changeset 3909:ef32c8bdee81

c, mmx versions separated. a52 style runtime stuff
author arpi
date Sun, 30 Dec 2001 21:44:10 +0000
parents 0cc94b1eec0f
children db1d556fcf58
files liba52/a52.h liba52/resample.c liba52/resample_c.c liba52/resample_mmx.c
diffstat 4 files changed, 189 insertions(+), 1262 deletions(-) [+]
line wrap: on
line diff
--- a/liba52/a52.h	Sun Dec 30 21:38:53 2001 +0000
+++ b/liba52/a52.h	Sun Dec 30 21:44:10 2001 +0000
@@ -119,7 +119,7 @@
 		 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);
+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);
--- a/liba52/resample.c	Sun Dec 30 21:38:53 2001 +0000
+++ b/liba52/resample.c	Sun Dec 30 21:44:10 2001 +0000
@@ -1,20 +1,9 @@
-// this code come from a52dec/libao/audio_out_oss.c
-
-// FIXME FIXME FIXME
 
 // a52_resample_init should find the requested converter (from type flags ->
 // given number of channels) and set up some function pointers...
 
 // a52_resample() should do the conversion.
 
-// MMX optimizations from Michael Niedermayer (michaelni@gmx.at) (under GPL)
-
-/* optimization TODO / NOTES 
-    movntq is slightly faster (0.5% with the current test.c benchmark) 
-	(but thats just test.c so that needs to be testd in reallity)
-	and it would mean (C / MMX2 / MMX / 3DNOW) versions 
-*/
-
 #include <inttypes.h>
 #include <stdio.h>
 #include "a52.h"
@@ -23,584 +12,33 @@
 
 int (* a52_resample) (float * _f, int16_t * s16)=NULL;
 
+#include "resample_c.c"
+
 #ifdef ARCH_X86
-static uint64_t __attribute__((aligned(8))) magicF2W= 0x43c0000043c00000LL;
-static uint64_t __attribute__((aligned(8))) wm1010= 0xFFFF0000FFFF0000LL;
-static uint64_t __attribute__((aligned(8))) wm0101= 0x0000FFFF0000FFFFLL;
-static uint64_t __attribute__((aligned(8))) wm1100= 0xFFFFFFFF00000000LL;
+#include "resample_mmx.c"
 #endif
 
-static inline int16_t convert (int32_t i)
-{
-    if (i > 0x43c07fff)
-	return 32767;
-    else if (i < 0x43bf8000)
-	return -32768;
-    else
-	return i - 0x43c00000;
-}
-
-static int chans=2;
-static int flags=0;
-
-int a52_resample_C(float * _f, int16_t * s16)
-{
-    int i;
-    int32_t * f = (int32_t *) _f;
-
-    switch (flags) {
-    case A52_MONO:
-	for (i = 0; i < 256; i++) {
-	    s16[5*i] = s16[5*i+1] = s16[5*i+2] = s16[5*i+3] = 0;
-	    s16[5*i+4] = convert (f[i]);
-	}
-	break;
-    case A52_CHANNEL:
-    case A52_STEREO:
-    case A52_DOLBY:
-	for (i = 0; i < 256; i++) {
-	    s16[2*i] = convert (f[i]);
-	    s16[2*i+1] = convert (f[i+256]);
-	}
-	break;
-    case A52_3F:
-	for (i = 0; i < 256; i++) {
-	    s16[5*i] = convert (f[i]);
-	    s16[5*i+1] = convert (f[i+512]);
-	    s16[5*i+2] = s16[5*i+3] = 0;
-	    s16[5*i+4] = convert (f[i+256]);
-	}
-	break;
-    case A52_2F2R:
-	for (i = 0; i < 256; i++) {
-	    s16[4*i] = convert (f[i]);
-	    s16[4*i+1] = convert (f[i+256]);
-	    s16[4*i+2] = convert (f[i+512]);
-	    s16[4*i+3] = convert (f[i+768]);
-	}
-	break;
-    case A52_3F2R:
-	for (i = 0; i < 256; i++) {
-	    s16[5*i] = convert (f[i]);
-	    s16[5*i+1] = convert (f[i+512]);
-	    s16[5*i+2] = convert (f[i+768]);
-	    s16[5*i+3] = convert (f[i+1024]);
-	    s16[5*i+4] = convert (f[i+256]);
-	}
-	break;
-    case A52_MONO | A52_LFE:
-	for (i = 0; i < 256; i++) {
-	    s16[6*i] = s16[6*i+1] = s16[6*i+2] = s16[6*i+3] = 0;
-	    s16[6*i+4] = convert (f[i+256]);
-	    s16[6*i+5] = convert (f[i]);
-	}
-	break;
-    case A52_CHANNEL | A52_LFE:
-    case A52_STEREO | A52_LFE:
-    case A52_DOLBY | A52_LFE:
-	for (i = 0; i < 256; i++) {
-	    s16[6*i] = convert (f[i+256]);
-	    s16[6*i+1] = convert (f[i+512]);
-	    s16[6*i+2] = s16[6*i+3] = s16[6*i+4] = 0;
-	    s16[6*i+5] = convert (f[i]);
-	}
-	break;
-    case A52_3F | A52_LFE:
-	for (i = 0; i < 256; i++) {
-	    s16[6*i] = convert (f[i+256]);
-	    s16[6*i+1] = convert (f[i+768]);
-	    s16[6*i+2] = s16[6*i+3] = 0;
-	    s16[6*i+4] = convert (f[i+512]);
-	    s16[6*i+5] = convert (f[i]);
-	}
-	break;
-    case A52_2F2R | A52_LFE:
-	for (i = 0; i < 256; i++) {
-	    s16[6*i] = convert (f[i+256]);
-	    s16[6*i+1] = convert (f[i+512]);
-	    s16[6*i+2] = convert (f[i+768]);
-	    s16[6*i+3] = convert (f[i+1024]);
-	    s16[6*i+4] = 0;
-	    s16[6*i+5] = convert (f[i]);
-	}
-	break;
-    case A52_3F2R | A52_LFE:
-	for (i = 0; i < 256; i++) {
-	    s16[6*i] = convert (f[i+256]);
-	    s16[6*i+1] = convert (f[i+768]);
-	    s16[6*i+2] = convert (f[i+1024]);
-	    s16[6*i+3] = convert (f[i+1280]);
-	    s16[6*i+4] = convert (f[i+512]);
-	    s16[6*i+5] = convert (f[i]);
-	}
-	break;
-    }
-    return chans*256;
-}
+void* a52_resample_init(uint32_t mm_accel,int flags,int chans){
+void* tmp;
 
 #ifdef ARCH_X86
-int a52_resample_MMX(float * _f, int16_t * s16)
-{
-    int i;
-    int32_t * f = (int32_t *) _f;
+    if(mm_accel&MM_ACCEL_X86_MMX){
+	tmp=a52_resample_MMX(flags,chans);
+	if(tmp){
+	    if(a52_resample==NULL) fprintf(stderr, "Using MMX optimized resampler\n");
+	    a52_resample=tmp;
+	    return tmp;
+	}
+    }
+#endif
 
-    switch (flags) {
-    case A52_MONO:
-	asm volatile(
-		"movl $-512, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-		"movq wm1100, %%mm3		\n\t"
-		"movq wm0101, %%mm4		\n\t"
-		"movq wm1010, %%mm5		\n\t"
-		"pxor %%mm6, %%mm6		\n\t"
-		"1:				\n\t"
-		"movq (%1, %%esi, 2), %%mm0	\n\t"
-		"movq 8(%1, %%esi, 2), %%mm1	\n\t"
-		"leal (%%esi, %%esi, 4), %%edi	\n\t"
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"packssdw %%mm1, %%mm0		\n\t"
-		"movq %%mm0, %%mm1		\n\t"
-		"pand %%mm4, %%mm0		\n\t"
-		"pand %%mm5, %%mm1		\n\t"
-		"movq %%mm6, (%0, %%edi)	\n\t" // 0 0 0 0
-		"movd %%mm0, 8(%0, %%edi)	\n\t" // A 0
-		"pand %%mm3, %%mm0		\n\t"
-		"movd %%mm6, 12(%0, %%edi)	\n\t" // 0 0
-		"movd %%mm1, 16(%0, %%edi)	\n\t" // 0 B
-		"pand %%mm3, %%mm1		\n\t"
-		"movd %%mm6, 20(%0, %%edi)	\n\t" // 0 0
-		"movq %%mm0, 24(%0, %%edi)	\n\t" // 0 0 C 0
-		"movq %%mm1, 32(%0, %%edi)	\n\t" // 0 0 0 B
-		"addl $8, %%esi			\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1280), "r" (f+256)
-		:"%esi", "%edi", "memory"
-	);
-	break;
-    case A52_CHANNEL:
-    case A52_STEREO:
-    case A52_DOLBY:
-/* benchmark scores are 0.3% better with SSE but we would need to set bias=0 and premultiply it
-#ifdef HAVE_SSE
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"1:				\n\t"
-		"cvtps2pi (%1, %%esi), %%mm0	\n\t"
-		"cvtps2pi 1024(%1, %%esi), %%mm2\n\t"
-		"movq %%mm0, %%mm1		\n\t"
-		"punpcklwd %%mm2, %%mm0		\n\t"
-		"punpckhwd %%mm2, %%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"
-		"emms				\n\t"
-		:: "r" (s16+512), "r" (f+256)
-		:"%esi", "memory"
-	);*/
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-		"1:				\n\t"
-		"movq (%1, %%esi), %%mm0	\n\t"
-		"movq 8(%1, %%esi), %%mm1	\n\t"
-		"movq 1024(%1, %%esi), %%mm2	\n\t"
-		"movq 1032(%1, %%esi), %%mm3	\n\t"
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"packssdw %%mm1, %%mm0		\n\t"
-		"packssdw %%mm3, %%mm2		\n\t"
-		"movq %%mm0, %%mm1		\n\t"
-		"punpcklwd %%mm2, %%mm0		\n\t"
-		"punpckhwd %%mm2, %%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"
-		"emms				\n\t"
-		:: "r" (s16+512), "r" (f+256)
-		:"%esi", "memory"
-	);
-	break;
-    case A52_3F: 
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-		"pxor %%mm6, %%mm6		\n\t"
-		"movq %%mm7, %%mm5		\n\t"
-		"punpckldq %%mm6, %%mm5		\n\t"
-		"1:				\n\t"
-		"movd (%1, %%esi), %%mm0	\n\t"
-		"punpckldq 2048(%1, %%esi), %%mm0\n\t"
-		"movd 1024(%1, %%esi), %%mm1	\n\t"
-		"punpckldq 4(%1, %%esi), %%mm1	\n\t"
-		"movd 2052(%1, %%esi), %%mm2	\n\t"
-		"movq %%mm7, %%mm3		\n\t"
-		"punpckldq 1028(%1, %%esi), %%mm3\n\t"
-		"movd 8(%1, %%esi), %%mm4	\n\t"
-		"punpckldq 2056(%1, %%esi), %%mm4\n\t"
-		"leal (%%esi, %%esi, 4), %%edi	\n\t"
-		"sarl $1, %%edi			\n\t"
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm5, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"psubd %%mm7, %%mm4		\n\t"
-		"packssdw %%mm6, %%mm0		\n\t"
-		"packssdw %%mm2, %%mm1		\n\t"
-		"packssdw %%mm4, %%mm3		\n\t"
-		"movq %%mm0, (%0, %%edi)	\n\t"
-		"movq %%mm1, 8(%0, %%edi)	\n\t"
-		"movq %%mm3, 16(%0, %%edi)	\n\t"
-		
-		"movd 1032(%1, %%esi), %%mm1	\n\t"
-		"punpckldq 12(%1, %%esi), %%mm1\n\t"
-		"movd 2060(%1, %%esi), %%mm2	\n\t"
-		"movq %%mm7, %%mm3		\n\t"
-		"punpckldq 1036(%1, %%esi), %%mm3\n\t"
-		"pxor %%mm0, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm5, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"packssdw %%mm1, %%mm0		\n\t"
-		"packssdw %%mm3, %%mm2		\n\t"
-		"movq %%mm0, 24(%0, %%edi)	\n\t"
-		"movq %%mm2, 32(%0, %%edi)	\n\t"
-				
-		"addl $16, %%esi		\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1280), "r" (f+256)
-		:"%esi", "%edi", "memory"
-	);
-	break;
-    case A52_2F2R:
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-		"1:				\n\t"
-		"movq (%1, %%esi), %%mm0	\n\t"
-		"movq 8(%1, %%esi), %%mm1	\n\t"
-		"movq 1024(%1, %%esi), %%mm2	\n\t"
-		"movq 1032(%1, %%esi), %%mm3	\n\t"
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"packssdw %%mm1, %%mm0		\n\t"
-		"packssdw %%mm3, %%mm2		\n\t"
-		"movq 2048(%1, %%esi), %%mm3	\n\t"
-		"movq 2056(%1, %%esi), %%mm4	\n\t"
-		"movq 3072(%1, %%esi), %%mm5	\n\t"
-		"movq 3080(%1, %%esi), %%mm6	\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"psubd %%mm7, %%mm4		\n\t"
-		"psubd %%mm7, %%mm5		\n\t"
-		"psubd %%mm7, %%mm6		\n\t"
-		"packssdw %%mm4, %%mm3		\n\t"
-		"packssdw %%mm6, %%mm5		\n\t"
-		"movq %%mm0, %%mm1		\n\t"
-		"movq %%mm3, %%mm4		\n\t"
-		"punpcklwd %%mm2, %%mm0		\n\t"
-		"punpckhwd %%mm2, %%mm1		\n\t"
-		"punpcklwd %%mm5, %%mm3		\n\t"
-		"punpckhwd %%mm5, %%mm4		\n\t"
-		"movq %%mm0, %%mm2		\n\t"
-		"movq %%mm1, %%mm5		\n\t"
-		"punpckldq %%mm3, %%mm0		\n\t"
-		"punpckhdq %%mm3, %%mm2		\n\t"
-		"punpckldq %%mm4, %%mm1		\n\t"
-		"punpckhdq %%mm4, %%mm5		\n\t"
-		"movq %%mm0, (%0, %%esi,2)	\n\t"
-		"movq %%mm2, 8(%0, %%esi,2)	\n\t"
-		"movq %%mm1, 16(%0, %%esi,2)	\n\t"
-		"movq %%mm5, 24(%0, %%esi,2)	\n\t"
-		"addl $16, %%esi		\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1024), "r" (f+256)
-		:"%esi", "memory"
-	);
-	break;
-    case A52_3F2R: 
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-		"1:				\n\t"
-		"movd (%1, %%esi), %%mm0	\n\t"
-		"punpckldq 2048(%1, %%esi), %%mm0\n\t"
-		"movd 3072(%1, %%esi), %%mm1	\n\t"
-		"punpckldq 4096(%1, %%esi), %%mm1\n\t"
-		"movd 1024(%1, %%esi), %%mm2	\n\t"
-		"punpckldq 4(%1, %%esi), %%mm2	\n\t"
-		"movd 2052(%1, %%esi), %%mm3	\n\t"
-		"punpckldq 3076(%1, %%esi), %%mm3\n\t"
-		"movd 4100(%1, %%esi), %%mm4	\n\t"
-		"punpckldq 1028(%1, %%esi), %%mm4\n\t"
-		"movd 8(%1, %%esi), %%mm5	\n\t"
-		"punpckldq 2056(%1, %%esi), %%mm5\n\t"
-		"leal (%%esi, %%esi, 4), %%edi	\n\t"
-		"sarl $1, %%edi			\n\t"
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"psubd %%mm7, %%mm4		\n\t"
-		"psubd %%mm7, %%mm5		\n\t"
-		"packssdw %%mm1, %%mm0		\n\t"
-		"packssdw %%mm3, %%mm2		\n\t"
-		"packssdw %%mm5, %%mm4		\n\t"
-		"movq %%mm0, (%0, %%edi)	\n\t"
-		"movq %%mm2, 8(%0, %%edi)	\n\t"
-		"movq %%mm4, 16(%0, %%edi)	\n\t"
-		
-		"movd 3080(%1, %%esi), %%mm0	\n\t"
-		"punpckldq 4104(%1, %%esi), %%mm0\n\t"
-		"movd 1032(%1, %%esi), %%mm1	\n\t"
-		"punpckldq 12(%1, %%esi), %%mm1\n\t"
-		"movd 2060(%1, %%esi), %%mm2	\n\t"
-		"punpckldq 3084(%1, %%esi), %%mm2\n\t"
-		"movd 4108(%1, %%esi), %%mm3	\n\t"
-		"punpckldq 1036(%1, %%esi), %%mm3\n\t"
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"packssdw %%mm1, %%mm0		\n\t"
-		"packssdw %%mm3, %%mm2		\n\t"
-		"movq %%mm0, 24(%0, %%edi)	\n\t"
-		"movq %%mm2, 32(%0, %%edi)	\n\t"
-				
-		"addl $16, %%esi		\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1280), "r" (f+256)
-		:"%esi", "%edi", "memory"
-	);
-	break;
-    case A52_MONO | A52_LFE:
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-		"pxor %%mm6, %%mm6		\n\t"
-		"1:				\n\t"
-		"movq 1024(%1, %%esi), %%mm0	\n\t"
-		"movq 1032(%1, %%esi), %%mm1	\n\t"
-		"movq (%1, %%esi), %%mm2	\n\t"
-		"movq 8(%1, %%esi), %%mm3	\n\t"
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"packssdw %%mm1, %%mm0		\n\t"
-		"packssdw %%mm3, %%mm2		\n\t"
-		"movq %%mm0, %%mm1		\n\t"
-		"punpcklwd %%mm2, %%mm0		\n\t"
-		"punpckhwd %%mm2, %%mm1		\n\t"
-		"leal (%%esi, %%esi, 2), %%edi	\n\t"
-		"movq %%mm6, (%0, %%edi)	\n\t"
-		"movd %%mm0, 8(%0, %%edi)	\n\t"
-		"punpckhdq %%mm0, %%mm0		\n\t"
-		"movq %%mm6, 12(%0, %%edi)	\n\t"
-		"movd %%mm0, 20(%0, %%edi)	\n\t"
-		"movq %%mm6, 24(%0, %%edi)	\n\t"
-		"movd %%mm1, 32(%0, %%edi)	\n\t"
-		"punpckhdq %%mm1, %%mm1		\n\t"
-		"movq %%mm6, 36(%0, %%edi)	\n\t"
-		"movd %%mm1, 44(%0, %%edi)	\n\t"
-		"addl $16, %%esi		\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1536), "r" (f+256)
-		:"%esi", "%edi", "memory"
-	);
-	break;
-    case A52_CHANNEL | A52_LFE:
-    case A52_STEREO | A52_LFE:
-    case A52_DOLBY | A52_LFE:
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-		"pxor %%mm6, %%mm6		\n\t"
-		"1:				\n\t"
-		"movq 1024(%1, %%esi), %%mm0	\n\t"
-		"movq 2048(%1, %%esi), %%mm1	\n\t"
-		"movq (%1, %%esi), %%mm5	\n\t" 
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm5		\n\t"
-		"leal (%%esi, %%esi, 2), %%edi	\n\t"
-		
-		"pxor %%mm4, %%mm4		\n\t"
-		"packssdw %%mm5, %%mm0		\n\t" // FfAa
-		"packssdw %%mm4, %%mm1		\n\t" // 00Bb
-		"punpckhwd %%mm0, %%mm4		\n\t" // F0f0
-		"punpcklwd %%mm1, %%mm0		\n\t" // BAba
-		"movq %%mm0, %%mm1		\n\t" // BAba
-		"punpckldq %%mm4, %%mm3		\n\t" // f0XX
-		"punpckldq %%mm6, %%mm0		\n\t" // 00ba
-		"punpckhdq %%mm1, %%mm3		\n\t" // BAf0
-		
-		"movq %%mm0, (%0, %%edi)	\n\t" // 00ba
-		"punpckhdq %%mm4, %%mm0		\n\t" // F000
-		"movq %%mm3, 8(%0, %%edi)	\n\t" // BAf0
-		"movq %%mm0, 16(%0, %%edi)	\n\t" // F000
-		"addl $8, %%esi			\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1536), "r" (f+256)
-		:"%esi", "%edi", "memory"
-	);
-	break;
-    case A52_3F | A52_LFE:
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-		"pxor %%mm6, %%mm6		\n\t"
-		"1:				\n\t"
-		"movq 1024(%1, %%esi), %%mm0	\n\t"
-		"movq 3072(%1, %%esi), %%mm1	\n\t"
-		"movq 2048(%1, %%esi), %%mm4	\n\t"
-		"movq (%1, %%esi), %%mm5	\n\t" 
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm4		\n\t"
-		"psubd %%mm7, %%mm5		\n\t"
-		"leal (%%esi, %%esi, 2), %%edi	\n\t"
-		
-		"packssdw %%mm4, %%mm0		\n\t" // EeAa
-		"packssdw %%mm5, %%mm1		\n\t" // FfBb
-		"movq %%mm0, %%mm2		\n\t" // EeAa
-		"punpcklwd %%mm1, %%mm0		\n\t" // BAba
-		"punpckhwd %%mm1, %%mm2		\n\t" // FEfe
-		"movq %%mm0, %%mm1		\n\t" // BAba
-		"punpckldq %%mm6, %%mm0		\n\t" // 00ba
-		"punpckhdq %%mm1, %%mm1		\n\t" // BABA
-		
-		"movq %%mm0, (%0, %%edi)	\n\t"
-		"punpckhdq %%mm2, %%mm0		\n\t" // FE00
-		"punpckldq %%mm1, %%mm2		\n\t" // BAfe
-		"movq %%mm2, 8(%0, %%edi)	\n\t"
-		"movq %%mm0, 16(%0, %%edi)	\n\t"
-		"addl $8, %%esi			\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1536), "r" (f+256)
-		:"%esi", "%edi", "memory"
-	);
-	break;
-    case A52_2F2R | A52_LFE:
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-//		"pxor %%mm6, %%mm6		\n\t"
-		"1:				\n\t"
-		"movq 1024(%1, %%esi), %%mm0	\n\t"
-		"movq 2048(%1, %%esi), %%mm1	\n\t"
-		"movq 3072(%1, %%esi), %%mm2	\n\t"
-		"movq 4096(%1, %%esi), %%mm3	\n\t"
-		"movq (%1, %%esi), %%mm5	\n\t" 
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"psubd %%mm7, %%mm5		\n\t"
-		"leal (%%esi, %%esi, 2), %%edi	\n\t"
-		
-		"packssdw %%mm2, %%mm0		\n\t" // CcAa
-		"packssdw %%mm3, %%mm1		\n\t" // DdBb
-		"packssdw %%mm5, %%mm5		\n\t" // FfFf
-		"movq %%mm0, %%mm2		\n\t" // CcAa
-		"punpcklwd %%mm1, %%mm0		\n\t" // BAba
-		"punpckhwd %%mm1, %%mm2		\n\t" // DCdc
-		"pxor %%mm4, %%mm4		\n\t" // 0000
-		"punpcklwd %%mm5, %%mm4		\n\t" // F0f0
-		"movq %%mm0, %%mm1		\n\t" // BAba
-		"movq %%mm4, %%mm3		\n\t" // F0f0
-		"punpckldq %%mm2, %%mm0		\n\t" // dcba
-		"punpckhdq %%mm1, %%mm1		\n\t" // BABA
-		"punpckldq %%mm1, %%mm4		\n\t" // BAf0
-		"punpckhdq %%mm3, %%mm2		\n\t" // F0DC
-		
-		"movq %%mm0, (%0, %%edi)	\n\t"
-		"movq %%mm4, 8(%0, %%edi)	\n\t"
-		"movq %%mm2, 16(%0, %%edi)	\n\t"
-		"addl $8, %%esi			\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1536), "r" (f+256)
-		:"%esi", "%edi", "memory"
-	);
-	break;
-    case A52_3F2R | A52_LFE:
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-//		"pxor %%mm6, %%mm6		\n\t"
-		"1:				\n\t"
-		"movq 1024(%1, %%esi), %%mm0	\n\t"
-		"movq 3072(%1, %%esi), %%mm1	\n\t"
-		"movq 4096(%1, %%esi), %%mm2	\n\t"
-		"movq 5120(%1, %%esi), %%mm3	\n\t"
-		"movq 2048(%1, %%esi), %%mm4	\n\t"
-		"movq (%1, %%esi), %%mm5	\n\t" 
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"psubd %%mm7, %%mm4		\n\t"
-		"psubd %%mm7, %%mm5		\n\t"
-		"leal (%%esi, %%esi, 2), %%edi	\n\t"
-		
-		"packssdw %%mm2, %%mm0		\n\t" // CcAa
-		"packssdw %%mm3, %%mm1		\n\t" // DdBb
-		"packssdw %%mm4, %%mm4		\n\t" // EeEe
-		"packssdw %%mm5, %%mm5		\n\t" // FfFf
-		"movq %%mm0, %%mm2		\n\t" // CcAa
-		"punpcklwd %%mm1, %%mm0		\n\t" // BAba
-		"punpckhwd %%mm1, %%mm2		\n\t" // DCdc
-		"punpcklwd %%mm5, %%mm4		\n\t" // FEfe
-		"movq %%mm0, %%mm1		\n\t" // BAba
-		"movq %%mm4, %%mm3		\n\t" // FEfe
-		"punpckldq %%mm2, %%mm0		\n\t" // dcba
-		"punpckhdq %%mm1, %%mm1		\n\t" // BABA
-		"punpckldq %%mm1, %%mm4		\n\t" // BAfe
-		"punpckhdq %%mm3, %%mm2		\n\t" // FEDC
-		
-		"movq %%mm0, (%0, %%edi)	\n\t"
-		"movq %%mm4, 8(%0, %%edi)	\n\t"
-		"movq %%mm2, 16(%0, %%edi)	\n\t"
-		"addl $8, %%esi			\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1536), "r" (f+256)
-		:"%esi", "%edi", "memory"
-	);
-	break;
-    }
-    return chans*256;
-}
-#endif //arch_x86
-
-void a52_resample_init(uint32_t mm_accel,int _flags,int _chans){
-    chans=_chans;
-    flags=_flags;
-
-    if(a52_resample==NULL) // only once please ;)
-    {
-	    if(mm_accel & MM_ACCEL_X86_MMX) 	fprintf(stderr, "Using MMX optimized resampler\n");
-	    else				fprintf(stderr, "No accelerated resampler found\n");
+    tmp=a52_resample_C(flags,chans);
+    if(tmp){
+	if(a52_resample==NULL) fprintf(stderr, "No accelerated resampler found\n");
+	a52_resample=tmp;
+	return tmp;
     }
     
-#ifdef ARCH_X86
-    if(mm_accel & MM_ACCEL_X86_MMX) a52_resample= a52_resample_MMX;
-#else
-    if(0);
-#endif
-    else		a52_resample= a52_resample_C;
+    fprintf(stderr, "Unimplemented resampler for mode 0x%X -> %d channels conversion - Contact MPlayer developers!\n", flags, chans);
+    return NULL;
 }
-
--- a/liba52/resample_c.c	Sun Dec 30 21:38:53 2001 +0000
+++ b/liba52/resample_c.c	Sun Dec 30 21:44:10 2001 +0000
@@ -1,34 +1,4 @@
-// this code come from a52dec/libao/audio_out_oss.c
-
-// FIXME FIXME FIXME
-
-// a52_resample_init should find the requested converter (from type flags ->
-// given number of channels) and set up some function pointers...
-
-// a52_resample() should do the conversion.
-
-// MMX optimizations from Michael Niedermayer (michaelni@gmx.at) (under GPL)
-
-/* optimization TODO / NOTES 
-    movntq is slightly faster (0.5% with the current test.c benchmark) 
-	(but thats just test.c so that needs to be testd in reallity)
-	and it would mean (C / MMX2 / MMX / 3DNOW) versions 
-*/
-
-#include <inttypes.h>
-#include <stdio.h>
-#include "a52.h"
-#include "mm_accel.h"
-#include "../config.h"
-
-int (* a52_resample) (float * _f, int16_t * s16)=NULL;
-
-#ifdef ARCH_X86
-static uint64_t __attribute__((aligned(8))) magicF2W= 0x43c0000043c00000LL;
-static uint64_t __attribute__((aligned(8))) wm1010= 0xFFFF0000FFFF0000LL;
-static uint64_t __attribute__((aligned(8))) wm0101= 0x0000FFFF0000FFFFLL;
-static uint64_t __attribute__((aligned(8))) wm1100= 0xFFFFFFFF00000000LL;
-#endif
+// this code is based on a52dec/libao/audio_out_oss.c
 
 static inline int16_t convert (int32_t i)
 {
@@ -40,46 +10,53 @@
 	return i - 0x43c00000;
 }
 
-static int chans=2;
-static int flags=0;
-
-int a52_resample_C(float * _f, int16_t * s16)
-{
+static int a52_resample_MONO_to_5_C(float * _f, int16_t * s16){
     int i;
     int32_t * f = (int32_t *) _f;
-
-    switch (flags) {
-    case A52_MONO:
 	for (i = 0; i < 256; i++) {
 	    s16[5*i] = s16[5*i+1] = s16[5*i+2] = s16[5*i+3] = 0;
 	    s16[5*i+4] = convert (f[i]);
 	}
-	break;
-    case A52_CHANNEL:
-    case A52_STEREO:
-    case A52_DOLBY:
+    return 5*256;
+}
+
+static int a52_resample_STEREO_to_2_C(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 	for (i = 0; i < 256; i++) {
 	    s16[2*i] = convert (f[i]);
 	    s16[2*i+1] = convert (f[i+256]);
 	}
-	break;
-    case A52_3F:
+    return 2*256;
+}
+
+static int a52_resample_3F_to_5_C(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 	for (i = 0; i < 256; i++) {
 	    s16[5*i] = convert (f[i]);
 	    s16[5*i+1] = convert (f[i+512]);
 	    s16[5*i+2] = s16[5*i+3] = 0;
 	    s16[5*i+4] = convert (f[i+256]);
 	}
-	break;
-    case A52_2F2R:
+    return 5*256;
+}
+
+static int a52_resample_2F_2R_to_4_C(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 	for (i = 0; i < 256; i++) {
 	    s16[4*i] = convert (f[i]);
 	    s16[4*i+1] = convert (f[i+256]);
 	    s16[4*i+2] = convert (f[i+512]);
 	    s16[4*i+3] = convert (f[i+768]);
 	}
-	break;
-    case A52_3F2R:
+    return 4*256;
+}
+
+static int a52_resample_3F_2R_to_5_C(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 	for (i = 0; i < 256; i++) {
 	    s16[5*i] = convert (f[i]);
 	    s16[5*i+1] = convert (f[i+512]);
@@ -87,25 +64,35 @@
 	    s16[5*i+3] = convert (f[i+1024]);
 	    s16[5*i+4] = convert (f[i+256]);
 	}
-	break;
-    case A52_MONO | A52_LFE:
+    return 5*256;
+}
+
+static int a52_resample_MONO_LFE_to_6_C(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 	for (i = 0; i < 256; i++) {
 	    s16[6*i] = s16[6*i+1] = s16[6*i+2] = s16[6*i+3] = 0;
 	    s16[6*i+4] = convert (f[i+256]);
 	    s16[6*i+5] = convert (f[i]);
 	}
-	break;
-    case A52_CHANNEL | A52_LFE:
-    case A52_STEREO | A52_LFE:
-    case A52_DOLBY | A52_LFE:
+    return 6*256;
+}
+
+static int a52_resample_STEREO_LFE_to_6_C(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 	for (i = 0; i < 256; i++) {
 	    s16[6*i] = convert (f[i+256]);
 	    s16[6*i+1] = convert (f[i+512]);
 	    s16[6*i+2] = s16[6*i+3] = s16[6*i+4] = 0;
 	    s16[6*i+5] = convert (f[i]);
 	}
-	break;
-    case A52_3F | A52_LFE:
+    return 6*256;
+}
+
+static int a52_resample_3F_LFE_to_6_C(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 	for (i = 0; i < 256; i++) {
 	    s16[6*i] = convert (f[i+256]);
 	    s16[6*i+1] = convert (f[i+768]);
@@ -113,8 +100,12 @@
 	    s16[6*i+4] = convert (f[i+512]);
 	    s16[6*i+5] = convert (f[i]);
 	}
-	break;
-    case A52_2F2R | A52_LFE:
+    return 6*256;
+}
+
+static int a52_resample_2F_2R_LFE_to_6_C(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 	for (i = 0; i < 256; i++) {
 	    s16[6*i] = convert (f[i+256]);
 	    s16[6*i+1] = convert (f[i+512]);
@@ -123,8 +114,12 @@
 	    s16[6*i+4] = 0;
 	    s16[6*i+5] = convert (f[i]);
 	}
-	break;
-    case A52_3F2R | A52_LFE:
+    return 6*256;
+}
+
+static int a52_resample_3F_2R_LFE_to_6_C(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 	for (i = 0; i < 256; i++) {
 	    s16[6*i] = convert (f[i+256]);
 	    s16[6*i+1] = convert (f[i+768]);
@@ -133,474 +128,46 @@
 	    s16[6*i+4] = convert (f[i+512]);
 	    s16[6*i+5] = convert (f[i]);
 	}
-	break;
-    }
-    return chans*256;
+    return 6*256;
 }
 
-#ifdef ARCH_X86
-int a52_resample_MMX(float * _f, int16_t * s16)
-{
-    int i;
-    int32_t * f = (int32_t *) _f;
 
+static void* a52_resample_C(int flags, int ch){
     switch (flags) {
     case A52_MONO:
-	asm volatile(
-		"movl $-512, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-		"movq wm1100, %%mm3		\n\t"
-		"movq wm0101, %%mm4		\n\t"
-		"movq wm1010, %%mm5		\n\t"
-		"pxor %%mm6, %%mm6		\n\t"
-		"1:				\n\t"
-		"movq (%1, %%esi, 2), %%mm0	\n\t"
-		"movq 8(%1, %%esi, 2), %%mm1	\n\t"
-		"leal (%%esi, %%esi, 4), %%edi	\n\t"
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"packssdw %%mm1, %%mm0		\n\t"
-		"movq %%mm0, %%mm1		\n\t"
-		"pand %%mm4, %%mm0		\n\t"
-		"pand %%mm5, %%mm1		\n\t"
-		"movq %%mm6, (%0, %%edi)	\n\t" // 0 0 0 0
-		"movd %%mm0, 8(%0, %%edi)	\n\t" // A 0
-		"pand %%mm3, %%mm0		\n\t"
-		"movd %%mm6, 12(%0, %%edi)	\n\t" // 0 0
-		"movd %%mm1, 16(%0, %%edi)	\n\t" // 0 B
-		"pand %%mm3, %%mm1		\n\t"
-		"movd %%mm6, 20(%0, %%edi)	\n\t" // 0 0
-		"movq %%mm0, 24(%0, %%edi)	\n\t" // 0 0 C 0
-		"movq %%mm1, 32(%0, %%edi)	\n\t" // 0 0 0 B
-		"addl $8, %%esi			\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1280), "r" (f+256)
-		:"%esi", "%edi", "memory"
-	);
+	if(ch==5) return a52_resample_MONO_to_5_C;
 	break;
     case A52_CHANNEL:
     case A52_STEREO:
     case A52_DOLBY:
-/* benchmark scores are 0.3% better with SSE but we would need to set bias=0 and premultiply it
-#ifdef HAVE_SSE
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"1:				\n\t"
-		"cvtps2pi (%1, %%esi), %%mm0	\n\t"
-		"cvtps2pi 1024(%1, %%esi), %%mm2\n\t"
-		"movq %%mm0, %%mm1		\n\t"
-		"punpcklwd %%mm2, %%mm0		\n\t"
-		"punpckhwd %%mm2, %%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"
-		"emms				\n\t"
-		:: "r" (s16+512), "r" (f+256)
-		:"%esi", "memory"
-	);*/
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-		"1:				\n\t"
-		"movq (%1, %%esi), %%mm0	\n\t"
-		"movq 8(%1, %%esi), %%mm1	\n\t"
-		"movq 1024(%1, %%esi), %%mm2	\n\t"
-		"movq 1032(%1, %%esi), %%mm3	\n\t"
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"packssdw %%mm1, %%mm0		\n\t"
-		"packssdw %%mm3, %%mm2		\n\t"
-		"movq %%mm0, %%mm1		\n\t"
-		"punpcklwd %%mm2, %%mm0		\n\t"
-		"punpckhwd %%mm2, %%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"
-		"emms				\n\t"
-		:: "r" (s16+512), "r" (f+256)
-		:"%esi", "memory"
-	);
+	if(ch==2) return a52_resample_STEREO_to_2_C;
 	break;
-    case A52_3F: 
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-		"pxor %%mm6, %%mm6		\n\t"
-		"movq %%mm7, %%mm5		\n\t"
-		"punpckldq %%mm6, %%mm5		\n\t"
-		"1:				\n\t"
-		"movd (%1, %%esi), %%mm0	\n\t"
-		"punpckldq 2048(%1, %%esi), %%mm0\n\t"
-		"movd 1024(%1, %%esi), %%mm1	\n\t"
-		"punpckldq 4(%1, %%esi), %%mm1	\n\t"
-		"movd 2052(%1, %%esi), %%mm2	\n\t"
-		"movq %%mm7, %%mm3		\n\t"
-		"punpckldq 1028(%1, %%esi), %%mm3\n\t"
-		"movd 8(%1, %%esi), %%mm4	\n\t"
-		"punpckldq 2056(%1, %%esi), %%mm4\n\t"
-		"leal (%%esi, %%esi, 4), %%edi	\n\t"
-		"sarl $1, %%edi			\n\t"
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm5, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"psubd %%mm7, %%mm4		\n\t"
-		"packssdw %%mm6, %%mm0		\n\t"
-		"packssdw %%mm2, %%mm1		\n\t"
-		"packssdw %%mm4, %%mm3		\n\t"
-		"movq %%mm0, (%0, %%edi)	\n\t"
-		"movq %%mm1, 8(%0, %%edi)	\n\t"
-		"movq %%mm3, 16(%0, %%edi)	\n\t"
-		
-		"movd 1032(%1, %%esi), %%mm1	\n\t"
-		"punpckldq 12(%1, %%esi), %%mm1\n\t"
-		"movd 2060(%1, %%esi), %%mm2	\n\t"
-		"movq %%mm7, %%mm3		\n\t"
-		"punpckldq 1036(%1, %%esi), %%mm3\n\t"
-		"pxor %%mm0, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm5, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"packssdw %%mm1, %%mm0		\n\t"
-		"packssdw %%mm3, %%mm2		\n\t"
-		"movq %%mm0, 24(%0, %%edi)	\n\t"
-		"movq %%mm2, 32(%0, %%edi)	\n\t"
-				
-		"addl $16, %%esi		\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1280), "r" (f+256)
-		:"%esi", "%edi", "memory"
-	);
+    case A52_3F:
+	if(ch==5) return a52_resample_3F_to_5_C;
 	break;
     case A52_2F2R:
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-		"1:				\n\t"
-		"movq (%1, %%esi), %%mm0	\n\t"
-		"movq 8(%1, %%esi), %%mm1	\n\t"
-		"movq 1024(%1, %%esi), %%mm2	\n\t"
-		"movq 1032(%1, %%esi), %%mm3	\n\t"
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"packssdw %%mm1, %%mm0		\n\t"
-		"packssdw %%mm3, %%mm2		\n\t"
-		"movq 2048(%1, %%esi), %%mm3	\n\t"
-		"movq 2056(%1, %%esi), %%mm4	\n\t"
-		"movq 3072(%1, %%esi), %%mm5	\n\t"
-		"movq 3080(%1, %%esi), %%mm6	\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"psubd %%mm7, %%mm4		\n\t"
-		"psubd %%mm7, %%mm5		\n\t"
-		"psubd %%mm7, %%mm6		\n\t"
-		"packssdw %%mm4, %%mm3		\n\t"
-		"packssdw %%mm6, %%mm5		\n\t"
-		"movq %%mm0, %%mm1		\n\t"
-		"movq %%mm3, %%mm4		\n\t"
-		"punpcklwd %%mm2, %%mm0		\n\t"
-		"punpckhwd %%mm2, %%mm1		\n\t"
-		"punpcklwd %%mm5, %%mm3		\n\t"
-		"punpckhwd %%mm5, %%mm4		\n\t"
-		"movq %%mm0, %%mm2		\n\t"
-		"movq %%mm1, %%mm5		\n\t"
-		"punpckldq %%mm3, %%mm0		\n\t"
-		"punpckhdq %%mm3, %%mm2		\n\t"
-		"punpckldq %%mm4, %%mm1		\n\t"
-		"punpckhdq %%mm4, %%mm5		\n\t"
-		"movq %%mm0, (%0, %%esi,2)	\n\t"
-		"movq %%mm2, 8(%0, %%esi,2)	\n\t"
-		"movq %%mm1, 16(%0, %%esi,2)	\n\t"
-		"movq %%mm5, 24(%0, %%esi,2)	\n\t"
-		"addl $16, %%esi		\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1024), "r" (f+256)
-		:"%esi", "memory"
-	);
+	if(ch==4) return a52_resample_2F_2R_to_4_C;
 	break;
-    case A52_3F2R: 
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-		"1:				\n\t"
-		"movd (%1, %%esi), %%mm0	\n\t"
-		"punpckldq 2048(%1, %%esi), %%mm0\n\t"
-		"movd 3072(%1, %%esi), %%mm1	\n\t"
-		"punpckldq 4096(%1, %%esi), %%mm1\n\t"
-		"movd 1024(%1, %%esi), %%mm2	\n\t"
-		"punpckldq 4(%1, %%esi), %%mm2	\n\t"
-		"movd 2052(%1, %%esi), %%mm3	\n\t"
-		"punpckldq 3076(%1, %%esi), %%mm3\n\t"
-		"movd 4100(%1, %%esi), %%mm4	\n\t"
-		"punpckldq 1028(%1, %%esi), %%mm4\n\t"
-		"movd 8(%1, %%esi), %%mm5	\n\t"
-		"punpckldq 2056(%1, %%esi), %%mm5\n\t"
-		"leal (%%esi, %%esi, 4), %%edi	\n\t"
-		"sarl $1, %%edi			\n\t"
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"psubd %%mm7, %%mm4		\n\t"
-		"psubd %%mm7, %%mm5		\n\t"
-		"packssdw %%mm1, %%mm0		\n\t"
-		"packssdw %%mm3, %%mm2		\n\t"
-		"packssdw %%mm5, %%mm4		\n\t"
-		"movq %%mm0, (%0, %%edi)	\n\t"
-		"movq %%mm2, 8(%0, %%edi)	\n\t"
-		"movq %%mm4, 16(%0, %%edi)	\n\t"
-		
-		"movd 3080(%1, %%esi), %%mm0	\n\t"
-		"punpckldq 4104(%1, %%esi), %%mm0\n\t"
-		"movd 1032(%1, %%esi), %%mm1	\n\t"
-		"punpckldq 12(%1, %%esi), %%mm1\n\t"
-		"movd 2060(%1, %%esi), %%mm2	\n\t"
-		"punpckldq 3084(%1, %%esi), %%mm2\n\t"
-		"movd 4108(%1, %%esi), %%mm3	\n\t"
-		"punpckldq 1036(%1, %%esi), %%mm3\n\t"
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"packssdw %%mm1, %%mm0		\n\t"
-		"packssdw %%mm3, %%mm2		\n\t"
-		"movq %%mm0, 24(%0, %%edi)	\n\t"
-		"movq %%mm2, 32(%0, %%edi)	\n\t"
-				
-		"addl $16, %%esi		\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1280), "r" (f+256)
-		:"%esi", "%edi", "memory"
-	);
+    case A52_3F2R:
+	if(ch==5) return a52_resample_3F_2R_to_5_C;
 	break;
     case A52_MONO | A52_LFE:
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-		"pxor %%mm6, %%mm6		\n\t"
-		"1:				\n\t"
-		"movq 1024(%1, %%esi), %%mm0	\n\t"
-		"movq 1032(%1, %%esi), %%mm1	\n\t"
-		"movq (%1, %%esi), %%mm2	\n\t"
-		"movq 8(%1, %%esi), %%mm3	\n\t"
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"packssdw %%mm1, %%mm0		\n\t"
-		"packssdw %%mm3, %%mm2		\n\t"
-		"movq %%mm0, %%mm1		\n\t"
-		"punpcklwd %%mm2, %%mm0		\n\t"
-		"punpckhwd %%mm2, %%mm1		\n\t"
-		"leal (%%esi, %%esi, 2), %%edi	\n\t"
-		"movq %%mm6, (%0, %%edi)	\n\t"
-		"movd %%mm0, 8(%0, %%edi)	\n\t"
-		"punpckhdq %%mm0, %%mm0		\n\t"
-		"movq %%mm6, 12(%0, %%edi)	\n\t"
-		"movd %%mm0, 20(%0, %%edi)	\n\t"
-		"movq %%mm6, 24(%0, %%edi)	\n\t"
-		"movd %%mm1, 32(%0, %%edi)	\n\t"
-		"punpckhdq %%mm1, %%mm1		\n\t"
-		"movq %%mm6, 36(%0, %%edi)	\n\t"
-		"movd %%mm1, 44(%0, %%edi)	\n\t"
-		"addl $16, %%esi		\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1536), "r" (f+256)
-		:"%esi", "%edi", "memory"
-	);
+	if(ch==6) return a52_resample_MONO_LFE_to_6_C;
 	break;
     case A52_CHANNEL | A52_LFE:
     case A52_STEREO | A52_LFE:
     case A52_DOLBY | A52_LFE:
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-		"pxor %%mm6, %%mm6		\n\t"
-		"1:				\n\t"
-		"movq 1024(%1, %%esi), %%mm0	\n\t"
-		"movq 2048(%1, %%esi), %%mm1	\n\t"
-		"movq (%1, %%esi), %%mm5	\n\t" 
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm5		\n\t"
-		"leal (%%esi, %%esi, 2), %%edi	\n\t"
-		
-		"pxor %%mm4, %%mm4		\n\t"
-		"packssdw %%mm5, %%mm0		\n\t" // FfAa
-		"packssdw %%mm4, %%mm1		\n\t" // 00Bb
-		"punpckhwd %%mm0, %%mm4		\n\t" // F0f0
-		"punpcklwd %%mm1, %%mm0		\n\t" // BAba
-		"movq %%mm0, %%mm1		\n\t" // BAba
-		"punpckldq %%mm4, %%mm3		\n\t" // f0XX
-		"punpckldq %%mm6, %%mm0		\n\t" // 00ba
-		"punpckhdq %%mm1, %%mm3		\n\t" // BAf0
-		
-		"movq %%mm0, (%0, %%edi)	\n\t" // 00ba
-		"punpckhdq %%mm4, %%mm0		\n\t" // F000
-		"movq %%mm3, 8(%0, %%edi)	\n\t" // BAf0
-		"movq %%mm0, 16(%0, %%edi)	\n\t" // F000
-		"addl $8, %%esi			\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1536), "r" (f+256)
-		:"%esi", "%edi", "memory"
-	);
+	if(ch==6) return a52_resample_STEREO_LFE_to_6_C;
 	break;
     case A52_3F | A52_LFE:
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-		"pxor %%mm6, %%mm6		\n\t"
-		"1:				\n\t"
-		"movq 1024(%1, %%esi), %%mm0	\n\t"
-		"movq 3072(%1, %%esi), %%mm1	\n\t"
-		"movq 2048(%1, %%esi), %%mm4	\n\t"
-		"movq (%1, %%esi), %%mm5	\n\t" 
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm4		\n\t"
-		"psubd %%mm7, %%mm5		\n\t"
-		"leal (%%esi, %%esi, 2), %%edi	\n\t"
-		
-		"packssdw %%mm4, %%mm0		\n\t" // EeAa
-		"packssdw %%mm5, %%mm1		\n\t" // FfBb
-		"movq %%mm0, %%mm2		\n\t" // EeAa
-		"punpcklwd %%mm1, %%mm0		\n\t" // BAba
-		"punpckhwd %%mm1, %%mm2		\n\t" // FEfe
-		"movq %%mm0, %%mm1		\n\t" // BAba
-		"punpckldq %%mm6, %%mm0		\n\t" // 00ba
-		"punpckhdq %%mm1, %%mm1		\n\t" // BABA
-		
-		"movq %%mm0, (%0, %%edi)	\n\t"
-		"punpckhdq %%mm2, %%mm0		\n\t" // FE00
-		"punpckldq %%mm1, %%mm2		\n\t" // BAfe
-		"movq %%mm2, 8(%0, %%edi)	\n\t"
-		"movq %%mm0, 16(%0, %%edi)	\n\t"
-		"addl $8, %%esi			\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1536), "r" (f+256)
-		:"%esi", "%edi", "memory"
-	);
+	if(ch==6) return a52_resample_3F_LFE_to_6_C;
 	break;
     case A52_2F2R | A52_LFE:
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-//		"pxor %%mm6, %%mm6		\n\t"
-		"1:				\n\t"
-		"movq 1024(%1, %%esi), %%mm0	\n\t"
-		"movq 2048(%1, %%esi), %%mm1	\n\t"
-		"movq 3072(%1, %%esi), %%mm2	\n\t"
-		"movq 4096(%1, %%esi), %%mm3	\n\t"
-		"movq (%1, %%esi), %%mm5	\n\t" 
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"psubd %%mm7, %%mm5		\n\t"
-		"leal (%%esi, %%esi, 2), %%edi	\n\t"
-		
-		"packssdw %%mm2, %%mm0		\n\t" // CcAa
-		"packssdw %%mm3, %%mm1		\n\t" // DdBb
-		"packssdw %%mm5, %%mm5		\n\t" // FfFf
-		"movq %%mm0, %%mm2		\n\t" // CcAa
-		"punpcklwd %%mm1, %%mm0		\n\t" // BAba
-		"punpckhwd %%mm1, %%mm2		\n\t" // DCdc
-		"pxor %%mm4, %%mm4		\n\t" // 0000
-		"punpcklwd %%mm5, %%mm4		\n\t" // F0f0
-		"movq %%mm0, %%mm1		\n\t" // BAba
-		"movq %%mm4, %%mm3		\n\t" // F0f0
-		"punpckldq %%mm2, %%mm0		\n\t" // dcba
-		"punpckhdq %%mm1, %%mm1		\n\t" // BABA
-		"punpckldq %%mm1, %%mm4		\n\t" // BAf0
-		"punpckhdq %%mm3, %%mm2		\n\t" // F0DC
-		
-		"movq %%mm0, (%0, %%edi)	\n\t"
-		"movq %%mm4, 8(%0, %%edi)	\n\t"
-		"movq %%mm2, 16(%0, %%edi)	\n\t"
-		"addl $8, %%esi			\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1536), "r" (f+256)
-		:"%esi", "%edi", "memory"
-	);
+	if(ch==6) return a52_resample_2F_2R_LFE_to_6_C;
 	break;
     case A52_3F2R | A52_LFE:
-	asm volatile(
-		"movl $-1024, %%esi		\n\t"
-		"movq magicF2W, %%mm7		\n\t"
-//		"pxor %%mm6, %%mm6		\n\t"
-		"1:				\n\t"
-		"movq 1024(%1, %%esi), %%mm0	\n\t"
-		"movq 3072(%1, %%esi), %%mm1	\n\t"
-		"movq 4096(%1, %%esi), %%mm2	\n\t"
-		"movq 5120(%1, %%esi), %%mm3	\n\t"
-		"movq 2048(%1, %%esi), %%mm4	\n\t"
-		"movq (%1, %%esi), %%mm5	\n\t" 
-		"psubd %%mm7, %%mm0		\n\t"
-		"psubd %%mm7, %%mm1		\n\t"
-		"psubd %%mm7, %%mm2		\n\t"
-		"psubd %%mm7, %%mm3		\n\t"
-		"psubd %%mm7, %%mm4		\n\t"
-		"psubd %%mm7, %%mm5		\n\t"
-		"leal (%%esi, %%esi, 2), %%edi	\n\t"
-		
-		"packssdw %%mm2, %%mm0		\n\t" // CcAa
-		"packssdw %%mm3, %%mm1		\n\t" // DdBb
-		"packssdw %%mm4, %%mm4		\n\t" // EeEe
-		"packssdw %%mm5, %%mm5		\n\t" // FfFf
-		"movq %%mm0, %%mm2		\n\t" // CcAa
-		"punpcklwd %%mm1, %%mm0		\n\t" // BAba
-		"punpckhwd %%mm1, %%mm2		\n\t" // DCdc
-		"punpcklwd %%mm5, %%mm4		\n\t" // FEfe
-		"movq %%mm0, %%mm1		\n\t" // BAba
-		"movq %%mm4, %%mm3		\n\t" // FEfe
-		"punpckldq %%mm2, %%mm0		\n\t" // dcba
-		"punpckhdq %%mm1, %%mm1		\n\t" // BABA
-		"punpckldq %%mm1, %%mm4		\n\t" // BAfe
-		"punpckhdq %%mm3, %%mm2		\n\t" // FEDC
-		
-		"movq %%mm0, (%0, %%edi)	\n\t"
-		"movq %%mm4, 8(%0, %%edi)	\n\t"
-		"movq %%mm2, 16(%0, %%edi)	\n\t"
-		"addl $8, %%esi			\n\t"
-		" jnz 1b			\n\t"
-		"emms				\n\t"
-		:: "r" (s16+1536), "r" (f+256)
-		:"%esi", "%edi", "memory"
-	);
+	if(ch==6) return a52_resample_3F_2R_LFE_to_6_C;
 	break;
     }
-    return chans*256;
+    return NULL;
 }
-#endif //arch_x86
-
-void a52_resample_init(uint32_t mm_accel,int _flags,int _chans){
-    chans=_chans;
-    flags=_flags;
-
-    if(a52_resample==NULL) // only once please ;)
-    {
-	    if(mm_accel & MM_ACCEL_X86_MMX) 	fprintf(stderr, "Using MMX optimized resampler\n");
-	    else				fprintf(stderr, "No accelerated resampler found\n");
-    }
-    
-#ifdef ARCH_X86
-    if(mm_accel & MM_ACCEL_X86_MMX) a52_resample= a52_resample_MMX;
-#else
-    if(0);
-#endif
-    else		a52_resample= a52_resample_C;
-}
-
--- a/liba52/resample_mmx.c	Sun Dec 30 21:38:53 2001 +0000
+++ b/liba52/resample_mmx.c	Sun Dec 30 21:44:10 2001 +0000
@@ -1,11 +1,3 @@
-// this code come from a52dec/libao/audio_out_oss.c
-
-// FIXME FIXME FIXME
-
-// a52_resample_init should find the requested converter (from type flags ->
-// given number of channels) and set up some function pointers...
-
-// a52_resample() should do the conversion.
 
 // MMX optimizations from Michael Niedermayer (michaelni@gmx.at) (under GPL)
 
@@ -15,137 +7,14 @@
 	and it would mean (C / MMX2 / MMX / 3DNOW) versions 
 */
 
-#include <inttypes.h>
-#include <stdio.h>
-#include "a52.h"
-#include "mm_accel.h"
-#include "../config.h"
-
-int (* a52_resample) (float * _f, int16_t * s16)=NULL;
-
-#ifdef ARCH_X86
 static uint64_t __attribute__((aligned(8))) magicF2W= 0x43c0000043c00000LL;
 static uint64_t __attribute__((aligned(8))) wm1010= 0xFFFF0000FFFF0000LL;
 static uint64_t __attribute__((aligned(8))) wm0101= 0x0000FFFF0000FFFFLL;
 static uint64_t __attribute__((aligned(8))) wm1100= 0xFFFFFFFF00000000LL;
-#endif
 
-static inline int16_t convert (int32_t i)
-{
-    if (i > 0x43c07fff)
-	return 32767;
-    else if (i < 0x43bf8000)
-	return -32768;
-    else
-	return i - 0x43c00000;
-}
-
-static int chans=2;
-static int flags=0;
-
-int a52_resample_C(float * _f, int16_t * s16)
-{
+static int a52_resample_MONO_to_5_MMX(float * _f, int16_t * s16){
     int i;
     int32_t * f = (int32_t *) _f;
-
-    switch (flags) {
-    case A52_MONO:
-	for (i = 0; i < 256; i++) {
-	    s16[5*i] = s16[5*i+1] = s16[5*i+2] = s16[5*i+3] = 0;
-	    s16[5*i+4] = convert (f[i]);
-	}
-	break;
-    case A52_CHANNEL:
-    case A52_STEREO:
-    case A52_DOLBY:
-	for (i = 0; i < 256; i++) {
-	    s16[2*i] = convert (f[i]);
-	    s16[2*i+1] = convert (f[i+256]);
-	}
-	break;
-    case A52_3F:
-	for (i = 0; i < 256; i++) {
-	    s16[5*i] = convert (f[i]);
-	    s16[5*i+1] = convert (f[i+512]);
-	    s16[5*i+2] = s16[5*i+3] = 0;
-	    s16[5*i+4] = convert (f[i+256]);
-	}
-	break;
-    case A52_2F2R:
-	for (i = 0; i < 256; i++) {
-	    s16[4*i] = convert (f[i]);
-	    s16[4*i+1] = convert (f[i+256]);
-	    s16[4*i+2] = convert (f[i+512]);
-	    s16[4*i+3] = convert (f[i+768]);
-	}
-	break;
-    case A52_3F2R:
-	for (i = 0; i < 256; i++) {
-	    s16[5*i] = convert (f[i]);
-	    s16[5*i+1] = convert (f[i+512]);
-	    s16[5*i+2] = convert (f[i+768]);
-	    s16[5*i+3] = convert (f[i+1024]);
-	    s16[5*i+4] = convert (f[i+256]);
-	}
-	break;
-    case A52_MONO | A52_LFE:
-	for (i = 0; i < 256; i++) {
-	    s16[6*i] = s16[6*i+1] = s16[6*i+2] = s16[6*i+3] = 0;
-	    s16[6*i+4] = convert (f[i+256]);
-	    s16[6*i+5] = convert (f[i]);
-	}
-	break;
-    case A52_CHANNEL | A52_LFE:
-    case A52_STEREO | A52_LFE:
-    case A52_DOLBY | A52_LFE:
-	for (i = 0; i < 256; i++) {
-	    s16[6*i] = convert (f[i+256]);
-	    s16[6*i+1] = convert (f[i+512]);
-	    s16[6*i+2] = s16[6*i+3] = s16[6*i+4] = 0;
-	    s16[6*i+5] = convert (f[i]);
-	}
-	break;
-    case A52_3F | A52_LFE:
-	for (i = 0; i < 256; i++) {
-	    s16[6*i] = convert (f[i+256]);
-	    s16[6*i+1] = convert (f[i+768]);
-	    s16[6*i+2] = s16[6*i+3] = 0;
-	    s16[6*i+4] = convert (f[i+512]);
-	    s16[6*i+5] = convert (f[i]);
-	}
-	break;
-    case A52_2F2R | A52_LFE:
-	for (i = 0; i < 256; i++) {
-	    s16[6*i] = convert (f[i+256]);
-	    s16[6*i+1] = convert (f[i+512]);
-	    s16[6*i+2] = convert (f[i+768]);
-	    s16[6*i+3] = convert (f[i+1024]);
-	    s16[6*i+4] = 0;
-	    s16[6*i+5] = convert (f[i]);
-	}
-	break;
-    case A52_3F2R | A52_LFE:
-	for (i = 0; i < 256; i++) {
-	    s16[6*i] = convert (f[i+256]);
-	    s16[6*i+1] = convert (f[i+768]);
-	    s16[6*i+2] = convert (f[i+1024]);
-	    s16[6*i+3] = convert (f[i+1280]);
-	    s16[6*i+4] = convert (f[i+512]);
-	    s16[6*i+5] = convert (f[i]);
-	}
-	break;
-    }
-    return chans*256;
-}
-
-#ifdef ARCH_X86
-int a52_resample_MMX(float * _f, int16_t * s16)
-{
-    int i;
-    int32_t * f = (int32_t *) _f;
-
-    switch (flags) {
-    case A52_MONO:
 	asm volatile(
 		"movl $-512, %%esi		\n\t"
 		"movq magicF2W, %%mm7		\n\t"
@@ -178,10 +47,12 @@
 		:: "r" (s16+1280), "r" (f+256)
 		:"%esi", "%edi", "memory"
 	);
-	break;
-    case A52_CHANNEL:
-    case A52_STEREO:
-    case A52_DOLBY:
+    return 5*256;
+}
+
+static int a52_resample_STEREO_to_2_MMX(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 /* benchmark scores are 0.3% better with SSE but we would need to set bias=0 and premultiply it
 #ifdef HAVE_SSE
 	asm volatile(
@@ -225,8 +96,12 @@
 		:: "r" (s16+512), "r" (f+256)
 		:"%esi", "memory"
 	);
-	break;
-    case A52_3F: 
+    return 2*256;
+}
+
+static int a52_resample_3F_to_5_MMX(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 	asm volatile(
 		"movl $-1024, %%esi		\n\t"
 		"movq magicF2W, %%mm7		\n\t"
@@ -277,8 +152,12 @@
 		:: "r" (s16+1280), "r" (f+256)
 		:"%esi", "%edi", "memory"
 	);
-	break;
-    case A52_2F2R:
+    return 5*256;
+}
+
+static int a52_resample_2F_2R_to_4_MMX(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 	asm volatile(
 		"movl $-1024, %%esi		\n\t"
 		"movq magicF2W, %%mm7		\n\t"
@@ -325,8 +204,12 @@
 		:: "r" (s16+1024), "r" (f+256)
 		:"%esi", "memory"
 	);
-	break;
-    case A52_3F2R: 
+    return 4*256;
+}
+
+static int a52_resample_3F_2R_to_5_MMX(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 	asm volatile(
 		"movl $-1024, %%esi		\n\t"
 		"movq magicF2W, %%mm7		\n\t"
@@ -381,8 +264,12 @@
 		:: "r" (s16+1280), "r" (f+256)
 		:"%esi", "%edi", "memory"
 	);
-	break;
-    case A52_MONO | A52_LFE:
+    return 5*256;
+}
+
+static int a52_resample_MONO_LFE_to_6_MMX(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 	asm volatile(
 		"movl $-1024, %%esi		\n\t"
 		"movq magicF2W, %%mm7		\n\t"
@@ -418,10 +305,12 @@
 		:: "r" (s16+1536), "r" (f+256)
 		:"%esi", "%edi", "memory"
 	);
-	break;
-    case A52_CHANNEL | A52_LFE:
-    case A52_STEREO | A52_LFE:
-    case A52_DOLBY | A52_LFE:
+    return 6*256;
+}
+
+static int a52_resample_STEREO_LFE_to_6_MMX(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 	asm volatile(
 		"movl $-1024, %%esi		\n\t"
 		"movq magicF2W, %%mm7		\n\t"
@@ -455,8 +344,12 @@
 		:: "r" (s16+1536), "r" (f+256)
 		:"%esi", "%edi", "memory"
 	);
-	break;
-    case A52_3F | A52_LFE:
+    return 6*256;
+}
+
+static int a52_resample_3F_LFE_to_6_MMX(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 	asm volatile(
 		"movl $-1024, %%esi		\n\t"
 		"movq magicF2W, %%mm7		\n\t"
@@ -492,8 +385,12 @@
 		:: "r" (s16+1536), "r" (f+256)
 		:"%esi", "%edi", "memory"
 	);
-	break;
-    case A52_2F2R | A52_LFE:
+    return 6*256;
+}
+
+static int a52_resample_2F_2R_LFE_to_6_MMX(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 	asm volatile(
 		"movl $-1024, %%esi		\n\t"
 		"movq magicF2W, %%mm7		\n\t"
@@ -535,8 +432,12 @@
 		:: "r" (s16+1536), "r" (f+256)
 		:"%esi", "%edi", "memory"
 	);
-	break;
-    case A52_3F2R | A52_LFE:
+    return 6*256;
+}
+
+static int a52_resample_3F_2R_LFE_to_6_MMX(float * _f, int16_t * s16){
+    int i;
+    int32_t * f = (int32_t *) _f;
 	asm volatile(
 		"movl $-1024, %%esi		\n\t"
 		"movq magicF2W, %%mm7		\n\t"
@@ -580,27 +481,48 @@
 		:: "r" (s16+1536), "r" (f+256)
 		:"%esi", "%edi", "memory"
 	);
+    return 6*256;
+}
+
+
+static void* a52_resample_MMX(int flags, int ch){
+    switch (flags) {
+    case A52_MONO:
+	if(ch==5) return a52_resample_MONO_to_5_MMX;
+	break;
+    case A52_CHANNEL:
+    case A52_STEREO:
+    case A52_DOLBY:
+	if(ch==2) return a52_resample_STEREO_to_2_MMX;
+	break;
+    case A52_3F:
+	if(ch==5) return a52_resample_3F_to_5_MMX;
+	break;
+    case A52_2F2R:
+	if(ch==4) return a52_resample_2F_2R_to_4_MMX;
+	break;
+    case A52_3F2R:
+	if(ch==5) return a52_resample_3F_2R_to_5_MMX;
+	break;
+    case A52_MONO | A52_LFE:
+	if(ch==6) return a52_resample_MONO_LFE_to_6_MMX;
+	break;
+    case A52_CHANNEL | A52_LFE:
+    case A52_STEREO | A52_LFE:
+    case A52_DOLBY | A52_LFE:
+	if(ch==6) return a52_resample_STEREO_LFE_to_6_MMX;
+	break;
+    case A52_3F | A52_LFE:
+	if(ch==6) return a52_resample_3F_LFE_to_6_MMX;
+	break;
+    case A52_2F2R | A52_LFE:
+	if(ch==6) return a52_resample_2F_2R_LFE_to_6_MMX;
+	break;
+    case A52_3F2R | A52_LFE:
+	if(ch==6) return a52_resample_3F_2R_LFE_to_6_MMX;
 	break;
     }
-    return chans*256;
-}
-#endif //arch_x86
-
-void a52_resample_init(uint32_t mm_accel,int _flags,int _chans){
-    chans=_chans;
-    flags=_flags;
-
-    if(a52_resample==NULL) // only once please ;)
-    {
-	    if(mm_accel & MM_ACCEL_X86_MMX) 	fprintf(stderr, "Using MMX optimized resampler\n");
-	    else				fprintf(stderr, "No accelerated resampler found\n");
-    }
-    
-#ifdef ARCH_X86
-    if(mm_accel & MM_ACCEL_X86_MMX) a52_resample= a52_resample_MMX;
-#else
-    if(0);
-#endif
-    else		a52_resample= a52_resample_C;
+    return NULL;
 }
 
+