PHP or SHX?

[Zach]

I’ve been working on a kernel where I want to clear bit 2 of VBLANK, but A, X, and Y are all being used for other purposes. I figured out a couple ways to accomplish this, use PHP and make sure the last value loaded was non-zero, or use SHX VBLANK - #OFFSET, Y.

 

Both approaches have pros and cons. What I’m wondering is whether it is a good idea to use SHX? Searching the forums for the SHX, I found that supercat had some ideas for using it (actually similar to the use I am describing). On the other hand, Thomas mentioned that $9E is unstable, and sure enough, that opcode is SHX.

 

Any thoughts?

[+batari]

I’ve been working on a kernel where I want to clear bit 2 of VBLANK, but A, X, and Y are all being used for other purposes. I figured out a couple ways to accomplish this, use PHP and make sure the last value loaded was non-zero, or use SHX VBLANK - #OFFSET, Y.

 

Both approaches have pros and cons. What I’m wondering is whether it is a good idea to use SHX? Searching the forums for the SHX, I found that supercat had some ideas for using it (actually similar to the use I am describing). On the other hand, Thomas mentioned that $9E is unstable, and sure enough, that opcode is SHX.

 

Any thoughts?

I thought the only "unstable" opcodes were those that can jam the CPU, (all $x2 except $A2) and ANE and LXA, as they involve an AND with a constant of unknown value.

 

My impression of opcodes like $9E is that they are stable, but unusual. Other unusual instructions are SHS, SHA, SHY, LAS, ANC, ARR, ASR, ANE, LXA, and SBX.

 

Personally, the problem I see with SHX is that it's anded with PCH+1 and indexed with Y, so it seems quite troublesome to use it.

[+batari]

I’ve been working on a kernel where I want to clear bit 2 of VBLANK, but A, X, and Y are all being used for other purposes. I figured out a couple ways to accomplish this, use PHP and make sure the last value loaded was non-zero, or use SHX VBLANK - #OFFSET, Y.

 

Both approaches have pros and cons. What I’m wondering is whether it is a good idea to use SHX? Searching the forums for the SHX, I found that supercat had some ideas for using it (actually similar to the use I am describing). On the other hand, Thomas mentioned that $9E is unstable, and sure enough, that opcode is SHX.

 

Any thoughts?

I thought the only "unstable" opcodes were those that can jam the CPU, (all $x2 except $A2) and ANE and LXA, as they involve an AND with a constant of unknown value.

 

My impression of opcodes like $9E is that they are stable, but unusual. Other unusual instructions are SHS, SHA, SHY, LAS, ANC, ARR, ASR, ANE, LXA, and SBX.

 

Personally, the problem I see with SHX is that it's anded with PCH+1 and indexed with Y, so it seems quite troublesome to use it.

More research indicates that apparently SHA, SHS, SHX and SHY sometimes don't get anded with PCH+1, and if the index crosses a page boundary, it isn't handled properly. Those problems aren't necessarily fatal.

[Zach]

More research indicates that apparently SHA, SHS, SHX and SHY sometimes don't get anded with PCH+1, and if the index crosses a page boundary, it isn't handled properly. Those problems aren't necessarily fatal.

Thanks Fred, unless Y is 0 or 1, I would have to cross a page boundary to reach VBLANK. So it seems that I should not use SHX for this purpose. No problem, I'll just use PHP.

[+batari]

More research indicates that apparently SHA, SHS, SHX and SHY sometimes don't get anded with PCH+1, and if the index crosses a page boundary, it isn't handled properly. Those problems aren't necessarily fatal.

Thanks Fred, unless Y is 0 or 1, I would have to cross a page boundary to reach VBLANK. So it seems that I should not use SHX for this purpose. No problem, I'll just use PHP.

You could use the TIA mirror of VBLANK at $41, which would give you more usable Y values. I'm not sure if that's enough, though.

Edited November 28, 2006 by batari

[Zach]

You could use the TIA mirror of VBLANK at $41, which would give you more usable Y values. I'm not sure if that's enough, though.

That's a good point, but it still limits the possible values of Y.

 

FWIW, I did test the SHX version of my kernel on a 4 switch system, and it looks good so far. I used the instruction SHX $FFC7,Y when Y=3A and X=06, and it did indeed clear VBLANK. Since bit 1 of X is 1, the bitwise AND with (FF+1) must have worked, even though the page boundary was crossed.

 

Maybe we would get different results on different systems. I guess SHX is not thoroughly understood, and it'll be safer to use PHP. Besides I haven't found anything useful to do with the 2 cycles I save in my kernel. (SHX takes 5 cycles, PHP takes 7 when you account for resetting the SP.)

Edited December 3, 2006 by Zach

[+batari]

You could use the TIA mirror of VBLANK at $41, which would give you more usable Y values. I'm not sure if that's enough, though.

That's a good point, but it still limits the possible values of Y.

 

FWIW, I did test the SHX version of my kernel on a 4 switch system, and it looks good so far. I used the instruction SHX $FFC7,Y when Y=3A and X=06, and it did indeed clear VBLANK. Since bit 2 of X is 1, the bitwise AND with (FF+1) must have worked, even though the page boundary was crossed.

 

Maybe we would get different results on different systems. I guess SHX is not thoroughly understood, and it'll be safer to use PHP. Besides I haven't found anything useful to do with the 2 cycles I save in my kernel. (SHX takes 5 cycles, PHP takes 7 when you account for resetting the SP.)

The only documentation I've seen about SHX says that, for instance, if X=$06 and you do SHX $FFC7,Y the actual bank where the value gets stored in a page-wrapping condition will be either $00 or $06, but since $06 happens to contain a TIA mirror, it should work just the same.