egcs-1.0.1 ColdFire (m5200) bug?

[Parag Patel]

Hello.  I'm not subscribed to this list (but am signed up for the 
egcs-announce list) but thought I should report this problem.

I was trying out the latest egcs-1.0.1 release with the uC/OS real-time 
OS for a Motorola Coldfire (68k-lite) SBC5206 evaluation board.

It seems to generate the right stuff most of the time, but this code 
snippet selects a 68k move.b instruction combo that is illegal on the 
Coldfire and rejected by gas.  Naturally, the compile fails.  The 
instruction appears to be "move.b %d?,%a?" or some combination thereof 
involving a move.b and a register for a local stack variable.

I can remove most any line from the function or from the struct and the 
compile will succeed.  Leaving the code as it is below, and the compile 
fails.  Changing the local var "i" below from a UBYTE to a UWORD or an 
int makes it work, so there is a workaround.


     -- Parag Patel <parag@cgt.com>



typedef unsigned char  BOOLEAN;
typedef unsigned char  UBYTE;
typedef unsigned int   UWORD;


extern UBYTE    OSLockNesting;
extern BOOLEAN  OSRunning;

typedef struct os_tcb {
    void          *OSTCBStkPtr;
    UBYTE          OSTCBStat;
    UBYTE          OSTCBPrio;
    UWORD          OSTCBDly;
    struct os_tcb *OSTCBNext;
    struct os_tcb *OSTCBPrev;
} OS_TCB;

extern OS_TCB   OSTCBTbl[];
extern OS_TCB  *OSTCBPrioTbl[];

void OSInit(void *idle_task_stk, UBYTE maxtasks)
{
    UBYTE i;

    OSLockNesting = 0;
    OSRunning     = 0;
    for (i = 0; i < 64; i++) {
        OSTCBPrioTbl[i] = (OS_TCB *)0;
    }
    OSTCBTbl[maxtasks-1].OSTCBNext = (OS_TCB *)0;
}