Can anybody explain below statment ?

Qt embedded developer

uint8_t Can1MessageRAM[712U] __attribute__((aligned(32)));

i want to understand what is purpose of attribute ?

if possible anybody can explain above sentence ?

Chris Kawa

All symbols in C++ that start with __ are implementation specified. In other words they are not portable. __attribute__((ATTRIBUTE)) is a GNU extension of C++ that lets you put non-standard attributes on types. One of those attributes used here is alignment. Other vendors have other means of doing that. For example MSVC has #pragma pack(32) for the same thing. Starting from C++11 we also have a standard way of specifying alignment with alignas(32), so these vendor specific extensions are no longer needed, but a lot of older code still uses them.

As for alignment itself that just means that the variable will be placed at an address aligned to the value specified. In other words if you print the address of the variable Can1MessageRAM it's going to be a multiple of 32. If you place two class members with alignment attribute the compiler might pad the space between them so that their address is the multiple of specified value.

For example if you had a struct like this:

struct A
{
    uint8_t Can1MessageRAM1[712U];
    uint8_t Can1MessageRAM2[712U];
};

and printed the offsets of its fields:

qDebug() << offsetof(A, Can1MessageRAM1) << offsetof(A, Can1MessageRAM2);

a possible output would be 0 712 i.e. the second member starts right after the first. If you align them like this:

struct A
{
    alignas(32) uint8_t Can1MessageRAM1[712U];
    alignas(32) uint8_t Can1MessageRAM2[712U];
};

the output may change to be like this: 0 736, so compiler inserts some space between the two members, so that the address of the second is aligned to 32.

Alignment has many uses. Some techniques, like using SSE requires aligned variables to work. On some hardware unaligned access is less performant because there's no dedicated silicon in the CPU, so it has to compute the unaligned address or split operation to aligned parts. Another usage is when you have a platform with fixed memory addresses mapping to specific hardware functions. Yet another use of aligned data is when you iterate over large dataset. Aligned elements might fit better into a cache line improving access performance. Many other uses exist.

DerReisende

@Qt-embedded-developer See e.g. here.

Chris Kawa

All symbols in C++ that start with __ are implementation specified. In other words they are not portable. __attribute__((ATTRIBUTE)) is a GNU extension of C++ that lets you put non-standard attributes on types. One of those attributes used here is alignment. Other vendors have other means of doing that. For example MSVC has #pragma pack(32) for the same thing. Starting from C++11 we also have a standard way of specifying alignment with alignas(32), so these vendor specific extensions are no longer needed, but a lot of older code still uses them.

As for alignment itself that just means that the variable will be placed at an address aligned to the value specified. In other words if you print the address of the variable Can1MessageRAM it's going to be a multiple of 32. If you place two class members with alignment attribute the compiler might pad the space between them so that their address is the multiple of specified value.

For example if you had a struct like this:

struct A
{
    uint8_t Can1MessageRAM1[712U];
    uint8_t Can1MessageRAM2[712U];
};

and printed the offsets of its fields:

qDebug() << offsetof(A, Can1MessageRAM1) << offsetof(A, Can1MessageRAM2);

a possible output would be 0 712 i.e. the second member starts right after the first. If you align them like this:

struct A
{
    alignas(32) uint8_t Can1MessageRAM1[712U];
    alignas(32) uint8_t Can1MessageRAM2[712U];
};

the output may change to be like this: 0 736, so compiler inserts some space between the two members, so that the address of the second is aligned to 32.

Alignment has many uses. Some techniques, like using SSE requires aligned variables to work. On some hardware unaligned access is less performant because there's no dedicated silicon in the CPU, so it has to compute the unaligned address or split operation to aligned parts. Another usage is when you have a platform with fixed memory addresses mapping to specific hardware functions. Yet another use of aligned data is when you iterate over large dataset. Aligned elements might fit better into a cache line improving access performance. Many other uses exist.