0.9u-12 --> Hiccups and Clicks

[Gerard]

Peter and all,

The problems are gone over here... I have set q1 on -2 for a few weeks.... Today i moved it a bit upwards (0)  and the cracks and hickups are gone....  

Maybe something for Edward to try.....

Grtz

[PeterSt]

Hmm ... if this is really so, you get the Nerd Award of the year, ok ?

In the mean time, I think I can adjust the Q1 temporarily then, which would only be during the load of the track (which typically is less than one second).

Thinking of this really being so, it would even make sense to me. But then there's another solution, and this is in fact the solution to something I pointed out some longer ago (over half a year) :
(actually, here it is : XXHighEnd Model 0.9m (Implements Processor Core Appointment))

So, at that time I wasn't able to appoint that -by now (!!) heavily cpu using thread- to the other core. Maybe with the increased knowledge () now I can ? Anyway, this now just *should* happen. Why ?

That half a year ago, the only thing what happened in that other thread was reading the file from disk. This is not able to use 100% cpu because of the I/O involved. Today, the cpu (better : that one core) will be used for 100% and that might take up to near a second. This is officially way too long to keep on filling a buffer each 1 millisecond, and as I indicated earlier : it looks like time slicing doesn't work properly anymore. Now I say : how can that ever be done properly with a cpu (core) being in use for 100% (and tasks between that core *have* to be divided) ?

As a sidenote I refer to what I said earlier about the copying from RAM disk (as Edward does it), which only makes this worse (the situation stays longer) ...

Gerard, 

[Gerard]

you get the Nerd Award of the year, ok ?

When this keeps on going, i will have to make a special showcase for all the awards i receive!

 

 

[edward]

Peter and all,

The problems are gone over here... I have set q1 on -2 for a few weeks.... Today i moved it a bit upwards (0)  and the cracks and hickups are gone....  

Maybe something for Edward to try.....

Grtz

I think we may have a winner!!

Well . . . at first attempt with Q1=0 there was more stumbling (and crackling) between tracks than before, so I decided instead of trying every one, I just went to 14 hoping that is most neutral. Success! I only played through the offending album once, but I think everything was smooth. And it makes sense, because I was eventually going to get to trying Q1 settings, but I first tried Schemes. No appointment was full of crackling. And the others were only mildly better. Scheme 1 is the only one that is perfect for me (which is just as well because it clearly sounds the best to me).

And let me just add, Peter, this success comes with all my usual tweaks (Virtual Memory off, limited services, etc.) in place. And also, this is with the original 0.9u-13 (not the modified XXEngine3.exe that you offered)

So, to be clear, I still clearly prefer Q1=-2, but if you have a way if tricking it to 14 during transitions then fantastic.



And finally, during all this, I stumbled onto something else - Peter check your PM.

[PeterSt]

I've been working the whole evening on the Appointment stuff for the loading thread, and eventually got it running. That is ...
I had been looking for two hours into an error that shouldn't be, prepared for bed, but got back. It suddenly jumped on me that the error could be caused by a thread not being allowed to have different Appointment (ok, Affinity) than its starting thread. Since I was working with Scheme-3, which appoints the audio to core 2, the threads coming from there couldn't go to core 1. It just errors.

Now I don't know of any logical plan to fix this. Yeah, maybe some technical complicated structure which uses a special scheduling thread (no Affinity), that synchronizing the playback and the load threads (each now can have their own Affinity).

Man, I must rize the price to 73 euro. Hahaha.

[Gerard]

Peter

First i like to say that u13 sounds just great overhere. Specially the bass!
Leaving q1 at -2 and changing to schema does the trick also like Edward sayd. But a bit different.  No appointment, schema 1, schema 2 dont give problems. Schema 3 and 4 does.


I allready have the 1 euro ready to send! 

 

[PeterSt]

Yes, I was looking into this excessively yesterday (because of wanting to tweak the appointment of the loading thread), and from theory (hence looking at the cpu graph, without sound) it is clear to me that Scheme-2 behaves how we want it. That is, the loading goes to core 1 BUT the playback can't be decided for (with Scheme-2 playback goes preferrably to core 2 but may go to core 1 when the OS likes that.

Scheme-1 is a different beast; this tells the audio to play in core 2, and all of the other stuff to be in core 1. But, the downside is that the loading goes to core 2 as well (it inherits from the playing thread). Also, knowing that Edward has (had) the problems with Scheme-1, from this we should derive that it is the loading itself. When you, Gerard, find that Scheme-1 helps making it better, the conclusion must be that it is stuff from the OS itself that interferes with all, and that it actually is a problem with switching between cores (the output towards the general bus or whatever) or something in that area I can't comprehend so far.
Scheme-3 indeed is the worse from normal theory, and Scheme-4 is the worse when switching between cores to the matter of dedicating threads doesn't work properly. On the last, keep in mind that each time a thread switches from core, the in-cpu cache memory has to be copied which is consuming by itself.

These matters are rather complex, and FYI many pro audio people see a degradation of sound (like glitches) with Quad core processors (or at least an unexpected lower throughput opposed to dual cores).
I think one can get a degree in getting this right.

[PeterSt]

FYI : So far I am not able to solve this by a different Appointment (Affinity) to the loading thread, but also the temporary change of Q1 cannot work. The latter is related to Exclusive Mode not being able to initialize more than once per Ausio Session; with 0.9u-14 I thought I could trick it (according to the concerned properties), but this turns out not to be so.

For now this remains unsolved for those who are bothered by it.

[PeterSt]

... but since I never give up, I managed to tweak the unallowed changing of the inherited Appointment (Affinity).



Below a few plots for Scheme-3 (must work out the others), noting that Scheme-3 plays sound on core 2, while the other processes are preferably on core 1.
The spikes represent the track load (hence the preprocessing of it), which went to core 2 with Scheme-3 before.

What I observe (or seem to, relative to how it was) is that there is plenty of headroom now for the 4-6 minute tracks which are loaded here. Normally the cpu shows a horizontal line at the top for the Doubled padded to 32 bits tracks (which for the before situation indeed would indicate some heavy working against itself e.g. heavy core cache switching). No squeezed toes in my shoes "will it do it" now. It just feels good.
Btw, Q1 is -2 here.

I'm near the end of Amused to Death now, and no single hiccup. Does this say anything ?

Oh, note the plots are at full throttle here, and TaskManager itself consumes the most cpu. The most flat lines in the right core are the most representative for what Engine#3 is doing. Where core 2 shows more activity, it's the "dying out" of other services being active at that moment. Picture 03 shows this from off the middle (the track load is not visible anymore (but for in the memory line, near the left).
In 01 vs. 04 (and 02 a bit) you can see that core 2 tends to help core 1. In 01 this apparently was not necessary. By itself this is good balance, because the paying thread now is on both cores temporarily ...

[PeterSt]

For those who were bothered by this, would you care to try the below version of XXEngine3.exe and report please ?

Put the current XXEngine3.exe in your XX folder aside, and replace it with the one in the zip below.
Thanks,
Peter

Edit : Removed the zip, while results are known from it by now, and it can create confusement (merely created by myself in other posts).

[Gerard]

For those who were bothered by this, would you care to try the below version of XXEngine3.exe and report please ?

Put the current XXEngine3.exe in your XX folder aside, and replace it with the one in the zip below.
Thanks,
Peter

I am really sorry to say Peter..... But nothing has changed.... For me it is exact the same like before.... From the beginning till the end of the song cracks.

Q1-2 schema 3 & 4:                          Cracks
Q1-2 schema 1 & 2 & No Appointment: No problemmo
I can bring my pc over... When that safe's time for you 

[PeterSt]

I am really sorry to say Peter..... But nothing has changed.... For me it is exact the same like before.... From the beginning till the end of the song cracks.

Q1-2 schema 3 & 4:                          Cracks
Q1-2 schema 1 & 2 & No Appointment: No problemmo
I can bring my pc over... When that safe's time for you 

But Gerard, I may have said it before ... you are bugged by something else. This is not about cracks throughout the tracks, this is about track boundaries.
Sorry ...

[edward]

For those who were bothered by this, would you care to try the below version of XXEngine3.exe and report please ?

Put the current XXEngine3.exe in your XX folder aside, and replace it with the one in the zip below.
Thanks,
Peter

OK Peter, I finally got around to trying this. I put this in with version 0.9u-14a. I played my "usual" set of DCD tracks and unfortunately I still heard a hiccup. Although, it occurred at a different point, so it sort of caught me off guard. Before I would start with track 2 (from Spiritchaser) and I would always hear a hiccup between tracks 3 & 4. And this is consistently repeatable. But with 0.9u-14a (and the new engine.exe) I didn't hear a hiccup until between tracks 5 & 6. I've only had a chance to listen once, so I will try again later to make sure this is repeatable.

But all is not lost!! There is a significant improvement in SQ over 0.9u-14! (Do you owe me one euro now or do I owe you?    )


Edit:
And BTW, I tried this with my new favorite Q1 = -1. (I figured this shouldn't make a difference since prior to 0.9u-14 Q1 = -2 was really -1.)

A bug has been found for the Q1 settings of -2 and -3;
-2 appeared to be set equal to -1;
-3 appeared to be set equal to -4.
Now both have their own values indeed.

[PeterSt]

Hmm ... I am playing with 0.9u-14a since it is there, Appointment = 2, and at least I never had a problem anymore.
I must say though that in 75% of cases I played with Q1 = 14.

Well, if you can find a pattern I'll be happy to dive into it again.

But with 0.9u-14a (and the new engine.exe)

In 0.9u-14 that new XXEngine3 is incorporated. But I guess that's what you were saying ? (hence don't copy that separately uploaded XXEngine3 over it ! the latest is working a tad differently again to this matter !).

[edward]

In 0.9u-14 that new XXEngine3 is incorporated.

But with 0.9u-14 (with Scheme 1) the track loading still incurs a CPU spike on core 2. With 0.9u-14a (and the separately uploaded XXEngine3.exe) the CPU usage during track load is with core 1. (I assume this accounts for the better SQ)

Is that what you mean by this:

. . . the latest is working a tad differently again to this matter !).

And if it's different, how can it also be already incorporated?