On 04/01/2013 12:38 PM, Tanu Kaskinen wrote:
> On Fri, 2013-03-29 at 16:56 +0100, Peter Meerwald wrote:
>> + case PA_SAMPLE_S16NE:
>> + {
>> + int16_t *d = (int16_t *) dst, *s = (int16_t *) src;
>> +
>> + for (i = n>> 2; i> 0; i--) {
>> + d[0] += (s[0] + s[1])/2;
>> + d[1] += (s[2] + s[3])/2;
>> + d[2] += (s[4] + s[5])/2;
>> + d[3] += (s[6] + s[7])/2;
>
> You probably meant '=', not '+='?
>
> Also, s[0] + s[1] can overflow, so the inputs should be divided
> individually before summing.
Or the inputs could be cast to uint32_t, which I guess would be better
than dividing multiple times.
--
Tanu
