Smoothly ramp changes in gain. #39

ywwg · 2013-07-08T02:02:51Z

https://bugs.launchpad.net/mixxx/+bug/854082

Add copyWithRampingGain, applyRampingGain and addWithRampingGain methods to SampleUtil and use them to ramp changes in gain in various places:

deck EQs
deck pregain
channel volume
channel orientation
crossfader
master volume
headphone volume
head/mix

Still remaining to be done:

balance

rryan · 2013-07-08T13:51:20Z

src/sampleutil.h

+    // buffers must be the same length.  pDest may be an alias of the source
+    // buffers. pSrc1 is the buffer that is fading out, pSrc2 is the buffer
+    // that is fading in.
+    static void crossfadeBuffers(CSAMPLE* pDest,


Maybe call it linearCrossfadeBuffers since it does linear crossfades.

daschuer · 2013-07-08T21:05:32Z

Thank you for the fix. I just want to clarify that the result is Audio-buffer-size dependent, right?
But It looks suitable.

rryan · 2013-07-08T21:09:15Z

Yea, the crossfade step is dependent on the audio buffer size so depending on the magnitude of the change in gain we could still get clicks (as we talked about on the bug report).

ywwg · 2013-07-09T19:15:05Z

The worst-case scenario I can imagine is 1ms latency at 44.1Khz, which is 44.1 samples per buffer. Even having that amount of space to interpolate is better than nothing. But also, as the buffer becomes smaller the delta gain will also become smaller because the user has less time to make the change. (Could a user twist a knob from 64 to 0 in 1ms? unlikely). So in practice I'd be surprised if it were possible to cause clicks.

rryan · 2013-07-09T19:19:47Z

An eq kill button or resetting an eq knob could take it from max value to 0 in one callback. There's really no perfect solution :). This one is a good one.

ywwg · 2013-07-09T19:48:00Z

(the most recent changes are untested, but it builds. Do not merge until I've actually done testing)

ywwg · 2013-07-11T14:21:03Z

the crossfader and gain knob additions are total fails. I will fix them.

ywwg · 2013-07-11T23:14:37Z

Ok pregain and master are working now, but headphone is tougher because it has a bunch of moving pieces that aren't all together (balance, gain, master mix)

ywwg · 2013-07-13T22:30:44Z

OK to merge?

rryan · 2013-07-13T22:43:08Z

src/engine/enginemaster.h

      private:
        double m_dVolume, m_dLeftGain, m_dCenterGain, m_dRightGain;
+        QMap<ChannelInfo*, double> m_cache;


Rather than use a QMap, can you keep the cached value in ChannelInfo? We already have ChannelInfo. It may be ugly but it avoids using a QMap in the callback.

Hm, the problem with that is it becomes difficult to do the comparison of the current gains to the previous gains to see if we need to do a crossfade. With the map, I can see if the two maps are equal. But without it, I'll have to iterate through the list of pchannels.

ywwg · 2013-07-14T22:31:39Z

notes addressed other than caching suggestion (see comment). (and I will squash some commits before merging so please let me do the final merge)

ywwg · 2013-07-14T22:32:25Z

src/engine/enginemaster.cpp

@@ -397,6 +403,20 @@ void EngineMaster::process(const CSAMPLE *, const CSAMPLE *pOut, const int iBuff
    // Perform the master mix
    mixChannels(masterOutput, maxChannels, m_pMaster, iBufferSize, &m_masterGain);

+    // Avoid soundwave discontinuties by interpolating from the old gains to new.
+    if (!m_masterGain.compare(m_prevMasterGain)) {
+        // We are reusing p_pPrevGainBuffer from above, but its data isn't needed


out of date comment

…gains. address review notes

abandon headphone buffer crossfading for now, it's got too many moving pieces

…cache every time.

…gains. address review notes

abandon headphone buffer crossfading for now, it's got too many moving pieces

…cache every time.

…tation.

Conflicts: src/engine/enginemaster.cpp src/engine/enginemaster.h src/sampleutil.cpp src/sampleutil.h

ywwg · 2013-07-18T14:22:46Z

Rebased, merged in RJ's changes.

ywwg · 2013-07-18T14:23:26Z

I will do some testing on this tonight before I commit the merge.

rryan · 2013-07-18T14:42:41Z

Cool -- thanks! There are still those outdated comments you marked to fix.

We should do some profiling. mixChannels is not an engine hot-spot (at least, it's nowhere near as hot as the EQs) -- it takes tens of microseconds last time I checked. But it would still be interesting to know the change ramping causes on various hardware.

ywwg · 2013-07-18T23:45:08Z

performance testing:

I loaded a track, started playing it, and then wiggled the mid eq really fast with the mouse.

trunk:
Debug [Main]: Stat("EngineBuffer::process_pauselock","count=326,sum=1.5881e+07ns,average=48714.8ns,min=4987ns,max=200271ns,variance=1.08776e+09ns^2,stddev=32981.2ns")
Debug [Main]: Stat("EngineMaster::mixChannels_0active","count=650,sum=1.80178e+06ns,average=2771.96ns,min=678ns,max=23088ns,variance=5.54639e+06ns^2,stddev=2355.08ns")
Debug [Main]: Stat("EngineMaster::mixChannels_1active","count=326,sum=1.0642e+06ns,average=3264.41ns,min=1082ns,max=11261ns,variance=3.05933e+06ns^2,stddev=1749.09ns")
Debug [Main]: Stat("EngineMaster::process","count=488,sum=1.56157e+08ns,average=319993ns,min=29216ns,max=1.20952e+06ns,variance=5.92068e+10ns^2,stddev=243325ns")
Debug [Main]: Stat("EngineMaster::process channels","count=488,sum=5.39613e+10ns,average=1.10577e+08ns,min=-2.14373e+09ns,max=2.12896e+09ns,variance=1.56943e+18ns^2,stddev=1.25277e+09ns")

crossfade:
Debug [Main]: Stat("EngineBuffer::process_pauselock","count=274,sum=9.736e+06ns,average=35532.8ns,min=6704ns,max=146136ns,variance=3.20477e+08ns^2,stddev=17901.9ns")
Debug [Main]: Stat("EngineMaster::mixChannels_0active","count=738,sum=1.36968e+06ns,average=1855.94ns,min=673ns,max=18869ns,variance=1.56249e+06ns^2,stddev=1250ns")
Debug [Main]: Stat("EngineMaster::mixChannels_1active","count=274,sum=714816ns,average=2608.82ns,min=1332ns,max=13053ns,variance=1.54973e+06ns^2,stddev=1244.88ns")
Debug [Main]: Stat("EngineMaster::process","count=506,sum=8.55766e+07ns,average=169124ns,min=25801ns,max=654444ns,variance=1.49361e+10ns^2,stddev=122213ns")
Debug [Main]: Stat("EngineMaster::process channels","count=506,sum=6.00099e+10ns,average=1.18597e+08ns,min=-2.14136e+09ns,max=2.13266e+09ns,variance=1.52487e+18ns^2,stddev=1.23486e+09ns")
Debug [Main]: Stat("EngineSideChain:writeSamples","count=506,sum=1.63087e+06ns,average=3223.07ns,min=881ns,max=346295ns,variance=2.41958e+08ns^2,stddev=15555ns")

I think it's better? I'm surprised about that.

ywwg · 2013-07-27T02:13:45Z

LGTM and stuff

Smoothly ramp changes in gain.

kain88-de · 2013-09-12T16:21:01Z

Ok I tracked down the first commit that makes the enginemaster test fail, it is 6d5d02f

After this enginemaster changes the channel buffer. When you later also merge in all the changes of rryan everytest starts to fail. I guess it has something to do with the changes you made in the gain calculation but I'm not sure since I don't understand the code here.

Add lp:1846966 to 2.2.3 changelog

rryan reviewed Jul 8, 2013
View reviewed changes

rryan reviewed Jul 13, 2013
View reviewed changes

ywwg reviewed Jul 14, 2013
View reviewed changes

ywwg added 11 commits July 14, 2013 19:01

Create crossfade utility function and apply it to EQs as an example

701eb61

style cleanup

598539a

make it clear that crossfader util is linear

7585587

Apply crossfade interpolation to master and headphone output, and pre…

27b578e

…gains. address review notes

Fix pregain buffer crossfading

6ea2624

Fix master mix buffer crossfading

f45bee0

abandon headphone buffer crossfading for now, it's got too many moving pieces

cleanup

3b1adbd

Tweak caching functionality so we aren't deleting and recreating the …

947e147

…cache every time.

start implementing ramping functions

166ba75

finish implementing ramping functions

03fc420

typos

9322505

ywwg added 12 commits July 18, 2013 10:08

make it clear that crossfader util is linear

25f4147

Apply crossfade interpolation to master and headphone output, and pre…

e609aab

…gains. address review notes

Fix pregain buffer crossfading

51ca6d9

Fix master mix buffer crossfading

9ffe2f3

abandon headphone buffer crossfading for now, it's got too many moving pieces

cleanup

1fcd2e5

Tweak caching functionality so we aren't deleting and recreating the …

408d1aa

…cache every time.

start implementing ramping functions

a3671b1

finish implementing ramping functions

87275a8

typos

ea9f506

delete all SSE hand-optimized functions from sampleutil

a33b58a

Keep old gain values directly in ChannelInfo. Simple example implemen…

6d5d02f

…tation.

Merge remote-tracking branch 'rryan/crossfade_util' into crossfade_util

62d3146

Conflicts: src/engine/enginemaster.cpp src/engine/enginemaster.h src/sampleutil.cpp src/sampleutil.h

clean up comments.

62dda7c

ywwg added a commit that referenced this pull request Jul 27, 2013

Merge pull request #39 from ywwg/crossfade_util

11f3c5a

Smoothly ramp changes in gain.

ywwg merged commit 11f3c5a into mixxxdj:master Jul 27, 2013

ywwg deleted the crossfade_util branch July 27, 2013 02:14

ywwg mentioned this pull request Dec 20, 2013

Refactor the engine to process samples in the range of [-1.0, 1.0]. Fixes Bug #1204039. #126

Merged

daschuer mentioned this pull request Dec 28, 2014

Spinny updates #440

Merged

daschuer mentioned this pull request Jun 11, 2018

Library Layout Redesign Single pane #1117

Closed

11 tasks

Be-ing added a commit that referenced this pull request Nov 27, 2019

Merge pull request #39 from uklotzde/release_2.2.3

d4edbce

Add lp:1846966 to 2.2.3 changelog

mixxxbot mentioned this pull request Aug 22, 2022

Smoothly ramp gain changes. #5992

Closed

ronso0 mentioned this pull request Sep 4, 2022

Effects: allow clearing chains with empty '---' preset #10859

Merged

ronso0 mentioned this pull request May 9, 2023

qopenglwindow-based replacement of qglwidget, full glsl shader based waveforms, vumeters and spinnies #10989

Merged

Swiftb0y mentioned this pull request Aug 7, 2023

GNOME 44 (Wayland): Context menu on tree items in side bar closes immediately #11576

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Smoothly ramp changes in gain. #39

Smoothly ramp changes in gain. #39

ywwg commented Jul 8, 2013

rryan Jul 8, 2013

daschuer commented Jul 8, 2013

rryan commented Jul 8, 2013

ywwg commented Jul 9, 2013

rryan commented Jul 9, 2013

ywwg commented Jul 9, 2013

ywwg commented Jul 11, 2013

ywwg commented Jul 11, 2013

ywwg commented Jul 13, 2013

rryan Jul 13, 2013

ywwg Jul 14, 2013

ywwg commented Jul 14, 2013

ywwg Jul 14, 2013

ywwg commented Jul 18, 2013

ywwg commented Jul 18, 2013

rryan commented Jul 18, 2013

ywwg commented Jul 18, 2013

ywwg commented Jul 27, 2013

kain88-de commented Sep 12, 2013

Smoothly ramp changes in gain. #39

Smoothly ramp changes in gain. #39

Conversation

ywwg commented Jul 8, 2013

rryan Jul 8, 2013

Choose a reason for hiding this comment

daschuer commented Jul 8, 2013

rryan commented Jul 8, 2013

ywwg commented Jul 9, 2013

rryan commented Jul 9, 2013

ywwg commented Jul 9, 2013

ywwg commented Jul 11, 2013

ywwg commented Jul 11, 2013

ywwg commented Jul 13, 2013

rryan Jul 13, 2013

Choose a reason for hiding this comment

ywwg Jul 14, 2013

Choose a reason for hiding this comment

ywwg commented Jul 14, 2013

ywwg Jul 14, 2013

Choose a reason for hiding this comment

ywwg commented Jul 18, 2013

ywwg commented Jul 18, 2013

rryan commented Jul 18, 2013

ywwg commented Jul 18, 2013

ywwg commented Jul 27, 2013

kain88-de commented Sep 12, 2013