On 24 Nov 1999 13:11:01 -0500, Nieminen Juha wrote:
>Ron Parker <ron### [at] povrayorg> wrote:
>: Not really.  FLD with a 32-bit real operand is 20 clocks, as opposed to the
>: 64-bit version at 25 clocks.  FST is 44 vs. 45 clocks.  FMUL is 27-35 clocks
>: for a float, or 32-57 for a double.  FDIV is 89 or 94.  So floats are faster,
>: under ideal conditions, but not by any significant amount.  (This data is for 
>: the 386.  Other processors vary, of course.  Intel no longer seems to publish
>: this data, for the obvious reason that pipelining and other optimizations 
>: make it useless.)
>
>  As someone already said, this is not true anymore for pentium-class
>processors.
>  In a plain pentium (if I remember right) the fld takes about 6 clocks
>and fst about 3 (for both float and double) and fmul takes 1 clock (amazingly)
>in most cases.
>  In a P-II it's even faster.

Y'all seem to be missing the point.  My point was that it is not true that 
there is a conversion penalty for using 32-bit floats on Intel x86 processors, 
and that that has been true since at least the 387.  Whether later processors
made the difference smaller or nonexistent is irrelevant: even when there was a
penalty, it favored using floats over doubles rather than the other way around
as stated, and even then it was really small (5 clocks on a 33-MHz 387 is 150ns.
It takes millions of FLDs - even on a 387 - before that even comes close to 
taking measurably more or less time.)

-- 
These are my opinions.  I do NOT speak for the POV-Team.
The superpatch: http://www2.fwi.com/~parkerr/superpatch/
My other stuff: http://www2.fwi.com/~parkerr/traces.html

Post a reply to this message