List of Figures

• ScaLAPACK software hierarchy
• Partitioning of global matrix A (s=19;c=3;a=1;l=12;p=16;k=11)
• Mapping of matrix A onto process grid (,). For example, note that process (0,0) contains a local array of size A(5,4).
• Mapping of matrix B onto process grid (,)
• Eight processes mapped to a process grid
• Gaussian elimination using Level 3 BLAS
• The one-dimensional block and cyclic column distributions
• The one-dimensional block-cyclic column- and the two-dimensional block-cyclic distributions
• A matrix decomposed into blocks mapped onto a process grid
• A matrix decomposed into blocks mapped onto a process grid
• Global view of the matrix operands
• The one-dimensional block-column and block-row distributions
• Mapping of local arrays for nonsymmetric band matrix A (no pivoting)
• Mapping of local arrays for nonsymmetric band matrix A (partial pivoting)
• Mapping of local arrays for symmetric positive definite band matrix A (UPLO='L')
• Mapping of local arrays for symmetric positive definite band matrix A (UPLO='U')
• Mapping of local arrays for nonsymmetric tridiagonal matrix A
• Mapping of local arrays for symmetric positive definite tridiagonal matrix A (UPLO='L')
• Mapping of local arrays for symmetric positive definite tridiagonal matrix A (UPLO='U')
• LU Performance per Intel XP/S MP Paragon node