Iphone – Performing iPhone optimization on externally downloaded PNGs

iphoneoptimizationpng

When a PNG is added to an XCode iPhone project, the compiler optimizes it using pngcrush. Once on the device, the image's rendering performance is very fast.

My problem is that my application downloads its PNGs from an external source at runtime (from Picasa Web albums, using the Google Data APIs). Unfortunately, these images' performance is quite bad. When I do custom rendering on top of the image, it seems 100x slower than its internally stored counterparts. I strongly suspect this is because the downloaded images haven't been optimized.

Does anyone know how I can optimize an externally downloaded PNG at runtime on the iPhone? I'm hoping for a class that does this. I even considered adding pngcrush's source code to my app, which seems drastic. I haven't been able to find an decent answer myself. I'd be very grateful for any help.

Thanks!

Update:
Some folks have suggested that it may be due to the file's size, but it isn't. During my tests, I added a toggle button to switch between using the embedded version and the downloaded version of exactly the same PNG. The only difference is that the embedded one was optimized by 'pngcrush' during compilation. This does some byte-swapping (from RGBA to BRGA) and pre-multiplication of alpha. (http://iphonedevelopment.blogspot.com/2008/10/iphone-optimized-pngs.html)

Also, the performance I'm referring to isn't the downloading, but the rendering. I superimpose custom painting on top of the image (overriding the drawRect method of the UIView), and it's very choppy when the background is the downloaded version, and very smooth when it's the embedded (and therefore optimized) version. Again, it's exactly the same file. The only difference is the optimization, which I'm hoping I can perform on the image at runtime, on the device, after downloading it.

Thanks again for everyone's help!

Best Answer

That link you posted pretty much answers your question.

During the build process XCode pre-processes your png so it's in a format that's more friendly to the graphics chip in the iPhone.

Png's that have not been processed like this will likely use a slower rendering path, one that deals with the non-native format and the fact that the alpha must be computed separately for each color.

So you have two options;

Perform the same work that pngcrush does and swap ordering/pre-multiply alpha. The speed up may be due to one or both of these.
After you have loaded your image, you can "create" a new image from it. This new image should be in the iPhone's native format and so should perform faster. The downside is it could potentially take up a bit more memory.

E.g.

CGRect area = CGRectMake(0, 0, width, height);
CGSize size = area.size;
UIGraphicsBeginImageContext(size);

[oldImage drawInRect:area];

UIImage *newImage = UIGraphicsGetImageFromCurrentImageContext();
UIGraphicsEndImageContext();

Related Solutions

Ios – How to develop for iPhone using a Windows development machine

It's certainly possible to develop on a Windows machine, in fact, my first application was exclusively developed on the old Dell Precision I had at the time :)

There are three routes;

Install OSx86 (aka iATKOS / Kalyway) on a second partition/disk and dual boot.
Run Mac OS X Server under VMWare (Mac OS X 10.7 (Lion) onwards, read the update below).
Use Delphi XE4 and the macincloud service. This is a commercial toolset, but the component and lib support is growing.

The first route requires modifying (or using a pre-modified) image of Leopard that can be installed on a regular PC. This is not as hard as you would think, although your success/effort ratio will depend upon how closely the hardware in your PC matches that in Mac hardware - e.g. if you're running a Core 2 Duo on an Intel Motherboard, with an NVidia graphics card you are laughing. If you're running an AMD machine or something without SSE3 it gets a little more involved.

If you purchase (or already own) a version of Leopard then this is a gray area since the Leopard EULA states you may only run it on an "Apple Labeled" machine. As many point out if you stick an Apple sticker on your PC you're probably covered.

The second option is more costly. The EULA for the workstation version of Leopard prevents it from being run under emulation and as a result, there's no support in VMWare for this. Leopard server, however, CAN be run under emulation and can be used for desktop purposes. Leopard server and VMWare are expensive, however.

If you're interested in option 1) I would suggest starting at Insanelymac and reading the OSx86 sections.

I do think you should consider whether the time you will invest is going to be worth the money you will save though. It was for me because I enjoy tinkering with this type of stuff and I started during the early iPhone betas, months before their App Store became available.

Alternatively, you could pick up a low-spec Mac Mini from eBay. You don't need much horsepower to run the SDK and you can always sell it on later if you decide to stop development or buy a better Mac.

Update: You cannot create a Mac OS X Client virtual machine for OS X 10.6 and earlier. Apple does not allow these Client OSes to be virtualized. With Mac OS X 10.7 (Lion) onwards, Apple has changed its licensing agreement in regards to virtualization. Source: VMWare KnowledgeBase

C++ – while (1) Vs. for (;;) Is there a speed difference

In perl, they result in the same opcodes:

$ perl -MO=Concise -e 'for(;;) { print "foo\n" }'
a  <@> leave[1 ref] vKP/REFC ->(end)
1     <0> enter ->2
2     <;> nextstate(main 2 -e:1) v ->3
9     <2> leaveloop vK/2 ->a
3        <{> enterloop(next->8 last->9 redo->4) v ->4
-        <@> lineseq vK ->9
4           <;> nextstate(main 1 -e:1) v ->5
7           <@> print vK ->8
5              <0> pushmark s ->6
6              <$> const[PV "foo\n"] s ->7
8           <0> unstack v ->4
-e syntax OK

$ perl -MO=Concise -e 'while(1) { print "foo\n" }'
a  <@> leave[1 ref] vKP/REFC ->(end)
1     <0> enter ->2
2     <;> nextstate(main 2 -e:1) v ->3
9     <2> leaveloop vK/2 ->a
3        <{> enterloop(next->8 last->9 redo->4) v ->4
-        <@> lineseq vK ->9
4           <;> nextstate(main 1 -e:1) v ->5
7           <@> print vK ->8
5              <0> pushmark s ->6
6              <$> const[PV "foo\n"] s ->7
8           <0> unstack v ->4
-e syntax OK

Likewise in GCC:

#include <stdio.h>

void t_while() {
    while(1)
        printf("foo\n");
}

void t_for() {
    for(;;)
        printf("foo\n");
}

    .file   "test.c"
    .section    .rodata
.LC0:
    .string "foo"
    .text
.globl t_while
    .type   t_while, @function
t_while:
.LFB2:
    pushq   %rbp
.LCFI0:
    movq    %rsp, %rbp
.LCFI1:
.L2:
    movl    $.LC0, %edi
    call    puts
    jmp .L2
.LFE2:
    .size   t_while, .-t_while
.globl t_for
    .type   t_for, @function
t_for:
.LFB3:
    pushq   %rbp
.LCFI2:
    movq    %rsp, %rbp
.LCFI3:
.L5:
    movl    $.LC0, %edi
    call    puts
    jmp .L5
.LFE3:
    .size   t_for, .-t_for
    .section    .eh_frame,"a",@progbits
.Lframe1:
    .long   .LECIE1-.LSCIE1
.LSCIE1:
    .long   0x0
    .byte   0x1
    .string "zR"
    .uleb128 0x1
    .sleb128 -8
    .byte   0x10
    .uleb128 0x1
    .byte   0x3
    .byte   0xc
    .uleb128 0x7
    .uleb128 0x8
    .byte   0x90
    .uleb128 0x1
    .align 8
.LECIE1:
.LSFDE1:
    .long   .LEFDE1-.LASFDE1
.LASFDE1:
    .long   .LASFDE1-.Lframe1
    .long   .LFB2
    .long   .LFE2-.LFB2
    .uleb128 0x0
    .byte   0x4
    .long   .LCFI0-.LFB2
    .byte   0xe
    .uleb128 0x10
    .byte   0x86
    .uleb128 0x2
    .byte   0x4
    .long   .LCFI1-.LCFI0
    .byte   0xd
    .uleb128 0x6
    .align 8
.LEFDE1:
.LSFDE3:
    .long   .LEFDE3-.LASFDE3
.LASFDE3:
    .long   .LASFDE3-.Lframe1
    .long   .LFB3
    .long   .LFE3-.LFB3
    .uleb128 0x0
    .byte   0x4
    .long   .LCFI2-.LFB3
    .byte   0xe
    .uleb128 0x10
    .byte   0x86
    .uleb128 0x2
    .byte   0x4
    .long   .LCFI3-.LCFI2
    .byte   0xd
    .uleb128 0x6
    .align 8
.LEFDE3:
    .ident  "GCC: (Ubuntu 4.3.3-5ubuntu4) 4.3.3"
    .section    .note.GNU-stack,"",@progbits

So I guess the answer is, they're the same in many compilers. Of course, for some other compilers this may not necessarily be the case, but chances are the code inside of the loop is going to be a few thousand times more expensive than the loop itself anyway, so who cares?

Best Answer

Related Solutions

Ios – How to develop for iPhone using a Windows development machine

C++ – while (1) Vs. for (;;) Is there a speed difference

Related Topic