Jose Ricardo Silva Junior

Fluid Simulation with Two-Way Interaction Rigid Body Using a Heterogeneous GPU and CPU Environment

Simulation of natural phenomena, such as water and smoke, is a very important topic to increase real time scene realism in video-games. Besides the graphical aspect, in order to achieve realism, it is necessary to correctly simulate and solve its complex governing equations, requiring an intense computational work.Fluid simulation is achieved by solving the Navier-Stokes set of equations, using a numerical method in CPU or GPU, independently, as these equations do not have an analytical solution. The real time simulacraon also requires the simulation of interaction of the particles with objects in the scene, requiring many collision and contact forces calculation, which may drastically increase the computational time. In this paper we propose an heterogeneous multicore CPU and GPU hybrid architecture for fluid simulation with two-ways of interaction between them, and with a fine granularity control over rigid bodys shape collision. We also show the impact of this heterogeneous architecture over GPU and CPU bounded simulations, which is commonly used for this kind of application. The heterogeneous architecture developed in this work is developed to best fit the Single Instruction Multiple Thread (SIMT) model used by GPUs in all simulation stages, allowing a high level performance increase.

Neighborhood grid

This paper introduces a novel and efficient data structure, called neighborhood grid, capable of supporting large number of particle based elements on GPUs (graphics processing units), and is used for optimizing fluid animation with the use of GPU computing. The presented fluid simulation approach is based on SPH (smoothed particle hydrodynamics) and uses a unique algorithm for the neighborhood gathering. The brute force approach to neighborhood gathering of n particles has complexity O ( n 2 ) , since it involves proximity queries of all pairs of fluid particles in order to compute the relevant mutual interactions. Usually, the algorithm is optimized by using spatial data structures which subdivide the environment in cells and then classify the particles among the cells based on their position, which is not efficient when a large number of particles are grouped in the same cell. Instead of using such approach, this work presents a novel and efficient data structure that maintains the particles into another form of proximity data structure, called neighborhood grid. In this structure, each cell contains only one particle and does not directly represent a discrete spatial subdivision. The neighborhood grid does process an approximate spatial neighborhood of the particles, yielding promising results for real time fluid animation, with results that goes up to 9 times speedup, when compared to traditional GPU approaches, and up to 100 times when compared against CPU implementations. We present a new data structure for the neighborhood gathering on fluid simulation, called neighborhood grid.The neighborhood grid has an expressive speedup against the uniform grid on GPUs.The neighborhood grid uses less memory when compared with the uniform grid.

A heterogeneous system based on GPU and multi-core CPU for real-time fluid and rigid body simulation

Computational fluid dynamics in simulation has become an important field not only for physics and engineering areas but also for simulation, computer graphics, virtual reality and even video game development. Many efficient models have been developed over the years, but when many contact interactions must be processed, most models present difficulties or cannot achieve real-time results when executed. The advent of parallel computing has enabled the development of many strategies for accelerating the simulations. Our work proposes a new system which uses some successful algorithms already proposed, as well as a data structure organisation based on a heterogeneous architecture using CPUs and GPUs, in order to process the simulation of the interaction of fluids and rigid bodies. This successfully results in a two-way interaction between them and their surrounding objects. As far as we know, this is the first work that presents a computational collaborative environment which makes use of two different paradigms of hardware architecture for this specific kind of problem. Since our method achieves real-time results, it is suitable for virtual reality, simulation and video game fluid simulation problems.

A GPU-based Architecture for Parallel Image-aware Version Control

Version control is considered a vital component for supporting professional software development and has been widely used for textual artifacts, like source code. However, binary artifacts have received small attention when compared to the former. This fact can impose huge restrictions for projects in the game and media industry, which use large amount of binary data, such as images, videos, graphics, 3D models, and animations, together with source code. For these kinds of artifacts, existing strategies, such as storing the file as a whole for each commit or performing conventional binary delta, consume significant storage space with duplicate data, and even worse, lose vital semantic information. As a response to this problem, this paper introduces an infrastructure to support version control of image artifacts. Due to the amount of data that must be processed, we implemented our proposal using a GPU architecture, allowing a massively parallel approach for version control. The proposed architecture provides speedup over 55 X if compared to the same implementation in CPU.

Efficient image-aware version control systems using GPU

Version control is considered to be a vital component for supporting professional software development. While it has been widely used for textual artifacts, such as source code or documentation, little attention has been given to binary artifacts. This omission can place huge restrictions on projects in the game and media industries as they contain large amounts of binary data, such as images, videos, three‐dimensional models, and animations, along with their source code. For these kinds of artifacts, existing strategies such as storing the file as a whole for each revision or saving conventional binary deltas consume significant storage space with duplicate data and, even worse, do not provide any understandable information on which modifications were made. As a response to this problem, this paper introduces a change‐set model infrastructure to support version control of image artifacts using a specialized data structure. Additionally, our approach can deal with the maintenance of duplicate nearly identical images through a merge operation. Because of the amount of data that has to be processed, we designed our solution based on a parallel architecture, which permits a massively parallel approach to version control. The paper also compares our approach with some popular open‐source version control systems, showing their repository growth in relation to ours as well as the time required to process image artifacts. Finally, we demonstrate that our architecture requires less storage space and runs much faster than current methods. Copyright

A New Gravitational N-body Simulation Algorithm for Investigation of Cosmological Chaotic Advection

Recently alternative approaches in cosmology seeks to explain the nature of dark matter as a direct result of the non-linear spacetime curvature due to different types of deformation potentials. In this context, a key test for this hypothesis is to examine the effects of deformation on the evolution of large scales structures. An important requirement for the fine analysis of this pure gravitational signature (without dark matter elements) is to characterize the position of a galaxy during its trajectory to the gravitational collapse of super clusters at low redshifts. In this context, each element in an gravitational N-body simulation behaves as a tracer of collapse governed by the process known as chaotic advection (or lagrangian turbulence). In order to develop a detailed study of this new approach we develop the COsmic LAgrangian TUrbulence Simulator (COLATUS) to perform gravitational N-body simulations based on Compute Unified Device Architecture (CUDA) for graphics processing units (GPUs). In this paper we report the first robust results obtained from COLATUS.Recently alternative approaches in cosmology seeks to explain the nature of dark matter as a direct result of the non-linear spacetime curvature due to different types of deformation potentials. In this context, a key test for this hypothesis is to examine the effects of deformation on the evolution of large scales structures. An important requirement for the fine analysis of this pure gravitational signature (without dark matter elements) is to characterize the position of a galaxy during its trajectory to the gravitational collapse of super clusters at low redshifts. In this context, each element in an gravitational N-body simulation behaves as a tracer of collapse governed by the process known as chaotic advection (or lagrangian turbulence). In order to develop a detailed study of this new approach we develop the COsmic LAgrangian TUrbulence Simulator (COLATUS) to perform gravitational N-body simulations based on Compute Unified Device Architecture (CUDA) for graphics processing units (GPUs). In this p...

Two-Way Real Time Fluid Simulation Using a Heterogeneous Multicore CPU and GPU Architecture

Natural phenomena simulation, such as water and smoke, is a very important topic to increase real time scene realism in video-games. However, the computational fluid simulation is an expensive task since we must numerically solve the Navier-Stokes equations. Additionally, an immersing simulation requires interaction between the flow and the objects in the scene, increasing even more the computational work. In this paper we propose an heterogeneous multicore CPU and GPU scalable architecture for fluid simulation with two-way interaction with solid objects. We also show the impact of this architecture over GPU and CPU bounded simulations and present results that can reproduce complex fluid behavior in real time applications like games.

Understanding game sessions through provenance

Abstract The outcome of a gameplay session is derived from a series of events, decisions, and interactions made during the game. Many techniques have been developed by the game industry to understand a gameplay session. A successful technique is game analytics, which aims at understanding behavior patterns to improve game quality. However, current methods are not sufficient to capture underlying cause-and-effect relationships that occur during a gameplay session, which would allow designers to better identify possible mistakes in the mechanics or fine-tune their game. Recently, it was proposed a conceptual framework based on provenance to capture these relationships. In this paper, we present a concrete framework to capture provenance data, allowing developers to add provenance gathering capabilities to their games. We instantiated our framework in two games, showing how it can be used in practice, and we developed a new game to demonstrate how provenance could be employed in early stages of game development to assist balancing the difficulty. We conducted an experiment with twelve volunteers and used the gathered provenance data to answer designers’ frequent questions when trying to understand game sessions and balancing the difficulty of their games. This supports the relevance of collecting provenance data from games.

NGrid: a proximity data structure for fluids animation with GPU computing

This paper introduces a novel and efficient data structure capable of supporting a large number of particle-based elements in a GPU architecture such as fluids animation. The presented fluid animation approach is based on SPH (Smoothed Particle Hydrodynamics) and uses a unique algorithm for the neighborhood gathering, required during a particle processing. Usually, this kind of information about neighborhood is provided by algorithm which use spatial data structures, subdividing the environment and classifying each particle among their position in space. Unfortunately, it does not provide efficiency for a large number of particles grouped closes to each other as most of them will fall in the same cell. Instead of using such approaches, this work presents a novel and efficient data structure that maintains the particles into another kind of proximity data structure, called NGrid. In this structure, each cell contains only one particle and does not directly represent a discrete spatial subdivision. The NGrid does process an approximate spatial neighborhood of the particles, yielding promising results for real time fluid animation, with results that goes up to 8× speedup, when compared to traditional GPU approaches, and up to 100× when compared against CPU implementations.

Using 3D Stereoscope for Detailed Breakout and Drilling Fractures Visualization

While 3D stereoscopy can drastically increase the immersion in virtual environments, many features that were not relevant in other fields become important and should be considered in scientific visualization. Particularly, in virtual environments where there is a single object, the lack of spatial references makes it difficult to navigate and manipulate virtual elements. This work proposes a methodology for borehole breakouts and fracture visualization and analysis, using 3D stereoscopy. This is important in order to increase the precision of the environmental manipulation, considering important design and visualization concepts for a pipeline. With our proposal, tools and system manipulations that were complex to be used, due the nature of single elements in the environment, become much easier to operate.