Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

Faster than 40d, no problems

42 (26.1%)

Faster than 40d, problems

72 (44.7%)

No slower than 40d, no problems

14 (8.7%)

No slower than 40d, problems

16 (9.9%)

Slower than 40d, no problems

2 (1.2%)

Slower than 40d, problems

3 (1.9%)

Doesn't work (please explain)

12 (7.5%)

Total Members Voted: 160

[Baughn]

If all you want is a playlist, the code for that should be already available. It's the environmental sound that isn't. Although..

It occurs to me, we could count the number of tiles of varying kinds that pass through the graphics routines, and pick an audio track based on that. I'd suggest you wait until I've got threading working, but it's a good idea, and would exercise the code until Toady picks it up.

Heck, we can even detect when sieges occur, or more precisely note when monsters go on-screen. Or pick up the siege popup when it happens.

[Mel_Vixen]

A bit hacky but in the right spirit . He we could make any message go "ding" like the doctor once said. 

[Grax]

Liquid pumping bug

Just imagine: Screw pump that pumps water or magma in the blind tunnel with channelled end (to lower level).

Liquid does go to the next reservoir through channelled square until it become full.
Then it start to disappear. Pumps simply takes liquid out of reservoir and that's all. It goes nowhere.

So you can drain a magma pipe or ocean to its lowest level.

[Baughn]

I'm still testing it to make sure it *works*, but I'm otherwise happy to inform you that the ICC non-commercial licence applies to DF/Linux.

So that's another 10-25% performance right there.

EDIT: That would be a yes. Works just fine. Let me get the improved build script off to toady..

[Janus]

ICC?

I tried a quick Google search, but naturally there are a lot of results which use that acronym.

[bhelyer]

I'm still testing it to make sure it *works*, but I'm otherwise happy to inform you that the ICC non-commercial licence applies to DF/Linux.

So that's another 10-25% performance right there.

EDIT: That would be a yes. Works just fine. Let me get the improved build script off to toady..

I was _just_ wondering about that this morning. That is good news. What was LLVM's performance improvement? LLVM won't build BC for me though, instead crashing, so that's more of a theoritical question.

Intel did at one time claim they got the Linux kernel building with icc. If that's true, then they must be pretty compatible with most GNU C sources.

@Janus: Intel C Compiler

[Baughn]

About the same as ICC. LLVM does similar optimizations, except that ICC is better at vectorizing. The thing is, most programs don't benefit greatly from vectorizing.

DF might, considering the huge datasets it likes to grind over. At any rate, it won't hurt.

Oh, and another win with ICC is that it'll emit multiple execution paths - one per CPU model, more or less, so you get SSE4 code for core 2 in the same executable as pentium 3-optimized code.. I've got the build script building every possible combination, and the executable is still only 440k, vs. gcc's 289k executable. It's an increase, but hopefully worth the cost.

(That's for BC. DF will obviously be larger, but it's still a win. Also, before you panic, note that the unused code paths on any given machine will not make it into the CPU cache, so no issues there.)

[bhelyer]

I assume you contacted the Intel guys directly, because I don't see anything dealing with donations in the non-commercial FAQ.

Oh, and another win with ICC is that it'll emit multiple execution paths - one per CPU model, more or less, so you get SSE4 code for core 2 in the same executable as pentium 3-optimized code.. I've got the build script building every possible combination, and the executable is still only 440k, vs. gcc's 289k executable. It's an increase, but hopefully worth the cost.

Very cool. I do recall building KQ with it when I was working on that. It's a very solid compiler. Though, it is their architecture so I guess you could expect that it would be.

[corvvs]

About the same as ICC. LLVM does similar optimizations, except that ICC is better at vectorizing. The thing is, most programs don't benefit greatly from vectorizing.

DF might, considering the huge datasets it likes to grind over. At any rate, it won't hurt.

Oh, and another win with ICC is that it'll emit multiple execution paths - one per CPU model, more or less, so you get SSE4 code for core 2 in the same executable as pentium 3-optimized code.. I've got the build script building every possible combination, and the executable is still only 440k, vs. gcc's 289k executable. It's an increase, but hopefully worth the cost.

(That's for BC. DF will obviously be larger, but it's still a win. Also, before you panic, note that the unused code paths on any given machine will not make it into the CPU cache, so no issues there.)

You might want to compare against GCC 4.4.0 that just came out. It can do lots of new loop optimizations to make cache-misses far less common. And IIRC, the version in development now will be able to parallelize certain kinds of loops, so you can get some sort of multi-processor usage without having to explicityly code multithreaded (although that would of course still be better).

[Baughn]

Yes, I emailed Intel and specifically asked them about DF's situation.

Donations are apparently fine; it's only if you actually draw a salary that you don't qualify for a non-commercial license.

[Jurph]

I love the idea of ICC boosting performance, and I'm sure each of our hard drives can handle it, but can Toady's host survive a client that's 33% larger?

[bhelyer]

Well, it still remains to be seen what the increase is on the DF binary, and plus it's just for the Linux version (Intel only offers the non-commercial version of their compiler on Linux, as far as I can remember), which would minimise the impact.

Donations are apparently fine; it's only if you actually draw a salary that you don't qualify for a non-commercial license.

The website gave me the impression the the requirements are quite strict, so that's good to know.

[Baughn]

He doesn't have to enable *every* possible code-path, either.

Also, if bandwidth is really a problem, I should think he'd switch to releasing via bittorrent. I'm sure nobody would mind.

[bhelyer]

I'm sure nobody would mind.

In these flats the internet is networked through a proxy/portal thing, and I agreed to not use bit-torrent, or they'd disconnect me. I have no idea if they try and detect the protocol (or even can), but I don't use it, even if it is for legal stuff.

Not to dissuade against BT, though. I'm sure it'd pop up in other mirrors. I thought the BT protocol wasn't too hot on small file sizes though.

[Baughn]

It depends on the block size you set, and DF isn't /that/ small.

Also, I may have forgotten to mention it, but it would be natural to have both a bittorrent and a b12g.com link. With the latter perhaps throttled.