[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-1-en-105":3,"doc-seo-278600-105":53,"doc-detail-278600-en":126},{"code":4,"msg":5,"data":6},0,"success",[7,14,19,24,29,34,39,44,49],{"id":8,"doc_module":9,"doc_module_name":10,"category_name":11,"show_sort_weight":12,"slug":13},11,1,"Template","Presentations",90,"presentations",{"id":15,"doc_module":9,"doc_module_name":10,"category_name":16,"show_sort_weight":17,"slug":18},12,"Resumes",80,"resumes",{"id":20,"doc_module":9,"doc_module_name":10,"category_name":21,"show_sort_weight":22,"slug":23},14,"Invoices",70,"invoices",{"id":25,"doc_module":9,"doc_module_name":10,"category_name":26,"show_sort_weight":27,"slug":28},15,"Posters",60,"posters",{"id":30,"doc_module":9,"doc_module_name":10,"category_name":31,"show_sort_weight":32,"slug":33},16,"Social Media",50,"social-media",{"id":35,"doc_module":9,"doc_module_name":10,"category_name":36,"show_sort_weight":37,"slug":38},17,"Forms",40,"forms",{"id":40,"doc_module":9,"doc_module_name":10,"category_name":41,"show_sort_weight":42,"slug":43},18,"Letters",30,"letters",{"id":45,"doc_module":9,"doc_module_name":10,"category_name":46,"show_sort_weight":47,"slug":48},21,"Paper Templates",5,"papers-templates",{"id":50,"doc_module":9,"doc_module_name":10,"category_name":51,"show_sort_weight":4,"slug":52},158,"General","general-158",{"code":4,"msg":54,"data":55},"ok",{"site_id":56,"language":57,"slug":58,"title":59,"keywords":60,"description":61,"schema_data":62,"social_meta":119,"head_meta":121,"extra_data":123,"updated_unix":125},105,"en","a-performance-study-for-iterative-stencil-loops-on-gpus-with-ghost-zone-optimizations-performance-model-framework-to-automatically-select-ghost-zone-size","A Performance Study for Iterative Stencil Loops on GPUs with Ghost Zone Optimizations - Performance model framework to automatically select ghost zone size","","Iterative stencil loops are commonly parallelized via tiling, but boundary halo regions require communication and synchronization, creating substantial overhead on shared-memory architectures such as GPUs. Ghost zones replicate stencil work to reduce halo exchange costs, yet choosing an optimal ghost zone size depends on both application characteristics and hardware behavior. This work develops a performance model and a code-generation framework using NVIDIA Tesla as a case study. Validation on four ISL applications shows predicted configurations achieve at least 95% of the optimal speedup.",{"@graph":63,"@context":118},[64,80,101],{"@type":65,"itemListElement":66},"BreadcrumbList",[67,71,74,77],{"item":68,"name":69,"@type":70,"position":9},"https://docshare.wps.com","Home","ListItem",{"item":72,"name":10,"@type":70,"position":73},"https://docshare.wps.com/template/",2,{"item":75,"name":11,"@type":70,"position":76},"https://docshare.wps.com/template/presentations/",3,{"item":78,"name":59,"@type":70,"position":79},"https://docshare.wps.com/template/a-performance-study-for-iterative-stencil-loops-on-gpus-with-ghost-zone-optimizations-performance-model-framework-to-automatically-select-ghost-zone-size/278600/",4,{"url":78,"name":59,"@type":81,"image":82,"author":87,"headline":59,"publisher":90,"fileFormat":93,"inLanguage":57,"description":61,"dateModified":94,"datePublished":95,"encodingFormat":93,"isAccessibleForFree":96,"interactionStatistic":97},"DigitalDocument",{"url":83,"@type":84,"width":85,"height":86},"https://docshare.wps.com/thumbnails/a-performance-study-for-iterative-stencil-loops-on-gpus-with-ghost-zone-optimizations-performance-model-framework-to-automatically-select-ghost-zone-size/278600.png","ImageObject",442,249,{"name":88,"@type":89},"Bulrr","Person",{"url":68,"name":91,"@type":92},"DocShare","Organization","application/pdf","2026-10-19","2026-09-15",true,{"@type":98,"interactionType":99,"userInteractionCount":8},"InteractionCounter",{"@type":100},"ViewAction",{"@type":102,"mainEntity":103},"FAQPage",[104,110,114],{"name":105,"@type":106,"acceptedAnswer":107},"Why do ghost zones reduce overhead in iterative stencil loops on GPUs?","Question",{"text":108,"@type":109},"Ghost zones enlarge tiles so each processing element can generate halo regions locally for several iterations, reducing inter-PE communication and synchronization costs.","Answer",{"name":111,"@type":106,"acceptedAnswer":112},"What determines the optimal ghost zone size?",{"text":113,"@type":109},"The best ghost zone size depends on the interaction between application stencil behavior and the underlying architecture, including how synchronization affects GPU thread execution and memory state.",{"name":115,"@type":106,"acceptedAnswer":116},"How does the proposed method select the ghost zone size automatically?",{"text":117,"@type":109},"It builds an analytical performance model and uses that model to automatically choose the ghost zone size that yields the best performance, then generates appropriate code based on the selected configuration.","https://schema.org",{"og:url":78,"og:type":120,"og:title":59,"og:site_name":91,"og:description":61},"article",{"robots":122,"canonical":78},"index,follow",{"doc_id":124,"site_id":56},278600,1791749738,{"code":4,"msg":5,"data":127},{"doc_id":124,"user_id":128,"nickname":88,"user_avatar":129,"doc_module":9,"category_id":8,"category_name":11,"doc_title":59,"doc_description":61,"doc_content":130,"file_id":131,"file_url":132,"file_type":133,"file_size":134,"view_count":8,"is_deleted":4,"is_public":9,"is_downloadable":9,"audit_status":9,"page_count":135,"language":136,"language_code":57,"site_id":56,"html_lang":57,"table_of_contents":137,"faqs":138,"seo_title":139,"seo_description":61,"update_tm":140,"read_time":141},5909892115043,"https://ap-avatar.wpscdn.com/davatar_6f874abed73319feea01a86fa6f0fab8","Preprint, to appear in the International Journal of Parallel Programming (Springer); after publication, the authoritative version will be available at [http://www.springerlink.com](http://www.springerlink.com)  \nA Performance Study for Iterative Stencil Loops on GPUs with Ghost Zone Optimizations  \nJiayuan Meng 􀀁 Kevin Skadron  \nAbstract Iterative stencil loops (ISLs) are used in many applications and tiling is a well-known technique to localize their computation. When ISLs are tiled across a parallel architecture, there are usually halo regions that need to be updated and exchanged among different processing elements (PEs) . In addition, synchronization is often used to signal the completion of halo exchanges. Both communication and synchronization may incur significant overhead on parallel architectures with shared memory. This is especially true in the case of graphics processors (GPUs), which do not preserve the state of the per-core L1 storage across global synchronizations. To reduce these overheads, ghost zones can be created to replicate stencil operations, reducing communication and synchronization costs at the expense of redundantly computing some values on multiple PEs. However, the selection of the optimal ghost zone size depends on the characteristics of both the architecture and the application, and it has only been studied for message-passing systems in distributed environments. To automate this process on shared memory systems, we establish a performance model using NVIDIA’s Tesla architecture as a case study and propose a framework that uses the performance model to automatically select the ghost zone size that performs best and generate appropriate code. The modeling is validated by four diverse ISL applications, for which the predicted ghost zone configurations are able to achieve a speedup no less than 95% of the optimal speedup.  \nJiayuan Meng  \nDepartment of Computer Science University of Virginia  \nTel.: +1-434-409-9430  \n[E-mail: jm6dg@virginia.edu](E-mail: jm6dg@virginia.edu)[ ](E-mail: jm6dg@virginia.edu)Kevin Skadron  \nDepartment of Computer Science  \nUniversity of Virginia Tel.: +1-434-982-2042  \nE-mail: [skadron@cs.virginia.edu](skadron@cs.virginia.edu)  \n1 Introduction  \nIterative stencil loops (ISL) [24] are widely used in image processing, data mining, and physical simulations. ISLs usually operate on multi-dimensional arrays, with each element computed as a function of some neighboring elements. These neighbors comprise the stencil. Multiple iterations across the array are usually required to achieve convergence and/or to simulate multiple time steps. Tiling [19][29] is often used to partition the stencil loops among multiple processing elements (PEs) for parallel execution, and we refer to a workload partition as a tile in this paper. Similar tiling techniques also help localize computation to optimize cache hit rate for an individual processor [13] .  \nFig. 1 (a) Iterative stencil loops and halo regions. (b) Ghost zones help reduce inter-loop communication.  \nTiling across multiple PEs introduces a problem because stencils along the boundary of a tile must obtain values that were computed remotely on other PEs, as shown in Figure 1(a) . This means that ISL algorithms may spend considerable time stalled due to inter-loop communication and synchronization delays to exchange these halo regions. Instead of incurring this overhead after every iteration, a tile can be enlarged to include a ghost zone. This ghost zone enlarges the tile with a perimeter overlapping neighboring tiles by multiple halo regions, as shown in Figure 1(b) . The overlap allows each PE to generate its halo regions locally [32] for a number of iterations proportional to the size of the ghost zone. As Figure 1(b) demonstrates, ghost zones group loops into stages, where each stage operates on overlapping stacks of tiles, which we refer to as trapezoids. Trapezoids still produce non-overlapping data at the end, and their height refle","cbCaii01trW9OHKH","https://ap.wps.com/l/cbCaii01trW9OHKH","pdf",623731,27,"English","# Introduction\n## Iterative stencil loops and tiling\n## Halo regions and communication/synchronization overhead\n## Ghost zones and the trade-off\n## Motivation for automatic selection on GPUs\n## Contributions and case study","[{\"question\":\"Why do ghost zones reduce overhead in iterative stencil loops on GPUs?\",\"answer\":\"Ghost zones enlarge tiles so each processing element can generate halo regions locally for several iterations, reducing inter-PE communication and synchronization costs.\"},{\"question\":\"What determines the optimal ghost zone size?\",\"answer\":\"The best ghost zone size depends on the interaction between application stencil behavior and the underlying architecture, including how synchronization affects GPU thread execution and memory state.\"},{\"question\":\"How does the proposed method select the ghost zone size automatically?\",\"answer\":\"It builds an analytical performance model and uses that model to automatically choose the ghost zone size that yields the best performance, then generates appropriate code based on the selected configuration.\"}]","A Performance Study for Iterative Stencil Loops on GPUs with Ghost Zone Optimizations - Performance model framework to automatically select ghost zone size | PDF",1789510954,9]