Automatically Identifying and Creating Accelerators Directly from C Code

Let’s say that you have been tasked to ensure that your company has an H.264 solution that supports high-definition video decoding at 30 frames per second. You are not a video expert. What do you do? You could get on the Internet and perform a Web search for H.264; before you know it, you’ll have the source code and be on your way. You managed to compile the code and get it running on the target, but it decodes at a whopping two frames per second. Now what? After sifting through pages and pages of profiling data, you find some hotspots, but you are not sure which parts to focus on to maximize the acceleration and you do not have enough time to try to optimize them all. Many of us have found ourselves in this situation at one time or another. Maybe you have even delivered a solution, but not without a lot of sweat and tears.