_Precision {   sdfg::math::blas::invalid = 0 , sdfg::math::blas::h = 'h' , sdfg::math::blas::s = 's' , sdfg::math::blas::d = 'd' ,   sdfg::math::blas::c = 'c' , sdfg::math::blas::z = 'z' }
	Precision/data type for BLAS operations. More...
enum	sdfg::math::blas::BLAS_Transpose { sdfg::math::blas::No = 111 , sdfg::math::blas::Trans = 112 , sdfg::math::blas::ConjTrans = 113 }
	Transpose operation for BLAS matrices. More...
enum	sdfg::math::blas::BLAS_Layout { sdfg::math::blas::RowMajor = 101 , sdfg::math::blas::ColMajor = 102 }
	Memory layout for BLAS matrices. More...

Functions
constexpr std::string_view	sdfg::math::blas::BLAS_Precision_to_string (BLAS_Precision precision)
constexpr std::string_view	sdfg::math::blas::BLAS_Transpose_to_string (BLAS_Transpose transpose)
constexpr char	sdfg::math::blas::BLAS_Transpose_to_char (BLAS_Transpose transpose)
constexpr std::string_view	sdfg::math::blas::BLAS_Layout_to_string (BLAS_Layout layout)
constexpr std::string_view	sdfg::math::blas::BLAS_Layout_to_short_string (BLAS_Layout layout)

Variables
data_flow::ImplementationType	sdfg::math::blas::ImplementationType_BLAS {"BLAS"}
	BLAS implementation type Uses standard CPU BLAS libraries (OpenBLAS, Intel MKL, etc.).

Detailed Description

BLAS (Basic Linear Algebra Subprograms) library node definitions.

This file defines base classes and enumerations for BLAS operations. BLAS nodes are mathematical library nodes that can be dispatched to various BLAS implementations (CPU BLAS, CUBLAS, etc.) based on their implementation type.

BLAS Implementation Types

BLAS nodes support multiple implementation types:

• ImplementationType_BLAS: Standard CPU BLAS (e.g., OpenBLAS, MKL)
• ImplementationType_CUDAWithTransfers: CUDA with automatic data transfers
• ImplementationType_CUDAWithoutTransfers: CUDA assuming data is on GPU

The implementation type is used by dispatchers to generate appropriate library calls.

BLAS Operations

Common BLAS operations include:

• Level 1: Vector operations (dot product)
• Level 2: Matrix-vector operations
• Level 3: Matrix-matrix operations (GEMM)

See also

math::blas::GemmNode for matrix multiplication

math::blas::DotNode for dot product

math::MathNode for expansion interface