[FFmpeg-cvslog] dirac/x86: Fix handling blocksizes that are not a multiple of 4

[Michael Niedermayer]

ffmpeg | branch: master | Michael Niedermayer <michaelni at gmx.at> | Mon Jan 28 20:40:13 2013 +0100| [94ef1667bb04ed81ff10f7ba4b8d7e54bd8bc76b] | committer: Michael Niedermayer

dirac/x86: Fix handling blocksizes that are not a multiple of 4

Fixes out of array accesses

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michaelni at gmx.at>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=94ef1667bb04ed81ff10f7ba4b8d7e54bd8bc76b
---

 libavcodec/x86/dsputil_mmx.c |   32 ++++++++++++++++++++++++++++----
 1 file changed, 28 insertions(+), 4 deletions(-)

diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
index 303d36d..dcf2345 100644
--- a/libavcodec/x86/dsputil_mmx.c
+++ b/libavcodec/x86/dsputil_mmx.c
@@ -1612,16 +1612,26 @@ void ff_avg_vc1_mspel_mc00_mmxext(uint8_t *dst, const uint8_t *src,
 #define DIRAC_PIXOP(OPNAME2, OPNAME, EXT)\
 void ff_ ## OPNAME2 ## _dirac_pixels8_ ## EXT(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
 {\
-    OPNAME ## _pixels8_ ## EXT(dst, src[0], stride, h);\
+    if (h&3)\
+        ff_ ## OPNAME2 ## _dirac_pixels8_c(dst, src, stride, h);\
+    else\
+        OPNAME ## _pixels8_ ## EXT(dst, src[0], stride, h);\
 }\
 void ff_ ## OPNAME2 ## _dirac_pixels16_ ## EXT(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
 {\
-    OPNAME ## _pixels16_ ## EXT(dst, src[0], stride, h);\
+    if (h&3)\
+        ff_ ## OPNAME2 ## _dirac_pixels16_c(dst, src, stride, h);\
+    else\
+        OPNAME ## _pixels16_ ## EXT(dst, src[0], stride, h);\
 }\
 void ff_ ## OPNAME2 ## _dirac_pixels32_ ## EXT(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
 {\
-    OPNAME ## _pixels16_ ## EXT(dst   , src[0]   , stride, h);\
-    OPNAME ## _pixels16_ ## EXT(dst+16, src[0]+16, stride, h);\
+    if (h&3) {\
+        ff_ ## OPNAME2 ## _dirac_pixels32_c(dst, src, stride, h);\
+    } else {\
+        OPNAME ## _pixels16_ ## EXT(dst   , src[0]   , stride, h);\
+        OPNAME ## _pixels16_ ## EXT(dst+16, src[0]+16, stride, h);\
+    }\
 }
 
 DIRAC_PIXOP(put, put, mmx)
@@ -1631,21 +1641,35 @@ DIRAC_PIXOP(avg, ff_avg, mmxext)
 #if HAVE_YASM
 void ff_put_dirac_pixels16_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h)
 {
+    if (h&3)
+        ff_put_dirac_pixels16_c(dst, src, stride, h);
+    else
     ff_put_pixels16_sse2(dst, src[0], stride, h);
 }
 void ff_avg_dirac_pixels16_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h)
 {
+    if (h&3)
+        ff_avg_dirac_pixels16_c(dst, src, stride, h);
+    else
     ff_avg_pixels16_sse2(dst, src[0], stride, h);
 }
 void ff_put_dirac_pixels32_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h)
 {
+    if (h&3) {
+        ff_put_dirac_pixels32_c(dst, src, stride, h);
+    } else {
     ff_put_pixels16_sse2(dst   , src[0]   , stride, h);
     ff_put_pixels16_sse2(dst+16, src[0]+16, stride, h);
+    }
 }
 void ff_avg_dirac_pixels32_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h)
 {
+    if (h&3) {
+        ff_avg_dirac_pixels32_c(dst, src, stride, h);
+    } else {
     ff_avg_pixels16_sse2(dst   , src[0]   , stride, h);
     ff_avg_pixels16_sse2(dst+16, src[0]+16, stride, h);
+    }
 }
 #endif
 #endif