]> git.sesse.net Git - ffmpeg/commitdiff
intreadwrite: Add intermediate variables in the byteswise AV_W*() macros
authorMartin Storsjö <martin@martin.st>
Mon, 1 Aug 2016 06:07:48 +0000 (09:07 +0300)
committerMartin Storsjö <martin@martin.st>
Tue, 2 Aug 2016 08:32:35 +0000 (11:32 +0300)
This avoids issues with expanding the argument multiple times,
and makes sure that it is of the right type for the following shifts.

Even if the caller of a macro could be expected not to pass parameters
that have side effects if expanded multiple times, these fallback
codepaths are rarely, if ever, tested, so it is expected that such
issues can arise.

Thefore, for safety, make sure the fallback codepaths only expand
the arguments once.

Signed-off-by: Martin Storsjö <martin@martin.st>
libavutil/intreadwrite.h

index cdf1ef47ec7bbae2c63939fc5250685abc9e2fe7..5f65957ce5f3b1125ca26e1d28386a5218b0905f 100644 (file)
@@ -210,7 +210,8 @@ union unaligned_16 { uint16_t l; } __attribute__((packed)) av_alias;
       ((const uint8_t*)(x))[1])
 #endif
 #ifndef AV_WB16
-#   define AV_WB16(p, d) do {                   \
+#   define AV_WB16(p, val) do {                 \
+        uint16_t d = val;                       \
         ((uint8_t*)(p))[1] = (d);               \
         ((uint8_t*)(p))[0] = (d)>>8;            \
     } while(0)
@@ -222,7 +223,8 @@ union unaligned_16 { uint16_t l; } __attribute__((packed)) av_alias;
       ((const uint8_t*)(x))[0])
 #endif
 #ifndef AV_WL16
-#   define AV_WL16(p, d) do {                   \
+#   define AV_WL16(p, val) do {                 \
+        uint16_t d = val;                       \
         ((uint8_t*)(p))[0] = (d);               \
         ((uint8_t*)(p))[1] = (d)>>8;            \
     } while(0)
@@ -236,7 +238,8 @@ union unaligned_16 { uint16_t l; } __attribute__((packed)) av_alias;
                 ((const uint8_t*)(x))[3])
 #endif
 #ifndef AV_WB32
-#   define AV_WB32(p, d) do {                   \
+#   define AV_WB32(p, val) do {                 \
+        uint32_t d = val;                       \
         ((uint8_t*)(p))[3] = (d);               \
         ((uint8_t*)(p))[2] = (d)>>8;            \
         ((uint8_t*)(p))[1] = (d)>>16;           \
@@ -252,7 +255,8 @@ union unaligned_16 { uint16_t l; } __attribute__((packed)) av_alias;
                 ((const uint8_t*)(x))[0])
 #endif
 #ifndef AV_WL32
-#   define AV_WL32(p, d) do {                   \
+#   define AV_WL32(p, val) do {                 \
+        uint32_t d = val;                       \
         ((uint8_t*)(p))[0] = (d);               \
         ((uint8_t*)(p))[1] = (d)>>8;            \
         ((uint8_t*)(p))[2] = (d)>>16;           \
@@ -272,7 +276,8 @@ union unaligned_16 { uint16_t l; } __attribute__((packed)) av_alias;
       (uint64_t)((const uint8_t*)(x))[7])
 #endif
 #ifndef AV_WB64
-#   define AV_WB64(p, d) do {                   \
+#   define AV_WB64(p, val) do {                 \
+        uint64_t d = val;                       \
         ((uint8_t*)(p))[7] = (d);               \
         ((uint8_t*)(p))[6] = (d)>>8;            \
         ((uint8_t*)(p))[5] = (d)>>16;           \
@@ -296,7 +301,8 @@ union unaligned_16 { uint16_t l; } __attribute__((packed)) av_alias;
       (uint64_t)((const uint8_t*)(x))[0])
 #endif
 #ifndef AV_WL64
-#   define AV_WL64(p, d) do {                   \
+#   define AV_WL64(p, val) do {                 \
+        uint64_t d = val;                       \
         ((uint8_t*)(p))[0] = (d);               \
         ((uint8_t*)(p))[1] = (d)>>8;            \
         ((uint8_t*)(p))[2] = (d)>>16;           \