FLACCL: CUDA-enabled FLAC encoder by Gregory S. Chudov (prev. FlaCuda), Formerly "lossless codecs and CUDA" Options

[Wombat]

0.8 is basically a re-branded 0.7 with default compression mode changed from -5 to -7 and default mode set to --gpu-only, to provide better results for casual user who doesn't want to bother with command line switches It doesn't deserve separate testing.

flac test suite is available in flac's sources, but has to be adapted for FlaCuda. This should be easy, it was done once with flake. I'm just very lazy and didn't find time to do it.

Thanks for explaining. I should have known that since it is mentioned on the Flac pages.
btw. the first time i understand exactly what you are doing -> Being lazy is one of my favorite ambitions. Enjoy!

[glebe]

Some addition to my previous post. If use "--verify" option then

1. 4 threads are faster than 3 threads
2. --gpu-only also gives speed boost

So on 0.8 for me optimal settings are "-8 --cpu-threads 3 --verify".

Thanks and happy new year!

[skamp]

Any plans for a CUDA port of LossyWAV? Direct encoding to lossyFLAC (from .wav to .lossy.flac in one pass) would be awesome.

[Gregory S. Chudo...]

This can be done, but not very easy. Some of the features (like noise shaping) are not quite suitable for parallel processing, and quality control will be an issue, as the output will most likely be not bit-identical to the reference implementation, so some ABXing will be required. I should probably wait for a stable version.

[skamp]

I should probably wait for a stable version.

…of LossyWAV? Latest stable (1.2.0) was released last month. Also, there's a CUDA implementation (more than one? can't recall) of FFT, shouldn't that help?

This post has been edited by skamp: Jan 5 2010, 16:02

[Gregory S. Chudo...]

Oops, i missed that release. FFT is the easy part. Ok, i'll try when i have the time.

[Nick.C]

Shaping is off by default in 1.2.0. A speedup would be lost as the second 1024 point FFT for one 512 sample codec-block is the same as the first 1024 point FFT for the next codec-block. Other than that, I see no particular problems (other than keeping the processed blocks in the correct order if that level of parallelism is reached).

[Gregory S. Chudo...]

Good then, because CUDA version is likely to support only default options.

[Gregory S. Chudo...]

Thanks to Wombat for finding quite a serious bug in FlaCuda 07-08. lossyWAV preprocessed files and other audio with blocks with zero lower bits were encoded incorrectly. Bugfix release (FlaCuda 09) is now available. Please, remember not to reencode anything with FlaCuda without verifying results, until it's no longer in alpha stage.

[skamp]

Thanks, but it fails under linux again.

[Gregory S. Chudo...]

Replace CUDA.NET.dll again.
This a third-party library: http://www.hoopoe-cloud.com/Solutions/CUDA.NET/Default.aspx
Unfortunately they've chosen to make two separate versions, for windows and linux/macos.

[arri]

Hello,

problems here with CUETools.FlaCuda v0.9 on Vista

tried to convert wav to flac:

CODE

1 out of 1 tracks converted with major problems.

Source: "H:\U2 - Achtung Baby.wav"
  An error occurred while finalizing the encoding process (Object not found) : "F:\U2 - Achtung Baby.flac"
  Conversion failed: Object not found

Conversion was made by foobar 1.0 with settings:



My previous version - Fla.Cuda v.06 has worked perfectly

Any ideas ?

[Gregory S. Chudo...]

Thanks, i think i fixed fb2k pipe input, try this one.

This post has been edited by Gregory S. Chudov: Jan 10 2010, 17:32

[arri]

Thanks, i think i fixed fb2k pipe input, try this one.

Thanks!

It worked like a charm

[Wombat]

One thing i am curious about.
Flac has more or less the "Variable Blocksize" implemented in its standard but never really realized it even if my hardware players decoder can manage it. With flake once there was a small advantage in using this.
Now with Flacuda i noticed it doesn´t matter what variable blocksize option i use it leads to bigger files.

[Gregory S. Chudo...]

That's because it's not implemented yet.

[odyssey]

Thanks for this great project!

I'm very impressed by the compressed filesize using flacuda (-11)! However, I'm still unable to make it work on my Geforce 9300 mainboard

Error : Exception of type 'GASS.CUDA.CUDAException' was thrown.

[Gregory S. Chudo...]

Can you please send me results of the following deviceQuery utility:  deviceQuery.rar ( 139.56K ) Number of downloads: 132

[odyssey]

Can you please send me results of the following deviceQuery utility:  deviceQuery.rar ( 139.56K ) Number of downloads: 132

CODE

CUDA Device Query (Driver API) statically linked version
Cuda driver error 3 in file 'd:/bld_sdk10.pl/devtools/SDK10/CUDA_2.3/SDK10/Compute/C/src/deviceQueryDrv/deviceQueryDrv.cpp' in line 38.

[Gregory S. Chudo...]

Hm, either you have very old drivers or there is something wrong with your hardware. Even the most basic function cuDeviceGetCount, which should've returned the number of CUDA-capable devices in your system, fails.

[odyssey]

I'm using the latest available drivers from the nVidia homepage. Just reinstalled the entire PC a few days ago. Win7 x64.

nForce driver v15.49 WHQL containing:
GPU Driver (v190.38) WHQL

[Gregory S. Chudo...]

Well, error 3 means: CUDA_ERROR_NOT_INITIALIZED = 3, ///< Driver not initialized

[odyssey]

So, where do I go from here?

[Gregory S. Chudo...]

No idea, to be honest. If you are sure that drivers are installed correctly and that your hardware is CUDA-capable, this leaves us only with NVIDIA customer support. Because if even the basic test (deviceQuery) from NVIDIA SDK doesn't work, FlaCuda certainly won't.

[Maggi]

I'm using the latest available drivers from the nVidia homepage. Just reinstalled the entire PC a few days ago. Win7 x64.

nForce driver v15.49 WHQL containing:
GPU Driver (v190.38) WHQL

according to nVidia's driver page, you are not using the latest drivers .. try these:

http://www.nvidia.com/object/win7_winvista...96.21_whql.html

as for your graphics card, it should be able to handle CUDA, if it has at least 256MB of local memory, but to be honest, I wouldn't expect any miracles, since it seems to be equipped with a mere of 16 cores

HTH,
Maggi