Stereo Tool

Now building a new version where the other part of the declipper is also separated into 2 threads. (Note: I should later combine them, that will also improve throughput).

BETA007:
Converted some more declipper code to use 2 cores.

Winamp DSP plugin: http://www.stereotool.com/download/dsp_ ... 21-007.exe
Stand alone version: http://www.stereotool.com/download/ster ... 21-007.exe
VST version: http://www.stereotool.com/download/vst_ ... 21-007.dll
VST version (No SSE2): http://www.stereotool.com/download/vst_ ... 21-007.dll
Command line version: http://www.stereotool.com/download/ster ... 21-007.exe
Linux command line version: http://www.stereotool.com/download/ster ... ETA621-007 NOT AVAILABLE
Linux GUI version: http://www.stereotool.com/download/ster ... ETA621-007 NOT AVAILABLE
DECLIPPER Winamp DSP plugin: http://www.stereotool.com/download/dsp_ ... 21-007.exe
DECLIPPER VST version: http://www.stereotool.com/download/vst_ ... 21-007.dll

TODO:
- Create dual core support flag in Audio class
- I need to check if the new thread has the right priority
- Currently I create a thread of each portion of data that arrives, it's probably better to have a single thread process all the data.
- Found a bug in the non-SSE2 version of the clipper. Should be easy to fix.
- Now I have this, it's probably better to use Loudness as limiter of the perfect declipper output instead of the final limiter which is used now, but is less good.
- Convert declipper to use 2 cores as well. - Detection part too
- Fix memory bug in declipper clipping detection.
- Declipper: Combine 2 multicore parts into 1 (avoids an expensive 'join' phase)
- Combine 2 multicore threads into 1. (Move them up 1 level from Audio to Common)
- If needed (hiccups?), make it possible to disable multicore.


Odd, I have more multithreading but now I'm only measuring a 25% performance increase. Can anyone else confirm this?

Hallo Hans,

i tryed the last version you build. The performance is better thats true, now i can use a ASIO latency from 2048 without any problems, but 1024 would be better.

The Number of Threads doesn't change as you can see in the Image:



The maximum CPU loading is between 15 - 20 Percent on a Quadcore Processor.

Do you Think you can optimice it more that it would be works on a ASIO System with a latency of 1024? I dont want to move the Latency in the StereoTool itself, because i find in 4096 it sounds best.

@near05: I have one more (small) improvement that I want to add tonight, but I think the effect will be marginal. (Basically, what I'll do is multithread one extra - small - piece of code, which then allows me to combine 2 others, which in turn means that I can get rid of a step where one thread needs to wait for the other to finish. This *might* help a bit in making the behavior a bit more constant).

Now building version where 2 steps are combined into 1 step. (Number of threads is still equal).

BETA008:
Combined 2 processing steps in the declipper to remove a synchronisation step. Performance might be (marginally) better.

Winamp DSP plugin: http://www.stereotool.com/download/dsp_ ... 21-008.exe
Stand alone version: http://www.stereotool.com/download/ster ... 21-008.exe
VST version: http://www.stereotool.com/download/vst_ ... 21-008.dll
VST version (No SSE2): http://www.stereotool.com/download/vst_ ... 21-008.dll
Command line version: http://www.stereotool.com/download/ster ... 21-008.exe
Linux command line version: http://www.stereotool.com/download/ster ... ETA621-008 NOT AVAILABLE
Linux GUI version: http://www.stereotool.com/download/ster ... ETA621-008 NOT AVAILABLE
DECLIPPER Winamp DSP plugin: http://www.stereotool.com/download/dsp_ ... 21-008.exe
DECLIPPER VST version: http://www.stereotool.com/download/vst_ ... 21-008.dll

TODO:
- Create dual core support flag in Audio class
- I need to check if the new thread has the right priority
- Currently I create a thread of each portion of data that arrives, it's probably better to have a single thread process all the data.
- Found a bug in the non-SSE2 version of the clipper. Should be easy to fix.
- Now I have this, it's probably better to use Loudness as limiter of the perfect declipper output instead of the final limiter which is used now, but is less good.
- Convert declipper to use 2 cores as well. - Detection part too
- Fix memory bug in declipper clipping detection.
- Declipper: Combine 2 multicore parts into 1 (avoids an expensive 'join' phase)
- Check if thread priority is copied correctly (also for ASIO threads called from driver level).
- Combine 2 multicore threads into 1. (Move them up 1 level from Audio to Common)
- If needed (hiccups?), make it possible to disable multicore.


I'm now measuring a 38% increase in throughput.

Hello Hans,

i tried your last version. The Performance is really great. Now i can use a ASIO Latency of 1024 without any Problems and crackling. The CPU Usage in Adobe Audition is ~ 20 Percent.

The core equilibration is well-balanced. Not only one core is used very busy.

As mentioned in thread viewtopic.php?f=2&t=3950

Al testing is done on the same pc
Preset: FM Europe stereo
RDS/50us/stereo/pre-emphasize enabled
Audio is always Frank Sinatra - that's live

When using ST6.2Final DSP:



When using ST6.2Final SA:



When using ST6.21-008 DSP:



When using ST6.21-008 SA:



The beta is using more CPU power???
Sound is still crackled....

Greetings, Robert

Yeah it should, before it was limited by the fact that it only used one core - so at most 25% (+ a bit for the GUI etc.). Now it uses 2 cores so it runs faster - not fast enough though