Question

I’m writing a small painting app for iOS. I’m subclassing a UIView ad performing computations inside its drawRect: method. It’s been all good until I started having lots of objects (actually polylines), then performance started to degrade.

- (void)drawRect:(CGRect)rect {
    [super drawRect:rect];

    CGContextRef imageContext = UIGraphicsGetCurrentContext();

    int i, j;
    for(i = 0; i < [_strokes count]; i++) {
        NSMutableArray *stroke = [_strokes objectAtIndex:i];
        if([stroke count] < 2) {
            continue;
        }
        CGContextSetStrokeColorWithColor(imageContext, [[_penColours objectAtIndex:i] CGColor]);
        for(j = 0; j < [stroke count]-1; j++) {
            CGPoint line[] = {
                [[stroke objectAtIndex:j] CGPointValue],
                [[stroke objectAtIndex:j+1] CGPointValue]
            };

            CGContextSetLineWidth(imageContext, _brushSize);
            CGContextSetLineCap(imageContext, kCGLineCapRound);
            CGContextSetLineJoin(imageContext, kCGLineJoinRound);

            CGContextAddLines(imageContext, line, 2);
            CGContextStrokePath(imageContext);
        }
    }

    CGContextFlush(imageContext);
}

where _strokes is an array of arrays of points. My first idea was to make an image as an ivar, paint it on the context, paint the additional strokes to the context (change j = 0 with j = [stroke count] - 2), and grab the context to an image and store it back to the ivar. Didn’t work.

Then I tried many other roads not really worth mentioning, until I found this other question on SO ( Quartz2D performance – how to improve ). Unfortunately it didn’t work as I expected as I had to retain the image, resulting in memory warning 1, 2, crash.

- (void)drawRect:(CGRect)rect {
    [super drawRect:rect];

    UIGraphicsBeginImageContext(CGSizeMake(1024, 768));
    CGContextRef imageContext = UIGraphicsGetCurrentContext();

    if(_image != nil) {
        CGContextDrawImage(imageContext, CGRectMake(0, 0, 1024, 768), [_image CGImage]);
    }

    int i, j;
    for(i = 0; i < [_strokes count]; i++) {
        NSMutableArray *stroke = [_strokes objectAtIndex:i];
        if([stroke count] < 2) {
            continue;
        }
        CGContextSetStrokeColorWithColor(imageContext, [[_penColours objectAtIndex:i] CGColor]);
        for(j = [stroke count]-2; j < [stroke count]-1; j++) {
            CGPoint line[] = {
                [[stroke objectAtIndex:j] CGPointValue],
                [[stroke objectAtIndex:j+1] CGPointValue]
            };

            CGContextSetLineWidth(imageContext, _brushSize);
            CGContextSetLineCap(imageContext, kCGLineCapRound);
            CGContextSetLineJoin(imageContext, kCGLineJoinRound);

            CGContextAddLines(imageContext, line, 2);
            CGContextStrokePath(imageContext);
        }
    }

    _image = UIGraphicsGetImageFromCurrentImageContext();
    [_image retain];
    UIGraphicsEndImageContext();

    CGContextRef context = UIGraphicsGetCurrentContext();
    CGContextDrawImage(context, CGRectMake(0, 0, 1024, 768), [_image CGImage]);
}

Answer 1

It looks like at least part of the problem is that you retain _image but you never release it. Each time you assign a new value to _image, you’re leaking the previous image, and that’ll fill up memory very quickly.

Beyond that, you should probably be using a CGLayer for your offscreen drawing rather than a bitmap. You can draw into the layer whenever you like, and then just copy it to the screen in your -drawRect: method.