<html> <head> <title>Yadex advanced uses</title> </head> <body> <div align="center"> <img src="logo_small.png" alt="Fancy logo"> <br>Yadex 1.5.2 (2001-06-30) <h1>Advanced user's guide</h1> <table><tr><td><td width="50%" align="center"> Under a silly name, this document discusses some issues that only "advanced" users might want to know about. If you're a new user, you can probably skip this document for the moment. <td></td></table> </div> <br> <br> <br> <h2>Scalability</h2> <h3>Level size</h3> Very large levels, that is levels with large numbers of things, vertices, linedefs, sidedefs or sectors, affect both memory consumption and responsiveness. <ul> <p><li>In theory, the amount of memory needed by Yadex to hold a level open for editing is exactly equal to the cumulated size of the <code>THINGS</code>, <code>VERTEXES</code>, <code>LINEDEFS</code>, <code>SIDEDEFS</code> and <code>SECTORS</code> lumps of said level. In practice, it probably will be a little higher because of the overhead in the memory allocation system. <p>In addition, each object in the selection consumes about 8 bytes. <p>Finally, during certain operations, Yadex creates bit vectors. Fortunately, these consume very little memory, about 1 bit for each object. If you had a level with 32,767 linedefs, creating a bit vector for the linedefs would require about 4 kB, which is not much, especially by comparison with the 447 kB that would be needed to store the linedefs themselves. <p><li>The worst problem that large levels cause is not memory but CPU cycles. Yadex spends a lot of its time browsing through the entire level data. For example, if you are in linedefs mode, every time you move the pointer, Yadex scans the entire linedefs table to see which linedef is under the pointer. As the number of objects increases, the responsiveness is bound to decrease. <p>There are several places in DEU where there were algorithms in O(2). For example, when you dragged a vertex, the piece of code that checked whether you dragged it on top of another vertex made about 1/2 N<sup>2</sup> tests, which is unreasonably high past a few hundred vertices. I've tried to replace such algorithms by ones in N×log(N) but some of them might remain. <p>The drag-and-drop code is awfully slow when dragging a large number of objects. I'm still not sure why. <p>The worst offender by far in terms of CPU cycles is the display itself, for obvious reasons. Zooming in will make it better since Yadex avoids drawing objects that are off-screen. Using a pixmap (as is the default) will make it <em>much</em> worse as (at least on my system) drawing to a pixmap is way slower than drawing to a window. With a large number of objects in sight, the difference can be very noticeable. </ul> <h3>Wad size</h3> <p>Since the directory of all loaded wads is kept in memory all the time, in pathological cases this could cause trouble. However, I don't think the problem could arise in practice as even the biggest wads have directories no larger than about 100 kB. <p>Wads with a large number of entries tend to cause delays because loading them generates several screenfuls of messages. I've made those load-time messages much more compact than they used to be but there's still enough left to be annoying with certain megawads. And, no, there's no way to suppress them (I mean the messages <g>). <p>If and when in-place saving is ever implemented, there'll be something to say about it here. <h3>Window size</h3> In the present implementation, the window is probably the worst memory and CPU hog. Unless the <code>-P</code> option is used, a pixmap is created for the window. For large window sizes, that pixmap can be quite large. For example, for a 1024×768 window on a 16-bit display, it would be about 1.5 MB large. <p>On the CPU side, every times the screen is redrawn completely (which happens quite often, particularly when dragging objects), the pixmap is filled with black and then blitted back to the frame buffer. In the above example, that amounts to reading or writing no less than about 4.5 million bytes for <em>every</em> refresh. This alone could explain why Yadex becomes so sluggish when the window size grows. <p>I plan to remove the pixmap completely and always display directly onto the window, like is done when <code>-P</code> is given. That would eliminate the need for the blitting step, which would cut two thirds of the load. What's more, I strongly suspect that clearing and drawing to a window is significantly faster than doing the same with a pixmap (because the X server typically can't use the video hardware on pixmaps). The problem is doing that in a flicker-free fashion. <h2>How Yadex deals with wad files</h2> <p>Yadex keeps the iwad open for the whole duration of the Yadex session. The iwad is closed only when Yadex exits. Pwads are kept open as long as they are loaded. A pwad is closed only when another pwad with exactly the same resources is loaded or, if it contains only a level, when that level is saved. <p>The wad files are opened with a simple <code>fopen()</code>. Yadex make no attempt to control the locking. However, even though it would probably be possible for another process to open a file while Yadex keeps it open, it would nontheless be a bad idea. <p>For more high-level information on how Yadex treats wad files, see <a href="wad_specs.html">there</a>. </p><hr>AYM 1999-08-13 </body> </html>
distrib
>
Mandriva
>
8.2
>
i586
>
media
>
contrib
>
by-pkgid
>
9383028e5b5e6b6c2824e168b46388d9
>
files
>
16
