Double the memory address

[AestheticDebris]

The same trick works entirely with offsets though.

E.g. the two buffers are stored starting at $xxxx, let's assume they're 16 bytes long

Buffer1 = $xxxx + 0
Buffer2 = $xxxx + 16

Thus SUM = 16 (this will always be the length of the buffer)

First time, X= 0, i.e. first buffer

X = SUM - X = 16-0 = 16

Second time X = 16, i.e. using second buffer

X = SUM - X = 16-16 = 0

No doubling of anything. We don't even ever have to care what $xxxx actually is for the maths to work out.

[phx]

Originally Posted by ross

It's not a problem with LoadSeg or the Amiga Hunks, nothing in the relocation table prevents you from 'rearranging' the same longword in memory twice.

Correct. But even then it would rarely help you.

So I can also create one for you by 'replacing' the linker job, but I don't think that's the purpose

I doubt that.

What you would achieve here is that a section's load-address is added twice to the 32-bit addend stored at a specific section-offset. This can only work when

buffer

is at offset 0. As soon as you want to double a label from a different section offset it fails.

EDIT: Hmm. Thinking about it, it could indeed work, if you also double the section-offset part of the addend at that section location. I.e. for buffer+buffer+24, when buffer resides at section offset 100, if would become 100+100+24.

[Bruce Abbott]

Originally Posted by remz

1) Rotareneg: I personally hate self-modifying code,

Me too, on any platform (very bad when you need to put a routine into ROM).

Using self-modifying code would make it slightly faster, but you should always look at the big picture before going for such micro-optimizations. Chances are the buffer address will not be held in a register continuously but be stored in a memory location for later use, and then you have to consider the time taken to store and retrieve it. Of course you will also have more code that uses the buffer. The more code is involved the less important that micro-optimization becomes.

Generally it is better to concentrate on overall efficiency and not sweat about little stuff like this. I usually start by writing code that does what I want in the obvious and logical way, then optimize it later only if necessary (when other avenues have already been exhausted).

[ross]

Originally Posted by phx

EDIT: Hmm. Thinking about it, it could indeed work, if you also double the section-offset part of the addend at that section location. I.e. for buffer+buffer+24, when buffer resides at section offset 100, if would become 100+100+24.

eheh, I was about to upload the exe to clear your doubt, but you understood what I intended to do