What's the most efficient way to make bitwise operations in a C array

Question

I have a C array like:

char byte_array[10];

And another one that acts as a mask:

char byte_mask[10];

I would

Answer 1

\#define CHAR_ARRAY_SIZE    (10)
\#define INT_ARRAY_SIZE     ((CHAR_ARRAY_SIZE/ (sizeof (unsigned int)) + 1)

typedef union _arr_tag_ {

    char          byte_array [CHAR_ARRAY_SIZE];
    unsigned int  int_array [INT_ARRAY_SIZE]; 

} arr_tag;

Now int_array for masking. This might work for both 32bit and 64 bit processors.

arr_tag arr_src, arr_result, arr_mask;

for (int i = 0; i < INT_ARRAY_SIZE; i ++) {
    arr_result.int_array [i] = arr_src.int_array[i] & arr_mask.int_array [i];
}

Try this, code might also look clean.

Answer 2

for ( i = 10 ; i-- > 0 ; )
    result_array[i] = byte_array[i] & byte_mask[i];

• Going backwards pre-loads processor cache-lines.
• Including the decrement in the compare can save some instructions.

This will work for all arrays and processors. However, if you know your arrays are word-aligned, a faster method is to cast to a larger type and do the same calculation.

For example, let's say n=16 instead of n=10. Then this would be much faster:

uint32_t* input32 = (uint32_t*)byte_array;
uint32_t* mask32 = (uint32_t*)byte_mask;
uint32_t* result32 = (uint32_t*)result_array;
for ( i = 4 ; i-- > 0 ; )
    result32[i] = input32[i] & mask32[i];

(Of course you need a proper type for uint32_t, and if n is not a power of 2 you need to clean up the beginning and/or ending so that the 32-bit stuff is aligned.)

Variation: The question specifically calls for the results to be placed in a separate array, however it would almost certainly be faster to modify the input array in-place.

Answer 3

If you want to make it faster, make sure that byte_array has length that is multiple of 4 (8 on 64-bit machines), and then:

char byte_array[12];
char byte_mask[12];
/* Checks for proper alignment */
assert(((unsigned int)(void *)byte_array) & 3 == 0);
assert(((unsigned int)(void *)byte_mask) & 3 == 0);
for (i = 0; i < (10+3)/4; i++) {
  ((unsigned int *)(byte_array))[i] &= ((unsigned int *)(byte_mask))[i];
}

This is much faster than doing it byte per byte.

(Note that this is in-place mutation; if you want to keep the original byte_array also, then you obviously need to store the results in another array instead.)