Renwei Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Renwei Zhang has a total of 18 co-authors. Here are the top co-authors by number of shared publications:
- Jiahui Hu
Huawei Technologies Co. Ltd., China
2 shared publications - Jie Zhao
Hunan University
2 shared publications - Wang Nie
Huawei Technologies Co. Ltd., China
2 shared publications - Yanzhi Yi
Huawei Technologies Co. Ltd., China
2 shared publications - Zhen Geng
Huawei Technologies Co. Ltd., China
2 shared publications - Zhiliang Gan
Huawei Technologies Co. Ltd., China
2 shared publications
Latest Publications
- Open Access
- 30 November 2023
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Computer Systems
- Vol. 41 (1-4) , 1-45
Modeling the Interplay between Loop Tiling and Fusion in Optimizing Compilers Using Affine Relations
Loop tiling and fusion are two essential transformations in optimizing compilers to enhance the data locality of programs. Existing heuristics either perform loop tiling and fusion in a particular order, missing some of their profitable compositions, or execute ad-hoc implementations for domain-specific applications, calling for a generalized and systematic solution in optimizing compilers. In this article, we present a so-called basteln (an abbreviation for backward slicing of tiled loop nests) strategy in polyhedral compilation to better model the interplay between loop tiling and fusion. The basteln strategy first groups loop nests by preserving their parallelism/tilability and next performs rectangular/parallelogram tiling to the output groups that produce data consumed outside the considered program fragment. The memory footprints required by each tile are then computed, from which the upward exposed data are extracted to determine the tile shapes of the remaining fusion groups. Such a tiling mechanism can construct complex tile shapes imposed by the dependences between these groups, which are further merged by a post-tiling fusion algorithm for enhancing data locality without losing the parallelism/tilability of the output groups. The basteln strategy also takes into account the amount of redundant computations and the fusion of independent groups, exhibiting a general applicability. We integrate the basteln strategy into two optimizing compilers, with one a general-purpose optimizer and the other a domain-specific compiler for deploying deep learning models. The experiments are conducted on CPU, GPU, and a deep learning accelerator to demonstrate the effectiveness of the approach for a wide class of application domains, including deep learning, image processing, sparse matrix computation, and linear algebra. In particular, the basteln strategy achieves a mean speedup of 1.8× over cuBLAS/cuDNN and 1.1× over TVM on GPU when used to optimize deep learning models; it also outperforms PPCG and TVM by 11% and 20%, respectively, when generating code for the deep learning accelerator.
- 30 November 2023
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Computer Systems
- Vol. 41 (1-4) , 1-45
Open AccessModeling the Interplay between Loop Tiling and Fusion in Optimizing Compilers Using Affine Relations
Loop tiling and fusion are two essential transformations in optimizing compilers to enhance the data locality of programs. Existing heuristics either perform loop tiling and fusion in a particular order, missing some of their profitable compositions, or execute ad-hoc implementations for domain-specific applications, calling for a generalized and systematic solution in optimizing compilers. In this article, we present a so-called basteln (an abbreviation for backward slicing of tiled loop nests) strategy in polyhedral compilation to better model the interplay between loop tiling and fusion. The basteln strategy first groups loop nests by preserving their parallelism/tilability and next performs rectangular/parallelogram tiling to the output groups that produce data consumed outside the considered program fragment. The memory footprints required by each tile are then computed, from which the upward exposed data are extracted to determine the tile shapes of the remaining fusion groups. Such a tiling mechanism can construct complex tile shapes imposed by the dependences between these groups, which are further merged by a post-tiling fusion algorithm for enhancing data locality without losing the parallelism/tilability of the output groups. The basteln strategy also takes into account the amount of redundant computations and the fusion of independent groups, exhibiting a general applicability. We integrate the basteln strategy into two optimizing compilers, with one a general-purpose optimizer and the other a domain-specific compiler for deploying deep learning models. The experiments are conducted on CPU, GPU, and a deep learning accelerator to demonstrate the effectiveness of the approach for a wide class of application domains, including deep learning, image processing, sparse matrix computation, and linear algebra. In particular, the basteln strategy achieves a mean speedup of 1.8× over cuBLAS/cuDNN and 1.1× over TVM on GPU when used to optimize deep learning models; it also outperforms PPCG and TVM by 11% and 20%, respectively, when generating code for the deep learning accelerator.
- Open Access
- 8 October 2022
- Published by Association for Computing Machinery (ACM)
- p. 451-466
Parallelizing Neural Network Models Effectively on GPU by Implementing Reductions Atomically
Due to the missing of a good orchestration of loop transformations, existing optimizing compilers for deploying neural networks on GPU either parallelize reductions ineffectively or miss the fusion opportunities with other operators. Neural network models thus exhibit sub-optimal performance on GPU. We present a practical approach called Panamera for the effective parallelization of reductions in neural networks on GPU. Panamera first leverages loop coalescing to flatten the loop dimensions of reductions, converting all reduction operators into canonical forms eligible for the polyhedral model. Next, Panamera uses polyhedral transformations to reduce the data movements caused by unfused reductions and perform multi-block hardware binding not considered by many compilers. Finally, Panamera embeds a highly optimized routine implemented using GPU atomic instructions, further improving the performance of neural network models while guaranteeing the correctness of parallel reductions. The experimental results demonstrate the effectiveness of our approach: for single operators our code obtains a mean speedup of 33.7×, 3.5×, 5.4× and 9.6× over cuDNN, CUB, TVM and Ansor, for sub-graphs our approach outperforms cuDNN, TVM and Ansor by 9.5×, 2.6× and 2.7×, and for end-to-end workloads, a tensor compiler integrated with our approach outperforms them by 122.5%, 19.3% and 15.2%.
- 8 October 2022
- Published by Association for Computing Machinery (ACM)
- p. 451-466
Open AccessParallelizing Neural Network Models Effectively on GPU by Implementing Reductions Atomically
Due to the missing of a good orchestration of loop transformations, existing optimizing compilers for deploying neural networks on GPU either parallelize reductions ineffectively or miss the fusion opportunities with other operators. Neural network models thus exhibit sub-optimal performance on GPU. We present a practical approach called Panamera for the effective parallelization of reductions in neural networks on GPU. Panamera first leverages loop coalescing to flatten the loop dimensions of reductions, converting all reduction operators into canonical forms eligible for the polyhedral model. Next, Panamera uses polyhedral transformations to reduce the data movements caused by unfused reductions and perform multi-block hardware binding not considered by many compilers. Finally, Panamera embeds a highly optimized routine implemented using GPU atomic instructions, further improving the performance of neural network models while guaranteeing the correctness of parallel reductions. The experimental results demonstrate the effectiveness of our approach: for single operators our code obtains a mean speedup of 33.7×, 3.5×, 5.4× and 9.6× over cuDNN, CUB, TVM and Ansor, for sub-graphs our approach outperforms cuDNN, TVM and Ansor by 9.5×, 2.6× and 2.7×, and for end-to-end workloads, a tensor compiler integrated with our approach outperforms them by 122.5%, 19.3% and 15.2%.
- Open Access
- 9 January 2017
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 5, 771-780
Detection of Composite Operation in Model Management
Model management systems become increasingly critical in model-driven engineering. One of the main tasks of these systems is to record the operations performed on model elements. While most systems support the record of primitive model change operations, complex composite model change operations are neglected, which may result in the lack of understandability. In this paper, we propose an approach to capture model transformation from primitive operations to composite ones. First, based on the low-level operation, we define some general high-level operations with hierarchical structures. Then, a matching algorithm is designed to compare primitive operations with the hierarchical structures from the bottom up. If matching successfully, the primitive operations would be lifted to a composited operation. The algorithm is iterative and ensures that all operations are lifted. The evaluation results on real-world cases show that both precision and recall of composite operation detection are improved when compared with the EMF Modeling Operations (EMO) and Complex Change Detection Engine (CCDE) algorithms.
- 9 January 2017
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 5, 771-780
Open AccessDetection of Composite Operation in Model Management
Model management systems become increasingly critical in model-driven engineering. One of the main tasks of these systems is to record the operations performed on model elements. While most systems support the record of primitive model change operations, complex composite model change operations are neglected, which may result in the lack of understandability. In this paper, we propose an approach to capture model transformation from primitive operations to composite ones. First, based on the low-level operation, we define some general high-level operations with hierarchical structures. Then, a matching algorithm is designed to compare primitive operations with the hierarchical structures from the bottom up. If matching successfully, the primitive operations would be lifted to a composited operation. The algorithm is iterative and ensures that all operations are lifted. The evaluation results on real-world cases show that both precision and recall of composite operation detection are improved when compared with the EMF Modeling Operations (EMO) and Complex Change Detection Engine (CCDE) algorithms.