Hi,
I'm wondering how the DCT and inverse DCT are implemented in IPP. Did you implement DCT through FFT?
For instance Matlab's implementation of DCT and invDCT are done through FFT.
If not, would it be faster to write my own function that does 2D DCT through FFT,
or DCT is already optimized?
Thanks
Andriy
I'm wondering how the DCT and inverse DCT are implemented in IPP. Did you implement DCT through FFT?
For instance Matlab's implementation of DCT and invDCT are done through FFT.
If not, would it be faster to write my own function that does 2D DCT through FFT,
or DCT is already optimized?
Thanks
Andriy
連結已複製
3 回應
Quoting - Vladimir Dudnik (Intel)
Hi Andriy,
which exactly IPP functions you are looking for?
IPP DCT functions, especially used in media processing (work on 8x8 blocks) are highly optimized in IPP.
Regards,
Vladimir
which exactly IPP functions you are looking for?
IPP DCT functions, especially used in media processing (work on 8x8 blocks) are highly optimized in IPP.
Regards,
Vladimir
I'm asking about DCTFwd and DCTInv functions for DCT of general size images. Have you used FFT subroutine to implement them?
PS: From my experiments DCTFwd with images 256x256 is 3 times faster than with images 257x257.
