Forums before death by AOL, social media and spammers... "We can't have nice things"
|    comp.arch    |    Apparently more than just beeps & boops    |    131,241 messages    |
[   << oldest   |   < older   |   list   |   newer >   |   newest >>   ]
|    Message 130,545 of 131,241    |
|    Anton Ertl to John Dallman    |
|    Re: Lessons from the ARM Architecture    |
|    17 Dec 25 16:21:36    |
      From: anton@mips.complang.tuwien.ac.at              jgd@cix.co.uk (John Dallman) writes:       >] * "Trap and Emulate” is an illusion of compatibility"       >] * Performance differential is too great for most applications              I disagree. Trap-and-emulate may be too slow for a feature that you       want programmers to use in hot paths on current CPUs, but there are       other cases. In particular, for a feature that cannot be implemented       properly yet, if you provide it as a trap-and-emulated instruction in       the current generation, and a faster implementation of the instruction       in a future implementation of the architecture, programmers will be       much less reluctant to use that instruction when its implementation is       fast on the dominant implementation of the day than if it just       produces a SIGILL or somesuch on older chips (or new, but       feature-reduced chips). Intel's marketing does not understand that,       that's why they are selling feature-reduced versions of chips that       have AVX and AVX-512 in hardware.              - anton       --       'Anyone trying for "industrial quality" ISA should avoid undefined behavior.'        Mitch Alsup, |
[   << oldest   |   < older   |   list   |   newer >   |   newest >>   ]
(c) 1994, bbs@darkrealms.ca