Native ARM Co Pro: OSBYTE SWI returns unexpected value in R1

[hoglet67]

Result from PiTubeDirect is different to the ARM7TDMI Co Pro, likely due to this:

    reg[1] = x | (y << 8);      // JGH
    reg[2] = y;

[BigEd]

this change:
3abda3a
"JGH: tube-swi: Corrected R1 return from OSBYTE &80+"

[dudleysoft]

This seems to be at odds with the behaviour of the ARM7TDMI co-processor's OS_Byte return values, which simply returns the values of X and Y in R1 and R2.

This tripped me up when I tried to implement reading the mouse values in BQUAKE using OS_Byte 128 to read the ADVAL values from the host.

Reading the values in BeebEm in the ARM7TDMI co-processor was returning a different result to the PiNative co-processor.

[hoglet67]

I would suggest just anding the result with &FF in each case., i.e. treat the contents of bits 8-31 as undefined.

@jgharston would you care to explain what's going on here?

[dudleysoft]

I would suggest just anding the result with &FF in each case., i.e. treat the contents of bits 8-31 as undefined.

@jgharston would you care to explain what's going on here?

This is what I did in my code, I'd tested under BeebEm with the ARM7TDMI and was wondering why my mouse coordinates seemed wrong, which led me to looking at the code and spotting it was doing this. The funny thing is I was already doing this in my keyboard scanning code so I must have noticed it before, presumably at that point I didn't realise BeebEm was behaving differently.

I'm not sure which way is the correct way, if RISC OS does it this way then perhaps that's why the change was made.

[hoglet67]

Just to be clear, if you do AND with &FF, does the same code work on both platforms?

[jgharston]

That's odd, I was sure RISC OS returned R1=x OR (y<<8) so that R1 returned the full value so the caller didn't need to do R1 OR (R2<<8), and I have a memory rattling around about "ADVAL returning full 32 bits" from somewhere.
But a quick test:

PRINT ~ADVAL7
3B6
SYS "OS_Byte",128,7,0 TO r0,r1,r2
PRINT ~r0,r1,r2
80 B6 3
*FX0
RISC OS 3.11

So, change it to reg[1] = x;
(I'll check some other ARM systems later)

"This tripped me up when I tried to implement reading the mouse values in BQUAKE using OS_Byte 128 to read the ADVAL values from the host."

How did it trip up? ORR R1,R1,R2,LSL#8 gives the same result regardless of whether R1 already has R2 in bit 8 upwards or not.

[dudleysoft]

Just to be clear, if you do AND with &FF, does the same code work on both platforms?

Yes it did.

"This tripped me up when I tried to implement reading the mouse values in BQUAKE using OS_Byte 128 to read the ADVAL values from the host."

I was writing in C, so I used x + (y<<8), because I tend to think in terms of regular maths rather than bitwise maths.

[jgharston]

Yep, ARM CoPro, 1984ish:
BL L00000D24 ; Wait for returned byte
STMDB R13!,{R0} ; Save returned Carry
BL &03000D24 ; Wait for returned byte
MOV R2,R0 ; Return OSBYTE Y value
BL &03000D24 ; Wait for returned byte
MOV R1,R0 ; Return OSBYTE X value
LDMIA R13!,{R11} ; Restore saved Carry value to R11
LDMIA R13!,{R0,R14} ; Restore saved link,R0
BIC R14,R14,#&20000000 ; Clear link's carry flag
TST R11,#&80 ; Check carry value in R11
ORRNE R14,R14,#&20000000 ; Set link's carry flag if set
B L000000F8 ; Jump to exit

RISC OS Kernel (end of BUFFV called by OSBYTE &80), 1997ish:
TEQ R1, #0 ; don't stamp on carry
ADDMI R1, R1, R5 ; wrap number of chars if negative
SUBCS R1, R5, R1 ; C=1 => convert to spaces
SUBCS R1, R1, #1 ; one fewer spaces than BuffSizes
MOV R2, R1, LSR #8 ; make R2 = hi-byte
AND R1, R1, #&FF ; and R1 = lo-byte
Pull "R3-R5,PC"

...but I'm sure I've seen somewhere some OSBYTE calls returning larger values in R1. But BASIC does ORR dest,R1,R2,LSL #8 when reading a large OSBYTE (eg ADVAL and USR&FFF4) and AND dest,R1,#255 when reading a byte value (eg VPOS) anyway.

[hoglet67]

Fixed in hognose-dev.