Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

I think that statement suffers from hindsight bias: x86 is great now because of enormous leaps in compiler technology to work. Yes, compilers do need to evolve with CPUs, they don't just grow on trees!

Put another way: given a choice between stuffing an electric engine in a Ford Fiesta, and a Tesla, Itanium could have been the Tesla in the hands of another company, but the market went after the Fiesta retrofit because it was here-and-now, not coming-soon.

EDIT: changed from "you" to "that statement"; added analogy.



I don't believe that's historically accurate at all. x86 compiler performance has been good enough for a long, long time. Yes, newer versions are clearly a lot better than ones from the mid 90s, but the performance improvements would be described in terms of percentages, not orders of magnitude.

IA-64 required a complete rethinking of compiler design to solve NP-hard problems on a large scale just to get passable performance. Translating C to ASM that keeps execution units busy is radically different on IA-64 than pretty much anything else.


> x86 compiler performance has been good enough for a long, long time.

> NP-hard problems on a large scale just to get passable performance.

First, it is hard to refute a claim, or find meaning in it, when you start using terms like "passable" and "good enough".

Second, I think this overstates the need for optimal scheduling because Itanium SPEC performance was "passable" out of the gate.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: