huangjie18 / shineenc Goto Github PK

libShine-fxp Fixed-point MP3 Encoder. 

---------From Original Readme----------

Shine (RISC OS fixed point port)

This port of Gabriel Bouvigne's Shine has been converted to fixed point
to reduce encoding time on Acorn/RISC OS computers without hardware
floating point coprocessors.

For use on other platforms, the RISC OS specifics in main.c should be
undefined, and the assembler fractional multiply functions will need to
be changed. Big endian support may also have to be re-instated.

Pete Everett. 9/4/01

email: [email protected]
web: http://www.everett9981.freeserve.co.uk


---------------------------------------------


Ported to GCC StrongARM Linux by Patrick Roberts for Lower Mars, LLC.   This
port will also work on X86, X86_64 and probably other platforms and CPUs, but
the mult_noarch_gcc.h needs to be fixed as the math is bad. (I only need x86 to
make testing easier... ARM code is 100% correct).


Tested Compiler:

    GCC VERSION 3.3.6
    --target=arm-linux
    --with-cpu=iwmmxt
    --with-float=soft (no float in Shine anyway)
    
    Compiled with the dev-ezx toolchain from the dev-ezx directory.
    

CHANGES from Peter Everett's version:

    - Removed global data from all but main.c
    - Created a DSO and static library from encoding functions
    - Ported assembly to gcc 3.6 StrongARM inline format
    - Changed file read/write functions so no direct file i/o
      is done by library functions (function pointers now)
    - All printf's in lib blocked if #ifdef DEBUG
    - Makefile for StrongARM/EZX and for generic platforms
    - Restored missing COPYING from original floating point Shine source
    - Moved RISC_OS-specific files to "unused" directory
    - Removed RISC_OS stuff from Main.c (shineenc), which is now more
      an example for library usage.

What could be improved:

    - "noarch-gcc" defines are not correct math (just close...)
    - Vectorize code for WMMX-capable CPUs.
    - Optimized assembly (mult_XXX_YYY.h) files