So is a 24 bit color model (for use in a MemoryImageSource) slower than a 32 bit color model?
I’ve read around and people say that 32 bit is faster because the processors are able to handle 32 bit numbers faster than 24 bit ones, and also I read that not using the 8 bits at the end of the 24 bits (because modern registers are 32 bit) is a waste of memory.
If anything I’d venture they’re both as fast.
More importantly though the situations where the minimal speed difference between the two, if any, makes an impact is probably around 40% of the time always exactly 0%.
That’s a bit of a misleading way to look at it, though. It’s not 24-bit vs 32-bit numbers you should be thinking about, but data types. And unless there is some darkened corner of Java I’m oblivious to, there’s no 24-bit data types to worry about. If the color values weren’t stored in 32-bit integers, then they’d have to be separated into 3 8-bit values or 1 16-bit and 1 8-bit (which would just be weird). A 24-bit RGB in a 32-bit data type (int) is just an RGBA value whos alpha component is ignored.
The argument holds true to some extent for lesser bit depths, though. A 16-bit color value would be more efficiently manipulated in the native word size (32-bit or 64-bit) than as a short.
[quote]and also I read that not using the 8 bits at the end of the 24 bits (because modern registers are 32 bit) is a waste of memory.
[/quote]
Which in practice is really no concern. It’s not like you have much of a choice for 24-bit color values, though. Either you use a tightly-packed array of bytes where each byte represents one component of an RGB triplet, or you use an array of ints where each int represents a full 24-bit color value with 8-bits unused. There are performance vs. memory implications for both approaches, but I don’t see that either would be statistically relevant in any but the most extreme circumstances (such as a heavily resource constrained system – in which case, what the hell are you running a JVM for in the first place?). If I were implementing a graphics package myself, I’d would certainly use int to represent 24-bit color values unless I knew the more common use case would be in implementing individual color components rather than full color values.