RE: Unaligned access

[johill@lanl.gov (Jeff Hill)]

Geoff,

EPICS has been tested on the alpha before, but not on a true 64 bit OS.
How does it feel to have 64 bit longs that the rest of us are lacking?

During revisions for EPICS R3.14 the routine issue_ca_array_put () has
been eliminated and there is now a more efficient replacement. Are you 
interested in beta testing R3.14?

I also back patched the EPICS R3.13 CVS branch so that this will be fixed 
if there is another R3.13 release.

Thanks for bringing this to my attention.

Jeff

> -----Original Message-----
> From: savage@fnal.gov [mailto:savage@fnal.gov]
> Sent: Tuesday, August 29, 2000 11:56 AM
> To: tech-talk
> Subject: Unaligned access
> 
> 
> Hi,
> 
> When using  CA puts on our 64-bit alpha machines running Tru64 Unix I am
> getting unaligned access faults.  Here's the reason.
> 
> In src/ca/access.c issue_ca_array_put are the lines
>     case    DBR_LONG:
>         *(long *)pdest = htonl (*(dbr_long_t *)pvalue);
>         break;
> 
> in src/db_db_access.h
> 
> typedef epicsInt32      dbr_long_t;
> 
> in src/include/epicsTypes.h
> 
> typedef int             epicsInt32;
> 
> The code is attempting to equate a long with an int.  This works on a 32
> bit machine but not a 64 bit machine.
> 
> All the native C data types (long, char, ...) in the issue_ca_array_put
> routine should be changed to be the epics dbr_xxxx data types for
> consistency.
> 
> Would someone please correct this in the next release so that I don't
> have to keep patching our code.  Are we the only ones using 64-bit
> computers?  Has anyone else encoutered this problem?
> 
> Thanks
> 
> Geoff
> 
>