> "Watching this program execute (c / ctrl-c) at adjustable speed (ctrl-t / alt-t) it becomes clearer that the process of printing an image is basically a pipeline that goes: IDCT → Y′CbCr to RGB → decimate → sharpen → block render. Thanks to Blinkenlights we can also see that the RGB conversion is going slower than it should, because code isn't benefiting from SSE register vectorization. Many other common issues concerning micro-optimization, such as register spillage, become super apparent as well."
Could anyone ELI5 how this information is obvious?
It is possible to guess that from symbol names (e.g. stbi__YCbCr_to_RGB_row function represents the second step) and currently running instructions (e.g. you don't see many SSE instructions during that step). But to be honest, Justine wrote the program in question (printimage) so that information should be obvious to its author ;-)
