Yeongkyu Lim

A Selective Ahead-Of-Time Compiler on Android Device

The runtime environment of Android is based on its own Java Virtual Machine(JVM) called Dalvik Virtual Machine (DVM) which is said to be having overhead of interpreting every bytecode to machine code during runtime just like other JVMs. There are already well known techniques to overcome runtime overhead of interpreting and Google has adapted one of them which is Just-In-Time Compiler (JITC) since Android 2.2, Froyo. Google chose trace based JITC to JIT compile only the hottest of hot code by using lazy fashion with counting strategy to minimize memory usage, thus applying various optimization techniques during JIT compilation to generate more efficient machine code is limited. In order to minimize runtime interpreting and compiling overhead, here a selective Ahead-Of-Time Compiler (AOTC) which generates machine code at static compile time with abundant optimization techniques for the selected hot methods by profiling will be introduced. The experimented results show that AOT compiling hot methods at static compile time and letting the others to aim the benefit of runtime JITC gives 5% of performance upgrade in average on ARM11 600MHz target environment.

Adaptive undersampling for efficient mobile ray tracing

Aiming to develop an efficient ray tracer for a mobile platform, we present an adaptive undersampling method that enhances the rendering speed by effectively replacing expensive ray-tracing operations with cheap interpolation whenever possible. Our method explores both object- and image-space information gathered during ray tracing to detect possibly problematic pixels. Rays are fired only for these pixels. We also present a postcorrection algorithm that minimizes annoying artifacts inevitably caused by undersampling. Our implementation on a mobile GPU demonstrates that this method can speed up the rendering computation significantly, while retaining almost the same visual quality of the rendering.

L-Bench: An Android benchmark set for low-power mobile GPUs ☆

Abstract In recent years GPUs have become one of the most important components in mobile application processors (APs). Thus, performance measurement and analysis of mobile GPUs are crucial to mobile AP manufacturers, device manufacturers, graphics application programmers, and end users. However, it is hard to analyze mobile GPUs in depth via existing high-level (with frames per second) or low-level benchmarks (with a fill rate, ALU performance, etc.). To bridge the gap between the benchmarks, we present a novel Android benchmark set for low-power GPUs, called L-Bench. This benchmark set consists of mid-level micro-benchmarks implemented on OpenGL ES 3.1, which are carefully chosen for different workloads. By analyzing the results, this benchmark suite provides not only frames per second of each benchmark but also performance of each GPU subsystem (geometry units, ALUs, texture mapping units, raster operations pipelines, caches/memory units, and tessellators) and overall GPU performance. For experiments, we perform our benchmark suite on five representative mobile devices that have different mobile GPUs, after that, we describe comprehensive analysis of each GPU architecture.

Performance optimization of 3D applications by OpenGL ES library hooking in mobile devices

The mobile GPU (Graphic Processing Unit) market has grown steadily due to expansion of the mobile game industry. Despite the rapid computation capability of mobile devices, handling a large amount of high-quality graphics in real-time is difficult. Therefore, effective technologies for improving mobile GPU in smartphones are required. In this thesis, we examine the trade-off between quality and performance, and address the benefits of graphic performance improvement by degrading quality. To implement this idea, we propose performance optimization methodologies for 3D applications using an OpenGL ES library hooking method. Our methodologies do not require any source code from 3D applications, and can be applied to any Android phones that use OpenGL ES in real-time. To demonstrate the benefits of our methodology, we conducted performance verifications of five well-known benchmarks using a smartphone, and measured the quality in accordance with each methodology. In addition, we showed the optimal trade-offs between quality and performance. By using the proposed technique, the performance of mobile GPU can be significantly improved to achieve a better trade-off between quality and performance.

A Performance Prediction Model of Parallel DCT on Mobile Embedded Systems

The recent development on semiconductor process and design technologies enables multi-core processors to become a dominant market trend in mobile devices. The parallel programming enabled by multi-core CPU can provide a great opportunity to increase the processing performance. This paper explores a performance prediction model of parallel DCT on heterogeneous mobile systems by measuring power dissipation. For our simulation, we implemented the fast DCT algorithm on various computing environments and the simulation results show the feasibility of the proposed method to estimate the performance gain in terms of power consumption on heterogeneous embedded systems.

Classified texture resizing for mobile devices

Power consumption is one of the most important factors in mobile computing. Especially for high-quality games, it takes a lot of computing power to render visual effects. In order to reduce this, some rendering techniques (e.g., Samsung Game Tuner) adjust rendering parameters (screen resolution, frame rates, and texture sizes) to improve power efficiency or performance. Among them, the texture resizing reduces power consumption in some cases, but it sometimes results in poor rendering quality or no energy saving. To improve the texture resizing, we present the classified texture resizing technique. Our main idea is to classify textures into certain types and to apply a different approach to each type. As a result, our approach minimizes degradation of rendering quality and can be applied to wider applications. Our experimental results show up to 16% power reduction of a GPU and DRAM.

Z 2 traversal order: An interleaving approach for VR stereo rendering on tile-based GPUs

With increasing demands of virtual reality (VR) applications, efficient VR rendering techniques are becoming essential. Because VR stereo rendering has increased computational costs to separately render views for the left and right eyes, to reduce the rendering cost in VR applications, we present a novel traversal order for tile-based mobile GPU architectures: Z2 traversal order. In tile-based mobile GPU architectures, a tile traversal order that maximizes spatial locality can increase GPU cache efficiency. For VR applications, our approach improves upon the traditional Z order curve. We render corresponding screen tiles in left and right views in turn, or simultaneously, and as a result, we can exploit spatial adjacency of the two tiles. To evaluate our approach, we conducted a trace-driven hardware simulation using Mesa and a hardware simulator. Our experimental results show that Z2 traversal order can reduce external memory bandwidth requirements and increase rendering performance.

Z 2 traversal order for VR stereo rendering on tile-based mobile GPUs

With increasing demands of virtual reality (VR) applications, efficient VR rendering techniques are becoming essential because VR stereo rendering requires increased computational costs to separately render views for the left and right eyes. To reduce the rendering cost in VR applications, we present a novel traversal order for tile-based mobile GPU architectures, called the Z2 traversal order. In tile-based mobile GPU architectures, a tile traversal order that maximizes spatial locality can increase the GPU cache efficiency. For VR applications, our approach improves the traditional Z-curve order; we render two screen tiles in the left and right views by turns or simultaneously, as a result, we can exploit spatial locality between the two tiles. To evaluate our approach, we conducted a trace-driven hardware simulation using Mesa and a hardware simulator. The experimental results show that the Z2 traversal order can reduce external memory bandwidth requirements and can increase rendering performance.

AXAA: adaptive approximate anti-aliasing

Post-processing anti-aliasing algorithms are widely used now for real-time rendering because of their simplicity, performance, and suitability for deferred shading. Fast approximate anti-aliasing (FXAA) [Lottes 2009] is the fastest method among them, so many games support FXAA to get anti-aliased images. However, FXAA can easily lose texture details and text sharpness due to its excessive blurring. To alleviate those problems of FXAA, we present adaptive approximate anti-aliasing (AXAA). Our approach adds three contributions to FXAA in order to avoid unnecessary filtering. First, we stop further anti-aliasing processes if the current pixel or its neighbors are judged as pixels on already filtered textures or fonts. Second, we try to maintain thin lines as much as possible in order to avoid blurring fonts and lines. Third, for higher performance, we adaptively set the search range of each pixel according to luma contrast. Our experiments show that AXAA provides significantly better image quality than FXAA, in terms of texture, text, and geometry details. Nevertheless, processing overhead of AXAA is still similar to that of FXAA.

Effective stereoscopic rendering for mobile VR

Lately, the VR market is expanding rapidly with the advent of mobile phone based VR such as Google Cardboard, Samsung Gear VR, LG 360VR starting with VR(Virtual Reality) HMD(Head Mount Device) PC based VR using Stereoscopic rendering such as Oculus Rift, Vive.