Templatized Fused Vector Floating-Point Dot Product for High-Level Synthesis
Machine-learning accelerators rely on floating-point matrix and vector multiplication kernels. To reduce their cost, customized many-term fused architectures are preferred, which improve the latency, power, and area of the designs. In this work, we design a parameterized fused many-term floating-poi...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2022-10-01
|
| Series: | Journal of Low Power Electronics and Applications |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2079-9268/12/4/56 |
| _version_ | 1797456919741333504 |
|---|---|
| author | Dionysios Filippas Chrysostomos Nicopoulos Giorgos Dimitrakopoulos |
| author_facet | Dionysios Filippas Chrysostomos Nicopoulos Giorgos Dimitrakopoulos |
| author_sort | Dionysios Filippas |
| collection | DOAJ |
| description | Machine-learning accelerators rely on floating-point matrix and vector multiplication kernels. To reduce their cost, customized many-term fused architectures are preferred, which improve the latency, power, and area of the designs. In this work, we design a parameterized fused many-term floating-point dot product architecture that is ready for high-level synthesis. In this way, we can exploit the efficiency offered by a well-structured fused dot-product architecture and the freedom offered by high-level synthesis in tuning the design’s pipeline to the selected floating-point format and architectural constraints. When compared with optimized dot-product units implemented directly in RTL, the proposed design offers lower-latency implementations under the same clock frequency with marginal area savings. This result holds for a variety of floating-point formats, including standard and reduced-precision representations. |
| first_indexed | 2024-03-09T16:14:46Z |
| format | Article |
| id | doaj.art-4380dfeaa95a42c6bb88960b53e04301 |
| institution | Directory Open Access Journal |
| issn | 2079-9268 |
| language | English |
| last_indexed | 2024-03-09T16:14:46Z |
| publishDate | 2022-10-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Journal of Low Power Electronics and Applications |
| spelling | doaj.art-4380dfeaa95a42c6bb88960b53e043012023-11-24T15:53:26ZengMDPI AGJournal of Low Power Electronics and Applications2079-92682022-10-011245610.3390/jlpea12040056Templatized Fused Vector Floating-Point Dot Product for High-Level SynthesisDionysios Filippas0Chrysostomos Nicopoulos1Giorgos Dimitrakopoulos2Electrical and Computer Engineering, Democritus University of Thrace, 67100 Xanthi, GreeceElectrical and Computer Engineering, University of Cyprus, 1678 Nicosia, CyprusElectrical and Computer Engineering, Democritus University of Thrace, 67100 Xanthi, GreeceMachine-learning accelerators rely on floating-point matrix and vector multiplication kernels. To reduce their cost, customized many-term fused architectures are preferred, which improve the latency, power, and area of the designs. In this work, we design a parameterized fused many-term floating-point dot product architecture that is ready for high-level synthesis. In this way, we can exploit the efficiency offered by a well-structured fused dot-product architecture and the freedom offered by high-level synthesis in tuning the design’s pipeline to the selected floating-point format and architectural constraints. When compared with optimized dot-product units implemented directly in RTL, the proposed design offers lower-latency implementations under the same clock frequency with marginal area savings. This result holds for a variety of floating-point formats, including standard and reduced-precision representations.https://www.mdpi.com/2079-9268/12/4/56floating point arithmeticvector dot producthigh level synthesis |
| spellingShingle | Dionysios Filippas Chrysostomos Nicopoulos Giorgos Dimitrakopoulos Templatized Fused Vector Floating-Point Dot Product for High-Level Synthesis Journal of Low Power Electronics and Applications floating point arithmetic vector dot product high level synthesis |
| title | Templatized Fused Vector Floating-Point Dot Product for High-Level Synthesis |
| title_full | Templatized Fused Vector Floating-Point Dot Product for High-Level Synthesis |
| title_fullStr | Templatized Fused Vector Floating-Point Dot Product for High-Level Synthesis |
| title_full_unstemmed | Templatized Fused Vector Floating-Point Dot Product for High-Level Synthesis |
| title_short | Templatized Fused Vector Floating-Point Dot Product for High-Level Synthesis |
| title_sort | templatized fused vector floating point dot product for high level synthesis |
| topic | floating point arithmetic vector dot product high level synthesis |
| url | https://www.mdpi.com/2079-9268/12/4/56 |
| work_keys_str_mv | AT dionysiosfilippas templatizedfusedvectorfloatingpointdotproductforhighlevelsynthesis AT chrysostomosnicopoulos templatizedfusedvectorfloatingpointdotproductforhighlevelsynthesis AT giorgosdimitrakopoulos templatizedfusedvectorfloatingpointdotproductforhighlevelsynthesis |