In the Galaxy S3 (GT-I9300), the video output buffer pitch is equal to the requested (width*4).When the width is a multiple of 2 but not 4. The buffer is not 128 bit aligned. This results in an unaligned write to the buffer causing a SIGBUS code 128. This is fixed by setting the vst4 as 64 bit aligned (2 pixels). I believe it isn't possible to have a width which isn't a multiple of 2.
In arm_neon/1420_rgb.S, the WIDTH is rounded up to a multiple of 16. This will Overflow the buffer at the end of the of the row if the OPITCH is not greater than the WIDTH rounded up.
My current fix would stop the crashes, but would leave up to 14 pixels blank at the end of the row if the OPITCH is not a multiple of 16.
PATCH: http://pastebin.com/gRKewmZF