GEMM-Like Convolution for Deep Learning Inference on the Xilinx Versal