[flac-dev] "keep qlp coeff precision such that only 32-bit math is required"

[Brian Willoughby]

Hmm, I don't know the history of the code, but flac 1.2.1 stream_encoder.c has

									min_qlp_coeff_precision = FLAC__MIN_QLP_COEFF_PRECISION;
									/* try to ensure a 32-bit datapath throughout for 16bps(+1bps for side channel) or less */
									if(subframe_bps <= 17) {
										max_qlp_coeff_precision = min(32 - subframe_bps - lpc_order, FLAC__MAX_QLP_COEFF_PRECISION);
										max_qlp_coeff_precision = max(max_qlp_coeff_precision, min_qlp_coeff_precision);
									}

in process_subframe_(), whereas evaluate_lpc_subframe_() has the comments that you refer to.

I've not done an analysis to determine whether these constants should match, or if maybe they should remain distinct.

Brian Willoughby


On Apr 18, 2015, at 8:05 AM, lvqcl <lvqcl.mail at gmail.com> wrote:
> stream_encoder.c has the following code:
> 
> 
> 	/* try to keep qlp coeff precision such that only 32-bit math is required for decode of <=16bps streams */
> 	if(subframe_bps <= 16) {
> ...
> 
> 
> But FLAC can convert 16-bit input to 17-bit if mid-side coding is used.
> So, does it make sense to compare subframe_bps with 17?
>