Matti Aarnio writes:
> On Wed, Apr 17, 2002 at 02:01:42AM -0400, Robert Love wrote:
>> On Wed, 2002-04-17 at 01:34, Linus Torvalds wrote:
>>> No, it also makes it much easier to convert to/from the standard UNIX time
>>> formats (ie "struct timeval" and "struct timespec") without any surprises,
>>> because a jiffy is exactly representable in both if you have a HZ value
>>> of 100 or 100, but not if your HZ is 1024.
>>
>> Exactly - this was my issue. So what _was_ the rationale behind Alpha
>> picking 1024 (and others following)? More importantly, can we change to
>> 1000?
>
> Alpha processors don't have full division hardware, they have to
> iterate it one bit at the time. They do have a flash multiplier,
> and a barrel-shifter. Shifts take one pipeline cycle, like to
> addition and substraction. Multiply takes 6-12 depending on model,
> but division takes 64...
Division by 1000 is a UMULH followed by a right shift.
So maybe it costs you one cycle more than division by 1024 would.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Tue Apr 23 2002 - 22:00:36 EST