{"paper":{"title":"ParamSpMM: Adaptive and Efficient Sparse Matrix-Matrix Multiplication on GPUs for GNNs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"ParamSpMM adapts SpMM on GPUs to graph inputs via a flexible data structure and ML predictor for better GNN efficiency.","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Guanhua Ye, Hongzheng Li, Lixing Zhang, Shigang Li, Yingxia Shao","submitted_at":"2026-05-15T07:38:04Z","abstract_excerpt":"Fueled by the ability to mine real-world graph data, GNN applications have experienced phenomenal growth. Sparse Matrix-Matrix Multiplication (SpMM) is a critical operator in GNNs. However, existing SpMM designs for GNNs struggle to adapt to diverse input characteristics. In this paper, we first conduct a comprehensive analysis of existing SpMM optimizations, revealing their limitations through statistical and empirical evidence. Based on this analysis, we introduce ParamSpMM, a parametric approach for highly adaptive and efficient SpMM computation in GNNs. It incorporates a new data structure"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our evaluations demonstrate that ParamSpMM outperforms Nvidia cuSPARSE with an average speedup of 1.92x, significantly enhancing GNN training efficiency.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The ML-based SpMM-decider can reliably predict optimal configurations from the crafted input features across diverse graph characteristics.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"ParamSpMM uses a Parameterized Compressed Sparse Row format and an ML-based decider to adapt SpMM optimizations for GNNs, delivering 1.92x average speedup over cuSPARSE.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"ParamSpMM adapts SpMM on GPUs to graph inputs via a flexible data structure and ML predictor for better GNN efficiency.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"1ef72aa7f8cfcdcb524961bc97319c0ecb18915187d17475352c995600ee0205"},"source":{"id":"2605.15695","kind":"arxiv","version":1},"verdict":{"id":"491632c6-e661-44e5-b607-0fadba193968","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T19:32:01.448336Z","strongest_claim":"Our evaluations demonstrate that ParamSpMM outperforms Nvidia cuSPARSE with an average speedup of 1.92x, significantly enhancing GNN training efficiency.","one_line_summary":"ParamSpMM uses a Parameterized Compressed Sparse Row format and an ML-based decider to adapt SpMM optimizations for GNNs, delivering 1.92x average speedup over cuSPARSE.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The ML-based SpMM-decider can reliably predict optimal configurations from the crafted input features across diverse graph characteristics.","pith_extraction_headline":"ParamSpMM adapts SpMM on GPUs to graph inputs via a flexible data structure and ML predictor for better GNN efficiency."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.15695/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T20:01:19.221162Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T19:41:03.222476Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T19:33:29.218344Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T17:21:56.038578Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"330779e39c764c3bd12423d104a466ca4e27f623312535f180a4da5bad3531f0"},"references":{"count":32,"sample":[{"doi":"","year":2016,"title":"Arai, J., Shiokawa, H., Yamamuro, T., Onizuka, M., Iwamura, S.: Rabbit order: Just-in-time parallel reordering for fast graph analysis. In: (IPDPS). pp. 22–31 (2016)","work_id":"a8461479-56bf-4291-890a-ad2880deca74","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2011,"title":"Bader, D.A., Meyerhenke, H., Sanders, P., Wagner, D.: 10th dimacs implementa- tion challenge-graph partitioning and graph clustering (2011)","work_id":"8d88f590-5041-4ca3-a655-03354b5de463","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2009,"title":"Waschington University in St","work_id":"a8c415f1-dac3-4b00-96b1-bbfed3ba3b6b","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Dai, G., Huang, G., Yang, S., Yu, Z., Zhang, H., Ding, Y., Xie, Y., Yang, H., Wang, Y.: Heuristic adaptability to input dynamics for spmm on gpus. p. 595–600. DAC ’22 (2022)","work_id":"fe2bd51e-5ac6-4c29-9fdc-79800ea0f10e","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Fan, R., Wang, W., Chu, X.: Fast sparse gpu kernels for accelerated training of graph neural networks. In: (IPDPS). pp. 501–511 (2023)","work_id":"5b6017c0-6221-409d-984a-750aa5c96c8c","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":32,"snapshot_sha256":"0ed6b87ff91dd5047cf9386db7b26c4bfaa3c0afa27b98073bb0ef30de88dd6f","internal_anchors":0},"formal_canon":{"evidence_count":1,"snapshot_sha256":"25666ce2ca02b5688b2a440549a4caf793bf62101f96a3f6ee3ec7976492b6aa"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}