Matches in SemOpenAlex for { <https://semopenalex.org/work/W4312122010> ?p ?o ?g. }
Showing items 1 to 61 of
61
with 100 items per page.
- W4312122010 abstract "CPU-based inference can be an alternative to off-chip accelerators, and vector architectures are a promising option due to their efficiency. However, the large design space of convolutional algorithms and hardware implementations makes it challenging to select the best options. This paper presents ongoing research into co-designing vector architectures for CPU-based CNN inference, focusing on the im2col+GEMM and Winograd kernels. Using the Gem5 simulator, we examine the impact of various hardware microarchitectural features on RISC-V Vector and ARM-SVE ISAs. We also study the impact of several BLIS-like algorithmic optimizations on im2col+GEMM. Our co-design study shows that longer vector lengths and larger caches can improve performance by 5x with our optimized CNN kernels, compared to a vector length of 512-bit and 1MB of L2 cache. For Winograd, we present a novel approach of inter-tile parallelization that exploits longer vector lengths and offers high memory reuse, resulting in up to 2.4x performance improvement for non-strided convolutional layers with 3x3 kernel size. Our study also shows that Winograd requires smaller cache sizes compared to im2col+GEMM." @default.
- W4312122010 created "2023-01-04" @default.
- W4312122010 creator A5020534755 @default.
- W4312122010 creator A5057776164 @default.
- W4312122010 creator A5074518408 @default.
- W4312122010 date "2022-12-22" @default.
- W4312122010 modified "2023-10-16" @default.
- W4312122010 title "Accelerating CNN inference on long vector architectures via co-design" @default.
- W4312122010 doi "https://doi.org/10.48550/arxiv.2212.11574" @default.
- W4312122010 hasPublicationYear "2022" @default.
- W4312122010 type Work @default.
- W4312122010 citedByCount "0" @default.
- W4312122010 crossrefType "posted-content" @default.
- W4312122010 hasAuthorship W4312122010A5020534755 @default.
- W4312122010 hasAuthorship W4312122010A5057776164 @default.
- W4312122010 hasAuthorship W4312122010A5074518408 @default.
- W4312122010 hasBestOaLocation W43121220101 @default.
- W4312122010 hasConcept C113775141 @default.
- W4312122010 hasConcept C11413529 @default.
- W4312122010 hasConcept C114614502 @default.
- W4312122010 hasConcept C115537543 @default.
- W4312122010 hasConcept C118524514 @default.
- W4312122010 hasConcept C149635348 @default.
- W4312122010 hasConcept C154945302 @default.
- W4312122010 hasConcept C173608175 @default.
- W4312122010 hasConcept C189783530 @default.
- W4312122010 hasConcept C2776214188 @default.
- W4312122010 hasConcept C2776221188 @default.
- W4312122010 hasConcept C33923547 @default.
- W4312122010 hasConcept C41008148 @default.
- W4312122010 hasConcept C74193536 @default.
- W4312122010 hasConceptScore W4312122010C113775141 @default.
- W4312122010 hasConceptScore W4312122010C11413529 @default.
- W4312122010 hasConceptScore W4312122010C114614502 @default.
- W4312122010 hasConceptScore W4312122010C115537543 @default.
- W4312122010 hasConceptScore W4312122010C118524514 @default.
- W4312122010 hasConceptScore W4312122010C149635348 @default.
- W4312122010 hasConceptScore W4312122010C154945302 @default.
- W4312122010 hasConceptScore W4312122010C173608175 @default.
- W4312122010 hasConceptScore W4312122010C189783530 @default.
- W4312122010 hasConceptScore W4312122010C2776214188 @default.
- W4312122010 hasConceptScore W4312122010C2776221188 @default.
- W4312122010 hasConceptScore W4312122010C33923547 @default.
- W4312122010 hasConceptScore W4312122010C41008148 @default.
- W4312122010 hasConceptScore W4312122010C74193536 @default.
- W4312122010 hasLocation W43121220101 @default.
- W4312122010 hasOpenAccess W4312122010 @default.
- W4312122010 hasPrimaryLocation W43121220101 @default.
- W4312122010 hasRelatedWork W1484089092 @default.
- W4312122010 hasRelatedWork W1558545464 @default.
- W4312122010 hasRelatedWork W1571368810 @default.
- W4312122010 hasRelatedWork W1598062406 @default.
- W4312122010 hasRelatedWork W1616582327 @default.
- W4312122010 hasRelatedWork W1784146144 @default.
- W4312122010 hasRelatedWork W1796231360 @default.
- W4312122010 hasRelatedWork W1981926925 @default.
- W4312122010 hasRelatedWork W2379400621 @default.
- W4312122010 hasRelatedWork W2506225776 @default.
- W4312122010 isParatext "false" @default.
- W4312122010 isRetracted "false" @default.
- W4312122010 workType "article" @default.