Abstract
Dynamic adaptation is a post-silicon optimization technique that adapts the hardware to workload phases. However, current adaptive approaches are oblivious to implicit phases that arise from operating on irregular data, such as sparse linear algebra operations. Implicit phases are short-lived and do not exhibit consistent behavior throughout execution. This calls for a high-accuracy, low overhead runtime mechanism for adaptation at a fine granularity. Moreover, adopting such techniques for reconfigurable manycore hardware, such as coarse-grained reconfigurable architectures (CGRAs), adds complexity due to synchronization and resource contention. We propose a lightweight machine learning-based adaptive framework called SparseAdapt. It enables low-overhead control of configuration parameters to tailor the hardware to both implicit (data-driven) and explicit (code-driven) phase changes. SparseAdapt is implemented within the runtime of a recently-proposed CGRA called Transmuter, which has been shown to deliver high performance for irregular sparse operations. SparseAdapt can adapt configuration parameters such as resource sharing, cache capacities, prefetcher aggressiveness, and dynamic voltage-frequency scaling (DVFS). Moreover, it can operate under the constraints of either (i) high energy-efficiency (maximal GFLOPS/W), or (ii) high power-performance (maximal GFLOPS3/W). We evaluate SparseAdapt with sparse matrix-matrix and matrix-vector multiplication (SpMSpM and SpMSpV) routines across a suite of uniform random, power-law and real-world matrices, in addition to end-to-end evaluation on two graph algorithms. SparseAdapt achieves similar performance on SpMSpM as the largest static configuration, with 5.3× better energy-efficiency. Furthermore, on both performance and efficiency, SparseAdapt is at most within 13% of an Oracle that adapts the configuration of each phase with global knowledge of the entire program execution. Finally, SparseAdapt is able to outperform the state-of-the-art approach for runtime reconfiguration by up to 2.9× in terms of energy-efficiency.
Highlights
Sparse linear algebra operations are key components of a plethora of modern applications, from graph analytics to scientific computing and machine learning [4, 5, 9, 10, 23, 50, 55, 64, 65, 68,69,70,71]
In order to tackle these challenges, we propose an adaptive runtime framework, SparseAdapt, that reconfigures a coarse-grained reconfigurable architectures (CGRAs) to adapt to evolving phases in sparse computation kernels
We evaluated our proposed framework first against the Baseline, Max Cfg and Best Avg configurations, followed by upper-bound studies
Summary
Sparse linear algebra operations are key components of a plethora of modern applications, from graph analytics to scientific computing and machine learning [4, 5, 9, 10, 23, 50, 55, 64, 65, 68,69,70,71]. Recent work have led to a myriad of proposals on optimizing sparse computation through fixed-function accelerator designs [3, 40, 57, 58, 72] While these demonstrate energy-efficiency improvements of the order of 100 of times over a GPU, there is an important trade-off in terms of loss of flexibility, i.e. such designs are only applicable to a few kernels. CGRAs incorporate word-granular operations to overcome the energy inefficiency of field-programmable gate arrays (FPGAs), while retaining programmability They allow for hardware reconfiguration at the granularity of the processing element (PE) array, network fabric, or the memory subsystem.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
