For the typical graph algorithms encountered most frequently in practice (such as those introduced in typical entry-level algorithms courses: graph searching\/traversals, shortest paths problems, strongly connected components and minimum spanning trees) we want to consider practical non-sequential platforms such as the emergence of cost effective General-Purpose computation on Graphics Processing Units (GPGPU). In this paper we provide two simple design techniques that allow a nonspecialist computer scientist to harness the power of their GPUs as parallel compute devices. These two natural ideas are (a) using a host CPU script to synchronize a distributed view of a graph algorithm where each node of the input graph is associated with a unique processing thread ID and (b) using GPU atomic operations to synchronize a single kernel launch where a set of threads, upper-bounded by at most the number of streaming processing units available, continuously stay active and time-slice the total workload until the algorithm completes. We give concrete comparative implementations of both of these approaches for the simple problem of exploring a graph using breadth-first search. Finally we conclude that OpenCL, in addition to CUDA, is a natural tool for modern graph algorithm designers, especially those who are not experts of GPU hardware architecture, to develop real-world usable graph applications.<\/p>\n","protected":false},"excerpt":{"rendered":"
For the typical graph algorithms encountered most frequently in practice (such as those introduced in typical entry-level algorithms courses: graph searching\/traversals, shortest paths problems, strongly connected components and minimum spanning trees) we want to consider practical non-sequential platforms such as the emergence of cost effective General-Purpose computation on Graphics Processing Units (GPGPU). In this paper […]<\/p>\n","protected":false},"author":351,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_publicize_message":"","jetpack_publicize_feature_enabled":true,"jetpack_social_post_already_shared":false,"jetpack_social_options":{"image_generator_settings":{"template":"highway","default_image_id":0,"font":"","enabled":false},"version":2}},"categories":[36,11,89,90,3],"tags":[1787,1782,14,158,20,1793,378],"class_list":["post-5998","post","type-post","status-publish","format-standard","hentry","category-algorithms","category-computer-science","category-nvidia-cuda","category-opencl","category-paper","tag-algorithms","tag-computer-science","tag-cuda","tag-graph-theory","tag-nvidia","tag-opencl","tag-tesla-c2050"],"views":2654,"jetpack_publicize_connections":[],"jetpack_featured_media_url":"","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/hgpu.org\/index.php?rest_route=\/wp\/v2\/posts\/5998","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/hgpu.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/hgpu.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/hgpu.org\/index.php?rest_route=\/wp\/v2\/users\/351"}],"replies":[{"embeddable":true,"href":"https:\/\/hgpu.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=5998"}],"version-history":[{"count":0,"href":"https:\/\/hgpu.org\/index.php?rest_route=\/wp\/v2\/posts\/5998\/revisions"}],"wp:attachment":[{"href":"https:\/\/hgpu.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5998"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/hgpu.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5998"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/hgpu.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5998"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}