@spun_off this seems slightly interesting https://godbolt.org/g/RUk2AR
-
-
Replying to @johnregehr
It is interesting. Is ICC documented as having fast-math on by default?
2 replies 0 retweets 0 likes -
Replying to @volatile_void @johnregehr
ICC is famous for being buggy junk that's tuned for benchmarks...
1 reply 0 retweets 1 like -
Replying to @RichFelker @johnregehr
These should be self-apparent: GPU vendors should not be writing drivers. CPU vendors should not be writing compilers.
1 reply 0 retweets 2 likes -
Replying to @RichFelker @johnregehr
I agree in theory. But, in practice, the CPU vendor needs support for new features before silicon exists.
1 reply 0 retweets 0 likes -
Replying to @alastair_d_reid @RichFelker
meeting Rich's goal would require very early openness with specs, a hard sell I suspect
1 reply 0 retweets 0 likes -
Replying to @johnregehr @RichFelker
At ARM, we start on the compiler 2 years before the architecture is finalised and upstream 2 yrs before Si
1 reply 0 retweets 2 likes -
Can't really see how you can feasibly do architecture without compiler...
2 replies 0 retweets 0 likes -
Replying to @erincandescent @oshepherd and
You can do a backend for an existing compiler.
2 replies 0 retweets 0 likes
The big problem is making bad high-level choices (like buggy math by default).
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.