Exploring GPU acceleration framework for climate based daylight modeling  

作　　者：Sida Du Yongqing Zhao Zhen Tian David Geisler-Moroder Wei Wang 

机构地区：[1]School of Architecture and Planning,Hunan University,Changsha,410082,China [2]Austria-China Low Carbon Building and Energy Joint Laboratory,Changsha,410082,China [3]University of Innsbruck,Unit of Energy Efficient Building,A-6020,Innsbruck,Austria

出　　处：《Building Simulation》2025年第1期33-46,共14页建筑模拟(英文)

基　　金：funded by the National Key R&D Program of China(Grant No.2023YFE0106700);National Natural Science Foundation of China(Grant No.52378096);the Hunan Provincial Natural Science Foundation(2023JJ30151);the Austria-China Low Carbon Building and Energy Joint Laboratory,the Hunan Key Laboratory of Science of Urban and Rural Human Settlements at Hilly Areas(2020TP1009);the project“CFS4LowCarb”in the TECXPORT Bilateral Cooperation Austria-PRC/MOST 2022 program under project FO999903877 financed by the Federal Ministry of Austria BMK managed by the FFG.

摘　　要：Indoor glare significantly affects the visual comfort and health of occupants,daylighting simulation can act as an effective method to analyze daylight glare issues during building design.To address the extensive computational costs associated with calculating annual daylight glare metrics with existing methods,this study introduces an acceleration framework,which can be widely applied.The framework integrates a newly developed daylight matrix multiplication program(DMM4GPU)for Graphics Processing Unit(GPU)computation acceleration and the previously developed Accelerad program,which can accelerate the calculation of Daylight Coefficient(DC)matrices in the Two-phase Method(2-PM),the View matrices in the Three-phase Method(3-PM)and Five-phase Method(5-PM).By comparing with standard Radiance Central Processing Unit(CPU)calculations,the study validated the acceleration framework’s simulation accuracy and significantly reduced computation time in daylight glare metrics calculations.It also analyzed the impact of various simulation parameters on the framework’s performance.Results indicate that the acceleration framework’s error in calculating Daylight Glare Probability(DGP)is minimal(RMSE<0.004),and the computation times for the 2-PM,3-PM,and 5-PM are reduced by 94.8%,93.9%,and 83.0%,respectively.Furthermore,this study discussed modeling techniques to avoid possible errors in the daylight GPU computations.

关 键 词：graphics processing unit acceleration of matrix multiplication glare evaluation computation time single-precision ray tracing 

分 类 号：TU113[建筑科学—建筑理论]