AAC Encoder: Saturate quantizer shift value to prevent
undefined behaviour.
In very rare cases the shift value 'totalShift = (16-4)-(3*(totalShift>>2))'
can be greater than accu data width. If you apply a shift with more then
31 bit the result depends on the architecture and is not defined in C.
For certain platforms zeros are shifted in. That would be our desired behaviour.
On other platforms the shift will be applied as modulo. For example >>34
would be applied as >>2. To prevent this discrepancy the shift value
is limited/saturated to DFRACT_BITS-1. 'accu >>= fixMin(totalShift,DFRACT_BITS-1)'.
Bug 9428126
Change-Id: I27177654c4dc22cf899bc35dad9cdd040dccb02d
* AAC-Encoder
- Remove source snippets related to unsupported frame length of 960.
- Adjust minBits initialization for disabled VBR configuration.
Modified file(s):
libAACenc/src/aacenc.h
libAACenc/src/aacenc_lib.cpp
libAACenc/src/aacenc_tns.cpp
libAACenc/src/bandwidth.cpp
libAACenc/src/psy_const.h
libAACenc/src/transform.h
- Revise encoder parameter restrictions: "Error Resilient(ER) AAC Low Complexity"
is not supported. Where the AAC part of the FDK encoder threw an
initialization error, now the aacEncoder_SetParam() returns an error
before initialization if the application tries to configure this
audio object type.
Modified file(s):
libAACenc/src/aacenc_lib.cpp
* SYS-Library
- Portability fix for Mingw32.
Modified file(s):
libSYS/include/machine_type.h
- Minor changes.
Modified file(s):
libSYS/include/wav_file.h
libSYS/src/wav_file.cpp
libSYS/src/cmdl_parser.cpp
libSYS/src/genericStds.cpp
Bug 9428126
Change-Id: I9fa44923d34f033d0dc607f2b85abacb8a85eb61
- Update disclaimer in all source files.
- Increment library version in each FDK sub-module.
Bug 9428126
Change-Id: I490b96d4ee472246b01483202b0bb4f1e9c2a5d7
This makes sure we read the right array element, corresponding
to all other array accesses in the same function, and all other
use of the isBook array elsewhere in the library.
This makes the fix in the previous commit ("Avoid too big changes
in isScale") work as intended, fixing other samples that triggered
assertions.
These values are delta encoded, but only steps [-60,60] are
possible to encode. If the difference between two values are
bigger than that, we end up with a sequence we are unable to
encode.
This fixes crashes on certain stereo samples at higher
bitrates.
Shifting by more than (or equal to) the variable length is
undefined in C.
This caused the quantized spectrum values to remain unshifted,
causing loud beeps in some samples.
The exact same change was originally applied in
657456949 (for AOSP) and in 24021f190 (in the separate fdk-aac
repo), but was overwritten and silently reverted by the new
upstream code drop for Android 4.2 in 381d69840a. After the code
drop, I chose not to reapply this change since the crashes that
it fixed had been fixed in other ways upstream and I was unable to
reproduce them after the new code drop.
This reverts commit e036dbc8eb.
Upstream took __builtin_clz in use (in the new header
libFDK/include/x86/clz_x86.h, together with an alternative
implementation for MSVC), so this change can be dropped to
minimize the differences.
This reverts commit 839ae290c1.
This change is no longer necessary on top of the new upstream,
I cannot reproduce the crash that this fixed with the new upstream.
Mingw32 doesn't define the __int64 types.
This was implicitly reverted in AOSP (where this change was
already merged before) by the new code drop from Fraunhofer.
From Fraunhofer:
* AAC Decoder
- Stick to the written MPEG standard instead of the MPEG reference software
in terms of reference level normalization. Always set the program reference
level equal to the target level. This disables level normalization using a
default level for streams without embedded metadata.
Modified file(s):
libAACdec\src\aacdec_drc.cpp
- Fix downmix channel assignment when using a WAV output channel ordering.
Modified file(s):
libPCMutils\src\pcmutils_lib.cpp
- Retain signal accuracy and prevent LSB alteration when no level correction
needs to be done.
Modified file(s):
libAACdec\src\aacdec_drc.h
libAACdec\src\aacdec_drc.cpp
libSBRdec\src\sbrdecoder.cpp
libSBRdec\src\sbr_dec.cpp
libSBRdec\src\sbrdec_drc.cpp
- Align metadata processing with reference implementation.
Modified file(s):
libAACdec\src\aacdec_drc.h
libAACdec\src\aacdecoder.cpp
* AAC-Encoder
- Prevent potential overflow in energy calculation after TNS processing.
Modified file(s):
libAACenc\src\band_nrg.cpp
- Added saturation for number of relevant lines which are used in pe
calculation.
Modified file(s):
libAACenc\src\line_pe.cpp
- Removed obsolete files.
Deleded file(s):
libAACenc\src\tns_param.h
libAACenc\src\tns_param.cpp
* FDK-Library
- Added x86 Count Leading Zeros intrinsic.
Modified file(s):
libFDK\include\clz.h
Added file(s):
libFDK\include\x86\clz_x86.h
- Fixed compilation for MIPS GCC-4.4 and higher.
Modified file(s):
libFDK\include\mips\cplx_mul.h
libFDK\include\mips\fixmul_mips.h
Change-Id: I4be65f07f88d412224c7fddc3f054e8f451176cc
This reverts commit 923f3e95f3.
The optimized version differed from the original in a few cases
(in some cases where the return value turned out to be negative
while the original implementation returned a positive value),
so revert it for now until it has been analyzed completely.
Addresses bug 7140347.
When flushing the decoder, you need to make sure to zero out any
lingering spectral coefficients as well as zero out the overlap buffer
state after decoding the final access unit. Zeroing out the spec
coefficients *before* you decode the last block, however, seems like a
mistake. If the final block contains any useful information for the
stream, then you end up outputting the overlap portion of the second
to last decoded block combined with zeros. Because of the shape of
the second to last block's windows, this results in what appears to be
a fade out, but actually is a zero'ed out final block.
Signed-off-by: John Grossman <johngro@google.com>
Change-Id: I0fab9ee1a5899811d5519a91dc05631e9bf4963c
Latest code drop from Fraunhofer:
* AAC-Decoder
- Expanded AAC-LD/ELD decoder TNS max band tables to avoid wrong data
access for sampling rates <22kHz and >48kHz.
Modified file(s):
libAACdec\src\aacdec_tns.cpp
libAACdec\src\aac_rom.h
libAACdec\src\aac_rom.cpp
- Fixed ELD synthesis QMF filterbank scaling for downsampled SBR.
Modified file(s):
libFDK\src\qmf.cpp
* AAC-Encoder
- Fixed bit rate limiting for lower limit as introduced in the Delivery 2012-05-11.
Modified file(s):
libAACenc\include\aacenc_lib.h
libAACenc\src\aacenc.h
libAACenc\src\aacenc.cpp
libSBRenc\src\sbr_rom.h
libSBRenc\src\sbr_rom.cpp
- Allow negative prediction gain as meaning that there is no coding gain.
Make use of scaleValueSaturate in gauss window calculation.
Modified file(s):
libAACenc\src\aacenc_tns.cpp
- Added energy saturation to prevent overflow in short blocks.
Modified file(s):
libAACenc\src\grp_data.cpp
- Perform scalefactor adaption in case quantized lines are out of valid range.
Modified file(s):
libAACenc\src\quantize.cpp
- Interrupt quantization loop when no spectral data is available.
Adjustments in bitreservoir adaption.
Modified file(s):
libAACenc\src\qc_main.cpp
- Indroduced dynamic scaling in none missing harmonic energy lowering compensation.
Modified file(s):
libSBRenc\src\env_est.cpp
* FDK-Library
- Fixed saturation for negative values in scaleValueSaturate.
Modified file(s):
libFDK\include\scale.h
Change-Id: If830ea65caef6b5554281e4b7a77a8b2e08825ce
aac project forces the system not to include sys/types.h by defining
_SYS_TYPES_H_. This is a workaround for broken android toolchain.
This is no longer needed in latest android version and can be safely
removed. This will also lead to more errors if changes in bionic
headers are needed.
Change-Id: Icb76c207d6b47e811eacff9f9403177629bfb451
Signed-off-by: Irina Tirdea <irina.tirdea@intel.com>
The library contains architecture specific versions of the clz functions
for arm and mips, but the fallback C version for other architectures is
pretty slow.
By using __builtin_clz on GCC (available since GCC 3.4), we get a
significant (20-40 %) speedup of the total running time on x86.
Speed difference: 5.1 s instead of 8.6 s for AAC-LC 128 kbit/s and
3.8 s instead of 4.9 s for HE-AACv2 32 kbit/s, on an i7.
Shifting by more than (or equal to) the variable length is
undefined in C.
This caused the quantized spectrum values to go out of range,
triggering asserts later.
Change-Id: If81b6c8caa7b9c75941ad9d280b686d2069c968c
Jean-Michel Trivi
Martin Storsjo
Elliott Hughes
Jeffrey Knockel
Sergio Ammirata
John Grossman
Keun young Park
Jakub Stachowski
Jean-Baptiste Queru
Irina Tirdea
Raghu Gandham