Difference in coding for Real time systems, Embedded or Desktop

Having programmed all three kinds, I must say that the answer depends a lot on your target platform. For example, programming for iPhone definitely qualifies as embedded, yet the skills you need are nearly identical to what you need to program for Mac. On the other hand, programming custom hardware would often require a working knowledge of in-circuit emulators, logic analyzers, and perhaps even oscilloscopes. An ability to read assembly language and relative fluency in binary operations is often assumed as well.

In general, programmers of embedded systems must be a lot more conscious about their memory use: situations when the system memory is expressed in single-digit kilobytes are not unusual. Very often you would have no operating system, so you'd need to fully understand low-level hardware interactions, interrupts, hardware timers, etc.

Real-time programming requires a different mindset altogether. The language you use becomes secondary to the way you think about resources, synchronization, schedules, memory allocations, and especially algorithms. For example, amortized linear-time algorithms do not qualify as linear-time algorithms, quick sort becomes an O(N^2) in your mind, and mainstream things such as lazy initialization become completely irrelevant. Pre-allocation becomes the order of the day, and formal verification of your most critical code becomes a requirement, not a luxury. On top of that, real-time operating systems present services that are different from what you would find on a typical desktop, so a lot of additional learning is in order.