avcodec/huffyuvencdsp: Fix load of misaligned values

Affected many ffvhuff FATE tests.

Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>

libavcodec/huffyuvencdsp.c

@@ -18,16 +18,32 @@
 
 #include "config.h"
 #include "libavutil/attributes.h"
+#include "libavutil/intreadwrite.h"
 #include "huffyuvencdsp.h"
 #include "mathops.h"
 
+#if HAVE_FAST_64BIT
+#define BITS 64
+typedef uint64_t uint_native;
+#else
+#define BITS 32
+typedef uint32_t uint_native;
+#endif
+#define RN  AV_JOIN(AV_RN, BITS)
+#define RNA AV_JOIN(AV_JOIN(AV_RN, BITS), A)
+#define WNA AV_JOIN(AV_JOIN(AV_WN, BITS), A)
+
+// 0x7f7f7f7f or 0x7f7f7f7f7f7f7f7f or whatever, depending on the cpu's native arithmetic size
+#define pb_7f (~(uint_native)0 / 255 * 0x7f)
+#define pb_80 (~(uint_native)0 / 255 * 0x80)
+
 // 0x00010001 or 0x0001000100010001 or whatever, depending on the cpu's native arithmetic size
-#define pw_1 (ULONG_MAX / UINT16_MAX)
+#define pw_1 ((uint_native)-1 / UINT16_MAX)
 
 static void diff_int16_c(uint16_t *dst, const uint16_t *src1, const uint16_t *src2, unsigned mask, int w){
     long i;
 #if !HAVE_FAST_UNALIGNED
-    if((long)src2 & (sizeof(long)-1)){
+    if ((uintptr_t)src2 & (sizeof(uint_native) - 1)) {
         for(i=0; i+3<w; i+=4){
             dst[i+0] = (src1[i+0]-src2[i+0]) & mask;
             dst[i+1] = (src1[i+1]-src2[i+1]) & mask;
@@ -37,13 +53,13 @@ static void diff_int16_c(uint16_t *dst, const uint16_t *src1, const uint16_t *sr
     }else
 #endif
     {
-        unsigned long pw_lsb = (mask >> 1) * pw_1;
-        unsigned long pw_msb = pw_lsb +  pw_1;
+        uint_native pw_lsb = (mask >> 1) * pw_1;
+        uint_native pw_msb = pw_lsb +  pw_1;
 
-        for (i = 0; i <= w - (int)sizeof(long)/2; i += sizeof(long)/2) {
-            long a = *(long*)(src1+i);
-            long b = *(long*)(src2+i);
-            *(long*)(dst+i) = ((a|pw_msb) - (b&pw_lsb)) ^ ((a^b^pw_msb)&pw_msb);
+        for (i = 0; i <= w - (int)sizeof(uint_native)/2; i += sizeof(uint_native)/2) {
+            uint_native a = RNA(src1 + i);
+            uint_native b = RN (src2 + i);
+            WNA(dst + i, ((a | pw_msb) - (b & pw_lsb)) ^ ((a^b^pw_msb) & pw_msb));
         }
     }
     for (; i<w; i++)