Screenshots

[Cody]

It's exactly the time needed for this generation during system entry that hurts...

Ouch! Yes, I see what you mean!

[Cody]

Sure does look pretty though...

[Astrobe]

Alternatively, current texture model is generated upon system entry [snip]

It's exactly the time needed for this generation during system entry that hurts (plus the fact that sound effects that might be playing at that time (or when going on F7) get borked a bit).

I was certain that the texture generation stuff was using a separate thread. That annoyed me a lot because other parts of the game were running and emitting logs and it was hard to measure generation times, so I ended up forcing the F7 screen to make textures at full resolution.

[another_commander]

I was certain that the texture generation stuff was using a separate thread.

I think you are right on that, actually. I will need to check in the source to confirm. There are some tasks that are done on separate threads and most likely this could be one of them. Just to clarify, when I said about moving texture tasks to other threads I was referring more to the Perlin calculations, which is where all the delay happens and for which I am not sure what the thread situation is.

[RockDoctor]

Moar experiments. Lots of potential, if only we could optimize the texture generation times a bit...

Is there an explicit model for the atmosphere drop off? You might be able to trim some computation effort by swapping a physics-based model (I think it'd be an exponential drop off, but I'd have to check that) for something simpler.

[Redspear]

Alternatively, current texture model is generated upon system entry [snip]

It's exactly the time needed for this generation during system entry that hurts (plus the fact that sound effects that might be playing at that time (or when going on F7) get borked a bit).

So texture generation can neither be delayed (as I suggested) or prepared earlier (e.g. during or just prior to hyperspace sequence) for any significant gain?

If there must be a pause in game then is there a better place to 'hide' it?... or even to 'stagger' it?

For the F7 screen I believe the source currently uses the main texture in miniature. Could it not instead use a simpler and seperate texture generation method? Just as the graphics options in the game allow for shaders and extra detail, perhaps the F7 screen could do without either (and without the new fancy textures ilustrated above).

[Astrobe]

I wouldn't mind a loading screen either if it's for really more awesome textures (there's still the problem of the F7 screen; might need some "please wait..." too).

One dirty optimisation trick: I believe that the textures are generated so that they can be tiled at least in one dimension (check if it's even two dimensions and if we can ditch it - might save a few CPU cycles per pixel). But the player only see half of it at any given time. So one could generate a half-size texture and tile it two times.

Or maybe use a totally different algorithm. For instance, a cellular automaton based one (basically a more elaborate version of Conway's game of life). That's more a research topic and I'm not even sure it can be made more efficient than Perlin. Or a hybrid technique using a set of predefined textures that could be altered in different ways by a PRNG-based algorithm ... That could give us map "features" (forests, deserts, glaciers, cities, ...).

[another_commander]

So texture generation can neither be delayed (as I suggested) or prepared earlier (e.g. during or just prior to hyperspace sequence) for any significant gain?

Jens had written texture pre-loading routines years ago that attempted to do just that. They are still in the code but disabled due to crashes, according to the comments. Hmm...

[Diziet Sma]

HTF do you get HTML entities (e.g., the degree symbol) into this posting engine.

I do a google search on "degrees symbol", then copy & paste one of them from the results page.

360°

[Cody]

Why not simply hit Alt+248 - 360°

[Diziet Sma]

Why not simply hit Alt+248 - 360°

Because that's never worked for me, perhaps?

[pleiadian]

Why not simply hit Alt+248 - 360°

Because that's never worked for me, perhaps?

Here's a full list of HTML entities:
http://www.ascii.cl/htmlcodes.htm

[another_commander]

Hey look! Screenshots!

[RockDoctor]

Why not simply hit Alt+248 - 360°

Because that's never worked for me, perhaps?

I use the BBCodeXtra plug in on FireFox (seemingly unsupported, but it still works), and my "Maths" character collection uses the following entries (format : entity ; entity decomposed with spaces ; description including UNICODE codepoint) :

Code: Select all

<BR>&times; & times ; U+00D7 Multiplication symbol. 
<BR>&divide; & divide ; U+00F7 Divide symbol. 
<BR>&le; & le ; U+2264 Less-than-equal symbol.
<BR>&ge; & ge ; U+2265 Greater-than-equal symbol.
<BR>&deg; & deg ; U+00B0 ; degree symbol
<BR>&plusmn; & plusmn ; U+00B1 (177) ; plus-minus sign (plus-or-minus sign)
<BR>&sup2; & sup2 ; U+00B2 (178) superscript two (superscript digit two, squared)
<BR>&sup3; & sup3 ; U+00B3 (179) superscript three (superscript digit three, cubed)
<BR>&minus; & minus ; U+2212 (8722) ; minus operator
<BR>&lowast; & lowast ; U+2217 (8727) ; asterisk operator
<BR>&radic; & radic ; U+221A (8730) ; square root (radical sign)
<BR>&prop; & prop ; U+221D (8733) ; proportional to
<BR>&infin; & infin ; U+221E (8734) ; infinity
<BR>&ang; & ang ; U+2220 (8736) ; angle
<BR>&and; & and ; U+2227 (8743) ; logical and (wedge)
<BR>&or; & or ; U+2228 (8744) ; logical or (vee)
<BR>&cap; & cap ; U+2229 (8745) ; intersection (cap)
<BR>&cup; & cup ; U+222A (8746) ; union (cup)
<BR>&int; &  int ; U+222B (8747) ; integral
<BR>&there4; & there4 ; U+2234 (8756) ; therefore sign
<BR>&sim; & sim ; U+223C (8764) ; tilde operator (varies with, similar to)
<BR>&cong; & cong ; U+2245 (8773) ; congruent to
<BR>&asymp; & asymp ; U+2248 (8776) ; almost equal to (asymptotic to)
<BR>&ne; & ne ; U+2260 (8800) ; not equal to
<BR>&equiv; & equiv ; U+2261 (8801) ; identical to; sometimes used for 'equivalent to'
<BR>&sub; & sub ; U+2282 (8834) ; subset of
<BR>&sup; & sup ; U+2283 (8835) ; superset of
<BR>&nsub; & nsub ; U+2284 (8836) ; not a subset of
<BR>&sube; & sube; U+2286 (8838) ; subset of or equal to
<BR>&supe; & supe ; U+2287 (8839) ; superset of or equal to
<BR>&oplus; & oplus ; U+2295 (8853) ; circled plus (direct sum) EARTH
<BR>&otimes; & otimes ; U+2297 (8855) ; circled times (vector product)
<BR>&perp; & perp ; U+22A5 (8869) ; up tack (orthogonal to, perpendicular)
<BR>&sdot; & sdot ; U+22C5 (8901) ; dot operator
<BR>&lceil; & lceil ; U+2308 (8968) ; left ceiling (APL upstile)
<BR>&rceil; & rceil ; U+2309 (8969) ; right ceiling
<BR>&lfloor; & lfloor ; U+230A (8970) ; left floor (APL downstile)
<BR>&rfloor; & rfloor ; U+230B (8971) ; right floor
<BR>&lang; & lang ; U+2329 (9001) ; left-pointing angle bracket (bra)
<BR>&rang; & rang ; U+232A (9002) ; right-pointing angle bracket (ket)

I just paste the whole lot in (well, BBCodeXtra does), then delete what I don't want. It's clunky, but it works for me.
<P>I'd better Preview. Burned fingertips!
Well, bowdlerise me - even the entities were de-fanged.

[Cody]

Because that's never worked for me, perhaps?

<raises eyebrow>