Asplos 24 Lightning Talks Session 4a Aces Accelerating Sparse
![](https://i0.wp.com/sicepat.com/images/image-not-found.png)
Asplos'24: international conference on architectural support for programming languages and operating systems lightning talks session 4a: acceleratorspaper. 12:00 pdt – 13:00 pdt: session 1. 1a: synthesis for architectures. (location: grande a b) session chair: adrian sampson (cornell university) explainable port mapping inference with sparse performance counters for amd’s zen architectures. fabian ritter and sebastian hack (saarland university) paper .
![](https://i0.wp.com/sicepat.com/images/image-not-found.png)
Sparse patterns, resulting in inconsistent performance across different sparse matrices. for instance, in inp, the sparse pat tern critically influences the index intersection between the two input matrices, affecting input data fetching efficiency. in outp, the sparse pattern determines the size of the result. Aces adaptive execution flow detect sparse patterns select condensing degrees •detect sparse patterns •indicate sparse pattern changes by row length variations (spada [asplos’23]) •adjacent rows with similar distributions of non zero elements tend to have a stable row length (number of non zero elements). To address these issues, we introduce aces, a novel spmm accelerator in this study. first, aces features an adaptive execution flow that dynamically adjusts to diverse sparse patterns. the adaptive execution flow balances parallel computing efficiency and data reuse. •aces outperforms all other spmm accelerators •25.5× over sigma, 8.9× over sparch, and 2.1× over spada xiaoyang lu*1, boyu long*2,3, xiaoming chen2, yinhe han2, xian he sun1 1illinois institute of technology,2chinese academy of sciences, 3university of chinese academy of sciences aces: accelerating sparse matrix multiplication with adaptive.
![](https://i0.wp.com/sicepat.com/images/image-not-found.png)
To address these issues, we introduce aces, a novel spmm accelerator in this study. first, aces features an adaptive execution flow that dynamically adjusts to diverse sparse patterns. the adaptive execution flow balances parallel computing efficiency and data reuse. •aces outperforms all other spmm accelerators •25.5× over sigma, 8.9× over sparch, and 2.1× over spada xiaoyang lu*1, boyu long*2,3, xiaoming chen2, yinhe han2, xian he sun1 1illinois institute of technology,2chinese academy of sciences, 3university of chinese academy of sciences aces: accelerating sparse matrix multiplication with adaptive. The asplos'24 program consists of 193 papers: the 170 papers we accepted thus far and, in addition, 23 major revisions from the fall cycle of asplos'23, which were re reviewed and accepted. the full details are available in the pdf of the front matter. Aces, a novel spmm accelerator that features an adaptive execution flow that dynamically adjusts to diverse sparse patterns and incorporates locality concurrency co optimizations within the global cache, significantly outperforms existing solutions and marks a substantial advancement in spmm acceleration. sparse matrix matrix multiplication (spmm) is a critical computational kernel in numerous.
![](https://i0.wp.com/sicepat.com/images/image-not-found.png)
The asplos'24 program consists of 193 papers: the 170 papers we accepted thus far and, in addition, 23 major revisions from the fall cycle of asplos'23, which were re reviewed and accepted. the full details are available in the pdf of the front matter. Aces, a novel spmm accelerator that features an adaptive execution flow that dynamically adjusts to diverse sparse patterns and incorporates locality concurrency co optimizations within the global cache, significantly outperforms existing solutions and marks a substantial advancement in spmm acceleration. sparse matrix matrix multiplication (spmm) is a critical computational kernel in numerous.
Comments are closed.