next up previous contents
Next: Symmetric Matrix-Vector Multiplication Up: General Matrix-Vector Multiplication Previous: PLAPACK FORTRAN-C interface

PLAPACK calls

PLAPACK provides two more layers of level-2 BLAS: the local level-2 BLAS and the global level-2 BLAS, with identical parameter sequences. For the general matrix-vector multiply, the calling sequences are given by place HR here

figure6944

figure6948

figure6951

place HR here Parameter TRANS has been replaced by integer trans, which can take on the values indicated in Table gif. All other parameters are now passed as part of linear algebra objects alpha, a, x, beta, and y.

PLACE BEGIN HR HERE [tbp]

tabular6846

PLACE BEGIN HR HERE

Valid values for integer parameters uplo, trans, and diag.

PLACE END HR HERE

Local call:

When a node calls PLA_Local_gemv, the operation is equivalent to extracting the local information (datatype, dimensions, and pointer to the buffer) and locally performing the general matrix-vector multiply. The local dimensions of alpha and beta must be tex2html_wrap_inline14121 , but they can be of any object type. The local dimensions of a, x, and y must be such that the operation is well-defined. All objects must have the same datatype. To best understand the operation of this call, see the sample implementation given in Figure gif.

PLACE BEGIN HR HERE

figure7807

PLACE END HR HERE

Notice: no communication is performed as part of this call. Thus, all information must be locally available.

Global call:

All nodes must simultaneously call PLA_Gemv, which performs the BLAS call on the distributed objects. Parameter trans now indicates is the global matrix A is to be (conjugate) transposed for the operation. Parameters alpha and beta must be tex2html_wrap_inline14125 multiscalars. Parameter a must be of object type matrix. Parameters x and y must all be of object type vector or (duplicated) projected vector. The global dimensions of the different objects must be such that the operation is well-defined.


next up previous contents
Next: Symmetric Matrix-Vector Multiplication Up: General Matrix-Vector Multiplication Previous: PLAPACK FORTRAN-C interface

rvdg@cs.utexas.edu