[tremor] Ogg/Vorbis report, FFT optimizations

[Madhav]

Hi all,

1. I need the thesis report for the http://www.sandvall.nu/thesis.pdf, Now this website do not contain this pdf. Can anyone mail me this pdf.

2. I have some questions regarding tremor. I am intended to use this for arm946e-s platform.
I made the profiling on ads 1.2 for arm946e-s processot of the tremor and found following statstics:

Name                            time%
XPROD31                        12.28%
---------------------------------------------------------------------
XNPROD31                       11.82%
----------------------------------------------------------------------
_vorbis_apply_window            8.39%
----------------------------------------------------------------------
mdct_butterfly_generic       7.36%
----------------------------------------------------------------------
mdct_backward                 7.34%
----------------------------------------------------------------------
vorbis_synthesis_blockin        6.30%
----------------------------------------------------------------------
ov_read                         6.16%
----------------------------------------------------------------------
floor1_inverse2                 5.64%
----------------------------------------------------------------------
vorbis_book_decodevv_add        4.68%
----------------------------------------------------------------------
mdct_butterfly_16             3.95%
----------------------------------------------------------------------
mdct_bitreverse               3.95%
----------------------------------------------------------------------
decode_packed_entry_number      3.91%
----------------------------------------------------------------------
XPROD32                        2.88%
----------------------------------------------------------------------
mapping0_inverse              2.73%
----------------------------------------------------------------------
mdct_butterflies              1.75%
----------------------------------------------------------------------
oggpack_look                    1.62%
----------------------------------------------------------------------
oggpack_adv                     1.44%
----------------------------------------------------------------------
mdct_butterfly_8              1.10%
----------------------------------------------------------------------
__rt_memcpy_w                   1.06%
----------------------------------------------------------------------
res2_inverse                    0.96%
----------------------------------------------------------------------
floor1_inverse1                 0.88%
----------------------------------------------------------------------
__rt_memclr_w                   0.75%

Here functions in bold is for the mdct, it is taking more than 50 % of the ogg-vorbis decode time.
How fft based method is used to optimise this further?

3. I am trying to compute the mips? The sizes supported by ogg are 256 or 2048

So i compute the mips as:

  MIPS (taken by ov_read) = [cycles taken by ov_read ( ) / sample] * sampling frequency * 10^-6

    Example: cycles taken by ov_read ( ) is say 20,00,000

  MIPS (taken by ov_read) = [cycles taken by ov_read ( ) / sample] * sampling frequency * 10^-6

 MIPS taken = (20,00,000 / 2048) * 44100 * 10^-6
                    = 43.06640625 MIPS


Is this interpretation correct?

I am waiting for the help from you all.

With Regards,

Madhav Kulkarni





 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.xiph.org/pipermail/tremor/attachments/20050302/2563e105/attachment.html