MPlayer-specific changes to liba52

git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@14991 b3059339-0415-0410-9bf9-f77b7e298cf2

1 files changed, 3023 insertions, 0 deletions

diff --git a/liba52/liba52_changes.diff b/liba52/liba52_changes.diff
new file mode 100644
index 0000000000..ceb1de2576
--- /dev/null
+++ b/liba52/liba52_changes.diff
@@ -0,0 +1,3023 @@
+--- include/a52.h	2005-03-22 19:58:53.000000000 +0100
++++ a52.h	2004-03-19 01:15:49.000000000 +0100
+@@ -19,6 +25,9 @@
+  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+  */
+ 
++#ifndef A52_H
++#define A52_H
++
+ #ifndef LIBA52_DOUBLE
+ typedef float sample_t;
+ #else
+@@ -113,3 +122,10 @@
+ void a52_dynrng (a52_state_t * state,
+ 		 sample_t (* call) (sample_t, void *), void * data);
+ int a52_block (a52_state_t * state, sample_t * samples);
++
++void* a52_resample_init(uint32_t mm_accel,int flags,int chans);
++extern int (* a52_resample) (float * _f, int16_t * s16);
++
++uint16_t crc16_block(uint8_t *data,uint32_t num_bytes);
++
++#endif /* A52_H */
+--- liba52/a52_internal.h	2005-03-22 19:59:35.000000000 +0100
++++ a52_internal.h	2004-03-19 01:15:49.000000000 +0100
+@@ -41,11 +43,12 @@
+ 
+ int downmix_init (int input, int flags, sample_t * level,
+ 		  sample_t clev, sample_t slev);
++void downmix_accel_init(uint32_t mm_accel);
+ int downmix_coeff (sample_t * coeff, int acmod, int output, sample_t level,
+ 		   sample_t clev, sample_t slev);
+-void downmix (sample_t * samples, int acmod, int output, sample_t bias,
++extern void (*downmix) (sample_t * samples, int acmod, int output, sample_t bias,
+ 	      sample_t clev, sample_t slev);
+-void upmix (sample_t * samples, int acmod, int output);
++extern void (*upmix) (sample_t * samples, int acmod, int output);
+ 
+ void imdct_init (uint32_t mm_accel);
+ extern void (* imdct_256) (sample_t * data, sample_t * delay, sample_t bias);
+--- liba52/bitstream.c	2005-03-22 19:59:35.000000000 +0100
++++ bitstream.c	2004-03-19 01:15:49.000000000 +0100
+@@ -29,7 +35,12 @@
+ 
+ #define BUFFER_SIZE 4096
+ 
++#ifdef ALT_BITSTREAM_READER
++int indx=0;
++uint32_t * buffer_start;
++#else
+ static uint32_t * buffer_start;
++#endif
+ 
+ uint32_t bits_left;
+ uint32_t current_word;
+@@ -41,6 +52,9 @@
+     align = (int)buf & 3;
+     buffer_start = (uint32_t *) (buf - align);
+     bits_left = 0;
++#ifdef ALT_BITSTREAM_READER
++    indx=0;
++#endif
+     bitstream_get (align * 8);
+ }
+ 
+--- liba52/bitstream.h	2005-03-22 19:59:35.000000000 +0100
++++ bitstream.h	2004-03-19 01:15:49.000000000 +0100
+@@ -19,6 +25,48 @@
+  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+  */
+ 
++/* code from ffmpeg/libavcodec */
++#if defined(__GNUC__) && (__GNUC__ > 3 || __GNUC_ == 3 && __GNUC_MINOR__ > 0)
++#    define always_inline __attribute__((always_inline)) inline
++#else
++#    define always_inline inline
++#endif
++
++#if defined(__sparc__) || defined(hpux)
++/*
++ * the alt bitstream reader performs unaligned memory accesses; that doesn't work
++ * on sparc/hpux.  For now, disable ALT_BITSTREAM_READER.
++ */
++#undef	ALT_BITSTREAM_READER
++#else
++// alternative (faster) bitstram reader (reades upto 3 bytes over the end of the input)
++#define ALT_BITSTREAM_READER
++
++/* used to avoid missaligned exceptions on some archs (alpha, ...) */
++#if defined (ARCH_X86) || defined(ARCH_ARMV4L)
++#    define unaligned32(a) (*(uint32_t*)(a))
++#else
++#    ifdef __GNUC__
++static always_inline uint32_t unaligned32(const void *v) {
++    struct Unaligned {
++	uint32_t i;
++    } __attribute__((packed));
++
++    return ((const struct Unaligned *) v)->i;
++}
++#    elif defined(__DECC)
++static inline uint32_t unaligned32(const void *v) {
++    return *(const __unaligned uint32_t *) v;
++}
++#    else
++static inline uint32_t unaligned32(const void *v) {
++    return *(const uint32_t *) v;
++}
++#    endif
++#endif //!ARCH_X86
++
++#endif
++ 
+ /* (stolen from the kernel) */
+ #ifdef WORDS_BIGENDIAN
+ 
+@@ -29,7 +77,7 @@
+ #	if defined (__i386__)
+ 
+ #	define swab32(x) __i386_swab32(x)
+-	static inline const uint32_t __i386_swab32(uint32_t x)
++	static always_inline const uint32_t __i386_swab32(uint32_t x)
+ 	{
+ 		__asm__("bswap %0" : "=r" (x) : "0" (x));
+ 		return x;
+@@ -37,25 +85,42 @@
+ 
+ #	else
+ 
+-#	define swab32(x)\
+-((((uint8_t*)&x)[0] << 24) | (((uint8_t*)&x)[1] << 16) |  \
+- (((uint8_t*)&x)[2] << 8)  | (((uint8_t*)&x)[3]))
+-
++#	define swab32(x) __generic_swab32(x)
++	static always_inline const uint32_t __generic_swab32(uint32_t x)
++	{
++		return ((((uint8_t*)&x)[0] << 24) | (((uint8_t*)&x)[1] << 16) |
++		 (((uint8_t*)&x)[2] << 8)  | (((uint8_t*)&x)[3]));
++	}
+ #	endif
+ #endif
+ 
++#ifdef ALT_BITSTREAM_READER
++extern uint32_t *buffer_start; 
++extern int indx;
++#else
+ extern uint32_t bits_left;
+ extern uint32_t current_word;
++#endif
+ 
+ void bitstream_set_ptr (uint8_t * buf);
+ uint32_t bitstream_get_bh(uint32_t num_bits);
+ int32_t bitstream_get_bh_2(uint32_t num_bits);
+ 
++
+ static inline uint32_t 
+-bitstream_get(uint32_t num_bits)
++bitstream_get(uint32_t num_bits) // note num_bits is practically a constant due to inlineing
+ {
++#ifdef ALT_BITSTREAM_READER
++    uint32_t result= swab32( unaligned32(((uint8_t *)buffer_start)+(indx>>3)) );
++
++    result<<= (indx&0x07);
++    result>>= 32 - num_bits;
++    indx+= num_bits;
++    
++    return result;
++#else
+     uint32_t result;
+-	
++    
+     if(num_bits < bits_left) {
+ 	result = (current_word << (32 - bits_left)) >> (32 - num_bits);
+ 	bits_left -= num_bits;
+@@ -63,11 +128,30 @@
+     }
+ 
+     return bitstream_get_bh(num_bits);
++#endif
++}
++
++static inline void bitstream_skip(int num_bits)
++{
++#ifdef ALT_BITSTREAM_READER
++	indx+= num_bits;
++#else
++	bitstream_get(num_bits);
++#endif
+ }
+ 
+ static inline int32_t 
+ bitstream_get_2(uint32_t num_bits)
+ {
++#ifdef ALT_BITSTREAM_READER
++    int32_t result= swab32( unaligned32(((uint8_t *)buffer_start)+(indx>>3)) );
++
++    result<<= (indx&0x07);
++    result>>= 32 - num_bits;
++    indx+= num_bits;
++        
++    return result;
++#else
+     int32_t result;
+ 	
+     if(num_bits < bits_left) {
+@@ -77,4 +161,5 @@
+     }
+ 
+     return bitstream_get_bh_2(num_bits);
++#endif
+ }
+--- liba52/downmix.c	2005-03-22 19:59:35.000000000 +0100
++++ downmix.c	2004-04-12 18:42:14.000000000 +0200
+@@ -17,18 +23,46 @@
+  * You should have received a copy of the GNU General Public License
+  * along with this program; if not, write to the Free Software
+  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
++ *
++ * SSE optimizations from Michael Niedermayer (michaelni@gmx.at)
+  */
+ 
+ #include "config.h"
+ 
+-#include <inttypes.h>
+ #include <string.h>
++#include <inttypes.h>
+ 
+ #include "a52.h"
+ #include "a52_internal.h"
++#include "mm_accel.h"
+ 
+ #define CONVERT(acmod,output) (((output) << 3) + (acmod))
+ 
++
++void (*downmix)(sample_t * samples, int acmod, int output, sample_t bias,
++	      sample_t clev, sample_t slev)= NULL;
++void (*upmix)(sample_t * samples, int acmod, int output)= NULL;
++
++static void downmix_SSE (sample_t * samples, int acmod, int output, sample_t bias,
++	      sample_t clev, sample_t slev);
++static void downmix_3dnow (sample_t * samples, int acmod, int output, sample_t bias,
++	      sample_t clev, sample_t slev);
++static void downmix_C (sample_t * samples, int acmod, int output, sample_t bias,
++	      sample_t clev, sample_t slev);
++static void upmix_MMX (sample_t * samples, int acmod, int output);
++static void upmix_C (sample_t * samples, int acmod, int output);
++
++void downmix_accel_init(uint32_t mm_accel)
++{
++    upmix= upmix_C;
++    downmix= downmix_C;
++#ifdef ARCH_X86    
++    if(mm_accel & MM_ACCEL_X86_MMX) upmix= upmix_MMX;
++    if(mm_accel & MM_ACCEL_X86_SSE) downmix= downmix_SSE;
++    if(mm_accel & MM_ACCEL_X86_3DNOW) downmix= downmix_3dnow;
++#endif
++}
++   
+ int downmix_init (int input, int flags, sample_t * level,
+ 		  sample_t clev, sample_t slev)
+ {
+@@ -61,7 +95,7 @@
+     output = flags & A52_CHANNEL_MASK;
+     if (output > A52_DOLBY)
+ 	return -1;
+-
++    
+     output = table[output][input & 7];
+ 
+     if ((output == A52_STEREO) &&
+@@ -145,7 +179,6 @@
+ 	    *level *= 1 / (1 + 3 * LEVEL_3DB);
+ 	    break;
+ 	}
+-
+     return output;
+ }
+ 
+@@ -440,12 +473,11 @@
+ static void zero (sample_t * samples)
+ {
+     int i;
+-
+     for (i = 0; i < 256; i++)
+ 	samples[i] = 0;
+ }
+ 
+-void downmix (sample_t * samples, int acmod, int output, sample_t bias,
++static void downmix_C (sample_t * samples, int acmod, int output, sample_t bias,
+ 	      sample_t clev, sample_t slev)
+ {
+     switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {
+@@ -557,7 +589,7 @@
+ 	break;
+ 
+     case CONVERT (A52_3F2R, A52_2F1R):
+-	mix3to2 (samples, bias);
++	mix3to2 (samples, bias); //FIXME possible bug? (output doesnt seem to be used)
+ 	move2to1 (samples + 768, samples + 512, bias);
+ 	break;
+ 
+@@ -581,12 +613,12 @@
+ 	break;
+ 
+     case CONVERT (A52_3F1R, A52_3F2R):
+-	memcpy (samples + 1027, samples + 768, 256 * sizeof (sample_t));
++	memcpy (samples + 1024, samples + 768, 256 * sizeof (sample_t));
+ 	break;
+     }
+ }
+ 
+-void upmix (sample_t * samples, int acmod, int output)
++static void upmix_C (sample_t * samples, int acmod, int output)
+ {
+     switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {
+ 
+@@ -651,3 +683,1137 @@
+ 	goto mix_31to21;
+     }
+ }
++
++#ifdef ARCH_X86
++static void mix2to1_SSE (sample_t * dest, sample_t * src, sample_t bias)
++{
++	asm volatile(
++	"movlps %2, %%xmm7		\n\t"
++	"shufps $0x00, %%xmm7, %%xmm7	\n\t"
++	"movl $-1024, %%esi		\n\t"
++	".balign 16\n\t"
++	"1:				\n\t"
++	"movaps (%0, %%esi), %%xmm0	\n\t" 
++	"movaps 16(%0, %%esi), %%xmm1	\n\t" 
++	"addps (%1, %%esi), %%xmm0	\n\t" 
++	"addps 16(%1, %%esi), %%xmm1	\n\t" 
++	"addps %%xmm7, %%xmm0		\n\t"
++	"addps %%xmm7, %%xmm1		\n\t"
++	"movaps %%xmm0, (%1, %%esi)	\n\t"
++	"movaps %%xmm1, 16(%1, %%esi)	\n\t"
++	"addl $32, %%esi		\n\t"
++	" jnz 1b			\n\t"
++	:: "r" (src+256), "r" (dest+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix3to1_SSE (sample_t * samples, sample_t bias)
++{
++	asm volatile(
++	"movlps %1, %%xmm7		\n\t"
++	"shufps $0x00, %%xmm7, %%xmm7	\n\t"
++	"movl $-1024, %%esi		\n\t"
++	".balign 16\n\t"
++	"1:				\n\t"
++	"movaps (%0, %%esi), %%xmm0	\n\t" 
++	"movaps 1024(%0, %%esi), %%xmm1	\n\t" 
++	"addps 2048(%0, %%esi), %%xmm0	\n\t" 
++	"addps %%xmm7, %%xmm1		\n\t"
++	"addps %%xmm1, %%xmm0		\n\t"
++	"movaps %%xmm0, (%0, %%esi)	\n\t"
++	"addl $16, %%esi		\n\t"
++	" jnz 1b			\n\t"
++	:: "r" (samples+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix4to1_SSE (sample_t * samples, sample_t bias)
++{
++	asm volatile(
++	"movlps %1, %%xmm7		\n\t"
++	"shufps $0x00, %%xmm7, %%xmm7	\n\t"
++	"movl $-1024, %%esi		\n\t"
++	".balign 16\n\t"
++	"1:				\n\t"
++	"movaps (%0, %%esi), %%xmm0	\n\t" 
++	"movaps 1024(%0, %%esi), %%xmm1	\n\t" 
++	"addps 2048(%0, %%esi), %%xmm0	\n\t" 
++	"addps 3072(%0, %%esi), %%xmm1	\n\t" 
++	"addps %%xmm7, %%xmm0		\n\t"
++	"addps %%xmm1, %%xmm0		\n\t"
++	"movaps %%xmm0, (%0, %%esi)	\n\t"
++	"addl $16, %%esi		\n\t"
++	" jnz 1b			\n\t"
++	:: "r" (samples+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix5to1_SSE (sample_t * samples, sample_t bias)
++{
++	asm volatile(
++	"movlps %1, %%xmm7		\n\t"
++	"shufps $0x00, %%xmm7, %%xmm7	\n\t"
++	"movl $-1024, %%esi		\n\t"
++	".balign 16\n\t"
++	"1:				\n\t"
++	"movaps (%0, %%esi), %%xmm0	\n\t" 
++	"movaps 1024(%0, %%esi), %%xmm1	\n\t" 
++	"addps 2048(%0, %%esi), %%xmm0	\n\t" 
++	"addps 3072(%0, %%esi), %%xmm1	\n\t" 
++	"addps %%xmm7, %%xmm0		\n\t"
++	"addps 4096(%0, %%esi), %%xmm1	\n\t" 
++	"addps %%xmm1, %%xmm0		\n\t"
++	"movaps %%xmm0, (%0, %%esi)	\n\t"
++	"addl $16, %%esi		\n\t"
++	" jnz 1b			\n\t"
++	:: "r" (samples+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix3to2_SSE (sample_t * samples, sample_t bias)
++{
++	asm volatile(
++	"movlps %1, %%xmm7		\n\t"
++	"shufps $0x00, %%xmm7, %%xmm7	\n\t"
++	"movl $-1024, %%esi		\n\t"
++	".balign 16\n\t"
++	"1:				\n\t"
++	"movaps 1024(%0, %%esi), %%xmm0	\n\t" 
++	"addps %%xmm7, %%xmm0		\n\t" //common
++	"movaps (%0, %%esi), %%xmm1	\n\t" 
++	"movaps 2048(%0, %%esi), %%xmm2	\n\t"
++	"addps %%xmm0, %%xmm1		\n\t"
++	"addps %%xmm0, %%xmm2		\n\t"
++	"movaps %%xmm1, (%0, %%esi)	\n\t"
++	"movaps %%xmm2, 1024(%0, %%esi)	\n\t"
++	"addl $16, %%esi		\n\t"
++	" jnz 1b			\n\t"
++	:: "r" (samples+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix21to2_SSE (sample_t * left, sample_t * right, sample_t bias)
++{
++	asm volatile(
++		"movlps %2, %%xmm7		\n\t"
++		"shufps $0x00, %%xmm7, %%xmm7	\n\t"
++		"movl $-1024, %%esi		\n\t"
++		".balign 16\n\t"
++		"1:				\n\t"
++		"movaps 1024(%1, %%esi), %%xmm0	\n\t" 
++		"addps %%xmm7, %%xmm0		\n\t" //common
++		"movaps (%0, %%esi), %%xmm1	\n\t" 
++		"movaps (%1, %%esi), %%xmm2	\n\t"
++		"addps %%xmm0, %%xmm1		\n\t"
++		"addps %%xmm0, %%xmm2		\n\t"
++		"movaps %%xmm1, (%0, %%esi)	\n\t"
++		"movaps %%xmm2, (%1, %%esi)	\n\t"
++		"addl $16, %%esi		\n\t"
++		" jnz 1b			\n\t"
++	:: "r" (left+256), "r" (right+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix21toS_SSE (sample_t * samples, sample_t bias)
++{
++	asm volatile(
++		"movlps %1, %%xmm7		\n\t"
++		"shufps $0x00, %%xmm7, %%xmm7	\n\t"
++		"movl $-1024, %%esi		\n\t"
++		".balign 16\n\t"
++		"1:				\n\t"
++		"movaps 2048(%0, %%esi), %%xmm0	\n\t"  // surround
++		"movaps (%0, %%esi), %%xmm1	\n\t" 
++		"movaps 1024(%0, %%esi), %%xmm2	\n\t"
++		"addps %%xmm7, %%xmm1		\n\t"
++		"addps %%xmm7, %%xmm2		\n\t"
++		"subps %%xmm0, %%xmm1		\n\t"
++		"addps %%xmm0, %%xmm2		\n\t"
++		"movaps %%xmm1, (%0, %%esi)	\n\t"
++		"movaps %%xmm2, 1024(%0, %%esi)	\n\t"
++		"addl $16, %%esi		\n\t"
++		" jnz 1b			\n\t"
++	:: "r" (samples+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix31to2_SSE (sample_t * samples, sample_t bias)
++{
++	asm volatile(
++		"movlps %1, %%xmm7		\n\t"
++		"shufps $0x00, %%xmm7, %%xmm7	\n\t"
++		"movl $-1024, %%esi		\n\t"
++		".balign 16\n\t"
++		"1:				\n\t"
++		"movaps 1024(%0, %%esi), %%xmm0	\n\t"  
++		"addps 3072(%0, %%esi), %%xmm0	\n\t"  
++		"addps %%xmm7, %%xmm0		\n\t" // common
++		"movaps (%0, %%esi), %%xmm1	\n\t" 
++		"movaps 2048(%0, %%esi), %%xmm2	\n\t"
++		"addps %%xmm0, %%xmm1		\n\t"
++		"addps %%xmm0, %%xmm2		\n\t"
++		"movaps %%xmm1, (%0, %%esi)	\n\t"
++		"movaps %%xmm2, 1024(%0, %%esi)	\n\t"
++		"addl $16, %%esi		\n\t"
++		" jnz 1b			\n\t"
++	:: "r" (samples+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix31toS_SSE (sample_t * samples, sample_t bias)
++{
++	asm volatile(
++		"movlps %1, %%xmm7		\n\t"
++		"shufps $0x00, %%xmm7, %%xmm7	\n\t"
++		"movl $-1024, %%esi		\n\t"
++		".balign 16\n\t"
++		"1:				\n\t"
++		"movaps 1024(%0, %%esi), %%xmm0	\n\t"  
++		"movaps 3072(%0, %%esi), %%xmm3	\n\t" // surround
++		"addps %%xmm7, %%xmm0		\n\t" // common
++		"movaps (%0, %%esi), %%xmm1	\n\t" 
++		"movaps 2048(%0, %%esi), %%xmm2	\n\t"
++		"addps %%xmm0, %%xmm1		\n\t"
++		"addps %%xmm0, %%xmm2		\n\t"
++		"subps %%xmm3, %%xmm1		\n\t"
++		"addps %%xmm3, %%xmm2		\n\t"
++		"movaps %%xmm1, (%0, %%esi)	\n\t"
++		"movaps %%xmm2, 1024(%0, %%esi)	\n\t"
++		"addl $16, %%esi		\n\t"
++		" jnz 1b			\n\t"
++	:: "r" (samples+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix22toS_SSE (sample_t * samples, sample_t bias)
++{
++	asm volatile(
++		"movlps %1, %%xmm7		\n\t"
++		"shufps $0x00, %%xmm7, %%xmm7	\n\t"
++		"movl $-1024, %%esi		\n\t"
++		".balign 16\n\t"
++		"1:				\n\t"
++		"movaps 2048(%0, %%esi), %%xmm0	\n\t"  
++		"addps 3072(%0, %%esi), %%xmm0	\n\t" // surround
++		"movaps (%0, %%esi), %%xmm1	\n\t" 
++		"movaps 1024(%0, %%esi), %%xmm2	\n\t"
++		"addps %%xmm7, %%xmm1		\n\t"
++		"addps %%xmm7, %%xmm2		\n\t"
++		"subps %%xmm0, %%xmm1		\n\t"
++		"addps %%xmm0, %%xmm2		\n\t"
++		"movaps %%xmm1, (%0, %%esi)	\n\t"
++		"movaps %%xmm2, 1024(%0, %%esi)	\n\t"
++		"addl $16, %%esi		\n\t"
++		" jnz 1b			\n\t"
++	:: "r" (samples+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix32to2_SSE (sample_t * samples, sample_t bias)
++{
++	asm volatile(
++	"movlps %1, %%xmm7		\n\t"
++	"shufps $0x00, %%xmm7, %%xmm7	\n\t"
++	"movl $-1024, %%esi		\n\t"
++	".balign 16\n\t"
++	"1:				\n\t"
++	"movaps 1024(%0, %%esi), %%xmm0	\n\t" 
++	"addps %%xmm7, %%xmm0		\n\t" // common
++	"movaps %%xmm0, %%xmm1		\n\t" // common
++	"addps (%0, %%esi), %%xmm0	\n\t" 
++	"addps 2048(%0, %%esi), %%xmm1	\n\t" 
++	"addps 3072(%0, %%esi), %%xmm0	\n\t" 
++	"addps 4096(%0, %%esi), %%xmm1	\n\t" 
++	"movaps %%xmm0, (%0, %%esi)	\n\t"
++	"movaps %%xmm1, 1024(%0, %%esi)	\n\t"
++	"addl $16, %%esi		\n\t"
++	" jnz 1b			\n\t"
++	:: "r" (samples+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix32toS_SSE (sample_t * samples, sample_t bias)
++{
++	asm volatile(
++	"movlps %1, %%xmm7		\n\t"
++	"shufps $0x00, %%xmm7, %%xmm7	\n\t"
++	"movl $-1024, %%esi		\n\t"
++	".balign 16\n\t"
++	"1:				\n\t"
++	"movaps 1024(%0, %%esi), %%xmm0	\n\t" 
++	"movaps 3072(%0, %%esi), %%xmm2	\n\t" 
++	"addps %%xmm7, %%xmm0		\n\t" // common
++	"addps 4096(%0, %%esi), %%xmm2	\n\t" // surround	
++	"movaps (%0, %%esi), %%xmm1	\n\t" 
++	"movaps 2048(%0, %%esi), %%xmm3	\n\t" 
++	"subps %%xmm2, %%xmm1		\n\t"	
++	"addps %%xmm2, %%xmm3		\n\t"	
++	"addps %%xmm0, %%xmm1		\n\t"	
++	"addps %%xmm0, %%xmm3		\n\t"	
++	"movaps %%xmm1, (%0, %%esi)	\n\t"
++	"movaps %%xmm3, 1024(%0, %%esi)	\n\t"
++	"addl $16, %%esi		\n\t"
++	" jnz 1b			\n\t"
++	:: "r" (samples+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void move2to1_SSE (sample_t * src, sample_t * dest, sample_t bias)
++{
++	asm volatile(
++		"movlps %2, %%xmm7		\n\t"
++		"shufps $0x00, %%xmm7, %%xmm7	\n\t"
++		"movl $-1024, %%esi		\n\t"
++		".balign 16\n\t"
++		"1:				\n\t"
++		"movaps (%0, %%esi), %%xmm0	\n\t"  
++		"movaps 16(%0, %%esi), %%xmm1	\n\t"  
++		"addps 1024(%0, %%esi), %%xmm0	\n\t"
++		"addps 1040(%0, %%esi), %%xmm1	\n\t"
++		"addps %%xmm7, %%xmm0		\n\t"
++		"addps %%xmm7, %%xmm1		\n\t"
++		"movaps %%xmm0, (%1, %%esi)	\n\t"
++		"movaps %%xmm1, 16(%1, %%esi)	\n\t"
++		"addl $32, %%esi		\n\t"
++		" jnz 1b			\n\t"
++	:: "r" (src+256), "r" (dest+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void zero_MMX(sample_t * samples)
++{
++	asm volatile(
++		"movl $-1024, %%esi		\n\t"
++		"pxor %%mm0, %%mm0		\n\t"
++		".balign 16\n\t"
++		"1:				\n\t"
++		"movq %%mm0, (%0, %%esi)	\n\t"
++		"movq %%mm0, 8(%0, %%esi)	\n\t"
++		"movq %%mm0, 16(%0, %%esi)	\n\t"
++		"movq %%mm0, 24(%0, %%esi)	\n\t"
++		"addl $32, %%esi		\n\t"
++		" jnz 1b			\n\t"
++		"emms"
++	:: "r" (samples+256)
++	: "%esi"
++	);
++}
++
++/*
++ I hope dest and src will be at least 8 byte aligned and size
++ will devide on 8 without remain
++ Note: untested and unused.
++*/
++static void copy_MMX(void *dest,const void *src,unsigned size)
++{
++  unsigned i;
++  size /= 64;
++	for(i=0;i<size;i++)
++	{
++	    __asm __volatile(
++		"movq	%0,   %%mm0\n\t"
++		"movq	8%0,  %%mm1\n\t"
++		"movq	16%0, %%mm2\n\t"
++		"movq	24%0, %%mm3\n\t"
++		"movq	32%0, %%mm4\n\t"
++		"movq	40%0, %%mm5\n\t"
++		"movq	48%0, %%mm6\n\t"
++		"movq	56%0, %%mm7\n\t"
++		"movq	%%mm0, %1\n\t"
++		"movq	%%mm1, 8%1\n\t"
++		"movq	%%mm2, 16%1\n\t"
++		"movq	%%mm3, 24%1\n\t"
++		"movq	%%mm4, 32%1\n\t"
++		"movq	%%mm5, 40%1\n\t"
++		"movq	%%mm6, 48%1\n\t"
++		"movq	%%mm7, 56%1\n\t"
++		:
++		:"m"(src),"m"(dest));
++	}
++}
++
++static void downmix_SSE (sample_t * samples, int acmod, int output, sample_t bias,
++	      sample_t clev, sample_t slev)
++{
++    switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {
++
++    case CONVERT (A52_CHANNEL, A52_CHANNEL2):
++	memcpy (samples, samples + 256, 256 * sizeof (sample_t));
++	break;
++
++    case CONVERT (A52_CHANNEL, A52_MONO):
++    case CONVERT (A52_STEREO, A52_MONO):
++    mix_2to1_SSE:
++	mix2to1_SSE (samples, samples + 256, bias);
++	break;
++
++    case CONVERT (A52_2F1R, A52_MONO):
++	if (slev == 0)
++	    goto mix_2to1_SSE;
++    case CONVERT (A52_3F, A52_MONO):
++    mix_3to1_SSE:
++	mix3to1_SSE (samples, bias);
++	break;
++
++    case CONVERT (A52_3F1R, A52_MONO):
++	if (slev == 0)
++	    goto mix_3to1_SSE;
++    case CONVERT (A52_2F2R, A52_MONO):
++	if (slev == 0)
++	    goto mix_2to1_SSE;
++	mix4to1_SSE (samples, bias);
++	break;
++
++    case CONVERT (A52_3F2R, A52_MONO):
++	if (slev == 0)
++	    goto mix_3to1_SSE;
++	mix5to1_SSE (samples, bias);
++	break;
++
++    case CONVERT (A52_MONO, A52_DOLBY):
++	memcpy (samples + 256, samples, 256 * sizeof (sample_t));
++	break;
++
++    case CONVERT (A52_3F, A52_STEREO):
++    case CONVERT (A52_3F, A52_DOLBY):
++    mix_3to2_SSE:
++	mix3to2_SSE (samples, bias);
++	break;
++
++    case CONVERT (A52_2F1R, A52_STEREO):
++	if (slev == 0)
++	    break;
++	mix21to2_SSE (samples, samples + 256, bias);
++	break;
++
++    case CONVERT (A52_2F1R, A52_DOLBY):
++	mix21toS_SSE (samples, bias);
++	break;
++
++    case CONVERT (A52_3F1R, A52_STEREO):
++	if (slev == 0)
++	    goto mix_3to2_SSE;
++	mix31to2_SSE (samples, bias);
++	break;
++
++    case CONVERT (A52_3F1R, A52_DOLBY):
++	mix31toS_SSE (samples, bias);
++	break;
++
++    case CONVERT (A52_2F2R, A52_STEREO):
++	if (slev == 0)
++	    break;
++	mix2to1_SSE (samples, samples + 512, bias);
++	mix2to1_SSE (samples + 256, samples + 768, bias);
++	break;
++
++    case CONVERT (A52_2F2R, A52_DOLBY):
++	mix22toS_SSE (samples, bias);
++	break;
++
++    case CONVERT (A52_3F2R, A52_STEREO):
++	if (slev == 0)
++	    goto mix_3to2_SSE;
++	mix32to2_SSE (samples, bias);
++	break;
++
++    case CONVERT (A52_3F2R, A52_DOLBY):
++	mix32toS_SSE (samples, bias);
++	break;
++
++    case CONVERT (A52_3F1R, A52_3F):
++	if (slev == 0)
++	    break;
++	mix21to2_SSE (samples, samples + 512, bias);
++	break;
++
++    case CONVERT (A52_3F2R, A52_3F):
++	if (slev == 0)
++	    break;
++	mix2to1_SSE (samples, samples + 768, bias);
++	mix2to1_SSE (samples + 512, samples + 1024, bias);
++	break;
++
++    case CONVERT (A52_3F1R, A52_2F1R):
++	mix3to2_SSE (samples, bias);
++	memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
++	break;
++
++    case CONVERT (A52_2F2R, A52_2F1R):
++	mix2to1_SSE (samples + 512, samples + 768, bias);
++	break;
++
++    case CONVERT (A52_3F2R, A52_2F1R):
++	mix3to2_SSE (samples, bias); //FIXME possible bug? (output doesnt seem to be used)
++	move2to1_SSE (samples + 768, samples + 512, bias);
++	break;
++
++    case CONVERT (A52_3F2R, A52_3F1R):
++	mix2to1_SSE (samples + 768, samples + 1024, bias);
++	break;
++
++    case CONVERT (A52_2F1R, A52_2F2R):
++	memcpy (samples + 768, samples + 512, 256 * sizeof (sample_t));
++	break;
++
++    case CONVERT (A52_3F1R, A52_2F2R):
++	mix3to2_SSE (samples, bias);
++	memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
++	break;
++
++    case CONVERT (A52_3F2R, A52_2F2R):
++	mix3to2_SSE (samples, bias);
++	memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
++	memcpy (samples + 768, samples + 1024, 256 * sizeof (sample_t));
++	break;
++
++    case CONVERT (A52_3F1R, A52_3F2R):
++	memcpy (samples + 1024, samples + 768, 256 * sizeof (sample_t));
++	break;
++    }
++}
++
++static void upmix_MMX (sample_t * samples, int acmod, int output)
++{
++    switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {
++
++    case CONVERT (A52_CHANNEL, A52_CHANNEL2):
++	memcpy (samples + 256, samples, 256 * sizeof (sample_t));
++	break;
++
++    case CONVERT (A52_3F2R, A52_MONO):
++	zero_MMX (samples + 1024);
++    case CONVERT (A52_3F1R, A52_MONO):
++    case CONVERT (A52_2F2R, A52_MONO):
++	zero_MMX (samples + 768);
++    case CONVERT (A52_3F, A52_MONO):
++    case CONVERT (A52_2F1R, A52_MONO):
++	zero_MMX (samples + 512);
++    case CONVERT (A52_CHANNEL, A52_MONO):
++    case CONVERT (A52_STEREO, A52_MONO):
++	zero_MMX (samples + 256);
++	break;
++
++    case CONVERT (A52_3F2R, A52_STEREO):
++    case CONVERT (A52_3F2R, A52_DOLBY):
++	zero_MMX (samples + 1024);
++    case CONVERT (A52_3F1R, A52_STEREO):
++    case CONVERT (A52_3F1R, A52_DOLBY):
++	zero_MMX (samples + 768);
++    case CONVERT (A52_3F, A52_STEREO):
++    case CONVERT (A52_3F, A52_DOLBY):
++    mix_3to2_MMX:
++	memcpy (samples + 512, samples + 256, 256 * sizeof (sample_t));
++	zero_MMX (samples + 256);
++	break;
++
++    case CONVERT (A52_2F2R, A52_STEREO):
++    case CONVERT (A52_2F2R, A52_DOLBY):
++	zero_MMX (samples + 768);
++    case CONVERT (A52_2F1R, A52_STEREO):
++    case CONVERT (A52_2F1R, A52_DOLBY):
++	zero_MMX (samples + 512);
++	break;
++
++    case CONVERT (A52_3F2R, A52_3F):
++	zero_MMX (samples + 1024);
++    case CONVERT (A52_3F1R, A52_3F):
++    case CONVERT (A52_2F2R, A52_2F1R):
++	zero_MMX (samples + 768);
++	break;
++
++    case CONVERT (A52_3F2R, A52_3F1R):
++	zero_MMX (samples + 1024);
++	break;
++
++    case CONVERT (A52_3F2R, A52_2F1R):
++	zero_MMX (samples + 1024);
++    case CONVERT (A52_3F1R, A52_2F1R):
++    mix_31to21_MMX:
++	memcpy (samples + 768, samples + 512, 256 * sizeof (sample_t));
++	goto mix_3to2_MMX;
++
++    case CONVERT (A52_3F2R, A52_2F2R):
++	memcpy (samples + 1024, samples + 768, 256 * sizeof (sample_t));
++	goto mix_31to21_MMX;
++    }
++}
++
++static void mix2to1_3dnow (sample_t * dest, sample_t * src, sample_t bias)
++{
++	asm volatile(
++	"movd  %2, %%mm7	\n\t"
++	"punpckldq %2, %%mm7	\n\t"
++	"movl  $-1024, %%esi	\n\t"
++	".balign 16\n\t"
++	"1:			\n\t"
++	"movq  (%0, %%esi), %%mm0	\n\t" 
++	"movq  8(%0, %%esi), %%mm1	\n\t"
++	"movq  16(%0, %%esi), %%mm2	\n\t" 
++	"movq  24(%0, %%esi), %%mm3	\n\t"
++	"pfadd (%1, %%esi), %%mm0	\n\t" 
++	"pfadd 8(%1, %%esi), %%mm1	\n\t"
++	"pfadd 16(%1, %%esi), %%mm2	\n\t" 
++	"pfadd 24(%1, %%esi), %%mm3	\n\t"
++	"pfadd %%mm7, %%mm0		\n\t"
++	"pfadd %%mm7, %%mm1		\n\t"
++	"pfadd %%mm7, %%mm2		\n\t"
++	"pfadd %%mm7, %%mm3		\n\t"
++	"movq  %%mm0, (%1, %%esi)	\n\t"
++	"movq  %%mm1, 8(%1, %%esi)	\n\t"
++	"movq  %%mm2, 16(%1, %%esi)	\n\t"
++	"movq  %%mm3, 24(%1, %%esi)	\n\t"
++	"addl $32, %%esi		\n\t"
++	" jnz 1b			\n\t"
++	:: "r" (src+256), "r" (dest+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix3to1_3dnow (sample_t * samples, sample_t bias)
++{
++	asm volatile(
++	"movd  %1, %%mm7	\n\t"
++	"punpckldq %1, %%mm7	\n\t"
++	"movl $-1024, %%esi	\n\t"
++	".balign 16\n\t"
++	"1:			\n\t"
++	"movq  (%0, %%esi), %%mm0	\n\t" 
++	"movq  8(%0, %%esi), %%mm1	\n\t"
++	"movq  1024(%0, %%esi), %%mm2	\n\t" 
++	"movq  1032(%0, %%esi), %%mm3	\n\t"
++	"pfadd 2048(%0, %%esi), %%mm0	\n\t" 
++	"pfadd 2056(%0, %%esi), %%mm1	\n\t"
++	"pfadd %%mm7, %%mm0		\n\t"
++	"pfadd %%mm7, %%mm1		\n\t"
++	"pfadd %%mm2, %%mm0		\n\t"
++	"pfadd %%mm3, %%mm1		\n\t"
++	"movq  %%mm0, (%0, %%esi)	\n\t"
++	"movq  %%mm1, 8(%0, %%esi)	\n\t"
++	"addl $16, %%esi		\n\t"
++	" jnz 1b			\n\t"
++	:: "r" (samples+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix4to1_3dnow (sample_t * samples, sample_t bias)
++{
++	asm volatile(
++	"movd  %1, %%mm7	\n\t"
++	"punpckldq %1, %%mm7	\n\t"
++	"movl $-1024, %%esi	\n\t"
++	".balign 16\n\t"
++	"1:			\n\t"
++	"movq  (%0, %%esi), %%mm0	\n\t" 
++	"movq  8(%0, %%esi), %%mm1	\n\t"
++	"movq  1024(%0, %%esi), %%mm2	\n\t" 
++	"movq  1032(%0, %%esi), %%mm3	\n\t"
++	"pfadd 2048(%0, %%esi), %%mm0	\n\t" 
++	"pfadd 2056(%0, %%esi), %%mm1	\n\t"
++	"pfadd 3072(%0, %%esi), %%mm2	\n\t" 
++	"pfadd 3080(%0, %%esi), %%mm3	\n\t"
++	"pfadd %%mm7, %%mm0		\n\t"
++	"pfadd %%mm7, %%mm1		\n\t"
++	"pfadd %%mm2, %%mm0		\n\t"
++	"pfadd %%mm3, %%mm1		\n\t"
++	"movq  %%mm0, (%0, %%esi)	\n\t"
++	"movq  %%mm1, 8(%0, %%esi)	\n\t"
++	"addl $16, %%esi		\n\t"
++	" jnz 1b			\n\t"
++	:: "r" (samples+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix5to1_3dnow (sample_t * samples, sample_t bias)
++{
++	asm volatile(
++	"movd  %1, %%mm7	\n\t"
++	"punpckldq %1, %%mm7	\n\t"
++	"movl $-1024, %%esi	\n\t"
++	".balign 16\n\t"
++	"1:			\n\t"
++	"movq  (%0, %%esi), %%mm0	\n\t" 
++	"movq  8(%0, %%esi), %%mm1	\n\t"
++	"movq  1024(%0, %%esi), %%mm2	\n\t" 
++	"movq  1032(%0, %%esi), %%mm3	\n\t"
++	"pfadd 2048(%0, %%esi), %%mm0	\n\t" 
++	"pfadd 2056(%0, %%esi), %%mm1	\n\t"
++	"pfadd 3072(%0, %%esi), %%mm2	\n\t" 
++	"pfadd 3080(%0, %%esi), %%mm3	\n\t"
++	"pfadd %%mm7, %%mm0		\n\t"
++	"pfadd %%mm7, %%mm1		\n\t"
++	"pfadd 4096(%0, %%esi), %%mm2	\n\t" 
++	"pfadd 4104(%0, %%esi), %%mm3	\n\t"
++	"pfadd %%mm2, %%mm0		\n\t"
++	"pfadd %%mm3, %%mm1		\n\t"
++	"movq  %%mm0, (%0, %%esi)	\n\t"
++	"movq  %%mm1, 8(%0, %%esi)	\n\t"
++	"addl $16, %%esi		\n\t"
++	" jnz 1b			\n\t"
++	:: "r" (samples+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix3to2_3dnow (sample_t * samples, sample_t bias)
++{
++	asm volatile(
++	"movd  %1, %%mm7	\n\t"
++	"punpckldq %1, %%mm7	\n\t"
++	"movl $-1024, %%esi	\n\t"
++	".balign 16\n\t"
++	"1:			\n\t"
++	"movq   1024(%0, %%esi), %%mm0	\n\t" 
++	"movq   1032(%0, %%esi), %%mm1	\n\t"
++	"pfadd  %%mm7, %%mm0		\n\t" //common
++	"pfadd  %%mm7, %%mm1		\n\t" //common
++	"movq   (%0, %%esi), %%mm2	\n\t" 
++	"movq   8(%0, %%esi), %%mm3	\n\t"
++	"movq   2048(%0, %%esi), %%mm4	\n\t"
++	"movq   2056(%0, %%esi), %%mm5	\n\t"
++	"pfadd  %%mm0, %%mm2		\n\t"
++	"pfadd  %%mm1, %%mm3		\n\t"
++	"pfadd  %%mm0, %%mm4		\n\t"
++	"pfadd  %%mm1, %%mm5		\n\t"
++	"movq   %%mm2, (%0, %%esi)	\n\t"
++	"movq   %%mm3, 8(%0, %%esi)	\n\t"
++	"movq   %%mm4, 1024(%0, %%esi)	\n\t"
++	"movq   %%mm5, 1032(%0, %%esi)	\n\t"
++	"addl $16, %%esi		\n\t"
++	" jnz 1b			\n\t"
++	:: "r" (samples+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix21to2_3dnow (sample_t * left, sample_t * right, sample_t bias)
++{
++	asm volatile(
++		"movd  %2, %%mm7	\n\t"
++		"punpckldq %2, %%mm7	\n\t"
++		"movl $-1024, %%esi	\n\t"
++		".balign 16\n\t"
++		"1:			\n\t"
++		"movq  1024(%1, %%esi), %%mm0	\n\t" 
++		"movq  1032(%1, %%esi), %%mm1	\n\t"
++		"pfadd %%mm7, %%mm0		\n\t" //common
++		"pfadd %%mm7, %%mm1		\n\t" //common
++		"movq  (%0, %%esi), %%mm2	\n\t" 
++		"movq  8(%0, %%esi), %%mm3	\n\t"
++		"movq  (%1, %%esi), %%mm4	\n\t"
++		"movq  8(%1, %%esi), %%mm5	\n\t"
++		"pfadd %%mm0, %%mm2		\n\t"
++		"pfadd %%mm1, %%mm3		\n\t"
++		"pfadd %%mm0, %%mm4		\n\t"
++		"pfadd %%mm1, %%mm5		\n\t"
++		"movq  %%mm2, (%0, %%esi)	\n\t"
++		"movq  %%mm3, 8(%0, %%esi)	\n\t"
++		"movq  %%mm4, (%1, %%esi)	\n\t"
++		"movq  %%mm5, 8(%1, %%esi)	\n\t"
++		"addl $16, %%esi		\n\t"
++		" jnz 1b			\n\t"
++	:: "r" (left+256), "r" (right+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix21toS_3dnow (sample_t * samples, sample_t bias)
++{
++	asm volatile(
++		"movd  %1, %%mm7	\n\t"
++		"punpckldq %1, %%mm7	\n\t"
++		"movl $-1024, %%esi	\n\t"
++		".balign 16\n\t"
++		"1:			\n\t"
++		"movq  2048(%0, %%esi), %%mm0	\n\t"  // surround
++		"movq  2056(%0, %%esi), %%mm1	\n\t"  // surround
++		"movq  (%0, %%esi), %%mm2	\n\t" 
++		"movq  8(%0, %%esi), %%mm3	\n\t"
++		"movq  1024(%0, %%esi), %%mm4	\n\t"
++		"movq  1032(%0, %%esi), %%mm5	\n\t"
++		"pfadd %%mm7, %%mm2		\n\t"
++		"pfadd %%mm7, %%mm3		\n\t"
++		"pfadd %%mm7, %%mm4		\n\t"
++		"pfadd %%mm7, %%mm5		\n\t"
++		"pfsub %%mm0, %%mm2		\n\t"
++		"pfsub %%mm1, %%mm3		\n\t"
++		"pfadd %%mm0, %%mm4		\n\t"
++		"pfadd %%mm1, %%mm5		\n\t"
++		"movq  %%mm2, (%0, %%esi)	\n\t"
++		"movq  %%mm3, 8(%0, %%esi)	\n\t"
++		"movq  %%mm4, 1024(%0, %%esi)	\n\t"
++		"movq  %%mm5, 1032(%0, %%esi)	\n\t"
++		"addl $16, %%esi		\n\t"
++		" jnz 1b			\n\t"
++	:: "r" (samples+256), "m" (bias)
++	: "%esi"
++	);
++}
++
++static void mix31to2_3dnow (sample_t * samples, sample_t bias)
++{
++	asm volatile(
++		"movd  %1, %%mm7	\n\t"
++		"punpckldq %1, %%mm7	\n\t"
++		"movl $-1024, %%esi	\n\t"
++		".balign 16\n\t"
++		"1:			\n\t"
++		"movq  1024(%0, %%esi), %%mm0	\n\t"  
++		"movq  1032(%0, %%esi), %%mm1	\n\t"
++		"pfadd 3072(%0, %%esi), %%mm0	\n\t"  
++		"pfadd 3080(%0, %%esi), %%mm1	\n\t"
++		"pfadd %%mm7, %%mm0		\n\t" // common
++		"pfadd %%mm7, %%mm1		\n\t" // common
++		"movq  (%0, %%esi), %%mm2	\n\t" 
++		"movq  8(%0, %%esi), %%mm3	\n\t"
++		"movq  2048(%0, %%esi), %%mm4	\n\t"
++		"movq  2056(%0, %%esi), %%mm5	\n\t"
++		"pfadd %%mm0, %%mm2		\n\t"
++		"pfadd %%mm1, %%mm3		\n\t"
++		"pfadd %%mm0, %%mm4		\n\t"
++		"pfadd %%mm1, %%mm5		\n\t"
++		"movq  %%mm2, (%0, %%esi)	\n\t"
++		"movq  %%mm3, 8(%0, %%esi)	\n\t"
++		"movq  %%mm4, 1024(%0, %%esi)	\n\t"
++		"movq  %%mm5, 1032(%0, %%esi)	\n\t"