Re: Question for Herbert


Wannabee skrev:

[about the prefetch...]
The prefetch itself takes its time ...

Only if it is needed.
[...]

Yes, sure.

Worst case are RD-modify-WR instructions like XOR
(for semi-transparent mouse cursors and my old clock arrows).
Regardless how I configured caching or tried to prefetch it,
V-RAM reads are always much slower than writes.
So I now save the background and redraw it (within the V-RAM)
instead of XORing the image twice.


...cycles were measured on single 512 px lines [500:66/CPU:AGP].
( Why did you not say "cs", "cy" for cycles :) )

It would interfiere with 'code-seg' and 'carry' ? :)
but 'px' is a more common abbreviations like 'mm'.


I guess 500:66 means 500 mhz, 66 mhz chipset, AGP.
The amd here runs at 1100mhz, 266mhz fsb, and has a 200mhz memoryclock.
PIII is 500mhz, 66, agp, just like above. So how can it appear faster?
It _isnt_. (confirmed by horizontal linedrawing and bitblt, and just about
every program I run on it) - but, it has some advantage for this kind of
code. I am guessing, but the codetests seems to support my claim.

It may also depend on the access type, my measured count is on
direct V-ram writes (no caching nor paging and all in ring0).

You once got an old KESYS demo pack and IIRC it should include
'kespeed.exe' which show the clock-cycles per dot for various
drawn components (lines,circels,cakelines and text).
It needs at least a VBE2.0 card with flat 1024x768,8 or 1152x864,8.
Also the 'circles.exe' reports line and circle speed info.

As for Herberts code this is ?????????????

Depends on the machine ?

[]
Yes, and it's hard to compare different machines by cyclecount
without having the BUS-speed ratios in the calculation.

Shouldnt the AMD be faster?
assuming the timings are correctly performed what would you say ?

V-RAM access speed is not limited by the CPU but by the memory
control chips and the BUS speed.
A CPU is usually 5 to 20 times faster than the busses.
I know one board with a 2.7 GHz Celeron and a 66 MHz BUS,
guess how this performs ....

Yes, I use the API bitblt.
Looks like windoze uses this hidden functions to become faster

ok. Can you explain how to use this "hidden" function?

I have some info for old Cirrus and S3-cards, from my new
ATI-cards I know only by coincidence that the screen refresh rate
can be alterd with a single byte at BAR0+0C (ie: 04D=60HZ 080=100Hz)
The whole BAR0 image is 16 KB large and somewhere in there must
be the command register and the parameters ...
I haven't analysed its windoze drivers yet,
because this is a very boring task.

__
wolfgang


.

Relevant Pages

  • Re: determinism, Bohm, Bell
    ... > between the predictions of quantum theory and what can be inferred, ... > might be called local hidden variables, but this fact is of little ... It is responsible for the misconception that Bell proved ... > Is 'hidden determinism' viable? ...
    (sci.physics)
  • determinism, Bohm, Bell
    ... The "problem" and "difficulty" to which Bell refers above is the conflict ... between the predictions of quantum theory and what can be inferred, ... might be called local hidden variables, but this fact is of little ...
    (sci.physics)
  • Re: Damn you, FEDEX! or Nikon D40 lost in Springfield, MO blackhole.
    ... the 2 mp Mavica he had been using with a Nikon D40. ... After shopping around, he got me to order one for him. ... The shipper had it insured, but from what I have read it could take weeks to sort this crap out. ... You may get your insurance from FedEx and a couple weeks later they find it and deliver it. ...
    (alt.photography)
  • Re: python 3 constant
    ... is require Python to change to accommodate your need. ... Can you see the slight difference? ... Best regards, ...
    (comp.lang.python)
  • Re: The Sci-Fi Rejection Letter That Time Forgot
    ... nations have stockpiled arsenals of these incredible bombs and the time the story is set. ...
    (rec.arts.sf.written)