GLARE: Accelerating Sparse DNN Inference Kernels with Global Memory Access Reduction | IEEE Conference Publication | IEEE Xplore