Cell processor implementation of a MILC lattice QCD application

Abstract

We present results of the implementation of one MILC lattice QCD application-simulation with dynamical clover fermions using the hybrid-molecular dynamics R algorithm-on the Cell Broadband Engine processor. Fifty-four individual computational kernels responsible for 98.8% of the overall execution time were ported to the Cell's Synergistic Processing Elements (SPEs). The remaining application framework, including MPI-based distributed code execution, was left to the Cell's PowerPC processor. We observe that we only infrequently achieve more than 10 GFLOPS with any of the kernels, which is just over 4% of the Cell's peak performance. At the same time, many of the kernels are sustaining a bandwidth close to 20 GB/s, which is 78% of the Cell's peak. This indicates that the application performance is limited by the bandwidth between the main memory and the SPEs. In spite of this limitation, speedups of 8.7× (for 8×8×16×16 lattice) and 9.6× (for 16×16×16×16 lattice) were achieved when comparing a 3.2 GHz Cell processor to a single core of a 2.33 GHz Intel Xeon processor. When comparing the code scaled up to execute on a dual-Cell blade and a quad-core dual-chip Intel Xeon blade, the speedups are 1.5× (8×8×16×16 lattice) and 4.1× (16×16×16×16 lattice).

Parallelizing the QUDA Library for Multi-GPU Calculations in Lattice Quantum Chromodynamics

“…LQCD has also been implemented on other heterogeneous devices, primarily on the Cell Broadband Engine. Efforts in this direction have been reported in [14], [15] as part of the "QCD Parallel Computing on the Cell Broadband Engine" (QPACE) project and elsewhere [16], [17].…”

QPACE -- a QCD parallel computer based on cell processors

“…Some groups which reported on porting lattice QCD kernels or full lattice QCD applications to the Cell processor either avoided some optimization problems, e.g., by restricting themselves to the on-chip memory [13], or have limited their efforts to optimize their data layout [14]. An analysis of different optimization strategies can be found in [15].…”

Implementation of Scientific Computing Applications on the Cell Broadband Engine

“…This was clearly demonstrated in the MILC implementation when accessing lattice site data in a strided manner. Thus, with an odd stride size (e.g., 1, 3, 5), the maximum bandwidth of 25.38 GB/s can be achieved, whereas with the stride size of 16, only 2.13 GB/s is achievable because only one out of 16 memory banks is used all the time [20].…”

“…The ERI kernel is an example of a problem whose computational complexity is O(N 4 ). Some of the results described in this paper have been presented in conference papers [19,20] while the quantum chemistry code results are new.…”