I found the solution. See also the last posts in this thread, where somebody supported me.http://www.hydrogenaud.io/forums/index.php?showtopic=64628&st=425&start=425
There are several issues which hindered the program to run properly inside of MusicBee:
1. besides the DLL files also flac.cl is required. Its looks like C Code, so I thought this is only for docu purposes, but its actually used during runtime
2. I copied the 4 DLL files to 4 different places in the hope to get it work...
C:\Program Files (x86)\MusicBee\Codec
C:\Program Files (x86)\MusicBee
I additionally copied now the flac.cl to these 4 location. Its still an open topic for me to identify, whether codec directory is sufficient.
3. it turned out when using the converter on the commandline, then also higher optimizations (-9 - -11) are possible. But then additionally the -lax option needds to be set. Higher optimization/compression than -8 turned out not to be possible when being called from within MusicBee.
4. when being called from within Musicbee then you need another commandline option, which was not required when being used in the DOS box. On the DOS box the converter can even read properly from stdin, without the need of the additional option --ignore-chunk-sizes.
5. inside of Musicbee you need to use --ignore-chunk-sizes as parameter, and higher levels than -8 are not possible, then conversion doesnt run.
The settings inside of MusicBee to make it work were for me (-> Settings -> File Converter):
- maximum Encoder Threads: 6 up to 9 (I would say this is CPU and Disk i/o dependend)
- use the converter from within MusicBees codec subdirectory where the converter, the DLLs and the flac.cl (!!) need to reside (remember my open topic, I need to validate, that it can be this directory exclusively, at the moment I have this spread around 4 locations):
C:\Program Files (x86)\MusicBee\Codec\CUETools.FLACCL.cmd.exe
- use the following arguments: -q -8 --ignore-chunk-sizes --opencl-type GPU --opencl-platform "NVIDIA CUDA" - -o [outputfile]
6. If you have an AMD graphic card, then you might need another cmd line option: --opencl-platform "AMD Accelerated Parallel Processing"
First findings in regards to performance.
With my GTX 980 I am getting an avg. GPU utilization up to 14% with few peaks up to 20%, depends on the number of encoder threads.
When converting the same APE files to FLAC with Encoders from 1..9 for my machine ~7 encoders utilize CPU and GPU best.
Using 8-9 encoders results still in some more files, but not that dramatic.
With 1 encoders I get 18, with 2 encoders I get 32 files converted within a minute.
The difference between using 6-9 encoders is only within 4-10 files more per additional encoder.
I made a 1st real life performance test by looking how many of my APE files I get converted within 4 minutes.
With FLACCL I get 314 files of 7,36 GB size in total. CPU utilization between 51-73, GPU between 2-14%.
With normal FLAC I get only 185 files of 4,47 GB size in total. CPU utilization between 37-68%, GPU of course 0%.
To convert 314 files, which FLACCL did in 4 minutes, the normal FLAC programm needs 6m35s, which is 65% more.
The size difference between the 314 files is very low 7,36 GB (FLACCL) vs 7,37 GB (flac).
So the only motivation to use this program is for speed if you have a big conversion job or if you simply like the idea being able to offload work to the GPU.
Thanks for your response and I hope you get it running as well, as it appears to me that you also spend some time on this.