16x16 and 8x8c x86 SIMD intra pred functions for VP8 and H.264

Originally committed as revision 23783 to svn://svn.ffmpeg.org/ffmpeg/trunk

7 files changed, 566 insertions, 2 deletions

diff --git a/libavcodec/h264pred.c b/libavcodec/h264pred.c
index 6e471949c7..3353a9a2c0 100644
--- a/libavcodec/h264pred.c
+++ b/libavcodec/h264pred.c
@@ -1299,4 +1299,5 @@ void ff_h264_pred_init(H264PredContext *h, int codec_id){
     h->pred16x16_add[ HOR_PRED8x8]= pred16x16_horizontal_add_c;
 
     if (ARCH_ARM) ff_h264_pred_init_arm(h, codec_id);
+    if (ARCH_X86) ff_h264_pred_init_x86(h, codec_id);
 }
diff --git a/libavcodec/h264pred.h b/libavcodec/h264pred.h
index e9f2e17b6b..b7fd2a2b7a 100644
--- a/libavcodec/h264pred.h
+++ b/libavcodec/h264pred.h
@@ -87,5 +87,6 @@ typedef struct H264PredContext{
 
 void ff_h264_pred_init(H264PredContext *h, int codec_id);
 void ff_h264_pred_init_arm(H264PredContext *h, int codec_id);
+void ff_h264_pred_init_x86(H264PredContext *h, int codec_id);
 
 #endif /* AVCODEC_H264PRED_H */
diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
index 5e8b67f855..3a0922337e 100644
--- a/libavcodec/x86/Makefile
+++ b/libavcodec/x86/Makefile
@@ -8,6 +8,7 @@ YASM-OBJS-$(CONFIG_FFT)                += x86/fft_mmx.o                 \
                                           $(YASM-OBJS-FFT-yes)
 YASM-OBJS-$(CONFIG_GPL)                += x86/h264_deblock_sse2.o       \
                                           x86/h264_idct_sse2.o          \
+                                          x86/h264_intrapred.o          \
 
 MMX-OBJS-$(CONFIG_CAVS_DECODER)        += x86/cavsdsp_mmx.o
 MMX-OBJS-$(CONFIG_MP1FLOAT_DECODER)    += x86/mpegaudiodec_mmx.o
diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
index cbdf0174d4..911a4de5fd 100644
--- a/libavcodec/x86/dsputil_mmx.c
+++ b/libavcodec/x86/dsputil_mmx.c
@@ -62,7 +62,7 @@ DECLARE_ALIGNED(8,  const uint64_t, ff_pw_128) = 0x0080008000800080ULL;
 DECLARE_ALIGNED(8,  const uint64_t, ff_pw_255) = 0x00ff00ff00ff00ffULL;
 
 DECLARE_ALIGNED(8,  const uint64_t, ff_pb_1  ) = 0x0101010101010101ULL;
-DECLARE_ALIGNED(8,  const uint64_t, ff_pb_3  ) = 0x0303030303030303ULL;
+DECLARE_ALIGNED(16, const xmm_reg,  ff_pb_3  ) = {0x0303030303030303ULL, 0x0303030303030303ULL};
 DECLARE_ALIGNED(8,  const uint64_t, ff_pb_7  ) = 0x0707070707070707ULL;
 DECLARE_ALIGNED(8,  const uint64_t, ff_pb_1F ) = 0x1F1F1F1F1F1F1F1FULL;
 DECLARE_ALIGNED(8,  const uint64_t, ff_pb_3F ) = 0x3F3F3F3F3F3F3F3FULL;
diff --git a/libavcodec/x86/dsputil_mmx.h b/libavcodec/x86/dsputil_mmx.h
index 7d1bf7fae6..51148e3a95 100644
--- a/libavcodec/x86/dsputil_mmx.h
+++ b/libavcodec/x86/dsputil_mmx.h
@@ -48,7 +48,7 @@ extern const uint64_t ff_pw_128;
 extern const uint64_t ff_pw_255;
 
 extern const uint64_t ff_pb_1;
-extern const uint64_t ff_pb_3;
+extern const xmm_reg  ff_pb_3;
 extern const uint64_t ff_pb_7;
 extern const uint64_t ff_pb_1F;
 extern const uint64_t ff_pb_3F;
diff --git a/libavcodec/x86/h264_intrapred.asm b/libavcodec/x86/h264_intrapred.asm
new file mode 100644
index 0000000000..4f0a43fc2b
--- /dev/null
+++ b/libavcodec/x86/h264_intrapred.asm
@@ -0,0 +1,486 @@
+;******************************************************************************
+;* H.264 intra prediction asm optimizations
+;* Copyright (c) 2010 Jason Garrett-Glaser
+;*
+;* 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
+;* 51, Inc., Foundation Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;******************************************************************************
+
+%include "x86inc.asm"
+
+SECTION_RODATA
+
+tm_shuf: times 8 db 0x03, 0x80
+
+SECTION .text
+
+cextern pb_3
+
+;-----------------------------------------------------------------------------
+; void pred16x16_vertical(uint8_t *src, int stride)
+;-----------------------------------------------------------------------------
+
+cglobal pred16x16_vertical_mmx, 2,3
+    sub   r0, r1
+    mov   r2, 8
+    movq mm0, [r0+0]
+    movq mm1, [r0+8]
+.loop:
+    movq [r0+r1*1+0], mm0
+    movq [r0+r1*1+8], mm1
+    movq [r0+r1*2+0], mm0
+    movq [r0+r1*2+8], mm1
+    lea   r0, [r0+r1*2]
+    dec   r2
+    jg .loop
+    REP_RET
+
+cglobal pred16x16_vertical_sse, 2,3
+    sub   r0, r1
+    mov   r2, 4
+    movaps xmm0, [r0]
+.loop:
+    movaps [r0+r1*1], xmm0
+    movaps [r0+r1*2], xmm0
+    lea   r0, [r0+r1*2]
+    movaps [r0+r1*1], xmm0
+    movaps [r0+r1*2], xmm0
+    lea   r0, [r0+r1*2]
+    dec   r2
+    jg .loop
+    REP_RET
+
+;-----------------------------------------------------------------------------
+; void pred16x16_horizontal(uint8_t *src, int stride)
+;-----------------------------------------------------------------------------
+
+%macro PRED16x16_H 1
+cglobal pred16x16_horizontal_%1, 2,3
+    mov       r2, 8
+%ifidn %1, ssse3
+    mova      m2, [pb_3]
+%endif
+.loop:
+    movd      m0, [r0+r1*0-4]
+    movd      m1, [r0+r1*1-4]
+
+%ifidn %1, ssse3
+    pshufb    m0, m2
+    pshufb    m1, m2
+%else
+    punpcklbw m0, m0
+    punpcklbw m1, m1
+%ifidn %1, mmxext
+    pshufw    m0, m0, 0xff
+    pshufw    m1, m1, 0xff
+%else
+    punpckhwd m0, m0
+    punpckhwd m1, m1
+    punpckhdq m0, m0
+    punpckhdq m1, m1
+%endif
+    mova [r0+r1*0+8], m0
+    mova [r0+r1*1+8], m1
+%endif
+
+    mova [r0+r1*0], m0
+    mova [r0+r1*1], m1
+    lea       r0, [r0+r1*2]
+    dec       r2
+    jg .loop
+    REP_RET
+%endmacro
+
+INIT_MMX
+PRED16x16_H mmx
+PRED16x16_H mmxext
+INIT_XMM
+PRED16x16_H ssse3
+
+;-----------------------------------------------------------------------------
+; void pred16x16_dc(uint8_t *src, int stride)
+;-----------------------------------------------------------------------------
+
+%macro PRED16x16_DC 2
+cglobal pred16x16_dc_%1, 2,7
+    mov       r4, r0
+    sub       r0, r1
+    pxor      mm0, mm0
+    pxor      mm1, mm1
+    psadbw    mm0, [r0+0]
+    psadbw    mm1, [r0+8]
+    dec        r0
+    movzx     r5d, byte [r0+r1*1]
+    paddw     mm0, mm1
+    movd      r6d, mm0
+    lea        r0, [r0+r1*2]
+%rep 7
+    movzx     r2d, byte [r0+r1*0]
+    movzx     r3d, byte [r0+r1*1]
+    add       r5d, r2d
+    add       r6d, r3d
+    lea        r0, [r0+r1*2]
+%endrep
+    movzx     r2d, byte [r0+r1*0]
+    add       r5d, r6d
+    lea       r2d, [r2+r5+16]
+    shr       r2d, 5
+%ifidn %1, mmx
+    movd       m0, r2d
+    punpcklbw  m0, m0
+    punpcklwd  m0, m0
+    punpckldq  m0, m0
+%elifidn %1, mmxext
+    movd       m0, r2d
+    punpcklbw  m0, m0
+    pshufw     m0, m0, 0
+%elifidn %1, sse
+    imul      r2d, 0x01010101
+    movd       m0, r2d
+    shufps     m0, m0, 0
+%elifidn %1, sse2
+    movd       m0, r2d
+    punpcklbw  m0, m0
+    pshuflw    m0, m0, 0
+    punpcklqdq m0, m0
+%elifidn %1, ssse3
+    pxor       m1, m1
+    movd       m0, r2d
+    pshufb     m0, m1
+%endif
+
+%if mmsize==8
+    mov       r3d, 8
+.loop:
+    %2 [r4+r1*0+0], m0
+    %2 [r4+r1*0+8], m0
+    %2 [r4+r1*1+0], m0
+    %2 [r4+r1*1+8], m0
+%else
+    mov       r3d, 4
+.loop:
+    %2 [r4+r1*0], m0
+    %2 [r4+r1*1], m0
+    lea   r4, [r4+r1*2]
+    %2 [r4+r1*0], m0
+    %2 [r4+r1*1], m0
+%endif
+    lea   r4, [r4+r1*2]
+    dec   r3d
+    jg .loop
+    REP_RET
+%endmacro
+
+INIT_MMX
+PRED16x16_DC    mmx, movq
+PRED16x16_DC mmxext, movq
+INIT_XMM
+PRED16x16_DC    sse, movaps
+PRED16x16_DC   sse2, movdqa
+PRED16x16_DC  ssse3, movdqa
+
+;-----------------------------------------------------------------------------
+; void pred16x16_tm_vp8(uint8_t *src, int stride)
+;-----------------------------------------------------------------------------
+
+%macro PRED16x16_TM_MMX 1
+cglobal pred16x16_tm_vp8_%1, 2,5
+    sub        r0, r1
+    pxor      mm7, mm7
+    movq      mm0, [r0+0]
+    movq      mm2, [r0+8]
+    movq      mm1, mm0
+    movq      mm3, mm2
+    punpcklbw mm0, mm7
+    punpckhbw mm1, mm7
+    punpcklbw mm2, mm7
+    punpckhbw mm3, mm7
+    movzx     r3d, byte [r0-1]
+    mov       r4d, 16
+.loop:
+    movzx     r2d, byte [r0+r1-1]
+    sub       r2d, r3d
+    movd      mm4, r2d
+%ifidn %1, mmx
+    punpcklwd mm4, mm4
+    punpckldq mm4, mm4
+%else
+    pshufw    mm4, mm4, 0
+%endif
+    movq      mm5, mm4
+    movq      mm6, mm4
+    movq      mm7, mm4
+    paddw     mm4, mm0
+    paddw     mm5, mm1
+    paddw     mm6, mm2
+    paddw     mm7, mm3
+    packuswb  mm4, mm5
+    packuswb  mm6, mm7
+    movq [r0+r1+0], mm4
+    movq [r0+r1+8], mm6
+    add        r0, r1
+    dec       r4d
+    jg .loop
+    REP_RET
+%endmacro
+
+PRED16x16_TM_MMX mmx
+PRED16x16_TM_MMX mmxext
+
+cglobal pred16x16_tm_vp8_sse2, 2,6,6
+    sub          r0, r1
+    pxor       xmm2, xmm2
+    movdqa     xmm0, [r0]
+    movdqa     xmm1, xmm0
+    punpcklbw  xmm0, xmm2
+    punpckhbw  xmm1, xmm2
+    movzx       r4d, byte [r0-1]
+    mov         r5d, 8
+.loop:
+    movzx       r2d, byte [r0+r1*1-1]
+    movzx       r3d, byte [r0+r1*2-1]
+    sub         r2d, r4d
+    sub         r3d, r4d
+    movd       xmm2, r2d
+    movd       xmm4, r3d
+    pshuflw    xmm2, xmm2, 0
+    pshuflw    xmm4, xmm4, 0
+    punpcklqdq xmm2, xmm2
+    punpcklqdq xmm4, xmm4
+    movdqa     xmm3, xmm2
+    movdqa     xmm5, xmm4
+    paddw      xmm2, xmm0
+    paddw      xmm3, xmm1
+    paddw      xmm4, xmm0
+    paddw      xmm5, xmm1
+    packuswb   xmm2, xmm3
+    packuswb   xmm4, xmm5
+    movdqa [r0+r1*1], xmm2
+    movdqa [r0+r1*2], xmm4
+    lea          r0, [r0+r1*2]
+    dec         r5d
+    jg .loop
+    REP_RET
+
+;-----------------------------------------------------------------------------
+; void pred8x8_vertical(uint8_t *src, int stride)
+;-----------------------------------------------------------------------------
+
+cglobal pred8x8_vertical_mmx, 2,2
+    sub    r0, r1
+    movq  mm0, [r0]
+%rep 3
+    movq [r0+r1*1], mm0
+    movq [r0+r1*2], mm0
+    lea    r0, [r0+r1*2]
+%endrep
+    movq [r0+r1*1], mm0
+    movq [r0+r1*2], mm0
+    RET
+
+;-----------------------------------------------------------------------------
+; void pred8x8_horizontal(uint8_t *src, int stride)
+;-----------------------------------------------------------------------------
+
+%macro PRED8x8_H 1
+cglobal pred8x8_horizontal_%1, 2,3
+    mov       r2, 4
+%ifidn %1, ssse3
+    mova      m2, [pb_3]
+%endif
+.loop:
+    movd      m0, [r0+r1*0-4]
+    movd      m1, [r0+r1*1-4]
+%ifidn %1, ssse3
+    pshufb    m0, m2
+    pshufb    m1, m2
+%else
+    punpcklbw m0, m0
+    punpcklbw m1, m1
+%ifidn %1, mmxext
+    pshufw    m0, m0, 0xff
+    pshufw    m1, m1, 0xff
+%else
+    punpckhwd m0, m0
+    punpckhwd m1, m1
+    punpckhdq m0, m0
+    punpckhdq m1, m1
+%endif
+%endif
+    mova [r0+r1*0], m0
+    mova [r0+r1*1], m1
+    lea       r0, [r0+r1*2]
+    dec       r2
+    jg .loop
+    REP_RET
+%endmacro
+
+INIT_MMX
+PRED8x8_H mmx
+PRED8x8_H mmxext
+PRED8x8_H ssse3
+
+;-----------------------------------------------------------------------------
+; void pred8x8_dc_rv40(uint8_t *src, int stride)
+;-----------------------------------------------------------------------------
+
+%macro PRED8x8_DC 1
+cglobal pred8x8_dc_rv40_%1, 2,7
+    mov       r4, r0
+    sub       r0, r1
+    pxor      mm0, mm0
+    psadbw    mm0, [r0]
+    dec        r0
+    movzx     r5d, byte [r0+r1*1]
+    movd      r6d, mm0
+    lea        r0, [r0+r1*2]
+%rep 3
+    movzx     r2d, byte [r0+r1*0]
+    movzx     r3d, byte [r0+r1*1]
+    add       r5d, r2d
+    add       r6d, r3d
+    lea        r0, [r0+r1*2]
+%endrep
+    movzx     r2d, byte [r0+r1*0]
+    add       r5d, r6d
+    lea       r2d, [r2+r5+8]
+    shr       r2d, 4
+%ifidn %1, mmx
+    movd      mm0, r2d
+    punpcklbw mm0, mm0
+    punpcklwd mm0, mm0
+    punpckldq mm0, mm0
+%else
+    movd      mm0, r2d
+    punpcklbw mm0, mm0
+    pshufw    mm0, mm0, 0
+%endif
+    mov       r3d, 4
+.loop:
+    movq [r4+r1*0], mm0
+    movq [r4+r1*1], mm0
+    lea   r4, [r4+r1*2]
+    dec   r3d
+    jg .loop
+    REP_RET
+%endmacro
+
+
+PRED8x8_DC mmx
+PRED8x8_DC mmxext
+
+;-----------------------------------------------------------------------------
+; void pred8x8_tm_vp8(uint8_t *src, int stride)
+;-----------------------------------------------------------------------------
+
+%macro PRED8x8_TM_MMX 1
+cglobal pred8x8_tm_vp8_%1, 2,6
+    sub        r0, r1
+    pxor      mm7, mm7
+    movq      mm0, [r0]
+    movq      mm1, mm0
+    punpcklbw mm0, mm7
+    punpckhbw mm1, mm7
+    movzx     r4d, byte [r0-1]
+    mov       r5d, 4
+.loop:
+    movzx     r2d, byte [r0+r1*1-1]
+    movzx     r3d, byte [r0+r1*2-1]
+    sub       r2d, r4d
+    sub       r3d, r4d
+    movd      mm2, r2d
+    movd      mm4, r3d
+%ifidn %1, mmx
+    punpcklwd mm2, mm2
+    punpcklwd mm4, mm4
+    punpckldq mm2, mm2
+    punpckldq mm4, mm4
+%else
+    pshufw    mm2, mm2, 0
+    pshufw    mm4, mm4, 0
+%endif
+    movq      mm3, mm2
+    movq      mm5, mm4
+    paddw     mm2, mm0
+    paddw     mm3, mm1
+    paddw     mm4, mm0
+    paddw     mm5, mm1
+    packuswb  mm2, mm3
+    packuswb  mm4, mm5
+    movq [r0+r1*1], mm2
+    movq [r0+r1*2], mm4
+    lea        r0, [r0+r1*2]
+    dec       r5d
+    jg .loop
+    REP_RET
+%endmacro
+
+PRED8x8_TM_MMX mmx
+PRED8x8_TM_MMX mmxext
+
+cglobal pred8x8_tm_vp8_sse2, 2,6,4
+    sub          r0, r1
+    pxor       xmm1, xmm1
+    movq       xmm0, [r0]
+    punpcklbw  xmm0, xmm1
+    movzx       r4d, byte [r0-1]
+    mov         r5d, 4
+.loop:
+    movzx       r2d, byte [r0+r1*1-1]
+    movzx       r3d, byte [r0+r1*2-1]
+    sub         r2d, r4d
+    sub         r3d, r4d
+    movd       xmm2, r2d
+    movd       xmm3, r3d
+    pshuflw    xmm2, xmm2, 0
+    pshuflw    xmm3, xmm3, 0
+    punpcklqdq xmm2, xmm2
+    punpcklqdq xmm3, xmm3
+    paddw      xmm2, xmm0
+    paddw      xmm3, xmm0
+    packuswb   xmm2, xmm3
+    movq   [r0+r1*1], xmm2
+    movhps [r0+r1*2], xmm2
+    lea          r0, [r0+r1*2]
+    dec         r5d
+    jg .loop
+    REP_RET
+
+cglobal pred8x8_tm_vp8_ssse3, 2,3,6
+    sub          r0, r1
+    movdqa     xmm4, [tm_shuf]
+    pxor       xmm1, xmm1
+    movq       xmm0, [r0]
+    punpcklbw  xmm0, xmm1
+    movd       xmm5, [r0-4]
+    pshufb     xmm5, xmm4
+    mov         r2d, 4
+.loop:
+    movd       xmm2, [r0+r1*1-4]
+    movd       xmm3, [r0+r1*2-4]
+    pshufb     xmm2, xmm4
+    pshufb     xmm3, xmm4
+    psubw      xmm2, xmm5
+    psubw      xmm3, xmm5
+    paddw      xmm2, xmm0
+    paddw      xmm3, xmm0
+    packuswb   xmm2, xmm3
+    movq   [r0+r1*1], xmm2
+    movhps [r0+r1*2], xmm2
+    lea          r0, [r0+r1*2]
+    dec         r2d
+    jg .loop
+    REP_RET
diff --git a/libavcodec/x86/h264dsp_mmx.c b/libavcodec/x86/h264dsp_mmx.c
index fd16a02d6a..8c00d5e0fb 100644
--- a/libavcodec/x86/h264dsp_mmx.c
+++ b/libavcodec/x86/h264dsp_mmx.c
@@ -19,6 +19,7 @@
  */
 
 #include "dsputil_mmx.h"
+#include "libavcodec/h264pred.h"
 
 DECLARE_ALIGNED(8, static const uint64_t, ff_pb_3_1  ) = 0x0103010301030103ULL;
 DECLARE_ALIGNED(8, static const uint64_t, ff_pb_7_3  ) = 0x0307030703070307ULL;
@@ -2322,3 +2323,77 @@ H264_WEIGHT( 4, 8)
 H264_WEIGHT( 4, 4)
 H264_WEIGHT( 4, 2)
 
+void ff_pred16x16_vertical_mmx     (uint8_t *src, int stride);
+void ff_pred16x16_vertical_sse     (uint8_t *src, int stride);
+void ff_pred16x16_horizontal_mmx   (uint8_t *src, int stride);
+void ff_pred16x16_horizontal_mmxext(uint8_t *src, int stride);
+void ff_pred16x16_horizontal_ssse3 (uint8_t *src, int stride);
+void ff_pred16x16_dc_mmx           (uint8_t *src, int stride);
+void ff_pred16x16_dc_mmxext        (uint8_t *src, int stride);
+void ff_pred16x16_dc_sse           (uint8_t *src, int stride);
+void ff_pred16x16_dc_sse2          (uint8_t *src, int stride);
+void ff_pred16x16_dc_ssse3         (uint8_t *src, int stride);
+void ff_pred16x16_tm_vp8_mmx       (uint8_t *src, int stride);
+void ff_pred16x16_tm_vp8_mmxext    (uint8_t *src, int stride);
+void ff_pred16x16_tm_vp8_sse2      (uint8_t *src, int stride);
+void ff_pred8x8_dc_rv40_mmx        (uint8_t *src, int stride);
+void ff_pred8x8_dc_rv40_mmxext     (uint8_t *src, int stride);
+void ff_pred8x8_vertical_mmx       (uint8_t *src, int stride);
+void ff_pred8x8_horizontal_mmx     (uint8_t *src, int stride);
+void ff_pred8x8_horizontal_mmxext  (uint8_t *src, int stride);
+void ff_pred8x8_horizontal_ssse3   (uint8_t *src, int stride);
+void ff_pred8x8_tm_vp8_mmx         (uint8_t *src, int stride);
+void ff_pred8x8_tm_vp8_mmxext      (uint8_t *src, int stride);
+void ff_pred8x8_tm_vp8_sse2        (uint8_t *src, int stride);
+void ff_pred8x8_tm_vp8_ssse3      (uint8_t *src, int stride);
+
+void ff_h264_pred_init_x86(H264PredContext *h, int codec_id)
+{
+#if HAVE_YASM
+    if (mm_flags & FF_MM_MMX) {
+        h->pred16x16[VERT_PRED8x8] = ff_pred16x16_vertical_mmx;
+        h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_mmx;
+        h->pred16x16[DC_PRED8x8  ] = ff_pred16x16_dc_mmx;
+        h->pred8x8  [VERT_PRED8x8] = ff_pred8x8_vertical_mmx;
+        h->pred8x8  [HOR_PRED8x8 ] = ff_pred8x8_horizontal_mmx;
+        if (codec_id == CODEC_ID_VP8) {
+            h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_tm_vp8_mmx;
+            h->pred8x8  [DC_PRED8x8   ] = ff_pred8x8_dc_rv40_mmx;
+            h->pred8x8  [PLANE_PRED8x8] = ff_pred8x8_tm_vp8_mmx;
+        }
+    }
+
+    if (mm_flags & FF_MM_MMXEXT) {
+        h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_mmxext;
+        h->pred16x16[DC_PRED8x8  ] = ff_pred16x16_dc_mmxext;
+        h->pred8x8  [HOR_PRED8x8 ] = ff_pred8x8_horizontal_mmxext;
+        if (codec_id == CODEC_ID_VP8) {
+            h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_tm_vp8_mmxext;
+            h->pred8x8  [DC_PRED8x8   ] = ff_pred8x8_dc_rv40_mmxext;
+            h->pred8x8  [PLANE_PRED8x8] = ff_pred8x8_tm_vp8_mmxext;
+        }
+    }
+
+    if (mm_flags & FF_MM_SSE) {
+        h->pred16x16[VERT_PRED8x8] = ff_pred16x16_vertical_sse;
+        h->pred16x16[DC_PRED8x8  ] = ff_pred16x16_dc_sse;
+    }
+
+    if (mm_flags & FF_MM_SSE2) {
+        h->pred16x16[DC_PRED8x8  ] = ff_pred16x16_dc_sse2;
+        if (codec_id == CODEC_ID_VP8) {
+            h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_tm_vp8_sse2;
+            h->pred8x8  [PLANE_PRED8x8] = ff_pred8x8_tm_vp8_sse2;
+        }
+    }
+
+    if (mm_flags & FF_MM_SSSE3) {
+        h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_ssse3;
+        h->pred16x16[DC_PRED8x8  ] = ff_pred16x16_dc_ssse3;
+        h->pred8x8  [HOR_PRED8x8 ] = ff_pred8x8_horizontal_ssse3;
+        if (codec_id == CODEC_ID_VP8) {
+            h->pred8x8  [PLANE_PRED8x8] = ff_pred8x8_tm_vp8_ssse3;
+        }
+    }
+#endif
+}