Open yijinsheng opened 1 week ago
It's possible nvCOMP allocates memory as work area that we need to deallocate some other way...
The memory you see is related to the CUDA state (the primary CUDA context, etc). You can see this memory get allocated when running cudaSetDevice()
, which initializes CUDA. If you don't call that, CUDA will automatically initialize itself upon the first call to a CUDA function (in this case the call to cudaMalloc()
).
To manually free this memory you can use cudaDeviceReset()
, though it doesn't seem to be necessary.
I have follow the example nvcompLZ4Example to compress my yuv data using 3090TI. but I have 2 questions about this.
I print out the compress and decompress time, while the decompress time is much longer than the comress time,this doesn't make sense Here is my code: