Dynamic Threads Research Articles

Leveraging Memory Copy Overlap for Efficient Sparse Matrix-Vector Multiplication on GPUs

Sparse matrix-vector multiplication (SpMV) is central to many scientific, engineering, and other applications, including machine learning. Compressed Sparse Row (CSR) is a widely used sparse matrix storage format. SpMV using the CSR format on GPU computing platforms is widely studied, where the access behavior of GPU is often the performance bottleneck. The Ampere GPU architecture recently from NVIDIA provides a new asynchronous memory copy instruction, memcpy_async, for more efficient data movement in shared memory. Leveraging the capability of this new memcpy_async instruction, we first propose the CSR-Partial-Overlap to carefully overlap the data copy from global memory to shared memory and computation, allowing us to take full advantage of the data transfer time. In addition, we design the dynamic batch partition and the dynamic threads distribution to achieve effective load balancing, avoid the overhead of fixing up partial sums, and improve thread utilization. Furthermore, we propose the CSR-Full-Overlap based on the CSR-Partial-Overlap, which takes the overlap of data transfer from host to device and SpMV kernel execution into account as well. The CSR-Full-Overlap unifies the two major overlaps in SpMV and hides the computation as much as possible in the two important access behaviors of the GPU. This allows CSR-Full-Overlap to achieve the best performance gains from both overlaps. As far as we know, this paper is the first in-depth study of how memcpy_async can be potentially applied to help accelerate SpMV computation in GPU platforms. We compare CSR-Full-Overlap to the current state-of-the-art cuSPARSE, where our experimental results show an average 2.03x performance gain and up to 2.67x performance gain.

Hesiod and Hávamál : Transitions and the Transmission of Wisdom

Hesiod and Hávamál:Transitions and the Transmission of Wisdom Lilah Grace Canevaro1 (bio) Study of Hesiod’s Works and Days has long profited from comparative analyses.2 Akkadian, Sumerian, Egyptian, and Hebrew wisdom literature has all been brought to bear on the archaic Greek poem. Many of the Works and Days’ maxims find parallels in, for example, the Akkadian Counsels of Wisdom, or the Egyptian Instruction texts. Hesiod’s myths about the creation of mankind recall stories such as the Babylonian Enuma Eliš and the first part of Atrahasis, or the Biblical narrative of Adam and Eve. Studies such as those of Penglase (1994), West (1997), and Haubold (2013) have tracked the influence of the Near East on Greek literature and culture, positing the fourteenth and the ninth centuries BCE as particular hot-beds of intercultural exchange. It is not too much of a stretch to posit that an archaic Greek poet would have been aware of Near Eastern poetry. However, this is not necessarily the case. Whether parallels between the Works and Days and extant Near Eastern wisdom literature indicate diachronic reception or synchronic cultural similarity is a bit of a grey area. For example, Hesiod is concerned throughout the Works and Days with ideas of measure and balance.3 Egyptian wisdom texts have the same preoccupation: The Instruction of Amen-em-Opet chapter 16 reads “Do not lean on the scales nor falsify the weights, / Nor damage the fractions of the measure …,” and indeed they have the ape god Thoth guarding the balance (“Which god is as great as Thoth?”). Was Hesiod’s interest piqued by the Egyptian wisdom tradition, or was due measure in all things simply a common cultural concern?4 It is not only the possibility of direct influence that makes these comparative studies so compelling. They also give us a glimpse into another tradition: a tradition developing along similar lines and at a similar stage, being guided by poets with similar preoccupations, and being shaped for audiences with similar concerns. They show us that the handing down of wisdom is a cultural inevitability and that certain forms of its expression are constants. In this article I too offer a comparative analysis. However, I step away from the Near East and away from any suggestion of a chain of transmission. I aim to offer fresh insights into Hesiod’s Works and Days by comparing it to the Eddic Hávamál, a poem far removed in terms of geography and date, but compellingly close in subject matter, construction, and transmission. Those who have studied Hávamál, just like Hesiodic scholars, have tied themselves in knots trying to disentangle the strands of authorship and the narrative threads. Hávamál is, like the Works and Days, a wisdom poem with a composite structure. It is made up not only of precepts and maxims but also elaborate mythological sections. It is associated with catalogic elements which may be original or later accretions, just like Hesiod’s Days, or the Catalogue of Women, or the Ornithomanteia. And most interestingly it is, like the Works and Days, a poem rooted in oral tradition, but poised at that crucial juncture: the advent of writing. Hesiod’s Works and Days is unique in archaic poetry. In particular, it is the balance between modes of reading which Hesiod maintains throughout the Works and Days that proves truly striking. Both wisdom texts and epic poems can be (and were) read in their entirety and excerpted. But the Works and Days is unique in inviting these two modes of reading in roughly equal measure. I have yet to find an archaic wisdom text from Greece or the Near East with such a strong narrative framework as the Works and Days—one with dynamic threads evolving over the course of the poem and with an addressee whose behaviour gradually changes or a focus which consistently and inexorably narrows.5 Wisdom texts may be read from beginning to end, but they definitely lend themselves more readily to division and cherry-picking. They fall apart far more easily than they hang together. At the other end of the scale is heroic epic. Homer was...

Concurrency control generation for dynamic threads using discrete-event systems

The application of Discrete-Event Systems (DES) theory to the problem of guaranteeably enforcing concurrency constraints in multithreaded applications has been studied under certain assumptions, namely, the assumption of a static pool of pre-existing instantiated threads, whose creation and termination are not modeled. This work presents an extension of this case to handle dynamically instantiated and terminated threads using a Petri net formalism and an online limited-lookahead state-space search technique.

A Critical Performance Study of Memory Mapping on Multi-Core Processors: An Experiment with k-means Algorithm with Large Data Mining Data Sets

Increased availability of Multi-Core processors is forcing us to re-design algorithms and applications so as to exploit the available computational power from multiple cores. It is not uncommon to employ memory mapping of files in applications involving huge I/O bandwidth to improve the response/service times. This paper mainly focuses on performance of memory mapped files on Multi-Core processors. Experiments are carried out with k-means algorithm, a popular Data mining (DM) clustering algorithm, to explore the potential of Multi-Core hardware under OpenMP API and POSIX threads. Observations are made both with static and dynamic threads of OpenMP. Experiments are also conducted with both simulated and real data sets. Experiments indicate that memory mapping of files gives considerable benefit on Multi-Core processors also. In addition, the benefit increased with increased physical memory size. Also, the benefit of memory mapping with the selected algorithm is increasing with number of cores.

Securing interaction between threads and the scheduler in the presence of synchronization

The problem of information flow in multithreaded programs remains an important open challenge. Existing approaches to specifying and enforcing information-flow security often suffer from over-restrictiveness, relying on nonstandard semantics, lack of compositionality, inability to handle dynamic threads, inability to handle synchronization, scheduler dependence, and efficiency overhead for the code that results from security-enforcing transformations. This paper suggests a remedy for some of these shortcomings by developing a novel treatment of the interaction between threads and the scheduler. As a result, we present a permissive noninterference-like security specification and a compositional security type system that provably enforces this specification. The type system guarantees security for a wide class of schedulers and provides a flexible and efficiency-friendly treatment of dynamic threads.

DprA/Smf protein localizes at the DNA uptake machinery in competent Bacillus subtilis cells

BackgroundDprA is a widely conserved bacterial protein and has been shown to confer an important function during transformation in competent cells, possibly through protection of incoming DNA. B. subtilis DprA (called Smf) and has been shown to play an important role during transformation with chromosomal DNA, but its mode of action is unknown.ResultsWe show that B. subtilis DprA/Smf is more important for transformation with plasmid DNA than with chromosomal DNA. A functional Smf-YFP fusion localized as discrete foci to the cell pole in a subset of cells grown to competence, dependent on the ComK master transcription factor. Smf-YFP foci colocalized with ComGA-CFP. However, a considerable number of cells having high ComK activity contained Smf dispersed throughout the cytosol and lacked a polar Smf assembly. The absence of polar Smf-YFP foci in these cells strongly correlated with the absence of ComGA-CFP foci, and comGA mutant cells mostly lacked polar Smf-YFP foci. Smf formed polar assemblies in the absence of RecA, and RecA formed dynamic threads after addition of DNA in a smf deletion strain. Upon addition of DNA, Smf-YFP foci relocalized from the poles to the cell centre, dependent on the presence of RecA protein.ConclusionOur data show that Smf is recruited to the polar competence machinery, and that polar Smf assembly requires a functional DNA uptake complex. High ComK levels drive expression of Smf in 20% of all cells grown to competence, but not all competent cells contain a polar DNA uptake machinery, showing that ComK activity is necessary but not sufficient to achieve assembly of the uptake machinery in all cells. Smf and RecA localize independently of each other, in agreement with our finding that Smf is much more important for plasmid transformation than RecA, but RecA influences the dynamic localization pattern of Smf. Our data show that DprA/Smf acts downstream of the DNA uptake machinery, and support the idea that Smf protects incoming ssDNA, possibly in conjunction with RecA.

A Dynamic Dextral-Sinistral Model for the Structure and Evolution of Prominence Magnetic Fields

AbstractWe propose the following coherent model for prominences.(i)They are composed of many fine threads whose mass we suggest has two kinds of origin. Some of the mass may be lifted from the upper photosphere in response to cancellation of magnetic flux in small flux tubes. If this rate of mass supply is slow then the mass needs to be supported, but if the mass supply rate is as large as the natural rate of draining then no support is necessary. However, much of the mass may not need to be supplied or supported at all, if it comprises dynamic threads of cool plasma that are in pressure balance with their hot surroundings and are continually forming from the corona and heating up again in response to a localised heating mechanism.(ii)The overall structure is of a large-scale flux tube along the polarity inversion line with an overlying arcade that links the network on either side of the filament channel. The structure of the channel and of barbs is explained by the distribution of flux sources along the channels, with majority polarity producing plagettes and minority polarity the barbs.(iii)The chirality of polar-crown prominences is suggested to be caused by a combination of subsurface flows, flux emergence and reconnection. By contrast, two generic types of low-latitude filament, namely those associated with unipolar flux tongues and with the emergence of new active regions near to remnant regions, are a natural consequence of nonuniform flux distribution and converging flux.

APRIL

Processors in large-scale multiprocessors must be able to tolerate large communication latencies and synchronization delays. This paper describes the architecture of a rapid-context-switching processor called APRIL with support for fine-grain threads and synchronization. APRIL achieves high single-thread performance and supports virtual dynamic threads. A commercial RISC-based implementation of APRIL and a run-time software system that can switch contexts in about 10 cycles is described. Measurements taken for several parallel applications on an APRIL simulator show that the overhead for supporting parallel tasks based on futures is reduced by a factor of two over a corresponding implementation on the Encore Multimax. The scalability of a multiprocessor based on APRIL is explored using a performance model. We show that the SPARC-based implementation of APRIL can achieve close to 80% processor utilization with as few as three resident threads per processor in a large-scale cache-based machine with an average base network latency of 55 cycles.