Stereo Tool

I'll quickly respond to a few of these: Order is the order in the menu. So, it goes from top to bottom. The menu essentially is a block diagram. True. Some people love it, some hate it. A real PEQ is on the todo list for the next version. True. It's gotten a lot better compared to - say - 2 years ago already, but more work is needed. To me at least, it shows everything I need in a very readable fashion. What are you missing or what makes it hard to read? If they are hidden completely you cannot click on them to open a specific page either - and I notice that I use them a lot more than I use the menu. Also the information that a specific filter is disabled wouldn't be so easy to see.
1. I think there are very few CPU's out there (correct me if I'm wrong) that have 4 cores and aren't fast enough to run Stereo Tool on 2. Maybe Intel Atom's, but I don't think anyone would buy those for a normal pc?
2. It would make the code a bit more complex. For one, the only way to use more cores would be to split the processing in multiple sections, but where I switch from one thread to the next would then have to depend on which filters are enabled and which settings they use, so I can split it up in a way that makes sense. That's not trivial.
3. For now, I need to optimize the clippers to use less CPU for a new Omnia product. So I have no other choice than to optimize it. Also, this is a high priority project so I need to do it now, before anything else. Once this is done, using more cores becomes only less useful because the total CPU load should be a lot lower than it is now.

Thnak you for a prompt reply What I meant by hard to read in terms of BS 412 is first of all - the graph size - it's just a bit too small, and the colour scheme. It's not self-explanatory too. Plus - it lacks a scale - there are no units

As for the order - since it's correct in the menu - it might be good to fix it in the meters too, otherwise it adds a bit of a mess

The CPU part - there's some fix needed for high end AMD chips. They're not a plain multithreaded cpus as Intels are. They're modular constructions with twice as much of ALU cores in comparison with the FPU ones. Tried FX6300 and it failed (i5-4430 match in passmark score). I totally agree about the optimization though - it's the way to go for now.

Just thinking - since clippers are so heavy on the CPU side - is there a way of running them in separate threads if necessary or is it completely impossible/impractical?

BTW - is there anything different in composite clipper/advanced clipper in comparison with 7.60 (I don't remember it being listed)? Some clipper induced artifacts appear to be less audible now.

Ah, every horizontal line = 1 dB.
Doesn't it match the menu? It should.....
O, yes, AMD....
We're working on making them less heavy now
Are you talking about the latest beta vs. 7.60 or 7.70 vs 7.60? I did change things in the latest beta when optimizing things; if you can tell me what you hear I'm very interested!

Oh, Is stereo enhancer after the AGC then? If yes, it would explain A LOT, e.g. why L-R midrange bass can be so annoying in ST (bass boost tends to boost midrange bass in the L-R area, thus making it very audible - some people like me find it not optimal, because basically when using headphones, excessive L-R bass can make you dizzy and certainly adds up to a general listening fatigue).

I know we have hpf and lpf in stereo enhancer, but here's where the general mid/side hpf filter for processing would help too

For the clipper part - I meant 7.60 vs 7.70 (beta too) - I seem to hear vibrating distortion less. Will take a closer look at this with regard to latest beta.

And yes - AMD> I know they're not the best chips around, but something seems wrong with thread managing as far as latest AMD CPU's are concerned - and it appears to be both - Windows and something in ST too. FX CPU's are generally uneven from application to application (it depends on whether app is using lots of FPU-dependent operations or not), but the difference between other apps and ST seems just too big.

As an example - I ran ST on Core2 Intels, like E6xx, E8xx with no problem as well as Athlon II's (dual cores, quads) - there were differences between Intels and AMD's in general performance (Athlons were a bit weaker) but the difference was not so huge in the first place, and the thread managing looked identical in both - AMDs and Intels. This is not the case with the FX AMD series, where everything looks packed in one thread and it just kills the CPU performance in ST. Its horsepower should be more than adequate though to run ST as smooth as my laptop at the very least (which gets half of the FX6300 passmark score).

It should definitely use 2 cores for processing, plus one for the GUI. It's starting separate threads for those... Can you try what happens if you set the affinity in Stereo Tool to different cores?

I did - then one core goes as high as 100% sharp and the second uses like 60%. Here are the screens:

7.70 - affinity off:



7.70 - affinity on (core 0 and 1 processing, 2 gui - but it doesn't matter how I assign them - it looks the same)



7.71 beta 001 - affinity off:



7.71 beta 001 - affinity on (as above):



CPU is FX6300 (overclocked to 4.2GHz on all six cores - it didn't help). The OS is Win7 Pro x64 with SP1. ST is always 32 bit, but I tested x64 - used slightly less CPU but still not less enough to run without buffer eating. Preset is stock Unveil FM with RDS settings saved (hence preset name ST 7_70)

My laptop has i5-3210m, it runs buttery smooth.

To explain this phenomenon:



see my point on more threading now? For AMD modular CPU's (Bulldozer, Piledriver, Steamroller cores) ST would need 4 threads so that floating point operations don't kill the CPU.

I agree. At least make it optional for filters like the Comp Clipper. It will make life a lot easier for a lot of lower end cpu's. (Yep I know Intel can handle the full ST but there is not only i5 and i7 in this world).

And one more suggestion. Make the buffer system better. It is very easy to run out of samples even when 400ms buffers are used.

Thanks a lot.
Mpex2006km - Konstantinos

_________________
Very proud user of Stereo Tool since 2010. Thanks Hans.

My radio station
http://amradio.ddns.net

From https://en.wikipedia.org/wiki/Bulldozer ... hitecture) : That means that when I use AVX the performance should be optimal. And... I do! But, it might be that Intels compiler (which should give the AVX-optimized functions to any AVX-capable processor) is 'acting up' again - they did that in the past (just let all AMD processors run unoptimized code, AMD sued them for that a number of years ago and they were convicted). I will build a version with only AVX support to test this.

Good point, indeed.

Versions 7.60; 7.51 work exactly the same. Version 7.50 though - the CPU usage seems identical BUT buffer is never lost, hence - it runs with no major glitches or interruptions. Strange...

Version 7.51 is when I switched to a new Intel compiler - which on most systems (including older AMD's) was reported to run more efficient. Apparently not on these newer ones...

7.71 beta should really run without any issues, since the CPU load is a lot lower than before.