Added simd patch (use sse) ~x86 masked. Closes #21585.

9 files changed, 1039 insertions, 110 deletions

diff --git a/media-libs/libvorbis/Manifest b/media-libs/libvorbis/Manifest
index 3aefcf7106cc..370f1349411c 100644
--- a/media-libs/libvorbis/Manifest
+++ b/media-libs/libvorbis/Manifest
@@ -1,13 +1,7 @@
-MD5 13ba2657dde7f84a0389ee0f31fd2523 ChangeLog 2029
-MD5 e9a0d9e24916cc5c07ae807bf644b5e8 libvorbis-1.0-r1.ebuild 1562
-MD5 547a992279bd48bc8c0ec30ba5a276f2 libvorbis-1.0.ebuild 1490
-MD5 41684381f21d3559c14320860478658c libvorbis-1.0-r2.ebuild 1710
-MD5 5603100e019622c95077292668929b6e libvorbis-1.0-r3.ebuild 1777
-MD5 0a0d3872bdf2c6da4e3fe5471fd3eb16 files/digest-libvorbis-1.0 65
-MD5 0a0d3872bdf2c6da4e3fe5471fd3eb16 files/digest-libvorbis-1.0-r1 65
-MD5 b8048905ed8638913c7b6302fdc3eb4b files/libvorbis-1.0-m4.patch 1032
+MD5 3df66b52d2f66cad749719b268ee3dce ChangeLog 2321
+MD5 073b0e5bb9e3c2061a1fa745f4130320 libvorbis-1.0-r2.ebuild 1707
+MD5 caf792187e22d3bcdee52fa25bf4283b libvorbis-1.0-r3.ebuild 1773
 MD5 0a0d3872bdf2c6da4e3fe5471fd3eb16 files/digest-libvorbis-1.0-r2 65
-MD5 069e26021a32d3d105c544229e071e5f files/libvorbis-1.0-r2-m4.patch 919
 MD5 0a0d3872bdf2c6da4e3fe5471fd3eb16 files/digest-libvorbis-1.0-r3 65
 MD5 174ab9630810bce8aac9eac4f4f20247 files/libvorbis-simd.patch 28887
-MD5 b8048905ed8638913c7b6302fdc3eb4b files/libvorbis-m4.patch 1032
+MD5 069e26021a32d3d105c544229e071e5f files/libvorbis-m4.patch 919
diff --git a/media-libs/libvorbis/files/digest-libvorbis-1.0-r1 b/media-libs/libvorbis/files/digest-libvorbis-1.0-r1
deleted file mode 100644
index 6c800c478006..000000000000
--- a/media-libs/libvorbis/files/digest-libvorbis-1.0-r1
+++ /dev/null
@@ -1 +0,0 @@
-MD5 d1ad94fe8e240269c790e18992171e53 libvorbis-1.0.tar.gz 749064
diff --git a/media-libs/libvorbis/files/digest-libvorbis-1.0 b/media-libs/libvorbis/files/digest-libvorbis-1.0-r3
index 6c800c478006..6c800c478006 100644
--- a/media-libs/libvorbis/files/digest-libvorbis-1.0
+++ b/media-libs/libvorbis/files/digest-libvorbis-1.0-r3
diff --git a/media-libs/libvorbis/files/libvorbis-1.0-m4.patch b/media-libs/libvorbis/files/libvorbis-1.0-m4.patch
deleted file mode 100644
index 43828e527c6a..000000000000
--- a/media-libs/libvorbis/files/libvorbis-1.0-m4.patch
+++ /dev/null
@@ -1,31 +0,0 @@
-diff -u -r libvorbis-1.0/vorbis.m4 libvorbis-cvs/vorbis.m4
---- libvorbis-1.0/vorbis.m4	2002-07-09 23:08:57.000000000 +1000
-+++ libvorbis-cvs/vorbis.m4	2002-08-07 02:01:40.000000000 +1000
-@@ -54,6 +54,7 @@
- #include <stdlib.h>
- #include <string.h>
- #include <vorbis/codec.h>
-+#include <vorbis/vorbisenc.h>
- 
- int main ()
- {
-@@ -62,7 +63,7 @@
-     vorbis_info		vi;
- 
-     vorbis_info_init (&vi);
--    vorbis_encode_init (&vi, 2, 44100, -1, 128, -1);
-+    vorbis_encode_init (&vi, 2, 44100, -1, 128000, -1);
-     vorbis_analysis_init (&vd, &vi);
-     vorbis_block_init (&vd, &vb);
-     /* this function was added in 1.0rc3, so this is what we're testing for */
-@@ -86,8 +87,8 @@
-        :
-      else
-        echo "*** Could not run Vorbis test program, checking why..."
--       CFLAGS="$CFLAGS $VORBIS_CFLAGS"
--       LIBS="$LIBS $VORBIS_LIBS $OGG_LIBS"
-+       CFLAGS="$CFLAGS $VORBIS_CFLAGS $OGG_CFLAGS"
-+       LIBS="$LIBS $VORBIS_LIBS $VORBISENC_LIBS $OGG_LIBS"
-        AC_TRY_LINK([
- #include <stdio.h>
- #include <vorbis/codec.h>
diff --git a/media-libs/libvorbis/files/libvorbis-1.0-r2-m4.patch b/media-libs/libvorbis/files/libvorbis-m4.patch
index 55a82bc98411..55a82bc98411 100644
--- a/media-libs/libvorbis/files/libvorbis-1.0-r2-m4.patch
+++ b/media-libs/libvorbis/files/libvorbis-m4.patch
diff --git a/media-libs/libvorbis/files/libvorbis-simd.patch b/media-libs/libvorbis/files/libvorbis-simd.patch
new file mode 100644
index 000000000000..85f1d1aef7be
--- /dev/null
+++ b/media-libs/libvorbis/files/libvorbis-simd.patch
@@ -0,0 +1,1019 @@
+diff -ur libvorbis-1.0/lib/block.c libvorbis-1.0-simd/lib/block.c
+--- libvorbis-1.0/lib/block.c	2002-07-11 08:40:48.000000000 +0200
++++ libvorbis-1.0-simd/lib/block.c	2003-04-26 20:32:07.000000000 +0200
+@@ -22,6 +22,7 @@
+ #include <stdlib.h>
+ #include <string.h>
+ #include <ogg/ogg.h>
++#include <assert.h>
+ #include "vorbis/codec.h"
+ #include "codec_internal.h"
+ 
+@@ -31,6 +32,10 @@
+ #include "registry.h"
+ #include "misc.h"
+ 
++#ifdef __SSE__
++#include <xmmintrin.h>
++#endif
++
+ static int ilog2(unsigned int v){
+   int ret=0;
+   if(v)--v;
+@@ -701,11 +706,95 @@
+       /* the overlap/add section */
+       if(v->lW){
+ 	if(v->W){
++#ifdef __SSE__
++	  /* large/large */
++	  float *pcm=v->pcm[j]+prevCenter;
++	  float *p=vb->pcm[j];
++	  unsigned long _pcm=(unsigned long)pcm&15;
++	  unsigned long _p=(unsigned long)p&15;
++	  register __m128* PCM,* P;
++	  i=0;
++	  /* n1 is always 1024, prevCenter is either 0 or 1024 */
++	  /* both pcm and p can be unaligned, and they usually are.
++	   * This code assumes unaligned addresses are still 8-byte
++	   * aligned (which is true because glibc's malloc does 8-byte
++	   * alignment */
++	  if (_pcm) {
++	    pcm[0]+=p[0]; pcm[1]+=p[1];
++	    i=2;
++	    PCM=(__m128*)(pcm+2);
++	    P=(__m128*)(p+2);
++	  } else {
++	    PCM=(__m128*)pcm;
++	    P=(__m128*)p;
++	  }
++	  if (_pcm ^ _p) {
++	    /* one is properly aligned, the other is not */
++	    register __m128 a;
++	    P=(__m128*)((float*)P-2);
++	    a=P[0];
++	    for (; i+7<n1; i+=8) {
++	      register __m128 b=P[1];
++	      /* now we need the upper most floats from a and the lower
++	      * most floats from b */
++	      *PCM=_mm_add_ps(_mm_movehl_ps(_mm_shuffle_ps(b,b,_MM_SHUFFLE(1,0,3,2)),a),*PCM);
++	      a=P[2];
++	      PCM[1]=_mm_add_ps(_mm_movehl_ps(_mm_shuffle_ps(a,a,_MM_SHUFFLE(1,0,3,2)),b),PCM[1]);
++	      PCM+=2; P+=2;
++	    }
++	  } else {
++	    for (; i+3<n1; i+=4) {
++	      *PCM=_mm_add_ps(*P,*PCM);
++	      ++PCM; ++P;
++	    }
++	  }
++	  /* strange, does not appear to happen */
++	  for(;i<n1;i++)
++	    pcm[i]+=p[i];
++#elif defined(simd_3dn)
++	  assert((n1&1)==0);
++      asm volatile("jecxz 2f\n\t"
++	  "1: "
++	  "movq (%1),%%mm0\n\t"
++	  "addl $8,%1\n\t"
++	  "pfadd (%2),%%mm0\n\t"
++	  "movq %%mm0,(%2)\n\t"
++	  "addl $8,%2\n\t"
++	  "subl $2,%%ecx\n\t"
++	  "jz 2f\n\t"
++	  "movq (%1),%%mm0\n\t"
++	  "addl $8,%1\n\t"
++	  "pfadd (%2),%%mm0\n\t"
++	  "movq %%mm0,(%2)\n\t"
++	  "addl $8,%2\n\t"
++	  "subl $2,%%ecx\n\t"
++	  "jz 2f\n\t"
++	  "movq (%1),%%mm0\n\t"
++	  "addl $8,%1\n\t"
++	  "pfadd (%2),%%mm0\n\t"
++	  "movq %%mm0,(%2)\n\t"
++	  "addl $8,%2\n\t"
++	  "subl $2,%%ecx\n\t"
++	  "jz 2f\n\t"
++	  "movq (%1),%%mm0\n\t"
++	  "addl $8,%1\n\t"
++	  "pfadd (%2),%%mm0\n\t"
++	  "movq %%mm0,(%2)\n\t"
++	  "addl $8,%2\n\t"
++	  "subl $2,%%ecx\n\t"
++	  "jz 2f\n\t"
++	  "jmp 1b\n\t"
++	  "2: femms\n\t"
++	  : : "c" (n1),
++	    "r" ((float*)(vb->pcm[j])),
++	    "r" ((float*)(v->pcm[j]+prevCenter)) : "memory" );
++#else
+ 	  /* large/large */
+ 	  float *pcm=v->pcm[j]+prevCenter;
+ 	  float *p=vb->pcm[j];
+ 	  for(i=0;i<n1;i++)
+ 	    pcm[i]+=p[i];
++#endif
+ 	}else{
+ 	  /* large/small */
+ 	  float *pcm=v->pcm[j]+prevCenter+n1/2-n0/2;
+@@ -732,12 +821,60 @@
+       }
+       
+       /* the copy section */
++#ifdef simd_3dn
++      assert((n&1)==0);
++      asm volatile("jecxz 2f\n\t"
++	  "testl $1,%%ecx\n\t"
++	  "jz 1f\n\t"
++	  "3: movl (%1),%%eax\n\t"
++	  "addl $4,%1\n\t"
++	  "movl %%eax,(%2)\n\t"
++	  "addl $4,%2\n\t"
++	  "decl %%ecx\n\t"
++	  "jz 2f\n\t"
++	  "1: "
++	  "movq (%1),%%mm0\n\t"
++	  "addl $8,%1\n\t"
++	  "movq %%mm0,(%2)\n\t"
++	  "addl $8,%2\n\t"
++	  "subl $2,%%ecx\n\t"
++	  "jc 3b\n\t"
++	  "jz 2f\n\t"
++	  "movq (%1),%%mm0\n\t"
++	  "addl $8,%1\n\t"
++	  "movq %%mm0,(%2)\n\t"
++	  "addl $8,%2\n\t"
++	  "subl $2,%%ecx\n\t"
++	  "jc 3b\n\t"
++	  "jz 2f\n\t"
++	  "movq (%1),%%mm0\n\t"
++	  "addl $8,%1\n\t"
++	  "movq %%mm0,(%2)\n\t"
++	  "addl $8,%2\n\t"
++	  "subl $2,%%ecx\n\t"
++	  "jc 3b\n\t"
++	  "jz 2f\n\t"
++	  "movq (%1),%%mm0\n\t"
++	  "addl $8,%1\n\t"
++	  "movq %%mm0,(%2)\n\t"
++	  "addl $8,%2\n\t"
++	  "subl $2,%%ecx\n\t"
++	  "jc 3b\n\t"
++	  "jz 2f\n\t"
++	  "jmp 1b\n\t"
++	  "2: femms\n\t"
++	  : : "c" (n),
++	    "r" ((float*)(vb->pcm[j]+n)),
++	    "r" ((float*)(v->pcm[j]+thisCenter))
++	  : "%eax", "memory" );
++#else
+       {
+ 	float *pcm=v->pcm[j]+thisCenter;
+ 	float *p=vb->pcm[j]+n;
+ 	for(i=0;i<n;i++)
+ 	  pcm[i]=p[i];
+       }
++#endif
+     }
+     
+     if(v->centerW)
+diff -ur libvorbis-1.0/lib/lsp.c libvorbis-1.0-simd/lib/lsp.c
+--- libvorbis-1.0/lib/lsp.c	2002-07-17 23:28:37.000000000 +0200
++++ libvorbis-1.0-simd/lib/lsp.c	2003-04-26 20:32:27.000000000 +0200
+@@ -54,7 +54,12 @@
+ #define   FLOAT_LOOKUP
+ #undef    INT_LOOKUP
+ 
++#ifdef __SSE__
++#include <xmmintrin.h>
++#endif
++
+ #ifdef FLOAT_LOOKUP
++
+ #include "lookup.c" /* catch this in the build system; we #include for
+                        compilers (like gcc) that can't inline across
+                        modules */
+@@ -73,17 +78,88 @@
+   while(i<n){
+     int k=map[i];
+     int qexp;
++#ifdef __SSE__
++    register __m128 pqpq;
++    static float __attribute__((aligned(16))) PQPQ[4];
++    register __m128 wwww;
++    float pq[2];
++#define p pq[1]
++#define q pq[0]
++#else
++#ifdef simd_3dn
++    float pq[2],ww[2];
++#define p pq[1]
++#define q pq[0]
++#define w ww[0]
++#else
+     float p=.7071067812f;
+     float q=.7071067812f;
+     float w=vorbis_coslook(wdel*k);
++#endif
++#endif
+     float *ftmp=lsp;
+     int c=m>>1;
+ 
++#ifdef __SSE__
++    static float __attribute__((aligned(16))) w;
++    w=vorbis_coslook(wdel*k);
++    PQPQ[0]=PQPQ[1]=.7071067812f;
++    if ((((long)ftmp)&15)==8) {
++      PQPQ[2]=ftmp[0]-w;
++      PQPQ[3]=ftmp[1]-w;
++      --c;
++      ftmp+=2;
++    } else {
++      PQPQ[2]=PQPQ[3]=1.f;
++    }
++    pqpq=*(__m128*)&(PQPQ[0]);
++    wwww=_mm_load1_ps(&w);
++#define UNROLL
++#ifdef UNROLL
++    while (c>3) {
++      pqpq=_mm_mul_ps(pqpq,_mm_sub_ps(*(__m128*)ftmp,wwww));
++      ftmp+=4;
++      pqpq=_mm_mul_ps(pqpq,_mm_sub_ps(*(__m128*)ftmp,wwww));
++      ftmp+=4;
++      c-=4;
++    }
++#endif
++    while (c>1) {
++      pqpq=_mm_mul_ps(pqpq,_mm_sub_ps(*(__m128*)ftmp,wwww));
++      ftmp+=4;
++      c-=2;
++    }
++    pqpq=_mm_mul_ps(pqpq,_mm_shuffle_ps(pqpq,pqpq,_MM_SHUFFLE(1,0,3,2)));
++    _mm_storel_pi((__m64*)(&(pq[0])),pqpq);
++    if (c) {
++      q*=ftmp[0]-w;
++      p*=ftmp[1]-w;
++      ftmp+=2;
++    };
++#else
++#ifdef simd_3dn
++    pq[0]=pq[1]=.7071067812f;
++    ww[0]=ww[1]=vorbis_coslook(wdel*k);
++
++    asm volatile("movq (%2),%%mm1\n\t"		/* mm1 = (w,w) */
++	"movq (%3),%%mm2\n\t"		/* mm2 = (p,q) */
++	"1: movq (%0),%%mm0\n\t"	/* mm0 = (ftmp[0],ftmp[1]) */
++	"pfsub %%mm1,%%mm0\n\t"		/* mm0 = (ftmp[0]-w,ftmp[1]-w) */
++	"pfmul %%mm0,%%mm2\n\t"		/* mm2 *= (ftmp[0]-w,ftmp[1]-w) */
++	"addl $8,%0\n\t"		/* ftmp += 2 */
++	"decl %1\n\t"			/* --c */
++	"jnz 1b\n\t"
++	"movq %%mm2,(%3)\n\t"		/* pq = mm0 */
++	"femms\n\t"
++	: "+r" (ftmp), "+r" (c) : "r" (ww), "r" (pq) : "memory" );
++#else
+     do{
+       q*=ftmp[0]-w;
+       p*=ftmp[1]-w;
+       ftmp+=2;
+     }while(--c);
++#endif
++#endif
+ 
+     if(m&1){
+       /* odd order filter; slightly assymetric */
+@@ -107,6 +183,9 @@
+       curve[i++]*=q;
+     }while(map[i]==k);
+   }
++#undef p
++#undef q
++#undef w
+   vorbis_fpu_restore(fpu);
+ }
+ 
+diff -ur libvorbis-1.0/lib/mdct.c libvorbis-1.0-simd/lib/mdct.c
+--- libvorbis-1.0/lib/mdct.c	2002-06-29 00:19:36.000000000 +0200
++++ libvorbis-1.0-simd/lib/mdct.c	2003-01-26 07:09:54.000000000 +0100
+@@ -271,41 +271,158 @@
+   REG_TYPE   r0;
+   REG_TYPE   r1;
+ 
++#if DATA_TYPE != float
++#undef simd_3dn
++#endif
++
+   do{
+     
++#ifdef simd_3dn
++    static unsigned long negxor[2]={0x80000000,0};
++    asm (			/* mm0 = (r0,r1) */
++	"movq %0,%%mm0\n\t"	/* mm0 = (x1[6],x1[7]) */
++	"movq %1,%%mm1\n\t"	/* mm1 = (x2[6],x2[7]) */
++	"movq %%mm0,%%mm2\n\t"	/* mm2 = (x1[6],x1[7]) */
++	"pfsub %%mm1,%%mm0\n\t"	/* mm0 = (x1[6]-x2[6],x1[7]-x2[7]) */
++	"pfadd %%mm1,%%mm2\n\t"	/* mm2 = (x1[6]+x2[6],x1[7]+x2[7]) */
++	"movq %%mm2,%0\n\t"
++
++	"movq %2,%%mm1\n\t"
++	"movq %%mm0,%%mm2\n\t"
++	"pfmul %%mm1,%%mm0\n\t"
++	"pxor %3,%%mm2\n\t"
++#ifdef simd_sse
++	"pshufw $0x4e,%%mm1,%%mm1\n\t"
++#else
++	"movq %%mm1,%%mm3\n\t"
++	"psllq $32,%%mm3\n\t"
++	"psrlq $32,%%mm1\n\t"
++	"por %%mm3,%%mm1\n\t"
++#endif
++	"pfmul %%mm2,%%mm1\n\t"
++	"pfacc %%mm1,%%mm0\n\t"
++	"movq %%mm0,%1\n\t"
++	"femms\n\t"
++	: : "m" (x1[6]), "m" (x2[6]), "m" (T[0]), "m" (negxor[0]) : "memory");
++#else
+                r0      = x1[6]      -  x2[6];
+ 	       r1      = x1[7]      -  x2[7];
+ 	       x1[6]  += x2[6];
+ 	       x1[7]  += x2[7];
+ 	       x2[6]   = MULT_NORM(r1 * T[1]  +  r0 * T[0]);
+ 	       x2[7]   = MULT_NORM(r1 * T[0]  -  r0 * T[1]);
++#endif
+ 	       
+ 	       T+=trigint;
+ 	       
++#ifdef simd_3dn
++    asm (			/* mm0 = (r0,r1) */
++	"movq %0,%%mm0\n\t"	/* mm0 = (x1[6],x1[7]) */
++	"movq %1,%%mm1\n\t"	/* mm1 = (x2[6],x2[7]) */
++	"movq %%mm0,%%mm2\n\t"	/* mm2 = (x1[6],x1[7]) */
++	"pfsub %%mm1,%%mm0\n\t"	/* mm0 = (x1[6]-x2[6],x1[7]-x2[7]) */
++	"pfadd %%mm1,%%mm2\n\t"	/* mm2 = (x1[6]+x2[6],x1[7]+x2[7]) */
++	"movq %%mm2,%0\n\t"
++
++	"movq %2,%%mm1\n\t"
++	"movq %%mm0,%%mm2\n\t"
++	"pfmul %%mm1,%%mm0\n\t"
++	"pxor %3,%%mm2\n\t"
++#ifdef simd_sse
++	"pshufw $0x4e,%%mm1,%%mm1\n\t"
++#else
++	"movq %%mm1,%%mm3\n\t"
++	"psllq $32,%%mm3\n\t"
++	"psrlq $32,%%mm1\n\t"
++	"por %%mm3,%%mm1\n\t"
++#endif
++	"pfmul %%mm2,%%mm1\n\t"
++	"pfacc %%mm1,%%mm0\n\t"
++	"movq %%mm0,%1\n\t"
++	"femms\n\t"
++	: : "m" (x1[4]), "m" (x2[4]), "m" (T[0]), "m" (negxor[0]) : "memory");
++#else
+ 	       r0      = x1[4]      -  x2[4];
+ 	       r1      = x1[5]      -  x2[5];
+ 	       x1[4]  += x2[4];
+ 	       x1[5]  += x2[5];
+ 	       x2[4]   = MULT_NORM(r1 * T[1]  +  r0 * T[0]);
+ 	       x2[5]   = MULT_NORM(r1 * T[0]  -  r0 * T[1]);
++#endif
+ 	       
+ 	       T+=trigint;
+ 	       
++#ifdef simd_3dn
++    asm (			/* mm0 = (r0,r1) */
++	"movq %0,%%mm0\n\t"	/* mm0 = (x1[6],x1[7]) */
++	"movq %1,%%mm1\n\t"	/* mm1 = (x2[6],x2[7]) */
++	"movq %%mm0,%%mm2\n\t"	/* mm2 = (x1[6],x1[7]) */
++	"pfsub %%mm1,%%mm0\n\t"	/* mm0 = (x1[6]-x2[6],x1[7]-x2[7]) */
++	"pfadd %%mm1,%%mm2\n\t"	/* mm2 = (x1[6]+x2[6],x1[7]+x2[7]) */
++	"movq %%mm2,%0\n\t"
++
++	"movq %2,%%mm1\n\t"
++	"movq %%mm0,%%mm2\n\t"
++	"pfmul %%mm1,%%mm0\n\t"
++	"pxor %3,%%mm2\n\t"
++#ifdef simd_sse
++	"pshufw $0x4e,%%mm1,%%mm1\n\t"
++#else
++	"movq %%mm1,%%mm3\n\t"
++	"psllq $32,%%mm3\n\t"
++	"psrlq $32,%%mm1\n\t"
++	"por %%mm3,%%mm1\n\t"
++#endif
++	"pfmul %%mm2,%%mm1\n\t"
++	"pfacc %%mm1,%%mm0\n\t"
++	"movq %%mm0,%1\n\t"
++	"femms\n\t"
++	: : "m" (x1[2]), "m" (x2[2]), "m" (T[0]), "m" (negxor[0]) : "memory");
++#else
+ 	       r0      = x1[2]      -  x2[2];
+ 	       r1      = x1[3]      -  x2[3];
+ 	       x1[2]  += x2[2];
+ 	       x1[3]  += x2[3];
+ 	       x2[2]   = MULT_NORM(r1 * T[1]  +  r0 * T[0]);
+ 	       x2[3]   = MULT_NORM(r1 * T[0]  -  r0 * T[1]);
++#endif
+ 	       
+ 	       T+=trigint;
+ 	       
++#ifdef simd_3dn
++    asm (			/* mm0 = (r0,r1) */
++	"movq %0,%%mm0\n\t"	/* mm0 = (x1[6],x1[7]) */
++	"movq %1,%%mm1\n\t"	/* mm1 = (x2[6],x2[7]) */
++	"movq %%mm0,%%mm2\n\t"	/* mm2 = (x1[6],x1[7]) */
++	"pfsub %%mm1,%%mm0\n\t"	/* mm0 = (x1[6]-x2[6],x1[7]-x2[7]) */
++	"pfadd %%mm1,%%mm2\n\t"	/* mm2 = (x1[6]+x2[6],x1[7]+x2[7]) */
++	"movq %%mm2,%0\n\t"
++
++	"movq %2,%%mm1\n\t"
++	"movq %%mm0,%%mm2\n\t"
++	"pfmul %%mm1,%%mm0\n\t"
++	"pxor %3,%%mm2\n\t"
++#ifdef simd_sse
++	"pshufw $0x4e,%%mm1,%%mm1\n\t"
++#else
++	"movq %%mm1,%%mm3\n\t"
++	"psllq $32,%%mm3\n\t"
++	"psrlq $32,%%mm1\n\t"
++	"por %%mm3,%%mm1\n\t"
++#endif
++	"pfmul %%mm2,%%mm1\n\t"
++	"pfacc %%mm1,%%mm0\n\t"
++	"movq %%mm0,%1\n\t"
++	"femms\n\t"
++	: : "m" (x1[0]), "m" (x2[0]), "m" (T[0]), "m" (negxor[0]) : "memory");
++#else
+ 	       r0      = x1[0]      -  x2[0];
+ 	       r1      = x1[1]      -  x2[1];
+ 	       x1[0]  += x2[0];
+ 	       x1[1]  += x2[1];
+ 	       x2[0]   = MULT_NORM(r1 * T[1]  +  r0 * T[0]);
+ 	       x2[1]   = MULT_NORM(r1 * T[0]  -  r0 * T[1]);
++#endif
+ 
+ 	       T+=trigint;
+     x1-=8;
+diff -ur libvorbis-1.0/lib/psy.c libvorbis-1.0-simd/lib/psy.c
+--- libvorbis-1.0/lib/psy.c	2002-07-13 12:18:33.000000000 +0200
++++ libvorbis-1.0-simd/lib/psy.c	2003-04-17 20:50:13.000000000 +0200
+@@ -29,9 +29,25 @@
+ #include "scales.h"
+ #include "misc.h"
+ 
++#ifdef __SSE__
++#include <mmintrin.h>
++#include <xmmintrin.h>
++#endif
++
+ #define NEGINF -9999.f
+ static double stereo_threshholds[]={0.0, .5, 1.0, 1.5, 2.5, 4.5, 8.5, 16.5, 9e10};
+ 
++#ifdef __SSE__
++static void* align(void* x) {
++  long l=(long)x;
++  int r=l&0xf;
++  if (r)
++    return (void*)(l+16-(l&0xf));
++  else
++    return x;
++}
++#endif
++
+ vorbis_look_psy_global *_vp_global_look(vorbis_info *vi){
+   codec_setup_info *ci=vi->codec_setup;
+   vorbis_info_psy_global *gi=&ci->psy_g_param;
+@@ -530,17 +546,31 @@
+   
+ }
+ 
++#ifdef __SSE__
++__m128 tmp __attribute__((aligned(16)));
++#endif
++
+ static void bark_noise_hybridmp(int n,const long *b,
+                                 const float *f,
+                                 float *noise,
+                                 const float offset,
+                                 const int fixed){
+   
++#ifdef __SSE__
++  /* we need to be 16-bytes aligned for SSE */
++  /* so we can be at most 16-4=12 bytes off, allocate three more floats */
++  float *N=align(alloca((n+4)*sizeof(*N)));
++  float *X=align(alloca((n+4)*sizeof(*N)));
++  float *XX=align(alloca((n+4)*sizeof(*N)));
++  float *Y=align(alloca((n+4)*sizeof(*N)));
++  float *XY=align(alloca((n+4)*sizeof(*N)));
++#else
+   float *N=alloca((n+1)*sizeof(*N));
+   float *X=alloca((n+1)*sizeof(*N));
+   float *XX=alloca((n+1)*sizeof(*N));
+   float *Y=alloca((n+1)*sizeof(*N));
+   float *XY=alloca((n+1)*sizeof(*N));
++#endif
+ 
+   float tN, tX, tXX, tY, tXY;
+   float fi;
+@@ -548,9 +578,131 @@
+ 
+   int lo, hi;
+   float R, A, B, D;
++#ifdef __SSE__
++  register __m128 offset4=_mm_set_ps1(offset);
++#endif
+   
+   tN = tX = tXX = tY = tXY = 0.f;
++#ifdef __SSE__
++  for (i=0, fi=0.f; i+4<n; i+=4, fi+=4.f) {
++    static float __attribute__((aligned(16))) c1111[4]={1.f,1.f,1.f,1.f};
++    static float __attribute__((aligned(16))) c0123[4]={0.f,1.f,2.f,3.f};
++    register __m128 xxxx = _mm_add_ps(_mm_set_ps1(fi),*(__m128*)c0123);
++    /* turns out that f may be unaligned and there is nothing I can do
++      * about it */
++    register __m128 yyyy=_mm_max_ps(*(__m128*)c1111,_mm_add_ps(_mm_loadu_ps((float*)f+i),offset4));
++    register __m128 wwww=_mm_mul_ps(yyyy,yyyy);
++    /* now it gets hairy */
++
++    float* a=(float*)&tmp;
++
++#ifdef OLD_AND_WORKING
++    tmp=wwww;
++    N[i]=tN;
++    N[i+1]=(tN+=a[0]);
++    N[i+2]=(tN+=a[1]);
++    N[i+3]=(tN+=a[2]);
++    tN+=a[3];
++
++    tmp=_mm_mul_ps(wwww,xxxx);
++    X[i]=tX;
++    X[i+1]=(tX+=a[0]);
++    X[i+2]=(tX+=a[1]);
++    X[i+3]=(tX+=a[2]);
++    tX+=a[3];
++
++    tmp=_mm_mul_ps(tmp,xxxx);
++    XX[i]=tXX;
++    XX[i+1]=(tXX+=a[0]);
++    XX[i+2]=(tXX+=a[1]);
++    XX[i+3]=(tXX+=a[2]);
++    tXX+=a[3];
++
++    tmp=_mm_mul_ps(wwww,yyyy);
++    Y[i]=tY;
++    Y[i+1]=(tY+=a[0]);
++    Y[i+2]=(tY+=a[1]);
++    Y[i+3]=(tY+=a[2]);
++    tY+=a[3];
++
++    tmp=_mm_mul_ps(tmp,xxxx);
++    XY[i]=tXY;
++    XY[i+1]=(tXY+=a[0]);
++    XY[i+2]=(tXY+=a[1]);
++    XY[i+3]=(tXY+=a[2]);
++    tXY+=a[3];
++#else
++    {
++      register __m128 o=_mm_set_ss(tN);
++      register __m128 x=wwww;
++      o=_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),x);
++      o=_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),_mm_shuffle_ps(x,x,_MM_SHUFFLE(0,0,0,1)));
++      o=_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),_mm_shuffle_ps(x,x,_MM_SHUFFLE(0,0,0,2)));
++      _mm_storeu_ps(N+i,_mm_shuffle_ps(o,o,_MM_SHUFFLE(0,1,2,3)));
++      _mm_store_ss(&tN,_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),_mm_shuffle_ps(x,x,_MM_SHUFFLE(0,0,0,3))));
++
++      o=_mm_set_ss(tX); x=_mm_mul_ps(wwww,xxxx);
++      o=_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),x);
++      o=_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),_mm_shuffle_ps(x,x,_MM_SHUFFLE(0,0,0,1)));
++      o=_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),_mm_shuffle_ps(x,x,_MM_SHUFFLE(0,0,0,2)));
++      _mm_storeu_ps(X+i,_mm_shuffle_ps(o,o,_MM_SHUFFLE(0,1,2,3)));
++      _mm_store_ss(&tX,_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),_mm_shuffle_ps(x,x,_MM_SHUFFLE(0,0,0,3))));
++
++      o=_mm_set_ss(tXX); x=_mm_mul_ps(x,xxxx);
++      o=_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),x);
++      o=_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),_mm_shuffle_ps(x,x,_MM_SHUFFLE(0,0,0,1)));
++      o=_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),_mm_shuffle_ps(x,x,_MM_SHUFFLE(0,0,0,2)));
++      _mm_storeu_ps(XX+i,_mm_shuffle_ps(o,o,_MM_SHUFFLE(0,1,2,3)));
++      _mm_store_ss(&tXX,_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),_mm_shuffle_ps(x,x,_MM_SHUFFLE(0,0,0,3))));
++
++      o=_mm_set_ss(tY); x=_mm_mul_ps(wwww,yyyy);
++      o=_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),x);
++      o=_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),_mm_shuffle_ps(x,x,_MM_SHUFFLE(0,0,0,1)));
++      o=_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),_mm_shuffle_ps(x,x,_MM_SHUFFLE(0,0,0,2)));
++      _mm_storeu_ps(Y+i,_mm_shuffle_ps(o,o,_MM_SHUFFLE(0,1,2,3)));
++      _mm_store_ss(&tY,_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),_mm_shuffle_ps(x,x,_MM_SHUFFLE(0,0,0,3))));
++
++      o=_mm_set_ss(tXY); x=_mm_mul_ps(x,xxxx);
++      o=_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),x);
++      o=_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),_mm_shuffle_ps(x,x,_MM_SHUFFLE(0,0,0,1)));
++      o=_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),_mm_shuffle_ps(x,x,_MM_SHUFFLE(0,0,0,2)));
++      _mm_storeu_ps(XY+i,_mm_shuffle_ps(o,o,_MM_SHUFFLE(0,1,2,3)));
++      _mm_store_ss(&tXY,_mm_add_ss(_mm_shuffle_ps(o,o,_MM_SHUFFLE(2,1,0,0)),_mm_shuffle_ps(x,x,_MM_SHUFFLE(0,0,0,3))));
++
++    }
++#endif
++
++#if 0
++    N[i] = tN;
++    N[i+1] = tN+w[0];
++    N[i+2] = tN+w[0]+w[1];
++    N[i+3] = tN+w[0]+w[1]+w[2];
++
++    X[i] = tX;
++    X[i+1] = tX+w[0]*x[0];
++    X[i+2] = tX+w[0]*x[0]+w[1]*x[1];
++    X[i+3] = tX+w[0]*x[0]+w[1]*x[1]+w[2]*x[2];
++
++    Y[i] = tY;
++    Y[i+1] = tY+w[0]*y[0];
++    Y[i+2] = tY+w[0]*y[0]+w[1]*y[1];
++    Y[i+3] = tY+w[0]*y[0]+w[1]*y[1]+w[2]*y[2];
++
++    XX[i] = tXX;
++    XX[i+1] = tXX+w[0]*x[0]*x[0];
++    XX[i+2] = tXX+w[0]*x[0]*x[0]+w[1]*x[1]*x[1];
++    XX[i+3] = tXX+w[0]*x[0]*x[0]+w[1]*x[1]*x[1]+w[2]*x[2]*x[2];
++
++    XY[i] = tXY;
++    XY[i+1] = tXY+w[0]*x[0]*y[0];
++    XY[i+2] = tXY+w[0]*x[0]*y[0]+w[1]*x[1]*y[1];
++    XY[i+3] = tXY+w[0]*x[0]*y[0]+w[1]*x[1]*y[1]+w[2]*x[2]*y[2];
++#endif
++  }
++  for (; i < n; i++, fi += 1.f) {
++#else
+   for (i = 0, fi = 0.f; i < n; i++, fi += 1.f) {
++#endif
+     float w, x, y;
+     
+     x = fi;
+@@ -597,6 +749,7 @@
+   }
+   
+   for ( ; hi < n; i++, fi += 1.f) {
++    /* TODO hotspot */
+     
+     lo = b[i] >> 16;
+     hi = b[i] & 0xffff;
+@@ -644,6 +797,7 @@
+     if (R > 0.f && R - offset < noise[i]) noise[i] = R - offset;
+   }
+   for ( ; hi < n; i++, fi += 1.f) {
++    /* TODO hotspot */
+     
+     hi = i + fixed / 2;
+     lo = hi - fixed;
+@@ -744,6 +898,7 @@
+  
+   if(sliding_lowpass>n)sliding_lowpass=n;
+   
++  /* TODO hotspot */
+   for(i=0;i<sliding_lowpass;i++){
+     residue[i]=
+       mdct[i]*FLOOR1_fromdB_INV_LOOKUP[codedflr[i]];
+@@ -792,6 +947,7 @@
+   }
+ #endif
+ 
++  /* TODO hotspot */
+   for(i=0;i<n;i++){
+     int dB=logmask[i]+.5;
+     if(dB>=NOISE_COMPAND_LEVELS)dB=NOISE_COMPAND_LEVELS-1;
+@@ -816,7 +972,28 @@
+      specified att) */
+   if(att<p->vi->ath_maxatt)att=p->vi->ath_maxatt;
+   
++  /* TODO hotspot */
++#ifdef __SSE__
++/* this optimization does not make a difference for me */
++  {
++    register __m128 att4=_mm_set_ps1(att);
++#if 0
++    for (i=0;i+4<n && (long)(p->ath+i)&0xf;++i)
++      logmask[i]=p->ath[i]+att;
++#endif
++    for (i=0;i+16<n;i+=16) {
++      _mm_storeu_ps(logmask+i,_mm_add_ps(_mm_loadu_ps(p->ath+i),att4));
++      _mm_storeu_ps(logmask+i+4,_mm_add_ps(_mm_loadu_ps(p->ath+i+4),att4));
++      _mm_storeu_ps(logmask+i+8,_mm_add_ps(_mm_loadu_ps(p->ath+i+8),att4));
++      _mm_storeu_ps(logmask+i+12,_mm_add_ps(_mm_loadu_ps(p->ath+i+12),att4));
++    }
++    for (;i+4<n;i+=4)
++      _mm_storeu_ps(logmask+i,_mm_add_ps(_mm_loadu_ps(p->ath+i),att4));
++  }
++  for(;i<n;i++)
++#else
+   for(i=0;i<n;i++)
++#endif
+     logmask[i]=p->ath[i]+att;
+ 
+   /* tone masking */
+@@ -833,7 +1010,21 @@
+   int i,n=p->n;
+   float toneatt=p->vi->tone_masteratt[offset_select];
+   
++  /* TODO hotspot */
++#ifdef __SSE__
++  register float* no=p->noiseoffset[offset_select];
++  register __m128 toneatt4=_mm_set_ps1(toneatt);
++  register __m128 noisemax4=_mm_set_ps1(p->vi->noisemaxsupp);
++  for(i=0;i+4<n;i+=4) {
++    register __m128 x=_mm_add_ps(_mm_loadu_ps(noise+i),_mm_loadu_ps(no+i));
++    x=_mm_min_ps(x,noisemax4);
++    x=_mm_max_ps(x,_mm_add_ps(_mm_loadu_ps(tone+i),toneatt4));
++    _mm_storeu_ps(logmask+i,x);
++  }
++  for(;i<n;i++){
++#else
+   for(i=0;i<n;i++){
++#endif
+     float val= noise[i]+p->noiseoffset[offset_select][i];
+     if(val>p->vi->noisemaxsupp)val=p->vi->noisemaxsupp;
+     logmask[i]=max(val,tone[i]+toneatt);
+@@ -883,6 +1074,7 @@
+   -0.159093, -0.175146, -0.192286, -0.210490, 
+   -0.229718, -0.249913, -0.271001, -0.292893};
+ 
++/* minor hotspot */
+ static void precomputed_couple_point(float premag,
+ 				     int floorA,int floorB,
+ 				     float *mag, float *ang){
+@@ -949,6 +1141,7 @@
+ }
+ 
+ /* this is for per-channel noise normalization */
++/* TODO hotspot */
+ static int apsort(const void *a, const void *b){
+   if(fabs(**(float **)a)>fabs(**(float **)b))return -1;
+   return 1;
+@@ -972,6 +1165,8 @@
+       for(j=0;j<n;j+=partition){
+ 	for(k=0;k<partition;k++)work[k]=mags[i]+k+j;
+ 	qsort(work,partition,sizeof(*work),apsort);
++	/* TODO hotspot */
++	/* the obvious MMX version is not faster */
+ 	for(k=0;k<partition;k++)ret[i][k+j]=work[k]-mags[i];
+       }
+     }
+@@ -1008,8 +1203,26 @@
+   if(start>n)start=n;
+ 
+   if(vi->normal_channel_p){
++    /* TODO hotspot */
++#ifdef __SSE__DISABLED
++    /* this optimization does not make a difference */
++    /* also, it appears to distort the results :( */
++    register __m128 half4=_mm_set1_ps(0.5f);
++    for(;j+4<start;j+=4) {
++      register __m128 x=_mm_add_ps(_mm_loadu_ps(in+j),half4);
++      register __m64 y=_mm_cvtps_pi32(x);
++      register __m64 z=_mm_cvtps_pi32(_mm_shuffle_ps(x,x,0x1b));
++      x=_mm_cvtpi32_ps(x,z);
++      x=_mm_cvtpi32_ps(_mm_shuffle_ps(x,x,0x1b),y);
++      _mm_storeu_ps(out+j,x);
++    }
++    _mm_empty();
++    for(;j<start;j++)
++      out[j]=rint(in[j]);
++#else
+     for(;j<start;j++)
+       out[j]=rint(in[j]);
++#endif
+     
+     for(;j+partition<=n;j+=partition){
+       float acc=0.;
+@@ -1018,6 +1231,7 @@
+       for(i=j;i<j+partition;i++)
+ 	acc+=in[i]*in[i];
+       
++      /* TODO hotspot */
+       for(i=0;i<partition;i++){
+ 	k=sortedindex[i+j-start];
+ 	
+@@ -1032,6 +1246,7 @@
+ 	}
+       }
+       
++      /* TODO hotspot */
+       for(;i<partition;i++){
+ 	k=sortedindex[i+j-start];
+ 	out[k]=0.;
+@@ -1093,6 +1308,7 @@
+       for(j=0;j<p->n;j+=partition){
+ 	float acc=0.f;
+ 
++	/* TODO hotspot */
+ 	for(k=0;k<partition;k++){
+ 	  int l=k+j;
+ 
+@@ -1116,6 +1332,7 @@
+ 	}
+ 	
+ 	if(p->vi->normal_point_p){
++	  /* TODO minor hotspot */
+ 	  for(k=0;k<partition && acc>=p->vi->normal_thresh;k++){
+ 	    int l=mag_sort[i][j+k];
+ 	    if(l<sliding_lowpass && l>=pointlimit && rint(qM[l])==0.f){
+diff -ur libvorbis-1.0/lib/vorbisfile.c libvorbis-1.0-simd/lib/vorbisfile.c
+--- libvorbis-1.0/lib/vorbisfile.c	2002-07-06 06:20:03.000000000 +0200
++++ libvorbis-1.0-simd/lib/vorbisfile.c	2003-04-26 21:55:49.000000000 +0200
+@@ -21,12 +21,19 @@
+ #include <string.h>
+ #include <math.h>
+ 
++#include <assert.h>
++
+ #include "vorbis/codec.h"
+ #include "vorbis/vorbisfile.h"
+ 
+ #include "os.h"
+ #include "misc.h"
+ 
++#ifdef __SSE__
++#include <xmmintrin.h>
++#include <mmintrin.h>
++#endif
++
+ /* A 'chained bitstream' is a Vorbis bitstream that contains more than
+    one logical bitstream arranged end to end (the only form of Ogg
+    multiplexing allowed in a Vorbis bitstream; grouping [parallel
+@@ -1500,7 +1507,72 @@
+ 	
+ 	if(host_endian==bigendianp){
+ 	  if(sgned){
+-	    
++#ifdef __SSE__
++	    /* Oh no!  On my box, exactly one of pcm[0][j] and pcm[1][j]
++	     * is always misaligned!  SSE required 16-byte alignment.
++	     * Also, the pcm[][] array layout is bad for vectorizing,
++	     * but we can fix that with mmx unpack magic for the common
++	     * case where channels==2. */
++	    if (channels==2 && samples>8) {
++	      register __m128 scale=_mm_set1_ps(32768.f);
++	      for (j=0; j+8<samples; j+=8) {
++		register __m128 x=_mm_loadu_ps(pcm[0]+j);
++		register __m128 y=_mm_loadu_ps(pcm[1]+j);
++
++		*(__m64*)buffer=_mm_cvtps_pi16(_mm_mul_ps(_mm_unpacklo_ps(x,y),scale));
++		*(__m64*)(buffer+8)=_mm_cvtps_pi16(_mm_mul_ps(_mm_unpackhi_ps(x,y),scale));
++
++		x=_mm_loadu_ps(pcm[0]+j+4);
++		y=_mm_loadu_ps(pcm[1]+j+4);
++
++		*(__m64*)(buffer+16)=_mm_cvtps_pi16(_mm_mul_ps(_mm_unpacklo_ps(x,y),scale));
++		*(__m64*)(buffer+24)=_mm_cvtps_pi16(_mm_mul_ps(_mm_unpackhi_ps(x,y),scale));
++
++		buffer+=32;
++	      }
++	      for (; j+4<samples; j+=4) {
++		register __m128 x=_mm_loadu_ps(pcm[0]+j);
++		register __m128 y=_mm_loadu_ps(pcm[1]+j);
++
++		*(__m64*)buffer=_mm_cvtps_pi16(_mm_mul_ps(_mm_unpacklo_ps(x,y),scale));
++		*(__m64*)(buffer+8)=_mm_cvtps_pi16(_mm_mul_ps(_mm_unpackhi_ps(x,y),scale));
++
++		buffer+=16;
++	      }
++	      _mm_empty();
++
++	      for (; j<samples; ++j) {
++		val=vorbis_ftoi(pcm[0][j]*32768.f);
++		if(val>32767)val=32767;
++		else if(val<-32768)val=-32768;
++		*(short*)buffer=val;
++		val=vorbis_ftoi(pcm[1][j]*32768.f);
++		if(val>32767)val=32767;
++		else if(val<-32768)val=-32768;
++		*(short*)(buffer+2)=val;
++		buffer+=4;
++	      }
++	    } else {
++#elif defined(simd_3dn)
++	  if (channels==2) {
++	    static float scale[2]={32768.f,32768.f};
++	    for(j=0;j<samples;j+=2) {
++	      asm("movq %0,%%mm0\n\t"	/* (pcm[0][j],pcm[0][j+1]) */
++		  "pfmul %1,%%mm0\n\t"
++		  "pf2id %%mm0,%%mm0\n\t"
++		  "packssdw %%mm0,%%mm0\n\t"
++		  "movq %3,%%mm1\n\t"	/* (pcm[1][j],pcm[1][j+1]) */
++		  "pfmul %1,%%mm1\n\t"
++		  "pf2id %%mm1,%%mm1\n\t"
++		  "packssdw %%mm1,%%mm1\n\t"
++		  "punpckhwd %%mm1,%%mm0\n\t"
++		  "movq %%mm0,(%2)\n\t"
++		  : : "m" (pcm[0][j]), "m" (scale), "r" (buffer), "m" (pcm[1][j]) );
++	      buffer += 8;
++	    }
++	    asm volatile("femms\n\t");
++	  } else {
++#endif
+ 	    vorbis_fpu_setround(&fpu);
+ 	    for(i=0;i<channels;i++) { /* It's faster in this order */
+ 	      float *src=pcm[i];
+@@ -1514,6 +1586,9 @@
+ 	      }
+ 	    }
+ 	    vorbis_fpu_restore(fpu);
++#if defined(__SSE__) || defined(simd_3dn)
++	  }
++#endif
+ 	    
+ 	  }else{
+ 	    
+@@ -1548,6 +1623,7 @@
+ 	  
+ 	}else{
+ 	  int val;
++	  write(1,"a",1);
+ 	  vorbis_fpu_setround(&fpu);
+ 	  for(j=0;j<samples;j++)
+ 	    for(i=0;i<channels;i++){
+diff -ur libvorbis-1.0/lib/window.c libvorbis-1.0-simd/lib/window.c
+--- libvorbis-1.0/lib/window.c	2002-03-23 04:17:34.000000000 +0100
++++ libvorbis-1.0-simd/lib/window.c	2003-02-08 02:53:04.000000000 +0100
+@@ -17,9 +17,14 @@
+ 
+ #include <stdlib.h>
+ #include <math.h>
++#include <stdio.h>
+ #include "os.h"
+ #include "misc.h"
+ 
++#ifdef __SSE__
++#include <xmmintrin.h>
++#endif
++
+ float *_vorbis_window(int type, int left){
+   float *ret=_ogg_calloc(left,sizeof(*ret));
+   int i;
+@@ -67,8 +72,68 @@
+     for(i=0;i<leftbegin;i++)
+       d[i]=0.f;
+     
++#ifdef __SSE__
++//    printf("%p %p\n",d+i,window[lW]);
++    /* alignment issues.  Again.  Who saw that one coming? :-/ */
++    p=0;
++    if (leftend-i>8) { /* make it worth our while */
++      unsigned long _d=(long)(d+i)&15;
++      unsigned long _win=(long)(window[lW])&15;
++      register __m128* D,* WIN;
++      if (_d) {
++	d[i]*=window[lW][p];
++	d[i+1]*=window[lW][p+1];
++	i+=2; p+=2;
++      }
++      D=(__m128*)(d+i);
++      WIN=(__m128*)(window[lW]+p);
++      if (_d ^ _win) {
++	/* one is properly aligned, the other is not */
++	register __m128 a;
++	WIN=(__m128*)((float*)WIN-2);
++	a=WIN[0];
++	for(;i+7<leftend;i+=8,p+=8) {
++	  /* now we need the upper most floats from a and the lower
++	  * most floats from b */
++	  register __m128 b=WIN[1];
++	  *D=_mm_mul_ps(_mm_movehl_ps(_mm_shuffle_ps(b,b,_MM_SHUFFLE(1,0,3,2)),a),*D);
++	  a=WIN[2];
++	  D[1]=_mm_mul_ps(_mm_movehl_ps(_mm_shuffle_ps(a,a,_MM_SHUFFLE(1,0,3,2)),b),D[1]);
++	  D+=2; WIN+=2;
++	}
++      } else {
++	for(;i+3<leftend;i+=4,p+=4) {
++	  *D=_mm_mul_ps(*D,*WIN);
++	  ++D; ++WIN;
++//	  d[i]*=window[lW][p];
++	}
++      }
++    }
++    for(;i<leftend;i++,p++)
++      d[i]*=window[lW][p];
++#elif defined(simd_3dn)
++    if ((i&1) || ((leftend-i)&1)) { /* d[i] is unaligned */
++      for(p=0;i<leftend;i++,p++)
++	d[i]*=window[lW][p];
++    } else {
++      asm("orl %0,%0\n\t"
++	  "jbe 2f\n\t"
++	  "1: movq (%1),%%mm0\n\t"
++	  "pfmul (%2),%%mm0\n\t"
++	  "addl $8,%2\n\t"
++	  "movq %%mm0,(%1)\n\t"
++	  "addl $8,%1\n\t"
++	  "decl %0\n\t"
++	  "jz 2f\n\t"
++	  "jmp 1b\n\t"
++	  "2: femms\n\t"
++	  : : "r" ((leftend-i)/2), "r" (&d[i]), "r" (&window[lW][0]) : "memory" );
++      i=leftend;
++    }
++#else
+     for(p=0;i<leftend;i++,p++)
+       d[i]*=window[lW][p];
++#endif
+     
+     for(i=rightbegin,p=rn/2-1;i<rightend;i++,p--)
+       d[i]*=window[nW][p];
+
diff --git a/media-libs/libvorbis/libvorbis-1.0-r2.ebuild b/media-libs/libvorbis/libvorbis-1.0-r2.ebuild
index ce958dcd9958..8827670acd33 100644
--- a/media-libs/libvorbis/libvorbis-1.0-r2.ebuild
+++ b/media-libs/libvorbis/libvorbis-1.0-r2.ebuild
@@ -1,6 +1,6 @@
 # Copyright 1999-2003 Gentoo Technologies, Inc.
 # Distributed under the terms of the GNU General Public License v2
-# $Header: /var/cvsroot/gentoo-x86/media-libs/libvorbis/libvorbis-1.0-r2.ebuild,v 1.3 2003/07/18 21:55:45 tester Exp $
+# $Header: /var/cvsroot/gentoo-x86/media-libs/libvorbis/libvorbis-1.0-r2.ebuild,v 1.4 2003/08/05 09:58:07 jje Exp $
 
 inherit libtool eutils
 
@@ -19,7 +19,7 @@ src_unpack() {
 	unpack ${A}
 
 	cd ${S}
-	epatch ${FILESDIR}/${PF}-m4.patch || die "Patching failed"
+	epatch ${FILESDIR}/${PN}-m4.patch || die "Patching failed"
 	# Fix a gcc crash.  With the new atexit patch to gcc, it
 	# seems it do not handle -mno-ieee-fp too well.
 	cp configure configure.orig
diff --git a/media-libs/libvorbis/libvorbis-1.0-r1.ebuild b/media-libs/libvorbis/libvorbis-1.0-r3.ebuild
index 0c573a249fcc..397c1c4573da 100644
--- a/media-libs/libvorbis/libvorbis-1.0-r1.ebuild
+++ b/media-libs/libvorbis/libvorbis-1.0-r3.ebuild
@@ -1,25 +1,30 @@
 # Copyright 1999-2003 Gentoo Technologies, Inc.
 # Distributed under the terms of the GNU General Public License v2
-# $Header: /var/cvsroot/gentoo-x86/media-libs/libvorbis/libvorbis-1.0-r1.ebuild,v 1.7 2003/02/13 12:51:32 vapier Exp $
+# $Header: /var/cvsroot/gentoo-x86/media-libs/libvorbis/libvorbis-1.0-r3.ebuild,v 1.1 2003/08/05 09:58:07 jje Exp $
 
-inherit libtool
+inherit libtool eutils
 
 S=${WORKDIR}/${P}
 DESCRIPTION="the Ogg Vorbis sound file format library"
 SRC_URI="http://fatpipe.vorbis.com/files/1.0/unix/${P}.tar.gz"
 HOMEPAGE="http://www.xiph.org/ogg/vorbis/index.html"
 
+IUSE="sse"
 DEPEND=">=media-libs/libogg-1.0"
 
 SLOT="0"
 LICENSE="as-is"
-KEYWORDS="x86 ppc sparc alpha"
+KEYWORDS="~x86"
 
 src_unpack() {
 	unpack ${A}
 
 	cd ${S}
-	patch -p1 < ${FILESDIR}/${P}-m4.patch || die "Patching failed"
+	if [ `use x86` ] ; then
+		use sse && epatch ${FILESDIR}/${PN}-simd.patch
+	fi
+
+	epatch ${FILESDIR}/${PN}-m4.patch || die "Patching failed"
 	# Fix a gcc crash.  With the new atexit patch to gcc, it
 	# seems it do not handle -mno-ieee-fp too well.
 	cp configure configure.orig
@@ -30,7 +35,7 @@ src_unpack() {
 src_compile() {
 	elibtoolize
 
-	export CFLAGS="${CFLAGS/-march=*/}"
+	#export CFLAGS="${CFLAGS/-march=*/}"
 
 	./configure --prefix=/usr \
 		--host=${CHOST} || die
@@ -41,6 +46,9 @@ src_compile() {
 src_install () {
 	make DESTDIR=${D} install || die
 
+	dosym /usr/lib/libvorbisfile.so.3.0.0 /usr/lib/libvorbisfile.so.0 
+	dosym /usr/lib/libvorbisenc.so.2.0.0 /usr/lib/libvorbisenc.so.0
+
 	echo "Removing docs installed by make install"
 	rm -rf ${D}/usr/share/doc
 
@@ -60,3 +68,4 @@ pkg_postinst() {
 	einfo "recompilation is needed for these things."
 	einfo
 }
+
diff --git a/media-libs/libvorbis/libvorbis-1.0.ebuild b/media-libs/libvorbis/libvorbis-1.0.ebuild
deleted file mode 100644
index 0114be22ee2b..000000000000
--- a/media-libs/libvorbis/libvorbis-1.0.ebuild
+++ /dev/null
@@ -1,61 +0,0 @@
-# Copyright 1999-2003 Gentoo Technologies, Inc.
-# Distributed under the terms of the GNU General Public License v2
-# $Header: /var/cvsroot/gentoo-x86/media-libs/libvorbis/libvorbis-1.0.ebuild,v 1.6 2003/02/13 12:51:36 vapier Exp $
-
-inherit libtool
-
-S=${WORKDIR}/${P}
-DESCRIPTION="the Ogg Vorbis sound file format library"
-SRC_URI="http://fatpipe.vorbis.com/files/1.0/unix/${P}.tar.gz"
-HOMEPAGE="http://www.xiph.org/ogg/vorbis/index.html"
-
-DEPEND=">=media-libs/libogg-1.0"
-
-SLOT="0"
-LICENSE="as-is"
-KEYWORDS="x86 ppc sparc "
-
-src_unpack() {
-	unpack ${A}
-
-	cd ${S}
-	# Fix a gcc crash.  With the new atexit patch to gcc, it
-	# seems it do not handle -mno-ieee-fp too well.
-	cp configure configure.orig
-	sed -e "s:-mno-ieee-fp::g" \
-		configure.orig >configure
-}
-
-src_compile() {
-	elibtoolize
-
-	export CFLAGS="${CFLAGS/-march=*/}"
-
-	./configure --prefix=/usr \
-		--host=${CHOST} || die
-
-	emake || die
-}
-
-src_install () {
-	make DESTDIR=${D} install || die
-
-	echo "Removing docs installed by make install"
-	rm -rf ${D}/usr/share/doc
-
-	dodoc AUTHORS COPYING README todo.txt
-	docinto txt
-	dodoc doc/*.txt
-	dohtml -r doc
-}
-
-pkg_postinst() {
-	einfo 
-	einfo "Note the 1.0 version of libvorbis has been installed"
-	einfo "Applications that used pre-1.0 vorbis libraries will"
-	einfo "need to be recompiled for the new version."
-	einfo "Now that the vorbis folks have finalized the API"
-	einfo "this should be the last time for a while that"
-	einfo "recompilation is needed for these things."
-	einfo
-}