RE: Compiler switches on RISC byte alignment

["Fairley, Diane" <dfairley@slac.stanford.edu>]

Thank you all for your insight.  We are indeed receiving network message packets of packed data.  Since this code must be portable to several processors, I like the idea of avoiding compiler directives by using byte pointers in unpack routines...and include handling byte-swap issues as well...
Diane

-----Original Message-----
From: Kay-Uwe Kasemir [mailto:kasemirk@ornl.gov] 
Sent: Thursday, September 02, 2004 12:28 PM
To: Lawrence T. Hoff
Cc: Fairley, Diane; EPICS Tech Talk
Subject: Re: Compiler switches on RISC byte alignment

Hi:

I'm surprised Jeff Hill hasn't jumped onto this thread.
He and I had talked about similar issues before.
Maybe he's getting some work done and ignoring tech-talk.

Anyway, the thing about using packed structures in order to access memory mapped I/O or raw network packages is that you often run into compiler-specific "pack" instructions.
Larry's copy works around the alignment, next you'd run into byte order issues when you change the CPU type.

The most portable way of coding might be to always use byte pointers and assemble/disassemble the data yourself. So instead of ...
typedef struct __attribute__ ((packed)){
      unsigned char  uCh;
      short          sInt;
....
you use
....
   volatile char *buf = <some address>;
   unsigned char uCh = *(buf++);
   unsigned short low = *(buf++);
   unsigned short high = *(buf++);
   short sInt = (high << 8) | low;
....

You can somewhat beautify this by defining routines unpack_short(), ...

-Kay

On Sep 2, 2004, at 14:56, Lawrence T. Hoff wrote:
>    As far as I know, that is a characteristic of the SPARC processor, 
> and cannot be influenced by compiler directives. You *can* use 
> byte-oriented instructions on data which is not on "natural" 
> boundaries. I.e. you could access "lInt" as follows (from a packed 
> array):
>
> unsigned long int temp;
> memcpy(&temp, &testdata.lInt, sizeof(temp));
>
> printf("testdata.lInt = %lu", temp)
>
> HTH -- Larry
>
>>
>> does not print the right value.  A pointer to an unaligned structure 
>> member results in a segment violation.
>>
>
>