<div dir="ltr">To me the real question is not whether that portion of the spec has been implemented by any existing encoders/decoders but whether the spec is broken (i.e. cannot be implemented as written). I don't know the rationale for making the LPC shift explicitly signed. In C a negative shift is undefined and it does seem in FLAC__lpc_restore_signal() for example that the LPC shift is used as the argument to a right shift operation. It's possible (generally/conceptually, not necessarily here) a negative shift value could be used to represent a left shift. However, I know very little about linear prediction and how coefficients are chosen and whether that makes sense. If it really is a design flaw in the spec then it makes sense to change it or document that negative shifts are not supported by any known implementation as you suggest.<div><br></div><div>Philosophically my objection to changing the spec based on lack of known implementations is that it could be artificially limiting, for example the same argument about breaking existing decoders could be made for 32-bit samples sizes.</div><div><br></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Mon, Jun 22, 2020 at 1:33 AM Martijn van Beurden <<a href="mailto:mvanb1@gmail.com">mvanb1@gmail.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-style:solid;border-left-color:rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div>Yes, this is such a case. However, implementing this in a future encoder/decoder would break compatibility with most (likely all) existing decoders, and only in some very, very rare cases where the material is such that the encoder chooses to use negative shifts, which makes it even harder to troubleshoot. Furthermore, as this can only be used in very rare cases, there is no benefit from allowing this.</div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">Op vr 19 jun. 2020 om 18:03 schreef Stephen F. Booth <<a href="mailto:me@sbooth.org" target="_blank">me@sbooth.org</a>>:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-style:solid;border-left-color:rgb(204,204,204);padding-left:1ex"><div dir="ltr">Is this a case where something allowed by the specification isn't implemented by the reference encoder/decoder (such as 25-32 bits per sample) but could be in a different implementation? If so, I am not sure whether it makes sense to change the specification based on the reference implementation.<div><br></div><div>Stephen</div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Wed, Jun 17, 2020 at 3:22 PM Martijn van Beurden <<a href="mailto:mvanb1@gmail.com" target="_blank">mvanb1@gmail.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-style:solid;border-left-color:rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div>Hi all,</div><div><br></div><div>When trying to better understand the way LPC exactly works, I stumbled upon something which, after some digging, was already reported and (partly) fixed: <a href="https://sourceforge.net/p/flac/bugs/424/" target="_blank">https://sourceforge.net/p/flac/bugs/424/</a></div><div><br></div><div>Apparently, the FLAC specification has a LPC shift that can be both positive and negative, but the encoder specifically makes sure that only positive shifts are encoded and the decoder only accepts positive shifts. The ffmpeg FLAC encoder and decoder show the same behaviour.</div><div><br></div><div>Now, the documentation in the source code is fixed, the documentation on the website (which I was looking at) isn't yet. The website format.html states: "Quantized linear predictor coefficient shift needed in bits (NOTE: this number is signed two's-complement)." The source code format.html says "Quantized linear predictor coefficient shift needed in bits (NOTE: this number is signed two's-complement;
but, due to implementation details, must be non-negative)."</div><div><br></div><div>I was thinking of submitting a patch to the FLAC website git to get the website format.html up-to-date (there have been more changes than just this one), but I feel the line above isn't clear enough. Maybe change it to something like this, to make the wording more similar to the rest of the specification<br></div><div><br></div><div>Quantized linear predictor coefficient shift needed in bits (NOTE: these bits must be 00000-01111. Originally this was a signed integer, but negative shifts were never implemented). <br></div><div><br></div><div>Or perhaps:</div><div><div><br></div><div>Quantized linear predictor coefficient shift needed in bits (NOTE: First bit must be zero. Originally this was a signed integer, but negative shifts were never implemented). <br></div><div><br></div><div>Any thoughts?</div><div><br></div><div>Kind regards, Martijn van Beurden<br></div></div></div>
_______________________________________________<br>
flac-dev mailing list<br>
<a href="mailto:flac-dev@xiph.org" target="_blank">flac-dev@xiph.org</a><br>
<a href="http://lists.xiph.org/mailman/listinfo/flac-dev" rel="noreferrer" target="_blank">http://lists.xiph.org/mailman/listinfo/flac-dev</a><br>
</blockquote></div>
</blockquote></div>
</blockquote></div>