Step 6 - Loop management code The desired algorithm is completed, but the loop management code is still missing: the assumption that the ECX register contains the number of iterations required to fully compute the difference image is made. The first step is declaring the loop entry point: since this simple project has no startup code, the first instruction of the loop is also the first of the source list, so you should select it then choose Edit|Mark Entry Point from the menu. This simple window associates a name to an entry point: this name will be required when adding jump instructions, so choosing a meaningful one will make the code more readable. The instruction associated to an entry point is now displayed with a red arrow on the left. To remove an entry point, you should just select the instruction and choose Edit|Remove Entry Point from the menu. The final step is adding the loop epilogue: the missing instructions update the pointers to the images, adjust the loop counter and eventually exit the loop. These are all scalar instructions, so you can find them by clicking the right mouse button, grouped at the end of the list. Clicking on Scalar instructions|Adjust pointer, the following window pops up: You can choose the x86 register to update from the drop-down list, and type the desired offset in the text field on top: you can enter both positive and negative numbers, since it is up to Quexal to choose the x86 instruction that optimally adjusts the pointer. After updating the EAX, EDX and EBX pointers, the value of the loop counter is decreased by adding –1 to the register ECX. The final instruction is the conditional jump to the top of the loop: clicking on Scalar instructions|Jump opens the Jump window. You can choose the entry and the condition: in this case the loop must be iterated until the ECX register reaches zero. Clicking the OK button will close the Jump window, and step back to the main window that shows now the final source code.  By now the benefits of having an interactive development environment that guides the developer step by step should be evident: you have been building the source code by point-and-click, choosing among options specific to each instruction in a visual way. You can forget both the mnemonics of MMX/SSE instructions, given that the source list uses a simple english-like language, and MMX registers, since you can work on variables. These improvements really flatten the learning curve, and the traditional way of writing MMX/SSE code should appear unpleasant at least.