I was a designer/producer in a Finnish demo group Halcyon
in the 90s. We started with Pascal, then started inlining some assembly in some inner loops to use memory aligment, registers, cache lines, unrolling, pipelining, then went full assembly (Saint
demo for '97 example, it's me touching the star and standing at the end, sprites digitized from VHS frames on Amiga I recall) and then to C with some handcoded asm inlining still (see hplus
64k intro for '98 example). It was lots of fun when you were 16-19 year old with meetings year around and allnighters before a demo party deadline. Back then about only libs "allowed" were a module player and perhaps some packer, so everybody wrote their own renderers (this was before GPUs) etc multiple times as they learned/discovered more. Profiling was simple -- we started on refresh vsync and changed the background color for important tasks during the frame, which resulted in colourful horizontal bands each indicating time spent on a task such as rotation, sorting, rendering etc, with black idle at the bottom (or difficult interpretation if things spanned multiple frames). Then we marked them with a pencil on the monitor frame, and celebrated whenever we could get them faster. Many of the lessons of memory alignment, pipelining etc could be discovered with this, as the inner loops were usually quite concentrated so you could experiment with changes easily. Then GPUs started becoming more mainstream and asset pipelines got more complex, so people who went to game industry could use their time and skills to keep up (Farbrausch for example), but from our group only one went to game development (Croaker designed Angry Birds later), rest studied engineering, business, arts etc without further involvement in this kind of creative performance critical coding.
Lately I stumbled upon Casey's posts on semantic compression and the rants and solutions reminded me how I liked that kind of pragmatic steps in coding back then. I know how effective some coders can be, and those pair-coding sessions were a delight. But when people got to industry one couldn't see them actually coding anymore, and I miss it. So now I have been keeping up with Handmade Hero streams, perhaps to find some structures and good practices to apply to my own work (I do research and prototypes on collaborative and performative web tools nowadays) and it is an added bonus that Casey has been in the game industry during this whole time, so he has a more experienced programming stance on things, but still keeping the creative spirit alive.
I find the streams very entertaining and educational, a bit similar to how the best physics lecturers explain and show their thought processes on the blackboard. This kind of visibility to actual development, not just some descriptions of it, is important generally, I think.