BTST.B with memory

[paraj]

Just a quick sanity check. I'm disassembling parts of RSI MegaDemo and it contains this small function, that the author clearly intended to wait for the blitter to be finished (it's called before blitter registers are updated):

00000a6e 0839 000e 00df f002      btst.b #$000e,$00dff002
00000a76 66f6                     bne.b #$f6 == $00000a6e
00000a78 4e75                     rts

It actually checks BLTEN though, and things just happen to work regardless, right?


(The copper isn't touching any blitter registers/dmacon, so it doesn't look like a "clever" way to synchronize with that).

[a/b]

Only 3 bottom bits are used if <ea> is memory (so bit 6 is tested), so it checks bit 6 of $002.b, which is bit 14 of $002.w = bltbusy.

[Photon]

The bit number is encoded correctly by the Assembler. BTST memory accesses are 8-bit (.b). When the CPU encounters the $000e bit number, it ANDs it with 7.

Because of this, bit 6 is tested whether or not the bit number in the instruction is $e or $6.

[paraj]

Doh, thanks guys, too much work on little endian systems has corrupted me it seems

[Rock'n Roll]

not btst.b, but bset.b with memory (test a bit and set)

I wondered how bset.b d2,(a0,d0.w) works.
where a0 is the bitplane, d0 is the offset byte to a
byte in the bitplane and d2 is a bit in that byte. (plot a point routine)

I found:
If a memory location is the destination, a byte is
read from that location, the bit operation performed using bit
number modulo 8, and the byte written back to the location.

e.g. d2.b=$fd

>?$fd
0x000000FD = %00000000000000000000000011111101 = 253 = 253
>?253/8
0x0000001F = %00000000000000000000000000011111 = 31 = 31
>?8*31
0x000000F8 = %00000000000000000000000011111000 = 248 = 248
>?253-248
0x00000005 = %00000000000000000000000000000101 = 5 = 5
>m ra0 1 ; where bit 5 is set	
0006EF80 2000 0000 0000 0000 0000 0000 0000 0000   ...............	
>? $20	 ; Bit 5	
0x00000020 = %00000000000000000000000000100000 = 32 = 32

very special...
I asked me, are there other interesting uses for this kind of bset with memory?

[Photon]

All four bit operations to memory are byte-size.

bclr, bchg, and bset have a built-in btst, and can therefore be used for any kind of polling.

[redblade]

@paraj: This also puzzled me as I saw different calls on different example sources. I managed to find a post on amiga.org by Piru which explained it and then it made sense especially because btst.b #14,$dff002 used to bring up error on Devpac 3. I think it was accepted in Devpac 2.

[paraj]

Originally Posted by redblade

@paraj: This also puzzled me as I saw different calls on different example sources. I managed to find a post on amiga.org by Piru which explained it and then it made sense especially because btst.b #14,$dff002 used to bring up error on Devpac 3. I think it was accepted in Devpac 2.

Ah, makes sense. I guess it's technically "fine" and does what the author intended here. It just set off my code smell detector due to the "unconventional" encoding, and because I've been bitten by doing e.g.

BTST #INTB_VERTB, intreqr(a6)

.

[Antiriad_UK]

All my old amiga code had btst #14 for blitwait. I got it from the Amiga System Programmer guide examples as I suspect many of us did - all the "BBUSY Test" commented lines were btst #14. Didn't realise it was wrong for a long time.