| Re: Message From RSI VP of Engineering [message #27619 is a reply to message #27491] |
Thu, 25 October 2001 07:51   |
btt
Messages: 345
Registered: December 2000
|
Senior Member |
|
|
JD
Thanks for assembling this comparison. This is the kind of meat-and-potatoes
information I have been lacking. The numerical performance issue is a weighty
one; that's not to diminish the importance of the native interface and display
speed/rendering. I don't want to ask you to compare apples and oranges, but (I
will anyway) how do you think the numerical performance of a Unix port of IDL to
OS X will compare to that we currently see in IDL on the G4 under OS 9? When we
switched from Unix IDL to Mac, I was blown away by the performance increase...
will I be blown 'back' to the slower performance?
Ben
JD Smith wrote:
> "Liam E. Gumley" wrote:
>>
>> JD Smith wrote:
>> [stuff deleted]
>>> The bigger trouble lies under the hood. IDL for MacOSX had some
>>> significant optimizations for display and within the core engine itself
>>> which are being tossed out with the bath water. The display speed will
>>> suffer, since in effect you're running through *two* levels of display
>>> (the X level, which translates drawing commands into the native display
>>> level). Any use of the much-improved OpenGL OS/hardware support will be
>>> impossible. The powerful AltiVec tuning already accomplished or planned
>>> for the OSX version will not be included.
>>>
>>> Here's a small sampling of a feature table comparison, far from
>>> complete:
>>>
>>> +=========================================================== ======+
>>> IDL feature comparison OSX Native OSX Straight Unix Port
>>> +=========================================================== ======+
>>> Interface Aqua X/Motif (server required)
>>> Display Speed Fast Slow
>>> 3D/OpenGL Optimization Yes No
>>> Altivec Vectorization Complete None, or limited
>>> Separate Core/IDE Threads Yes No
>>> Pervasive PDF Output Yes No
>>> +=========================================================== ======+
>>
>> Please correct me if I'm wrong, but I don't see why the display speed
>> would suffer.
>
> I'll put a fine point on it: running the RSI-supplied graphics_times3
> benchmark on a native OSX vs. a X11-based IDL would reveal the former to
> be much faster than the latter. This is a direct result of layering two
> display devices one atop another (which is different from your SGI, for
> which X11 is the native drawing layer).
>
> You could demonstrate this to yourself quite convincingly by running
> graphics_times3 in your IDL version running in an X-emulator under
> Windows, and on the Windows version directly. I think you'll find the
> latter to be a good deal faster.
>
> This may not be a *practical* limit for what you do, but certainly could
> impact others with more display-taxing applications.
>
> A similar story could be told for core routine performance and lack of
> Altivec tuning. Unless RSI is hiding a miracle up their sleeve, "IDL
> OSX--" will be noticeably slower in both display and computation than
> the aborted IDL OSX. Of course, we may never know the difference.
>
> JD
--
Ben Tupper
180 Mckown Point Road
West Boothbay Harbor, ME 04575
email: btupper@bigelow.org
telephone: (207) 633-9600
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
