[PDF][PDF] Fracture Mechanics on the Intel R ItaniumTM Architecture (A Case Study)
G Heber, AJ Dolgert, M Alt, KA Mazurkiewicz, L Stringer - 2001 - researchgate.net
G Heber, AJ Dolgert, M Alt, KA Mazurkiewicz, L Stringer
2001•researchgate.netWe optimized our fracture mechanics code to achieve a 3.4-fold speedup on Itanium
processors. The computational core of the code is a linear equation solver using
preconditioned conjugate-gradient method. The dense and sparse matrix-vector
computations are both floating point and memory bandwidth intensive. We found that the
EPIC architecture depends heavily on the compiler to find instruction level parallelism to
achieve maximum performance.
processors. The computational core of the code is a linear equation solver using
preconditioned conjugate-gradient method. The dense and sparse matrix-vector
computations are both floating point and memory bandwidth intensive. We found that the
EPIC architecture depends heavily on the compiler to find instruction level parallelism to
achieve maximum performance.
Abstract
We optimized our fracture mechanics code to achieve a 3.4-fold speedup on Itanium processors. The computational core of the code is a linear equation solver using preconditioned conjugate-gradient method. The dense and sparse matrix-vector computations are both floating point and memory bandwidth intensive. We found that the EPIC architecture depends heavily on the compiler to find instruction level parallelism to achieve maximum performance.
researchgate.net