I couldn't help myself... here is the option for a build using the generic decoder that produces 32bit float output.

You select it via --with-cpu=generic_float, which implies dummy output and also disables the buffer.
The decoded audio can be written raw or to a IEEE float wav file.

This demonstrates that it's doable and not too complicated for the generic code.
Next would be to a) make it into a runtime option and b) adapt assembler opts to that.
The latter will most likely not happen for every opt.



git-svn-id: svn://scm.orgis.org/mpg123/trunk@668 35dc7657-300d-0410-a2e5-dc2837fedb53

configure.ac

@@ -93,6 +93,7 @@ AC_ARG_ENABLE(fifo,
 
 AC_ARG_WITH([cpu], [
   --with-cpu=generic[_fpu]      Use generic processor code with floating point arithmetic
+  --with-cpu=generic_float      A special build with generic fpu code that produces 32bit float output
   --with-cpu=generic_nofpu      Use generic processor code with fixed point arithmetic (p.ex. ARM, experimental)
   --with-cpu=i386[_fpu]         Use code optimized for i386 processors with floating point arithmetic
   --with-cpu=i386_nofpu         Use code optimized for i386 processors with fixed point arithmetic (experimental)
@@ -393,6 +394,9 @@ case $cpu_type in
   generic)
     CFLAGS="$CFLAGS -DOPT_GENERIC -DREAL_IS_FLOAT"
   ;;
+  generic_float)
+    CFLAGS="$CFLAGS -DOPT_GENERIC -DFLOATOUT -DNOXFERMEM -DREAL_IS_FLOAT"
+  ;;
   generic_fpu)
     CFLAGS="$CFLAGS -DOPT_GENERIC -DREAL_IS_FLOAT"
   ;;
@@ -444,7 +448,11 @@ case $cpu_type in
   ;;
 esac
 
-CPU_TYPE_LIB="libcpu_$cpu_type.a"
+if test $cpu_type = "generic_float"; then
+	CPU_TYPE_LIB="libcpu_generic.a"
+else
+	CPU_TYPE_LIB="libcpu_$cpu_type.a"
+fi
 AC_SUBST( CPU_TYPE_LIB )
 
 
@@ -503,8 +511,10 @@ AC_DEFINE_UNQUOTED(INDEX_SIZE, $seektable, [size of the frame index seek table])
 dnl ############## Audio Sub-system Settings
 
 # Choose audio subsystem based on what we have.
-
+if test $cpu_type = "generic_float"; then
-if   test "x$HAVE_ALSA" = "xyes"; then
+	audio_system="dummy"
+	AC_MSG_WARN( Defaulting to dummy output with float samples, as the audio device system cannot handle floats. )
+elif   test "x$HAVE_ALSA" = "xyes"; then
 	audio_system="alsa"
 elif test "x$HAVE_OSS" = "xyes"; then
 	audio_system="oss"
@@ -683,7 +693,12 @@ echo ")
   Seek table size ......... $seektable
   FIFO support ............ $fifo
 "
-  
+if test $cpu_type = "generic_float"; then
+	echo "You chose the generic build with 32bit float output.
+That means output via -s, -O or -w to a file/pipe and no buffer.
+"
+fi
+
 if test "$debugging" = "enabled"; then
 	echo "CFLAGS='$CFLAGS'"
 	echo "LIBS='$LIBS'"

src/common.c

@@ -73,7 +73,7 @@ unsigned long track_frames = 0;
 #define TRACK_MAX_FRAMES ULONG_MAX/4/1152
 
 /* this could become a struct... */
-long lastscale = -1; /* last used scale */
+scale_t lastscale = -1; /* last used scale */
 int rva_level[2] = {-1,-1}; /* significance level of stored rva */
 float rva_gain[2] = {0,0}; /* mix, album */
 float rva_peak[2] = {0,0};
@@ -129,6 +129,9 @@ unsigned long samples_to_bytes(unsigned long s, struct frame *fr , struct audio_
 	samf = floor(sammy);
 	return (unsigned long)
 		(((ai->format & AUDIO_FORMAT_MASK) == AUDIO_FORMAT_16) ? 2 : 1)
+#ifdef FLOATOUT
+		* 2
+#endif
 		* ai->channels
 		* (int) (((sammy - samf) < 0.5) ? samf : ( sammy-samf > 0.5 ? samf+1 : ((unsigned long) samf % 2 == 0 ? samf : samf + 1)));
 }
@@ -247,9 +250,10 @@ void do_volume(double factor)
 /* adjust the volume, taking both outscale and rva values into account */
 void do_rva()
 {
-	float rvafact = 1;
+	double rvafact = 1;
 	float peak = 0;
-	long newscale;
+	scale_t newscale;
+
 	if(param.rva)
 	{
 		int rt = 0;
@@ -272,7 +276,7 @@ void do_rva()
 	/* if peak is unknown (== 0) this check won't hurt */
 	if((peak*newscale) > MAXOUTBURST)
 	{
-		newscale = (long) ((float) MAXOUTBURST/peak);
+		newscale = (scale_t) ((double) MAXOUTBURST/peak);
 		warning2("limiting scale value to %li to prevent clipping with indicated peak factor of %f", newscale, peak);
 	}
 	/* first rva setting is forced with lastscale < 0 */

src/common.h

@@ -102,7 +102,7 @@ void print_stat(struct frame *fr,unsigned long no,long buffsize,struct audio_inf
 void clear_stat();
 
 /* rva data, used in common.c, set in id3.c */
-extern long lastscale;
+extern scale_t lastscale;
 extern int rva_level[2];
 extern float rva_gain[2];
 extern float rva_peak[2];

src/decode.c

@@ -12,16 +12,14 @@
 
 #include "config.h"
 #include "mpg123.h"
+#include "decode.h"
 
-#define WRITE_SAMPLE(samples,sum,clip) \
+/* 8bit functions silenced for FLOATOUT */
-  if( (sum) > REAL_PLUS_32767) { *(samples) = 0x7fff; (clip)++; } \
-  else if( (sum) < REAL_MINUS_32768) { *(samples) = -0x8000; (clip)++; } \
-  else { *(samples) = REAL_TO_SHORT(sum); }
 
 int synth_1to1_8bit(real *bandPtr,int channel,unsigned char *samples,int *pnt)
 {
-  short samples_tmp[64];
+  sample_t samples_tmp[64];
-  short *tmp1 = samples_tmp + channel;
+  sample_t *tmp1 = samples_tmp + channel;
   int i,ret;
   int pnt1=0;
 
@@ -29,7 +27,11 @@ int synth_1to1_8bit(real *bandPtr,int channel,unsigned char *samples,int *pnt)
   samples += channel + *pnt;
 
   for(i=0;i<32;i++) {
+#ifdef FLOATOUT
+    *samples = 0;
+#else
     *samples = conv16to8[*tmp1>>AUSHIFT];
+#endif
     samples += 2;
     tmp1 += 2;
   }
@@ -40,8 +42,8 @@ int synth_1to1_8bit(real *bandPtr,int channel,unsigned char *samples,int *pnt)
 
 int synth_1to1_8bit_mono(real *bandPtr,unsigned char *samples,int *pnt)
 {
-  short samples_tmp[64];
+  sample_t samples_tmp[64];
-  short *tmp1 = samples_tmp;
+  sample_t *tmp1 = samples_tmp;
   int i,ret;
   int pnt1 = 0;
 
@@ -49,7 +51,11 @@ int synth_1to1_8bit_mono(real *bandPtr,unsigned char *samples,int *pnt)
   samples += *pnt;
 
   for(i=0;i<32;i++) {
+#ifdef FLOATOUT
+    *samples++ = 0;
+#else
     *samples++ = conv16to8[*tmp1>>AUSHIFT];
+#endif
     tmp1 += 2;
   }
   *pnt += 32;
@@ -59,8 +65,8 @@ int synth_1to1_8bit_mono(real *bandPtr,unsigned char *samples,int *pnt)
 
 int synth_1to1_8bit_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
 {
-  short samples_tmp[64];
+  sample_t samples_tmp[64];
-  short *tmp1 = samples_tmp;
+  sample_t *tmp1 = samples_tmp;
   int i,ret;
   int pnt1 = 0;
 
@@ -68,8 +74,13 @@ int synth_1to1_8bit_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
   samples += *pnt;
 
   for(i=0;i<32;i++) {
+#ifdef FLOATOUT
+    *samples++ = 0;
+    *samples++ = 0;
+#else
     *samples++ = conv16to8[*tmp1>>AUSHIFT];
     *samples++ = conv16to8[*tmp1>>AUSHIFT];
+#endif
     tmp1 += 2;
   }
   *pnt += 64;
@@ -79,8 +90,8 @@ int synth_1to1_8bit_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
 
 int synth_1to1_mono(real *bandPtr,unsigned char *samples,int *pnt)
 {
-  short samples_tmp[64];
+  sample_t samples_tmp[64];
-  short *tmp1 = samples_tmp;
+  sample_t *tmp1 = samples_tmp;
   int i,ret;
   int pnt1 = 0;
 
@@ -88,11 +99,11 @@ int synth_1to1_mono(real *bandPtr,unsigned char *samples,int *pnt)
   samples += *pnt;
 
   for(i=0;i<32;i++) {
-    *( (short *)samples) = *tmp1;
+    *( (sample_t *)samples) = *tmp1;
-    samples += 2;
+    samples += sizeof(sample_t);
     tmp1 += 2;
   }
-  *pnt += 64;
+  *pnt += 32*sizeof(sample_t);
 
   return ret;
 }
@@ -103,11 +114,11 @@ int synth_1to1_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
   int i,ret;
 
   ret = synth_1to1(bandPtr,0,samples,pnt);
-  samples = samples + *pnt - 128;
+  samples = samples + *pnt - 64*sizeof(sample_t);
 
   for(i=0;i<32;i++) {
-    ((short *)samples)[1] = ((short *)samples)[0];
+    ((sample_t *)samples)[1] = ((sample_t *)samples)[0];
-    samples+=4;
+    samples+=2*sizeof(sample_t);
   }
 
   return ret;
@@ -119,7 +130,7 @@ int synth_1to1(real *bandPtr,int channel,unsigned char *out,int *pnt)
   static real buffs[2][2][0x110];
   static const int step = 2;
   static int bo = 1;
-  short *samples = (short *) (out+*pnt);
+  sample_t *samples = (sample_t *) (out+*pnt);
 
   real *b0,(*buf)[0x110];
   int clip = 0; 
@@ -216,7 +227,7 @@ int synth_1to1(real *bandPtr,int channel,unsigned char *out,int *pnt)
     }
   }
 
-  *pnt += 128;
+  *pnt += 64*sizeof(sample_t);
 
   return clip;
 }

src/decode.h

@@ -0,0 +1,18 @@
+/*
+	decode.h: common definitions for decode functions
+
+	copyright 2007 by the mpg123 project - free software under the terms of the LGPL 2.1
+	see COPYING and AUTHORS files in distribution or http://mpg123.de
+	initially written by Thomas Orgis, taking WRITE_SAMPLE from decode.c
+*/
+
+#ifdef FLOATOUT
+#define WRITE_SAMPLE(samples,sum,clip) *(samples) = sum
+#define sample_t float
+#else
+#define WRITE_SAMPLE(samples,sum,clip) \
+  if( (sum) > REAL_PLUS_32767) { *(samples) = 0x7fff; (clip)++; } \
+  else if( (sum) < REAL_MINUS_32768) { *(samples) = -0x8000; (clip)++; } \
+  else { *(samples) = REAL_TO_SHORT(sum); }
+#define sample_t short
+#endif

src/decode_2to1.c

@@ -12,16 +12,12 @@
 
 #include "config.h"
 #include "mpg123.h"
-
+#include "decode.h"
-#define WRITE_SAMPLE(samples,sum,clip) \
-  if( (sum) > 32767.0) { *(samples) = 0x7fff; (clip)++; } \
-  else if( (sum) < -32768.0) { *(samples) = -0x8000; (clip)++; } \
-  else { *(samples) = sum; }
 
 int synth_2to1_8bit(real *bandPtr,int channel,unsigned char *samples,int *pnt)
 {
-  short samples_tmp[32];
+  sample_t samples_tmp[32];
-  short *tmp1 = samples_tmp + channel;
+  sample_t *tmp1 = samples_tmp + channel;
   int i,ret;
   int pnt1 = 0;
 
@@ -29,7 +25,11 @@ int synth_2to1_8bit(real *bandPtr,int channel,unsigned char *samples,int *pnt)
   samples += channel + *pnt;
 
   for(i=0;i<16;i++) {
+#ifdef FLOATOUT
+    *samples = 0;
+#else
     *samples = conv16to8[*tmp1>>AUSHIFT];
+#endif
     samples += 2;
     tmp1 += 2;
   }
@@ -40,8 +40,8 @@ int synth_2to1_8bit(real *bandPtr,int channel,unsigned char *samples,int *pnt)
 
 int synth_2to1_8bit_mono(real *bandPtr,unsigned char *samples,int *pnt)
 {
-  short samples_tmp[32];
+  sample_t samples_tmp[32];
-  short *tmp1 = samples_tmp;
+  sample_t *tmp1 = samples_tmp;
   int i,ret;
   int pnt1 = 0;
 
@@ -49,7 +49,11 @@ int synth_2to1_8bit_mono(real *bandPtr,unsigned char *samples,int *pnt)
   samples += *pnt;
 
   for(i=0;i<16;i++) {
+#ifdef FLOATOUT
+    *samples++ = 0;
+#else
     *samples++ = conv16to8[*tmp1>>AUSHIFT];
+#endif
     tmp1 += 2;
   }
   *pnt += 16;
@@ -60,8 +64,8 @@ int synth_2to1_8bit_mono(real *bandPtr,unsigned char *samples,int *pnt)
 
 int synth_2to1_8bit_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
 {
-  short samples_tmp[32];
+  sample_t samples_tmp[32];
-  short *tmp1 = samples_tmp;
+  sample_t *tmp1 = samples_tmp;
   int i,ret;
   int pnt1 = 0;
 
@@ -69,8 +73,13 @@ int synth_2to1_8bit_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
   samples += *pnt;
 
   for(i=0;i<16;i++) {
+#ifdef FLOATOUT
+    *samples++ = 0;
+    *samples++ = 0;
+#else
     *samples++ = conv16to8[*tmp1>>AUSHIFT];
     *samples++ = conv16to8[*tmp1>>AUSHIFT];
+#endif
     tmp1 += 2;
   }
   *pnt += 32;
@@ -80,8 +89,8 @@ int synth_2to1_8bit_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
 
 int synth_2to1_mono(real *bandPtr,unsigned char *samples,int *pnt)
 {
-  short samples_tmp[32];
+  sample_t samples_tmp[32];
-  short *tmp1 = samples_tmp;
+  sample_t *tmp1 = samples_tmp;
   int i,ret;
   int pnt1=0;
 
@@ -89,11 +98,11 @@ int synth_2to1_mono(real *bandPtr,unsigned char *samples,int *pnt)
   samples += *pnt;
 
   for(i=0;i<16;i++) {
-    *( (short *) samples) = *tmp1;
+    *( (sample_t *) samples) = *tmp1;
-    samples += 2;
+    samples += sizeof(sample_t);
     tmp1 += 2;
   }
-  *pnt += 32;
+  *pnt += 16*sizeof(sample_t);
 
   return ret;
 }
@@ -103,11 +112,11 @@ int synth_2to1_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
   int i,ret;
 
   ret = synth_2to1(bandPtr,0,samples,pnt);
-  samples = samples + *pnt - 64;
+  samples = samples + *pnt - 32*sizeof(sample_t);
 
   for(i=0;i<16;i++) {
-    ((short *)samples)[1] = ((short *)samples)[0];
+    ((sample_t *)samples)[1] = ((sample_t *)samples)[0];
-    samples+=4;
+    samples+=2*sizeof(sample_t);
   }
   
   return ret;
@@ -118,7 +127,7 @@ int synth_2to1(real *bandPtr,int channel,unsigned char *out,int *pnt)
   static real buffs[2][2][0x110];
   static const int step = 2;
   static int bo = 1;
-  short *samples = (short *) (out + *pnt);
+  sample_t *samples = (sample_t *) (out + *pnt);
 
   real *b0,(*buf)[0x110];
   int clip = 0; 
@@ -223,7 +232,7 @@ int synth_2to1(real *bandPtr,int channel,unsigned char *out,int *pnt)
     }
   }
 
-  *pnt += 64;
+  *pnt += 32*sizeof(sample_t);
 
   return clip;
 }

src/decode_4to1.c

@@ -15,16 +15,12 @@
 
 #include "config.h"
 #include "mpg123.h"
-
+#include "decode.h"
-#define WRITE_SAMPLE(samples,sum,clip) \
-  if( (sum) > 32767.0) { *(samples) = 0x7fff; (clip)++; } \
-  else if( (sum) < -32768.0) { *(samples) = -0x8000; (clip)++; } \
-  else { *(samples) = sum; }
 
 int synth_4to1_8bit(real *bandPtr,int channel,unsigned char *samples,int *pnt)
 {
-  short samples_tmp[16];
+  sample_t samples_tmp[16];
-  short *tmp1 = samples_tmp + channel;
+  sample_t *tmp1 = samples_tmp + channel;
   int i,ret;
   int pnt1 = 0;
 
@@ -32,7 +28,11 @@ int synth_4to1_8bit(real *bandPtr,int channel,unsigned char *samples,int *pnt)
   samples += channel + *pnt;
 
   for(i=0;i<8;i++) {
+#ifdef FLOATOUT
+    *samples = 0;
+#else
     *samples = conv16to8[*tmp1>>AUSHIFT];
+#endif
     samples += 2;
     tmp1 += 2;
   }
@@ -43,8 +43,8 @@ int synth_4to1_8bit(real *bandPtr,int channel,unsigned char *samples,int *pnt)
 
 int synth_4to1_8bit_mono(real *bandPtr,unsigned char *samples,int *pnt)
 {
-  short samples_tmp[16];
+  sample_t samples_tmp[16];
-  short *tmp1 = samples_tmp;
+  sample_t *tmp1 = samples_tmp;
   int i,ret;
   int pnt1 = 0;
 
@@ -52,7 +52,11 @@ int synth_4to1_8bit_mono(real *bandPtr,unsigned char *samples,int *pnt)
   samples += *pnt;
 
   for(i=0;i<8;i++) {
+#ifdef FLOATOUT
+    *samples++ = 0;
+#else
     *samples++ = conv16to8[*tmp1>>AUSHIFT];
+#endif
     tmp1 += 2;
   }
   *pnt += 8;
@@ -63,8 +67,8 @@ int synth_4to1_8bit_mono(real *bandPtr,unsigned char *samples,int *pnt)
 
 int synth_4to1_8bit_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
 {
-  short samples_tmp[16];
+  sample_t samples_tmp[16];
-  short *tmp1 = samples_tmp;
+  sample_t *tmp1 = samples_tmp;
   int i,ret;
   int pnt1 = 0;
 
@@ -72,8 +76,13 @@ int synth_4to1_8bit_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
   samples += *pnt;
 
   for(i=0;i<8;i++) {
+#ifdef FLOATOUT
+    *samples++ = 0;
+    *samples++ = 0;
+#else
     *samples++ = conv16to8[*tmp1>>AUSHIFT];
     *samples++ = conv16to8[*tmp1>>AUSHIFT];
+#endif
     tmp1 += 2;
   }
   *pnt += 16;
@@ -83,8 +92,8 @@ int synth_4to1_8bit_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
 
 int synth_4to1_mono(real *bandPtr,unsigned char *samples,int *pnt)
 {
-  short samples_tmp[16];
+  sample_t samples_tmp[16];
-  short *tmp1 = samples_tmp;
+  sample_t *tmp1 = samples_tmp;
   int i,ret;
   int pnt1 = 0;
 
@@ -92,11 +101,11 @@ int synth_4to1_mono(real *bandPtr,unsigned char *samples,int *pnt)
   samples += *pnt;
 
   for(i=0;i<8;i++) {
-    *( (short *)samples) = *tmp1;
+    *( (sample_t *)samples) = *tmp1;
-    samples += 2;
+    samples += sizeof(sample_t);
     tmp1 += 2;
   }
-  *pnt += 16;
+  *pnt += 8*sizeof(sample_t);
 
   return ret;
 }
@@ -106,11 +115,11 @@ int synth_4to1_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
   int i,ret;
 
   ret = synth_4to1(bandPtr,0,samples,pnt);
-  samples = samples + *pnt - 32;
+  samples = samples + *pnt - 16*sizeof(sample_t);
 
   for(i=0;i<8;i++) {
-    ((short *)samples)[1] = ((short *)samples)[0];
+    ((sample_t *)samples)[1] = ((sample_t *)samples)[0];
-    samples+=4;
+    samples+=2*sizeof(sample_t);
   }
 
   return ret;
@@ -121,7 +130,7 @@ int synth_4to1(real *bandPtr,int channel,unsigned char *out,int *pnt)
   static real buffs[2][2][0x110];
   static const int step = 2;
   static int bo = 1;
-  short *samples = (short *) (out + *pnt);
+  sample_t *samples = (sample_t *) (out + *pnt);
 
   real *b0,(*buf)[0x110];
   int clip = 0; 
@@ -232,7 +241,7 @@ int synth_4to1(real *bandPtr,int channel,unsigned char *out,int *pnt)
     }
   }
   
-  *pnt += 32;
+  *pnt += 16*sizeof(sample_t);
 
   return clip;
 }

src/decode_ntom.c

@@ -13,11 +13,7 @@
 #include "config.h"
 #include "mpg123.h"
 #include "debug.h"
-
+#include "decode.h"
-#define WRITE_SAMPLE(samples,sum,clip) \
-  if( (sum) > 32767.0) { *(samples) = 0x7fff; (clip)++; } \
-  else if( (sum) < -32768.0) { *(samples) = -0x8000; (clip)++; } \
-  else { *(samples) = sum; }
 
 #define NTOM_MUL (32768)
 static unsigned long ntom_val[2] = { NTOM_MUL>>1,NTOM_MUL>>1 };
@@ -48,8 +44,8 @@ int synth_ntom_set_step(long m,long n)
 
 int synth_ntom_8bit(real *bandPtr,int channel,unsigned char *samples,int *pnt)
 {
-  short samples_tmp[8*64];
+  sample_t samples_tmp[8*64];
-  short *tmp1 = samples_tmp + channel;
+  sample_t *tmp1 = samples_tmp + channel;
   int i,ret;
   int pnt1 = 0;
 
@@ -57,7 +53,11 @@ int synth_ntom_8bit(real *bandPtr,int channel,unsigned char *samples,int *pnt)
   samples += channel + *pnt;
 
   for(i=0;i<(pnt1>>2);i++) {
+#ifdef FLOATOUT
+    *samples = 0;
+#else
     *samples = conv16to8[*tmp1>>AUSHIFT];
+#endif
     samples += 2;
     tmp1 += 2;
   }
@@ -68,8 +68,8 @@ int synth_ntom_8bit(real *bandPtr,int channel,unsigned char *samples,int *pnt)
 
 int synth_ntom_8bit_mono(real *bandPtr,unsigned char *samples,int *pnt)
 {
-  short samples_tmp[8*64];
+  sample_t samples_tmp[8*64];
-  short *tmp1 = samples_tmp;
+  sample_t *tmp1 = samples_tmp;
   int i,ret;
   int pnt1 = 0;
 
@@ -77,7 +77,11 @@ int synth_ntom_8bit_mono(real *bandPtr,unsigned char *samples,int *pnt)
   samples += *pnt;
 
   for(i=0;i<(pnt1>>2);i++) {
+#ifdef FLOATOUT
+    *samples++ = 0;
+#else
     *samples++ = conv16to8[*tmp1>>AUSHIFT];
+#endif
     tmp1 += 2;
   }
   *pnt += pnt1 >> 2;
@@ -87,8 +91,8 @@ int synth_ntom_8bit_mono(real *bandPtr,unsigned char *samples,int *pnt)
 
 int synth_ntom_8bit_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
 {
-  short samples_tmp[8*64];
+  sample_t samples_tmp[8*64];
-  short *tmp1 = samples_tmp;
+  sample_t *tmp1 = samples_tmp;
   int i,ret;
   int pnt1 = 0;
 
@@ -96,8 +100,13 @@ int synth_ntom_8bit_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
   samples += *pnt;
 
   for(i=0;i<(pnt1>>2);i++) {
+#ifdef FLOATOUT
+    *samples++ = 0;
+    *samples++ = 0;
+#else
     *samples++ = conv16to8[*tmp1>>AUSHIFT];
     *samples++ = conv16to8[*tmp1>>AUSHIFT];
+#endif
     tmp1 += 2;
   }
   *pnt += pnt1 >> 1;
@@ -107,8 +116,8 @@ int synth_ntom_8bit_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
 
 int synth_ntom_mono(real *bandPtr,unsigned char *samples,int *pnt)
 {
-  short samples_tmp[8*64];
+  sample_t samples_tmp[8*64];
-  short *tmp1 = samples_tmp;
+  sample_t *tmp1 = samples_tmp;
   int i,ret;
   int pnt1 = 0;
 
@@ -116,11 +125,11 @@ int synth_ntom_mono(real *bandPtr,unsigned char *samples,int *pnt)
   samples += *pnt;
 
   for(i=0;i<(pnt1>>2);i++) {
-    *( (short *)samples) = *tmp1;
+    *( (sample_t *)samples) = *tmp1;
-    samples += 2;
+    samples += sizeof(sample_t);
     tmp1 += 2;
   }
-  *pnt += pnt1 >> 1;
+  *pnt += (pnt1>>2)*sizeof(sample_t);
 
   return ret;
 }
@@ -135,8 +144,8 @@ int synth_ntom_mono2stereo(real *bandPtr,unsigned char *samples,int *pnt)
   samples += pnt1;
   
   for(i=0;i<((*pnt-pnt1)>>2);i++) {
-    ((short *)samples)[1] = ((short *)samples)[0];
+    ((sample_t *)samples)[1] = ((sample_t *)samples)[0];
-    samples+=4;
+    samples+=2*sizeof(sample_t);
   }
 
   return ret;
@@ -148,7 +157,7 @@ int synth_ntom(real *bandPtr,int channel,unsigned char *out,int *pnt)
   static real buffs[2][2][0x110];
   static const int step = 2;
   static int bo = 1;
-  short *samples = (short *) (out + *pnt);
+  sample_t *samples = (sample_t *) (out + *pnt);
 
   real *b0,(*buf)[0x110];
   int clip = 0; 

src/getlopt.c

@@ -71,6 +71,7 @@ int performoption (int argc, char *argv[], topt *opt)
 				debug1("int at %p", opt->var);
 				*( (int *) opt->var ) = (int) opt->value;
 			}
+			/* GLO_DOUBLE is not supported here */
 			else prog_error();
 								
 			debug("casting assignment done");
@@ -90,6 +91,8 @@ int performoption (int argc, char *argv[], topt *opt)
 				*((long *) opt->var) = atol(loptarg);
 			else if(opt->flags & GLO_INT)
 				*((int *) opt->var) = atoi(loptarg);
+			else if(opt->flags & GLO_DOUBLE)
+				*((double *) opt->var) = atof(loptarg);
 			else prog_error();
 		}
 		else

src/getlopt.h

@@ -33,6 +33,7 @@ for .... no flag) */
 #define GLO_CHAR 2
 #define GLO_INT  4
 #define GLO_LONG 8
+#define GLO_DOUBLE 16
 
 /* flags:
  *	bit 0 = 0 - no argument

src/mpg123.c

@@ -107,7 +107,7 @@ char *prgName = NULL;
 char *equalfile = NULL;
 /* ThOr: pointers are not TRUE or FALSE */
 int have_eq_settings = FALSE;
-long outscale  = MAXOUTBURST;
+scale_t outscale  = MAXOUTBURST;
 long numframes = -1;
 long startFrame= 0;
 int buffer_fd[2];
@@ -364,7 +364,11 @@ topt opts[] = {
 	{0,   "speaker",     0,                  set_output_s, 0,0},
 	{0,   "lineout",     0,                  set_output_l, 0,0},
 	{'o', "output",      GLO_ARG | GLO_CHAR, set_output, 0,  0},
+#ifdef FLOATOUT
+	{'f', "scale",       GLO_ARG | GLO_DOUBLE, 0, &outscale,   0},
+#else
 	{'f', "scale",       GLO_ARG | GLO_LONG, 0, &outscale,   0},
+#endif
 	{'n', "frames",      GLO_ARG | GLO_LONG, 0, &numframes,  0},
 	#ifdef HAVE_TERMIOS
 	{'C', "control",     GLO_INT,  0, &param.term_ctrl, TRUE},
@@ -1120,7 +1124,7 @@ static void usage(int err)  /* print syntax & exit */
 	fprintf(o,"   -w <filename> write Output as WAV file\n");
 	fprintf(o,"   -k n  skip first n frames [0]        -n n  decode only n frames [all]\n");
 	fprintf(o,"   -c    check range violations         -y    DISABLE resync on errors\n");
-	fprintf(o,"   -b n  output buffer: n Kbytes [0]    -f n  change scalefactor [32768]\n");
+	fprintf(o,"   -b n  output buffer: n Kbytes [0]    -f n  change scalefactor [%g]\n", (double)outscale);
 	fprintf(o,"   -r n  set/force samplerate [auto]    -g n  set audio hardware output gain\n");
 	fprintf(o,"   -os,-ol,-oh  output to built-in speaker,line-out connector,headphones\n");
 	#ifdef NAS
@@ -1187,7 +1191,7 @@ static void long_usage(int err)
 	fprintf(o,"        --cdr <f>          write samples as CDR file in <f> (- is stdout)\n");
 	fprintf(o,"        --reopen           force close/open on audiodevice\n");
 	fprintf(o," -g     --gain             set audio hardware output gain\n");
-	fprintf(o," -f <n> --scale <n>        scale output samples (soft gain, default=%li)\n", outscale);
+	fprintf(o," -f <n> --scale <n>        scale output samples (soft gain, default=%g)\n", (double)outscale);
 	fprintf(o,"        --rva-mix,\n");
 	fprintf(o,"        --rva-radio        use RVA2/ReplayGain values for mix/radio mode\n");
 	fprintf(o,"        --rva-album,\n");

src/mpg123.h

@@ -138,8 +138,15 @@ typedef unsigned char byte;
 #define         MPG_MD_DUAL_CHANNEL     2
 #define         MPG_MD_MONO             3
 
+/* float output only for generic decoder! */
+#ifdef FLOATOUT
+#define MAXOUTBURST 1.0
+#define scale_t double
+#else
 /* I suspect that 32767 would be a better idea here, but Michael put this in... */
 #define MAXOUTBURST 32768
+#define scale_t long
+#endif
 
 /* Pre Shift fo 16 to 8 bit converter table */
 #define AUSHIFT (3)
@@ -417,7 +424,7 @@ extern struct parameter param;
 
 /* avoid the SIGINT in terminal control */
 void next_track(void);
-extern long outscale;
+extern scale_t outscale;
 
 #include "optimize.h"
 

src/optimize.h

@@ -34,7 +34,7 @@ typedef int (*func_synth_pent)(real *,int,unsigned char *);
 /* last headaches about getting mmx hardcode out */
 real init_layer3_gainpow2(int i);
 real* init_layer2_table(real *table, double m);
-void make_decode_tables(long scale);
+void make_decode_tables(scale_t scaleval);
 
 /* only 3dnow replaces that one, it's internal to layer3.c otherwise */
 void dct36(real *,real *,real *,real *,real *);

src/tabinit.c

@@ -60,7 +60,7 @@ static long intwinbase[] = {
  64019, 65290, 66494, 67629, 68692, 69679, 70590, 71420, 72169, 72835,
  73415, 73908, 74313, 74630, 74856, 74992, 75038 };
 
-void make_decode_tables(long scaleval)
+void make_decode_tables(scale_t scaleval)
 {
   int i,j,k,kr,divv;
   real *costab;

src/wav.c

@@ -20,36 +20,71 @@
 #include "mpg123.h"
 #include "debug.h"
 
+#ifdef FLOATOUT
+#define WAVE_FORMAT 3
+#else
+#define WAVE_FORMAT 1
+#endif
+
 struct
 {
-   byte riffheader[4];
+	byte riffheader[4];
-   byte WAVElen[4];
+	byte WAVElen[4]; /* should this include riffheader or not? */
-   struct
+	struct
-   {
+	{
-      byte fmtheader[8];
+		byte WAVEID[4];
-      byte fmtlen[4];
+		byte fmtheader[4];
-      struct
+		byte fmtlen[4];
-      {
+		struct
-         byte FormatTag[2];
+		{
-         byte Channels[2];
+			byte FormatTag[2];
-         byte SamplesPerSec[4];
+			byte Channels[2];
-         byte AvgBytesPerSec[4];
+			byte SamplesPerSec[4];
-         byte BlockAlign[2];
+			byte AvgBytesPerSec[4];
-         byte BitsPerSample[2]; /* format specific for PCM */
+			byte BlockAlign[2];
-      } fmt;
+			byte BitsPerSample[2]; /* format specific for PCM */
-      struct
+			#ifdef FLOATOUT
-      {
+			byte cbSize[2];
-         byte dataheader[4];
+			#endif
-         byte datalen[4];
+		} fmt;
-         /* from here you insert your PCM data */
+		#ifdef FLOATOUT
-      } data;
+		byte factheader[4];
-   } WAVE;
+		byte factlen[4];
+		struct
+		{
+			byte samplelen[4];
+		} fact;
+		#endif
+		struct
+		{
+			byte dataheader[4];
+			byte datalen[4];
+			/* from here you insert your PCM data */
+		} data;
+	} WAVE;
 } RIFF = 
-{  { 'R','I','F','F' } , { sizeof(RIFF.WAVE),0,0,0 } , 
+{
-   {  { 'W','A','V','E','f','m','t',' ' } , { sizeof(RIFF.WAVE.fmt),0,0,0} ,
+	{ 'R','I','F','F' } ,
-      { {1,0} , {0,0},{0,0,0,0},{0,0,0,0},{0,0},{0,0} } ,
+	{ sizeof(RIFF.WAVE),0,0,0 } , 
-      { { 'd','a','t','a' }  , {0,0,0,0} }
+	{
-   }
+		{ 'W','A','V','E' },
+		{ 'f','m','t',' ' },
+		{ sizeof(RIFF.WAVE.fmt),0,0,0 } ,
+		{
+			{WAVE_FORMAT,0} , {0,0},{0,0,0,0},{0,0,0,0},{0,0},{0,0}
+			#ifdef FLOATOUT
+			,{0,0}
+			#endif
+		} ,
+		#ifdef FLOATOUT
+		{ 'f','a','c','t' },
+		{ sizeof(RIFF.WAVE.fact),0,0,0 },
+		{
+			{0,0,0,0} /* to be filled later, like datalen and wavelen */
+		},
+		#endif
+		{ { 'd','a','t','a' }  , {0,0,0,0} }
+	}
 };
 
 struct auhead {
@@ -90,6 +125,15 @@ static void long2bigendian(long inval,byte *outval,int b)
   }
 }
 
+static long from_little(byte *inval, int b)
+{
+	long ret = 0;
+	int i;
+	for(i=0;i<b;++i) ret += ((long)inval[i])<<(i*8);
+
+	return ret;
+}
+
 static int testEndian(void) 
 {
   long i,a=0,b=0,c=0;
@@ -129,6 +173,10 @@ static int open_file(char *filename)
 
 int au_open(struct audio_info_struct *ai, char *aufilename)
 {
+#ifdef FLOATOUT
+	error("AU file support for float values not there yet");
+	return -1;
+#else
   flipendian = 0;
 
   switch(ai->format) {
@@ -161,10 +209,15 @@ int au_open(struct audio_info_struct *ai, char *aufilename)
   datalen = 0;
 
   return 0;
+#endif
 }
 
 int cdr_open(struct audio_info_struct *ai, char *cdrfilename)
 {
+#ifdef FLOATOUT
+	error("refusing to produce cdr file with float values");
+	return -1;
+#else
   param.force_stereo = 0;
   ai->format = AUDIO_FORMAT_SIGNED_16;
   ai->rate = 44100;
@@ -182,6 +235,7 @@ int cdr_open(struct audio_info_struct *ai, char *cdrfilename)
     return -1;
 
   return 0;
+#endif
 }
 
 int wav_open(struct audio_info_struct *ai, char *wavfilename)
@@ -191,8 +245,11 @@ int wav_open(struct audio_info_struct *ai, char *wavfilename)
    flipendian = 0;
 
    /* standard MS PCM, and its format specific is BitsPerSample */
-   long2littleendian(1,RIFF.WAVE.fmt.FormatTag,sizeof(RIFF.WAVE.fmt.FormatTag));
+   long2littleendian(WAVE_FORMAT,RIFF.WAVE.fmt.FormatTag,sizeof(RIFF.WAVE.fmt.FormatTag));
-
+#ifdef FLOATOUT
+   long2littleendian(bps=32,RIFF.WAVE.fmt.BitsPerSample,sizeof(RIFF.WAVE.fmt.BitsPerSample));
+   flipendian = testEndian();
+#else
    if(ai->format == AUDIO_FORMAT_SIGNED_16) {
       long2littleendian(bps=16,RIFF.WAVE.fmt.BitsPerSample,sizeof(RIFF.WAVE.fmt.BitsPerSample));
       flipendian = testEndian();
@@ -204,6 +261,7 @@ int wav_open(struct audio_info_struct *ai, char *wavfilename)
       error("Format not supported.");
       return -1;
    }
+#endif
 
    if(ai->rate < 0) ai->rate = 44100;
 
@@ -235,6 +293,20 @@ int wav_write(unsigned char *buf,int len)
      return 0;
   
    if(flipendian) {
+#ifdef FLOATOUT
+		if(len & 3)
+		{
+			error("Number of bytes no multiple of 4 (32bit)!");
+			return 0;
+		}
+		for(i=0;i<len;i+=4)
+		{
+			int j;
+			unsigned char tmp[4];
+			for(j = 0; j<=3; ++j) tmp[j] = buf[i+j];
+			for(j = 0; j<=3; ++j) buf[i+j] = tmp[3-j];
+		}
+#else
      if(len & 1) {
        error("Odd number of bytes!");
        return 0;
@@ -245,6 +317,7 @@ int wav_write(unsigned char *buf,int len)
        buf[i+0] = buf[i+1];
        buf[i+1] = tmp;
      }
+#endif
    }
 
    temp = fwrite(buf, 1, len, wavfp);
@@ -264,6 +337,10 @@ int wav_close(void)
    if(fseek(wavfp, 0L, SEEK_SET) >= 0) {
      long2littleendian(datalen,RIFF.WAVE.data.datalen,sizeof(RIFF.WAVE.data.datalen));
      long2littleendian(datalen+sizeof(RIFF.WAVE),RIFF.WAVElen,sizeof(RIFF.WAVElen));
+#ifdef FLOATOUT
+     long2littleendian(datalen/(from_little(RIFF.WAVE.fmt.Channels,2)*from_little(RIFF.WAVE.fmt.BitsPerSample,2)/8),
+                       RIFF.WAVE.fact.samplelen,sizeof(RIFF.WAVE.fact.samplelen));
+#endif
      fwrite(&RIFF, sizeof(RIFF),1,wavfp);
    }
    else {