{"id":79324,"created":"2025-01-18T23:34:05.435628+00:00","metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:00079324","sets":["1164:2240:6352:6629"]},"path":["6629"],"owner":"10","recid":"79324","title":["GPUによる3倍精度浮動小数点演算の検討"],"pubdate":{"attribute_name":"公開日","attribute_value":"2011-11-21"},"_buckets":{"deposit":"537e70d0-9e4a-4ff7-9e43-010587ece888"},"_deposit":{"id":"79324","pid":{"type":"depid","value":"79324","revision_id":0},"owners":[10],"status":"published","created_by":10},"item_title":"GPUによる3倍精度浮動小数点演算の検討","author_link":["0"],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"GPUによる3倍精度浮動小数点演算の検討"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"高精度計算","subitem_subject_scheme":"Other"}]},"item_type_id":"4","publish_date":"2011-11-21","item_4_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"筑波大学大学院システム情報工学研究科"},{"subitem_text_value":"筑波大学システム情報系"}]},"item_language":{"attribute_name":"言語","attribute_value_mlt":[{"subitem_language":"jpn"}]},"item_publisher":{"attribute_name":"出版者","attribute_value_mlt":[{"subitem_publisher":"情報処理学会","subitem_publisher_language":"ja"}]},"publish_status":"0","weko_shared_id":-1,"item_file_price":{"attribute_name":"Billing file","attribute_type":"file","attribute_value_mlt":[{"url":{"url":"https://ipsj.ixsq.nii.ac.jp/record/79324/files/IPSJ-HPC11132023.pdf"},"date":[{"dateType":"Available","dateValue":"2013-11-21"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-HPC11132023.pdf","filesize":[{"value":"427.9 kB"}],"mimetype":"application/pdf","priceinfo":[{"tax":["include_tax"],"price":"660","billingrole":"5"},{"tax":["include_tax"],"price":"330","billingrole":"6"},{"tax":["include_tax"],"price":"0","billingrole":"14"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"f5c543cb-ba32-4c17-b25d-782d6f96d879","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2011 by the Information Processing Society of Japan"}]},"item_4_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"椋木大地"},{"creatorName":"高橋大介"}],"nameIdentifiers":[{}]}]},"item_4_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AN10463942","subitem_source_identifier_type":"NCID"}]},"item_4_textarea_12":{"attribute_name":"Notice","attribute_value_mlt":[{"subitem_textarea_value":"SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc."}]},"item_resource_type":{"attribute_name":"資源タイプ","attribute_value_mlt":[{"resourceuri":"http://purl.org/coar/resource_type/c_18gh","resourcetype":"technical report"}]},"item_4_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"近年，プロセッサの演算性能向上に対してメモリやネットワークのバンド幅不足が問題となっている．浮動小数点演算において倍精度演算で精度が不足する場合，4 倍精度演算を用いることが検討されてきたが，データアクセス量が少なくて済む 3 倍精度演算が有効となるケースが存在すると考えられる．本研究では 3 倍精度数を倍精度数と単精度数に分けて格納する Double+Single 型 3 倍精度型 （D+S 型） および D+S 型 3 倍精度演算 （D+S 型演算） を提案し，GPU による 3 倍精度の BLAS（Basic Linear Algebra Subprograms） ルーチンを実装して，その性能をTesla C2050 で評価した．D+S 型演算には Double-Double 型 4 倍精度演算 （DD 型演算） のアルゴリズムにおいて一部演算を単精度演算で行う手法を実装したが，倍精度数-単精度数の型変換が多発し D+S 型演算は DD 型演算よりも高コストとなった．そのため BLAS の入出力を D+S 型で行い，演算には DD 型演算を用いる方式を実装した．Tesla C2050 では 3 倍精度 AXPY が CUBLAS の倍精度 AXPY の約 1.57 倍の実行時間，3 倍精度 GEMV が倍精度 GEMV の約 1.69 倍の実行時間となり，それぞれ 4 倍精度ルーチンよりも高速な性能を示した．本稿では GPU における D+S 型および D+S 型演算の有効性について議論する．","subitem_description_type":"Other"}]},"item_4_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"9","bibliographic_titles":[{"bibliographic_title":"研究報告ハイパフォーマンスコンピューティング（HPC）"}],"bibliographicPageStart":"1","bibliographicIssueDates":{"bibliographicIssueDate":"2011-11-21","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"23","bibliographicVolumeNumber":"2011-HPC-132"}]},"relation_version_is_last":true,"weko_creator_id":"10"},"updated":"2025-01-21T20:17:18.582487+00:00","links":{}}