Jean-Marc Valin
c4a40c65c5
doing spreading function and gain quantisation in-place
2008-03-05 23:22:00 +11:00
Jean-Marc Valin
6441379e57
fixed-point: part of stereo_mix() converted.
2008-03-05 23:00:21 +11:00
Jean-Marc Valin
ab69667495
oops, unb0rked testcelt output
2008-03-05 22:40:58 +11:00
Jean-Marc Valin
9d312b9756
Defining RADIX_TWO_ONLY removes all butterflies, except for radix 2 and 4.
2008-03-05 17:34:45 +11:00
Jean-Marc Valin
f93747c44a
fixed-point: changed find_spectral_pitch() to use single-precision (16-bit) FFT.
...
This involved adding kfft_single.[ch] that redefines kiss_fft a second time
with a different prefix. All this is still a bit of a mess now. The mask
had to be converted to 16-bit input, but we're still using floats to apply it.
2008-03-05 17:20:30 +11:00
Jean-Marc Valin
4024444811
nothing to see here
2008-03-05 15:18:53 +11:00
Jean-Marc Valin
b624cdd2ce
fixed-point: masking curve computation now converted. ***Fixed a bug in the
...
handling of the left side slope of the spreading function (right-side coef
was used)
2008-03-05 13:40:20 +11:00
Jean-Marc Valin
b0c93c66ea
oops, pgain_table now has half the number of elements
2008-03-05 13:12:16 +11:00
Jean-Marc Valin
afa5eb38b5
fixed-point: quantised the pitch gain table.
2008-03-05 12:22:37 +11:00
Jean-Marc Valin
92376697c6
fixed-point: converted the pitch gain quantisation, except for the codebook
...
itself
2008-03-05 11:31:57 +11:00
Jean-Marc Valin
a78dc946cd
Making 16-bit compilers happy
2008-03-05 08:15:55 +11:00
Jean-Marc Valin
d9de593104
Fixed a bunch of warnings
2008-03-05 08:11:57 +11:00
Jean-Marc Valin
a1bc18a388
compute_band_energies() merged with normalised_bands()
2008-03-04 22:52:52 +11:00
Jean-Marc Valin
069c3de65f
Removed deprecated mode interface and added missing include
2008-03-04 20:38:00 +11:00
Jean-Marc Valin
7137c26e4f
fixed-point: exp_rotation() now fully converted, using an approximation of the
...
cos() and sin() functions.
2008-03-04 17:43:09 +11:00
Jean-Marc Valin
7e94cc4d95
fixed-point: converted denormalise_bands()
2008-03-04 17:31:18 +11:00
Jean-Marc Valin
a31899f0e2
added suffix to real fft. The celtclient can now be built with an unmodified
...
speexdsp library.
2008-03-03 22:28:23 +11:00
Jean-Marc Valin
300cb5b86a
First attempt at adding a suffix to kiss_fft functions so we can have multiple
...
copies of the FFT.
2008-03-03 17:52:11 +11:00
Jean-Marc Valin
bc60f8b346
making sure __GNUC_PREREQ() is only used when __GNUC__ is defined
2008-03-03 17:23:24 +11:00
Jean-Marc Valin
0758689246
pre-emphasis coef now a constant
2008-03-03 17:19:17 +11:00
Jean-Marc Valin
312132780e
A bunch of const qualifyers and a few comments
2008-03-03 16:52:54 +11:00
Jean-Marc Valin
730c6e392d
Applying the window properly in find_spectral_pitch()
2008-03-03 16:38:29 +11:00
Jean-Marc Valin
ddb181b75e
A bit more reordering in stack memory allocation -- saved 1024 bytes on peak use
2008-03-03 14:53:47 +11:00
Jean-Marc Valin
137ec8e9b8
applying the pitch windowing directly in find_spectral_pitch()
2008-03-03 14:44:42 +11:00
Jean-Marc Valin
05080b4b9b
No longer storing the zero parts of the in[] array in the encoder
2008-03-03 14:23:51 +11:00
Jean-Marc Valin
632f218b4b
Pitch predictor now uses a larger range of offset by not being limited to the
...
window size when parts of the window are zero.
2008-03-03 14:08:21 +11:00
Jean-Marc Valin
0141723d30
compute_mdcts() no longer reads the part of the input that should be zeros
...
anyway (check-point for pitch improvements to follow)
2008-03-03 13:59:55 +11:00
Jean-Marc Valin
3e08a88389
Only the (rising) overlap part of the window is stored now. No need to hold the
...
zeros and ones (nor even the second half).
2008-03-03 13:49:20 +11:00
Jean-Marc Valin
f28062fe6a
Don't bother multiplying when the window's value is one or zero
2008-03-03 13:42:23 +11:00
Jean-Marc Valin
227ad2f82d
cwrs32-test should be 16-bit clean now.
2008-03-03 10:07:19 +11:00
Jean-Marc Valin
4c2787a45c
Laplace encoder now works with 16-bit frequencies
2008-03-03 10:04:48 +11:00
Jean-Marc Valin
eb5a245ac8
Hadn't realised the bitr* stuff wasn't needed.
2008-03-03 04:11:49 +11:00
Jean-Marc Valin
7ff2cbc1bd
oops, fixed the Laplace testcase
2008-03-02 21:25:22 +11:00
Jean-Marc Valin
37942649cc
Saved 4 kB of stack usage in find_spectral_pitch() by doing the FFT in-place
...
(sort of)
2008-03-01 22:55:27 +11:00
Jean-Marc Valin
7cdc5a34a4
Making it easier to debug laplace testcase
2008-03-01 20:56:17 +11:00
Jean-Marc Valin
a556694e12
Stereo pitch search no longer requires twice the memory/complexity.
2008-03-01 20:14:20 +11:00
Jean-Marc Valin
86cb7e8600
oops. Another 16-bit fix
2008-03-01 08:45:55 +11:00
Jean-Marc Valin
9e006b6dc0
16-bit fix
2008-03-01 08:33:51 +11:00
Jean-Marc Valin
d748cd5570
Another C90-fying pass. Fixed some warnings in the process.
2008-03-01 07:27:03 +11:00
Jean-Marc Valin
81b38c2295
Moved mdct state and window to the mode so it can be shared between multiple
...
encoders and decoders.
2008-02-29 21:08:49 +11:00
Jean-Marc Valin
37131e3ab3
Just realised half the twiddles in the real FFT weren't even used. Removing
...
them saves another 1kB for fixed-point (2kB for float or full precision)
2008-02-29 20:03:41 +11:00
Jean-Marc Valin
8d4ac155e6
Saved on stack usage by changing the order of the allocation
2008-02-29 17:24:02 +11:00
Jean-Marc Valin
9b9e98625c
Bit of memory reduction in the pitch search. More needed.
2008-02-29 17:07:53 +11:00
Jean-Marc Valin
cac91ec4db
Fixed temp arrays that were allocated too large (no change in peak mem)
2008-02-29 17:03:34 +11:00
Jean-Marc Valin
877b1975b6
fixed-point: converted intra prediction and folding, unb0rked mixed-precision
2008-02-29 16:40:39 +11:00
Jean-Marc Valin
2aaa0fee93
fixed-point: overflow debugging now works again.
2008-02-29 16:07:40 +11:00
Jean-Marc Valin
8600f69f79
Initial support for a managed stack/scratchpad. Still needs some work.
2008-02-29 15:14:12 +11:00
Jean-Marc Valin
8b2a59235f
fixed-point: energy quantisation is entirely converted now.
2008-02-29 00:32:51 +11:00
Jean-Marc Valin
adf87e261c
fixed-point: more energy quantisation stuff.
2008-02-29 00:15:10 +11:00
Jean-Marc Valin
47bf175f29
fixed-point: a few more bits converted from the energy quantisation. Almost done
2008-02-29 00:07:00 +11:00
