[FFmpeg-cvslog] avcodec/me_cmp: Fix crashes on ARM due to misalignment

[Michael Niedermayer]

ffmpeg | branch: release/3.3 | Michael Niedermayer <michael at niedermayer.cc> | Sat Aug 19 23:38:58 2017 +0200| [4a122a0879d5d237cdfc4087871643a884be5ce3] | committer: Michael Niedermayer

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=872503

Signed-off-by: Michael Niedermayer <michael at niedermayer.cc>
(cherry picked from commit bc488ec28aec4bc91ba47283c49c9f7f25696eaa)
Signed-off-by: Michael Niedermayer <michael at niedermayer.cc>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=4a122a0879d5d237cdfc4087871643a884be5ce3
---

 libavcodec/me_cmp.c               | 10 +++++-----
 libavcodec/pixblockdsp.c          |  1 +
 libavcodec/pixblockdsp.h          |  5 +++++
 libavcodec/x86/pixblockdsp_init.c |  2 ++
 4 files changed, 13 insertions(+), 5 deletions(-)

diff --git a/libavcodec/me_cmp.c b/libavcodec/me_cmp.c
index 6639b919ff..5e34a11593 100644
--- a/libavcodec/me_cmp.c
+++ b/libavcodec/me_cmp.c
@@ -628,7 +628,7 @@ 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);
 }
@@ -668,7 +668,7 @@ 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
@@ -695,7 +695,7 @@ 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++)
@@ -714,7 +714,7 @@ 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));
 
@@ -817,7 +817,7 @@ 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                               =
diff --git a/libavcodec/pixblockdsp.c b/libavcodec/pixblockdsp.c
index 417c944e00..50e1d1d735 100644
--- a/libavcodec/pixblockdsp.c
+++ b/libavcodec/pixblockdsp.c
@@ -82,6 +82,7 @@ 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) {
diff --git a/libavcodec/pixblockdsp.h b/libavcodec/pixblockdsp.h
index 3ba1596a88..e036700ff0 100644
--- a/libavcodec/pixblockdsp.h
+++ b/libavcodec/pixblockdsp.h
@@ -33,6 +33,11 @@ typedef struct PixblockDSPContext {
                         const uint8_t *s1 /* align 8 */,
                         const uint8_t *s2 /* align 8 */,
                         ptrdiff_t stride);
+    void (*diff_pixels_unaligned)(int16_t *av_restrict block /* align 16 */,
+                        const uint8_t *s1,
+                        const uint8_t *s2,
+                        ptrdiff_t stride);
+
 } PixblockDSPContext;
 
 void ff_pixblockdsp_init(PixblockDSPContext *c, AVCodecContext *avctx);
diff --git a/libavcodec/x86/pixblockdsp_init.c b/libavcodec/x86/pixblockdsp_init.c
index fa9578a2d3..ade55e01a3 100644
--- a/libavcodec/x86/pixblockdsp_init.c
+++ b/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;
     }
 }