Tau's True Audio (TTA) codec rewrite
TTA is a lossless audio codec designed around realtime hardware coding. The codec has no tunable parameters, and each input produces a unique output. It supports 8, 16, and 24-bit PCM.
http://tausoft.org/wiki/True_Audio_Codec_Overview
How ttaR is faster than ttaenc (2024-12-28) https://gist.github.com/stseelig/0b9afe795a547b5798587db996694f97
The library is platform agnostic. The program is POSIX native and has a mostly native (mostly working) Win32 port via MinGW.
2's complement integers are assumed.
While developing, I mostly tested on:
-
AMD Ryzen 7 1700 (primary tuning)
-
Intel Celeron N2830 (secondary tuning)
honorable mentions:
-
IBM PowerPC 7400 (intrinsics)
-
ARM Cortex-A53 (intrinsics)
Unix
$ clang -O3 -DNDEBUG -nolibc -ffreestanding -fPIC -shared ./src/build_lib.c -o libttaR.so
$ clang -O3 -DNDEBUG ./src/build_cli.c -o ttaR -L./ -lpthread -lttaR
Windows
$ mingw32-clang -O3 -DNDEBUG -ffreestanding -fPIC -shared ./src/build_lib.c -o libttaR.dll
$ mingw32-clang -O3 -DNDEBUG ./src/build_cli.c -o ttaR.exe -L./ -lttaR
gcc(1) would also work, but it produces a much slower binary than clang(1).
NDEBUG
- disables all debug assertions
LIBTTAr_OPT_DISABLE_UNROLLED_1CH
- disables the unrolled mono loop
LIBTTAr_OPT_DISABLE_UNROLLED_2CH
- disables the unrolled stereo loop
LIBTTAr_OPT_DISABLE_MCH
- disables the general/multichannel loop
LIBTTAr_OPT_DISABLE_SIMD_INTRINSICS
- disables SIMD intrinsics
LIBTTAr_OPT_SLOW_CPU
-
for weak and/or old CPUs (specifically the Intel Celeron N2830)
-
fine-grained suboptions in 'src/lib/common.h'
$ ttaR encode file.(wav|w64)
$ ttaR decode file.tta
By default, ttaR will multithread with the number of coder threads equal to the number of online processors.
Read the man page (./man/ttaR.1) for a full list of the options.
This project started as a fork of ttaenc 3.4.1-linux (2007). Any code from ttaenc was put into the library, and the CLI program was completely rewritten.
The library prioritizes speed. So, its memory usage is a bit bloated to avoid some bounds checks. (The safer alternative is maybe 10-20% slower.)
The library is just a collection of functions for reading from one buffer and writing to another with some support functions to calculate values. It does not allocate, print, nor make any other syscall. (Though it can abort, but should not, if NDEBUG is not defined.) It does not even need libc as long as your compiler has a builtin memset and memmove (both, ie, not gcc). So fairly bare-bones as far as codec libraries go.
The two codec functions are reentrant, so you can code as few samples at a time with buffers as small as you want (with some padding). ttaR (CLI util) 1.0 did this, but I replaced it with doing the whole frame at once in 1.1 because it's a little faster and it simplified the source a bit.