Conveners
Algorithms and Machines: Parallel 1F
- Mike Clark (NVIDIA)
Algorithms and Machines: Parallel 2F
- James Osborn (ANL)
Algorithms and Machines: Parallel 8F
- Steve Gottlieb (Indiana University)
Dr
Ran Zhou
(Fermilab)
23/06/2014, 14:15
Algorithms and Machines
Talk
Electromagnetic effects play an important role in many
phenomena such as isospin symmetry breaking in the
hadron spectrum, hadronic contributions to g-2, etc.
We have generalized the MILC QCD code to include the
electromagnetic field. In this work, we focus on
simulations including charged sea quarks using the RHMC algorithm.
We explain the algorithm, code performance, and results...
Dr
Meifeng Lin
(Brookhaven National Laboratory)
23/06/2014, 14:35
Algorithms and Machines
Talk
Lattice QCD has entered the era of simulations directly at the physical quark masses thanks to the increasing computing power and algorithmic advancement. But the computational cost for such simulations is still extremely high, both for the gauge field generation and the measurements of physical observables. Adaptive multi grid (MG) algorithm has proven to be quite effective in reducing the...
Mr
Greg McGlynn
(Columbia University)
23/06/2014, 14:55
Algorithms and Machines
Talk
We determine the scaling behavior of the autocorrelation times of observables constructed from the topological charge density on lattices with periodic and open boundary conditions using a series of high-statistics numerical simulations. The autocorrelation functions of such observables turn out to obey a simple differential equation which allows the motion of topological charge in hybrid...
Mr
Matthias Rottmann
(University of Wuppertal)
23/06/2014, 15:15
Algorithms and Machines
Talk
The overlap lattice discretization of the Dirac operator preserves the important physical property of chiral symmetry, at the expense of requiring much more effort when solving systems with this operator. We present a preconditioning technique based on another lattice discretization, the Wilson-Dirac operator. Recently, close to normal operators have gained popularity in numerical simulations....
James Osborn
(ANL)
23/06/2014, 15:35
Algorithms and Machines
Talk
I will give an introduction to the FUEL project for lattice field
theory code. The code being developed (called "qhmc") was initially
targeted for gauge field generation for beyond standard model
theories, and is now growing into a more general framework suitable
for analysis too. The design is based on using the Lua scripting
language as a wrapper for existing lattice field theory...
Dr
Piotr Korcyl
(DESY Zeuthen)
23/06/2014, 16:30
Algorithms and Machines
Talk
We report on the status of large volume simulations with 2+1 dynamical quarks which are being performed by the CLS collaboration. The algorithmic details include: open boundary conditions, twisted mass reweighting and the rational approximation for the strange quark, whereas the main feature of the simulation strategy is the approach to the physical point along a trajectory of constant trace...
Dr
M Clark
(NVIDIA)
23/06/2014, 16:50
Algorithms and Machines
Talk
There has been much progress to date in developing scalable sparse linear solver algorithms on GPUs, utilizing well-known mathematical methods such as mixed precision, domain decomposition and pipelining to improve performance, allowing efficient use of large GPU installations such as Blue Waters and Titan. However, there has been less focus on deploying 'mathematically optimal' linear...
Dr
Alexei Strelchenko
(Fermilab)
23/06/2014, 17:10
Algorithms and Machines
Talk
Realization of the Incremental eigCG algorithm for solving Hermitian linear systems with multiple right hand sides on the CUDA platform is reported. In particular, we present details of implementation of the algorithm in the QUDA library. Preliminary performance results are provided.
Dr
Justin Foley
(Microway and Nvidia)
23/06/2014, 17:30
Algorithms and Machines
Talk
We describe recent additions to the QUDA software library that are
aimed at extending strong scaling in multi-GPU lattice calculations.
These include the addition of CPU-thread support in order to increase
concurrency and improve the overlap of computation and communication
in Krylov solver routines, as well as the modifications needed to enable the GPUDirect RDMA feature recently...
Dr
Nuno Cardoso
(NCSA, University of Illinois)
23/06/2014, 17:50
Algorithms and Machines
Talk
In this work, we present the GPU implementation of the overrelaxation and steepest descent method with Fourier acceleration methods for Laudau and Coulomb gauge fixing using CUDA for SU(N) with N>2.
A multi-GPU implementation of the overrelaxation method is also presented using MPI and CUDA.
The GPU performance was measured on BlueWaters and compared against the gauge fixing of the CPU MILC code.
Ms
Ruizi Li
(Indiana University)
23/06/2014, 18:10
Algorithms and Machines
Talk
The conjugate gradient algorithm is among the most essential and time consuming parts of lattice calculations with staggered quarks. We test the performance of dslash, the key step in the CG algorithm, on the Intel Xeon Phi, also known as the many integrated cores (MIC) architecture, with different parallelization strategies using MPI,OpenMP, and the vector processing units (VPUs).
Dr
Francesco Di Renzo
(University of Parma and INFN)
27/06/2014, 14:15
Algorithms and Machines
Talk
In the context of thimble regularization of lattice field theories we are developing a new simulation algorithm. The main difficulty is to devise a sampling of configurations on a non-trivial manifold, which is defined as the hypersurface formed by the union of all paths of steepest descent of the complex action ending in a given saddle point. The main point with the new algorithm is the...
Mr
Matthias Puhr
(University Regensburg)
27/06/2014, 14:35
Algorithms and Machines
Talk
The overlap Dirac operator at finite chemical potential can be efficiently computed with the two-sided Lanczos algorithm. Compared with other methods the Lanczos algorithm has the advantage that it takes into account information about the source vector when constructing an approximation to the overlap operator.
In order to study transport coefficients in terms of correlators of conserved...
Dr
Renwick Hudspith
(York University)
27/06/2014, 14:55
Algorithms and Machines
Talk
We provide details of our implementation of a non-linear conjugate gradient method for Landau and Coulomb gauge fixing with Fourier acceleration. We find clear improvement over the Fourier accelerated steepest descent method, with the average time taken for our algorithm to converge to a fixed, high accuracy, being lower by a factor of 2 to 4 for several lattice volumes.
Dr
Keitaro Nagata
(KEK)
27/06/2014, 15:15
Algorithms and Machines
Talk
The Wilson fermion determinant can be written in the form of a series
expansion in fugacity z=exp(mu/T), provided that the eigenmodes of the
temporally reduced operator are obtained. Since the calculation of
all eigenmodes rapidly becomes prohibitive for larger volumes, we
develop a method to calculate only the low-energy eigenmodes of the
reduced matrix using a matrix filetering...
Mr
Matthew Spraggs
(University of Southampton)
27/06/2014, 15:35
Algorithms and Machines
Talk
I demonstrate pyQCD, a native Python library providing an extensible API for single-node lattice measurements and simulations. Boost.Python is used to wrap the underlying C++ code and expose an interface to Python for the generation of propagators and configurations, both of which are returned as numpy ndarray types. The library can take advantage of GPU technology by using CUDA where possible...
