memory alignment of a struct

Hello everyone,


I heard when define a struct, it is important to make the size of the struct memory aligned.

For example,

Code:

---

struct foo
{
    char p;
    char[3] align;
};

---

Code:

---

struct goo
{
    char p;
};

---

struct foo is better than struct goo, even if we only use member p.

Allocate 3 bytes to align with memory. But I do not know why make memory aligned is better. Any reasons? (for example, prevent memory from running out?)


thanks in advance,
George

Depends what you're doing with the struct I guess. For your goo example, a compiler is likely to allocate 4 or 8 bytes for this structure even though it technically only needs 1 byte (ignoring vtables etc). You can alter the alignment using pragmas if you need to do something like overlaying the structure on an area of memory (or raw bytes) or to address a hardware device.

For most applications though I personally don't worry too much about the alignment.

Regards
Alan

Hi,

Its easy and fast for accessing/fetching memory which is aligned to system words/bus size from hardware perspective i.e CPU

thats why memory alignment is used !

for unaligned memory CPU needs extra cycles for accessing/fetching.

-Anant

Hi,

here is a article...

http://www.devx.com/tips/Tip/13265

http://en.wikipedia.org/wiki/Data_structure_alignment


-Anant

Hi Anant,

Quote:

---

Originally Posted by anantwakode

Hi,

Its easy and fast for accessing/fetching memory which is aligned to system words/bus size from hardware perspective i.e CPU

thats why memory alignment is used !

for unaligned memory CPU needs extra cycles for accessing/fetching.

-Anant

---

I do not know why alignment will make memory reading operation faster. In my case, foo needs to fetch 4 bytes and goo needs to fetch 1 byte. Even if each time, CPU fetch 4 bytes, I think they should be the same performance, right?

Why do you think memory alignment will improve performance? An example?


regards,
George

Thanks Alan,

Quote:

---

Originally Posted by AlanGRutter

Depends what you're doing with the struct I guess. For your goo example, a compiler is likely to allocate 4 or 8 bytes for this structure even though it technically only needs 1 byte (ignoring vtables etc). You can alter the alignment using pragmas if you need to do something like overlaying the structure on an area of memory (or raw bytes) or to address a hardware device.

For most applications though I personally don't worry too much about the alignment.

Regards
Alan

---

I am interested in two points,

1. How to do memory overlapping?

2. How to address a H/W device?

You mean both of them could be implemented by #pragma? Could you show an example please?


regards,
George

Thanks Anant. It is a good article. But I do not understand why alignment will improve performance. Could you show an example please?

Quote:

---

Originally Posted by anantwakode

Hi,

here is a article...

http://www.devx.com/tips/Tip/13265

http://en.wikipedia.org/wiki/Data_structure_alignment


-Anant

---

regards,
George

Hi,

e.g

struct foo
{
char ch1;
};

struct soo
{
char ch2;
};

see if I create objects
foo f;
soo s;

memory structure...may be like this

| f.ch1,s.ch2 | | |

may some cpu instructions have to filterout/Ignore (will need extra work to do to leave s.ch2 untouched ) s.ch2 while processing f.ch1 as the most of the instructions are developed to read/work on words at a time


if these are aligned

|f.ch1 |s.ch2 | |

can be fetched and processed seprately .................



-Anant

Quote:

---

Originally Posted by anantwakode

Hi,

e.g

struct foo
{
char ch1;
};

struct soo
{
char ch2;
};

see if I create objects
foo f;
soo s;

memory structure...may be like this

| f.ch1,s.ch2 | | |

may some cpu instructions have to filterout/Ignore (will need extra work to do to leave s.ch2 untouched ) s.ch2 while processing f.ch1 as the most of the instructions are developed to read/work on words at a time


if these are aligned

|f.ch1 |s.ch2 | |

can be fetched and processed seprately .................



-Anant

---

What is your format mean? Like sign '|' ??

| f.ch1,s.ch2 | | |

|f.ch1 |s.ch2 | |


regards,
George

hi,

| means Word Boundry, i.e memory space

|1000|1001|1002|1003|..................
|Word|Word|Word|Word|Word|Word|Word|Word|Word|Word|


-Anant

What is being said about alignment is true however you generally do not need to specifically pad out your data structures so that they align on word boundaries as the compiler will do this for you.

Consider the following code

Code:

---

#include <iostream>
#include <string>

using namespace std;

struct goo
{
    char p;
    int i;
};

#pragma pack(1)

struct goo1
{
    char p;
    int i;
};

#pragma align()

int main()
{
    // character 'A' + Integer value 7979 (note byte ordering)
    char bytes[] = {0x41, 0x2b, 0x1f, 0x00, 0x00};

    cout << "Sizeof char: " << sizeof(char) << endl;
    cout << "Sizeof int: " << sizeof(int) << endl;

    cout << "Sizeof unpacked structure goo: " << sizeof(goo) << endl;
    cout << "Sizeof packed structure goo1: " << sizeof(goo1) << endl;

    //Overlay goo1 onto byte array
    goo1 *pStructGoo1 = reinterpret_cast<goo1 *>(bytes);

    cout << "Char p: " << pStructGoo1->p << endl;
    cout << "Integer i: " << pStructGoo1->i << endl;

    return 0;
}

---

In the first instance (goo) the compiler has allocated 4 bytes for the char and 4 bytes for the int. If I use the pragma to pack the structure to single byte boundaries, the extra space is removed.

By packing to single byte boundaries you can overlay a structure onto a stream of bytes as demonstrated above.

Regards
Alan