Hi, 2014-05-16 11:23 GMT+02:00 Janne Grunau <[email protected]>: > That would probably make sense. dequantizing took around 1% on limited set > of the samples I looked at.
On the other hand, I didn't think of the encoders, where multiple things may justify a SIMD implementation (quantized coeff location may not be known, rate/distortion-optimized quantization may follow, ...). Plus it affects so many tests that the effort of changing to on-the-fly dequantization is probably not worth the gain (code-wise, binary-size-wise or speed-wise). Best regards, -- Christophe _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
