Re: [mpg123-devel] fixed point decoders

[Taihei M. <tm...@ma...>]

On 2009/05/31, at 0:27, Thomas Orgis wrote:

> A notable exception is the layer3 test... there we barely scratch  
> over the limit (around 1.4e-4).
> And another notable exception is that one layer2 test that gets  
> around 5e-2 RMS error.
> It falls out of the usual pattern... I somehow hope that this means  
> we can fix this one.

I've tuned the integer decoder, and now the layer3 decoder reaches  
"limited" accuracy. Check the latest trunk.

==== Layer 1 ====
--> 16 bit signed integer output
fl1.bit:        RMS=9.153515e-05 (LIMITED) maxdiff=3.705621e-04 (FAIL)
fl2.bit:        RMS=9.362248e-05 (LIMITED) maxdiff=4.243851e-04 (FAIL)
fl3.bit:        RMS=9.387823e-05 (LIMITED) maxdiff=3.730059e-04 (FAIL)
fl4.bit:        RMS=7.200389e-05 (LIMITED) maxdiff=3.411770e-04 (FAIL)
fl5.bit:        RMS=1.051527e-04 (LIMITED) maxdiff=4.037619e-04 (FAIL)
fl6.bit:        RMS=1.180426e-04 (LIMITED) maxdiff=4.869699e-04 (FAIL)
fl7.bit:        RMS=8.529281e-05 (LIMITED) maxdiff=3.802776e-04 (FAIL)
fl8.bit:        RMS=9.816843e-05 (LIMITED) maxdiff=4.596114e-04 (FAIL)

==== Layer 2 ====
--> 16 bit signed integer output
fl10.bit:       RMS=1.146784e-04 (LIMITED) maxdiff=4.718304e-04 (FAIL)
fl11.bit:       RMS=1.132384e-04 (LIMITED) maxdiff=4.780293e-04 (FAIL)
fl12.bit:       RMS=1.120420e-04 (LIMITED) maxdiff=4.792809e-04 (FAIL)
fl13.bit:       RMS=1.054132e-04 (LIMITED) maxdiff=3.585219e-04 (FAIL)
fl14.bit:       RMS=1.204947e-04 (LIMITED) maxdiff=3.799021e-03 (FAIL)
fl15.bit:       RMS=5.289819e-02 (FAIL) maxdiff=3.792340e-01 (FAIL)
fl16.bit:       RMS=1.202716e-04 (LIMITED) maxdiff=8.831620e-04 (FAIL)

==== Layer 3 ====
--> 16 bit signed integer output
compl.bit:      RMS=1.303472e-04 (LIMITED) maxdiff=1.008749e-03 (FAIL)

fl15.bit is still weird... I should look into more.

> Obviously that build is not using the high-quality tuned MAD  
> library, but the fast setup with limited accuracy.
> What disturbs me about that: This code is faster than the generic  
> mpg123 code, be it with floats or integers!
> Granted, the floating point output of mpg123 is of higher quality,  
> but MAD kicks our ass on the quick-and-dirty track.
>
> Seems like we'd need to add fixed point assembler optimizations to  
> catch up (MAD has some assembler stuff, particulary for ARM, too).

Asm optimization will help, but I think using the same codebase for  
both integer and fp decoder is disadvantageous in terms of performance  
(of the int decoder).

Thanks,
Taihei Monma