more fixes

Original commit message from CVS:
more fixes

12 files changed, 3451 insertions, 4 deletions

diff --git a/gst-libs/gst/Makefile.am b/gst-libs/gst/Makefile.am
index b179ac30..e822561d 100644
--- a/gst-libs/gst/Makefile.am
+++ b/gst-libs/gst/Makefile.am
@@ -1,5 +1,3 @@
-SUBDIRS = audio resample riff
-# riff getbits putbits idct bytestream control resample
+SUBDIRS = audio idct resample riff
 
-DIST_SUBDIRS = audio resample riff
-# riff getbits putbits idct bytestream control resample
+DIST_SUBDIRS = audio idct resample riff
diff --git a/gst-libs/gst/idct/README b/gst-libs/gst/idct/README
new file mode 100644
index 00000000..600f3da9
--- /dev/null
+++ b/gst-libs/gst/idct/README
@@ -0,0 +1,48 @@
+This archive contains a quick & dirty implementation of the IEEE Standard
+1180-1990 accuracy test for inverse DCT.  It is not guaranteed to be
+correct ... but if you find any bugs, please let me know (by email to
+tgl@cs.cmu.edu).
+
+The test harness consists of the C program ieeetest.c and shell script
+doieee.  For comparison purposes I have also supplied a copy of jrevdct.c,
+the inverse DCT routine from release 4 of the Independent JPEG Group's
+free JPEG software.  (jrevdct.c is slightly modified from the IJG release
+so that it will compile without the IJG include files.)  jrevdct.c passes
+the 1180 test --- or at least, this program thinks so.  jrevdct.out is
+the output from a test run.
+
+Note that numerical results may vary somewhat across machines.  This appears
+to be mostly due to differing results from the cosine function.
+
+
+INSTALLATION:
+  Check the Makefile, change CC and CFLAGS if needed.  Then say "make".
+If your C compiler is non-ANSI, you may need to change includes and/or
+function headers.
+
+  To test a different IDCT routine, link with that routine instead of
+jrevdct.o.  You will need to modify dct.h and/or ieeetest.c if your
+routine's calling convention is not in-place modification of an array
+of 64 "short"s.
+
+
+USAGE:
+  The standard test procedure is
+	doieee ieeetest >outputfile
+Expect it to take a while (almost 80 minutes on my old 68030 box).
+Each of the six passes will emit a row of 100 dots as it runs.
+
+You can grep the output for the word FAILS if you just want to know
+yea or nay.
+
+
+LEGAL MUMBO-JUMBO:
+  I hereby release the test harness to the public domain.
+			Thomas G. Lane, 22 Nov 1993
+
+IMPORTANT: jrevdct.c is NOT public domain, but is copyrighted free software
+(not the same thing at all).  It is subject to IJG's distribution terms, which
+primarily state that if you incorporate it into a program you must acknowledge
+IJG's contribution in your program documentation.  For more details and the
+complete IJG software, see the IJG FTP archive at ftp.uu.net, in directory
+/graphics/jpeg.
diff --git a/gst-libs/gst/idct/dct.h b/gst-libs/gst/idct/dct.h
new file mode 100644
index 00000000..fcb7de37
--- /dev/null
+++ b/gst-libs/gst/idct/dct.h
@@ -0,0 +1,34 @@
+/* define DCT types */
+
+#include "config.h"
+
+/*
+ * DCTSIZE      underlying (1d) transform size
+ * DCTSIZE2     DCTSIZE squared
+ */
+
+#define DCTSIZE      (8)
+#define DCTSIZE2     (DCTSIZE*DCTSIZE)
+
+#define EIGHT_BIT_SAMPLES	/* needed in jrevdct.c */
+
+typedef short DCTELEM;		/* must be at least 16 bits */
+
+typedef DCTELEM DCTBLOCK[DCTSIZE2];
+
+typedef long INT32;		/* must be at least 32 bits */
+
+extern void gst_idct_int_idct();
+
+extern void gst_idct_init_fast_int_idct (void);
+extern void gst_idct_fast_int_idct (short *block);
+
+#ifdef HAVE_LIBMMX
+extern void gst_idct_mmx_idct (short *block);
+extern void gst_idct_mmx32_idct (short *block);
+extern void gst_idct_sse_idct (short *block);
+#endif /* HAVE_LIBMMX */
+
+extern void gst_idct_init_float_idct(void);
+extern void gst_idct_float_idct (short *block);
+
diff --git a/gst-libs/gst/idct/doieee b/gst-libs/gst/idct/doieee
new file mode 100755
index 00000000..1d5ff877
--- /dev/null
+++ b/gst-libs/gst/idct/doieee
@@ -0,0 +1,15 @@
+# perform IEEE 1180 test series
+# Typical usage:
+#    doieee  >outfile
+# where progname is ieeetest or a variant
+
+for i in 1 2 3 4 5; 
+do
+time ./ieeetest $i -256 255 1 10000
+time ./ieeetest $i -5 5     1 10000
+time ./ieeetest $i -300 300 1 10000
+
+time ./ieeetest $i -256 255 -1 10000
+time ./ieeetest $i -5 5     -1 10000
+time ./ieeetest $i -300 300 -1 10000
+done
diff --git a/gst-libs/gst/idct/fastintidct.c b/gst-libs/gst/idct/fastintidct.c
new file mode 100644
index 00000000..3c9e9bb9
--- /dev/null
+++ b/gst-libs/gst/idct/fastintidct.c
@@ -0,0 +1,207 @@
+/* idct.c, inverse fast discrete cosine transform                           */
+
+/* Copyright (C) 1996, MPEG Software Simulation Group. All Rights Reserved. */
+
+/*
+ * Disclaimer of Warranty
+ *
+ * These software programs are available to the user without any license fee or
+ * royalty on an "as is" basis.  The MPEG Software Simulation Group disclaims
+ * any and all warranties, whether express, implied, or statuary, including any
+ * implied warranties or merchantability or of fitness for a particular
+ * purpose.  In no event shall the copyright-holder be liable for any
+ * incidental, punitive, or consequential damages of any kind whatsoever
+ * arising from the use of these programs.
+ *
+ * This disclaimer of warranty extends to the user of these programs and user's
+ * customers, employees, agents, transferees, successors, and assigns.
+ *
+ * The MPEG Software Simulation Group does not represent or warrant that the
+ * programs furnished hereunder are free of infringement of any third-party
+ * patents.
+ *
+ * Commercial implementations of MPEG-1 and MPEG-2 video, including shareware,
+ * are subject to royalty fees to patent holders.  Many of these patents are
+ * general enough such that they are unavoidable regardless of implementation
+ * design.
+ *
+ */
+
+/**********************************************************/
+/* inverse two dimensional DCT, Chen-Wang algorithm       */
+/* (cf. IEEE ASSP-32, pp. 803-816, Aug. 1984)             */
+/* 32-bit integer arithmetic (8 bit coefficients)         */
+/* 11 mults, 29 adds per DCT                              */
+/*                                      sE, 18.8.91       */
+/**********************************************************/
+/* coefficients extended to 12 bit for IEEE1180-1990      */
+/* compliance                           sE,  2.1.94       */
+/**********************************************************/
+
+/* this code assumes >> to be a two's-complement arithmetic */
+/* right shift: (-2)>>1 == -1 , (-3)>>1 == -2               */
+
+#define W1 2841 /* 2048*sqrt(2)*cos(1*pi/16) */
+#define W2 2676 /* 2048*sqrt(2)*cos(2*pi/16) */
+#define W3 2408 /* 2048*sqrt(2)*cos(3*pi/16) */
+#define W5 1609 /* 2048*sqrt(2)*cos(5*pi/16) */
+#define W6 1108 /* 2048*sqrt(2)*cos(6*pi/16) */
+#define W7 565  /* 2048*sqrt(2)*cos(7*pi/16) */
+
+#include "dct.h"
+
+/* private data */
+static short iclip[1024]; /* clipping table */
+static short *iclp;
+
+/* private prototypes */
+static void idctrow (short *blk);
+static void idctcol (short *blk);
+
+/* row (horizontal) IDCT
+ *
+ *           7                       pi         1
+ * dst[k] = sum c[l] * src[l] * cos( -- * ( k + - ) * l )
+ *          l=0                      8          2
+ *
+ * where: c[0]    = 128
+ *        c[1..7] = 128*sqrt(2)
+ */
+
+static void idctrow(blk)
+short *blk;
+{
+  int x0, x1, x2, x3, x4, x5, x6, x7, x8;
+
+  /* shortcut */
+  if (!((x1 = blk[4]<<11) | (x2 = blk[6]) | (x3 = blk[2]) |
+        (x4 = blk[1]) | (x5 = blk[7]) | (x6 = blk[5]) | (x7 = blk[3])))
+  {
+    blk[0]=blk[1]=blk[2]=blk[3]=blk[4]=blk[5]=blk[6]=blk[7]=blk[0]<<3;
+    return;
+  }
+
+  x0 = (blk[0]<<11) + 128; /* for proper rounding in the fourth stage */
+
+  /* first stage */
+  x8 = W7*(x4+x5);
+  x4 = x8 + (W1-W7)*x4;
+  x5 = x8 - (W1+W7)*x5;
+  x8 = W3*(x6+x7);
+  x6 = x8 - (W3-W5)*x6;
+  x7 = x8 - (W3+W5)*x7;
+  
+  /* second stage */
+  x8 = x0 + x1;
+  x0 -= x1;
+  x1 = W6*(x3+x2);
+  x2 = x1 - (W2+W6)*x2;
+  x3 = x1 + (W2-W6)*x3;
+  x1 = x4 + x6;
+  x4 -= x6;
+  x6 = x5 + x7;
+  x5 -= x7;
+  
+  /* third stage */
+  x7 = x8 + x3;
+  x8 -= x3;
+  x3 = x0 + x2;
+  x0 -= x2;
+  x2 = (181*(x4+x5)+128)>>8;
+  x4 = (181*(x4-x5)+128)>>8;
+  
+  /* fourth stage */
+  blk[0] = (x7+x1)>>8;
+  blk[1] = (x3+x2)>>8;
+  blk[2] = (x0+x4)>>8;
+  blk[3] = (x8+x6)>>8;
+  blk[4] = (x8-x6)>>8;
+  blk[5] = (x0-x4)>>8;
+  blk[6] = (x3-x2)>>8;
+  blk[7] = (x7-x1)>>8;
+}
+
+/* column (vertical) IDCT
+ *
+ *             7                         pi         1
+ * dst[8*k] = sum c[l] * src[8*l] * cos( -- * ( k + - ) * l )
+ *            l=0                        8          2
+ *
+ * where: c[0]    = 1/1024
+ *        c[1..7] = (1/1024)*sqrt(2)
+ */
+static void idctcol(blk)
+short *blk;
+{
+  int x0, x1, x2, x3, x4, x5, x6, x7, x8;
+
+  /* shortcut */
+  if (!((x1 = (blk[8*4]<<8)) | (x2 = blk[8*6]) | (x3 = blk[8*2]) |
+        (x4 = blk[8*1]) | (x5 = blk[8*7]) | (x6 = blk[8*5]) | (x7 = blk[8*3])))
+  {
+    blk[8*0]=blk[8*1]=blk[8*2]=blk[8*3]=blk[8*4]=blk[8*5]=blk[8*6]=blk[8*7]=
+      iclp[(blk[8*0]+32)>>6];
+    return;
+  }
+
+  x0 = (blk[8*0]<<8) + 8192;
+
+  /* first stage */
+  x8 = W7*(x4+x5) + 4;
+  x4 = (x8+(W1-W7)*x4)>>3;
+  x5 = (x8-(W1+W7)*x5)>>3;
+  x8 = W3*(x6+x7) + 4;
+  x6 = (x8-(W3-W5)*x6)>>3;
+  x7 = (x8-(W3+W5)*x7)>>3;
+  
+  /* second stage */
+  x8 = x0 + x1;
+  x0 -= x1;
+  x1 = W6*(x3+x2) + 4;
+  x2 = (x1-(W2+W6)*x2)>>3;
+  x3 = (x1+(W2-W6)*x3)>>3;
+  x1 = x4 + x6;
+  x4 -= x6;
+  x6 = x5 + x7;
+  x5 -= x7;
+  
+  /* third stage */
+  x7 = x8 + x3;
+  x8 -= x3;
+  x3 = x0 + x2;
+  x0 -= x2;
+  x2 = (181*(x4+x5)+128)>>8;
+  x4 = (181*(x4-x5)+128)>>8;
+  
+  /* fourth stage */
+  blk[8*0] = iclp[(x7+x1)>>14];
+  blk[8*1] = iclp[(x3+x2)>>14];
+  blk[8*2] = iclp[(x0+x4)>>14];
+  blk[8*3] = iclp[(x8+x6)>>14];
+  blk[8*4] = iclp[(x8-x6)>>14];
+  blk[8*5] = iclp[(x0-x4)>>14];
+  blk[8*6] = iclp[(x3-x2)>>14];
+  blk[8*7] = iclp[(x7-x1)>>14];
+}
+
+/* two dimensional inverse discrete cosine transform */
+void gst_idct_fast_int_idct(block)
+short *block;
+{
+  int i;
+
+  for (i=0; i<8; i++)
+    idctrow(block+8*i);
+
+  for (i=0; i<8; i++)
+    idctcol(block+i);
+}
+
+void gst_idct_init_fast_int_idct()
+{
+  int i;
+
+  iclp = iclip+512;
+  for (i= -512; i<512; i++)
+    iclp[i] = (i<-256) ? -256 : ((i>255) ? 255 : i);
+}
diff --git a/gst-libs/gst/idct/floatidct.c b/gst-libs/gst/idct/floatidct.c
new file mode 100644
index 00000000..520c3913
--- /dev/null
+++ b/gst-libs/gst/idct/floatidct.c
@@ -0,0 +1,102 @@
+/* Reference_IDCT.c, Inverse Discrete Fourier Transform, double precision          */
+
+/* Copyright (C) 1996, MPEG Software Simulation Group. All Rights Reserved. */
+
+/*
+ * Disclaimer of Warranty
+ *
+ * These software programs are available to the user without any license fee or
+ * royalty on an "as is" basis.  The MPEG Software Simulation Group disclaims
+ * any and all warranties, whether express, implied, or statuary, including any
+ * implied warranties or merchantability or of fitness for a particular
+ * purpose.  In no event shall the copyright-holder be liable for any
+ * incidental, punitive, or consequential damages of any kind whatsoever
+ * arising from the use of these programs.
+ *
+ * This disclaimer of warranty extends to the user of these programs and user's
+ * customers, employees, agents, transferees, successors, and assigns.
+ *
+ * The MPEG Software Simulation Group does not represent or warrant that the
+ * programs furnished hereunder are free of infringement of any third-party
+ * patents.
+ *
+ * Commercial implementations of MPEG-1 and MPEG-2 video, including shareware,
+ * are subject to royalty fees to patent holders.  Many of these patents are
+ * general enough such that they are unavoidable regardless of implementation
+ * design.
+ *
+ */
+
+/*  Perform IEEE 1180 reference (64-bit floating point, separable 8x1
+ *  direct matrix multiply) Inverse Discrete Cosine Transform
+*/
+
+
+/* Here we use math.h to generate constants.  Compiler results may
+   vary a little */
+
+#include <math.h>
+
+#ifndef PI
+# ifdef M_PI
+#  define PI M_PI
+# else
+#  define PI 3.14159265358979323846
+# endif
+#endif
+
+/* private data */
+
+/* cosine transform matrix for 8x1 IDCT */
+static double gst_idct_float_c[8][8];
+
+/* initialize DCT coefficient matrix */
+
+void gst_idct_init_float_idct()
+{
+  int freq, time;
+  double scale;
+
+  for (freq=0; freq < 8; freq++)
+  {
+    scale = (freq == 0) ? sqrt(0.125) : 0.5;
+    for (time=0; time<8; time++)
+      gst_idct_float_c[freq][time] = scale*cos((PI/8.0)*freq*(time + 0.5));
+  }
+}
+
+/* perform IDCT matrix multiply for 8x8 coefficient block */
+
+void gst_idct_float_idct(block)
+short *block;
+{
+  int i, j, k, v;
+  double partial_product;
+  double tmp[64];
+
+  for (i=0; i<8; i++)
+    for (j=0; j<8; j++)
+    {
+      partial_product = 0.0;
+
+      for (k=0; k<8; k++)
+        partial_product+= gst_idct_float_c[k][j]*block[8*i+k];
+
+      tmp[8*i+j] = partial_product;
+    }
+
+  /* Transpose operation is integrated into address mapping by switching 
+     loop order of i and j */
+
+  for (j=0; j<8; j++)
+    for (i=0; i<8; i++)
+    {
+      partial_product = 0.0;
+
+      for (k=0; k<8; k++)
+        partial_product+= gst_idct_float_c[k][i]*tmp[8*k+j];
+
+      v = (int) floor(partial_product+0.5);
+      block[8*i+j] = (v<-256) ? -256 : ((v>255) ? 255 : v);
+    }
+}
diff --git a/gst-libs/gst/idct/idct.h b/gst-libs/gst/idct/idct.h
new file mode 100644
index 00000000..b5654737
--- /dev/null
+++ b/gst-libs/gst/idct/idct.h
@@ -0,0 +1,54 @@
+/* Gnome-Streamer
+ * Copyright (C) <1999> Erik Walthinsen <omega@cse.ogi.edu>
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Library General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Library General Public License for more details.
+ *
+ * You should have received a copy of the GNU Library General Public
+ * License along with this library; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 02111-1307, USA.
+ */
+
+
+#ifndef __GST_IDCT_H__
+#define __GST_IDCT_H__
+
+#include <glib.h>
+
+typedef enum {
+  GST_IDCT_DEFAULT,    
+  GST_IDCT_INT,	
+  GST_IDCT_FAST_INT, 
+  GST_IDCT_FLOAT,   
+  GST_IDCT_MMX,	
+  GST_IDCT_MMX32,
+  GST_IDCT_SSE,
+} GstIDCTMethod;
+
+typedef struct _GstIDCT GstIDCT;
+typedef void (*GstIDCTFunction) (gshort *block);
+
+#define GST_IDCT_TRANSPOSE(idct) ((idct)->need_transpose)
+
+struct _GstIDCT {
+  /* private */
+  GstIDCTFunction convert;
+  GstIDCTFunction convert_sparse;
+  gboolean need_transpose;
+};
+
+
+GstIDCT *gst_idct_new(GstIDCTMethod method);
+#define gst_idct_convert(idct, blocks) (idct)->convert((blocks))
+#define gst_idct_convert_sparse(idct, blocks) (idct)->convert_sparse((blocks))
+void gst_idct_destroy(GstIDCT *idct);
+
+#endif /* __GST_IDCT_H__ */
diff --git a/gst-libs/gst/idct/ieeetest.c b/gst-libs/gst/idct/ieeetest.c
new file mode 100644
index 00000000..d78158e4
--- /dev/null
+++ b/gst-libs/gst/idct/ieeetest.c
@@ -0,0 +1,335 @@
+/*
+ * ieeetest.c --- test IDCT code against the IEEE Std 1180-1990 spec
+ *
+ * Note that this does only one pass of the test.
+ * Six invocations of ieeetest are needed to complete the entire spec.
+ * The shell script "doieee" performs the complete test.
+ *
+ * Written by Tom Lane (tgl@cs.cmu.edu).
+ * Released to public domain 11/22/93.
+ */
+
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+#include <math.h>
+
+#include <gst/gst.h>
+#include "gstidct.h"
+#include "dct.h"
+
+
+/* prototypes */
+
+void usage (char *msg);
+long ieeerand (long L, long H);
+void dct_init(void);
+void ref_fdct(DCTELEM block[8][8]);
+void ref_idct(DCTELEM block[8][8]);
+
+/* error stat accumulators -- assume initialized to 0 */
+
+long sumerrs[DCTSIZE2];
+long sumsqerrs[DCTSIZE2];
+int maxerr[DCTSIZE2];
+
+
+char * meets (double val, double limit)
+{
+  return ((fabs(val) <= limit) ? "meets" : "FAILS");
+}
+
+int
+main(int argc, char **argv)
+{
+  long minpix, maxpix, sign;
+  long curiter, niters;
+  int i, j;
+  double max, total;
+  int method;
+  DCTELEM   block[DCTSIZE2];	/* random source data */
+  DCTELEM   refcoefs[DCTSIZE2]; /* coefs from reference FDCT */
+  DCTELEM   refout[DCTSIZE2];	/* output from reference IDCT */
+  DCTELEM   testout[DCTSIZE2]; /* output from test IDCT */
+  GstIDCT   *idct;
+  guint64  tscstart, tscmin = ~0, tscmax = 0;
+  guint64  tscstop;
+
+  /* Argument parsing --- not very bulletproof at all */
+
+  if (argc != 6) usage(NULL);
+
+  method = atoi(argv[1]);
+  minpix = atoi(argv[2]);
+  maxpix = atoi(argv[3]);
+  sign   = atoi(argv[4]);
+  niters = atol(argv[5]);
+
+  gst_library_load("gstidct");
+
+  idct = gst_idct_new(method);
+  if (idct == 0) {
+    printf("method not available\n\n\n");
+
+    return 0;
+  }
+
+  dct_init();
+
+  /* Loop once per generated random-data block */
+
+  for (curiter = 0; curiter < niters; curiter++) {
+
+    /* generate a pseudo-random block of data */
+    for (i = 0; i < DCTSIZE2; i++)
+      block[i] = (DCTELEM) (ieeerand(-minpix,maxpix) * sign);
+
+    /* perform reference FDCT */
+    memcpy(refcoefs, block, sizeof(DCTELEM)*DCTSIZE2);
+    ref_fdct(refcoefs);
+    /* clip */
+    for (i = 0; i < DCTSIZE2; i++) {
+      if (refcoefs[i] < -2048) refcoefs[i] = -2048;
+      else if (refcoefs[i] > 2047) refcoefs[i] = 2047;
+    }
+
+    /* perform reference IDCT */
+    memcpy(refout, refcoefs, sizeof(DCTELEM)*DCTSIZE2);
+    ref_idct(refout);
+    /* clip */
+    for (i = 0; i < DCTSIZE2; i++) {
+      if (refout[i] < -256) refout[i] = -256;
+      else if (refout[i] > 255) refout[i] = 255;
+    }
+
+    /* perform test IDCT */
+	 if (GST_IDCT_TRANSPOSE(idct)) {
+      for (j = 0; j < DCTSIZE; j++) {
+        for (i = 0; i < DCTSIZE; i++) {
+		    testout[i*DCTSIZE+j] = refcoefs[j*DCTSIZE+i];
+		  }
+	   }
+	 }
+	 else {
+      memcpy(testout, refcoefs, sizeof(DCTELEM)*DCTSIZE2);
+	 }
+
+	 gst_trace_read_tsc(&tscstart);
+    gst_idct_convert(idct, testout);
+	 gst_trace_read_tsc(&tscstop);
+	 //printf("time %llu, %llu %lld\n", tscstart, tscstop, tscstop-tscstart);
+	 if (tscstop - tscstart < tscmin) tscmin = tscstop-tscstart;
+	 if (tscstop - tscstart > tscmax) tscmax = tscstop-tscstart;
+
+    /* clip */
+    for (i = 0; i < DCTSIZE2; i++) {
+      if (testout[i] < -256) testout[i] = -256;
+      else if (testout[i] > 255) testout[i] = 255;
+    }
+
+    /* accumulate error stats */
+    for (i = 0; i < DCTSIZE2; i++) {
+      register int err = testout[i] - refout[i];
+      sumerrs[i] += err;
+      sumsqerrs[i] += err * err;
+      if (err < 0) err = -err;
+      if (maxerr[i] < err) maxerr[i] = err;
+    }
+
+    if (curiter % 100 == 99) {
+      fprintf(stderr, ".");
+      fflush(stderr);
+    }
+  }
+  fprintf(stderr, "\n");
+
+  /* print results */
+
+  printf("IEEE test conditions: -L = %ld, +H = %ld, sign = %ld, #iters = %ld\n",
+	 minpix, maxpix, sign, niters);
+
+  printf("Speed, min time %lld, max %lld\n", tscmin, tscmax);
+
+  printf("Peak absolute values of errors:\n");
+  for (i = 0, j = 0; i < DCTSIZE2; i++) {
+    if (j < maxerr[i]) j = maxerr[i];
+    printf("%4d", maxerr[i]);
+    if ((i%DCTSIZE) == DCTSIZE-1) printf("\n");
+  }
+  printf("Worst peak error = %d  (%s spec limit 1)\n\n", j,
+	 meets((double) j, 1.0));
+
+  printf("Mean square errors:\n");
+  max = total = 0.0;
+  for (i = 0; i < DCTSIZE2; i++) {
+    double err = (double) sumsqerrs[i]  / ((double) niters);
+    total += (double) sumsqerrs[i];
+    if (max < err) max = err;
+    printf(" %8.4f", err);
+    if ((i%DCTSIZE) == DCTSIZE-1) printf("\n");
+  }
+  printf("Worst pmse = %.6f  (%s spec limit 0.06)\n", max, meets(max, 0.06));
+  total /= (double) (64*niters);
+  printf("Overall mse = %.6f  (%s spec limit 0.02)\n\n", total,
+	 meets(total, 0.02));
+
+  printf("Mean errors:\n");
+  max = total = 0.0;
+  for (i = 0; i < DCTSIZE2; i++) {
+    double err = (double) sumerrs[i]  / ((double) niters);
+    total += (double) sumerrs[i];
+    printf(" %8.4f", err);
+    if (err < 0.0) err = -err;
+    if (max < err) max = err;
+    if ((i%DCTSIZE) == DCTSIZE-1) printf("\n");
+  }
+  printf("Worst mean error = %.6f  (%s spec limit 0.015)\n", max,
+	 meets(max, 0.015));
+  total /= (double) (64*niters);
+  printf("Overall mean error = %.6f  (%s spec limit 0.0015)\n\n", total,
+	 meets(total, 0.0015));
+
+  /* test for 0 input giving 0 output */
+  memset(testout, 0, sizeof(DCTELEM)*DCTSIZE2);
+  gst_idct_convert(idct, testout);
+  for (i = 0, j=0; i < DCTSIZE2; i++) {
+    if (testout[i]) {
+      printf("Position %d of IDCT(0) = %d (FAILS)\n", i, testout[i]);
+      j++;
+    }
+  }
+  printf("%d elements of IDCT(0) were not zero\n\n\n", j);
+
+  exit(0);
+  return 0;
+}
+
+
+void usage (char *msg)
+{
+  if (msg != NULL)
+    fprintf(stderr, "\nerror: %s\n", msg);
+
+  fprintf(stderr, "\n");
+  fprintf(stderr, "usage: ieeetest minpix maxpix sign niters\n");
+  fprintf(stderr, "\n");
+  fprintf(stderr, "  test = 1 - 5\n");
+  fprintf(stderr, "  minpix = -L value per IEEE spec\n");
+  fprintf(stderr, "  maxpix =  H value per IEEE spec\n");
+  fprintf(stderr, "  sign = +1 for normal, -1 to run negated test\n");
+  fprintf(stderr, "  niters = # iterations (10000 for full test)\n");
+  fprintf(stderr, "\n");
+
+  exit(1);
+}
+
+
+/* Pseudo-random generator specified by IEEE 1180 */
+
+long ieeerand (long L, long H)
+{
+  static long randx = 1;
+  static double z = (double) 0x7fffffff;
+
+  long i,j;
+  double x;
+
+  randx = (randx * 1103515245) + 12345;
+  i = randx & 0x7ffffffe;
+  x = ((double) i) / z;
+  x *= (L+H+1);
+  j = x;
+  return j-L;
+}
+
+
+/* Reference double-precision FDCT and IDCT */
+
+
+/* The cosine lookup table */
+/* coslu[a][b] = C(b)/2 * cos[(2a+1)b*pi/16] */
+double coslu[8][8];
+
+
+/* Routine to initialise the cosine lookup table */
+void dct_init(void)
+{
+  int a,b;
+  double tmp;
+
+  for(a=0;a<8;a++)
+    for(b=0;b<8;b++) {
+      tmp = cos((double)((a+a+1)*b) * (3.14159265358979323846 / 16.0));
+      if(b==0)
+	tmp /= sqrt(2.0);
+      coslu[a][b] = tmp * 0.5;
+    }
+}
+
+
+void ref_fdct (DCTELEM block[8][8])
+{
+  int x,y,u,v;
+  double tmp, tmp2;
+  double res[8][8];
+
+  for (v=0; v<8; v++) {
+    for (u=0; u<8; u++) {
+      tmp = 0.0;
+      for (y=0; y<8; y++) {
+	tmp2 = 0.0;
+	for (x=0; x<8; x++) {
+	  tmp2 += (double) block[y][x] * coslu[x][u];
+	}
+	tmp += coslu[y][v] * tmp2;
+      }
+      res[v][u] = tmp;
+    }
+  }
+
+  for (v=0; v<8; v++) {
+    for (u=0; u<8; u++) {
+      tmp = res[v][u];
+      if (tmp < 0.0) {
+	x = - ((int) (0.5 - tmp));
+      } else {
+	x = (int) (tmp + 0.5);
+      }
+      block[v][u] = (DCTELEM) x;
+    }
+  }
+}
+
+
+void ref_idct (DCTELEM block[8][8])
+{
+  int x,y,u,v;
+  double tmp, tmp2;
+  double res[8][8];
+
+  for (y=0; y<8; y++) {
+    for (x=0; x<8; x++) {
+      tmp = 0.0;
+      for (v=0; v<8; v++) {
+	tmp2 = 0.0;
+	for (u=0; u<8; u++) {
+	  tmp2 += (double) block[v][u] * coslu[x][u];
+	}
+	tmp += coslu[y][v] * tmp2;
+      }
+      res[y][x] = tmp;
+    }
+  }
+
+  for (v=0; v<8; v++) {
+    for (u=0; u<8; u++) {
+      tmp = res[v][u];
+      if (tmp < 0.0) {
+	x = - ((int) (0.5 - tmp));
+      } else {
+	x = (int) (tmp + 0.5);
+      }
+      block[v][u] = (DCTELEM) x;
+    }
+  }
+}
diff --git a/gst-libs/gst/idct/intidct.c b/gst-libs/gst/idct/intidct.c
new file mode 100644
index 00000000..119b7fd1
--- /dev/null
+++ b/gst-libs/gst/idct/intidct.c
@@ -0,0 +1,391 @@
+/*
+ * jrevdct.c
+ *
+ * Copyright (C) 1991, 1992, Thomas G. Lane.
+ * This file is part of the Independent JPEG Group's software.
+ * For conditions of distribution and use, see the accompanying README file.
+ *
+ * This file contains the basic inverse-DCT transformation subroutine.
+ *
+ * This implementation is based on an algorithm described in
+ *   C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
+ *   Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
+ *   Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
+ * The primary algorithm described there uses 11 multiplies and 29 adds.
+ * We use their alternate method with 12 multiplies and 32 adds.
+ * The advantage of this method is that no data path contains more than one
+ * multiplication; this allows a very simple and accurate implementation in
+ * scaled fixed-point arithmetic, with a minimal number of shifts.
+ */
+
+#include "dct.h"
+
+/* We assume that right shift corresponds to signed division by 2 with
+ * rounding towards minus infinity.  This is correct for typical "arithmetic
+ * shift" instructions that shift in copies of the sign bit.  But some
+ * C compilers implement >> with an unsigned shift.  For these machines you
+ * must define RIGHT_SHIFT_IS_UNSIGNED.
+ * RIGHT_SHIFT provides a proper signed right shift of an INT32 quantity.
+ * It is only applied with constant shift counts.  SHIFT_TEMPS must be
+ * included in the variables of any routine using RIGHT_SHIFT.
+ */
+
+#ifdef RIGHT_SHIFT_IS_UNSIGNED
+#define SHIFT_TEMPS	INT32 shift_temp;
+#define RIGHT_SHIFT(x,shft)  \
+	((shift_temp = (x)) < 0 ? \
+	 (shift_temp >> (shft)) | ((~((INT32) 0)) << (32-(shft))) : \
+	 (shift_temp >> (shft)))
+#else
+#define SHIFT_TEMPS
+#define RIGHT_SHIFT(x,shft)	((x) >> (shft))
+#endif
+
+
+/*
+ * This routine is specialized to the case DCTSIZE = 8.
+ */
+
+#if DCTSIZE != 8
+  Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
+#endif
+
+
+/*
+ * A 2-D IDCT can be done by 1-D IDCT on each row followed by 1-D IDCT
+ * on each column.  Direct algorithms are also available, but they are
+ * much more complex and seem not to be any faster when reduced to code.
+ *
+ * The poop on this scaling stuff is as follows:
+ *
+ * Each 1-D IDCT step produces outputs which are a factor of sqrt(N)
+ * larger than the true IDCT outputs.  The final outputs are therefore
+ * a factor of N larger than desired; since N=8 this can be cured by
+ * a simple right shift at the end of the algorithm.  The advantage of
+ * this arrangement is that we save two multiplications per 1-D IDCT,
+ * because the y0 and y4 inputs need not be divided by sqrt(N).
+ *
+ * We have to do addition and subtraction of the integer inputs, which
+ * is no problem, and multiplication by fractional constants, which is
+ * a problem to do in integer arithmetic.  We multiply all the constants
+ * by CONST_SCALE and convert them to integer constants (thus retaining
+ * CONST_BITS bits of precision in the constants).  After doing a
+ * multiplication we have to divide the product by CONST_SCALE, with proper
+ * rounding, to produce the correct output.  This division can be done
+ * cheaply as a right shift of CONST_BITS bits.  We postpone shifting
+ * as long as possible so that partial sums can be added together with
+ * full fractional precision.
+ *
+ * The outputs of the first pass are scaled up by PASS1_BITS bits so that
+ * they are represented to better-than-integral precision.  These outputs
+ * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
+ * with the recommended scaling.  (To scale up 12-bit sample data further, an
+ * intermediate INT32 array would be needed.)
+ *
+ * To avoid overflow of the 32-bit intermediate results in pass 2, we must
+ * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26.  Error analysis
+ * shows that the values given below are the most effective.
+ */
+
+#ifdef EIGHT_BIT_SAMPLES
+#define CONST_BITS  13
+#define PASS1_BITS  2
+#else
+#define CONST_BITS  13
+#define PASS1_BITS  1		/* lose a little precision to avoid overflow */
+#endif
+
+#define ONE	((INT32) 1)
+
+#define CONST_SCALE (ONE << CONST_BITS)
+
+/* Convert a positive real constant to an integer scaled by CONST_SCALE. */
+
+#define FIX(x)	((INT32) ((x) * CONST_SCALE + 0.5))
+
+/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
+ * causing a lot of useless floating-point operations at run time.
+ * To get around this we use the following pre-calculated constants.
+ * If you change CONST_BITS you may want to add appropriate values.
+ * (With a reasonable C compiler, you can just rely on the FIX() macro...)
+ */
+
+#if CONST_BITS == 13
+#define FIX_0_298631336  ((INT32)  2446)	/* FIX(0.298631336) */
+#define FIX_0_390180644  ((INT32)  3196)	/* FIX(0.390180644) */
+#define FIX_0_541196100  ((INT32)  4433)	/* FIX(0.541196100) */
+#define FIX_0_765366865  ((INT32)  6270)	/* FIX(0.765366865) */
+#define FIX_0_899976223  ((INT32)  7373)	/* FIX(0.899976223) */
+#define FIX_1_175875602  ((INT32)  9633)	/* FIX(1.175875602) */
+#define FIX_1_501321110  ((INT32)  12299)	/* FIX(1.501321110) */
+#define FIX_1_847759065  ((INT32)  15137)	/* FIX(1.847759065) */
+#define FIX_1_961570560  ((INT32)  16069)	/* FIX(1.961570560) */
+#define FIX_2_053119869  ((INT32)  16819)	/* FIX(2.053119869) */
+#define FIX_2_562915447  ((INT32)  20995)	/* FIX(2.562915447) */
+#define FIX_3_072711026  ((INT32)  25172)	/* FIX(3.072711026) */
+#else
+#define FIX_0_298631336  FIX(0.298631336)
+#define FIX_0_390180644  FIX(0.390180644)
+#define FIX_0_541196100  FIX(0.541196100)
+#define FIX_0_765366865  FIX(0.765366865)
+#define FIX_0_899976223  FIX(0.899976223)
+#define FIX_1_175875602  FIX(1.175875602)
+#define FIX_1_501321110  FIX(1.501321110)
+#define FIX_1_847759065  FIX(1.847759065)
+#define FIX_1_961570560  FIX(1.961570560)
+#define FIX_2_053119869  FIX(2.053119869)
+#define FIX_2_562915447  FIX(2.562915447)
+#define FIX_3_072711026  FIX(3.072711026)
+#endif
+
+
+/* Descale and correctly round an INT32 value that's scaled by N bits.
+ * We assume RIGHT_SHIFT rounds towards minus infinity, so adding
+ * the fudge factor is correct for either sign of X.
+ */
+
+#define DESCALE(x,n)  RIGHT_SHIFT((x) + (ONE << ((n)-1)), n)
+
+/* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
+ * For 8-bit samples with the recommended scaling, all the variable
+ * and constant values involved are no more than 16 bits wide, so a
+ * 16x16->32 bit multiply can be used instead of a full 32x32 multiply;
+ * this provides a useful speedup on many machines.
+ * There is no way to specify a 16x16->32 multiply in portable C, but
+ * some C compilers will do the right thing if you provide the correct
+ * combination of casts.
+ * NB: for 12-bit samples, a full 32-bit multiplication will be needed.
+ */
+
+#ifdef EIGHT_BIT_SAMPLES
+#ifdef SHORTxSHORT_32		/* may work if 'int' is 32 bits */
+#define MULTIPLY(var,const)  (((INT16) (var)) * ((INT16) (const)))
+#endif
+#ifdef SHORTxLCONST_32		/* known to work with Microsoft C 6.0 */
+#define MULTIPLY(var,const)  (((INT16) (var)) * ((INT32) (const)))
+#endif
+#endif
+
+#ifndef MULTIPLY		/* default definition */
+#define MULTIPLY(var,const)  ((var) * (const))
+#endif
+
+
+/*
+ * Perform the inverse DCT on one block of coefficients.
+ */
+
+void
+gst_idct_int_idct (DCTBLOCK data)
+{
+  INT32 tmp0, tmp1, tmp2, tmp3;
+  INT32 tmp10, tmp11, tmp12, tmp13;
+  INT32 z1, z2, z3, z4, z5;
+  register DCTELEM *dataptr;
+  int rowctr;
+  SHIFT_TEMPS
+
+  /* Pass 1: process rows. */
+  /* Note results are scaled up by sqrt(8) compared to a true IDCT; */
+  /* furthermore, we scale the results by 2**PASS1_BITS. */
+
+  dataptr = data;
+  for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--) {
+    /* Due to quantization, we will usually find that many of the input
+     * coefficients are zero, especially the AC terms.  We can exploit this
+     * by short-circuiting the IDCT calculation for any row in which all
+     * the AC terms are zero.  In that case each output is equal to the
+     * DC coefficient (with scale factor as needed).
+     * With typical images and quantization tables, half or more of the
+     * row DCT calculations can be simplified this way.
+     */
+
+    if ((dataptr[1] | dataptr[2] | dataptr[3] | dataptr[4] |
+	 dataptr[5] | dataptr[6] | dataptr[7]) == 0) {
+      /* AC terms all zero */
+      DCTELEM dcval = (DCTELEM) (dataptr[0] << PASS1_BITS);
+      
+      dataptr[0] = dcval;
+      dataptr[1] = dcval;
+      dataptr[2] = dcval;
+      dataptr[3] = dcval;
+      dataptr[4] = dcval;
+      dataptr[5] = dcval;
+      dataptr[6] = dcval;
+      dataptr[7] = dcval;
+      
+      dataptr += DCTSIZE;	/* advance pointer to next row */
+      continue;
+    }
+
+    /* Even part: reverse the even part of the forward DCT. */
+    /* The rotator is sqrt(2)*c(-6). */
+
+    z2 = (INT32) dataptr[2];
+    z3 = (INT32) dataptr[6];
+
+    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
+    tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
+    tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
+
+    tmp0 = ((INT32) dataptr[0] + (INT32) dataptr[4]) << CONST_BITS;
+    tmp1 = ((INT32) dataptr[0] - (INT32) dataptr[4]) << CONST_BITS;
+
+    tmp10 = tmp0 + tmp3;
+    tmp13 = tmp0 - tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp12 = tmp1 - tmp2;
+    
+    /* Odd part per figure 8; the matrix is unitary and hence its
+     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
+     */
+
+    tmp0 = (INT32) dataptr[7];
+    tmp1 = (INT32) dataptr[5];
+    tmp2 = (INT32) dataptr[3];
+    tmp3 = (INT32) dataptr[1];
+
+    z1 = tmp0 + tmp3;
+    z2 = tmp1 + tmp2;
+    z3 = tmp0 + tmp2;
+    z4 = tmp1 + tmp3;
+    z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
+    
+    tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
+    tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
+    tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
+    tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
+    z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
+    z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
+    z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
+    z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
+    
+    z3 += z5;
+    z4 += z5;
+    
+    tmp0 += z1 + z3;
+    tmp1 += z2 + z4;
+    tmp2 += z2 + z3;
+    tmp3 += z1 + z4;
+
+    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
+
+    dataptr[0] = (DCTELEM) DESCALE(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
+    dataptr[7] = (DCTELEM) DESCALE(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
+    dataptr[1] = (DCTELEM) DESCALE(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
+    dataptr[6] = (DCTELEM) DESCALE(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
+    dataptr[2] = (DCTELEM) DESCALE(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
+    dataptr[5] = (DCTELEM) DESCALE(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
+    dataptr[3] = (DCTELEM) DESCALE(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
+    dataptr[4] = (DCTELEM) DESCALE(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
+
+    dataptr += DCTSIZE;		/* advance pointer to next row */
+  }
+
+  /* Pass 2: process columns. */
+  /* Note that we must descale the results by a factor of 8 == 2**3, */
+  /* and also undo the PASS1_BITS scaling. */
+
+  dataptr = data;
+  for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--) {
+    /* Columns of zeroes can be exploited in the same way as we did with rows.
+     * However, the row calculation has created many nonzero AC terms, so the
+     * simplification applies less often (typically 5% to 10% of the time).
+     * On machines with very fast multiplication, it's possible that the
+     * test takes more time than it's worth.  In that case this section
+     * may be commented out.
+     */
+
+#ifndef NO_ZERO_COLUMN_TEST
+    if ((dataptr[DCTSIZE*1] | dataptr[DCTSIZE*2] | dataptr[DCTSIZE*3] |
+	 dataptr[DCTSIZE*4] | dataptr[DCTSIZE*5] | dataptr[DCTSIZE*6] |
+	 dataptr[DCTSIZE*7]) == 0) {
+      /* AC terms all zero */
+      DCTELEM dcval = (DCTELEM) DESCALE((INT32) dataptr[0], PASS1_BITS+3);
+      
+      dataptr[DCTSIZE*0] = dcval;
+      dataptr[DCTSIZE*1] = dcval;
+      dataptr[DCTSIZE*2] = dcval;
+      dataptr[DCTSIZE*3] = dcval;
+      dataptr[DCTSIZE*4] = dcval;
+      dataptr[DCTSIZE*5] = dcval;
+      dataptr[DCTSIZE*6] = dcval;
+      dataptr[DCTSIZE*7] = dcval;
+      
+      dataptr++;		/* advance pointer to next column */
+      continue;
+    }
+#endif
+
+    /* Even part: reverse the even part of the forward DCT. */
+    /* The rotator is sqrt(2)*c(-6). */
+
+    z2 = (INT32) dataptr[DCTSIZE*2];
+    z3 = (INT32) dataptr[DCTSIZE*6];
+
+    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
+    tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
+    tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
+
+    tmp0 = ((INT32) dataptr[DCTSIZE*0] + (INT32) dataptr[DCTSIZE*4]) << CONST_BITS;
+    tmp1 = ((INT32) dataptr[DCTSIZE*0] - (INT32) dataptr[DCTSIZE*4]) << CONST_BITS;
+
+    tmp10 = tmp0 + tmp3;
+    tmp13 = tmp0 - tmp3;
+    tmp11 = tmp1 + tmp2;
+    tmp12 = tmp1 - tmp2;
+    
+    /* Odd part per figure 8; the matrix is unitary and hence its
+     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
+     */
+
+    tmp0 = (INT32) dataptr[DCTSIZE*7];
+    tmp1 = (INT32) dataptr[DCTSIZE*5];
+    tmp2 = (INT32) dataptr[DCTSIZE*3];
+    tmp3 = (INT32) dataptr[DCTSIZE*1];
+
+    z1 = tmp0 + tmp3;
+    z2 = tmp1 + tmp2;
+    z3 = tmp0 + tmp2;
+    z4 = tmp1 + tmp3;
+    z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
+    
+    tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
+    tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
+    tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
+    tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
+    z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
+    z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
+    z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
+    z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
+    
+    z3 += z5;
+    z4 += z5;
+    
+    tmp0 += z1 + z3;
+    tmp1 += z2 + z4;
+    tmp2 += z2 + z3;
+    tmp3 += z1 + z4;
+
+    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
+
+    dataptr[DCTSIZE*0] = (DCTELEM) DESCALE(tmp10 + tmp3,
+					   CONST_BITS+PASS1_BITS+3);
+    dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp10 - tmp3,
+					   CONST_BITS+PASS1_BITS+3);
+    dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp11 + tmp2,
+					   CONST_BITS+PASS1_BITS+3);
+    dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(tmp11 - tmp2,
+					   CONST_BITS+PASS1_BITS+3);
+    dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(tmp12 + tmp1,
+					   CONST_BITS+PASS1_BITS+3);
+    dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp12 - tmp1,
+					   CONST_BITS+PASS1_BITS+3);
+    dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp13 + tmp0,
+					   CONST_BITS+PASS1_BITS+3);
+    dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp13 - tmp0,
+					   CONST_BITS+PASS1_BITS+3);
+    
+    dataptr++;			/* advance pointer to next column */
+  }
+}
diff --git a/gst-libs/gst/idct/mmx32idct.c b/gst-libs/gst/idct/mmx32idct.c
new file mode 100644
index 00000000..78bf45bf
--- /dev/null
+++ b/gst-libs/gst/idct/mmx32idct.c
@@ -0,0 +1,783 @@
+/* 

+ *  idctmmx32.cpp 

+ *

+ *	Copyright (C) Alberto Vigata - January 2000 - ultraflask@yahoo.com

+ *

+ *  This file is part of FlasKMPEG, a free MPEG to MPEG/AVI converter

+ *	

+ *  FlasKMPEG is free software; you can redistribute it and/or modify

+ *  it under the terms of the GNU General Public License as published by

+ *  the Free Software Foundation; either version 2, or (at your option)

+ *  any later version.

+ *   

+ *  FlasKMPEG is distributed in the hope that it will be useful,

+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of

+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the

+ *  GNU General Public License for more details.

+ *   

+ *  You should have received a copy of the GNU General Public License

+ *  along with GNU Make; see the file COPYING.  If not, write to

+ *  the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. 

+ *

+ */

+

+

+// MMX32 iDCT algorithm  (IEEE-1180 compliant) :: idct_mmx32()

+//

+// MPEG2AVI

+// --------

+//  v0.16B33 initial release

+//

+// This was one of the harder pieces of work to code.

+// Intel's app-note focuses on the numerical issues of the algorithm, but

+// assumes the programmer is familiar with IDCT mathematics, leaving the

+// form of the complete function up to the programmer's imagination.

+//

+//  ALGORITHM OVERVIEW

+//  ------------------

+// I played around with the code for quite a few hours.  I came up

+// with *A* working IDCT algorithm, however I'm not sure whether my routine

+// is "the correct one."  But rest assured, my code passes all six IEEE 

+// accuracy tests with plenty of margin.

+//

+//   My IDCT algorithm consists of 4 steps:

+//

+//   1) IDCT-row transformation (using the IDCT-row function) on all 8 rows

+//      This yields an intermediate 8x8 matrix.

+//

+//   2) intermediate matrix transpose (mandatory)

+//

+//   3) IDCT-row transformation (2nd time) on all 8 rows of the intermediate

+//      matrix.  The output is the final-result, in transposed form.

+//

+//   4) post-transformation matrix transpose 

+//      (not necessary if the input-data is already transposed, this could

+//       be done during the MPEG "zig-zag" scan, but since my algorithm

+//       requires at least one transpose operation, why not re-use the

+//       transpose-code.)

+//

+//   Although the (1st) and (3rd) steps use the SAME row-transform operation,

+//   the (3rd) step uses different shift&round constants (explained later.)

+//

+//   Also note that the intermediate transpose (2) would not be neccessary,

+//   if the subsequent operation were a iDCT-column transformation.  Since

+//   we only have the iDCT-row transform, we transpose the intermediate

+//   matrix and use the iDCT-row transform a 2nd time.

+//

+//   I had to change some constants/variables for my method to work :

+//

+//      As given by Intel, the #defines for SHIFT_INV_COL and RND_INV_COL are

+//      wrong.  Not surprising since I'm not using a true column-transform 

+//      operation, but the row-transform operation (as mentioned earlier.)

+//      round_inv_col[], which is given as "4 short" values, should have the

+//      same dimensions as round_inv_row[].  The corrected variables are 

+//      shown.

+//

+//      Intel's code defines a different table for each each row operation.

+//      The tables given are 0/4, 1/7, 2/6, and 5/3.  My code only uses row#0.

+//      Using the other rows messes up the overall transform.

+//

+//   IMPLEMENTATION DETAILs

+//   ----------------------

+// 

+//   I divided the algorithm's work into two subroutines,

+//    1) idct_mmx32_rows() - transforms 8 rows, then transpose

+//    2) idct_mmx32_cols() - transforms 8 rows, then transpose

+//       yields final result ("drop-in" direct replacement for INT32 IDCT)

+//

+//   The 2nd function is a clone of the 1st, with changes made only to the

+//   shift&rounding instructions.

+//

+//      In the 1st function (rows), the shift & round instructions use 

+//       SHIFT_INV_ROW & round_inv_row[] (renamed to r_inv_row[])

+//

+//      In the 2nd function (cols)-> r_inv_col[], and

+//       SHIFT_INV_COL & round_inv_col[] (renamed to r_inv_col[])

+//

+//   Each function contains an integrated transpose-operator, which comes

+//   AFTER the primary transformation operation.  In the future, I'll optimize

+//   the code to do more of the transpose-work "in-place".  Right now, I've

+//   left the code as two subroutines and a main calling function, so other

+//   people can read the code more easily.

+//

+//   liaor@umcc.ais.org  http://members.tripod.com/~liaor

+//  

+

+

+//;=============================================================================

+//;

+//;  AP-922   http://developer.intel.com/vtune/cbts/strmsimd

+//; These examples contain code fragments for first stage iDCT 8x8

+//; (for rows) and first stage DCT 8x8 (for columns)

+//;

+//;=============================================================================

+/*

+mword typedef qword

+qword ptr equ mword ptr */

+

+#include <mmx.h>

+

+#define BITS_INV_ACC	4	//; 4 or 5 for IEEE

+	// 5 yields higher accuracy, but lessens dynamic range on the input matrix

+#define SHIFT_INV_ROW	(16 - BITS_INV_ACC)

+#define SHIFT_INV_COL	(1 + BITS_INV_ACC +14 )  // changed from Intel's val)

+//#define SHIFT_INV_COL	(1 + BITS_INV_ACC )

+

+#define RND_INV_ROW		(1 << (SHIFT_INV_ROW-1))

+#define RND_INV_COL		(1 << (SHIFT_INV_COL-1)) 

+#define RND_INV_CORR	(RND_INV_COL - 1)		//; correction -1.0 and round

+//#define RND_INV_ROW		(1024 * (6 - BITS_INV_ACC)) //; 1 << (SHIFT_INV_ROW-1)

+//#define RND_INV_COL		(16 * (BITS_INV_ACC - 3)) //; 1 << (SHIFT_INV_COL-1)

+

+

+//.data

+//Align 16

+const static long r_inv_row[2] = { RND_INV_ROW, RND_INV_ROW};

+const static long r_inv_col[2] = {RND_INV_COL, RND_INV_COL};

+const static long r_inv_corr[2] = {RND_INV_CORR, RND_INV_CORR };

+

+//const static short r_inv_col[4] = 

+//	{RND_INV_COL, RND_INV_COL, RND_INV_COL, RND_INV_COL};

+//const static short r_inv_corr[4] =

+//	{RND_INV_CORR, RND_INV_CORR, RND_INV_CORR, RND_INV_CORR};

+

+/* constants for the forward DCT

+

+//#define BITS_FRW_ACC	3 //; 2 or 3 for accuracy

+//#define SHIFT_FRW_COL	BITS_FRW_ACC

+//#define SHIFT_FRW_ROW	(BITS_FRW_ACC + 17)

+//#define RND_FRW_ROW		(262144 * (BITS_FRW_ACC - 1)) //; 1 << (SHIFT_FRW_ROW-1)

+

+const static __int64 one_corr = 0x0001000100010001;

+const static long r_frw_row[2] = {RND_FRW_ROW, RND_FRW_ROW };

+

+//const static short tg_1_16[4] = {13036, 13036, 13036, 13036 }; //tg * (2<<16) + 0.5

+//const static short tg_2_16[4] = {27146, 27146, 27146, 27146 }; //tg * (2<<16) + 0.5

+//const static short tg_3_16[4] = {-21746, -21746, -21746, -21746 }; //tg * (2<<16) + 0.5

+//const static short cos_4_16[4] = {-19195, -19195, -19195, -19195 }; //cos * (2<<16) + 0.5

+//const static short ocos_4_16[4] = {23170, 23170, 23170, 23170 }; //cos * (2<<15) + 0.5

+

+//concatenated table, for forward DCT transformation

+const static short tg_all_16[] = {

+	13036, 13036, 13036, 13036,		// tg * (2<<16) + 0.5

+	27146, 27146, 27146, 27146,		//tg * (2<<16) + 0.5

+	-21746, -21746, -21746, -21746,	// tg * (2<<16) + 0.5

+	-19195, -19195, -19195, -19195,	//cos * (2<<16) + 0.5

+	23170, 23170, 23170, 23170 };	//cos * (2<<15) + 0.5

+

+#define tg_1_16 (tg_all_16 + 0)

+#define tg_2_16 (tg_all_16 + 8)

+#define tg_3_16 (tg_all_16 + 16)

+#define cos_4_16 (tg_all_16 + 24)

+#define ocos_4_16 (tg_all_16 + 32)

+*/

+/*

+;=============================================================================

+;

+; The first stage iDCT 8x8 - inverse DCTs of rows

+;

+;-----------------------------------------------------------------------------

+; The 8-point inverse DCT direct algorithm

+;-----------------------------------------------------------------------------

+;

+; static const short w[32] = {

+; FIX(cos_4_16), FIX(cos_2_16), FIX(cos_4_16), FIX(cos_6_16),

+; FIX(cos_4_16), FIX(cos_6_16), -FIX(cos_4_16), -FIX(cos_2_16),

+; FIX(cos_4_16), -FIX(cos_6_16), -FIX(cos_4_16), FIX(cos_2_16),

+; FIX(cos_4_16), -FIX(cos_2_16), FIX(cos_4_16), -FIX(cos_6_16),

+; FIX(cos_1_16), FIX(cos_3_16), FIX(cos_5_16), FIX(cos_7_16),

+; FIX(cos_3_16), -FIX(cos_7_16), -FIX(cos_1_16), -FIX(cos_5_16),

+; FIX(cos_5_16), -FIX(cos_1_16), FIX(cos_7_16), FIX(cos_3_16),

+; FIX(cos_7_16), -FIX(cos_5_16), FIX(cos_3_16), -FIX(cos_1_16) };

+;

+; #define DCT_8_INV_ROW(x, y)

+

+;{

+; int a0, a1, a2, a3, b0, b1, b2, b3;

+;

+; a0 =x[0]*w[0]+x[2]*w[1]+x[4]*w[2]+x[6]*w[3];

+; a1 =x[0]*w[4]+x[2]*w[5]+x[4]*w[6]+x[6]*w[7];

+; a2 = x[0] * w[ 8] + x[2] * w[ 9] + x[4] * w[10] + x[6] * w[11];

+; a3 = x[0] * w[12] + x[2] * w[13] + x[4] * w[14] + x[6] * w[15];

+; b0 = x[1] * w[16] + x[3] * w[17] + x[5] * w[18] + x[7] * w[19];

+; b1 = x[1] * w[20] + x[3] * w[21] + x[5] * w[22] + x[7] * w[23];

+; b2 = x[1] * w[24] + x[3] * w[25] + x[5] * w[26] + x[7] * w[27];

+; b3 = x[1] * w[28] + x[3] * w[29] + x[5] * w[30] + x[7] * w[31];

+;

+; y[0] = SHIFT_ROUND ( a0 + b0 );

+; y[1] = SHIFT_ROUND ( a1 + b1 );

+; y[2] = SHIFT_ROUND ( a2 + b2 );

+; y[3] = SHIFT_ROUND ( a3 + b3 );

+; y[4] = SHIFT_ROUND ( a3 - b3 );

+; y[5] = SHIFT_ROUND ( a2 - b2 );

+; y[6] = SHIFT_ROUND ( a1 - b1 );

+; y[7] = SHIFT_ROUND ( a0 - b0 );

+;}

+;

+;-----------------------------------------------------------------------------

+;

+; In this implementation the outputs of the iDCT-1D are multiplied

+; for rows 0,4 - by cos_4_16,

+; for rows 1,7 - by cos_1_16,

+; for rows 2,6 - by cos_2_16,

+; for rows 3,5 - by cos_3_16

+; and are shifted to the left for better accuracy

+;

+; For the constants used,

+; FIX(float_const) = (short) (float_const * (1<<15) + 0.5)

+;

+;=============================================================================

+;=============================================================================

+IF _MMX ; MMX code

+;=============================================================================

+

+//; Table for rows 0,4 - constants are multiplied by cos_4_16

+const short tab_i_04[] = {

+	16384, 16384, 16384, -16384,	// ; movq-> w06 w04 w02 w00

+	21407, 8867, 8867, -21407,		// w07 w05 w03 w01

+	16384, -16384, 16384, 16384,	//; w14 w12 w10 w08

+	-8867, 21407, -21407, -8867,	//; w15 w13 w11 w09

+	22725, 12873, 19266, -22725,	//; w22 w20 w18 w16

+	19266, 4520, -4520, -12873,		//; w23 w21 w19 w17

+	12873, 4520, 4520, 19266,		//; w30 w28 w26 w24

+	-22725, 19266, -12873, -22725 };//w31 w29 w27 w25

+

+//; Table for rows 1,7 - constants are multiplied by cos_1_16

+const short tab_i_17[] = {

+	22725, 22725, 22725, -22725,	// ; movq-> w06 w04 w02 w00

+	29692, 12299, 12299, -29692,	//	; w07 w05 w03 w01

+	22725, -22725, 22725, 22725,	//; w14 w12 w10 w08

+	-12299, 29692, -29692, -12299,	//; w15 w13 w11 w09

+	31521, 17855, 26722, -31521,	//; w22 w20 w18 w16

+	26722, 6270, -6270, -17855,		//; w23 w21 w19 w17

+	17855, 6270, 6270, 26722,		//; w30 w28 w26 w24

+	-31521, 26722, -17855, -31521};	// w31 w29 w27 w25

+

+//; Table for rows 2,6 - constants are multiplied by cos_2_16

+const short tab_i_26[] = {

+	21407, 21407, 21407, -21407,	// ; movq-> w06 w04 w02 w00

+	27969, 11585, 11585, -27969,	// ; w07 w05 w03 w01

+	21407, -21407, 21407, 21407,	// ; w14 w12 w10 w08

+	-11585, 27969, -27969, -11585,	//  ;w15 w13 w11 w09

+	29692, 16819, 25172, -29692, 	// ;w22 w20 w18 w16

+	25172, 5906, -5906, -16819, 	// ;w23 w21 w19 w17

+	16819, 5906, 5906, 25172, 		// ;w30 w28 w26 w24

+	-29692, 25172, -16819, -29692};	//  ;w31 w29 w27 w25

+

+

+//; Table for rows 3,5 - constants are multiplied by cos_3_16

+const short tab_i_35[] = {

+	19266, 19266, 19266, -19266,	//; movq-> w06 w04 w02 w00

+	25172, 10426, 10426, -25172,	//; w07 w05 w03 w01

+	19266, -19266, 19266, 19266,	//; w14 w12 w10 w08

+	-10426, 25172, -25172, -10426,	//; w15 w13 w11 w09

+	26722, 15137, 22654, -26722,	//; w22 w20 w18 w16

+	22654, 5315, -5315, -15137,		//; w23 w21 w19 w17

+	15137, 5315, 5315, 22654,		//; w30 w28 w26 w24

+	-26722, 22654, -15137, -26722};	//; w31 w29 w27 w25

+*/

+

+// CONCATENATED TABLE, rows 0,1,2,3,4,5,6,7 (in order )

+//

+// In our implementation, however, we only use row0 !

+//

+static const short tab_i_01234567[] = {

+	//row0, this row is required

+	16384, 16384, 16384, -16384,	// ; movq-> w06 w04 w02 w00

+	21407, 8867, 8867, -21407,		// w07 w05 w03 w01

+	16384, -16384, 16384, 16384,	//; w14 w12 w10 w08

+	-8867, 21407, -21407, -8867,	//; w15 w13 w11 w09

+	22725, 12873, 19266, -22725,	//; w22 w20 w18 w16

+	19266, 4520, -4520, -12873,		//; w23 w21 w19 w17

+	12873, 4520, 4520, 19266,		//; w30 w28 w26 w24

+	-22725, 19266, -12873, -22725,  //w31 w29 w27 w25

+

+	// the rest of these rows (1-7), aren't used !

+

+	//row1

+	22725, 22725, 22725, -22725,	// ; movq-> w06 w04 w02 w00

+	29692, 12299, 12299, -29692,	//	; w07 w05 w03 w01

+	22725, -22725, 22725, 22725,	//; w14 w12 w10 w08

+	-12299, 29692, -29692, -12299,	//; w15 w13 w11 w09

+	31521, 17855, 26722, -31521,	//; w22 w20 w18 w16

+	26722, 6270, -6270, -17855,		//; w23 w21 w19 w17

+	17855, 6270, 6270, 26722,		//; w30 w28 w26 w24

+	-31521, 26722, -17855, -31521,	// w31 w29 w27 w25

+

+	//row2

+	21407, 21407, 21407, -21407,	// ; movq-> w06 w04 w02 w00

+	27969, 11585, 11585, -27969,	// ; w07 w05 w03 w01

+	21407, -21407, 21407, 21407,	// ; w14 w12 w10 w08

+	-11585, 27969, -27969, -11585,	//  ;w15 w13 w11 w09

+	29692, 16819, 25172, -29692, 	// ;w22 w20 w18 w16

+	25172, 5906, -5906, -16819, 	// ;w23 w21 w19 w17

+	16819, 5906, 5906, 25172, 		// ;w30 w28 w26 w24

+	-29692, 25172, -16819, -29692,	//  ;w31 w29 w27 w25

+

+	//row3

+	19266, 19266, 19266, -19266,	//; movq-> w06 w04 w02 w00

+	25172, 10426, 10426, -25172,	//; w07 w05 w03 w01

+	19266, -19266, 19266, 19266,	//; w14 w12 w10 w08

+	-10426, 25172, -25172, -10426,	//; w15 w13 w11 w09

+	26722, 15137, 22654, -26722,	//; w22 w20 w18 w16

+	22654, 5315, -5315, -15137,		//; w23 w21 w19 w17

+	15137, 5315, 5315, 22654,		//; w30 w28 w26 w24

+	-26722, 22654, -15137, -26722,	//; w31 w29 w27 w25

+

+	//row4

+	16384, 16384, 16384, -16384,	// ; movq-> w06 w04 w02 w00

+	21407, 8867, 8867, -21407,		// w07 w05 w03 w01

+	16384, -16384, 16384, 16384,	//; w14 w12 w10 w08

+	-8867, 21407, -21407, -8867,	//; w15 w13 w11 w09

+	22725, 12873, 19266, -22725,	//; w22 w20 w18 w16

+	19266, 4520, -4520, -12873,		//; w23 w21 w19 w17

+	12873, 4520, 4520, 19266,		//; w30 w28 w26 w24

+	-22725, 19266, -12873, -22725,  //w31 w29 w27 w25

+

+	//row5

+	19266, 19266, 19266, -19266,	//; movq-> w06 w04 w02 w00

+	25172, 10426, 10426, -25172,	//; w07 w05 w03 w01

+	19266, -19266, 19266, 19266,	//; w14 w12 w10 w08

+	-10426, 25172, -25172, -10426,	//; w15 w13 w11 w09

+	26722, 15137, 22654, -26722,	//; w22 w20 w18 w16

+	22654, 5315, -5315, -15137,		//; w23 w21 w19 w17

+	15137, 5315, 5315, 22654,		//; w30 w28 w26 w24

+	-26722, 22654, -15137, -26722,	//; w31 w29 w27 w25

+

+	//row6

+	21407, 21407, 21407, -21407,	// ; movq-> w06 w04 w02 w00

+	27969, 11585, 11585, -27969,	// ; w07 w05 w03 w01

+	21407, -21407, 21407, 21407,	// ; w14 w12 w10 w08

+	-11585, 27969, -27969, -11585,	//  ;w15 w13 w11 w09

+	29692, 16819, 25172, -29692, 	// ;w22 w20 w18 w16

+	25172, 5906, -5906, -16819, 	// ;w23 w21 w19 w17

+	16819, 5906, 5906, 25172, 		// ;w30 w28 w26 w24

+	-29692, 25172, -16819, -29692,	//  ;w31 w29 w27 w25

+

+	//row7

+	22725, 22725, 22725, -22725,	// ; movq-> w06 w04 w02 w00

+	29692, 12299, 12299, -29692,	//	; w07 w05 w03 w01

+	22725, -22725, 22725, 22725,	//; w14 w12 w10 w08

+	-12299, 29692, -29692, -12299,	//; w15 w13 w11 w09

+	31521, 17855, 26722, -31521,	//; w22 w20 w18 w16

+	26722, 6270, -6270, -17855,		//; w23 w21 w19 w17

+	17855, 6270, 6270, 26722,		//; w30 w28 w26 w24

+	-31521, 26722, -17855, -31521};	// w31 w29 w27 w25

+

+

+#define INP eax		// pointer to (short *blk)

+#define OUT ecx		// pointer to output (temporary store space qwTemp[])

+#define TABLE ebx	// pointer to tab_i_01234567[]

+#define round_inv_row edx

+#define round_inv_col edx

+

+#define ROW_STRIDE 8 // for 8x8 matrix transposer

+

+// private variables and functions

+

+//temporary storage space, 8x8 of shorts

+

+__inline static void idct_mmx32_rows( short *blk ); // transform rows

+__inline static void idct_mmx32_cols( short *blk ); // transform "columns"

+	// the "column" transform actually transforms rows, it is

+	// identical to the row-transform except for the ROUNDING

+	// and SHIFTING coefficients.

+

+ 

+static void 

+idct_mmx32_rows( short *blk )	// transform all 8 rows of 8x8 iDCT block

+{

+  int x;

+  short qwTemp[64];

+  short *out = &qwTemp[0];

+  short *inptr = blk;

+  // this subroutine performs two operations

+  // 1) iDCT row transform

+  //		for( i = 0; i < 8; ++ i)

+  //			DCT_8_INV_ROW_1( blk[i*8], qwTemp[i] );

+  //

+  // 2) transpose the matrix (which was stored in qwTemp[])

+  //        qwTemp[] -> [8x8 matrix transpose] -> blk[]

+

+  for (x=0; x<8; x++) {  // transform one row per iteration

+	 movq_m2r(*(inptr), mm0);		// 0 ; x3 x2 x1 x0

+

+	 movq_m2r(*(inptr+4), mm1);	// 1 ; x7 x6 x5 x4

+	 movq_r2r(mm0, mm2);				// 2 ; x3 x2 x1 x0

+

+	 movq_m2r(*(tab_i_01234567), mm3);	// 3 ; w06 w04 w02 w00

+	 punpcklwd_r2r(mm1, mm0);			// x5 x1 x4 x0

+

+    // ----------

+	 movq_r2r(mm0, mm5);					// 5 ; x5 x1 x4 x0

+	 punpckldq_r2r(mm0, mm0);			// x4 x0 x4 x0

+

+	 movq_m2r(*(tab_i_01234567+4), mm4);	// 4 ; w07 w05 w03 w01

+	 punpckhwd_r2r(mm1, mm2);			// 1 ; x7 x3 x6 x2

+

+	 pmaddwd_r2r(mm0, mm3);				// x4*w06+x0*w04 x4*w02+x0*w00

+	 movq_r2r(mm2, mm6);				// 6 ; x7 x3 x6 x2

+

+	 movq_m2r(*(tab_i_01234567+16), mm1);// 1 ; w22 w20 w18 w16

+	 punpckldq_r2r(mm2, mm2);			// x6 x2 x6 x2

+

+	 pmaddwd_r2r(mm2, mm4);				// x6*w07+x2*w05 x6*w03+x2*w01

+	 punpckhdq_r2r(mm5, mm5);			// x5 x1 x5 x1

+

+	 pmaddwd_m2r(*(tab_i_01234567+8), mm0);// x4*w14+x0*w12 x4*w10+x0*w08

+	 punpckhdq_r2r(mm6, mm6);			// x7 x3 x7 x3

+

+	 movq_m2r(*(tab_i_01234567+20), mm7);// 7 ; w23 w21 w19 w17

+	 pmaddwd_r2r(mm5, mm1);				// x5*w22+x1*w20 x5*w18+x1*w16

+

+	 paddd_m2r(*(r_inv_row), mm3);// +rounder

+	 pmaddwd_r2r(mm6, mm7);				// x7*w23+x3*w21 x7*w19+x3*w17

+

+	 pmaddwd_m2r(*(tab_i_01234567+12), mm2);// x6*w15+x2*w13 x6*w11+x2*w09

+	 paddd_r2r(mm4, mm3);				// 4 ; a1=sum(even1) a0=sum(even0)

+

+	 pmaddwd_m2r(*(tab_i_01234567+24), mm5);// x5*w30+x1*w28 x5*w26+x1*w24

+	 movq_r2r(mm3, mm4);				// 4 ; a1 a0

+

+	 pmaddwd_m2r(*(tab_i_01234567+28), mm6);// x7*w31+x3*w29 x7*w27+x3*w25

+	 paddd_r2r(mm7, mm1);				// 7 ; b1=sum(odd1) b0=sum(odd0)

+

+	 paddd_m2r(*(r_inv_row), mm0);// +rounder

+	 psubd_r2r(mm1, mm3);				// a1-b1 a0-b0

+

+	 psrad_i2r(SHIFT_INV_ROW, mm3);		// y6=a1-b1 y7=a0-b0

+	 paddd_r2r(mm4, mm1);				// 4 ; a1+b1 a0+b0

+

+	 paddd_r2r(mm2, mm0);				// 2 ; a3=sum(even3) a2=sum(even2)

+	 psrad_i2r(SHIFT_INV_ROW, mm1);		// y1=a1+b1 y0=a0+b0

+

+	 paddd_r2r(mm6, mm5);				// 6 ; b3=sum(odd3) b2=sum(odd2)

+	 movq_r2r(mm0, mm4);				// 4 ; a3 a2

+

+	 paddd_r2r(mm5, mm0);				// a3+b3 a2+b2

+	 psubd_r2r(mm5, mm4);				// 5 ; a3-b3 a2-b2

+

+	 psrad_i2r(SHIFT_INV_ROW, mm4);		// y4=a3-b3 y5=a2-b2

+	 psrad_i2r(SHIFT_INV_ROW, mm0);		// y3=a3+b3 y2=a2+b2

+

+	 packssdw_r2r(mm3, mm4);				// 3 ; y6 y7 y4 y5

+

+	 packssdw_r2r(mm0, mm1);				// 0 ; y3 y2 y1 y0

+	 movq_r2r(mm4, mm7);				// 7 ; y6 y7 y4 y5

+

+	 psrld_i2r(16, mm4);					// 0 y6 0 y4

+

+	 movq_r2m(mm1, *(out));	// 1 ; save y3 y2 y1 y0

+	 pslld_i2r(16, mm7);					// y7 0 y5 0

+

+	 por_r2r(mm4, mm7);					// 4 ; y7 y6 y5 y4

+

+   // begin processing row 1

+	 movq_r2m(mm7, *(out+4));	// 7 ; save y7 y6 y5 y4

+

+	 inptr += 8;

+	 out += 8;

+  }

+	 

+

+	// done with the iDCT row-transformation

+

+	// now we have to transpose the output 8x8 matrix

+	// 8x8 (OUT) -> 8x8't' (IN)

+	// the transposition is implemented as 4 sub-operations.

+	// 1) transpose upper-left quad

+	// 2) transpose lower-right quad

+	// 3) transpose lower-left quad

+	// 4) transpose upper-right quad

+

+ 

+	// mm0 = 1st row [ A B C D ] row1

+	// mm1 = 2nd row [ E F G H ] 2

+	// mm2 = 3rd row [ I J K L ] 3

+	// mm3 = 4th row [ M N O P ] 4

+

+	// 1) transpose upper-left quad

+  out = &qwTemp[0];

+

+  movq_m2r(*(out + ROW_STRIDE * 0), mm0);

+

+  movq_m2r(*(out + ROW_STRIDE * 1), mm1);

+  movq_r2r(mm0, mm4);	// mm4 = copy of row1[A B C D]

+	

+  movq_m2r(*(out + ROW_STRIDE * 2), mm2);

+  punpcklwd_r2r(mm1, mm0); // mm0 = [ 0 4 1 5]

+	

+  movq_m2r(*(out + ROW_STRIDE * 3), mm3);

+  punpckhwd_r2r(mm1, mm4); // mm4 = [ 2 6 3 7]

+

+  movq_r2r(mm2, mm6);

+  punpcklwd_r2r(mm3, mm2);	// mm2 = [ 8 12 9 13]

+

+  punpckhwd_r2r(mm3, mm6);	// mm6 = 10 14 11 15]

+  movq_r2r(mm0, mm1);	// mm1 = [ 0 4 1 5]

+

+  inptr = blk;

+

+  punpckldq_r2r(mm2, mm0);	// final result mm0 = row1 [0 4 8 12]

+

+  movq_r2r(mm4, mm3);	// mm3 = [ 2 6 3 7]

+  punpckhdq_r2r(mm2, mm1); // mm1 = final result mm1 = row2 [1 5 9 13]

+

+  movq_r2m(mm0, *(inptr + ROW_STRIDE * 0)); // store row 1

+  punpckldq_r2r(mm6, mm4); // final result mm4 = row3 [2 6 10 14]

+

+// begin reading next quadrant (lower-right)

+  movq_m2r(*(out + ROW_STRIDE*4 + 4), mm0); 

+  punpckhdq_r2r(mm6, mm3); // final result mm3 = row4 [3 7 11 15]

+

+  movq_r2m(mm4, *(inptr + ROW_STRIDE * 2)); // store row 3

+  movq_r2r(mm0, mm4);	// mm4 = copy of row1[A B C D]

+

+  movq_r2m(mm1, *(inptr + ROW_STRIDE * 1)); // store row 2

+

+  movq_m2r(*(out + ROW_STRIDE*5 + 4), mm1);

+

+  movq_r2m(mm3, *(inptr + ROW_STRIDE * 3)); // store row 4

+  punpcklwd_r2r(mm1, mm0); // mm0 = [ 0 4 1 5]

+

+	// 2) transpose lower-right quadrant

+

+//	movq mm0, qword ptr [OUT + ROW_STRIDE*4 + 8]

+

+//	movq mm1, qword ptr [OUT + ROW_STRIDE*5 + 8]

+//	 movq mm4, mm0;	// mm4 = copy of row1[A B C D]

+	

+  movq_m2r(*(out + ROW_STRIDE*6 + 4), mm2);

+//	 punpcklwd mm0, mm1; // mm0 = [ 0 4 1 5]

+  punpckhwd_r2r(mm1, mm4); // mm4 = [ 2 6 3 7]

+	

+  movq_m2r(*(out + ROW_STRIDE*7 + 4), mm3);

+  movq_r2r(mm2, mm6);

+

+  punpcklwd_r2r(mm3, mm2);	// mm2 = [ 8 12 9 13]

+  movq_r2r(mm0, mm1);	// mm1 = [ 0 4 1 5]

+

+  punpckhwd_r2r(mm3, mm6);	// mm6 = 10 14 11 15]

+  movq_r2r(mm4, mm3);	// mm3 = [ 2 6 3 7]

+

+  punpckldq_r2r(mm2, mm0);	// final result mm0 = row1 [0 4 8 12]

+

+  punpckhdq_r2r(mm2, mm1); // mm1 = final result mm1 = row2 [1 5 9 13]

+  ; // slot

+

+  movq_r2m(mm0, *(inptr + ROW_STRIDE*4 + 4)); // store row 1

+  punpckldq_r2r(mm6, mm4); // final result mm4 = row3 [2 6 10 14]

+

+  movq_m2r(*(out + ROW_STRIDE * 4 ), mm0);

+  punpckhdq_r2r(mm6, mm3); // final result mm3 = row4 [3 7 11 15]

+  

+  movq_r2m(mm4, *(inptr + ROW_STRIDE*6 + 4)); // store row 3

+  movq_r2r(mm0, mm4);	// mm4 = copy of row1[A B C D]

+

+  movq_r2m(mm1, *(inptr + ROW_STRIDE*5 + 4)); // store row 2

+  ; // slot

+  

+  movq_m2r(*(out + ROW_STRIDE * 5 ), mm1);

+  ; // slot

+

+  movq_r2m(mm3, *(inptr + ROW_STRIDE*7 + 4)); // store row 4

+  punpcklwd_r2r(mm1, mm0); // mm0 = [ 0 4 1 5]

+

+  // 3) transpose lower-left

+//	movq mm0, qword ptr [OUT + ROW_STRIDE * 4 ]

+

+//	movq mm1, qword ptr [OUT + ROW_STRIDE * 5 ]

+//	 movq mm4, mm0;	// mm4 = copy of row1[A B C D]

+	

+  movq_m2r(*(out + ROW_STRIDE * 6 ), mm2);

+//	 punpcklwd mm0, mm1; // mm0 = [ 0 4 1 5]

+  punpckhwd_r2r(mm1, mm4); // mm4 = [ 2 6 3 7]

+	

+  movq_m2r(*(out + ROW_STRIDE * 7 ), mm3);

+  movq_r2r(mm2, mm6);

+

+  punpcklwd_r2r(mm3, mm2);	// mm2 = [ 8 12 9 13]

+  movq_r2r(mm0, mm1);	// mm1 = [ 0 4 1 5]

+

+  punpckhwd_r2r(mm3, mm6);	// mm6 = 10 14 11 15]

+  movq_r2r(mm4, mm3);	// mm3 = [ 2 6 3 7]

+

+  punpckldq_r2r(mm2, mm0);	// final result mm0 = row1 [0 4 8 12]

+

+  punpckhdq_r2r(mm2, mm1); // mm1 = final result mm1 = row2 [1 5 9 13]

+  ;//slot

+

+  movq_r2m(mm0, *(inptr + ROW_STRIDE * 0 + 4 )); // store row 1

+  punpckldq_r2r(mm6, mm4); // final result mm4 = row3 [2 6 10 14]

+

+// begin reading next quadrant (upper-right)

+  movq_m2r(*(out + ROW_STRIDE*0 + 4), mm0);

+  punpckhdq_r2r(mm6, mm3); // final result mm3 = row4 [3 7 11 15]

+

+  movq_r2m(mm4, *(inptr + ROW_STRIDE * 2 + 4)); // store row 3

+  movq_r2r(mm0, mm4);	// mm4 = copy of row1[A B C D]

+

+  movq_r2m(mm1, *(inptr + ROW_STRIDE * 1 + 4)); // store row 2

+  movq_m2r(*(out + ROW_STRIDE*1 + 4), mm1);

+

+  movq_r2m(mm3, *(inptr + ROW_STRIDE * 3 + 4)); // store row 4

+  punpcklwd_r2r(mm1, mm0); // mm0 = [ 0 4 1 5]

+

+

+	// 2) transpose lower-right quadrant

+

+//	movq mm0, qword ptr [OUT + ROW_STRIDE*4 + 8]

+

+//	movq mm1, qword ptr [OUT + ROW_STRIDE*5 + 8]

+//	 movq mm4, mm0;	// mm4 = copy of row1[A B C D]

+	

+  movq_m2r(*(out + ROW_STRIDE*2 + 4), mm2);

+//	 punpcklwd mm0, mm1; // mm0 = [ 0 4 1 5]

+  punpckhwd_r2r(mm1, mm4); // mm4 = [ 2 6 3 7]

+	

+  movq_m2r(*(out + ROW_STRIDE*3 + 4), mm3);

+  movq_r2r(mm2, mm6);

+

+  punpcklwd_r2r(mm3, mm2);	// mm2 = [ 8 12 9 13]

+  movq_r2r(mm0, mm1);	// mm1 = [ 0 4 1 5]

+

+  punpckhwd_r2r(mm3, mm6);	// mm6 = 10 14 11 15]

+  movq_r2r(mm4, mm3);	// mm3 = [ 2 6 3 7]

+

+  punpckldq_r2r(mm2, mm0);	// final result mm0 = row1 [0 4 8 12]

+

+  punpckhdq_r2r(mm2, mm1); // mm1 = final result mm1 = row2 [1 5 9 13]

+  ; // slot

+

+  movq_r2m(mm0, *(inptr + ROW_STRIDE*4)); // store row 1

+  punpckldq_r2r(mm6, mm4); // final result mm4 = row3 [2 6 10 14]

+

+  movq_r2m(mm1, *(inptr + ROW_STRIDE*5)); // store row 2

+  punpckhdq_r2r(mm6, mm3); // final result mm3 = row4 [3 7 11 15]

+

+  movq_r2m(mm4, *(inptr + ROW_STRIDE*6)); // store row 3

+  ; // slot

+

+  movq_r2m(mm3, *(inptr + ROW_STRIDE*7)); // store row 4

+  ; // slot

+ 

+}

+

+

+static void 

+idct_mmx32_cols( short *blk )	// transform all 8 cols of 8x8 iDCT block

+{

+  int x;

+  short *inptr = blk;

+

+	// Despite the function's name, the matrix is transformed

+	// row by row.  This function is identical to idct_mmx32_rows(),

+	// except for the SHIFT amount and ROUND_INV amount.

+

+	// this subroutine performs two operations

+	// 1) iDCT row transform

+	//		for( i = 0; i < 8; ++ i)

+	//			DCT_8_INV_ROW_1( blk[i*8], qwTemp[i] );

+	//

+	// 2) transpose the matrix (which was stored in qwTemp[])

+	//        qwTemp[] -> [8x8 matrix transpose] -> blk[]

+

+

+  for (x=0; x<8; x++) {  // transform one row per iteration

+

+    movq_m2r(*(inptr), mm0);		// 0 ; x3 x2 x1 x0

+

+    movq_m2r(*(inptr+4), mm1);	// 1 ; x7 x6 x5 x4

+	 movq_r2r(mm0, mm2);				// 2 ; x3 x2 x1 x0

+

+	 movq_m2r(*(tab_i_01234567), mm3);	// 3 ; w06 w04 w02 w00

+	 punpcklwd_r2r(mm1, mm0);			// x5 x1 x4 x0

+

+// ----------

+	 movq_r2r(mm0, mm5);					// 5 ; x5 x1 x4 x0

+	 punpckldq_r2r(mm0, mm0);			// x4 x0 x4 x0

+

+	 movq_m2r(*(tab_i_01234567+4), mm4);	// 4 ; w07 w05 w03 w01

+	 punpckhwd_r2r(mm1, mm2);			// 1 ; x7 x3 x6 x2

+

+	 pmaddwd_r2r(mm0, mm3);				// x4*w06+x0*w04 x4*w02+x0*w00

+	 movq_r2r(mm2, mm6);				// 6 ; x7 x3 x6 x2

+

+	 movq_m2r(*(tab_i_01234567+16), mm1);// 1 ; w22 w20 w18 w16

+	 punpckldq_r2r(mm2, mm2);			// x6 x2 x6 x2

+

+	 pmaddwd_r2r(mm2, mm4);				// x6*w07+x2*w05 x6*w03+x2*w01

+	 punpckhdq_r2r(mm5, mm5);			// x5 x1 x5 x1

+

+	 pmaddwd_m2r(*(tab_i_01234567+8), mm0);// x4*w14+x0*w12 x4*w10+x0*w08

+	 punpckhdq_r2r(mm6, mm6);			// x7 x3 x7 x3

+

+	 movq_m2r(*(tab_i_01234567+20), mm7);// 7 ; w23 w21 w19 w17

+	 pmaddwd_r2r(mm5, mm1);				// x5*w22+x1*w20 x5*w18+x1*w16

+

+	 paddd_m2r(*(r_inv_col), mm3);// +rounder

+	 pmaddwd_r2r(mm6, mm7);				// x7*w23+x3*w21 x7*w19+x3*w17

+

+	 pmaddwd_m2r(*(tab_i_01234567+12), mm2);// x6*w15+x2*w13 x6*w11+x2*w09

+	 paddd_r2r(mm4, mm3);				// 4 ; a1=sum(even1) a0=sum(even0)

+

+	 pmaddwd_m2r(*(tab_i_01234567+24), mm5);// x5*w30+x1*w28 x5*w26+x1*w24

+	 movq_r2r(mm3, mm4);				// 4 ; a1 a0

+

+	 pmaddwd_m2r(*(tab_i_01234567+28), mm6);// x7*w31+x3*w29 x7*w27+x3*w25

+	 paddd_r2r(mm7, mm1);				// 7 ; b1=sum(odd1) b0=sum(odd0)

+

+	 paddd_m2r(*(r_inv_col), mm0);// +rounder

+	 psubd_r2r(mm1, mm3);				// a1-b1 a0-b0

+

+	 psrad_i2r(SHIFT_INV_COL, mm3);		// y6=a1-b1 y7=a0-b0

+	 paddd_r2r(mm4, mm1);				// 4 ; a1+b1 a0+b0

+

+	 paddd_r2r(mm2, mm0);				// 2 ; a3=sum(even3) a2=sum(even2)

+	 psrad_i2r(SHIFT_INV_COL, mm1);		// y1=a1+b1 y0=a0+b0

+

+	 paddd_r2r(mm6, mm5);				// 6 ; b3=sum(odd3) b2=sum(odd2)

+	 movq_r2r(mm0, mm4);				// 4 ; a3 a2

+

+	 paddd_r2r(mm5, mm0);				// a3+b3 a2+b2

+	 psubd_r2r(mm5, mm4);				// 5 ; a3-b3 a2-b2

+

+

+	 psrad_i2r(SHIFT_INV_COL, mm4);		// y4=a3-b3 y5=a2-b2

+	 psrad_i2r(SHIFT_INV_COL, mm0);		// y3=a3+b3 y2=a2+b2

+

+	 packssdw_r2r(mm3, mm4);				// 3 ; y6 y7 y4 y5

+

+	 packssdw_r2r(mm0, mm1);				// 0 ; y3 y2 y1 y0

+	 movq_r2r(mm4, mm7);				// 7 ; y6 y7 y4 y5

+

+	 psrld_i2r(16, mm4);					// 0 y6 0 y4

+

+	 movq_r2m(mm1, *(inptr));	// 1 ; save y3 y2 y1 y0

+	 pslld_i2r(16, mm7);					// y7 0 y5 0

+

+	 por_r2r(mm4, mm7);					// 4 ; y7 y6 y5 y4

+

+   // begin processing row 1

+	 movq_r2m(mm7, *(inptr+4));	// 7 ; save y7 y6 y5 y4

+

+	 inptr += 8;

+  }

+  // done with the iDCT column-transformation

+}

+

+//	

+// public interface to MMX32 IDCT 8x8 operation

+//

+void

+gst_idct_mmx32_idct( short *blk )

+{

+	// 1) iDCT row transformation

+	idct_mmx32_rows( blk ); // 1) transform iDCT row, and transpose

+

+	// 2) iDCT column transformation

+	idct_mmx32_cols( blk ); // 2) transform iDCT row, and transpose

+

+	emms();  // restore processor state

+	// all done

+}

diff --git a/gst-libs/gst/idct/mmxidct.S b/gst-libs/gst/idct/mmxidct.S
new file mode 100644
index 00000000..1b15be57
--- /dev/null
+++ b/gst-libs/gst/idct/mmxidct.S
@@ -0,0 +1,740 @@
+/*
+ * the input data is tranposed and each 16 bit element in the 8x8 matrix
+ * is left aligned:
+ * for example in 11...1110000 format
+ * If the iDCT is of I macroblock then 0.5 needs to be added to the;DC Component
+ * (element[0][0] of the matrix)
+ *
+ * Notes:
+ *  - the scratchN variables should be put on the stack to avoid
+ *    reentrancy problems
+ */
+
+#ifdef PIC
+#define pic_offset(a) a@GOTOFF(%ebx)
+#else
+#define pic_offset(a) a
+#endif
+
+/* extrn re_matrix */
+
+.data
+	.align 16
+	.type	 preSC,@object
+preSC:  .short  16384,22725,21407,19266,16384,12873,8867,4520
+        .short  22725,31521,29692,26722,22725,17855,12299,6270
+        .short  21407,29692,27969,25172,21407,16819,11585,5906
+        .short  19266,26722,25172,22654,19266,15137,10426,5315
+        .short  16384,22725,21407,19266,16384,12873,8867,4520
+        .short  12873,17855,16819,15137,25746,20228,13933,7103
+        .short  17734,24598,23170,20853,17734,13933,9597,4892
+        .short  18081,25080,23624,21261,18081,14206,9785,4988
+	.size	 preSC,128
+	.align 8
+	.type	x0005000200010001,@object
+	.size	x0005000200010001,8
+x0005000200010001:
+	.long	0x00010001,0x00050002
+	.align 8
+	.type	x0040000000000000,@object
+	.size	x0040000000000000,8
+x0040000000000000:
+	.long	0, 0x00400000
+	.align 8
+	.type	x5a825a825a825a82,@object
+	.size	x5a825a825a825a82,8
+x5a825a825a825a82:
+	.long	0x5a825a82, 0x5a825a82
+	.align 8
+	.type	x539f539f539f539f,@object
+	.size	x539f539f539f539f,8
+x539f539f539f539f:
+	.long	0x539f539f,0x539f539f
+	.align 8
+	.type	x4546454645464546,@object
+	.size	x4546454645464546,8
+x4546454645464546:
+	.long	0x45464546,0x45464546
+	.align 8
+	.type	x61f861f861f861f8,@object
+	.size	x61f861f861f861f8,8
+x61f861f861f861f8:
+	.long	0x61f861f8,0x61f861f8
+	.type	x0004000000000000,@object
+	.size	x0004000000000000,8
+x0004000000000000:
+	.long	0x00000000,0x00040000
+	.type	x0000000000000004,@object
+	.size	x0000000000000004,8
+x0000000000000004:
+	.long	0x00000004,0x00000000
+	.align 8
+	.type	 scratch1,@object
+	.size	 scratch1,8
+scratch1:
+	.long 0,0
+	.align 8
+	.type	 scratch3,@object
+	.size	 scratch3,8
+scratch3:
+	.long 0,0
+	.align 8
+	.type	 scratch5,@object
+	.size	 scratch5,8
+scratch5:
+	.long 0,0
+	.align 8
+	.type	 scratch7,@object
+	.size	 scratch7,8
+scratch7:
+	.long 0,0
+	.type	 x0,@object
+	.size	 x0,8
+x0:
+	.long 0,0
+	.align 8
+.text
+	.align 4
+.globl gst_idct_mmx_idct
+	.type	 gst_idct_mmx_idct,@function
+gst_idct_mmx_idct:
+	pushl %ebp
+	movl %esp,%ebp
+	pushl %ebx
+	pushl %ecx
+	pushl %edx
+	pushl %esi
+	pushl %edi
+#ifdef PIC
+	call here
+here:	popl %ebx
+	addl $_GLOBAL_OFFSET_TABLE_+[.-here],%ebx
+#endif
+	movl 8(%ebp),%esi		/* source matrix */
+	movq (%esi), %mm0
+	paddw pic_offset(x0000000000000004), %mm0
+	movq 8(%esi), %mm1
+	psllw $4, %mm0
+	movq 16(%esi), %mm2
+	psllw $4, %mm1
+	movq 24(%esi), %mm3
+	psllw $4, %mm2
+	movq 32(%esi), %mm4
+	psllw $4, %mm3
+	movq 40(%esi), %mm5
+	psllw $4, %mm4
+	movq 48(%esi), %mm6
+	psllw $4, %mm5
+	movq 56(%esi), %mm7
+	psllw $4, %mm6
+	psllw $4, %mm7
+	movq %mm0,  (%esi)
+	movq %mm1, 8(%esi)
+	movq %mm2,16(%esi)
+	movq %mm3,24(%esi)
+	movq %mm4,32(%esi)
+	movq %mm5,40(%esi)
+	movq %mm6,48(%esi)
+	movq %mm7,56(%esi)
+	movq 64(%esi), %mm0
+	movq 72(%esi), %mm1
+	psllw $4, %mm0
+	movq 80(%esi), %mm2
+	psllw $4, %mm1
+	movq 88(%esi), %mm3
+	psllw $4, %mm2
+	movq 96(%esi), %mm4
+	psllw $4, %mm3
+	movq 104(%esi), %mm5
+	psllw $4, %mm4
+	movq 112(%esi), %mm6
+	psllw $4, %mm5
+	movq 120(%esi), %mm7
+	psllw $4, %mm6
+	psllw $4, %mm7
+	movq %mm0,64(%esi)
+	movq %mm1,72(%esi)
+	movq %mm2,80(%esi)
+	movq %mm3,88(%esi)
+	movq %mm4,96(%esi)
+	movq %mm5,104(%esi)
+	movq %mm6,112(%esi)
+	movq %mm7,120(%esi)
+	leal pic_offset(preSC), %ecx
+/* column 0: even part
+ * use V4, V12, V0, V8 to produce V22..V25
+ */
+	movq 8*12(%ecx), %mm0	/* maybe the first mul can be done together */
+				/* with the dequantization in iHuff module */
+	pmulhw 8*12(%esi), %mm0		/* V12 */
+	movq 8*4(%ecx), %mm1
+	pmulhw 8*4(%esi), %mm1		/* V4 */
+	movq (%ecx), %mm3
+	psraw $1, %mm0			/* t64=t66 */
+	pmulhw (%esi), %mm3		/* V0 */
+	movq 8*8(%ecx), %mm5		/* duplicate V4 */
+	movq %mm1, %mm2			/* added 11/1/96 */
+	pmulhw 8*8(%esi),%mm5		/* V8 */
+	psubsw %mm0, %mm1		/* V16 */
+	pmulhw pic_offset(x5a825a825a825a82), %mm1	/* 23170 ->V18 */
+	paddsw %mm0, %mm2		/* V17 */
+	movq %mm2, %mm0			/* duplicate V17 */
+	psraw $1, %mm2			/* t75=t82 */
+	psraw $2, %mm0			/* t72 */
+	movq %mm3, %mm4			/* duplicate V0 */
+	paddsw %mm5, %mm3		/* V19 */
+	psubsw %mm5, %mm4		/* V20 ;mm5 free */
+/* moved from the block below */
+	movq 8*10(%ecx), %mm7
+	psraw $1, %mm3			/* t74=t81 */
+	movq %mm3, %mm6			/* duplicate t74=t81 */
+	psraw $2, %mm4			/* t77=t79 */
+	psubsw %mm0, %mm1		/* V21 ; mm0 free */
+	paddsw %mm2, %mm3		/* V22 */
+	movq %mm1, %mm5			/* duplicate V21 */
+	paddsw %mm4, %mm1		/* V23 */
+	movq %mm3, 8*4(%esi)		/* V22 */
+	psubsw %mm5, %mm4		/* V24; mm5 free */
+	movq %mm1, 8*12(%esi)		/* V23 */
+	psubsw %mm2, %mm6		/* V25; mm2 free */
+	movq %mm4, (%esi)		/* V24 */
+/* keep mm6 alive all along the next block */
+	/* movq %mm6, 8*8(%esi) 	V25 */
+/* column 0: odd part
+ * use V2, V6, V10, V14 to produce V31, V39, V40, V41
+ */
+/* moved above: movq 8*10(%ecx), %mm7 */
+
+	pmulhw 8*10(%esi), %mm7		/* V10 */
+	movq 8*6(%ecx), %mm0
+	pmulhw 8*6(%esi), %mm0		/* V6 */
+	movq 8*2(%ecx), %mm5
+	movq %mm7, %mm3			/* duplicate V10 */
+	pmulhw 8*2(%esi), %mm5		/* V2 */
+	movq 8*14(%ecx), %mm4
+	psubsw %mm0, %mm7		/* V26 */
+	pmulhw 8*14(%esi), %mm4		/* V14 */
+	paddsw %mm0, %mm3		/* V29 ; free mm0 */
+	movq %mm7, %mm1			/* duplicate V26 */
+	psraw $1, %mm3			/* t91=t94 */
+	pmulhw pic_offset(x539f539f539f539f),%mm7	/* V33 */
+	psraw $1, %mm1			/* t96 */
+	movq %mm5, %mm0			/* duplicate V2 */
+	psraw $2, %mm4			/* t85=t87 */
+	paddsw %mm4,%mm5		/* V27 */
+	psubsw %mm4, %mm0		/* V28 ; free mm4 */
+	movq %mm0, %mm2			/* duplicate V28 */
+	psraw $1, %mm5			/* t90=t93 */
+	pmulhw pic_offset(x4546454645464546),%mm0	/* V35 */
+	psraw $1, %mm2			/* t97 */
+	movq %mm5, %mm4			/* duplicate t90=t93 */
+	psubsw %mm2, %mm1		/* V32 ; free mm2 */
+	pmulhw pic_offset(x61f861f861f861f8),%mm1	/* V36 */
+	psllw $1, %mm7			/* t107 */
+	paddsw %mm3, %mm5		/* V31 */
+	psubsw %mm3, %mm4		/* V30 ; free mm3 */
+	pmulhw pic_offset(x5a825a825a825a82),%mm4	/* V34 */
+	nop
+	psubsw %mm1, %mm0		/* V38 */
+	psubsw %mm7, %mm1		/* V37 ; free mm7 */
+	psllw $1, %mm1			/* t114 */
+/* move from the next block */
+	movq %mm6, %mm3			/* duplicate V25 */
+/* move from the next block */
+	movq 8*4(%esi), %mm7		/* V22 */
+	psllw $1, %mm0			/* t110 */
+	psubsw %mm5, %mm0		/* V39 (mm5 needed for next block) */
+	psllw $2, %mm4			/* t112 */
+/* moved from the next block */
+	movq 8*12(%esi), %mm2		/* V23 */
+	psubsw %mm0, %mm4		/* V40 */
+	paddsw %mm4, %mm1		/* V41; free mm0 */
+/* moved from the next block */
+	psllw $1, %mm2			/* t117=t125 */
+/* column 0: output butterfly */
+/* moved above:
+ * movq %mm6, %mm3			duplicate V25
+ * movq 8*4(%esi), %mm7			V22
+ * movq 8*12(%esi), %mm2		V23
+ * psllw $1, %mm2			t117=t125
+ */
+	psubsw %mm1, %mm6		/* tm6 */
+	paddsw %mm1, %mm3		/* tm8; free mm1 */
+	movq %mm7, %mm1			/* duplicate V22 */
+	paddsw %mm5, %mm7		/* tm0 */
+	movq %mm3, 8*8(%esi)		/* tm8; free mm3 */
+	psubsw %mm5, %mm1		/* tm14; free mm5 */
+	movq %mm6, 8*6(%esi)		/* tm6; free mm6 */
+	movq %mm2, %mm3			/* duplicate t117=t125 */
+	movq (%esi), %mm6		/* V24 */
+	paddsw %mm0, %mm2		/* tm2 */
+	movq %mm7, (%esi)		/* tm0; free mm7 */
+	psubsw %mm0, %mm3		/* tm12; free mm0 */
+	movq %mm1, 8*14(%esi)		/* tm14; free mm1 */
+	psllw $1, %mm6			/* t119=t123 */
+	movq %mm2, 8*2(%esi)		/* tm2; free mm2 */
+	movq %mm6, %mm0			/* duplicate t119=t123 */
+	movq %mm3, 8*12(%esi)		/* tm12; free mm3 */
+	paddsw %mm4, %mm6		/* tm4 */
+/* moved from next block */
+	movq 8*5(%ecx), %mm1
+	psubsw %mm4, %mm0		/* tm10; free mm4 */
+/* moved from next block */
+	pmulhw 8*5(%esi), %mm1		/* V5 */
+	movq %mm6, 8*4(%esi)		/* tm4; free mm6 */
+	movq %mm0, 8*10(%esi)		/* tm10; free mm0 */
+/* column 1: even part
+ * use V5, V13, V1, V9 to produce V56..V59
+ */
+/* moved to prev block:
+ *	movq 8*5(%ecx), %mm1
+ *	pmulhw 8*5(%esi), %mm1		 V5
+ */
+	movq 8*13(%ecx), %mm7
+	psllw $1, %mm1			/* t128=t130 */
+	pmulhw 8*13(%esi), %mm7		/* V13 */
+	movq %mm1, %mm2			/* duplicate t128=t130 */
+	movq 8(%ecx), %mm3
+	pmulhw 8(%esi), %mm3		/* V1 */
+	movq 8*9(%ecx), %mm5
+	psubsw %mm7, %mm1		/* V50 */
+	pmulhw 8*9(%esi), %mm5		/* V9 */
+	paddsw %mm7, %mm2		/* V51 */
+	pmulhw pic_offset(x5a825a825a825a82), %mm1	/* 23170 ->V52 */
+	movq %mm2, %mm6			/* duplicate V51 */
+	psraw $1, %mm2			/* t138=t144 */
+	movq %mm3, %mm4			/* duplicate V1 */
+	psraw $2, %mm6			/* t136 */
+	paddsw %mm5, %mm3		/* V53 */
+	psubsw %mm5, %mm4		/* V54 ;mm5 free */
+	movq %mm3, %mm7			/* duplicate V53 */
+/* moved from next block */
+	movq 8*11(%ecx), %mm0
+	psraw $1, %mm4			/* t140=t142 */
+	psubsw %mm6, %mm1		/* V55 ; mm6 free */
+	paddsw %mm2, %mm3		/* V56 */
+	movq %mm4, %mm5			/* duplicate t140=t142 */
+	paddsw %mm1, %mm4		/* V57 */
+	movq %mm3, 8*5(%esi)		/* V56 */
+	psubsw %mm1, %mm5		/* V58; mm1 free */
+	movq %mm4, 8*13(%esi)		/* V57 */
+	psubsw %mm2, %mm7		/* V59; mm2 free */
+	movq %mm5, 8*9(%esi)		/* V58 */
+/* keep mm7 alive all along the next block
+ *	movq %mm7, 8(%esi)		V59
+ * moved above
+ *	movq 8*11(%ecx), %mm0
+ */
+	pmulhw 8*11(%esi), %mm0		/* V11 */
+	movq 8*7(%ecx), %mm6
+	pmulhw 8*7(%esi), %mm6		/* V7 */
+	movq 8*15(%ecx), %mm4
+	movq %mm0, %mm3			/* duplicate V11 */
+	pmulhw 8*15(%esi), %mm4		/* V15 */
+	movq 8*3(%ecx), %mm5
+	psllw $1, %mm6			/* t146=t152 */
+	pmulhw 8*3(%esi), %mm5		/* V3 */
+	paddsw %mm6, %mm0		/* V63 */
+/* note that V15 computation has a correction step: 
+ * this is a 'magic' constant that rebiases the results to be closer to the
+ * expected result.  this magic constant can be refined to reduce the error
+ * even more by doing the correction step in a later stage when the number
+ * is actually multiplied by 16
+ */
+	paddw pic_offset(x0005000200010001), %mm4
+	psubsw %mm6, %mm3		/* V60 ; free mm6 */
+	psraw $1, %mm0			/* t154=t156 */
+	movq %mm3, %mm1			/* duplicate V60 */
+	pmulhw pic_offset(x539f539f539f539f), %mm1	/* V67 */
+	movq %mm5, %mm6			/* duplicate V3 */
+	psraw $2, %mm4			/* t148=t150 */
+	paddsw %mm4, %mm5		/* V61 */
+	psubsw %mm4, %mm6		/* V62 ; free mm4 */
+	movq %mm5, %mm4			/* duplicate V61 */
+	psllw $1, %mm1			/* t169 */
+	paddsw %mm0, %mm5		/* V65 -> result */
+	psubsw %mm0, %mm4		/* V64 ; free mm0 */
+	pmulhw pic_offset(x5a825a825a825a82), %mm4	/* V68 */
+	psraw $1, %mm3			/* t158 */
+	psubsw %mm6, %mm3		/* V66 */
+	movq %mm5, %mm2			/* duplicate V65 */
+	pmulhw pic_offset(x61f861f861f861f8), %mm3	/* V70 */
+	psllw $1, %mm6			/* t165 */
+	pmulhw pic_offset(x4546454645464546), %mm6	/* V69 */
+	psraw $1, %mm2			/* t172 */
+/* moved from next block */
+	movq 8*5(%esi), %mm0		/* V56 */
+	psllw $1, %mm4			/* t174 */
+/* moved from next block */
+	psraw $1, %mm0			/* t177=t188 */
+	nop
+	psubsw %mm3, %mm6		/* V72 */
+	psubsw %mm1, %mm3		/* V71 ; free mm1 */
+	psubsw %mm2, %mm6		/* V73 ; free mm2 */
+/* moved from next block */
+	psraw $1, %mm5			/* t178=t189 */
+	psubsw %mm6, %mm4		/* V74 */
+/* moved from next block */
+	movq %mm0, %mm1			/* duplicate t177=t188 */
+	paddsw %mm4, %mm3		/* V75 */
+/* moved from next block */
+	paddsw %mm5, %mm0		/* tm1 */
+/* location
+ *  5 - V56
+ * 13 - V57
+ *  9 - V58
+ *  X - V59, mm7
+ *  X - V65, mm5
+ *  X - V73, mm6
+ *  X - V74, mm4
+ *  X - V75, mm3
+ * free mm0, mm1 & mm2
+ * moved above
+ *	movq 8*5(%esi), %mm0		V56
+ *	psllw $1, %mm0			t177=t188 ! new !!
+ *	psllw $1, %mm5			t178=t189 ! new !!
+ *	movq %mm0, %mm1			duplicate t177=t188
+ *	paddsw %mm5, %mm0		tm1
+ */
+	movq 8*13(%esi), %mm2		/* V57 */
+	psubsw %mm5, %mm1		/* tm15; free mm5 */
+	movq %mm0, 8(%esi)		/* tm1; free mm0 */
+	psraw $1, %mm7			/* t182=t184 ! new !! */
+/* save the store as used directly in the transpose
+ *	movq %mm1, 120(%esi)		tm15; free mm1
+ */
+	movq %mm7, %mm5			/* duplicate t182=t184 */
+	psubsw %mm3, %mm7		/* tm7 */
+	paddsw %mm3, %mm5		/* tm9; free mm3 */
+	movq 8*9(%esi), %mm0		/* V58 */
+	movq %mm2, %mm3			/* duplicate V57 */
+	movq %mm7, 8*7(%esi)		/* tm7; free mm7 */
+	psubsw %mm6, %mm3		/* tm13 */
+	paddsw %mm6, %mm2		/* tm3 ; free mm6 */
+/* moved up from the transpose */
+	movq %mm3, %mm7
+/* moved up from the transpose */
+	punpcklwd %mm1, %mm3
+	movq %mm0, %mm6			/* duplicate V58 */
+	movq %mm2, 8*3(%esi)		/* tm3; free mm2 */
+	paddsw %mm4, %mm0		/* tm5 */
+	psubsw %mm4, %mm6		/* tm11; free mm4 */
+/* moved up from the transpose */
+	punpckhwd %mm1, %mm7
+	movq %mm0, 8*5(%esi)		/* tm5; free mm0 */
+/* moved up from the transpose */
+	movq %mm5, %mm2
+/* transpose - M4 part
+ *  ---------       ---------
+ * | M1 | M2 |     | M1'| M3'|
+ *  ---------  -->  ---------
+ * | M3 | M4 |     | M2'| M4'|
+ *  ---------       ---------
+ * Two alternatives: use full mmword approach so the following code can be
+ * scheduled before the transpose is done without stores, or use the faster
+ * half mmword stores (when possible)
+ */
+	movd %mm3, 8*9+4(%esi)		/* MS part of tmt9 */
+	punpcklwd %mm6, %mm5
+	movd %mm7, 8*13+4(%esi)		/* MS part of tmt13 */
+	punpckhwd %mm6, %mm2
+	movd %mm5, 8*9(%esi)		/* LS part of tmt9 */
+	punpckhdq %mm3, %mm5		/* free mm3 */
+	movd %mm2, 8*13(%esi)		/* LS part of tmt13 */
+	punpckhdq %mm7, %mm2		/* free mm7 */
+/* moved up from the M3 transpose */
+	movq 8*8(%esi), %mm0
+/* moved up from the M3 transpose */
+	movq 8*10(%esi), %mm1
+/* moved up from the M3 transpose */
+	movq %mm0, %mm3
+/* shuffle the rest of the data, and write it with 2 mmword writes */
+	movq %mm5, 8*11(%esi)		/* tmt11 */
+/* moved up from the M3 transpose */
+	punpcklwd %mm1, %mm0
+	movq %mm2, 8*15(%esi)		/* tmt15 */
+/* moved up from the M3 transpose */
+	punpckhwd %mm1, %mm3
+/* transpose - M3 part
+ * moved up to previous code section
+ *	movq 8*8(%esi), %mm0
+ *	movq 8*10(%esi), %mm1
+ *	movq %mm0, %mm3
+ *	punpcklwd %mm1, %mm0
+ *	punpckhwd %mm1, %mm3
+ */
+	movq 8*12(%esi), %mm6
+	movq 8*14(%esi), %mm4
+	movq %mm6, %mm2
+/* shuffle the data and write the lower parts of the transposed in 4 dwords */
+	punpcklwd %mm4, %mm6
+	movq %mm0, %mm1
+	punpckhdq %mm6, %mm1
+	movq %mm3, %mm7
+	punpckhwd %mm4, %mm2		/* free mm4 */
+	punpckldq %mm6, %mm0		/* free mm6 */
+/* moved from next block */
+	movq 8*13(%esi), %mm4		/* tmt13 */
+	punpckldq %mm2, %mm3
+	punpckhdq %mm2, %mm7		/* free mm2 */
+/* moved from next block */
+	movq %mm3, %mm5			/* duplicate tmt5 */
+/* column 1: even part (after transpose)
+* moved above
+*	movq %mm3, %mm5			duplicate tmt5
+*	movq 8*13(%esi), %mm4		tmt13
+*/
+	psubsw %mm4, %mm3		/* V134 */
+	pmulhw pic_offset(x5a825a825a825a82), %mm3	/* 23170 ->V136 */
+	movq 8*9(%esi), %mm6		/* tmt9 */
+	paddsw %mm4, %mm5		/* V135 ; mm4 free */
+	movq %mm0, %mm4			/* duplicate tmt1 */
+	paddsw %mm6, %mm0		/* V137 */
+	psubsw %mm6, %mm4		/* V138 ; mm6 free */
+	psllw $2, %mm3			/* t290 */
+	psubsw %mm5, %mm3		/* V139 */
+	movq %mm0, %mm6			/* duplicate V137 */
+	paddsw %mm5, %mm0		/* V140 */
+	movq %mm4, %mm2			/* duplicate V138 */
+	paddsw %mm3, %mm2		/* V141 */
+	psubsw %mm3, %mm4		/* V142 ; mm3 free */
+	movq %mm0, 8*9(%esi)		/* V140 */
+	psubsw %mm5, %mm6		/* V143 ; mm5 free */
+/* moved from next block */
+	movq 8*11(%esi), %mm0		/* tmt11 */
+	movq %mm2, 8*13(%esi)		/* V141 */
+/* moved from next block */
+	movq %mm0, %mm2			/* duplicate tmt11 */
+/* column 1: odd part (after transpose) */
+/* moved up to the prev block
+ *	movq 8*11(%esi), %mm0		tmt11
+ *	movq %mm0, %mm2			duplicate tmt11
+ */
+	movq 8*15(%esi), %mm5		/* tmt15 */
+	psubsw %mm7, %mm0		/* V144 */
+	movq %mm0, %mm3			/* duplicate V144 */
+	paddsw %mm7, %mm2		/* V147 ; free mm7 */
+	pmulhw pic_offset(x539f539f539f539f), %mm0	/* 21407-> V151 */
+	movq %mm1, %mm7			/* duplicate tmt3 */
+	paddsw %mm5, %mm7		/* V145 */
+	psubsw %mm5, %mm1		/* V146 ; free mm5 */
+	psubsw %mm1, %mm3		/* V150 */
+	movq %mm7, %mm5			/* duplicate V145 */
+	pmulhw pic_offset(x4546454645464546), %mm1	/* 17734-> V153 */
+	psubsw %mm2, %mm5		/* V148 */
+	pmulhw pic_offset(x61f861f861f861f8), %mm3	/* 25080-> V154 */
+	psllw $2, %mm0			/* t311 */
+	pmulhw pic_offset(x5a825a825a825a82), %mm5	/* 23170-> V152 */
+	paddsw %mm2, %mm7		/* V149 ; free mm2 */
+	psllw $1, %mm1			/* t313 */
+	nop	/* without the nop - freeze here for one clock */
+	movq %mm3, %mm2			/* duplicate V154 */
+	psubsw %mm0, %mm3		/* V155 ; free mm0 */
+	psubsw %mm2, %mm1		/* V156 ; free mm2 */
+/* moved from the next block */
+	movq %mm6, %mm2			/* duplicate V143 */
+/* moved from the next block */
+	movq 8*13(%esi), %mm0		/* V141 */
+	psllw $1, %mm1			/* t315 */
+	psubsw %mm7, %mm1		/* V157 (keep V149) */
+	psllw $2, %mm5			/* t317 */
+	psubsw %mm1, %mm5		/* V158 */
+	psllw $1, %mm3			/* t319 */
+	paddsw %mm5, %mm3		/* V159 */
+/* column 1: output butterfly (after transform)
+ * moved to the prev block
+ *	movq %mm6, %mm2			duplicate V143
+ *	movq 8*13(%esi), %mm0		V141
+ */
+	psubsw %mm3, %mm2		/* V163 */
+	paddsw %mm3, %mm6		/* V164 ; free mm3 */
+	movq %mm4, %mm3			/* duplicate V142 */
+	psubsw %mm5, %mm4		/* V165 ; free mm5 */
+	movq %mm2, pic_offset(scratch7)		/* out7 */
+	psraw $4, %mm6
+	psraw $4, %mm4
+	paddsw %mm5, %mm3		/* V162 */
+	movq 8*9(%esi), %mm2		/* V140 */
+	movq %mm0, %mm5			/* duplicate V141 */
+/* in order not to perculate this line up,
+ * we read 72(%esi) very near to this location
+ */
+	movq %mm6, 8*9(%esi)		/* out9 */
+	paddsw %mm1, %mm0		/* V161 */
+	movq %mm3, pic_offset(scratch5)		/* out5 */
+	psubsw %mm1, %mm5		/* V166 ; free mm1 */
+	movq %mm4, 8*11(%esi)		/* out11 */
+	psraw $4, %mm5
+	movq %mm0, pic_offset(scratch3)		/* out3 */
+	movq %mm2, %mm4			/* duplicate V140 */
+	movq %mm5, 8*13(%esi)		/* out13 */
+	paddsw %mm7, %mm2		/* V160 */
+/* moved from the next block */
+	movq 8(%esi), %mm0
+	psubsw %mm7, %mm4		/* V167 ; free mm7 */
+/* moved from the next block */
+	movq 8*3(%esi), %mm7
+	psraw $4, %mm4
+	movq %mm2, pic_offset(scratch1)		/* out1 */
+/* moved from the next block */
+	movq %mm0, %mm1
+	movq %mm4, 8*15(%esi)		/* out15 */
+/* moved from the next block */
+	punpcklwd %mm7, %mm0
+/* transpose - M2 parts
+ * moved up to the prev block
+ *	movq 8(%esi), %mm0
+ *	movq 8*3(%esi), %mm7
+ *	movq %mm0, %mm1
+ *	punpcklwd %mm7, %mm0
+ */
+	movq 8*5(%esi), %mm5
+	punpckhwd %mm7, %mm1
+	movq 8*7(%esi), %mm4
+	movq %mm5, %mm3
+/* shuffle the data and write the lower parts of the trasposed in 4 dwords */
+	movd %mm0, 8*8(%esi)		/* LS part of tmt8 */
+	punpcklwd %mm4, %mm5
+	movd %mm1, 8*12(%esi)		/* LS part of tmt12 */
+	punpckhwd %mm4, %mm3
+	movd %mm5, 8*8+4(%esi)		/* MS part of tmt8 */
+	punpckhdq %mm5, %mm0		/* tmt10 */
+	movd %mm3, 8*12+4(%esi)		/* MS part of tmt12 */
+	punpckhdq %mm3, %mm1		/* tmt14 */
+/* transpose - M1 parts */
+	movq (%esi), %mm7
+	movq 8*2(%esi), %mm2
+	movq %mm7, %mm6
+	movq 8*4(%esi), %mm5
+	punpcklwd %mm2, %mm7
+	movq 8*6(%esi), %mm4
+	punpckhwd %mm2, %mm6		/* free mm2 */
+	movq %mm5, %mm3
+	punpcklwd %mm4, %mm5
+	punpckhwd %mm4, %mm3		/* free mm4 */
+	movq %mm7, %mm2
+	movq %mm6, %mm4
+	punpckldq %mm5, %mm7		/* tmt0 */
+	punpckhdq %mm5, %mm2		/* tmt2 ; free mm5 */
+/* shuffle the rest of the data, and write it with 2 mmword writes */
+	punpckldq %mm3, %mm6		/* tmt4 */
+/* moved from next block */
+	movq %mm2, %mm5			/* duplicate tmt2 */
+	punpckhdq %mm3, %mm4		/* tmt6 ; free mm3 */
+/* moved from next block */
+	movq %mm0, %mm3			/* duplicate tmt10 */
+/* column 0: odd part (after transpose)
+ *moved up to prev block
+ *	movq %mm0, %mm3			duplicate tmt10
+ *	movq %mm2, %mm5			duplicate tmt2
+ */
+	psubsw %mm4, %mm0		/* V110 */
+	paddsw %mm4, %mm3		/* V113 ; free mm4 */
+	movq %mm0, %mm4			/* duplicate V110 */
+	paddsw %mm1, %mm2		/* V111 */
+	pmulhw pic_offset(x539f539f539f539f), %mm0	/* 21407-> V117 */
+	psubsw %mm1, %mm5		/* V112 ; free mm1 */
+	psubsw %mm5, %mm4		/* V116 */
+	movq %mm2, %mm1			/* duplicate V111 */
+	pmulhw pic_offset(x4546454645464546), %mm5	/* 17734-> V119 */
+	psubsw %mm3, %mm2		/* V114 */
+	pmulhw pic_offset(x61f861f861f861f8), %mm4	/* 25080-> V120 */
+	paddsw %mm3, %mm1		/* V115 ; free mm3 */
+	pmulhw pic_offset(x5a825a825a825a82), %mm2	/* 23170-> V118 */
+	psllw $2, %mm0			/* t266 */
+	movq %mm1, (%esi)		/* save V115 */
+	psllw $1, %mm5			/* t268 */
+	psubsw %mm4, %mm5		/* V122 */
+	psubsw %mm0, %mm4		/* V121 ; free mm0 */
+	psllw $1, %mm5			/* t270 */
+	psubsw %mm1, %mm5		/* V123 ; free mm1 */
+	psllw $2, %mm2			/* t272 */
+	psubsw %mm5, %mm2		/* V124 (keep V123) */
+	psllw $1, %mm4			/* t274 */
+	movq %mm5, 8*2(%esi)		/* save V123 ; free mm5 */
+	paddsw %mm2, %mm4		/* V125 (keep V124) */
+/* column 0: even part (after transpose) */
+	movq 8*12(%esi), %mm0		/* tmt12 */
+	movq %mm6, %mm3			/* duplicate tmt4 */
+	psubsw %mm0, %mm6		/* V100 */
+	paddsw %mm0, %mm3		/* V101 ; free mm0 */
+	pmulhw pic_offset(x5a825a825a825a82), %mm6	/* 23170 ->V102 */
+	movq %mm7, %mm5			/* duplicate tmt0 */
+	movq 8*8(%esi), %mm1		/* tmt8 */
+	paddsw %mm1, %mm7		/* V103 */
+	psubsw %mm1, %mm5		/* V104 ; free mm1 */
+	movq %mm7, %mm0			/* duplicate V103 */
+	psllw $2, %mm6			/* t245 */
+	paddsw %mm3, %mm7		/* V106 */
+	movq %mm5, %mm1			/* duplicate V104 */
+	psubsw %mm3, %mm6		/* V105 */
+	psubsw %mm3, %mm0		/* V109; free mm3 */
+	paddsw %mm6, %mm5		/* V107 */
+	psubsw %mm6, %mm1		/* V108 ; free mm6 */
+/* column 0: output butterfly (after transform) */
+	movq %mm1, %mm3			/* duplicate V108 */
+	paddsw %mm2, %mm1		/* out4 */
+	psraw $4, %mm1
+	psubsw %mm2, %mm3		/* out10 ; free mm2 */
+	psraw $4, %mm3
+	movq %mm0, %mm6			/* duplicate V109 */
+	movq %mm1, 8*4(%esi)		/* out4 ; free mm1 */
+	psubsw %mm4, %mm0		/* out6 */
+	movq %mm3, 8*10(%esi)		/* out10 ; free mm3 */
+	psraw $4, %mm0
+	paddsw %mm4, %mm6		/* out8 ; free mm4 */
+	movq %mm7, %mm1			/* duplicate V106 */
+	movq %mm0, 8*6(%esi)		/* out6 ; free mm0 */
+	psraw $4, %mm6
+	movq (%esi), %mm4		/* V115 */
+	movq %mm6, 8*8(%esi)		/* out8 ; free mm6 */
+	movq %mm5, %mm2			/* duplicate V107 */
+	movq 8*2(%esi), %mm3		/* V123 */
+	paddsw %mm4, %mm7		/* out0 */
+/* moved up from next block */
+	movq pic_offset(scratch3), %mm0
+	psraw $4, %mm7
+/* moved up from next block */
+	movq pic_offset(scratch5), %mm6 
+	psubsw %mm4, %mm1		/* out14 ; free mm4 */
+	paddsw %mm3, %mm5		/* out2 */
+	psraw $4, %mm1
+	movq %mm7, (%esi)		/* out0 ; free mm7 */
+	psraw $4, %mm5
+	movq %mm1, 8*14(%esi)		/* out14 ; free mm1 */
+	psubsw %mm3, %mm2		/* out12 ; free mm3 */
+	movq %mm5, 8*2(%esi)		/* out2 ; free mm5 */
+	psraw $4, %mm2
+/* moved up to the prev block */
+	movq pic_offset(scratch7), %mm4
+/* moved up to the prev block */
+	psraw $4, %mm0
+	movq %mm2, 8*12(%esi)		/* out12 ; free mm2 */
+/* moved up to the prev block */
+	psraw $4, %mm6
+/* move back the data to its correct place
+* moved up to the prev block
+ *	movq pic_offset(scratch3), %mm0
+ *	movq pic_offset(scratch5), %mm6
+ *	movq pic_offset(scratch7), %mm4
+ *	psraw $4, %mm0
+ *	psraw $4, %mm6
+*/
+	movq pic_offset(scratch1), %mm1
+	psraw $4, %mm4
+	movq %mm0, 8*3(%esi)		/* out3 */
+	psraw $4, %mm1
+	movq %mm6, 8*5(%esi)		/* out5 */
+	movq %mm4, 8*7(%esi)		/* out7 */
+	movq %mm1, 8(%esi)		/* out1 */
+	emms
+	popl %edi
+	popl %esi
+	popl %edx
+	popl %ecx
+	popl %ebx
+	movl %ebp,%esp
+	popl %ebp
+	ret
+.Lfe1:
+	.size	 gst_idct_mmx_idct,.Lfe1-gst_idct_mmx_idct
diff --git a/gst-libs/gst/idct/sseidct.S b/gst-libs/gst/idct/sseidct.S
new file mode 100644
index 00000000..99cda4f2
--- /dev/null
+++ b/gst-libs/gst/idct/sseidct.S
@@ -0,0 +1,740 @@
+.data
+	.align 4
+	.type	 rounder0,@object
+rounder0:
+	.long 65536
+	.long 65536
+	.size	 rounder0,8
+	.align 4
+	.type	 rounder4,@object
+rounder4:
+	.long 1024
+	.long 1024
+	.size	 rounder4,8
+	.align 4
+	.type	 rounder1,@object
+rounder1:
+	.long 3597
+	.long 3597
+	.size	 rounder1,8
+	.align 4
+	.type	 rounder7,@object
+rounder7:
+	.long 512
+	.long 512
+	.size	 rounder7,8
+	.align 4
+	.type	 rounder2,@object
+rounder2:
+	.long 2260
+	.long 2260
+	.size	 rounder2,8
+	.align 4
+	.type	 rounder6,@object
+rounder6:
+	.long 512
+	.long 512
+	.size	 rounder6,8
+	.align 4
+	.type	 rounder3,@object
+rounder3:
+	.long 1203
+	.long 1203
+	.size	 rounder3,8
+	.align 4
+	.type	 rounder5,@object
+rounder5:
+	.long 120
+	.long 120
+	.size	 rounder5,8
+	.align 2
+	.type	 _T1.46,@object
+_T1.46:
+	.value 13036
+	.value 13036
+	.value 13036
+	.value 13036
+	.align 2
+	.type	 _T2.47,@object
+_T2.47:
+	.value 27146
+	.value 27146
+	.value 27146
+	.value 27146
+	.align 2
+	.type	 _T3.48,@object
+_T3.48:
+	.value -21746
+	.value -21746
+	.value -21746
+	.value -21746
+	.align 2
+	.type	 _C4.49,@object
+_C4.49:
+	.value 23170
+	.value 23170
+	.value 23170
+	.value 23170
+	.local	scratch0.50
+	.comm	scratch0.50,8,4
+	.local	scratch1.51
+	.comm	scratch1.51,8,4
+	.align 2
+	.type	 table04.54,@object
+table04.54:
+	.value 16384
+	.value 21407
+	.value -16384
+	.value -21407
+	.value 16384
+	.value 8867
+	.value 16384
+	.value 8867
+	.value 22725
+	.value 19266
+	.value -22725
+	.value -12873
+	.value 12873
+	.value 4520
+	.value 19266
+	.value -4520
+	.value 16384
+	.value -8867
+	.value 16384
+	.value -8867
+	.value -16384
+	.value 21407
+	.value 16384
+	.value -21407
+	.value 12873
+	.value -22725
+	.value 19266
+	.value -22725
+	.value 4520
+	.value 19266
+	.value 4520
+	.value -12873
+	.align 2
+	.type	 table17.55,@object
+table17.55:
+	.value 22725
+	.value 29692
+	.value -22725
+	.value -29692
+	.value 22725
+	.value 12299
+	.value 22725
+	.value 12299
+	.value 31521
+	.value 26722
+	.value -31521
+	.value -17855
+	.value 17855
+	.value 6270
+	.value 26722
+	.value -6270
+	.value 22725
+	.value -12299
+	.value 22725
+	.value -12299
+	.value -22725
+	.value 29692
+	.value 22725
+	.value -29692
+	.value 17855
+	.value -31521
+	.value 26722
+	.value -31521
+	.value 6270
+	.value 26722
+	.value 6270
+	.value -17855
+	.align 2
+	.type	 table26.56,@object
+table26.56:
+	.value 21407
+	.value 27969
+	.value -21407
+	.value -27969
+	.value 21407
+	.value 11585
+	.value 21407
+	.value 11585
+	.value 29692
+	.value 25172
+	.value -29692
+	.value -16819
+	.value 16819
+	.value 5906
+	.value 25172
+	.value -5906
+	.value 21407
+	.value -11585
+	.value 21407
+	.value -11585
+	.value -21407
+	.value 27969
+	.value 21407
+	.value -27969
+	.value 16819
+	.value -29692
+	.value 25172
+	.value -29692
+	.value 5906
+	.value 25172
+	.value 5906
+	.value -16819
+	.align 2
+	.type	 table35.57,@object
+table35.57:
+	.value 19266
+	.value 25172
+	.value -19266
+	.value -25172
+	.value 19266
+	.value 10426
+	.value 19266
+	.value 10426
+	.value 26722
+	.value 22654
+	.value -26722
+	.value -15137
+	.value 15137
+	.value 5315
+	.value 22654
+	.value -5315
+	.value 19266
+	.value -10426
+	.value 19266
+	.value -10426
+	.value -19266
+	.value 25172
+	.value 19266
+	.value -25172
+	.value 15137
+	.value -26722
+	.value 22654
+	.value -26722
+	.value 5315
+	.value 22654
+	.value 5315
+	.value -15137
+.text
+	.align 4
+.globl gst_idct_sse_idct
+	.type	 gst_idct_sse_idct,@function
+gst_idct_sse_idct:
+	subl $8,%esp
+	pushl %ebp
+	pushl %edi
+	pushl %esi
+	pushl %ebx
+	call .L51
+.L51:
+	popl %ebx
+	addl $_GLOBAL_OFFSET_TABLE_+[.-.L51],%ebx
+	movl 28(%esp),%edx
+	leal table04.54@GOTOFF(%ebx),%eax
+	movq (%edx), %mm2
+	movq 8(%edx), %mm5
+	movq %mm2, %mm0
+	movq (%eax), %mm3
+	movq %mm5, %mm6
+	movq 8(%eax), %mm4
+	pmaddwd %mm0, %mm3
+	pshufw $78, %mm2, %mm2
+	leal rounder0@GOTOFF(%ebx),%ecx
+	movq 16(%eax), %mm1
+	pmaddwd %mm2, %mm4
+	pmaddwd 32(%eax), %mm0
+	pshufw $78, %mm6, %mm6
+	movq 24(%eax), %mm7
+	pmaddwd %mm5, %mm1
+	paddd (%ecx), %mm3
+	pmaddwd %mm6, %mm7
+	pmaddwd 40(%eax), %mm2
+	paddd %mm4, %mm3
+	pmaddwd 48(%eax), %mm5
+	movq %mm3, %mm4
+	pmaddwd 56(%eax), %mm6
+	paddd %mm7, %mm1
+	paddd (%ecx), %mm0
+	psubd %mm1, %mm3
+	psrad $11, %mm3
+	paddd %mm4, %mm1
+	paddd %mm2, %mm0
+	psrad $11, %mm1
+	paddd %mm6, %mm5
+	movq %mm0, %mm4
+	paddd %mm5, %mm0
+	psubd %mm5, %mm4
+	movq 64(%edx), %mm2
+	psrad $11, %mm0
+	movq 72(%edx), %mm5
+	psrad $11, %mm4
+	packssdw %mm0, %mm1
+	movq %mm5, %mm6
+	packssdw %mm3, %mm4
+	movq %mm2, %mm0
+	movq %mm1, (%edx)
+	pshufw $177, %mm4, %mm4
+	movq (%eax), %mm3
+	movq %mm4, 8(%edx)
+	pmaddwd %mm0, %mm3
+	movq 8(%eax), %mm4
+	pshufw $78, %mm2, %mm2
+	leal rounder4@GOTOFF(%ebx),%ecx
+	movq 16(%eax), %mm1
+	pmaddwd %mm2, %mm4
+	pmaddwd 32(%eax), %mm0
+	pshufw $78, %mm6, %mm6
+	movq 24(%eax), %mm7
+	pmaddwd %mm5, %mm1
+	paddd (%ecx), %mm3
+	pmaddwd %mm6, %mm7
+	pmaddwd 40(%eax), %mm2
+	paddd %mm4, %mm3
+	pmaddwd 48(%eax), %mm5
+	movq %mm3, %mm4
+	pmaddwd 56(%eax), %mm6
+	paddd %mm7, %mm1
+	paddd (%ecx), %mm0
+	psubd %mm1, %mm3
+	psrad $11, %mm3
+	paddd %mm4, %mm1
+	paddd %mm2, %mm0
+	psrad $11, %mm1
+	paddd %mm6, %mm5
+	movq %mm0, %mm4
+	paddd %mm5, %mm0
+	psubd %mm5, %mm4
+	leal table17.55@GOTOFF(%ebx),%eax
+	movq 16(%edx), %mm2
+	psrad $11, %mm0
+	movq 24(%edx), %mm5
+	psrad $11, %mm4
+	packssdw %mm0, %mm1
+	movq %mm5, %mm6
+	packssdw %mm3, %mm4
+	movq %mm2, %mm0
+	movq %mm1, 64(%edx)
+	pshufw $177, %mm4, %mm4
+	movq (%eax), %mm3
+	movq %mm4, 72(%edx)
+	pmaddwd %mm0, %mm3
+	movq 8(%eax), %mm4
+	pshufw $78, %mm2, %mm2
+	leal rounder1@GOTOFF(%ebx),%ecx
+	movq 16(%eax), %mm1
+	pmaddwd %mm2, %mm4
+	pmaddwd 32(%eax), %mm0
+	pshufw $78, %mm6, %mm6
+	movq 24(%eax), %mm7
+	pmaddwd %mm5, %mm1
+	paddd (%ecx), %mm3
+	pmaddwd %mm6, %mm7
+	pmaddwd 40(%eax), %mm2
+	paddd %mm4, %mm3
+	pmaddwd 48(%eax), %mm5
+	movq %mm3, %mm4
+	pmaddwd 56(%eax), %mm6
+	paddd %mm7, %mm1
+	paddd (%ecx), %mm0
+	psubd %mm1, %mm3
+	psrad $11, %mm3
+	paddd %mm4, %mm1
+	paddd %mm2, %mm0
+	psrad $11, %mm1
+	paddd %mm6, %mm5
+	movq %mm0, %mm4
+	paddd %mm5, %mm0
+	psubd %mm5, %mm4
+	movq 112(%edx), %mm2
+	psrad $11, %mm0
+	movq 120(%edx), %mm5
+	psrad $11, %mm4
+	packssdw %mm0, %mm1
+	movq %mm5, %mm6
+	packssdw %mm3, %mm4
+	movq %mm2, %mm0
+	movq %mm1, 16(%edx)
+	pshufw $177, %mm4, %mm4
+	movq (%eax), %mm3
+	movq %mm4, 24(%edx)
+	pmaddwd %mm0, %mm3
+	movq 8(%eax), %mm4
+	pshufw $78, %mm2, %mm2
+	leal rounder7@GOTOFF(%ebx),%ecx
+	movq 16(%eax), %mm1
+	pmaddwd %mm2, %mm4
+	pmaddwd 32(%eax), %mm0
+	pshufw $78, %mm6, %mm6
+	movq 24(%eax), %mm7
+	pmaddwd %mm5, %mm1
+	paddd (%ecx), %mm3
+	pmaddwd %mm6, %mm7
+	pmaddwd 40(%eax), %mm2
+	paddd %mm4, %mm3
+	pmaddwd 48(%eax), %mm5
+	movq %mm3, %mm4
+	pmaddwd 56(%eax), %mm6
+	paddd %mm7, %mm1
+	paddd (%ecx), %mm0
+	psubd %mm1, %mm3
+	psrad $11, %mm3
+	paddd %mm4, %mm1
+	paddd %mm2, %mm0
+	psrad $11, %mm1
+	paddd %mm6, %mm5
+	movq %mm0, %mm4
+	paddd %mm5, %mm0
+	psubd %mm5, %mm4
+	leal table26.56@GOTOFF(%ebx),%eax
+	movq 32(%edx), %mm2
+	psrad $11, %mm0
+	movq 40(%edx), %mm5
+	psrad $11, %mm4
+	packssdw %mm0, %mm1
+	movq %mm5, %mm6
+	packssdw %mm3, %mm4
+	movq %mm2, %mm0
+	movq %mm1, 112(%edx)
+	pshufw $177, %mm4, %mm4
+	movq (%eax), %mm3
+	movq %mm4, 120(%edx)
+	pmaddwd %mm0, %mm3
+	movq 8(%eax), %mm4
+	pshufw $78, %mm2, %mm2
+	leal rounder2@GOTOFF(%ebx),%ecx
+	movq 16(%eax), %mm1
+	pmaddwd %mm2, %mm4
+	pmaddwd 32(%eax), %mm0
+	pshufw $78, %mm6, %mm6
+	movq 24(%eax), %mm7
+	pmaddwd %mm5, %mm1
+	paddd (%ecx), %mm3
+	pmaddwd %mm6, %mm7
+	pmaddwd 40(%eax), %mm2
+	paddd %mm4, %mm3
+	pmaddwd 48(%eax), %mm5
+	movq %mm3, %mm4
+	pmaddwd 56(%eax), %mm6
+	paddd %mm7, %mm1
+	paddd (%ecx), %mm0
+	psubd %mm1, %mm3
+	psrad $11, %mm3
+	paddd %mm4, %mm1
+	paddd %mm2, %mm0
+	psrad $11, %mm1
+	paddd %mm6, %mm5
+	movq %mm0, %mm4
+	paddd %mm5, %mm0
+	psubd %mm5, %mm4
+	movq 96(%edx), %mm2
+	psrad $11, %mm0
+	movq 104(%edx), %mm5
+	psrad $11, %mm4
+	packssdw %mm0, %mm1
+	movq %mm5, %mm6
+	packssdw %mm3, %mm4
+	movq %mm2, %mm0
+	movq %mm1, 32(%edx)
+	pshufw $177, %mm4, %mm4
+	movq (%eax), %mm3
+	movq %mm4, 40(%edx)
+	pmaddwd %mm0, %mm3
+	movq 8(%eax), %mm4
+	pshufw $78, %mm2, %mm2
+	leal rounder6@GOTOFF(%ebx),%ecx
+	movq 16(%eax), %mm1
+	pmaddwd %mm2, %mm4
+	pmaddwd 32(%eax), %mm0
+	pshufw $78, %mm6, %mm6
+	movq 24(%eax), %mm7
+	pmaddwd %mm5, %mm1
+	paddd (%ecx), %mm3
+	pmaddwd %mm6, %mm7
+	pmaddwd 40(%eax), %mm2
+	paddd %mm4, %mm3
+	pmaddwd 48(%eax), %mm5
+	movq %mm3, %mm4
+	pmaddwd 56(%eax), %mm6
+	paddd %mm7, %mm1
+	paddd (%ecx), %mm0
+	psubd %mm1, %mm3
+	psrad $11, %mm3
+	paddd %mm4, %mm1
+	paddd %mm2, %mm0
+	psrad $11, %mm1
+	paddd %mm6, %mm5
+	movq %mm0, %mm4
+	paddd %mm5, %mm0
+	psubd %mm5, %mm4
+	leal table35.57@GOTOFF(%ebx),%eax
+	movq 48(%edx), %mm2
+	psrad $11, %mm0
+	movq 56(%edx), %mm5
+	psrad $11, %mm4
+	packssdw %mm0, %mm1
+	movq %mm5, %mm6
+	packssdw %mm3, %mm4
+	movq %mm2, %mm0
+	movq %mm1, 96(%edx)
+	pshufw $177, %mm4, %mm4
+	movq (%eax), %mm3
+	movq %mm4, 104(%edx)
+	pmaddwd %mm0, %mm3
+	movq 8(%eax), %mm4
+	pshufw $78, %mm2, %mm2
+	leal rounder3@GOTOFF(%ebx),%ecx
+	movq 16(%eax), %mm1
+	pmaddwd %mm2, %mm4
+	pmaddwd 32(%eax), %mm0
+	pshufw $78, %mm6, %mm6
+	movq 24(%eax), %mm7
+	pmaddwd %mm5, %mm1
+	paddd (%ecx), %mm3
+	pmaddwd %mm6, %mm7
+	pmaddwd 40(%eax), %mm2
+	paddd %mm4, %mm3
+	pmaddwd 48(%eax), %mm5
+	movq %mm3, %mm4
+	pmaddwd 56(%eax), %mm6
+	paddd %mm7, %mm1
+	paddd (%ecx), %mm0
+	psubd %mm1, %mm3
+	psrad $11, %mm3
+	paddd %mm4, %mm1
+	paddd %mm2, %mm0
+	psrad $11, %mm1
+	paddd %mm6, %mm5
+	movq %mm0, %mm4
+	paddd %mm5, %mm0
+	psubd %mm5, %mm4
+	movq 80(%edx), %mm2
+	psrad $11, %mm0
+	movq 88(%edx), %mm5
+	psrad $11, %mm4
+	packssdw %mm0, %mm1
+	movq %mm5, %mm6
+	packssdw %mm3, %mm4
+	movq %mm2, %mm0
+	movq %mm1, 48(%edx)
+	pshufw $177, %mm4, %mm4
+	movq (%eax), %mm3
+	movq %mm4, 56(%edx)
+	pmaddwd %mm0, %mm3
+	movq 8(%eax), %mm4
+	pshufw $78, %mm2, %mm2
+	leal rounder5@GOTOFF(%ebx),%ecx
+	movq 16(%eax), %mm1
+	pmaddwd %mm2, %mm4
+	pmaddwd 32(%eax), %mm0
+	pshufw $78, %mm6, %mm6
+	movq 24(%eax), %mm7
+	pmaddwd %mm5, %mm1
+	paddd (%ecx), %mm3
+	pmaddwd %mm6, %mm7
+	pmaddwd 40(%eax), %mm2
+	paddd %mm4, %mm3
+	pmaddwd 48(%eax), %mm5
+	movq %mm3, %mm4
+	pmaddwd 56(%eax), %mm6
+	paddd %mm7, %mm1
+	paddd (%ecx), %mm0
+	psubd %mm1, %mm3
+	psrad $11, %mm3
+	paddd %mm4, %mm1
+	paddd %mm2, %mm0
+	psrad $11, %mm1
+	paddd %mm6, %mm5
+	movq %mm0, %mm4
+	paddd %mm5, %mm0
+	psubd %mm5, %mm4
+	psrad $11, %mm0
+	psrad $11, %mm4
+	packssdw %mm0, %mm1
+	packssdw %mm3, %mm4
+	movq %mm1, 80(%edx)
+	pshufw $177, %mm4, %mm4
+	movq %mm4, 88(%edx)
+	leal _T1.46@GOTOFF(%ebx),%edi
+	movq (%edi), %mm0
+	movq 16(%edx), %mm1
+	movq %mm0, %mm2
+	movq 112(%edx), %mm4
+	pmulhw %mm1, %mm0
+	leal _T3.48@GOTOFF(%ebx),%esi
+	movl %esi,16(%esp)
+	movq (%esi), %mm5
+	pmulhw %mm4, %mm2
+	movq 80(%edx), %mm6
+	movq %mm5, %mm7
+	movq 48(%edx), %mm3
+	psubsw %mm4, %mm0
+	leal _T2.47@GOTOFF(%ebx),%ecx
+	movq (%ecx), %mm4
+	pmulhw %mm3, %mm5
+	paddsw %mm2, %mm1
+	pmulhw %mm6, %mm7
+	movq %mm4, %mm2
+	paddsw %mm3, %mm5
+	pmulhw 32(%edx), %mm4
+	paddsw %mm6, %mm7
+	psubsw %mm6, %mm5
+	paddsw %mm3, %mm7
+	movq 96(%edx), %mm3
+	movq %mm0, %mm6
+	pmulhw %mm3, %mm2
+	psubsw %mm5, %mm0
+	psubsw %mm3, %mm4
+	paddsw %mm6, %mm5
+	leal scratch0.50@GOTOFF(%ebx),%esi
+	movl %esi,20(%esp)
+	movq %mm0, scratch0.50@GOTOFF(%ebx)
+	movq %mm1, %mm6
+	paddsw 32(%edx), %mm2
+	paddsw %mm7, %mm6
+	psubsw %mm7, %mm1
+	movq %mm1, %mm7
+	movq (%edx), %mm3
+	paddsw %mm5, %mm1
+	leal _C4.49@GOTOFF(%ebx),%eax
+	movq (%eax), %mm0
+	psubsw %mm5, %mm7
+	leal scratch1.51@GOTOFF(%ebx),%ebp
+	movq %mm6, scratch1.51@GOTOFF(%ebx)
+	pmulhw %mm0, %mm1
+	movq %mm4, %mm6
+	pmulhw %mm0, %mm7
+	movq 64(%edx), %mm5
+	movq %mm3, %mm0
+	psubsw %mm5, %mm3
+	paddsw %mm5, %mm0
+	paddsw %mm3, %mm4
+	movq %mm0, %mm5
+	psubsw %mm6, %mm3
+	paddsw %mm2, %mm5
+	paddsw %mm1, %mm1
+	psubsw %mm2, %mm0
+	paddsw %mm7, %mm7
+	movq %mm3, %mm2
+	movq %mm4, %mm6
+	paddsw %mm7, %mm3
+	psraw $6, %mm3
+	paddsw %mm1, %mm4
+	psraw $6, %mm4
+	psubsw %mm1, %mm6
+	movq (%ebp), %mm1
+	psubsw %mm7, %mm2
+	psraw $6, %mm6
+	movq %mm5, %mm7
+	movq %mm4, 16(%edx)
+	psraw $6, %mm2
+	movq %mm3, 32(%edx)
+	paddsw %mm1, %mm5
+	movq (%esi), %mm4
+	psubsw %mm1, %mm7
+	psraw $6, %mm5
+	movq %mm0, %mm3
+	movq %mm2, 80(%edx)
+	psubsw %mm4, %mm3
+	psraw $6, %mm7
+	paddsw %mm0, %mm4
+	movq %mm5, (%edx)
+	psraw $6, %mm3
+	movq %mm6, 96(%edx)
+	psraw $6, %mm4
+	movq %mm7, 112(%edx)
+	movq %mm3, 64(%edx)
+	movq %mm4, 48(%edx)
+	movq (%edi), %mm0
+	movq 24(%edx), %mm1
+	movq %mm0, %mm2
+	movq 120(%edx), %mm4
+	pmulhw %mm1, %mm0
+	movl 16(%esp),%esi
+	movq (%esi), %mm5
+	pmulhw %mm4, %mm2
+	movq 88(%edx), %mm6
+	movq %mm5, %mm7
+	movq 56(%edx), %mm3
+	psubsw %mm4, %mm0
+	movq (%ecx), %mm4
+	pmulhw %mm3, %mm5
+	paddsw %mm2, %mm1
+	pmulhw %mm6, %mm7
+	movq %mm4, %mm2
+	paddsw %mm3, %mm5
+	pmulhw 40(%edx), %mm4
+	paddsw %mm6, %mm7
+	psubsw %mm6, %mm5
+	paddsw %mm3, %mm7
+	movq 104(%edx), %mm3
+	movq %mm0, %mm6
+	pmulhw %mm3, %mm2
+	psubsw %mm5, %mm0
+	psubsw %mm3, %mm4
+	paddsw %mm6, %mm5
+	movq %mm0, scratch0.50@GOTOFF(%ebx)
+	movq %mm1, %mm6
+	paddsw 40(%edx), %mm2
+	paddsw %mm7, %mm6
+	psubsw %mm7, %mm1
+	movq %mm1, %mm7
+	movq 8(%edx), %mm3
+	paddsw %mm5, %mm1
+	movq (%eax), %mm0
+	psubsw %mm5, %mm7
+	movq %mm6, scratch1.51@GOTOFF(%ebx)
+	pmulhw %mm0, %mm1
+	movq %mm4, %mm6
+	pmulhw %mm0, %mm7
+	movq 72(%edx), %mm5
+	movq %mm3, %mm0
+	psubsw %mm5, %mm3
+	paddsw %mm5, %mm0
+	paddsw %mm3, %mm4
+	movq %mm0, %mm5
+	psubsw %mm6, %mm3
+	paddsw %mm2, %mm5
+	paddsw %mm1, %mm1
+	psubsw %mm2, %mm0
+	paddsw %mm7, %mm7
+	movq %mm3, %mm2
+	movq %mm4, %mm6
+	paddsw %mm7, %mm3
+	psraw $6, %mm3
+	paddsw %mm1, %mm4
+	psraw $6, %mm4
+	psubsw %mm1, %mm6
+	movq (%ebp), %mm1
+	psubsw %mm7, %mm2
+	psraw $6, %mm6
+	movq %mm5, %mm7
+	movq %mm4, 24(%edx)
+	psraw $6, %mm2
+	movq %mm3, 40(%edx)
+	paddsw %mm1, %mm5
+	movl 20(%esp),%esi
+	movq (%esi), %mm4
+	psubsw %mm1, %mm7
+	psraw $6, %mm5
+	movq %mm0, %mm3
+	movq %mm2, 88(%edx)
+	psubsw %mm4, %mm3
+	psraw $6, %mm7
+	paddsw %mm0, %mm4
+	movq %mm5, 8(%edx)
+	psraw $6, %mm3
+	movq %mm6, 104(%edx)
+	psraw $6, %mm4
+	movq %mm7, 120(%edx)
+	movq %mm3, 72(%edx)
+	movq %mm4, 56(%edx)
+	popl %ebx
+	popl %esi
+	popl %edi
+	popl %ebp
+	addl $8,%esp
+	ret