{"627068":{"#nid":"627068","#data":{"type":"event","title":"PACE Seminar","body":[{"value":"\u003Cp\u003E\u003Cstrong\u003EPresenter:\u003C\/strong\u003E\u0026nbsp;Dr.\u0026nbsp;Christopher Stone\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003EDate and Time:\u003C\/strong\u003E\u0026nbsp;Thursday,\u0026nbsp;October\u0026nbsp;3\u003Csup\u003Eth\u003C\/sup\u003E\u0026nbsp;11:00am \u0026ndash; 12:00pm\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003ELocation:\u003C\/strong\u003E\u0026nbsp;CODA Room 113\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003EHost:\u0026nbsp;\u003C\/strong\u003EPACE\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003ETitle:\u0026nbsp;\u003C\/strong\u003EAccelerating non-linear multiphysics models with SIMD vectorization:\u0026nbsp;\u0026nbsp;what to try when the compiler won\u0026rsquo;t vectorize your code automatically\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003E\u003Cstrong\u003EAbstract:\u0026nbsp;\u003C\/strong\u003E\u003C\/strong\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003ESingle-instruction, multiple data (SIMD) parallelism plays a significant role on modern high-performance computing (HPC) systems. For example, 8 to 16 double-precision operations can be completed each clock cycle on each core of an Intel Xeon Skylake CPU. Explicitly programming for SIMD (i.e., vector) parallelism is complex and challenging and so is mostly left to auto-vectorizing compilers. However, due to the potential 8x performance boost available from SIMD, explicit vectorization may be the only recourse when the compiler refuses to vectorize for performance-critical kernels in scientific applications. In this talk, we shall examine several explicit vectorization methods and apply them to multiphysics kernels commonly found in high-fidelity combustion simulations. We\u0026rsquo;ll also look at how the SIMD principles apply to GPU computing and will examine the performance of SIMD kernels on various hardware platforms.\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003EBiography\u003C\/strong\u003E\u003Cstrong\u003E:\u003C\/strong\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003EDr. Stone is a high-performance computing\u0026nbsp;(HPC).\u0026nbsp;consultant with the Department of Defense HPCMP PETTT Program. He recievd his PhD in Aerospace Engineering from Georgia Institute of Technology in 2003, where he studied reacting flows (e.g., combustion) using computational fluid dynamics (CFD) and high-performance computing. He has sixteen years of postgraduate, professional research and development experience in computational science and HPC. His experiences encompass a broad spectrum of application domains, technologies, and parallel computing methodologies. Additionally, he has several years of teaching experience at the graduate and undergraduate levels in mechanical engineering and computer science.\u0026nbsp;\u003C\/p\u003E\r\n","summary":null,"format":"limited_html"}],"field_subtitle":"","field_summary":[{"value":"\u003Cp\u003E\u003Cstrong\u003EPresenter:\u003C\/strong\u003E\u0026nbsp;Dr.\u0026nbsp;Christopher Stone\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003EDate and Time:\u003C\/strong\u003E\u0026nbsp;Thursday,\u0026nbsp;October\u0026nbsp;3\u003Csup\u003Eth\u003C\/sup\u003E\u0026nbsp;11:00am \u0026ndash; 12:00pm\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003ELocation:\u003C\/strong\u003E\u0026nbsp;CODA Room 113\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003EHost:\u0026nbsp;\u003C\/strong\u003EPACE\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003ETitle:\u0026nbsp;\u003C\/strong\u003EAccelerating non-linear multiphysics models with SIMD vectorization:\u0026nbsp;\u0026nbsp;what to try when the compiler won\u0026rsquo;t vectorize your code automatically\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003E\u003Cstrong\u003EAbstract:\u0026nbsp;\u003C\/strong\u003E\u003C\/strong\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003ESingle-instruction, multiple data (SIMD) parallelism plays a significant role on modern high-performance computing (HPC) systems. For example, 8 to 16 double-precision operations can be completed each clock cycle on each core of an Intel Xeon Skylake CPU. Explicitly programming for SIMD (i.e., vector) parallelism is complex and challenging and so is mostly left to auto-vectorizing compilers. However, due to the potential 8x performance boost available from SIMD, explicit vectorization may be the only recourse when the compiler refuses to vectorize for performance-critical kernels in scientific applications. In this talk, we shall examine several explicit vectorization methods and apply them to multiphysics kernels commonly found in high-fidelity combustion simulations. We\u0026rsquo;ll also look at how the SIMD principles apply to GPU computing and will examine the performance of SIMD kernels on various hardware platforms.\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003EBiography\u003C\/strong\u003E\u003Cstrong\u003E:\u003C\/strong\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003EDr. Stone is a high-performance computing\u0026nbsp;(HPC).\u0026nbsp;consultant with the Department of Defense HPCMP PETTT Program. He recievd his PhD in Aerospace Engineering from Georgia Institute of Technology in 2003, where he studied reacting flows (e.g., combustion) using computational fluid dynamics (CFD) and high-performance computing. He has sixteen years of postgraduate, professional research and development experience in computational science and HPC. His experiences encompass a broad spectrum of application domains, technologies, and parallel computing methodologies. Additionally, he has several years of teaching experience at the graduate and undergraduate levels in mechanical engineering and computer science.\u0026nbsp;\u003C\/p\u003E\r\n","format":"limited_html"}],"field_summary_sentence":[{"value":"Accelerating non-linear multiphysics models with SIMD vectorization:  what to try when the compiler won\u2019t vectorize your code automatically"}],"uid":"35120","created_gmt":"2019-10-02 21:54:58","changed_gmt":"2019-10-02 21:57:49","author":"mweiner3","boilerplate_text":"","field_publication":"","field_article_url":"","field_event_time":{"event_time_start":"2019-10-03T12:00:00-04:00","event_time_end":"2019-10-03T13:00:00-04:00","event_time_end_last":"2019-10-03T13:00:00-04:00","gmt_time_start":"2019-10-03 16:00:00","gmt_time_end":"2019-10-03 17:00:00","gmt_time_end_last":"2019-10-03 17:00:00","rrule":null,"timezone":"America\/New_York"},"extras":[],"groups":[{"id":"337231","name":"Georgia Tech High Performance Computing (PACE)"}],"categories":[],"keywords":[{"id":"178925","name":"PACE Special Event"}],"core_research_areas":[],"news_room_topics":[],"event_categories":[{"id":"1795","name":"Seminar\/Lecture\/Colloquium"}],"invited_audience":[{"id":"78761","name":"Faculty\/Staff"},{"id":"177814","name":"Postdoc"},{"id":"174045","name":"Graduate students"},{"id":"78751","name":"Undergraduate students"}],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[{"value":"\u003Cp\u003E\u003Cstrong\u003EOrganizer:\u003C\/strong\u003E PACE\/ART\/OIT\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003EContact:\u0026nbsp;\u003C\/strong\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003ENuyun Zhang, nuyun@gatech.edu\u003C\/p\u003E\r\n","format":"limited_html"}],"email":[],"slides":[],"orientation":[],"userdata":""}}}