Re: Rapid gfx display Qs

Schmidt wrote:
"Jim Mack" schrieb...

Thanks very much. It's going to take a while to decipher and absorb
the lessons in there.

Sorry, just put it together somewhat, without many comments.

Oh, that wasn't a complaint, it was an acknowlegement that it's
complex code and worth studying. My uses of DirectDraw to date have
been much more primitive.

I changed the test slightly, to load and display 100 random
1280x720 JPGs (obviously with incorrect colors, but sized OK) from
my larger pool.

That's caused by Color-swapping (changed Blue and Reg-Channels),
because the ijl15 sadly does not support a 32Bit-Destination-Format
in BGRA, instead it only offers RGBA - that's why I wrote the
Save-Routine, which already saves the jpg with changed Red and Blue
Channels. That means, you will have to reencode all your current
BMPs using my ijl-save-routine, which needs a 24Bit-DIB in normal
Windows BGR-Order as Input.

I won't have the actual images for a couple of weeks, so I'm just
using a completely random set of images that I resized to 1280x720.
Some are rather surprising when they pop up, no matter what the
colors. (-:

I have in my library some very fast ASM code to swap red/blue on
entire bitmaps. At least it will be fast enough to do a one-time
swap-and-encode on the (now I'm told 85,000) images in my final set.


The result, with no VBlankWait, is about 30-31 ms per frame on my
system: Intel Celeron 2.2GHz with NVidia GeForce 200 AGP. Waiting
for VBlank (@60hz refresh) gives 35 ms average.

That's not bad - in my timing-loop I read from Disk-Cache -
if you jump around between 100 different Files, then your 30msec
are not that much slower than my 14msec + 9msec-Head-Tracking-
time = 23msec.

BTW, which code do you use - the one with the additional
Memory-Surface - or the older version with the AGP-Surface?

I haven't tried the newer version yet, and if it's deliberately
slower, I probably won't. I'll have the advantage of a high-end video
card with lots of surface memory on board.


Ok, if you wait for VBlank, then one shouldn't force any other
screen- updates - and I made a mistake with drawing directly to the
Forms-hwnd in my example - normally in my Apps I draw to
a PictureBox.hWnd on a borderless PictureBox. In the
example as is, I use GetWindowRect against the Form (inside .Blt) -
and this includes the caption and Form-Borders of course, forcing
the system to an additional redraw. That's why you got it working
better with your borderless form.

It's my understanding that using a flipping chain with 2 backbuffers
can eliminate the need to delay updates while waiting for the VBlank
to flip. I don't have the display memory now to support that (I don't
think), but when I do, I'll give that a try too. In any case the
target system will use a full-screen borderless form, so I don't
foresee any problems with shear.


Would be interested in your results, if your new hardware
arrives - and maybe you should spend your system a 10000
RPM SCSI-Disk - this reduces your DiskAccess-Time
from 8-9 to 5-6msec - but a fast 7200Sata-Drive should
also do fine, at least the JPEG-Decoding-time scales
lineary with the Processor-speed - as said, currently
on a 1.8GHz Intel-DualCore ca. 10msec - on a 2.6GHz
Intel-DualCore this should result in ca. 7msec for the
JPG-Decoding-Task.

If I know my client, the disk will be two cheap SATA3 in RAID0. That
should be more than fast enough for loading JPGs. I'll know more in a
week or so.

--
Jim

.

Relevant Pages

  • Re: HUGE problem with A7V8X-LA and data corruption issues! (Hewlett Packard System)
    ... I create images to a seperate drive in the machine. ... Large file transfers across the network no matter what ... Ran memory tests, removed memory, used different ... for awhile than crapped again(I've tried various NICS and the ones I've ...
    (alt.comp.periphs.mainboard.asus)
  • Re: How do I put a sprite into a dynamic area ?
    ... My worries over 'protecting' my dynamic area stem from all the stuff ... They're basically for use if you wanted to use the Dynamic Area to be resizeable and behave like a cache in a similar manner to the Font Manager. ... What the Font Manager does is to cache a lot of its images into the font cache. ... When you resize it, these calls are used to allow the font manager to choose how to rearrange its memory before and/or after the memory has been reduced in size, to make sure that it keeps the most useful bits. ...
    (comp.sys.acorn.programmer)
  • Re: Canon 17 - 40mm F/4 L lens - Help PLEASE
    ... > They were JPGs! ... do you have another lens, ... > All images are un retouched exactly as they came from the camera. ... Realize that your Nikon applied in-camera sharpening to its images ...
    (rec.photo.digital.slr-systems)
  • 2.6.8.1+patches: Still a memory leak with cdrecord
    ... and it happens with *data* images ... and DMA is turned off, memory is running ... Athlon MP2800+, the leaking is also visiable, but the memory ... Aug 26 14:13:42 wirth kernel: Normal per-cpu: ...
    (Linux-Kernel)
  • Re: virtual addresses
    ... memory has grown by something pretty close to Moore's ... How may jpeg images did you have on your PC in 1995? ... But at the same time we are doubling the resolution, ... configure a 1GB ramdisc for the swap file, ...
    (comp.programming)