On 2014-05-15 18:27:54 +0200, Christophe Gisquet wrote: > Hi, > > (message addressed to the people in the thread rather than just the poster) > > 2014-05-13 17:25 GMT+02:00 <[email protected]>: > > +static void inline ff_dct_unquantize_h263_neon(int qmul, int qadd, > > I think the vpx, rv40 and h264 decoders at least prefer dequantizing > while entropy decoding/parsing coeffs, because there are so few of > them and they are sparsely distributed. > > Therefore, I'm wondering if a SIMD implementation improves that much > things, and so, if an alternative is just to have the remaining codecs > dequantize on the fly like the aforementioned ones, and scrap the SIMD > dequantizing code.
That would probably make sense. dequantizing took around 1% on limited set of the samples I looked at. I didn't say anything because choosing simple functions for the introduction of intrinsics was imho a good idea. Janne _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
