Merge commit 'c166148409fe8f0dbccef2fe684286a40ba1e37d'

* commit 'c166148409fe8f0dbccef2fe684286a40ba1e37d':
  dsputil: Move pix_sum, pix_norm1, shrink function pointers to mpegvideoenc

Conflicts:
	libavcodec/dsputil.c
	libavcodec/mpegvideo_enc.c
	libavcodec/x86/dsputilenc.asm
	libavcodec/x86/dsputilenc_mmx.c

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

21 files changed, 498 insertions, 347 deletions

diff --git a/libavcodec/arm/Makefile b/libavcodec/arm/Makefile
index 575e79fe5e..fbbd0696b7 100644
--- a/libavcodec/arm/Makefile
+++ b/libavcodec/arm/Makefile
@@ -22,6 +22,7 @@ OBJS-$(CONFIG_IDCTDSP)                 += arm/idctdsp_init_arm.o        \
 OBJS-$(CONFIG_LLAUDDSP)                += arm/lossless_audiodsp_init_arm.o
 OBJS-$(CONFIG_MPEGAUDIODSP)            += arm/mpegaudiodsp_init_arm.o
 OBJS-$(CONFIG_MPEGVIDEO)               += arm/mpegvideo_arm.o
+OBJS-$(CONFIG_MPEGVIDEOENC)            += arm/mpegvideoencdsp_init_arm.o
 OBJS-$(CONFIG_NEON_CLOBBER_TEST)       += arm/neontest.o
 OBJS-$(CONFIG_VIDEODSP)                += arm/videodsp_init_arm.o
 OBJS-$(CONFIG_VP3DSP)                  += arm/vp3dsp_init_arm.o
@@ -61,6 +62,7 @@ ARMV6-OBJS-$(CONFIG_IDCTDSP)           += arm/idctdsp_init_armv6.o      \
                                           arm/idctdsp_armv6.o           \
                                           arm/simple_idct_armv6.o
 ARMV6-OBJS-$(CONFIG_MPEGAUDIODSP)      += arm/mpegaudiodsp_fixed_armv6.o
+ARMV6-OBJS-$(CONFIG_MPEGVIDEOENC)      += arm/mpegvideoencdsp_armv6.o
 
 ARMV6-OBJS-$(CONFIG_MLP_DECODER)       += arm/mlpdsp_armv6.o
 ARMV6-OBJS-$(CONFIG_VC1_DECODER)       += arm/startcode_armv6.o
diff --git a/libavcodec/arm/dsputil_armv6.S b/libavcodec/arm/dsputil_armv6.S
index 8876d5fa18..60232243e5 100644
--- a/libavcodec/arm/dsputil_armv6.S
+++ b/libavcodec/arm/dsputil_armv6.S
@@ -297,58 +297,3 @@ function ff_sse16_armv6, export=1
 
         pop             {r4-r9, pc}
 endfunc
-
-function ff_pix_norm1_armv6, export=1
-        push            {r4-r6, lr}
-        mov             r12, #16
-        mov             lr,  #0
-1:
-        ldm             r0,  {r2-r5}
-        uxtb16          r6,  r2
-        uxtb16          r2,  r2,  ror #8
-        smlad           lr,  r6,  r6,  lr
-        uxtb16          r6,  r3
-        smlad           lr,  r2,  r2,  lr
-        uxtb16          r3,  r3,  ror #8
-        smlad           lr,  r6,  r6,  lr
-        uxtb16          r6,  r4
-        smlad           lr,  r3,  r3,  lr
-        uxtb16          r4,  r4,  ror #8
-        smlad           lr,  r6,  r6,  lr
-        uxtb16          r6,  r5
-        smlad           lr,  r4,  r4,  lr
-        uxtb16          r5,  r5,  ror #8
-        smlad           lr,  r6,  r6,  lr
-        subs            r12, r12, #1
-        add             r0,  r0,  r1
-        smlad           lr,  r5,  r5,  lr
-        bgt             1b
-
-        mov             r0,  lr
-        pop             {r4-r6, pc}
-endfunc
-
-function ff_pix_sum_armv6, export=1
-        push            {r4-r7, lr}
-        mov             r12, #16
-        mov             r2,  #0
-        mov             r3,  #0
-        mov             lr,  #0
-        ldr             r4,  [r0]
-1:
-        subs            r12, r12, #1
-        ldr             r5,  [r0, #4]
-        usada8          r2,  r4,  lr,  r2
-        ldr             r6,  [r0, #8]
-        usada8          r3,  r5,  lr,  r3
-        ldr             r7,  [r0, #12]
-        usada8          r2,  r6,  lr,  r2
-        beq             2f
-        ldr_pre         r4,  r0,  r1
-        usada8          r3,  r7,  lr,  r3
-        bgt             1b
-2:
-        usada8          r3,  r7,  lr,  r3
-        add             r0,  r2,  r3
-        pop             {r4-r7, pc}
-endfunc
diff --git a/libavcodec/arm/dsputil_init_armv6.c b/libavcodec/arm/dsputil_init_armv6.c
index 57b90daa1e..1cfad42183 100644
--- a/libavcodec/arm/dsputil_init_armv6.c
+++ b/libavcodec/arm/dsputil_init_armv6.c
@@ -43,9 +43,6 @@ int ff_pix_abs8_armv6(MpegEncContext *s, uint8_t *blk1, uint8_t *blk2,
 int ff_sse16_armv6(MpegEncContext *s, uint8_t *blk1, uint8_t *blk2,
                    int line_size, int h);
 
-int ff_pix_norm1_armv6(uint8_t *pix, int line_size);
-int ff_pix_sum_armv6(uint8_t *pix, int line_size);
-
 av_cold void ff_dsputil_init_armv6(DSPContext *c, AVCodecContext *avctx,
                                    unsigned high_bit_depth)
 {
@@ -63,7 +60,4 @@ av_cold void ff_dsputil_init_armv6(DSPContext *c, AVCodecContext *avctx,
     c->sad[1] = ff_pix_abs8_armv6;
 
     c->sse[0] = ff_sse16_armv6;
-
-    c->pix_norm1 = ff_pix_norm1_armv6;
-    c->pix_sum   = ff_pix_sum_armv6;
 }
diff --git a/libavcodec/arm/mpegvideoencdsp_armv6.S b/libavcodec/arm/mpegvideoencdsp_armv6.S
new file mode 100644
index 0000000000..ab0dad7b18
--- /dev/null
+++ b/libavcodec/arm/mpegvideoencdsp_armv6.S
@@ -0,0 +1,76 @@
+/*
+ * Copyright (c) 2009 Mans Rullgard <mans@mansr.com>
+ *
+ * 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
+ */
+
+#include "libavutil/arm/asm.S"
+
+function ff_pix_norm1_armv6, export=1
+        push            {r4-r6, lr}
+        mov             r12, #16
+        mov             lr,  #0
+1:
+        ldm             r0,  {r2-r5}
+        uxtb16          r6,  r2
+        uxtb16          r2,  r2,  ror #8
+        smlad           lr,  r6,  r6,  lr
+        uxtb16          r6,  r3
+        smlad           lr,  r2,  r2,  lr
+        uxtb16          r3,  r3,  ror #8
+        smlad           lr,  r6,  r6,  lr
+        uxtb16          r6,  r4
+        smlad           lr,  r3,  r3,  lr
+        uxtb16          r4,  r4,  ror #8
+        smlad           lr,  r6,  r6,  lr
+        uxtb16          r6,  r5
+        smlad           lr,  r4,  r4,  lr
+        uxtb16          r5,  r5,  ror #8
+        smlad           lr,  r6,  r6,  lr
+        subs            r12, r12, #1
+        add             r0,  r0,  r1
+        smlad           lr,  r5,  r5,  lr
+        bgt             1b
+
+        mov             r0,  lr
+        pop             {r4-r6, pc}
+endfunc
+
+function ff_pix_sum_armv6, export=1
+        push            {r4-r7, lr}
+        mov             r12, #16
+        mov             r2,  #0
+        mov             r3,  #0
+        mov             lr,  #0
+        ldr             r4,  [r0]
+1:
+        subs            r12, r12, #1
+        ldr             r5,  [r0, #4]
+        usada8          r2,  r4,  lr,  r2
+        ldr             r6,  [r0, #8]
+        usada8          r3,  r5,  lr,  r3
+        ldr             r7,  [r0, #12]
+        usada8          r2,  r6,  lr,  r2
+        beq             2f
+        ldr_pre         r4,  r0,  r1
+        usada8          r3,  r7,  lr,  r3
+        bgt             1b
+2:
+        usada8          r3,  r7,  lr,  r3
+        add             r0,  r2,  r3
+        pop             {r4-r7, pc}
+endfunc
diff --git a/libavcodec/arm/mpegvideoencdsp_init_arm.c b/libavcodec/arm/mpegvideoencdsp_init_arm.c
new file mode 100644
index 0000000000..4bfe835684
--- /dev/null
+++ b/libavcodec/arm/mpegvideoencdsp_init_arm.c
@@ -0,0 +1,38 @@
+/*
+ * 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
+ */
+
+#include <stdint.h>
+
+#include "libavutil/cpu.h"
+#include "libavutil/arm/cpu.h"
+#include "libavcodec/avcodec.h"
+#include "libavcodec/mpegvideoencdsp.h"
+
+int ff_pix_norm1_armv6(uint8_t *pix, int line_size);
+int ff_pix_sum_armv6(uint8_t *pix, int line_size);
+
+av_cold void ff_mpegvideoencdsp_init_arm(MpegvideoEncDSPContext *c,
+                                         AVCodecContext *avctx)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (have_armv6(cpu_flags)) {
+        c->pix_norm1 = ff_pix_norm1_armv6;
+        c->pix_sum   = ff_pix_sum_armv6;
+    }
+}
diff --git a/libavcodec/dnxhdenc.c b/libavcodec/dnxhdenc.c
index 850427e3fc..45b882da40 100644
--- a/libavcodec/dnxhdenc.c
+++ b/libavcodec/dnxhdenc.c
@@ -323,6 +323,7 @@ static av_cold int dnxhd_encode_init(AVCodecContext *avctx)
 
     ff_blockdsp_init(&ctx->bdsp, avctx);
     ff_idctdsp_init(&ctx->m.idsp, avctx);
+    ff_mpegvideoencdsp_init(&ctx->m.mpvencdsp, avctx);
     ff_dct_common_init(&ctx->m);
     ff_dct_encode_init(&ctx->m);
 
@@ -733,8 +734,8 @@ static int dnxhd_mb_var_thread(AVCodecContext *avctx, void *arg,
             int varc;
 
             if (!partial_last_row && mb_x * 16 <= avctx->width - 16) {
-                sum  = ctx->m.dsp.pix_sum(pix, ctx->m.linesize);
-                varc = ctx->m.dsp.pix_norm1(pix, ctx->m.linesize);
+                sum  = ctx->m.mpvencdsp.pix_sum(pix, ctx->m.linesize);
+                varc = ctx->m.mpvencdsp.pix_norm1(pix, ctx->m.linesize);
             } else {
                 int bw = FFMIN(avctx->width - 16 * mb_x, 16);
                 int bh = FFMIN((avctx->height >> ctx->interlaced) - 16 * mb_y, 16);
diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c
index fe3c135713..640a4bfa9c 100644
--- a/libavcodec/dsputil.c
+++ b/libavcodec/dsputil.c
@@ -26,7 +26,6 @@
  */
 
 #include "libavutil/attributes.h"
-#include "libavutil/imgutils.h"
 #include "libavutil/internal.h"
 #include "avcodec.h"
 #include "copy_block.h"
@@ -34,8 +33,6 @@
 #include "dsputil.h"
 #include "simple_idct.h"
 #include "faandct.h"
-#include "imgconvert.h"
-#include "mathops.h"
 #include "mpegvideo.h"
 #include "config.h"
 
@@ -48,74 +45,6 @@ uint32_t ff_square_tab[512] = { 0, };
 #define BIT_DEPTH 8
 #include "dsputilenc_template.c"
 
-static int pix_sum_c(uint8_t *pix, int line_size)
-{
-    int s = 0, i, j;
-
-    for (i = 0; i < 16; i++) {
-        for (j = 0; j < 16; j += 8) {
-            s   += pix[0];
-            s   += pix[1];
-            s   += pix[2];
-            s   += pix[3];
-            s   += pix[4];
-            s   += pix[5];
-            s   += pix[6];
-            s   += pix[7];
-            pix += 8;
-        }
-        pix += line_size - 16;
-    }
-    return s;
-}
-
-static int pix_norm1_c(uint8_t *pix, int line_size)
-{
-    int s = 0, i, j;
-    uint32_t *sq = ff_square_tab + 256;
-
-    for (i = 0; i < 16; i++) {
-        for (j = 0; j < 16; j += 8) {
-#if 0
-            s += sq[pix[0]];
-            s += sq[pix[1]];
-            s += sq[pix[2]];
-            s += sq[pix[3]];
-            s += sq[pix[4]];
-            s += sq[pix[5]];
-            s += sq[pix[6]];
-            s += sq[pix[7]];
-#else
-#if HAVE_FAST_64BIT
-            register uint64_t x = *(uint64_t *) pix;
-            s += sq[x         & 0xff];
-            s += sq[(x >>  8) & 0xff];
-            s += sq[(x >> 16) & 0xff];
-            s += sq[(x >> 24) & 0xff];
-            s += sq[(x >> 32) & 0xff];
-            s += sq[(x >> 40) & 0xff];
-            s += sq[(x >> 48) & 0xff];
-            s += sq[(x >> 56) & 0xff];
-#else
-            register uint32_t x = *(uint32_t *) pix;
-            s += sq[x         & 0xff];
-            s += sq[(x >>  8) & 0xff];
-            s += sq[(x >> 16) & 0xff];
-            s += sq[(x >> 24) & 0xff];
-            x  = *(uint32_t *) (pix + 4);
-            s += sq[x         & 0xff];
-            s += sq[(x >>  8) & 0xff];
-            s += sq[(x >> 16) & 0xff];
-            s += sq[(x >> 24) & 0xff];
-#endif
-#endif
-            pix += 8;
-        }
-        pix += line_size - 16;
-    }
-    return s;
-}
-
 static int sse4_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
                   int line_size, int h)
 {
@@ -1094,9 +1023,6 @@ av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)
 
     c->sum_abs_dctelem = sum_abs_dctelem_c;
 
-    c->pix_sum   = pix_sum_c;
-    c->pix_norm1 = pix_norm1_c;
-
     /* TODO [0] 16  [1] 8 */
     c->pix_abs[0][0] = pix_abs16_c;
     c->pix_abs[0][1] = pix_abs16_x2_c;
@@ -1141,11 +1067,6 @@ av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)
     ff_dsputil_init_dwt(c);
 #endif
 
-    c->shrink[0] = av_image_copy_plane;
-    c->shrink[1] = ff_shrink22;
-    c->shrink[2] = ff_shrink44;
-    c->shrink[3] = ff_shrink88;
-
     c->draw_edges = draw_edges_8_c;
 
     switch (avctx->bits_per_raw_sample) {
diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h
index 8633c90056..8dafbbd9d7 100644
--- a/libavcodec/dsputil.h
+++ b/libavcodec/dsputil.h
@@ -72,9 +72,6 @@ typedef struct DSPContext {
                         int stride);
     int (*sum_abs_dctelem)(int16_t *block /* align 16 */);
 
-    int (*pix_sum)(uint8_t *pix, int line_size);
-    int (*pix_norm1)(uint8_t *pix, int line_size);
-
     me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
     me_cmp_func sse[6];
     me_cmp_func hadamard8_diff[6];
@@ -108,9 +105,6 @@ typedef struct DSPContext {
 #define EDGE_WIDTH 16
 #define EDGE_TOP    1
 #define EDGE_BOTTOM 2
-
-    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src,
-                      int src_wrap, int width, int height);
 } DSPContext;
 
 void ff_dsputil_static_init(void);
diff --git a/libavcodec/motion_est.c b/libavcodec/motion_est.c
index b093580e8f..6b3cd61e8a 100644
--- a/libavcodec/motion_est.c
+++ b/libavcodec/motion_est.c
@@ -903,8 +903,9 @@ void ff_estimate_p_frame_motion(MpegEncContext * s,
 
     /* intra / predictive decision */
     pix = c->src[0][0];
-    sum = s->dsp.pix_sum(pix, s->linesize);
-    varc = s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)sum*sum)>>8) + 500;
+    sum  = s->mpvencdsp.pix_sum(pix, s->linesize);
+    varc = s->mpvencdsp.pix_norm1(pix, s->linesize) -
+           (((unsigned) sum * sum) >> 8) + 500;
 
     pic->mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
     pic->mb_var [s->mb_stride * mb_y + mb_x] = (varc+128)>>8;
diff --git a/libavcodec/mpegvideo_enc.c b/libavcodec/mpegvideo_enc.c
index 4a453f3723..0d630df4ea 100644
--- a/libavcodec/mpegvideo_enc.c
+++ b/libavcodec/mpegvideo_enc.c
@@ -1010,7 +1010,7 @@ static int get_intra_count(MpegEncContext *s, uint8_t *src,
             int offset = x + y * stride;
             int sad  = s->dsp.sad[0](NULL, src + offset, ref + offset, stride,
                                      16);
-            int mean = (s->dsp.pix_sum(src + offset, stride) + 128) >> 8;
+            int mean = (s->mpvencdsp.pix_sum(src + offset, stride) + 128) >> 8;
             int sae  = get_sae(src + offset, mean, stride);
 
             acc += sae + 500 < sad;
@@ -1278,15 +1278,21 @@ static int estimate_best_b_count(MpegEncContext *s)
                 data[2] += INPLACE_OFFSET;
             }
 
-            s->dsp.shrink[scale](s->tmp_frames[i]->data[0], s->tmp_frames[i]->linesize[0],
-                                 data[0], pre_input.f->linesize[0],
-                                 c->width,      c->height);
-            s->dsp.shrink[scale](s->tmp_frames[i]->data[1], s->tmp_frames[i]->linesize[1],
-                                 data[1], pre_input.f->linesize[1],
-                                 c->width >> 1, c->height >> 1);
-            s->dsp.shrink[scale](s->tmp_frames[i]->data[2], s->tmp_frames[i]->linesize[2],
-                                 data[2], pre_input.f->linesize[2],
-                                 c->width >> 1, c->height >> 1);
+            s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[0],
+                                       s->tmp_frames[i]->linesize[0],
+                                       data[0],
+                                       pre_input.f->linesize[0],
+                                       c->width, c->height);
+            s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[1],
+                                       s->tmp_frames[i]->linesize[1],
+                                       data[1],
+                                       pre_input.f->linesize[1],
+                                       c->width >> 1, c->height >> 1);
+            s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[2],
+                                       s->tmp_frames[i]->linesize[2],
+                                       data[2],
+                                       pre_input.f->linesize[2],
+                                       c->width >> 1, c->height >> 1);
         }
     }
 
@@ -2585,9 +2591,10 @@ static int mb_var_thread(AVCodecContext *c, void *arg){
             int yy = mb_y * 16;
             uint8_t *pix = s->new_picture.f->data[0] + (yy * s->linesize) + xx;
             int varc;
-            int sum = s->dsp.pix_sum(pix, s->linesize);
+            int sum = s->mpvencdsp.pix_sum(pix, s->linesize);
 
-            varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)sum*sum)>>8) + 500 + 128)>>8;
+            varc = (s->mpvencdsp.pix_norm1(pix, s->linesize) -
+                    (((unsigned) sum * sum) >> 8) + 500 + 128) >> 8;
 
             s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc;
             s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8;
diff --git a/libavcodec/mpegvideoencdsp.c b/libavcodec/mpegvideoencdsp.c
index c5e0b4874c..bde4345750 100644
--- a/libavcodec/mpegvideoencdsp.c
+++ b/libavcodec/mpegvideoencdsp.c
@@ -22,7 +22,10 @@
 #include "config.h"
 #include "libavutil/avassert.h"
 #include "libavutil/attributes.h"
+#include "libavutil/imgutils.h"
 #include "avcodec.h"
+#include "dsputil.h"
+#include "imgconvert.h"
 #include "mpegvideoencdsp.h"
 
 static int try_8x8basis_c(int16_t rem[64], int16_t weight[64],
@@ -54,12 +57,92 @@ static void add_8x8basis_c(int16_t rem[64], int16_t basis[64], int scale)
                   (BASIS_SHIFT - RECON_SHIFT);
 }
 
+static int pix_sum_c(uint8_t *pix, int line_size)
+{
+    int s = 0, i, j;
+
+    for (i = 0; i < 16; i++) {
+        for (j = 0; j < 16; j += 8) {
+            s   += pix[0];
+            s   += pix[1];
+            s   += pix[2];
+            s   += pix[3];
+            s   += pix[4];
+            s   += pix[5];
+            s   += pix[6];
+            s   += pix[7];
+            pix += 8;
+        }
+        pix += line_size - 16;
+    }
+    return s;
+}
+
+static int pix_norm1_c(uint8_t *pix, int line_size)
+{
+    int s = 0, i, j;
+    uint32_t *sq = ff_square_tab + 256;
+
+    for (i = 0; i < 16; i++) {
+        for (j = 0; j < 16; j += 8) {
+#if 0
+            s += sq[pix[0]];
+            s += sq[pix[1]];
+            s += sq[pix[2]];
+            s += sq[pix[3]];
+            s += sq[pix[4]];
+            s += sq[pix[5]];
+            s += sq[pix[6]];
+            s += sq[pix[7]];
+#else
+#if HAVE_FAST_64BIT
+            register uint64_t x = *(uint64_t *) pix;
+            s += sq[x         & 0xff];
+            s += sq[(x >>  8) & 0xff];
+            s += sq[(x >> 16) & 0xff];
+            s += sq[(x >> 24) & 0xff];
+            s += sq[(x >> 32) & 0xff];
+            s += sq[(x >> 40) & 0xff];
+            s += sq[(x >> 48) & 0xff];
+            s += sq[(x >> 56) & 0xff];
+#else
+            register uint32_t x = *(uint32_t *) pix;
+            s += sq[x         & 0xff];
+            s += sq[(x >>  8) & 0xff];
+            s += sq[(x >> 16) & 0xff];
+            s += sq[(x >> 24) & 0xff];
+            x  = *(uint32_t *) (pix + 4);
+            s += sq[x         & 0xff];
+            s += sq[(x >>  8) & 0xff];
+            s += sq[(x >> 16) & 0xff];
+            s += sq[(x >> 24) & 0xff];
+#endif
+#endif
+            pix += 8;
+        }
+        pix += line_size - 16;
+    }
+    return s;
+}
+
 av_cold void ff_mpegvideoencdsp_init(MpegvideoEncDSPContext *c,
                                      AVCodecContext *avctx)
 {
     c->try_8x8basis = try_8x8basis_c;
     c->add_8x8basis = add_8x8basis_c;
 
+    c->shrink[0] = av_image_copy_plane;
+    c->shrink[1] = ff_shrink22;
+    c->shrink[2] = ff_shrink44;
+    c->shrink[3] = ff_shrink88;
+
+    c->pix_sum   = pix_sum_c;
+    c->pix_norm1 = pix_norm1_c;
+
+    if (ARCH_ARM)
+        ff_mpegvideoencdsp_init_arm(c, avctx);
+    if (ARCH_PPC)
+        ff_mpegvideoencdsp_init_ppc(c, avctx);
     if (ARCH_X86)
         ff_mpegvideoencdsp_init_x86(c, avctx);
 }
diff --git a/libavcodec/mpegvideoencdsp.h b/libavcodec/mpegvideoencdsp.h
index a7bc2ae2c5..81e3fe67b0 100644
--- a/libavcodec/mpegvideoencdsp.h
+++ b/libavcodec/mpegvideoencdsp.h
@@ -31,10 +31,19 @@ typedef struct MpegvideoEncDSPContext {
                         int16_t basis[64], int scale);
     void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
 
+    int (*pix_sum)(uint8_t *pix, int line_size);
+    int (*pix_norm1)(uint8_t *pix, int line_size);
+
+    void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src,
+                      int src_wrap, int width, int height);
 } MpegvideoEncDSPContext;
 
 void ff_mpegvideoencdsp_init(MpegvideoEncDSPContext *c,
                              AVCodecContext *avctx);
+void ff_mpegvideoencdsp_init_arm(MpegvideoEncDSPContext *c,
+                                 AVCodecContext *avctx);
+void ff_mpegvideoencdsp_init_ppc(MpegvideoEncDSPContext *c,
+                                 AVCodecContext *avctx);
 void ff_mpegvideoencdsp_init_x86(MpegvideoEncDSPContext *c,
                                  AVCodecContext *avctx);
 
diff --git a/libavcodec/ppc/Makefile b/libavcodec/ppc/Makefile
index 88aaf2644a..c357dafbac 100644
--- a/libavcodec/ppc/Makefile
+++ b/libavcodec/ppc/Makefile
@@ -13,6 +13,7 @@ OBJS-$(CONFIG_IDCTDSP)                 += ppc/idctdsp.o
 OBJS-$(CONFIG_MPEGAUDIODSP)            += ppc/mpegaudiodsp_altivec.o
 OBJS-$(CONFIG_MPEGVIDEO)               += ppc/mpegvideo_altivec.o      \
                                           ppc/mpegvideodsp.o
+OBJS-$(CONFIG_MPEGVIDEOENC)            += ppc/mpegvideoencdsp.o
 OBJS-$(CONFIG_VIDEODSP)                += ppc/videodsp_ppc.o
 OBJS-$(CONFIG_VP3DSP)                  += ppc/vp3dsp_altivec.o
 
diff --git a/libavcodec/ppc/dsputil_altivec.c b/libavcodec/ppc/dsputil_altivec.c
index 20b15b019e..5ab1b51e2b 100644
--- a/libavcodec/ppc/dsputil_altivec.c
+++ b/libavcodec/ppc/dsputil_altivec.c
@@ -308,34 +308,6 @@ static int sad8_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
     return s;
 }
 
-static int pix_norm1_altivec(uint8_t *pix, int line_size)
-{
-    int i, s = 0;
-    const vector unsigned int zero =
-        (const vector unsigned int) vec_splat_u32(0);
-    vector unsigned char perm = vec_lvsl(0, pix);
-    vector unsigned int sv = (vector unsigned int) vec_splat_u32(0);
-    vector signed int sum;
-
-    for (i = 0; i < 16; i++) {
-        /* Read the potentially unaligned pixels. */
-        vector unsigned char pixl = vec_ld(0,  pix);
-        vector unsigned char pixr = vec_ld(15, pix);
-        vector unsigned char pixv = vec_perm(pixl, pixr, perm);
-
-        /* Square the values, and add them to our sum. */
-        sv = vec_msum(pixv, pixv, sv);
-
-        pix += line_size;
-    }
-    /* Sum up the four partial sums, and put the result into s. */
-    sum = vec_sums((vector signed int) sv, (vector signed int) zero);
-    sum = vec_splat(sum, 3);
-    vec_ste(sum, 0, &s);
-
-    return s;
-}
-
 /* Sum of Squared Errors for an 8x8 block, AltiVec-enhanced.
  * It's the sad8_altivec code above w/ squaring added. */
 static int sse8_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
@@ -430,35 +402,6 @@ static int sse16_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
     return s;
 }
 
-static int pix_sum_altivec(uint8_t *pix, int line_size)
-{
-    int i, s;
-    const vector unsigned int zero =
-        (const vector unsigned int) vec_splat_u32(0);
-    vector unsigned char perm = vec_lvsl(0, pix);
-    vector unsigned int sad = (vector unsigned int) vec_splat_u32(0);
-    vector signed int sumdiffs;
-
-    for (i = 0; i < 16; i++) {
-        /* Read the potentially unaligned 16 pixels into t1. */
-        vector unsigned char pixl = vec_ld(0,  pix);
-        vector unsigned char pixr = vec_ld(15, pix);
-        vector unsigned char t1   = vec_perm(pixl, pixr, perm);
-
-        /* Add each 4 pixel group together and put 4 results into sad. */
-        sad = vec_sum4s(t1, sad);
-
-        pix += line_size;
-    }
-
-    /* Sum up the four partial sums, and put the result into s. */
-    sumdiffs = vec_sums((vector signed int) sad, (vector signed int) zero);
-    sumdiffs = vec_splat(sumdiffs, 3);
-    vec_ste(sumdiffs, 0, &s);
-
-    return s;
-}
-
 static void get_pixels_altivec(int16_t *restrict block, const uint8_t *pixels,
                                int line_size)
 {
@@ -911,9 +854,6 @@ av_cold void ff_dsputil_init_altivec(DSPContext *c, AVCodecContext *avctx,
     c->sse[0] = sse16_altivec;
     c->sse[1] = sse8_altivec;
 
-    c->pix_norm1 = pix_norm1_altivec;
-    c->pix_sum   = pix_sum_altivec;
-
     c->diff_pixels = diff_pixels_altivec;
 
     if (!high_bit_depth) {
diff --git a/libavcodec/ppc/mpegvideoencdsp.c b/libavcodec/ppc/mpegvideoencdsp.c
new file mode 100644
index 0000000000..00ae2a6f30
--- /dev/null
+++ b/libavcodec/ppc/mpegvideoencdsp.c
@@ -0,0 +1,103 @@
+/*
+ * 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
+ */
+
+#include "config.h"
+#include <stdint.h>
+#if HAVE_ALTIVEC_H
+#include <altivec.h>
+#endif
+
+#include "libavutil/attributes.h"
+#include "libavutil/cpu.h"
+#include "libavutil/ppc/cpu.h"
+#include "libavutil/ppc/types_altivec.h"
+#include "libavutil/ppc/util_altivec.h"
+#include "libavcodec/mpegvideoencdsp.h"
+
+#if HAVE_ALTIVEC
+
+static int pix_norm1_altivec(uint8_t *pix, int line_size)
+{
+    int i, s = 0;
+    const vector unsigned int zero =
+        (const vector unsigned int) vec_splat_u32(0);
+    vector unsigned char perm = vec_lvsl(0, pix);
+    vector unsigned int sv = (vector unsigned int) vec_splat_u32(0);
+    vector signed int sum;
+
+    for (i = 0; i < 16; i++) {
+        /* Read the potentially unaligned pixels. */
+        vector unsigned char pixl = vec_ld(0,  pix);
+        vector unsigned char pixr = vec_ld(15, pix);
+        vector unsigned char pixv = vec_perm(pixl, pixr, perm);
+
+        /* Square the values, and add them to our sum. */
+        sv = vec_msum(pixv, pixv, sv);
+
+        pix += line_size;
+    }
+    /* Sum up the four partial sums, and put the result into s. */
+    sum = vec_sums((vector signed int) sv, (vector signed int) zero);
+    sum = vec_splat(sum, 3);
+    vec_ste(sum, 0, &s);
+
+    return s;
+}
+
+static int pix_sum_altivec(uint8_t *pix, int line_size)
+{
+    int i, s;
+    const vector unsigned int zero =
+        (const vector unsigned int) vec_splat_u32(0);
+    vector unsigned char perm = vec_lvsl(0, pix);
+    vector unsigned int sad = (vector unsigned int) vec_splat_u32(0);
+    vector signed int sumdiffs;
+
+    for (i = 0; i < 16; i++) {
+        /* Read the potentially unaligned 16 pixels into t1. */
+        vector unsigned char pixl = vec_ld(0,  pix);
+        vector unsigned char pixr = vec_ld(15, pix);
+        vector unsigned char t1   = vec_perm(pixl, pixr, perm);
+
+        /* Add each 4 pixel group together and put 4 results into sad. */
+        sad = vec_sum4s(t1, sad);
+
+        pix += line_size;
+    }
+
+    /* Sum up the four partial sums, and put the result into s. */
+    sumdiffs = vec_sums((vector signed int) sad, (vector signed int) zero);
+    sumdiffs = vec_splat(sumdiffs, 3);
+    vec_ste(sumdiffs, 0, &s);
+
+    return s;
+}
+
+#endif /* HAVE_ALTIVEC */
+
+av_cold void ff_mpegvideoencdsp_init_ppc(MpegvideoEncDSPContext *c,
+                                         AVCodecContext *avctx)
+{
+#if HAVE_ALTIVEC
+    if (!PPC_ALTIVEC(av_get_cpu_flags()))
+        return;
+
+    c->pix_norm1 = pix_norm1_altivec;
+    c->pix_sum   = pix_sum_altivec;
+#endif /* HAVE_ALTIVEC */
+}
diff --git a/libavcodec/svq1enc.c b/libavcodec/svq1enc.c
index 6d1b397fd2..9ff690dce2 100644
--- a/libavcodec/svq1enc.c
+++ b/libavcodec/svq1enc.c
@@ -517,6 +517,7 @@ static av_cold int svq1_encode_init(AVCodecContext *avctx)
 
     ff_dsputil_init(&s->dsp, avctx);
     ff_hpeldsp_init(&s->hdsp, avctx->flags);
+    ff_mpegvideoencdsp_init(&s->m.mpvencdsp, avctx);
 
     avctx->coded_frame = av_frame_alloc();
     s->current_picture = av_frame_alloc();
diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
index f757be177d..ac336c7d86 100644
--- a/libavcodec/x86/Makefile
+++ b/libavcodec/x86/Makefile
@@ -109,6 +109,7 @@ YASM-OBJS-$(CONFIG_IDCTDSP)            += x86/idctdsp.o
 YASM-OBJS-$(CONFIG_LLAUDDSP)           += x86/lossless_audiodsp.o
 YASM-OBJS-$(CONFIG_LLVIDDSP)           += x86/lossless_videodsp.o
 YASM-OBJS-$(CONFIG_MPEGAUDIODSP)       += x86/imdct36.o
+YASM-OBJS-$(CONFIG_MPEGVIDEOENC)       += x86/mpegvideoencdsp.o
 YASM-OBJS-$(CONFIG_QPELDSP)            += x86/qpeldsp.o                 \
                                           x86/fpel.o                    \
                                           x86/qpel.o
diff --git a/libavcodec/x86/dsputilenc.asm b/libavcodec/x86/dsputilenc.asm
index 84cb7b363b..13682ba5d4 100644
--- a/libavcodec/x86/dsputilenc.asm
+++ b/libavcodec/x86/dsputilenc.asm
@@ -23,10 +23,6 @@
 
 %include "libavutil/x86/x86util.asm"
 
-SECTION_RODATA
-
-cextern pw_1
-
 SECTION .text
 
 %macro DIFF_PIXELS_1 4
@@ -465,113 +461,6 @@ cglobal diff_pixels, 4, 5, 5
     jne .loop
     RET
 
-; int ff_pix_sum16_mmx(uint8_t *pix, int line_size)
-; %1 = number of xmm registers used
-; %2 = number of loops
-; %3 = number of GPRs used
-%macro PIX_SUM16 4
-cglobal pix_sum16, 2, %3, %1
-    movsxdifnidn r1, r1d
-    mov          r2, %2
-%if cpuflag(xop)
-    lea          r3, [r1*3]
-%else
-    pxor         m5, m5
-%endif
-    pxor         m4, m4
-.loop:
-%if cpuflag(xop)
-    vphaddubq    m0, [r0]
-    vphaddubq    m1, [r0+r1]
-    vphaddubq    m2, [r0+r1*2]
-    vphaddubq    m3, [r0+r3]
-%else
-    mova         m0, [r0]
-%if mmsize == 8
-    mova         m1, [r0+8]
-%else
-    mova         m1, [r0+r1]
-%endif
-    punpckhbw    m2, m0, m5
-    punpcklbw    m0, m5
-    punpckhbw    m3, m1, m5
-    punpcklbw    m1, m5
-%endif ; cpuflag(xop)
-    paddw        m1, m0
-    paddw        m3, m2
-    paddw        m3, m1
-    paddw        m4, m3
-%if mmsize == 8
-    add          r0, r1
-%else
-    lea          r0, [r0+r1*%4]
-%endif
-    dec r2
-    jne .loop
-%if cpuflag(xop)
-    pshufd       m0, m4, q0032
-    paddd        m4, m0
-%else
-    HADDW        m4, m5
-%endif
-    movd        eax, m4
-    RET
-%endmacro
-
-INIT_MMX mmx
-PIX_SUM16 0, 16, 3, 0
-INIT_XMM sse2
-PIX_SUM16 6, 8,  3, 2
-%if HAVE_XOP_EXTERNAL
-INIT_XMM xop
-PIX_SUM16 5, 4,  4, 4
-%endif
-
-; int ff_pix_norm1_mmx(uint8_t *pix, int line_size)
-; %1 = number of xmm registers used
-; %2 = number of loops
-%macro PIX_NORM1 2
-cglobal pix_norm1, 2, 3, %1
-    movsxdifnidn r1, r1d
-    mov          r2, %2
-    pxor         m0, m0
-    pxor         m5, m5
-.loop:
-    mova         m2, [r0+0]
-%if mmsize == 8
-    mova         m3, [r0+8]
-%else
-    mova         m3, [r0+r1]
-%endif
-    punpckhbw    m1, m2, m0
-    punpcklbw    m2, m0
-    punpckhbw    m4, m3, m0
-    punpcklbw    m3, m0
-    pmaddwd      m1, m1
-    pmaddwd      m2, m2
-    pmaddwd      m3, m3
-    pmaddwd      m4, m4
-    paddd        m2, m1
-    paddd        m4, m3
-    paddd        m5, m2
-    paddd        m5, m4
-%if mmsize == 8
-    add          r0, r1
-%else
-    lea          r0, [r0+r1*2]
-%endif
-    dec r2
-    jne .loop
-    HADDD        m5, m1
-    movd        eax, m5
-    RET
-%endmacro
-
-INIT_MMX mmx
-PIX_NORM1 0, 16
-INIT_XMM sse2
-PIX_NORM1 6, 8
-
 ;-----------------------------------------------
 ;int ff_sum_abs_dctelem(int16_t *block)
 ;-----------------------------------------------
diff --git a/libavcodec/x86/dsputilenc_mmx.c b/libavcodec/x86/dsputilenc_mmx.c
index f235ad0a53..9e3078b144 100644
--- a/libavcodec/x86/dsputilenc_mmx.c
+++ b/libavcodec/x86/dsputilenc_mmx.c
@@ -37,11 +37,6 @@ void ff_diff_pixels_mmx(int16_t *block, const uint8_t *s1, const uint8_t *s2,
                         int stride);
 void ff_diff_pixels_sse2(int16_t *block, const uint8_t *s1, const uint8_t *s2,
                          int stride);
-int ff_pix_sum16_mmx(uint8_t *pix, int line_size);
-int ff_pix_sum16_sse2(uint8_t *pix, int line_size);
-int ff_pix_sum16_xop(uint8_t *pix, int line_size);
-int ff_pix_norm1_mmx(uint8_t *pix, int line_size);
-int ff_pix_norm1_sse2(uint8_t *pix, int line_size);
 int ff_sum_abs_dctelem_mmx(int16_t *block);
 int ff_sum_abs_dctelem_mmxext(int16_t *block);
 int ff_sum_abs_dctelem_sse2(int16_t *block);
@@ -364,8 +359,6 @@ av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx,
         if (!high_bit_depth)
             c->get_pixels = ff_get_pixels_mmx;
         c->diff_pixels = ff_diff_pixels_mmx;
-        c->pix_sum     = ff_pix_sum16_mmx;
-        c->pix_norm1   = ff_pix_norm1_mmx;
     }
 
     if (EXTERNAL_SSE2(cpu_flags))
@@ -431,8 +424,6 @@ av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx,
         c->sse[0] = ff_sse16_sse2;
         c->sum_abs_dctelem   = ff_sum_abs_dctelem_sse2;
         c->diff_pixels = ff_diff_pixels_sse2;
-        c->pix_sum     = ff_pix_sum16_sse2;
-        c->pix_norm1   = ff_pix_norm1_sse2;
 
 #if HAVE_ALIGNED_STACK
         c->hadamard8_diff[0] = ff_hadamard8_diff16_sse2;
@@ -448,9 +439,5 @@ av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx,
 #endif
     }
 
-    if (EXTERNAL_XOP(cpu_flags)) {
-        c->pix_sum           = ff_pix_sum16_xop;
-    }
-
     ff_dsputil_init_pix_mmx(c, avctx);
 }
diff --git a/libavcodec/x86/mpegvideoencdsp.asm b/libavcodec/x86/mpegvideoencdsp.asm
new file mode 100644
index 0000000000..4fe6cfe5a6
--- /dev/null
+++ b/libavcodec/x86/mpegvideoencdsp.asm
@@ -0,0 +1,137 @@
+;*****************************************************************************
+;* SIMD-optimized MPEG encoding functions
+;*****************************************************************************
+;* Copyright (c) 2000, 2001 Fabrice Bellard
+;* Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
+;*
+;* 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
+;*****************************************************************************
+
+%include "libavutil/x86/x86util.asm"
+
+SECTION_RODATA
+
+cextern pw_1
+
+SECTION .text
+; int ff_pix_sum16_mmx(uint8_t *pix, int line_size)
+; %1 = number of xmm registers used
+; %2 = number of loops
+; %3 = number of GPRs used
+%macro PIX_SUM16 4
+cglobal pix_sum16, 2, %3, %1
+    movsxdifnidn r1, r1d
+    mov          r2, %2
+%if cpuflag(xop)
+    lea          r3, [r1*3]
+%else
+    pxor         m5, m5
+%endif
+    pxor         m4, m4
+.loop:
+%if cpuflag(xop)
+    vphaddubq    m0, [r0]
+    vphaddubq    m1, [r0+r1]
+    vphaddubq    m2, [r0+r1*2]
+    vphaddubq    m3, [r0+r3]
+%else
+    mova         m0, [r0]
+%if mmsize == 8
+    mova         m1, [r0+8]
+%else
+    mova         m1, [r0+r1]
+%endif
+    punpckhbw    m2, m0, m5
+    punpcklbw    m0, m5
+    punpckhbw    m3, m1, m5
+    punpcklbw    m1, m5
+%endif ; cpuflag(xop)
+    paddw        m1, m0
+    paddw        m3, m2
+    paddw        m3, m1
+    paddw        m4, m3
+%if mmsize == 8
+    add          r0, r1
+%else
+    lea          r0, [r0+r1*%4]
+%endif
+    dec r2
+    jne .loop
+%if cpuflag(xop)
+    pshufd       m0, m4, q0032
+    paddd        m4, m0
+%else
+    HADDW        m4, m5
+%endif
+    movd        eax, m4
+    RET
+%endmacro
+
+INIT_MMX mmx
+PIX_SUM16 0, 16, 3, 0
+INIT_XMM sse2
+PIX_SUM16 6, 8,  3, 2
+%if HAVE_XOP_EXTERNAL
+INIT_XMM xop
+PIX_SUM16 5, 4,  4, 4
+%endif
+
+; int ff_pix_norm1_mmx(uint8_t *pix, int line_size)
+; %1 = number of xmm registers used
+; %2 = number of loops
+%macro PIX_NORM1 2
+cglobal pix_norm1, 2, 3, %1
+    movsxdifnidn r1, r1d
+    mov          r2, %2
+    pxor         m0, m0
+    pxor         m5, m5
+.loop:
+    mova         m2, [r0+0]
+%if mmsize == 8
+    mova         m3, [r0+8]
+%else
+    mova         m3, [r0+r1]
+%endif
+    punpckhbw    m1, m2, m0
+    punpcklbw    m2, m0
+    punpckhbw    m4, m3, m0
+    punpcklbw    m3, m0
+    pmaddwd      m1, m1
+    pmaddwd      m2, m2
+    pmaddwd      m3, m3
+    pmaddwd      m4, m4
+    paddd        m2, m1
+    paddd        m4, m3
+    paddd        m5, m2
+    paddd        m5, m4
+%if mmsize == 8
+    add          r0, r1
+%else
+    lea          r0, [r0+r1*2]
+%endif
+    dec r2
+    jne .loop
+    HADDD        m5, m1
+    movd        eax, m5
+    RET
+%endmacro
+
+INIT_MMX mmx
+PIX_NORM1 0, 16
+INIT_XMM sse2
+PIX_NORM1 6, 8
+
diff --git a/libavcodec/x86/mpegvideoencdsp_init.c b/libavcodec/x86/mpegvideoencdsp_init.c
index d7650ec0e1..16841893a4 100644
--- a/libavcodec/x86/mpegvideoencdsp_init.c
+++ b/libavcodec/x86/mpegvideoencdsp_init.c
@@ -22,6 +22,12 @@
 #include "libavcodec/avcodec.h"
 #include "libavcodec/mpegvideoencdsp.h"
 
+int ff_pix_sum16_mmx(uint8_t *pix, int line_size);
+int ff_pix_sum16_sse2(uint8_t *pix, int line_size);
+int ff_pix_sum16_xop(uint8_t *pix, int line_size);
+int ff_pix_norm1_mmx(uint8_t *pix, int line_size);
+int ff_pix_norm1_sse2(uint8_t *pix, int line_size);
+
 #if HAVE_INLINE_ASM
 
 #define PHADDD(a, t)                            \
@@ -95,9 +101,24 @@
 av_cold void ff_mpegvideoencdsp_init_x86(MpegvideoEncDSPContext *c,
                                          AVCodecContext *avctx)
 {
-#if HAVE_INLINE_ASM
     int cpu_flags = av_get_cpu_flags();
 
+    if (EXTERNAL_MMX(cpu_flags)) {
+        c->pix_sum   = ff_pix_sum16_mmx;
+        c->pix_norm1 = ff_pix_norm1_mmx;
+    }
+
+    if (EXTERNAL_SSE2(cpu_flags)) {
+        c->pix_sum     = ff_pix_sum16_sse2;
+        c->pix_norm1   = ff_pix_norm1_sse2;
+    }
+
+    if (EXTERNAL_XOP(cpu_flags)) {
+        c->pix_sum     = ff_pix_sum16_xop;
+    }
+
+#if HAVE_INLINE_ASM
+
     if (INLINE_MMX(cpu_flags)) {
         if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
             c->try_8x8basis = try_8x8basis_mmx;