The article notes the "maturity" of the support software. Indeed when I was writing manuals for SGI in the 90s, I was trying to document what seemed like extrem

The article notes the "maturity" of the support software. Indeed when I was writing manuals for SGI in the 90s, I was trying to document what seemed like extremely sophisticated compiler optimization techniques, including link-time optimization. (Due to customer demand there was also a screaming Fortran IV compiler, but it was bought in from a vendor.) I also remember how one range of the MIPS line (4000?) had the "4K page bug" where a page fault induced by the last instruction on a page caused some nasty problem. IRIX had special code to handle it.