avcodec/jpeg2000dec: Use 32x32->64bit for 9/7i dequantization
authorMichael Niedermayer <michaelni@gmx.at>
Wed, 24 Jun 2015 15:07:27 +0000 (17:07 +0200)
committerMichael Niedermayer <michaelni@gmx.at>
Wed, 24 Jun 2015 15:09:21 +0000 (17:09 +0200)
Like the 5/3 case this is needed to avoid overflows and similarly for 16bpp
output pure 32bit operations are insufficient if high quality is wanted

Note, this code-path is only used in bitexact mode, so this should not
affect the speed of any real use-case

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
libavcodec/jpeg2000dec.c

index 11f4cd6..43568ea 100644 (file)
@@ -1457,7 +1457,7 @@ static void dequantization_int_97(int x, int y, Jpeg2000Cblk *cblk,
         int32_t *datap = &comp->i_data[(comp->coord[0][1] - comp->coord[0][0]) * (y + j) + x];
         int *src = t1->data[j];
         for (i = 0; i < w; ++i)
-            datap[i] = (src[i] * band->i_stepsize + (1<<14)) >> 15;
+            datap[i] = (src[i] * (int64_t)band->i_stepsize + (1<<14)) >> 15;
     }
 }