Operator fusion, as a key performance optimization technique in the deployment of AI models, significantly improves execution efficiency and has been widely adopted in modern AI compilers. However, for cascaded reduction operations involving multiple loops with inter-loop data dependencies, such as the safe softmax followed by GEMM within attention mechanisms, existing compilers lack effective automated fusion and kernel generation capabilities. Although some works have addressed specific instances through hand-crafted fusion strategies, their solutions are limited in generality and difficult to extend to other similar structures. Given the prevalence of such computational patterns in deep learning models, there remains significant untapped potential in achieving general and automated fusion optimization. In this paper, we present a formal theoretical methodology for analyzing cascaded reductions which can fuse them into a single loop and introduce an incremental computation form. Based on this methodology, we design Red uction Fuser (RedFuser), a framework that automatically identifies supported cascaded reduction patterns and generates optimized fused kernels. Experiments show that RedFuser successfully fuses diverse workloads, achieving up to 2× to 5× speedup over state-of-the-art AI compilers and matching the performance of highly optimized hand-written kernels.

In this paper, the deep convolutional neural network (DCNN) is applied to locating and recognizing complex workpieces automatically for the vision-based sorting robot in industrial production process. Firstly, in order to obtain the location of workpieces, the pixel projection algorithm (PPA), which consists of pre-procession and pixel projection operation, is presented to eliminate uneven illumination, and locate and segment workpieces images. Then, we get the objective information and identify the object by training DCNN, which is used to recognize the rational degree and type of workpieces at a high rate of speed. Finally, experimental results prove the validity of the location-recognition algorithms for the vision-based sorting robot. The location error and recognition accuracy can be significantly improved in the experimental environment.