The Experts below are selected from a list of 10986 Experts worldwide ranked by ideXlab platform
Olivier Tardieu - One of the best experts on this subject based on the ideXlab platform.
-
work stealing without the baggage
Conference on Object-Oriented Programming Systems Languages and Applications, 2012Co-Authors: Vivek Kumar, Stephen M Blackburn, Daniel Frampton, David Grove, Olivier TardieuAbstract:Work-stealing is a promising approach for effectively exploiting software Parallelism on parallel hardware. A programmer who uses work-stealing explicitly identifies Potential Parallelism and the runtime then schedules work, keeping otherwise idle hardware busy while relieving overloaded hardware of its burden. Prior work has demonstrated that work-stealing is very effective in practice. However, work-stealing comes with a substantial overhead: as much as 2x to 12x slowdown over orthodox sequential code.In this paper we identify the key sources of overhead in work-stealing schedulers and present two significant refinements to their implementation. We evaluate our work-stealing designs using a range of benchmarks, four different work-stealing implementations, including the popular fork-join framework, and a range of architectures. On these benchmarks, compared to orthodox sequential Java, our fastest design has an overhead of just 15%. By contrast, fork-join has a 2.3x overhead and the previous implementation of the system we use has an overhead of 4.1x. These results and our insight into the sources of overhead for work-stealing implementations give further hope to an already promising technique for exploiting increasingly available hardware Parallelism.
Yousry Y. Azmy - One of the best experts on this subject based on the ideXlab platform.
-
parametric study of parallel block jacobi source iteration hybrid methods in 2 d cartesian geometry and construction of the integral transport matrix method matrices via green s functions
Epj Web of Conferences, 2021Co-Authors: Dylan S. Hoagland, Yousry Y. AzmyAbstract:Parallel Block Jacobi (PBJ ) [1] is an asynchronous spatial domain decomposition with application in solving the neutron transport equation due to its extendibility to massively parallel solution in unstructured spatial meshes (grids) without the use of the computationally complex and expensive sweeps required by the Source Iteration (SI ) method in these applications. [2] However, PBJ iterative methods suffer a lack of iterative robustness in problems with optically thin cells, [1] which we have previously demonstrated to be a consequence of PBJ ’s asynchronicity. To mitigate this effect, we have developed multiple PBJ / SI hybrid methods which employ a PBJ method (Parallel Block Jacobi - Integral Transport Matrix Method (PBJ-ITMM ) or Inexact Parallel Block Jacobi (IPBJ )) along with SI . [3,4] In this work, we perform a parametric study to determine performance of numerous PBJ / SI hybrid methods as a function of multiple problem parameters. This parametric study reached 5 main conclusions: 1) our hybrid approach is more effective with PBJ-ITMM than with IPBJ , 2) for PBJ-ITMM , there is a hybrid method that mitigates the aforementioned iterative slowdown in optically thin cells without diminishing the method’s Potential Parallelism in unstructured grids, 3) this hybrid method is most effective in problems with large, continuous regions of very thin cells, 4) the best performing hybrid method consistently executes within a factor of ten slower than current state-of-the-art acceleration methods that are not efficiently extendable to the massively parallel regime, and 5) both PBJ -ITMM and IPBJ are observed to be viable approaches for our desired applications. In the pursuit of implementing PBJ-ITMM in unstructured grids, we conclude with a description of the Green’s Function ITMM Construction (GFIC ) algorithm, which allows for the ITMM matrices to be constructed using the pre-existing SI sweep algorithm already present in unstructured grid SN transport codes.
-
PARAMETRIC STUDY OF PARALLEL BLOCK JACOBI / SOURCE ITERATION HYBRID METHODS IN 2-D CARTESIAN GEOMETRY AND CONSTRUCTION OF THE INTEGRAL TRANSPORT MATRIX METHOD MATRICES VIA GREEN’S FUNCTIONS
'EDP Sciences', 2021Co-Authors: Dylan S. Hoagland, Yousry Y. AzmyAbstract:Parallel Block Jacobi (PBJ) [1] is an asynchronous spatial domain decomposition with application in solving the neutron transport equation due to its extendibility to massively parallel solution in unstructured spatial meshes (grids) without the use of the computationally complex and expensive sweeps required by the Source Iteration (SI) method in these applications. [2] However, PBJ iterative methods suffer a lack of iterative robustness in problems with optically thin cells, [1] which we have previously demonstrated to be a consequence of PBJ’s asynchronicity. To mitigate this effect, we have developed multiple PBJ / SI hybrid methods which employ a PBJ method (Parallel Block Jacobi - Integral Transport Matrix Method (PBJ-ITMM) or Inexact Parallel Block Jacobi (IPBJ)) along with SI. [3,4] In this work, we perform a parametric study to determine performance of numerous PBJ / SI hybrid methods as a function of multiple problem parameters. This parametric study reached 5 main conclusions: 1) our hybrid approach is more effective with PBJ-ITMM than with IPBJ, 2) for PBJ-ITMM, there is a hybrid method that mitigates the aforementioned iterative slowdown in optically thin cells without diminishing the method’s Potential Parallelism in unstructured grids, 3) this hybrid method is most effective in problems with large, continuous regions of very thin cells, 4) the best performing hybrid method consistently executes within a factor of ten slower than current state-of-the-art acceleration methods that are not efficiently extendable to the massively parallel regime, and 5) both PBJ-ITMM and IPBJ are observed to be viable approaches for our desired applications. In the pursuit of implementing PBJ-ITMM in unstructured grids, we conclude with a description of the Green’s Function ITMM Construction (GFIC) algorithm, which allows for the ITMM matrices to be constructed using the pre-existing SI sweep algorithm already present in unstructured grid SN transport codes
Pierre Boulet - One of the best experts on this subject based on the ideXlab platform.
-
array ol with delays a domain specific specification language for multidimensional intensive signal processing
Multidimensional Systems and Signal Processing, 2010Co-Authors: Calin Glitia, Philippe Dumont, Pierre BouletAbstract:Intensive signal processing applications appear in many application domains such as video processing or detection systems. These applications handle multidimensional data structures (mainly arrays) to deal with the various dimensions of the data (space, time, frequency). A specification language allowing the direct manipulation of these different dimensions with a high level of abstraction is a key to handling the complexity of these applications and to benefit from their massive Potential Parallelism. The Array-OL specification language is designed to do just that. We introduce here an extension of Array-OL to deal with states or delays by the way of uniform inter-repetition dependences. We show that this specification language is able to express the main patterns of computation of the intensive signal processing domain.
-
Projection of the Array-OL Specification Language onto the Kahn Process Network Computation Model
2005Co-Authors: Abdelkader Amar, Pierre Boulet, Pierre DumontAbstract:The Array-OL specification model has been introduced to model systematic signal processing applications. This model is multidimensional and allows to express the full Potential Parallelism of an application: both task and data Parallelism. The Array-OL language is an expression of data-dependences and thus allows many execution orders. In order to execute Array-OL applications on distributed architectures, we show here how to project such specification onto the Kahn process network model of computation. We show how Array-OL code transformations allow to choose a projection adapted to the target architecture. An experiment on a distributed process network implementation based on CORBA concludes this article.
-
Projection of the Array-OL Specification Language onto the Kahn Process Network Computation Model
2005Co-Authors: Abdelkader Amar, Pierre Boulet, Philippe DumontAbstract:The Array-OL specification model has been introduced to model systematic signal processing applications. This model is multidimensional and allows to express the full Potential Parallelism of an application: both task and data Parallelism. The Array-OL language is an expression of data dependences and thus allows many execution orders. par In order to execute Array-OL applications on distributed architectures, we show here how to project such specification onto the Kahn process network model of computation. We show how Array-OL code transformations allow to choose a projection adapted to the target architecture.
-
Projection of the . . . Process Network Computation Model
2005Co-Authors: Abdelkader Amar, Pierre Boulet, Philippe DumontAbstract:The Array-OL specification model has been introduced to model systematic signal processing applications. This model is multidimensional and allows to express the full Potential Parallelism of an application: both task and data Parallelism. The Array-OL language is an expression of data-dependences and thus allows many execution orders. In order to execute Array-OL applications on distributed architectures, we show here how to project such specification onto the Kahn process network model of computation. We show how Array-OL code transformations allow to choose a projection adapted to the target architecture. An experiment on a distributed process network implementation based on CORBA concludes this article
Philippe Dumont - One of the best experts on this subject based on the ideXlab platform.
-
array ol with delays a domain specific specification language for multidimensional intensive signal processing
Multidimensional Systems and Signal Processing, 2010Co-Authors: Calin Glitia, Philippe Dumont, Pierre BouletAbstract:Intensive signal processing applications appear in many application domains such as video processing or detection systems. These applications handle multidimensional data structures (mainly arrays) to deal with the various dimensions of the data (space, time, frequency). A specification language allowing the direct manipulation of these different dimensions with a high level of abstraction is a key to handling the complexity of these applications and to benefit from their massive Potential Parallelism. The Array-OL specification language is designed to do just that. We introduce here an extension of Array-OL to deal with states or delays by the way of uniform inter-repetition dependences. We show that this specification language is able to express the main patterns of computation of the intensive signal processing domain.
-
Projection of the Array-OL Specification Language onto the Kahn Process Network Computation Model
2005Co-Authors: Abdelkader Amar, Pierre Boulet, Philippe DumontAbstract:The Array-OL specification model has been introduced to model systematic signal processing applications. This model is multidimensional and allows to express the full Potential Parallelism of an application: both task and data Parallelism. The Array-OL language is an expression of data dependences and thus allows many execution orders. par In order to execute Array-OL applications on distributed architectures, we show here how to project such specification onto the Kahn process network model of computation. We show how Array-OL code transformations allow to choose a projection adapted to the target architecture.
-
Projection of the . . . Process Network Computation Model
2005Co-Authors: Abdelkader Amar, Pierre Boulet, Philippe DumontAbstract:The Array-OL specification model has been introduced to model systematic signal processing applications. This model is multidimensional and allows to express the full Potential Parallelism of an application: both task and data Parallelism. The Array-OL language is an expression of data-dependences and thus allows many execution orders. In order to execute Array-OL applications on distributed architectures, we show here how to project such specification onto the Kahn process network model of computation. We show how Array-OL code transformations allow to choose a projection adapted to the target architecture. An experiment on a distributed process network implementation based on CORBA concludes this article
Shaojun Wei - One of the best experts on this subject based on the ideXlab platform.
-
low computing load high Parallelism detection method based on chebyshev iteration for massive mimo systems with vlsi architecture
IEEE Transactions on Signal Processing, 2017Co-Authors: Guiqiang Peng, Leibo Liu, Peng Zhang, Shouyi Yin, Shaojun WeiAbstract:Minimum-mean-square-error (MMSE) detection is becoming increasingly relevant in signal detection for massive multiple-input-multiple-output systems because of the increasing numbers of both users and antennas. This paper proposes a signal detection method called parallelizable Chebyshev iteration (PCI) that reduces the computing load and explores the Potential Parallelism of matrix inversions and multiplications, which are both major issues in MMSE detection. First, an eigenvalue-approximation-based method is used to obtain an initial solution. Then, optimized Chebyshev iteration is applied for the approximate computation of matrix inversions and multiplications. The number of multiplications is reduced from $\mathcal {O}(BU^2+U^3)$ to $\mathcal {O}(KBU)$ , where $B$ , $U$ , and $K$ are the numbers of antennas, users and iterations, respectively. The PCI method eliminates the correlations in large-scale matrix inversions and multiplications, thereby improving the Parallelism among elements of the estimated vector. These improvements are achieved at the cost of a minor reduction in detection accuracy. Based on the PCI method, a very-large-scale-integration fully pipelined architecture is proposed to realize 128 $\times$ 16 64-QAM MMSE detection. Here, the iterative parameters are obtained through approximate computations and are used repeatedly, and the user-level pipeline processing pattern achieves an optimal tradeoff among the throughput, area, and power. This architecture was verified on an FPGA, and the layout was implemented using TSMC 65 nm 1P9M CMOS technology. Results of 2.46 Gbps/W (throughput/power) and 0.53 Gbps/mm $^2$ (throughput/area) were obtained, which represent increases of 4.56 $\times$ and 3.79 $\times$ , respectively, compared with current state-of-the-art designs.