xGEMM (General Matrix-Matrix Product) VI

Owning Palette: LVCUBLAS

Requires: GPU Analysis Toolkit

Calculates the rectangular, or general, matrix expression alpha*op(A)*op(B) + bC for matrices A, B (if applicable), and C, for scalars alpha and beta, and for matrix operation op(X). When you wire data to A in, B in (if available), and C, this VI automatically selects the first available instance.

To use a different instance, you must manually select the polymorphic instance you want to use.

Details  

xGEMM (General Matrix-Matrix Product) (aAA + bC)

The connector pane displays the default data types for this polymorphic instance.

	op(B) specifies the operation the VI performs on matrix B, where matrix op(B) can equal B, B' or conj(B'). 0 CUBLAS_OP_N (default)—Specifies B. 1 CUBLAS_OP_T—Specifies B'. 2 CUBLAS_OP_C—Specifies conj(B').
	op(A) specifies the operation the VI performs on matrix A, where matrix op(A) can equal A, A' or conj(A'). 0 CUBLAS_OP_N (default)—Specifies A. 1 CUBLAS_OP_T—Specifies A'. 2 CUBLAS_OP_C—Specifies conj(A').
	CUBLAS Handle in specifies the initialized CUBLAS library to use for the BLAS calculation. For example, you can wire the CUBLAS Handle output from the Initialize Library VI to specify the CUBLAS handle to the CUBLAS library you already initialized. This input also determines the device that executes the function.
	C specifies the matrix C stored on the device. This input specifies a class that can contain the following data types: Single-precision, floating-point numeric Double-precision, floating-point numeric Complex single-precision, floating-point numeric Complex double-precision, floating-point numeric
	A in specifies the square matrix A stored on the device. This input specifies a class that can contain the following data types: Single-precision, floating-point numeric Double-precision, floating-point numeric Complex single-precision, floating-point numeric Complex double-precision, floating-point numeric
	a represents alpha and specifies the scalar operand in the product alpha*op(A)*op(B). The default is 1.
	b represents beta and specifies the scalar operand in the product beta*C. The default is 0.
	error in describes error conditions that occur before this node runs. This input provides standard error in functionality.
	m specifies the number of rows to use in matrix op(A) and C.
	n specifies the number of columns to use in matrix op(B) and C.
	k specifies the number of columns to use in matrix op(A) and the number of rows to use in matrix op(B).
	leading dimensions specifies the column dimension to index consecutive rows. Use lda, ldb, and ldc for A, A, and C, respectively.
	CUBLAS Handle out returns the handle that defines the BLAS operation.
	aAA+bC returns a matrix with the same dimensions as C. For the elements of the first m rows and m columns, the VI returns the result of alpha*op(A)op(B) + bC. For any remaining elements, the VI returns the value of the element with the same index in C.
	A out returns the square matrix A stored on the device.
	error out contains error information. This output provides standard error out functionality.

xGEMM (General Matrix-Matrix Product) (aAB + bC)

The connector pane displays the default data types for this polymorphic instance.

	op(B) specifies the operation the VI performs on matrix B, where matrix op(B) can equal B, B' or conj(B'). 0 CUBLAS_OP_N (default)—Specifies B. 1 CUBLAS_OP_T—Specifies B'. 2 CUBLAS_OP_C—Specifies conj(B').
	op(A) specifies the operation the VI performs on matrix A, where matrix op(A) can equal A, A' or conj(A'). 0 CUBLAS_OP_N (default)—Specifies A. 1 CUBLAS_OP_T—Specifies A'. 2 CUBLAS_OP_C—Specifies conj(A').
	CUBLAS Handle in specifies the initialized CUBLAS library to use for the BLAS calculation. For example, you can wire the CUBLAS Handle output from the Initialize Library VI to specify the CUBLAS handle to the CUBLAS library you already initialized. This input also determines the device that executes the function.
	C specifies the matrix C stored on the device. This input specifies a class that can contain the following data types: Single-precision, floating-point numeric Double-precision, floating-point numeric Complex single-precision, floating-point numeric Complex double-precision, floating-point numeric
	A in specifies the matrix A stored on the device. This input specifies a class that can contain the following data types: Single-precision, floating-point numeric Double-precision, floating-point numeric Complex single-precision, floating-point numeric Complex double-precision, floating-point numeric
	B in specifies the matrix B stored on the device. This input specifies a class that can contain the following data types: Single-precision, floating-point numeric Double-precision, floating-point numeric Complex single-precision, floating-point numeric Complex double-precision, floating-point numeric
	a represents alpha and specifies the scalar operand in the product alpha*op(A)*op(B). The default is 1.
	b represents beta and specifies the scalar operand in the product beta*C. The default is 0.
	error in describes error conditions that occur before this node runs. This input provides standard error in functionality.
	m specifies the number of rows to use in matrix op(A) and C.
	n specifies the number of columns to use in matrix op(B) and C.
	k specifies the number of columns to use in matrix op(A) and the number of rows to use in matrix op(B).
	leading dimensions specifies the column dimension to index consecutive rows. Use lda, ldb, and ldc for A, B, and C, respectively.
	CUBLAS Handle out returns the handle that defines the BLAS operation.
	aAB+bC returns a matrix with the same dimensions as C. For the elements of the first m rows and n columns of the matrix you select, the VI replaces the result of alpha*op(A)op(B) + bC. For any remaining elements, the VI returns the value of the element with the same index in C.
	A out returns the square matrix A stored on the device.
	B out returns the matrix B stored on the device.
	error out contains error information. This output provides standard error out functionality.

xGEMM (General Matrix-Matrix Product) Details

For more information on how to use this VI, refer to the Designing the Block Diagram to Compute on a GPU Device topic.

For more information about the CUBLAS library and BLAS operations, refer to the NVIDIA GPU Computing Documentation website at nvidia.com and download the CUBLAS Library User Guide.

Refer to the BLAS (Basic Linear Algebra Subprograms) website at netlib.org for more information on BLAS functions.