Commits · d9fb8a6651ea66f18f034fc70a51cfed17d3e12f · Alexander Traud / Opus

Jan 30, 2014
- Optimizing divisions with a signed numerator · d9fb8a66
  Jean-Marc Valin authored 11 years ago
  
  d9fb8a66
Jan 29, 2014
- Speeding up ec_tell_frac() · 51231750
  Jean-Marc Valin authored 11 years ago
  
  51231750
Jan 23, 2014
- Speeding up extract_collapse_mask() slightly · c543ec1b
  Jean-Marc Valin authored 11 years ago
  
  c543ec1b
- Using SSAT in SIG2WORD16() on ARMv6 · 2e6876c0
  Jean-Marc Valin authored 11 years ago
  
  2e6876c0
Jan 21, 2014
- Unrolled version of the comb filter for ARM (eliminates MOVs) · 9b3a3ad7
  Jean-Marc Valin authored 11 years ago
  
  9b3a3ad7
- Making decode_pulses() compute the L2-norm on the fly · 6218cba2
  Jean-Marc Valin authored 11 years ago
  
  6218cba2
- Speed up the comb filter on ARM by using MAC16_32_Q16() · 17b19783
  Jean-Marc Valin authored 11 years ago
  
  17b19783
- Save more integer divisions on ARM when we know the operands are positive · 29354ff6
  Jean-Marc Valin authored 11 years ago
  
  29354ff6
Jan 20, 2014
- Fixes use of uninitialized values in dynalloc_analysis() · ce1173c7
  Jean-Marc Valin authored 11 years ago
  
  ce1173c7
- Using a table on ARM for unsigned division by small (<=256) integers. · ec5d01cb
  Jean-Marc Valin authored 11 years ago
  
  Saves 0.6% for 64 kb/s and 1.8% for 128 kb/s when decoding on arm7tdmi.
  ec5d01cb
Jan 19, 2014
- Minor cleanup in dynalloc_analysis() · 379af35f
  Jean-Marc Valin authored 11 years ago
  
  379af35f
- Adds a median filter to make dynalloc_analysis() more conservative · 05548fa1
  Jean-Marc Valin authored 11 years ago
  
  This should prevent extreme dynalloc behaviour in cases where some bands are heavily attenuated.
  05548fa1
Jan 08, 2014
- Fix declaration after statement in fixed point. · a65db56e
  Gregory Maxwell authored 11 years ago
  
  a65db56e
- pseudostack instrumentation (off by default) · e7750904
  Jean-Marc Valin authored 11 years ago
  
  e7750904
Jan 07, 2014
- Fixes SMALL_FOOTPRINT for float · 9134e96c
  Jean-Marc Valin authored 11 years ago
  
  9134e96c
- Don't allocate pulses on the stack when calling the SILK PLC. · e17ca256
  Jean-Marc Valin authored 11 years ago
  
  Also minor C89 fix for the previous commit
  e17ca256
- Moves CELT PLC pitch search to a separate function to reduce peak stack · b63e7110
  Jean-Marc Valin authored 11 years ago
  
  b63e7110
- Adds SMALL_FOOTPRINT hack to the CELT PLC too · 5f807c17
  Jean-Marc Valin authored 11 years ago
  
  5f807c17
- Moves deemphasis() call out of celt_decode_lost() to reduce peak stack · 9d1b6fef
  Jean-Marc Valin authored 11 years ago
  
  9d1b6fef
Jan 06, 2014
- Cleaning up leftovers of "freq" in celt_decode_with_ec() · ad8371d1
  Jean-Marc Valin authored 11 years ago
  
  ad8371d1
- Reduces the decoder stack use by removing the pcm_silk buffer in fixed-point · 4d07b135
  Jean-Marc Valin authored 11 years ago
  
  We only keep when concealing less than 10ms with SILK.
  4d07b135
- Moves the remains of compute_inv_mdcts() to celt_synthesis() · 14ca4ed6
  Jean-Marc Valin authored 11 years ago
  
  14ca4ed6
- Hack that makes the SMALL_FOOTPRINT CELT decoder use only 4.25 kB of stack. · 32454dca
  Jean-Marc Valin authored 11 years ago
  
  32454dca
- Reduces decoder stack usage by only storing one channel of denormalized MDCT · bdc7b933
  Jean-Marc Valin authored 11 years ago
  
  bdc7b933
- Some cleaning up of the synthesis code. · 4a6744a4
  Jean-Marc Valin authored 11 years ago
  
  4a6744a4
Jan 05, 2014
- Making exp_rotation1() use MAC16_16(), which saves a few cycles on ARM · ed01a596
  Jean-Marc Valin authored 11 years ago
  
  ed01a596
Jan 04, 2014
- Silences unused parameter warning · ccec752a
  Jean-Marc Valin authored 11 years ago
  
  ccec752a
- Moving the radix-2 to expose trivial twiddle factors · ef0eac49
  Jean-Marc Valin authored 11 years ago
  
  ef0eac49
Dec 31, 2013
- Improving the accuracy of the fixed-point radix-3 and radix-5 · c8f62a4a
  Jean-Marc Valin authored 11 years ago
  
  c8f62a4a
Dec 29, 2013
- Minor cleanup -- nothing to see here · e1f84620
  Jean-Marc Valin authored 11 years ago
  
  e1f84620
- Fixed-point: slight accuracy improvement in the comb filter · 05291fd6
  Jean-Marc Valin authored 11 years ago
  
  05291fd6
- Remove a SAVE_STACK that was pasted accidentally in the previous commit · 30f52cbe
  Jean-Marc Valin authored 11 years ago
  
  30f52cbe
- Unifying scaling of fixed-point and float FFT · e1dc1e22
  Jean-Marc Valin authored 11 years ago
  
  e1dc1e22
- Fixes C89 issue · dbb96ab5
  Jean-Marc Valin authored 11 years ago
  
  dbb96ab5
- Getting rid of some negations · 4c1a90a8
  Jean-Marc Valin authored 11 years ago
  
  Since we're doing two rotations, we can invert the sign on both. Also adding a few comments for optimizing the FFT.
  4c1a90a8
- Slightly improving the accuracy of the fixed-point MDCT downscale · cc344fb8
  Jean-Marc Valin authored 11 years ago
  
  Also simplifying the code
  cc344fb8
Dec 27, 2013
- Commit 99968abb was causing us to allocate too much stack in the MDCT · e0c00e27
  Jean-Marc Valin authored 11 years ago
  
  e0c00e27
- Removes the separate 1/8N rotation in the (I)MDCT and unmerges the MDCT sizes · e43a0abe
  Jean-Marc Valin authored 11 years ago
  
  Undoes commits f7547a4e and 72513f3c
  e43a0abe
Dec 23, 2013
- Inverse MDCT no longer requires any scratch space · a5e3c8a6
  Jean-Marc Valin authored 11 years ago
  
  a5e3c8a6
Dec 22, 2013

Reverse the ordering of the FFT stage to optimize a degenerate radix-4 case. · e2bcb3fe

Jean-Marc Valin authored 11 years ago

This also happens to increase the accuracy since it appears that the new
ordering is optimal (at least for 20 ms frames), whereas the previous ordering
was pessimal.

e2bcb3fe