A89: Re: Re: TIOS functions disobey register conventions!
[Prev][Next][Index][Thread]
A89: Re: Re: TIOS functions disobey register conventions!
Hi!
> | I am replying to my own answer...
>
> Which answer?
Oops. I should tell "to my own message...".
>
> | move.w performs SIGN EXTENDING into the high word!!! I checked this
> | on real Motorola!!! VTI works correctly!!! So, all TIOS routines
> | which uses movem.w for saving registers may cause trashing of high
> | word of these registers!!!
>
> Sorry about the lack of technical understanding, but do you have any
> idea about how many there are, and if this applies only to d0-d4?
1) I have not any idea. But searching for them is terrible. Simply
speaking, "problematic" routines starts and ends with movem.w.
2) Not only to d0-d4, but to ALL regs. Anyway, nobody use movem.w with
address regs, so the problem is with d0-d6.
> Maybe we should tell TI.
What we will get if we tell this to them (anyway, I believe that someone
from TI reads these messages too). They may correct this in say AMS 2.06,
but what with existing AMSs?
> I suppose they don't know; otherwise they wouldn't be using these
> instructions.
I think that they know this (I found just now that this behaviour is
documented in official Motorola manual), but that they made a "bug":
they forgot the exact interpretation.
> If I understand you correctly, only one bit is trashed, right?
No :(( The whole high word! More precise:
if low_word<0x7FFF then set high word to 0
else set high word to 0xFFFF
Zeljko