pointer arithmetic for alignment on short or int boundry

pointer arithmetic for alignment on short or int boundry

[Andy Howell]

I need to calculate an address on a short or int boundry. For some 
reason this is not working:

char *a;

a = (a + 1) & ((char*) -2));

Gcc 3.3 on Solaris complains:

error: invalid operands of types `char*' and `
    char*' to binary `operator&'

Why shouldn't I be able to mask off the low bit?

Thanks,

	Andy

Re: pointer arithmetic for alignment on short or int boundry

[Burak Serdar]

Try

a = (char *)((unsigned)(a + 1) & (unsigned)((char*) -2))

You can only add or subtract pointers for pointer arithmetic.

Andy Howell wrote:
> I need to calculate an address on a short or int boundry. For some 
> reason this is not working:
> 
> char *a;
> 
> a = (a + 1) & ((char*) -2));
> 
> Gcc 3.3 on Solaris complains:
> 
> error: invalid operands of types `char*' and `
>    char*' to binary `operator&'
> 
> Why shouldn't I be able to mask off the low bit?
> 
> Thanks,
> 
>     Andy
> 
> 

Re: pointer arithmetic for alignment on short or int boundry

[Andy Howell]

Burak,

	Thanks. You saved me from having to find that "do-as-I-mean" compiler 
switch again.

Andy

> Try
> 
> a = (char *)((unsigned)(a + 1) & (unsigned)((char*) -2))
> 
> You can only add or subtract pointers for pointer arithmetic.
> 
> Andy Howell wrote:
> 
>> I need to calculate an address on a short or int boundry. For some 
>> reason this is not working:
>>
>> char *a;
>>
>> a = (a + 1) & ((char*) -2));
>>
>> Gcc 3.3 on Solaris complains:
>>
>> error: invalid operands of types `char*' and `
>>    char*' to binary `operator&'
>>
>> Why shouldn't I be able to mask off the low bit?
>>
>> Thanks,
>>
>>     Andy
>>
>>
> 
> 

Re: pointer arithmetic for alignment on short or int boundry

[Eljay Love-Jensen]

Hi Andy,

You need to do mathematical operations on non-pointer types.  Use <stdint.h> (part of C99) to assist.

char* alignToIntPtr(char* a) {
  const int lowBitsMask = sizeof(int) - 1;
  int lowBits = reinterpret_cast<intptr_t>(a) & lowBitsMask;

  // Rounding UP to next appropriate address, or stay same.
  int delta = lowBits ? (sizeof(int) - lowBits) : 0;
  a += delta;

  // ...OR...
  // Rounding DOWN to previous appropriate address, or stay same.
  a -= lowBits;

  return a;
}

I trust the optimizer to make the actual code nice and efficient.  However, if this is in a really tight loop (as indicated by the profiler), you may want to investigate some bit-twiddling algorithms to make it even more efficient.

HTH,
--Eljay