avcodec/me_cmp: Fix crashes on ARM due to misalignment
authorMichael Niedermayer <michael@niedermayer.cc>
Sat, 19 Aug 2017 21:38:58 +0000 (23:38 +0200)
committerMichael Niedermayer <michael@niedermayer.cc>
Sat, 9 Sep 2017 23:33:28 +0000 (01:33 +0200)
Adds a diff_pixels_unaligned()

Fixes: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=872503

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
(cherry picked from commit bc488ec28aec4bc91ba47283c49c9f7f25696eaa)
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
libavcodec/me_cmp.c
libavcodec/pixblockdsp.c
libavcodec/pixblockdsp.h
libavcodec/x86/pixblockdsp_init.c

index 6639b91..5e34a11 100644 (file)
@@ -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                               =
index 417c944..50e1d1d 100644 (file)
@@ -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) {
index 3ba1596..e036700 100644 (file)
@@ -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);
index fa9578a..ade55e0 100644 (file)
@@ -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;
     }
 }