Comment by cm3 | Hacker News Reader

Comment by cm3 | Hacker News Reader

Parent Full thread

Burroughs line of machines and Intel iAPX 432 and it's derivative i960 (which is actually in use). The project for which the Intel CPU was designed together with Siemens would have given us a safer foundation due to using this processor in an Ada based operating system in the 1980s. But then UNIX won and gave us the hegemony of C and all the avoidable security bugs associated with it. Lisp machines had similar designs. These days there are RISC-V designs that have similar features. Just like DJB has been wishing for a few extra arithmetic instructions for crypto in x86 processors, I wish for instructions that would make garbage collectors more efficient and also provide support for memory safety features.

https://en.wikipedia.org/wiki/Intel_iAPX_432

https://en.wikipedia.org/wiki/Intel_i960

https://en.wikipedia.org/wiki/Burroughs_large_systems#Tagged...

http://www.smecc.org/The%20Architecture%20%20of%20the%20Burr...

I'm sure there were similar design in the 1990s, but I don't have references ready off the top of my head like the above.

Thanks. I will definitely look into these when I have the time. I'm actually a grad student in EE but this topic interests me. Maybe I can get something published on hardware accelerated GCs... one day!

Genuinely curious, do they not teach historic designs in EE course plans? To learn from and improve, but not reinvent something half-way or worse. Also, older designs are much easier to study completely compared to the super complex logic inside your current day x86.

Personally, I would call it hardware-assisted gc if we consider hardware acceleration to be things like GPUs, crypto accelerators, etc.

An undergrad computer architecture course would typically gloss over the history of CPUs and focus either on MIPS or x86 (or both). For example, I did two courses on x86, starting from 8086 (and 8088), through Pentium, and some bits of Itanium.

You're right that starting with a simple architecture makes things much easier. 8086 for example operated in 16-bit real mode (segmented memory), and so the memory layout was trivial compared to 32-bit protected mode in x86.

I haven't taken any graduate architecture courses yet, but my assumption that they would go into more detail on the development of CPUs through history.

Reply on news.ycombinator.com