dsputil: make add_hfyu_left_prediction_sse4() support unaligned src.

This makes add_hfyu_left_prediction_sse4() handle sources that are not
16-byte aligned in its own function rather than by proxying the call to
add_hfyu_left_prediction_ssse3(). This fixes a crash on Win64, since the
sse4 version clobberes xmm6, but the ssse3 version (which uses MMX regs)
does not restore it, thus leading to XMM clobbering and RSP being off.

Fixes bug 342.

libavcodec/x86/dsputil_yasm.asm

@@ -388,12 +388,16 @@ cglobal add_hfyu_median_prediction_mmx2, 6,6,0, dst, top, diff, w, left, left_to
     RET
 
 
-%macro ADD_HFYU_LEFT_LOOP 1 ; %1 = is_aligned
+%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
@@ -435,7 +439,7 @@ cglobal add_hfyu_left_prediction_ssse3, 3,3,7, dst, src, w, left
     mova    m3, [pb_zz11zz55zz99zzdd]
     movd    m0, leftm
     psllq   m0, 56
-    ADD_HFYU_LEFT_LOOP 1
+    ADD_HFYU_LEFT_LOOP 1, 1
 
 INIT_XMM
 cglobal add_hfyu_left_prediction_sse4, 3,3,7, dst, src, w, left
@@ -446,12 +450,14 @@ cglobal add_hfyu_left_prediction_sse4, 3,3,7, dst, src, w, left
     movd    m0, leftm
     pslldq  m0, 15
     test    srcq, 15
-    jnz add_hfyu_left_prediction_ssse3.skip_prologue
+    jnz .src_unaligned
     test    dstq, 15
-    jnz .unaligned
-    ADD_HFYU_LEFT_LOOP 1
-.unaligned:
-    ADD_HFYU_LEFT_LOOP 0
+    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
 
 
 ; float scalarproduct_float_sse(const float *v1, const float *v2, int len)