next up previous
Next: Deriving a Block Algorithm Up: No Title Previous: The BLAS as the

Block Algorithms and Their Derivation

  It is comparatively straightforward to recode many of the algorithms in LINPACK and EISPACK so that they call Level 2 BLAS. Indeed, in the simplest cases the same floating-point operations are done, possibly even in the same order: it is just a matter of reorganizing the software. To illustrate this point, we consider the Cholesky factorization algorithm used in the LINPACK routine SPOFA, which factorizes a symmetric positive definite matrix as tex2html_wrap_inline1761. We consider Cholesky factorization because the algorithm is simple, and no pivoting is required. In Section 4 we shall consider the slightly more complicated example of LU factorization.

Suppose that after j-1 steps the block tex2html_wrap_inline1765 in the upper lefthand corner of A has been factored as tex2html_wrap_inline1769. The next row and column of the factorization can then be computed by writing tex2html_wrap_inline1761 as


displaymath1773
where tex2html_wrap_inline1775, tex2html_wrap_inline1777, tex2html_wrap_inline1779, and tex2html_wrap_inline1781 are column vectors of length j-1, and tex2html_wrap_inline1785 and tex2html_wrap_inline1787 are scalars. Equating coefficients of the tex2html_wrap_inline1789 column, we obtain
eqnarray141
Since tex2html_wrap_inline1791 has already been computed, we can compute tex2html_wrap_inline1779 and tex2html_wrap_inline1787 from the equations
eqnarray149

The body of the code of the LINPACK routine SPOFA that implements the above method is shown in Figure 1. The same computation recoded in ``LAPACK-style'' to use the Level 2 BLAS routine STRSV (which solves a triangular system of equations) is shown in Figure 2. The call to STRSV has replaced the loop over K which made several calls to the Level 1 BLAS routine SDOT. (For reasons given below, this is not the actual code used in LAPACK -- hence the term ``LAPACK-style''.)

This change by itself is sufficient to result in large gains in performance on a number of machines--for example, from 72 to 251 megaflops for a matrix of order 500 on one processor of a CRAY Y-MP. Since this is 81% of the peak speed of matrix-matrix multiplication on this processor, we cannot hope to do very much better by using Level 3 BLAS.

We can, however, restructure the algorithm at a deeper level to exploit the faster speed of the Level 3 BLAS. This restructuring involves recasting the algorithm as a block algorithm--that is, an algorithm that operates on blocks or submatrices of the original matrix.




next up previous
Next: Deriving a Block Algorithm Up: No Title Previous: The BLAS as the

Jack Dongarra
Sun Feb 9 10:05:05 EST 1997