Is there any way to avoid a huge memory spike when using CGContextDrawImage for a high resolution image?

Question

Hi,
I'm trying to fix the orientation of an UIImage in my app (iPad 2, iOS 7). I'm using the method described here: https://gist.github.com/alex-cellcity/1531596, but a huge memory spike appears when CGContextDrawImage is called. The image size I'm trying to fix is 2848x4288.
Anybody knows any algorithm to avoid this huge spike?
Thanks,
Roberto.

Answer 1

While I never done iOS code it appears that the code you link to basically does the following:
takes a large image A
generates a transformation matrix by doing rotation in 90 degree increments and by mirroring
creates a large image B
draws image A into B using the transformation matrix
releases memory for image A
goes on using image B
This results in a memory peak as you have two large images in RAM, until image A gets deallocated, which can't be done until B has been drawn.
I can see two possible solutions to this, both which probably require you to explicitly modify individual pixels:
1) Try to simplify the solution - if 180 degree rotation and mirroring is enough to solve your image issues, then a simple in-place (same bitmap) pixel swap could be done.
2) Find more memory -  if you need 90, 270 or arbitrary degrees of rotation as well, you could instead keep either image A or B on disc.
Keep image A on disc: read pixels from disc and place appropriately in image B. This may require you to first load image A and write it back to disc as ARGB data (one byte for each color) for easy access.
Keep image B on disc: get pixel from image A and place it at the right location in image (file) B - this will require pre-creating a file of the correct size for Bs ARGB data and then moving the insertion point for writing around. When done dealloc image A and load image B.
Keep image A and B on disc: combine using disc for image A and B - this will minimize RAM usage but will probably be the slowest solution and will increases disc (SSD/flash) load and wear.