PSADBW is not a a CISC style instruction, but a SIMD instruction. Modern RISC processor have those as well. You got similar instruction in Arms Neon instruction-set as well as the newer Scalable Vector Extensions (SVE2).
I think you are still mixing up what CISC and RISC is about. It is not like as soon as an instruction seemingly does a lot of work it is magically a CISC instruction. Vector and SIMD instructions will typically do a lot of work.
The key thing about RISC instructions is that they are fixed-length and designed in a manner that allows you to split them easily into four steps: fetch, decode, execute and writeback. That makes it easier to pipeline them.
The problem with a lot of classic CISC stuff is that they often did stuff requring a very variable number of clock cycles. Take something like the 6502. It is a very simple chips with just 3000 transistors. Yet it is very CISC like. Pretty much any arithmetic operation whether adding or subtracting requires a memory read.
RISC instructions are designed to work almost exclusively with registers to avoid stalling while waiting for memory to be read. They also have fixed width so you can easily decode many instructions in parallel and easily fit them into pipelines.
Even if these SIMD instructions look really powerful they follow the same principles: They are fixed width, work primilary with registers and can easily be split into clearly defined steps.
Also not true they are not used in modern software. They are used a lot by people working on performance sensitive software. Video codecs use them a lot. I used to work a lot with a guy who did that kind of stuff when I worked on video conferencing sofware. He would look at the assembly generated from his C code. He used Intel C compiler. If his code didn't compile into these SIMD instructions that he believed the compiler should be able to figure out he called up Intel and explained the problem to them. They would add fixes to their compiler so the could spit out those instructions when the right code patterns got reckognized.
Multimedia, graphics, video, image processing and all sorts of scientific data use these kinds of instructions a lot. You can use them directly through compiler intrinsics.
