How to decode with CUDA hardware instead of CPU? #43

I'll look into it closer later, but I think I made a mistake above. It should be hwaccel_in as hwaccel FFmpeg option is an input option. Also, could you fill in on what you mean by "it couldn't work to sample block"? If there is an error message, please post it. Thanks.

Confirmed.

import ffmpegio as ff

hls_url = "tests/assets/sample.mp4"
BLOCK_SIZE = 10
SAMPLE_RATE = 8000

with ff.open(
    hls_url,
    "rva",
    blocksize=BLOCK_SIZE,
    s=(640, 360),
    ac=1,
    sample_rate=SAMPLE_RATE,
    sample_fmt="flt",
    hwaccel_in="cuda",
    show_log=True,
    loglevel='debug'
) as stream:
    for f in stream: # audio & video stream raw data are stored in a dict with stream ids as the keys
        print(f.keys()) 
        exit()

If you turn on show_log and set the FFmpeg loglevel to 'debug', it confirms that you are loading the NV decoder:

[AVHWDeviceContext @ 000002a7c70f7e40] Loaded lib: nvcuda.dll
[AVHWDeviceContext @ 000002a7c70f7e40] Loaded sym: cuInit
[AVHWDeviceContext @ 000002a7c70f7e40] Loaded sym: cuDriverGetVersion
[AVHWDeviceContext @ 000002a7c70f7e40] Loaded sym: cuDeviceGetCount
[AVHWDeviceContext @ 000002a7c70f7e40] Loaded sym: cuDeviceGet
[AVHWDeviceContext @ 000002a7c70f7e40] Loaded sym: cuDeviceGetAttribute
...

hwaccel_in must be used instead of hwaccel or else read errors out due to:

Option hwaccel (use HW accelerated decoding) cannot be applied to output url - -- you are trying to apply an input option to an output file or vice versa. Move this option before the file it belongs to.

Thank you. It's working properly. I couldn't see that in the official documentation, so it would be nice to see it added.

👍

It is actually mentioned in the documentation... but yeah, I'm expecting the users to be well-versed in FFmpeg options, but they are definitely not most obvious for those who have not used it before. Maybe I should create a FAQ covering hardware acceleration...