avcodec/me_cmp: Fix crashes on ARM due to misalignment

Adds a diff_pixels_unaligned()

Fixes: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=872503Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
(cherry picked from commit bc488ec2)
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

libavcodec/me_cmp.c

@@ -555,7 +555,7 @@ static int dct_sad8x8_c(MpegEncContext *s, uint8_t *src1,
 
     av_assert2(h == 8);
 
-    s->pdsp.diff_pixels(temp, src1, src2, stride);
+    s->pdsp.diff_pixels_unaligned(temp, src1, src2, stride);
     s->fdsp.fdct(temp);
     return s->mecc.sum_abs_dctelem(temp);
 }
@@ -595,7 +595,7 @@ static int dct264_sad8x8_c(MpegEncContext *s, uint8_t *src1,
     int16_t dct[8][8];
     int i, sum = 0;
 
-    s->pdsp.diff_pixels(dct[0], src1, src2, stride);
+    s->pdsp.diff_pixels_unaligned(dct[0], src1, src2, stride);
 
 #define SRC(x) dct[i][x]
 #define DST(x, v) dct[i][x] = v
@@ -622,7 +622,7 @@ static int dct_max8x8_c(MpegEncContext *s, uint8_t *src1,
 
     av_assert2(h == 8);
 
-    s->pdsp.diff_pixels(temp, src1, src2, stride);
+    s->pdsp.diff_pixels_unaligned(temp, src1, src2, stride);
     s->fdsp.fdct(temp);
 
     for (i = 0; i < 64; i++)
@@ -641,7 +641,7 @@ static int quant_psnr8x8_c(MpegEncContext *s, uint8_t *src1,
     av_assert2(h == 8);
     s->mb_intra = 0;
 
-    s->pdsp.diff_pixels(temp, src1, src2, stride);
+    s->pdsp.diff_pixels_unaligned(temp, src1, src2, stride);
 
     memcpy(bak, temp, 64 * sizeof(int16_t));
 
@@ -744,7 +744,7 @@ static int bit8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,
 
     av_assert2(h == 8);
 
-    s->pdsp.diff_pixels(temp, src1, src2, stride);
+    s->pdsp.diff_pixels_unaligned(temp, src1, src2, stride);
 
     s->block_last_index[0 /* FIXME */] =
     last                               =

libavcodec/pixblockdsp.c

@@ -55,6 +55,7 @@ av_cold void ff_pixblockdsp_init(PixblockDSPContext *c, AVCodecContext *avctx)
 {
     const unsigned high_bit_depth = avctx->bits_per_raw_sample > 8;
 
+    c->diff_pixels_unaligned =
     c->diff_pixels = diff_pixels_c;
 
     switch (avctx->bits_per_raw_sample) {

libavcodec/pixblockdsp.h

@@ -31,6 +31,11 @@ typedef struct PixblockDSPContext {
                         const uint8_t *s1 /* align 8 */,
                         const uint8_t *s2 /* align 8 */,
                         int stride);
+    void (*diff_pixels_unaligned)(int16_t *av_restrict block /* align 16 */,
+                        const uint8_t *s1,
+                        const uint8_t *s2,
+                        int stride);
+
 } PixblockDSPContext;
 
 void ff_pixblockdsp_init(PixblockDSPContext *c, AVCodecContext *avctx);

libavcodec/x86/pixblockdsp_init.c

@@ -39,12 +39,14 @@ av_cold void ff_pixblockdsp_init_x86(PixblockDSPContext *c,
     if (EXTERNAL_MMX(cpu_flags)) {
         if (!high_bit_depth)
             c->get_pixels = ff_get_pixels_mmx;
+        c->diff_pixels_unaligned =
         c->diff_pixels = ff_diff_pixels_mmx;
     }
 
     if (EXTERNAL_SSE2(cpu_flags)) {
         if (!high_bit_depth)
             c->get_pixels = ff_get_pixels_sse2;
+        c->diff_pixels_unaligned =
         c->diff_pixels = ff_diff_pixels_sse2;
     }
 }