This issue was created by revision r339.

When assembling tiles, we first decode the tile data (first alloc), then copy this data into the large image (already allocated) that will be scaled up or down into the resulting image. In total, we allocate double the memory of the tile data. The image decoding code needs to be updated so it can directly populate an allocated image, rather than aloocating it itself.

This issue was closed by revision r362.