Criticality guided energy aware speculation for speculative multithreaded processors

Unending quest for performance improvement coupled with the advancements in integrated circuit technology have led to the development of new architectural paradigm. Speculative multithreaded architecture (SpMT) philosophy relies on aggressive speculative execution for improved performance. However, aggressive speculative execution comes with a mixed flavor of improving performance, when successful, and adversely affecting the energy consumption (and performance) because of useless computation in the event of mis-speculation. Dynamic instruction criticality information can be usefully applied to control and guide such an aggressive speculative execution. In this paper, we present a model of micro-execution for SpMT architecture that we have developed to determine the dynamic instruction criticality. We have also developed two novel techniques utilizing the criticality information namely delaying the non-critical loads and the criticality based thread-prediction for reducing useless computations and energy consumption. Experimental results showing break-up of critical instructions and effectiveness of proposed techniques in reducing energy consumption are presented in the context of multiscalar processor that implements SpMT architecture. Our experiments show 17.7% and 11.6% reduction in dynamic energy for criticality based thread prediction and criticality based delayed load scheme respectively while the improvement in dynamic energy delay product is 13.9% and 5.5%, respectively. (c) 2012 Published by Elsevier B.V.

Long-range Coulomb interactions and nanoscale electronic inhomogeneities in correlated oxides

Electronic, magnetic, or structural inhomogeneities ranging in size from nanoscopic to mesoscopic scales seem endemic and are possibly generic to colossal magnetoresistance manganites and other transition metal oxides. They are hence of great current interest and understanding them is of fundamental importance. We show here that an extension, to include long-range Coulomb interactions, of a quantum two-fluid l-b model proposed recently for manganites [Phys. Rev. Lett. 92, 157203 (2004)] leads to an excellent description of such inhomogeneities. In the l-b model two very different kinds of electronic states, one localized and polaronic (l) and the other extended or broad band (b) coexist. For model parameters appropriate to manganites and even within a simple dynamical mean-field theory (DMFT) framework, it describes many of the unusual phenomena seen in manganites, including colossal magnetoresistance (CMR), qualitatively and quantitatively. However, in the absence of long-ranged Coulomb interaction, a system described by such a model would actually phase separate, into macroscopic regions of l and b electrons, respectively. As we show in this paper, in the presence of Coulomb interactions, the macroscopic phase separation gets suppressed and instead nanometer scale regions of polarons interspersed with band electron puddles appear, constituting a kind of quantum Coulomb glass. We characterize the size scales and distribution of the inhomogeneity using computer simulations. For realistic values of the long-range Coulomb interaction parameter V-0, our results for the thresholds for occupancy of the b states are in agreement with, and hence support, the earlier approach mentioned above based on a configuration averaged DMFT treatment which neglects V-0; but the present work has features that cannot be addressed in the DMFT framework. Our work points to an interplay of strong correlations, long-range Coulomb interaction, and dopant ion disorder, all inevitably present in transition metal oxides as the origin of nanoscale inhomogeneities rather than disorder frustrated phase competition as is generally believed. As regards manganites, it argues against explanations for CMR based on disorder frustrated phase separation and for an intrinsic origin of CMR. Based on this, we argue that the observed micrometer (meso) scale inhomogeneities owe their existence to extrinsic causes, e.g., strain due to cracks and defects. We suggest possible experiments to validate our speculation.

Paleoseismological evidence of surface faulting along the northeastern Himalayan front, India: Timing, size, and spatial extent of great earthquakes

The similar to 2500 km long Himalayan arc has experienced three large to great earthquakes of M-w 7.8 to 8.4 during the past century, but none produced surface rupture. Paleoseismic studies have been conducted during the last decade to begin understanding the timing, size, rupture extent, return period, and mechanics of the faulting associated with the occurrence of large surface rupturing earthquakes along the similar to 2500 km long Himalayan Frontal Thrust (HFT) system of India and Nepal. The previous studies have been limited to about nine sites along the western two-thirds of the HFT extending through northwest India and along the southern border of Nepal. We present here the results of paleoseismic investigations at three additional sites further to the northeast along the HFT within the Indian states of West Bengal and Assam. The three sites reside between the meizoseismal areas of the 1934 Bihar-Nepal and 1950 Assam earthquakes. The two westernmost of the sites, near the village of Chalsa and near the Nameri Tiger Preserve, show that offsets during the last surface rupture event were at minimum of about 14 m and 12 m, respectively. Limits on the ages of surface rupture at Chalsa (site A) and Nameri (site B), though broad, allow the possibility that the two sites record the same great historical rupture reported in Nepal around A.D. 1100. The correlation between the two sites is supported by the observation that the large displacements as recorded at Chalsa and Nameri would most likely be associated with rupture lengths of hundreds of kilometers or more and are on the same order as reported for a surface rupture earthquake reported in Nepal around A.D. 1100. Assuming the offsets observed at Chalsa and Nameri occurred synchronously with reported offsets in Nepal, the rupture length of the event would approach 700 to 800 km. The easternmost site is located within Harmutty Tea Estate (site C) at the edges of the 1950 Assam earthquake meizoseismal area. Here the most recent event offset is relatively much smaller (<2.5 m), and radiocarbon dating shows it to have occurred after A.D. 1100 (after about A.D. 1270). The location of the site near the edge of the meizoseismal region of the 1950 Assam earthquake and the relatively lesser offset allows speculation that the displacement records the 1950 M-w 8.4 Assam earthquake. Scatter in radiocarbon ages on detrital charcoal has not resulted in a firm bracket on the timing of events observed in the trenches. Nonetheless, the observations collected here, when taken together, suggest that the largest of thrust earthquakes along the Himalayan arc have rupture lengths and displacements of similar scale to the largest that have occurred historically along the world's subduction zones.

Anomalous viscous loss in emulsions

We propose a model for concentrated emulsions based on the speculation that a macroscopic shear strain does not produce an affine deformation in the randomly close-packed droplet structure. The model yields an anomalous contribution to the complex dynamic shear modulus that varies as the square root of frequency. We test this prediction using a novel light scattering technique to measure the dynamic shear modulus, and directly observe the predicted behavior over six decades of frequency and a wide range of volume fractions.

Simultaneous MultiStreaming for complexity-effective VLIW architectures

Very Long Instruction Word (VLIW) architectures exploit instruction level parallelism (ILP) with the help of the compiler to achieve higher instruction throughput with minimal hardware. However, control and data dependencies between operations limit the available ILP, which not only hinders the scalability of VLIW architectures, but also result in code size expansion. Although speculation and predicated execution mitigate ILP limitations due to control dependencies to a certain extent, they increase hardware cost and exacerbate code size expansion. Simultaneous multistreaming (SMS) can significantly improve operation throughput by allowing interleaved execution of operations from multiple instruction streams. In this paper we study SMS for VLIW architectures and quantify the benefits associated with it using a case study of the MPEG-2 video decoder. We also propose the notion of virtual resources for VLIW architectures, which decouple architectural resources (resources exposed to the compiler) from the microarchitectural resources, to limit code size expansion. Our results for a VLIW architecture demonstrate that: (1) SMS delivers much higher throughput than that achieved by speculation and predicated execution, (2) the increase in performance due to the addition of speculation and predicated execution support over SMS averages around 12%. The minor increase in performance might not warrant the additional hardware complexity involved, and (3) the notion of virtual resources is very effective in reducing no-operations (NOPs) and consequently reduce code size with little or no impact on performance.