lossless_videodsp: move shared functions from huffyuvdsp

Several codecs other than huffyuv use them.

Signed-off-by: James Almer <jamrial@gmail.com>

4 files changed, 265 insertions, 262 deletions

diff --git a/libavcodec/x86/huffyuvdsp.asm b/libavcodec/x86/huffyuvdsp.asm
index 0dbe598421..0befd3baa8 100644
--- a/libavcodec/x86/huffyuvdsp.asm
+++ b/libavcodec/x86/huffyuvdsp.asm
@@ -22,196 +22,8 @@
 
 %include "libavutil/x86/x86util.asm"
 
-SECTION_RODATA
-cextern pb_15
-pb_zzzzzzzz77777777: times 8 db -1
-pb_7: times 8 db 7
-pb_zzzz3333zzzzbbbb: db -1,-1,-1,-1,3,3,3,3,-1,-1,-1,-1,11,11,11,11
-pb_zz11zz55zz99zzdd: db -1,-1,1,1,-1,-1,5,5,-1,-1,9,9,-1,-1,13,13
-
 SECTION .text
 
-; void ff_add_hfyu_median_pred_mmxext(uint8_t *dst, const uint8_t *top,
-;                                     const uint8_t *diff, int w,
-;                                     int *left, int *left_top)
-%macro HFYU_MEDIAN 0
-cglobal add_hfyu_median_pred, 6,6,8, dst, top, diff, w, left, left_top
-    movu    m0, [topq]
-    mova    m2, m0
-    movd    m4, [left_topq]
-    LSHIFT  m2, 1
-    mova    m1, m0
-    por     m4, m2
-    movd    m3, [leftq]
-    psubb   m0, m4 ; t-tl
-    add    dstq, wq
-    add    topq, wq
-    add   diffq, wq
-    neg      wq
-    jmp .skip
-.loop:
-    movu    m4, [topq+wq]
-    mova    m0, m4
-    LSHIFT  m4, 1
-    por     m4, m1
-    mova    m1, m0 ; t
-    psubb   m0, m4 ; t-tl
-.skip:
-    movu    m2, [diffq+wq]
-%assign i 0
-%rep mmsize
-    mova    m4, m0
-    paddb   m4, m3 ; t-tl+l
-    mova    m5, m3
-    pmaxub  m3, m1
-    pminub  m5, m1
-    pminub  m3, m4
-    pmaxub  m3, m5 ; median
-    paddb   m3, m2 ; +residual
-%if i==0
-    mova    m7, m3
-    LSHIFT  m7, mmsize-1
-%else
-    mova    m6, m3
-    RSHIFT  m7, 1
-    LSHIFT  m6, mmsize-1
-    por     m7, m6
-%endif
-%if i<mmsize-1
-    RSHIFT  m0, 1
-    RSHIFT  m1, 1
-    RSHIFT  m2, 1
-%endif
-%assign i i+1
-%endrep
-    movu [dstq+wq], m7
-    add      wq, mmsize
-    jl .loop
-    movzx   r2d, byte [dstq-1]
-    mov [leftq], r2d
-    movzx   r2d, byte [topq-1]
-    mov [left_topq], r2d
-    RET
-%endmacro
-
-%if ARCH_X86_32
-INIT_MMX mmxext
-HFYU_MEDIAN
-%endif
-INIT_XMM sse2
-HFYU_MEDIAN
-
-
-%macro ADD_HFYU_LEFT_LOOP 2 ; %1 = dst_is_aligned, %2 = src_is_aligned
-    add     srcq, wq
-    add     dstq, wq
-    neg     wq
-%%.loop:
-%if %2
-    mova    m1, [srcq+wq]
-%else
-    movu    m1, [srcq+wq]
-%endif
-    mova    m2, m1
-    psllw   m1, 8
-    paddb   m1, m2
-    mova    m2, m1
-    pshufb  m1, m3
-    paddb   m1, m2
-    pshufb  m0, m5
-    mova    m2, m1
-    pshufb  m1, m4
-    paddb   m1, m2
-%if mmsize == 16
-    mova    m2, m1
-    pshufb  m1, m6
-    paddb   m1, m2
-%endif
-    paddb   m0, m1
-%if %1
-    mova    [dstq+wq], m0
-%else
-    movq    [dstq+wq], m0
-    movhps  [dstq+wq+8], m0
-%endif
-    add     wq, mmsize
-    jl %%.loop
-    mov     eax, mmsize-1
-    sub     eax, wd
-    movd    m1, eax
-    pshufb  m0, m1
-    movd    eax, m0
-    RET
-%endmacro
-
-; int ff_add_hfyu_left_pred(uint8_t *dst, const uint8_t *src, int w, int left)
-INIT_MMX ssse3
-cglobal add_hfyu_left_pred, 3,3,7, dst, src, w, left
-.skip_prologue:
-    mova    m5, [pb_7]
-    mova    m4, [pb_zzzz3333zzzzbbbb]
-    mova    m3, [pb_zz11zz55zz99zzdd]
-    movd    m0, leftm
-    psllq   m0, 56
-    ADD_HFYU_LEFT_LOOP 1, 1
-
-INIT_XMM sse4
-cglobal add_hfyu_left_pred, 3,3,7, dst, src, w, left
-    mova    m5, [pb_15]
-    mova    m6, [pb_zzzzzzzz77777777]
-    mova    m4, [pb_zzzz3333zzzzbbbb]
-    mova    m3, [pb_zz11zz55zz99zzdd]
-    movd    m0, leftm
-    pslldq  m0, 15
-    test    srcq, 15
-    jnz .src_unaligned
-    test    dstq, 15
-    jnz .dst_unaligned
-    ADD_HFYU_LEFT_LOOP 1, 1
-.dst_unaligned:
-    ADD_HFYU_LEFT_LOOP 0, 1
-.src_unaligned:
-    ADD_HFYU_LEFT_LOOP 0, 0
-
-%macro ADD_BYTES 0
-cglobal add_bytes, 3,4,2, dst, src, w, size
-    mov  sizeq, wq
-    and  sizeq, -2*mmsize
-    jz  .2
-    add   dstq, sizeq
-    add   srcq, sizeq
-    neg  sizeq
-.1:
-    mova    m0, [srcq + sizeq]
-    mova    m1, [srcq + sizeq + mmsize]
-    paddb   m0, [dstq + sizeq]
-    paddb   m1, [dstq + sizeq + mmsize]
-    mova   [dstq + sizeq], m0
-    mova   [dstq + sizeq + mmsize], m1
-    add  sizeq, 2*mmsize
-    jl .1
-.2:
-    and     wq, 2*mmsize-1
-    jz    .end
-    add   dstq, wq
-    add   srcq, wq
-    neg     wq
-.3:
-    mov  sizeb, [srcq + wq]
-    add [dstq + wq], sizeb
-    inc     wq
-    jl .3
-.end:
-    REP_RET
-%endmacro
-
-%if ARCH_X86_32
-INIT_MMX mmx
-ADD_BYTES
-%endif
-INIT_XMM sse2
-ADD_BYTES
-
 ; void add_hfyu_left_pred_bgr32(uint8_t *dst, const uint8_t *src,
 ;                               intptr_t w, uint8_t *left)
 %macro LEFT_BGR32 0
diff --git a/libavcodec/x86/huffyuvdsp_init.c b/libavcodec/x86/huffyuvdsp_init.c
index 3ced3c0a1c..fc87c3844b 100644
--- a/libavcodec/x86/huffyuvdsp_init.c
+++ b/libavcodec/x86/huffyuvdsp_init.c
@@ -25,93 +25,20 @@
 #include "libavutil/x86/cpu.h"
 #include "libavcodec/huffyuvdsp.h"
 
-void ff_add_bytes_mmx(uint8_t *dst, uint8_t *src, intptr_t w);
-void ff_add_bytes_sse2(uint8_t *dst, uint8_t *src, intptr_t w);
-
-void ff_add_hfyu_median_pred_mmxext(uint8_t *dst, const uint8_t *top,
-                                    const uint8_t *diff, intptr_t w,
-                                    int *left, int *left_top);
-void ff_add_hfyu_median_pred_sse2(uint8_t *dst, const uint8_t *top,
-                                  const uint8_t *diff, intptr_t w,
-                                  int *left, int *left_top);
-
-int  ff_add_hfyu_left_pred_ssse3(uint8_t *dst, const uint8_t *src,
-                                 intptr_t w, int left);
-int  ff_add_hfyu_left_pred_sse4(uint8_t *dst, const uint8_t *src,
-                                intptr_t w, int left);
-
 void ff_add_hfyu_left_pred_bgr32_mmx(uint8_t *dst, const uint8_t *src,
                                      intptr_t w, uint8_t *left);
 void ff_add_hfyu_left_pred_bgr32_sse2(uint8_t *dst, const uint8_t *src,
                                       intptr_t w, uint8_t *left);
 
-#if HAVE_INLINE_ASM && HAVE_7REGS && ARCH_X86_32
-static void add_hfyu_median_pred_cmov(uint8_t *dst, const uint8_t *top,
-                                      const uint8_t *diff, intptr_t w,
-                                      int *left, int *left_top)
-{
-    x86_reg w2 = -w;
-    x86_reg x;
-    int l  = *left     & 0xff;
-    int tl = *left_top & 0xff;
-    int t;
-    __asm__ volatile (
-        "mov          %7, %3            \n"
-        "1:                             \n"
-        "movzbl (%3, %4), %2            \n"
-        "mov          %2, %k3           \n"
-        "sub         %b1, %b3           \n"
-        "add         %b0, %b3           \n"
-        "mov          %2, %1            \n"
-        "cmp          %0, %2            \n"
-        "cmovg        %0, %2            \n"
-        "cmovg        %1, %0            \n"
-        "cmp         %k3, %0            \n"
-        "cmovg       %k3, %0            \n"
-        "mov          %7, %3            \n"
-        "cmp          %2, %0            \n"
-        "cmovl        %2, %0            \n"
-        "add    (%6, %4), %b0           \n"
-        "mov         %b0, (%5, %4)      \n"
-        "inc          %4                \n"
-        "jl           1b                \n"
-        : "+&q"(l), "+&q"(tl), "=&r"(t), "=&q"(x), "+&r"(w2)
-        : "r"(dst + w), "r"(diff + w), "rm"(top + w)
-    );
-    *left     = l;
-    *left_top = tl;
-}
-#endif
-
 av_cold void ff_huffyuvdsp_init_x86(HuffYUVDSPContext *c)
 {
     int cpu_flags = av_get_cpu_flags();
 
-#if HAVE_INLINE_ASM && HAVE_7REGS && ARCH_X86_32
-    if (cpu_flags & AV_CPU_FLAG_CMOV)
-        c->add_hfyu_median_pred = add_hfyu_median_pred_cmov;
-#endif
-
     if (ARCH_X86_32 && EXTERNAL_MMX(cpu_flags)) {
-        c->add_bytes = ff_add_bytes_mmx;
         c->add_hfyu_left_pred_bgr32 = ff_add_hfyu_left_pred_bgr32_mmx;
     }
 
-    if (ARCH_X86_32 && EXTERNAL_MMXEXT(cpu_flags)) {
-        /* slower than cmov version on AMD */
-        if (!(cpu_flags & AV_CPU_FLAG_3DNOW))
-            c->add_hfyu_median_pred = ff_add_hfyu_median_pred_mmxext;
-    }
-
     if (EXTERNAL_SSE2(cpu_flags)) {
-        c->add_bytes            = ff_add_bytes_sse2;
-        c->add_hfyu_median_pred = ff_add_hfyu_median_pred_sse2;
         c->add_hfyu_left_pred_bgr32 = ff_add_hfyu_left_pred_bgr32_sse2;
     }
-
-    if (EXTERNAL_SSSE3(cpu_flags)) {
-        c->add_hfyu_left_pred = ff_add_hfyu_left_pred_ssse3;
-        if (cpu_flags & AV_CPU_FLAG_SSE4) // not really SSE4, just slow on Conroe
-            c->add_hfyu_left_pred = ff_add_hfyu_left_pred_sse4;
-    }
 }
diff --git a/libavcodec/x86/lossless_videodsp.asm b/libavcodec/x86/lossless_videodsp.asm
index f06fcdf7cf..a6ce5fe62b 100644
--- a/libavcodec/x86/lossless_videodsp.asm
+++ b/libavcodec/x86/lossless_videodsp.asm
@@ -24,13 +24,199 @@
 
 SECTION_RODATA
 
+cextern pb_15
+pb_zzzzzzzz77777777: times 8 db -1
+pb_7: times 8 db 7
 pb_ef: times 8 db 14,15
 pb_67: times 8 db  6, 7
+pb_zzzz3333zzzzbbbb: db -1,-1,-1,-1,3,3,3,3,-1,-1,-1,-1,11,11,11,11
+pb_zz11zz55zz99zzdd: db -1,-1,1,1,-1,-1,5,5,-1,-1,9,9,-1,-1,13,13
 pb_zzzz2323zzzzabab: db -1,-1,-1,-1, 2, 3, 2, 3,-1,-1,-1,-1,10,11,10,11
 pb_zzzzzzzz67676767: db -1,-1,-1,-1,-1,-1,-1,-1, 6, 7, 6, 7, 6, 7, 6, 7
 
 SECTION .text
 
+; void ff_add_median_pred_mmxext(uint8_t *dst, const uint8_t *top,
+;                                const uint8_t *diff, int w,
+;                                int *left, int *left_top)
+%macro MEDIAN_PRED 0
+cglobal add_median_pred, 6,6,8, dst, top, diff, w, left, left_top
+    movu    m0, [topq]
+    mova    m2, m0
+    movd    m4, [left_topq]
+    LSHIFT  m2, 1
+    mova    m1, m0
+    por     m4, m2
+    movd    m3, [leftq]
+    psubb   m0, m4 ; t-tl
+    add    dstq, wq
+    add    topq, wq
+    add   diffq, wq
+    neg      wq
+    jmp .skip
+.loop:
+    movu    m4, [topq+wq]
+    mova    m0, m4
+    LSHIFT  m4, 1
+    por     m4, m1
+    mova    m1, m0 ; t
+    psubb   m0, m4 ; t-tl
+.skip:
+    movu    m2, [diffq+wq]
+%assign i 0
+%rep mmsize
+    mova    m4, m0
+    paddb   m4, m3 ; t-tl+l
+    mova    m5, m3
+    pmaxub  m3, m1
+    pminub  m5, m1
+    pminub  m3, m4
+    pmaxub  m3, m5 ; median
+    paddb   m3, m2 ; +residual
+%if i==0
+    mova    m7, m3
+    LSHIFT  m7, mmsize-1
+%else
+    mova    m6, m3
+    RSHIFT  m7, 1
+    LSHIFT  m6, mmsize-1
+    por     m7, m6
+%endif
+%if i<mmsize-1
+    RSHIFT  m0, 1
+    RSHIFT  m1, 1
+    RSHIFT  m2, 1
+%endif
+%assign i i+1
+%endrep
+    movu [dstq+wq], m7
+    add      wq, mmsize
+    jl .loop
+    movzx   r2d, byte [dstq-1]
+    mov [leftq], r2d
+    movzx   r2d, byte [topq-1]
+    mov [left_topq], r2d
+    RET
+%endmacro
+
+%if ARCH_X86_32
+INIT_MMX mmxext
+MEDIAN_PRED
+%endif
+INIT_XMM sse2
+MEDIAN_PRED
+
+
+%macro ADD_LEFT_LOOP 2 ; %1 = dst_is_aligned, %2 = src_is_aligned
+    add     srcq, wq
+    add     dstq, wq
+    neg     wq
+%%.loop:
+%if %2
+    mova    m1, [srcq+wq]
+%else
+    movu    m1, [srcq+wq]
+%endif
+    mova    m2, m1
+    psllw   m1, 8
+    paddb   m1, m2
+    mova    m2, m1
+    pshufb  m1, m3
+    paddb   m1, m2
+    pshufb  m0, m5
+    mova    m2, m1
+    pshufb  m1, m4
+    paddb   m1, m2
+%if mmsize == 16
+    mova    m2, m1
+    pshufb  m1, m6
+    paddb   m1, m2
+%endif
+    paddb   m0, m1
+%if %1
+    mova    [dstq+wq], m0
+%else
+    movq    [dstq+wq], m0
+    movhps  [dstq+wq+8], m0
+%endif
+    add     wq, mmsize
+    jl %%.loop
+    mov     eax, mmsize-1
+    sub     eax, wd
+    movd    m1, eax
+    pshufb  m0, m1
+    movd    eax, m0
+    RET
+%endmacro
+
+; int ff_add_left_pred(uint8_t *dst, const uint8_t *src, int w, int left)
+INIT_MMX ssse3
+cglobal add_left_pred, 3,3,7, dst, src, w, left
+.skip_prologue:
+    mova    m5, [pb_7]
+    mova    m4, [pb_zzzz3333zzzzbbbb]
+    mova    m3, [pb_zz11zz55zz99zzdd]
+    movd    m0, leftm
+    psllq   m0, 56
+    ADD_LEFT_LOOP 1, 1
+
+INIT_XMM sse4
+cglobal add_left_pred, 3,3,7, dst, src, w, left
+    mova    m5, [pb_15]
+    mova    m6, [pb_zzzzzzzz77777777]
+    mova    m4, [pb_zzzz3333zzzzbbbb]
+    mova    m3, [pb_zz11zz55zz99zzdd]
+    movd    m0, leftm
+    pslldq  m0, 15
+    test    srcq, 15
+    jnz .src_unaligned
+    test    dstq, 15
+    jnz .dst_unaligned
+    ADD_LEFT_LOOP 1, 1
+.dst_unaligned:
+    ADD_LEFT_LOOP 0, 1
+.src_unaligned:
+    ADD_LEFT_LOOP 0, 0
+
+%macro ADD_BYTES 0
+cglobal add_bytes, 3,4,2, dst, src, w, size
+    mov  sizeq, wq
+    and  sizeq, -2*mmsize
+    jz  .2
+    add   dstq, sizeq
+    add   srcq, sizeq
+    neg  sizeq
+.1:
+    mova    m0, [srcq + sizeq]
+    mova    m1, [srcq + sizeq + mmsize]
+    paddb   m0, [dstq + sizeq]
+    paddb   m1, [dstq + sizeq + mmsize]
+    mova   [dstq + sizeq], m0
+    mova   [dstq + sizeq + mmsize], m1
+    add  sizeq, 2*mmsize
+    jl .1
+.2:
+    and     wq, 2*mmsize-1
+    jz    .end
+    add   dstq, wq
+    add   srcq, wq
+    neg     wq
+.3:
+    mov  sizeb, [srcq + wq]
+    add [dstq + wq], sizeb
+    inc     wq
+    jl .3
+.end:
+    REP_RET
+%endmacro
+
+%if ARCH_X86_32
+INIT_MMX mmx
+ADD_BYTES
+%endif
+INIT_XMM sse2
+ADD_BYTES
+
 %macro INT16_LOOP 2 ; %1 = a/u (aligned/unaligned), %2 = add/sub
     movd    m4, maskd
     SPLATW  m4, m4
@@ -86,9 +272,11 @@ SECTION .text
     RET
 %endmacro
 
+%if ARCH_X86_32
 INIT_MMX mmx
 cglobal add_int16, 4,4,5, dst, src, mask, w, tmp
     INT16_LOOP a, add
+%endif
 
 INIT_XMM sse2
 cglobal add_int16, 4,4,5, dst, src, mask, w, tmp
@@ -100,9 +288,11 @@ cglobal add_int16, 4,4,5, dst, src, mask, w, tmp
 .unaligned:
     INT16_LOOP u, add
 
+%if ARCH_X86_32
 INIT_MMX mmx
 cglobal diff_int16, 5,5,5, dst, src1, src2, mask, w, tmp
     INT16_LOOP a, sub
+%endif
 
 INIT_XMM sse2
 cglobal diff_int16, 5,5,5, dst, src1, src2, mask, w, tmp
diff --git a/libavcodec/x86/lossless_videodsp_init.c b/libavcodec/x86/lossless_videodsp_init.c
index 548d0433e1..465feef81b 100644
--- a/libavcodec/x86/lossless_videodsp_init.c
+++ b/libavcodec/x86/lossless_videodsp_init.c
@@ -18,10 +18,27 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
+#include "config.h"
+#include "libavutil/x86/asm.h"
 #include "../lossless_videodsp.h"
 #include "libavutil/pixdesc.h"
 #include "libavutil/x86/cpu.h"
 
+void ff_add_bytes_mmx(uint8_t *dst, uint8_t *src, intptr_t w);
+void ff_add_bytes_sse2(uint8_t *dst, uint8_t *src, intptr_t w);
+
+void ff_add_median_pred_mmxext(uint8_t *dst, const uint8_t *top,
+                               const uint8_t *diff, intptr_t w,
+                               int *left, int *left_top);
+void ff_add_median_pred_sse2(uint8_t *dst, const uint8_t *top,
+                             const uint8_t *diff, intptr_t w,
+                             int *left, int *left_top);
+
+int  ff_add_left_pred_ssse3(uint8_t *dst, const uint8_t *src,
+                            intptr_t w, int left);
+int  ff_add_left_pred_sse4(uint8_t *dst, const uint8_t *src,
+                            intptr_t w, int left);
+
 void ff_add_int16_mmx(uint16_t *dst, const uint16_t *src, unsigned mask, int w);
 void ff_add_int16_sse2(uint16_t *dst, const uint16_t *src, unsigned mask, int w);
 void ff_diff_int16_mmx (uint16_t *dst, const uint16_t *src1, const uint16_t *src2, unsigned mask, int w);
@@ -31,28 +48,85 @@ int ff_add_hfyu_left_pred_int16_sse4(uint16_t *dst, const uint16_t *src, unsigne
 void ff_add_hfyu_median_pred_int16_mmxext(uint16_t *dst, const uint16_t *top, const uint16_t *diff, unsigned mask, int w, int *left, int *left_top);
 void ff_sub_hfyu_median_pred_int16_mmxext(uint16_t *dst, const uint16_t *src1, const uint16_t *src2, unsigned mask, int w, int *left, int *left_top);
 
+#if HAVE_INLINE_ASM && HAVE_7REGS && ARCH_X86_32
+static void add_median_pred_cmov(uint8_t *dst, const uint8_t *top,
+                                 const uint8_t *diff, intptr_t w,
+                                 int *left, int *left_top)
+{
+    x86_reg w2 = -w;
+    x86_reg x;
+    int l  = *left     & 0xff;
+    int tl = *left_top & 0xff;
+    int t;
+    __asm__ volatile (
+        "mov          %7, %3            \n"
+        "1:                             \n"
+        "movzbl (%3, %4), %2            \n"
+        "mov          %2, %k3           \n"
+        "sub         %b1, %b3           \n"
+        "add         %b0, %b3           \n"
+        "mov          %2, %1            \n"
+        "cmp          %0, %2            \n"
+        "cmovg        %0, %2            \n"
+        "cmovg        %1, %0            \n"
+        "cmp         %k3, %0            \n"
+        "cmovg       %k3, %0            \n"
+        "mov          %7, %3            \n"
+        "cmp          %2, %0            \n"
+        "cmovl        %2, %0            \n"
+        "add    (%6, %4), %b0           \n"
+        "mov         %b0, (%5, %4)      \n"
+        "inc          %4                \n"
+        "jl           1b                \n"
+        : "+&q"(l), "+&q"(tl), "=&r"(t), "=&q"(x), "+&r"(w2)
+        : "r"(dst + w), "r"(diff + w), "rm"(top + w)
+    );
+    *left     = l;
+    *left_top = tl;
+}
+#endif
 
 void ff_llviddsp_init_x86(LLVidDSPContext *c, AVCodecContext *avctx)
 {
     int cpu_flags = av_get_cpu_flags();
     const AVPixFmtDescriptor *pix_desc = av_pix_fmt_desc_get(avctx->pix_fmt);
 
-    if (EXTERNAL_MMX(cpu_flags)) {
+#if HAVE_INLINE_ASM && HAVE_7REGS && ARCH_X86_32
+    if (cpu_flags & AV_CPU_FLAG_CMOV)
+        c->add_median_pred = add_median_pred_cmov;
+#endif
+
+    if (ARCH_X86_32 && EXTERNAL_MMX(cpu_flags)) {
+        c->add_bytes = ff_add_bytes_mmx;
+
         c->add_int16 = ff_add_int16_mmx;
         c->diff_int16 = ff_diff_int16_mmx;
     }
 
+    if (ARCH_X86_32 && EXTERNAL_MMXEXT(cpu_flags)) {
+        /* slower than cmov version on AMD */
+        if (!(cpu_flags & AV_CPU_FLAG_3DNOW))
+            c->add_median_pred = ff_add_median_pred_mmxext;
+    }
+
     if (EXTERNAL_MMXEXT(cpu_flags) && pix_desc && pix_desc->comp[0].depth<16) {
         c->add_hfyu_median_pred_int16 = ff_add_hfyu_median_pred_int16_mmxext;
         c->sub_hfyu_median_pred_int16 = ff_sub_hfyu_median_pred_int16_mmxext;
     }
 
     if (EXTERNAL_SSE2(cpu_flags)) {
+        c->add_bytes       = ff_add_bytes_sse2;
+        c->add_median_pred = ff_add_median_pred_sse2;
+
         c->add_int16 = ff_add_int16_sse2;
         c->diff_int16 = ff_diff_int16_sse2;
     }
 
     if (EXTERNAL_SSSE3(cpu_flags)) {
+        c->add_left_pred = ff_add_left_pred_ssse3;
+        if (cpu_flags & AV_CPU_FLAG_SSE4) // not really SSE4, just slow on Conroe
+            c->add_left_pred = ff_add_left_pred_sse4;
+
         c->add_hfyu_left_pred_int16 = ff_add_hfyu_left_pred_int16_ssse3;
     }