Merge commit '7a1d6ddd2c6b2d66fbc1afa584cf506930a26453'

* commit '7a1d6ddd2c6b2d66fbc1afa584cf506930a26453':
  xvid: Add C IDCT

Conflicts:
	libavcodec/dct-test.c
	libavcodec/xvididct.c

See: 298b3b6c1f8f66b9bc6de53a7b51d3de745d946b
Merged-by: Michael Niedermayer <michaelni@gmx.at>

8 files changed, 324 insertions, 346 deletions

diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index 8173449c21..b0ce0a6249 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -325,7 +325,7 @@ OBJS-$(CONFIG_MPEG1VIDEO_DECODER)      += mpeg12dec.o mpeg12.o mpeg12data.o
 OBJS-$(CONFIG_MPEG1VIDEO_ENCODER)      += mpeg12enc.o mpeg12.o
 OBJS-$(CONFIG_MPEG2VIDEO_DECODER)      += mpeg12dec.o mpeg12.o mpeg12data.o
 OBJS-$(CONFIG_MPEG2VIDEO_ENCODER)      += mpeg12enc.o mpeg12.o
-OBJS-$(CONFIG_MPEG4_DECODER)           += xvididct.o xvid_c_idct.o
+OBJS-$(CONFIG_MPEG4_DECODER)           += xvididct.o
 OBJS-$(CONFIG_MPL2_DECODER)            += mpl2dec.o ass.o
 OBJS-$(CONFIG_MSMPEG4V1_DECODER)       += msmpeg4dec.o msmpeg4.o msmpeg4data.o
 OBJS-$(CONFIG_MSMPEG4V2_DECODER)       += msmpeg4dec.o msmpeg4.o msmpeg4data.o
diff --git a/libavcodec/dct-test.c b/libavcodec/dct-test.c
index bc6b07744a..f4f8abd309 100644
--- a/libavcodec/dct-test.c
+++ b/libavcodec/dct-test.c
@@ -42,6 +42,7 @@
 #include "dct.h"
 #include "idctdsp.h"
 #include "simple_idct.h"
+#include "xvididct.h"
 #include "aandcttab.h"
 #include "faandct.h"
 #include "faanidct.h"
@@ -81,6 +82,9 @@ static const struct algo idct_tab[] = {
     { "INT",         ff_j_rev_dct,         FF_IDCT_PERM_LIBMPEG2 },
     { "SIMPLE-C",    ff_simple_idct_8,     FF_IDCT_PERM_NONE },
     { "PR-C",        ff_prores_idct_wrap,  FF_IDCT_PERM_NONE, 0, 1 },
+#if CONFIG_MPEG4_DECODER
+    { "XVID",        ff_xvid_idct,         FF_IDCT_PERM_NONE, 0, 1 },
+#endif /* CONFIG_MPEG4_DECODER */
 };
 
 #if ARCH_ARM
diff --git a/libavcodec/x86/xvididct_init.c b/libavcodec/x86/xvididct_init.c
index be0f700766..8c227844b5 100644
--- a/libavcodec/x86/xvididct_init.c
+++ b/libavcodec/x86/xvididct_init.c
@@ -26,10 +26,16 @@
 #include "idctdsp.h"
 #include "xvididct.h"
 
-av_cold void ff_xvid_idct_init_x86(IDCTDSPContext *c)
+av_cold void ff_xvid_idct_init_x86(IDCTDSPContext *c, AVCodecContext *avctx,
+                                   unsigned high_bit_depth)
 {
     int cpu_flags = av_get_cpu_flags();
 
+    if (high_bit_depth ||
+        !(avctx->idct_algo == FF_IDCT_AUTO ||
+          avctx->idct_algo == FF_IDCT_XVID))
+        return;
+
     if (INLINE_MMX(cpu_flags)) {
         c->idct_put  = ff_xvid_idct_mmx_put;
         c->idct_add  = ff_xvid_idct_mmx_add;
diff --git a/libavcodec/xvid_c_idct.c b/libavcodec/xvid_c_idct.c
deleted file mode 100644
index a2a5b9c693..0000000000
--- a/libavcodec/xvid_c_idct.c
+++ /dev/null
@@ -1,330 +0,0 @@
-/*****************************************************************************
- *
- *  XVID MPEG-4 VIDEO CODEC
- *  - Inverse DCT  -
- *
- *  Copyright (C) 2006-2011 Xvid Solutions GmbH
- *
- * This file is part of FFmpeg.
- *
- * FFmpeg is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License as published by the Free Software Foundation; either
- * version 2.1 of the License, or (at your option) any later version.
- *
- * FFmpeg 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
- * Lesser General Public License for more details.
- *
- * You should have received a copy of the GNU Lesser General Public
- * License along with FFmpeg; if not, write to the Free Software
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
- *
- *
- ****************************************************************************/
-
-/**
- * @file
- * Walken IDCT
- * Alternative idct implementations for decoding compatibility
- *
- * @author Skal
- * @note this "C" version is not the original one, but is modified to
- *       yield the same error profile as the MMX version.
- */
-
-
-#include "xvididct.h"
-
-
-#define ROW_SHIFT 11
-#define COL_SHIFT 6
-
-// #define FIX(x)   (int)((x) * (1<<ROW_SHIFT))
-#define Rnd0 65536 // 1<<(COL_SHIFT+ROW_SHIFT-1);
-#define Rnd1 3597  // FIX (1.75683487303);
-#define Rnd2 2260  // FIX (1.10355339059);
-#define Rnd3 1203  // FIX (0.587788325588);
-#define Rnd4 0
-#define Rnd5 120   // FIX (0.058658283817);
-#define Rnd6 512   // FIX (0.25);
-#define Rnd7 512   // FIX (0.25);
-
-static const int Tab04[] = { 22725, 21407, 19266, 16384, 12873,  8867, 4520 };
-static const int Tab17[] = { 31521, 29692, 26722, 22725, 17855, 12299, 6270 };
-static const int Tab26[] = { 29692, 27969, 25172, 21407, 16819, 11585, 5906 };
-static const int Tab35[] = { 26722, 25172, 22654, 19266, 15137, 10426, 5315 };
-
-static int Idct_Row(short * In, const int * const Tab, int Rnd)
-{
-  const int C1 = Tab[0];
-  const int C2 = Tab[1];
-  const int C3 = Tab[2];
-  const int C4 = Tab[3];
-  const int C5 = Tab[4];
-  const int C6 = Tab[5];
-  const int C7 = Tab[6];
-
-  const int Right = In[5]|In[6]|In[7];
-  const int Left  = In[1]|In[2]|In[3];
-  if (!(Right | In[4]))
-  {
-    const int K = C4*In[0] + Rnd;
-    if (Left)
-    {
-      const int a0 = K + C2*In[2];
-      const int a1 = K + C6*In[2];
-      const int a2 = K - C6*In[2];
-      const int a3 = K - C2*In[2];
-
-      const int b0 = C1*In[1] + C3*In[3];
-      const int b1 = C3*In[1] - C7*In[3];
-      const int b2 = C5*In[1] - C1*In[3];
-      const int b3 = C7*In[1] - C5*In[3];
-
-      In[0] = (a0 + b0) >> ROW_SHIFT;
-      In[1] = (a1 + b1) >> ROW_SHIFT;
-      In[2] = (a2 + b2) >> ROW_SHIFT;
-      In[3] = (a3 + b3) >> ROW_SHIFT;
-      In[4] = (a3 - b3) >> ROW_SHIFT;
-      In[5] = (a2 - b2) >> ROW_SHIFT;
-      In[6] = (a1 - b1) >> ROW_SHIFT;
-      In[7] = (a0 - b0) >> ROW_SHIFT;
-    }
-    else
-    {
-      const int a0 = K >> ROW_SHIFT;
-      if (a0) {
-        In[0] = In[1] = In[2] = In[3] =
-        In[4] = In[5] = In[6] = In[7] = a0;
-      }
-      else return 0;
-    }
-  }
-  else if (!(Left|Right))
-  {
-    const int a0 = (Rnd + C4*(In[0]+In[4])) >> ROW_SHIFT;
-    const int a1 = (Rnd + C4*(In[0]-In[4])) >> ROW_SHIFT;
-
-    In[0] = a0;
-    In[3] = a0;
-    In[4] = a0;
-    In[7] = a0;
-    In[1] = a1;
-    In[2] = a1;
-    In[5] = a1;
-    In[6] = a1;
-  }
-  else
-  {
-    const int K = C4*In[0] + Rnd;
-    const int a0 = K + C2*In[2] + C4*In[4] + C6*In[6];
-    const int a1 = K + C6*In[2] - C4*In[4] - C2*In[6];
-    const int a2 = K - C6*In[2] - C4*In[4] + C2*In[6];
-    const int a3 = K - C2*In[2] + C4*In[4] - C6*In[6];
-
-    const int b0 = C1*In[1] + C3*In[3] + C5*In[5] + C7*In[7];
-    const int b1 = C3*In[1] - C7*In[3] - C1*In[5] - C5*In[7];
-    const int b2 = C5*In[1] - C1*In[3] + C7*In[5] + C3*In[7];
-    const int b3 = C7*In[1] - C5*In[3] + C3*In[5] - C1*In[7];
-
-    In[0] = (a0 + b0) >> ROW_SHIFT;
-    In[1] = (a1 + b1) >> ROW_SHIFT;
-    In[2] = (a2 + b2) >> ROW_SHIFT;
-    In[3] = (a3 + b3) >> ROW_SHIFT;
-    In[4] = (a3 - b3) >> ROW_SHIFT;
-    In[5] = (a2 - b2) >> ROW_SHIFT;
-    In[6] = (a1 - b1) >> ROW_SHIFT;
-    In[7] = (a0 - b0) >> ROW_SHIFT;
-  }
-  return 1;
-}
-
-#define Tan1  0x32ec
-#define Tan2  0x6a0a
-#define Tan3  0xab0e
-#define Sqrt2 0x5a82
-
-#define MULT(c,x, n)  ( ((c) * (x)) >> (n) )
-// 12b version => #define MULT(c,x, n)  ( (((c)>>3) * (x)) >> ((n)-3) )
-// 12b zero-testing version:
-
-#define BUTF(a, b, tmp) \
-  (tmp) = (a)+(b);      \
-  (b)   = (a)-(b);      \
-  (a)   = (tmp)
-
-#define LOAD_BUTF(m1, m2, a, b, tmp, S) \
-  (m1) = (S)[(a)] + (S)[(b)];           \
-  (m2) = (S)[(a)] - (S)[(b)]
-
-static void Idct_Col_8(short * const In)
-{
-  int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, Spill;
-
-    // odd
-
-  mm4 = (int)In[7*8];
-  mm5 = (int)In[5*8];
-  mm6 = (int)In[3*8];
-  mm7 = (int)In[1*8];
-
-  mm0 = MULT(Tan1, mm4, 16) + mm7;
-  mm1 = MULT(Tan1, mm7, 16) - mm4;
-  mm2 = MULT(Tan3, mm5, 16) + mm6;
-  mm3 = MULT(Tan3, mm6, 16) - mm5;
-
-  mm7 = mm0 + mm2;
-  mm4 = mm1 - mm3;
-  mm0 = mm0 - mm2;
-  mm1 = mm1 + mm3;
-  mm6 = mm0 + mm1;
-  mm5 = mm0 - mm1;
-  mm5 = 2*MULT(Sqrt2, mm5, 16);  // 2*sqrt2
-  mm6 = 2*MULT(Sqrt2, mm6, 16);  // Watch out: precision loss but done to match
-                                 // the pmulhw used in mmx/sse versions
-
-    // even
-
-  mm1 = (int)In[2*8];
-  mm2 = (int)In[6*8];
-  mm3 = MULT(Tan2,mm2, 16) + mm1;
-  mm2 = MULT(Tan2,mm1, 16) - mm2;
-
-  LOAD_BUTF(mm0, mm1, 0*8, 4*8, Spill, In);
-
-  BUTF(mm0, mm3, Spill);
-  BUTF(mm0, mm7, Spill);
-  In[8*0] = (int16_t) (mm0 >> COL_SHIFT);
-  In[8*7] = (int16_t) (mm7 >> COL_SHIFT);
-  BUTF(mm3, mm4, mm0);
-  In[8*3] = (int16_t) (mm3 >> COL_SHIFT);
-  In[8*4] = (int16_t) (mm4 >> COL_SHIFT);
-
-  BUTF(mm1, mm2, mm0);
-  BUTF(mm1, mm6, mm0);
-  In[8*1] = (int16_t) (mm1 >> COL_SHIFT);
-  In[8*6] = (int16_t) (mm6 >> COL_SHIFT);
-  BUTF(mm2, mm5, mm0);
-  In[8*2] = (int16_t) (mm2 >> COL_SHIFT);
-  In[8*5] = (int16_t) (mm5 >> COL_SHIFT);
-}
-
-static void Idct_Col_4(short * const In)
-{
-  int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, Spill;
-
-    // odd
-
-  mm0 = (int)In[1*8];
-  mm2 = (int)In[3*8];
-
-  mm1 = MULT(Tan1, mm0, 16);
-  mm3 = MULT(Tan3, mm2, 16);
-
-  mm7 = mm0 + mm2;
-  mm4 = mm1 - mm3;
-  mm0 = mm0 - mm2;
-  mm1 = mm1 + mm3;
-  mm6 = mm0 + mm1;
-  mm5 = mm0 - mm1;
-  mm6 = 2*MULT(Sqrt2, mm6, 16);  // 2*sqrt2
-  mm5 = 2*MULT(Sqrt2, mm5, 16);
-
-    // even
-
-  mm0 = mm1 = (int)In[0*8];
-  mm3 = (int)In[2*8];
-  mm2 = MULT(Tan2,mm3, 16);
-
-  BUTF(mm0, mm3, Spill);
-  BUTF(mm0, mm7, Spill);
-  In[8*0] = (int16_t) (mm0 >> COL_SHIFT);
-  In[8*7] = (int16_t) (mm7 >> COL_SHIFT);
-  BUTF(mm3, mm4, mm0);
-  In[8*3] = (int16_t) (mm3 >> COL_SHIFT);
-  In[8*4] = (int16_t) (mm4 >> COL_SHIFT);
-
-  BUTF(mm1, mm2, mm0);
-  BUTF(mm1, mm6, mm0);
-  In[8*1] = (int16_t) (mm1 >> COL_SHIFT);
-  In[8*6] = (int16_t) (mm6 >> COL_SHIFT);
-  BUTF(mm2, mm5, mm0);
-  In[8*2] = (int16_t) (mm2 >> COL_SHIFT);
-  In[8*5] = (int16_t) (mm5 >> COL_SHIFT);
-}
-
-static void Idct_Col_3(short * const In)
-{
-  int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, Spill;
-
-    // odd
-
-  mm7 = (int)In[1*8];
-  mm4 = MULT(Tan1, mm7, 16);
-
-  mm6 = mm7 + mm4;
-  mm5 = mm7 - mm4;
-  mm6 = 2*MULT(Sqrt2, mm6, 16);  // 2*sqrt2
-  mm5 = 2*MULT(Sqrt2, mm5, 16);
-
-    // even
-
-  mm0 = mm1 = (int)In[0*8];
-  mm3 = (int)In[2*8];
-  mm2 = MULT(Tan2,mm3, 16);
-
-  BUTF(mm0, mm3, Spill);
-  BUTF(mm0, mm7, Spill);
-  In[8*0] = (int16_t) (mm0 >> COL_SHIFT);
-  In[8*7] = (int16_t) (mm7 >> COL_SHIFT);
-  BUTF(mm3, mm4, mm0);
-  In[8*3] = (int16_t) (mm3 >> COL_SHIFT);
-  In[8*4] = (int16_t) (mm4 >> COL_SHIFT);
-
-  BUTF(mm1, mm2, mm0);
-  BUTF(mm1, mm6, mm0);
-  In[8*1] = (int16_t) (mm1 >> COL_SHIFT);
-  In[8*6] = (int16_t) (mm6 >> COL_SHIFT);
-  BUTF(mm2, mm5, mm0);
-  In[8*2] = (int16_t) (mm2 >> COL_SHIFT);
-  In[8*5] = (int16_t) (mm5 >> COL_SHIFT);
-}
-
-#undef Tan1
-#undef Tan2
-#undef Tan3
-#undef Sqrt2
-
-#undef ROW_SHIFT
-#undef COL_SHIFT
-
-//////////////////////////////////////////////////////////
-
-void ff_idct_xvid(int16_t *const In)
-{
-  int i, Rows = 0x07;
-
-  Idct_Row(In + 0*8, Tab04, Rnd0);
-  Idct_Row(In + 1*8, Tab17, Rnd1);
-  Idct_Row(In + 2*8, Tab26, Rnd2);
-  if (Idct_Row(In + 3*8, Tab35, Rnd3)) Rows |= 0x08;
-  if (Idct_Row(In + 4*8, Tab04, Rnd4)) Rows |= 0x10;
-  if (Idct_Row(In + 5*8, Tab35, Rnd5)) Rows |= 0x20;
-  if (Idct_Row(In + 6*8, Tab26, Rnd6)) Rows |= 0x40;
-  if (Idct_Row(In + 7*8, Tab17, Rnd7)) Rows |= 0x80;
-
-  if (Rows&0xf0) {
-    for(i=0; i<8; i++)
-      Idct_Col_8(In + i);
-  }
-  else if (Rows&0x08) {
-    for(i=0; i<8; i++)
-      Idct_Col_4(In + i);
-  }
-  else {
-    for(i=0; i<8; i++)
-      Idct_Col_3(In + i);
-  }
-}
diff --git a/libavcodec/xvididct.c b/libavcodec/xvididct.c
index 9f3dab805c..e9fab70cbb 100644
--- a/libavcodec/xvididct.c
+++ b/libavcodec/xvididct.c
@@ -1,4 +1,8 @@
 /*
+ * Xvid MPEG-4 IDCT
+ *
+ * Copyright (C) 2006-2011 Xvid Solutions GmbH
+ *
  * This file is part of FFmpeg.
  *
  * FFmpeg is free software; you can redistribute it and/or
@@ -16,21 +20,313 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
+/**
+ * @file
+ * Walken IDCT
+ * Alternative IDCT implementation for decoding compatibility.
+ *
+ * @author Skal
+ * @note This C version is not the original IDCT, but a modified one that
+ *       yields the same error profile as the MMX/MMXEXT/SSE2 versions.
+ */
+
 #include "config.h"
 #include "libavutil/attributes.h"
 #include "avcodec.h"
 #include "idctdsp.h"
 #include "xvididct.h"
 
-static void idct_xvid_put(uint8_t *dest, int line_size, int16_t *block)
+#define ROW_SHIFT 11
+#define COL_SHIFT  6
+
+// #define FIX(x)   (int)((x) * (1 << ROW_SHIFT))
+#define RND0 65536 // 1 << (COL_SHIFT + ROW_SHIFT - 1);
+#define RND1 3597  // FIX (1.75683487303);
+#define RND2 2260  // FIX (1.10355339059);
+#define RND3 1203  // FIX (0.587788325588);
+#define RND4 0
+#define RND5 120   // FIX (0.058658283817);
+#define RND6 512   // FIX (0.25);
+#define RND7 512   // FIX (0.25);
+
+static const int TAB04[] = { 22725, 21407, 19266, 16384, 12873,  8867, 4520 };
+static const int TAB17[] = { 31521, 29692, 26722, 22725, 17855, 12299, 6270 };
+static const int TAB26[] = { 29692, 27969, 25172, 21407, 16819, 11585, 5906 };
+static const int TAB35[] = { 26722, 25172, 22654, 19266, 15137, 10426, 5315 };
+
+static int idct_row(short *in, const int *const tab, int rnd)
+{
+    const int c1 = tab[0];
+    const int c2 = tab[1];
+    const int c3 = tab[2];
+    const int c4 = tab[3];
+    const int c5 = tab[4];
+    const int c6 = tab[5];
+    const int c7 = tab[6];
+
+    const int right = in[5] | in[6] | in[7];
+    const int left  = in[1] | in[2] | in[3];
+    if (!(right | in[4])) {
+        const int k = c4 * in[0] + rnd;
+        if (left) {
+            const int a0 = k + c2 * in[2];
+            const int a1 = k + c6 * in[2];
+            const int a2 = k - c6 * in[2];
+            const int a3 = k - c2 * in[2];
+
+            const int b0 = c1 * in[1] + c3 * in[3];
+            const int b1 = c3 * in[1] - c7 * in[3];
+            const int b2 = c5 * in[1] - c1 * in[3];
+            const int b3 = c7 * in[1] - c5 * in[3];
+
+            in[0] = (a0 + b0) >> ROW_SHIFT;
+            in[1] = (a1 + b1) >> ROW_SHIFT;
+            in[2] = (a2 + b2) >> ROW_SHIFT;
+            in[3] = (a3 + b3) >> ROW_SHIFT;
+            in[4] = (a3 - b3) >> ROW_SHIFT;
+            in[5] = (a2 - b2) >> ROW_SHIFT;
+            in[6] = (a1 - b1) >> ROW_SHIFT;
+            in[7] = (a0 - b0) >> ROW_SHIFT;
+        } else {
+            const int a0 = k >> ROW_SHIFT;
+            if (a0) {
+                in[0] =
+                in[1] =
+                in[2] =
+                in[3] =
+                in[4] =
+                in[5] =
+                in[6] =
+                in[7] = a0;
+            } else
+                return 0;
+        }
+    } else if (!(left | right)) {
+        const int a0 = (rnd + c4 * (in[0] + in[4])) >> ROW_SHIFT;
+        const int a1 = (rnd + c4 * (in[0] - in[4])) >> ROW_SHIFT;
+
+        in[0] = a0;
+        in[3] = a0;
+        in[4] = a0;
+        in[7] = a0;
+        in[1] = a1;
+        in[2] = a1;
+        in[5] = a1;
+        in[6] = a1;
+    } else {
+        const int k  = c4 * in[0] + rnd;
+        const int a0 = k + c2 * in[2] + c4 * in[4] + c6 * in[6];
+        const int a1 = k + c6 * in[2] - c4 * in[4] - c2 * in[6];
+        const int a2 = k - c6 * in[2] - c4 * in[4] + c2 * in[6];
+        const int a3 = k - c2 * in[2] + c4 * in[4] - c6 * in[6];
+
+        const int b0 = c1 * in[1] + c3 * in[3] + c5 * in[5] + c7 * in[7];
+        const int b1 = c3 * in[1] - c7 * in[3] - c1 * in[5] - c5 * in[7];
+        const int b2 = c5 * in[1] - c1 * in[3] + c7 * in[5] + c3 * in[7];
+        const int b3 = c7 * in[1] - c5 * in[3] + c3 * in[5] - c1 * in[7];
+
+        in[0] = (a0 + b0) >> ROW_SHIFT;
+        in[1] = (a1 + b1) >> ROW_SHIFT;
+        in[2] = (a2 + b2) >> ROW_SHIFT;
+        in[3] = (a3 + b3) >> ROW_SHIFT;
+        in[4] = (a3 - b3) >> ROW_SHIFT;
+        in[5] = (a2 - b2) >> ROW_SHIFT;
+        in[6] = (a1 - b1) >> ROW_SHIFT;
+        in[7] = (a0 - b0) >> ROW_SHIFT;
+    }
+    return 1;
+}
+
+#define TAN1  0x32EC
+#define TAN2  0x6A0A
+#define TAN3  0xAB0E
+#define SQRT2 0x5A82
+
+#define MULT(c, x, n)  (((c) * (x)) >> (n))
+// 12b version => #define MULT(c,x, n)  ((((c) >> 3) * (x)) >> ((n) - 3))
+// 12b zero-testing version:
+
+#define BUTTERFLY(a, b, tmp)     \
+    (tmp) = (a) + (b);           \
+    (b)   = (a) - (b);           \
+    (a)   = (tmp)
+
+#define LOAD_BUTTERFLY(m1, m2, a, b, tmp, s)   \
+    (m1) = (s)[(a)] + (s)[(b)];                \
+    (m2) = (s)[(a)] - (s)[(b)]
+
+static void idct_col_8(short *const in)
+{
+    int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, spill;
+
+    // odd
+
+    mm4 = (int) in[7 * 8];
+    mm5 = (int) in[5 * 8];
+    mm6 = (int) in[3 * 8];
+    mm7 = (int) in[1 * 8];
+
+    mm0 = MULT(TAN1, mm4, 16) + mm7;
+    mm1 = MULT(TAN1, mm7, 16) - mm4;
+    mm2 = MULT(TAN3, mm5, 16) + mm6;
+    mm3 = MULT(TAN3, mm6, 16) - mm5;
+
+    mm7 = mm0 + mm2;
+    mm4 = mm1 - mm3;
+    mm0 = mm0 - mm2;
+    mm1 = mm1 + mm3;
+    mm6 = mm0 + mm1;
+    mm5 = mm0 - mm1;
+    mm5 = 2 * MULT(SQRT2, mm5, 16); // 2*sqrt2
+    mm6 = 2 * MULT(SQRT2, mm6, 16); // Watch out: precision loss but done to match
+                                    // the pmulhw used in MMX/MMXEXT/SSE2 versions
+
+    // even
+
+    mm1 = (int) in[2 * 8];
+    mm2 = (int) in[6 * 8];
+    mm3 = MULT(TAN2, mm2, 16) + mm1;
+    mm2 = MULT(TAN2, mm1, 16) - mm2;
+
+    LOAD_BUTTERFLY(mm0, mm1, 0 * 8, 4 * 8, spill, in);
+
+    BUTTERFLY(mm0, mm3, spill);
+    BUTTERFLY(mm0, mm7, spill);
+    in[8 * 0] = (int16_t) (mm0 >> COL_SHIFT);
+    in[8 * 7] = (int16_t) (mm7 >> COL_SHIFT);
+    BUTTERFLY(mm3, mm4, mm0);
+    in[8 * 3] = (int16_t) (mm3 >> COL_SHIFT);
+    in[8 * 4] = (int16_t) (mm4 >> COL_SHIFT);
+
+    BUTTERFLY(mm1, mm2, mm0);
+    BUTTERFLY(mm1, mm6, mm0);
+    in[8 * 1] = (int16_t) (mm1 >> COL_SHIFT);
+    in[8 * 6] = (int16_t) (mm6 >> COL_SHIFT);
+    BUTTERFLY(mm2, mm5, mm0);
+    in[8 * 2] = (int16_t) (mm2 >> COL_SHIFT);
+    in[8 * 5] = (int16_t) (mm5 >> COL_SHIFT);
+}
+
+static void idct_col_4(short *const in)
+{
+    int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, spill;
+
+    // odd
+
+    mm0 = (int) in[1 * 8];
+    mm2 = (int) in[3 * 8];
+
+    mm1 = MULT(TAN1, mm0, 16);
+    mm3 = MULT(TAN3, mm2, 16);
+
+    mm7 = mm0 + mm2;
+    mm4 = mm1 - mm3;
+    mm0 = mm0 - mm2;
+    mm1 = mm1 + mm3;
+    mm6 = mm0 + mm1;
+    mm5 = mm0 - mm1;
+    mm6 = 2 * MULT(SQRT2, mm6, 16); // 2*sqrt2
+    mm5 = 2 * MULT(SQRT2, mm5, 16);
+
+    // even
+
+    mm0 = mm1 = (int) in[0 * 8];
+    mm3 = (int) in[2 * 8];
+    mm2 = MULT(TAN2, mm3, 16);
+
+    BUTTERFLY(mm0, mm3, spill);
+    BUTTERFLY(mm0, mm7, spill);
+    in[8 * 0] = (int16_t) (mm0 >> COL_SHIFT);
+    in[8 * 7] = (int16_t) (mm7 >> COL_SHIFT);
+    BUTTERFLY(mm3, mm4, mm0);
+    in[8 * 3] = (int16_t) (mm3 >> COL_SHIFT);
+    in[8 * 4] = (int16_t) (mm4 >> COL_SHIFT);
+
+    BUTTERFLY(mm1, mm2, mm0);
+    BUTTERFLY(mm1, mm6, mm0);
+    in[8 * 1] = (int16_t) (mm1 >> COL_SHIFT);
+    in[8 * 6] = (int16_t) (mm6 >> COL_SHIFT);
+    BUTTERFLY(mm2, mm5, mm0);
+    in[8 * 2] = (int16_t) (mm2 >> COL_SHIFT);
+    in[8 * 5] = (int16_t) (mm5 >> COL_SHIFT);
+}
+
+static void idct_col_3(short *const in)
+{
+    int mm0, mm1, mm2, mm3, mm4, mm5, mm6, mm7, spill;
+
+    // odd
+
+    mm7 = (int) in[1 * 8];
+    mm4 = MULT(TAN1, mm7, 16);
+
+    mm6 = mm7 + mm4;
+    mm5 = mm7 - mm4;
+    mm6 = 2 * MULT(SQRT2, mm6, 16); // 2*sqrt2
+    mm5 = 2 * MULT(SQRT2, mm5, 16);
+
+    // even
+
+    mm0 = mm1 = (int) in[0 * 8];
+    mm3 = (int) in[2 * 8];
+    mm2 = MULT(TAN2, mm3, 16);
+
+    BUTTERFLY(mm0, mm3, spill);
+    BUTTERFLY(mm0, mm7, spill);
+    in[8 * 0] = (int16_t) (mm0 >> COL_SHIFT);
+    in[8 * 7] = (int16_t) (mm7 >> COL_SHIFT);
+    BUTTERFLY(mm3, mm4, mm0);
+    in[8 * 3] = (int16_t) (mm3 >> COL_SHIFT);
+    in[8 * 4] = (int16_t) (mm4 >> COL_SHIFT);
+
+    BUTTERFLY(mm1, mm2, mm0);
+    BUTTERFLY(mm1, mm6, mm0);
+    in[8 * 1] = (int16_t) (mm1 >> COL_SHIFT);
+    in[8 * 6] = (int16_t) (mm6 >> COL_SHIFT);
+    BUTTERFLY(mm2, mm5, mm0);
+    in[8 * 2] = (int16_t) (mm2 >> COL_SHIFT);
+    in[8 * 5] = (int16_t) (mm5 >> COL_SHIFT);
+}
+
+void ff_xvid_idct(int16_t *const in)
+{
+    int i, rows = 0x07;
+
+    idct_row(in + 0 * 8, TAB04, RND0);
+    idct_row(in + 1 * 8, TAB17, RND1);
+    idct_row(in + 2 * 8, TAB26, RND2);
+    if (idct_row(in + 3 * 8, TAB35, RND3))
+        rows |= 0x08;
+    if (idct_row(in + 4 * 8, TAB04, RND4))
+        rows |= 0x10;
+    if (idct_row(in + 5 * 8, TAB35, RND5))
+        rows |= 0x20;
+    if (idct_row(in + 6 * 8, TAB26, RND6))
+        rows |= 0x40;
+    if (idct_row(in + 7 * 8, TAB17, RND7))
+        rows |= 0x80;
+
+    if (rows & 0xF0) {
+        for (i = 0; i < 8; i++)
+            idct_col_8(in + i);
+    } else if (rows & 0x08) {
+        for (i = 0; i < 8; i++)
+            idct_col_4(in + i);
+    } else {
+        for (i = 0; i < 8; i++)
+            idct_col_3(in + i);
+    }
+}
+
+static void xvid_idct_put(uint8_t *dest, int line_size, int16_t *block)
 {
-    ff_idct_xvid(block);
+    ff_xvid_idct(block);
     ff_put_pixels_clamped(block, dest, line_size);
 }
 
-static void idct_xvid_add(uint8_t *dest, int line_size, int16_t *block)
+static void xvid_idct_add(uint8_t *dest, int line_size, int16_t *block)
 {
-    ff_idct_xvid(block);
+    ff_xvid_idct(block);
     ff_add_pixels_clamped(block, dest, line_size);
 }
 
@@ -44,14 +340,14 @@ av_cold void ff_xvid_idct_init(IDCTDSPContext *c, AVCodecContext *avctx)
         return;
 
     if (avctx->idct_algo == FF_IDCT_XVID) {
-        c->idct_put  = idct_xvid_put;
-        c->idct_add  = idct_xvid_add;
-        c->idct      = ff_idct_xvid;
+        c->idct_put  = xvid_idct_put;
+        c->idct_add  = xvid_idct_add;
+        c->idct      = ff_xvid_idct;
         c->perm_type = FF_IDCT_PERM_NONE;
     }
 
     if (ARCH_X86)
-        ff_xvid_idct_init_x86(c);
+        ff_xvid_idct_init_x86(c, avctx, high_bit_depth);
 
     ff_init_scantable_permutation(c->idct_permutation, c->perm_type);
 }
diff --git a/libavcodec/xvididct.h b/libavcodec/xvididct.h
index b0233c342c..f7dfba4886 100644
--- a/libavcodec/xvididct.h
+++ b/libavcodec/xvididct.h
@@ -19,13 +19,16 @@
 #ifndef AVCODEC_XVIDIDCT_H
 #define AVCODEC_XVIDIDCT_H
 
+#include <stdint.h>
+
 #include "avcodec.h"
 #include "idctdsp.h"
 
-void ff_xvid_idct_init(IDCTDSPContext *c, AVCodecContext *avctx);
+void ff_xvid_idct(int16_t *const in);
 
-void ff_xvid_idct_init_x86(IDCTDSPContext *c);
+void ff_xvid_idct_init(IDCTDSPContext *c, AVCodecContext *avctx);
 
-void ff_idct_xvid(int16_t *const In);
+void ff_xvid_idct_init_x86(IDCTDSPContext *c, AVCodecContext *avctx,
+                           unsigned high_bit_depth);
 
 #endif /* AVCODEC_XVIDIDCT_H */
diff --git a/tests/fate/xvid.mak b/tests/fate/xvid.mak
index 68f01b5827..877b2a100b 100644
--- a/tests/fate/xvid.mak
+++ b/tests/fate/xvid.mak
@@ -1,8 +1,7 @@
 fate-xvid-custom-matrix: CMD = framemd5 -flags +bitexact -idct simple  -i $(TARGET_SAMPLES)/mpeg4/xvid_vlc_trac7411.h263
-fate-xvid-idctmmx:       CMD = framemd5 -flags +bitexact -cpuflags all -i $(TARGET_SAMPLES)/mpeg4/xvid_vlc_trac7411.h263
+fate-xvid-idct:          CMD = framemd5 -flags +bitexact -cpuflags all -i $(TARGET_SAMPLES)/mpeg4/xvid_vlc_trac7411.h263
 
-FATE_XVID-$(call DEMDEC, M4V, MPEG4) += fate-xvid-custom-matrix
-FATE_XVID-$(filter $(HAVE_MMX_INLINE), $(call DEMDEC, M4V, MPEG4)) += fate-xvid-idctmmx
+FATE_XVID-$(call DEMDEC, M4V, MPEG4) += fate-xvid-custom-matrix fate-xvid-idct
 
 FATE_SAMPLES_AVCONV += $(FATE_XVID-yes)
 
diff --git a/tests/ref/fate/xvid-idctmmx b/tests/ref/fate/xvid-idct
index 2fc93ee684..2fc93ee684 100644
--- a/tests/ref/fate/xvid-idctmmx
+++ b/tests/ref/fate/xvid-idct