Taming Control Divergence in GPUs through Control Flow Linearization

Branch divergence is a very commonly occurring performance problem in GPGPU in which the execution of diverging branches is serialized to execute only one control flow path at a time. Existing hardware mechanism to reconverge threads using a stack causes duplicate execution of code for unstructured control flow graphs. Also the stack mechanism cannot effectively utilize the available parallelism among diverging branches. Further, the amount of nested divergence allowed is also limited by depth of the branch divergence stack. In this paper we propose a simple and elegant transformation to handle all of the above mentioned problems. The transformation converts an unstructured CFG to a structured CFG without duplicating user code. It incurs only a linear increase in the number of basic blocks and also the number of instructions. Our solution linearizes the CFG using a predicate variable. This mechanism reconverges the divergent threads as early as possible. It also reduces the depth of the reconvergence stack. The available parallelism in nested branches can be effectively extracted by scheduling the basic blocks to reduce the effect of stalls due to memory accesses. It can also increase execution efficiency of nested loops with different trip counts for different threads. We implemented the proposed transformation at PTX level using the Ocelot compiler infrastructure. We evaluated the technique using various benchmarks to show that it can be effective in handling the performance problem due to divergence in unstructured CFGs.

Two-Phase Flow With Phase Change in Porous Channels

Phase change heat transfer in porous media finds applications in various geological flows and modern heat pipes. We present a study to show the effect of phase change on heat transfer in a porous channel. We show that the ratio of Jakob numbers based on wall superheat and inlet fluid subcooling governs the liquid-vapor interface location in the porous channel and below a critical value of the ratio, the liquid penetrates all the way to the extent of the channel in the flow direction. In such cases, the Nusselt number is higher due to the proximity of the liquid-vapor interface to the heat loads. For higher heat loads or lower subcooling of the liquid, the liquid-vapor interface is pushed toward the inlet, and heat transfer occurs through a wider vapor region thus resulting in a lower Nusselt number. This study is relevant in the designing of efficient two-phase heat exchangers such as capillary suction based heat pipes where a prior estimation of the interface location for the maximum heat load is required to ensure that the liquid-vapor interface is always inside the porous block for its operation.

Unsteady flow and heat transfer between rotating coaxial disks

A study is made on the flow and heat transfer of a viscous fluid confined between two parallel disks. The disks are allowed to rotate with different time dependent angular velocities, and the upper disk is made to approach the lower one with a constant speed. Numerical solutions of the governing parabolic partial differential equations are obtained through a fourth-order accurate compact finite difference scheme. The normal forces and torques that the fluid exerts on the rotating surfaces are obtained at different nondimensional times for different values of the rate of squeezing and disk angular velocities. The temperature distribution and heat transfer are also investigated in the present analysis.

A water model study of the flow asymmetry inside a continuous slab casting mold

The work studies the extent of asymmetric flow in water models of continuous casting molds of two different configurations. In the molds where fluid is discharged through multiple holes at the bottom, the flow pattern in the lower portion depends on the size of the lower two recirculating domains. If they reach the mold bottom, the flow pattern in the lower portion is symmetrical about the central plane; otherwise, it is asymmetrical. On the other hand, in the molds where the fluid is discharged through the entire mold cross section, the flow pattern is always asymmetrical if the aspect ratio is 1:6.25 or more. The fluid jet swirls while emerging through the nozzle. The interaction of the swirling Jets with the wide sidewalls of the mold gives rise to asymmetrical flow inside the mold. In the molds with lower aspect ratios, where the jets do not touch the wide side walls, the flow pattern is symmetrical about the central plane.

Flow behavior of long chain plasticizing liquids: Segment size calculations in carboxylate esters

This is the first comprehensive report on the calculation of segment size, which signifies the asic unit of flow in long chain plasticizing liquids, by a novel multi-pronged approach. Unlike,low molecular weight liquids and high polymer melts these complex long chain liquids encompasses the least understood domain of the liquid state. In the present work the flow behaviour of carboxylate ester (300-900 Da) has been explained through segmental motion taking into account the independence of molecular weight region. The segment size have been calculated by various methods based on satistical thermodynamics, molecular dynamics and group additivity nd their merits analysed.

Flow behavior of LDPE and its blends with LLDPE I and II: A comparative study

Studies on melt rheological properties of blends of low density polyethylene (LDPE) with selected grades of linear low density polyethylene (LLDPE), which differ widely in their melt flow indices, are reported, The data obtained in a capillary rheometer are presented to describe the effects of blend composition and shear rate on flow behavior index, melt viscosity, and melt elasticity. In general, blending of LLDPE I that has a low melt flow index (2 g/10 min) with LDPE results in a decrease of its melt viscosity, processing temperature, and the tendency of extrudate distortion, depending on blending ratio. A blending ratio around 20-30% LLDPE I seems optimum from the point of view of desirable improvement in processability behavior. On the other hand, blending of LLDPE II that has a high melt flow index (10 g/10 min) with LDPE offers a distinct advantage in increasing the pseudoplasticity of LDPE/LLDPE II blends.

Unsteady flow with attenuation in a fluid filled elastic tube with a stenosis

An equation governing the excess pressure has been derived, for an axially tethered and stenosed elastic tube filled with viscous liquid, by introducing the elasticity of the tube through pressure-area relation. This equation is solved numerically for large Womersley parameter and the results are presented for different types of pressure-radius relations and geometries by prescribing an outgoing wave suffering attenuation at some axial point of the tube. For a locally constricted tube it is observed that the pressure oscillates more and generates sound on the down stream side of the constriction.

Flow of metal into the flash gap in the last stages of a plane-strain closed-die forging operation

In closed-die forging the flash geometry should be such as to ensure that the cavity is completely filled just as the two dies come into contact at the parting plane. If metal is caused to extrude through the flash gap as the dies approach the point of contact — a practice generally resorted to as a means of ensuring complete filling — dies are unnecessarily stressed in a high-stress regime (as the flash is quite thin and possibly cooled by then), which reduces the die life and unnecessarily increases the energy requirement of the operation. It is therefore necessary to carefully determine the dimensions of the flash land and flash thickness — the two parameters, apart from friction at the land, which control the lateral flow. The dimensions should be such that the flow into the longitudinal cavity is controlled throughout the operation, ensuring complete filling just as the dies touch at the parting plane. The design of the flash must be related to the shape and size of the forging cavity as the control of flow has to be exercised throughout the operation: it is possible to do this if the mechanics of how the lateral extrusion into the flash takes place is understood for specific cavity shapes and sizes. The work reported here is part of an ongoing programme investigating flow in closed-die forging. A simple closed shape (no longitudinal flow) which may correspond to the last stages of a real forging operation is analysed using the stress equilibrium approach. Metal from the cavity (flange) flows into the flash by shearing in the cavity in one of the three modes considered here: for a given cavity the mode with the least energy requirement is assumed to be the most realistic. On this basis a map has been developed which, given the depth and width of the cavity as well as the flash thickness, will tell the designer of the most likely mode (of the three modes considered) in which metal in the cavity will shear and then flow into the flash gap. The results of limited set of experiments, reported herein, validate this method of selecting the optimum model of flow into the flash gap.

Pressure sensitive flow and constraint factor in amorphous materials below glass transition

The constraint factor, C (given by the hardness-yield strength ratio H/Y in the fully lastic regime of indentation), in metallic glasses, is greater than three, a reflection of the sensitivity of their plastic flow to pressure. Furthermore, C increases with increasing temperature. In this work, we examine if this is true in amorphous polymers as well, through experiments on amorphous poly(methyl methacrylate) (PMMA). Uniaxial compression as well as spherical indentation tests were conducted in the 248-348 K range to construct H/Y versus indentation strain plots at each temperature and obtain the C-values. Results show that C increases with temperature in PMMA as well. Good correlation between the loss factors, measured using a dynamic mechanical analyzer, and C, suggest that the enhanced sensitivity to pressure is possibly due to beta-relaxation. We offer possible mechanistic reasons for the observed trends in amorphous materials in terms of relaxation processes.

A couple stress model of blood flow in the microcirculation

A simple mathematical model depicting blood flow in the capillary is developed with an emphasis on the permeability property of the blood vessel based on Starling's hypothesis. In this study the effect of inertia has been neglected in comparison with the viscosity on the basis of the smallness of the Reynolds number of the flow in the capillary. The capillary blood vessel is approximated by a circular cylindrical tube with a permeable wall. The blood is represented by a couple stress fluid. With such an ideal model the velocity and pressure fields are determined. It is shown that an increase in the couple stress parameter increases the resistance to the flow and thereby decreases the volume rate flow. A comparison of the results with those of the Newtonian case has also been made.

A hydrodynamical study of the flow in renal tubules

The hydrodynamical problem of flow in proximal renal tubule is investigated by considering axisymmetric flow of a viscous, incompressible fluid through a long narrow tube of varying cross-section with reabsorption at the wall. Two cases for reabsorption have been studied (i) when the bulk flow,Q, decays exponentially with the axial distancex, and (ii) whenQ is an arbitrary function ofx such thatQ-Q 0 can be expressed as a Fourier integral (whereQ 0 is the flux atx=0). The analytic expressions for flow variables have been obtained by applying perturbation method in terms of wall parameter ε. The effects of ε on pressure drop across the tube, radial velocity and wall shear have been studied in the case of exponentially decaying bulk flow and it has been found that the results are in agreement with the existing ones for the renal tubules.

DFL: A Data Flow Language

Many novel computer architectures like array and multiprocessors which achieve high performance through the use of concurrency exploit variations of the von Neumann model of computation. The effective utilization of the machines makes special demands on programmers and their programming languages, such as the structuring of data into vectors or the partitioning of programs into concurrent processes. In comparison, the data flow model of computation demands only that the principle of structured programming be followed. A data flow program, often represented as a data flow graph, is a program that expresses a computation by indicating the data dependencies among operators. A data flow computer is a machine designed to take advantage of concurrency in data flow graphs by executing data independent operations in parallel. In this paper, we discuss the design of a high level language (DFL: Data Flow Language) suitable for data flow computers. Some sample procedures in DFL are presented. The implementation aspects have not been discussed in detail since there are no new problems encountered. The language DFL embodies the concepts of functional programming, but in appearance closely resembles Pascal. The language is a better vehicle than the data flow graph for expressing a parallel algorithm. The compiler has been implemented on a DEC 1090 system in Pascal.

Studies on micro explosive driven blast wave propagation in confined domains using NONEL tubes

Explosive driven micro blast waves are generated in the laboratory using NONEL tubes. The explosive mixture coated to the inner walls of the plastic Nonel tube comprises of HMX and Aluminum ( 18mg/m). The detonation is triggered electrically to generate micro blast waves from the open end of the tube. Flow visualization and over pressure measurements have been carried out to understand the propagation dynamics of these micro-blast waves in both confined and unconfined domains. The classical cubic root law used for large scale blast correlation appears to hold good even for these micro-blasts generated in the laboratory.

A note on the movement of single large liquid bubbles in vertical tubes

The flow of single large liquid bubbles under gravity in closed tubes is studied here for the case when the liquid bubble exhibits micropolar behaviour. The film thickness, velocity profile in the bubble and film, and nonNewtonian effects are studied and compared with those for the correspondingNewtonian fluid. The investigation is restricted to the case where the bubble length is far greater than the tube radius.

Unsteady laminar boundary layers in a compressible stagnation flow

The unsteady laminar compressible boundary-layer flow in the immediate vicinity of a two-dimensional stagnation point due to an incident stream whose velocity varies arbitrarily with time is considered. The governing partial differential equations, involving both time and the independent similarity variable, are transformed into new co-ordinates with finite ranges by means of a transformation which maps an infinite interval into a finite one. The resulting equations are solved by converting them into a matrix equation through the application of implicit finite-difference formulae. Computations have been carried out for two particular unsteady free-stream velocity distributions: (1) a constantly accelerating stream and (2) a fluctuating stream. The results show that in the former case both the skin-friction and the heat-transfer parameter increase steadily with time after a certain instant, while in the latter they oscillate thus responding to the fluctuations in the free-stream velocity.