Commit FS#12130 - Fix ADX decoding on 64-bit systems by Sean Bartell. Fixes an incorrect assumption in the ADX decoder that sizeof(int) == 32 when performing fixed point math. Update comments in the fixed point library to correct this.

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@29927 a1c6a512-1295-4272-9138-f99709370657

apps/codecs/adx.c

@@ -129,7 +129,7 @@ enum codec_status codec_run(void)
         
         fp_sincos((unsigned long)phasemultiple,&z);
 
-        a = (M_SQRT2*big28)-(z*big28/LONG_MAX);
+        a = (M_SQRT2*big28) - (z >> 3);
 
         /**
          * In the long passed to fsqrt there are only 4 nonfractional bits,

apps/codecs/lib/fixedpoint.h

@@ -33,7 +33,7 @@
  *          where phase is a 32 bit unsigned integer with 0 representing 0
  *          and 0xFFFFFFFF representing 2*pi, and *cos is the address to
  *          a long signed integer.  Value returned is a long signed integer
- *          from LONG_MIN to LONG_MAX, representing -1 to 1 respectively.
+ *          from -0x80000000 to 0x7fffffff, representing -1 to 1 respectively.
  *          That is, value is a fixed point integer with 31 fractional bits.
  *
  *  Take square root of a fixed point number:

apps/plugins/lib/fixedpoint.h

@@ -33,7 +33,7 @@
  *          where phase is a 32 bit unsigned integer with 0 representing 0
  *          and 0xFFFFFFFF representing 2*pi, and *cos is the address to
  *          a long signed integer.  Value returned is a long signed integer
- *          from LONG_MIN to LONG_MAX, representing -1 to 1 respectively.
+ *          from -0x80000000 to 0x7fffffff, representing -1 to 1 respectively.
  *          That is, value is a fixed point integer with 31 fractional bits.
  *
  *  Take square root of a fixed point number: