What ever happened to Tandem and NonStop OS 2016

OTOH, we had a situation where the floating point context switch wasn't complete due to a poorly documented fpu register, leading to a case where a particular sequence of task switches and arithmetic events made a branch on fp state fail when it should have succeeded. It was big fun tracking that one down. I hope the cpu engineers and tech writers felt the pins we stuck in their effigies... but thats an example of the OS weenies having to get their physical model right where the compiler weenies didn't have such a hard time- they "only" had to use the registers to achieve the math.

Which isn't to say I don't think the compiler weenies have it tough. Having looked thru quite enough object dumps I can see its a nasty subtle job at best and over the top drive-you-crazy when things are tougher.

The compiler weenies doing our compilers didn't have much of any cpu state issues, just ABI stuff and no cache-context-tlb-prediction things to worry about beyond whatever brokenness exists in the cpu. The OS and library weenies had to make it all go, which to your point, is perhaps an example of a fairly old-fashioned architecture.

It'd be fun to mess with that kind of thing at some point- none to likely in the forseeable future given the cpu's we get to work with lag behind industry by 5-10 years or so- we're happy with PowerPC 750's these days.

Gregm