Many of the successful implementations of the OpenCL standard are for GPUs, namely NVDIA and AMD GPUs. In this paper, we resort to a pattern-based compressing application named Multimedia Multiscale Parser (MMP), which has been ported to OpenCL, to assess the performance and limitations of the OpenCL implementations from NVIDIA and AMD for GPUs. Specifically, we run the OpenCL version of MMP on two NVIDIA GPUs (GTX 680 and Titan Black Edition) and on two AMD GPUs (HD 7970 and R9-290x). The results show that for this application, OpenCL executions on NVIDIA GPUs are roughly twice as fast as executions performed on AMD GPUs. Also, we observe that the speedup relatively to the sequential version ranges from 5.12 (R9-290x) to 10.87 (Titan Black Edition).