My suggestion would be to use a lookup table to speed up the processing.
Another thing to note is with the current definition of your matrix the maximum size will be 8x8 bits. This fits into a uint64_t so we can use this to our advantage especially when using a 64-bit platform.
I have worked out a simple example using a lookup table which you can find below and run using: http://www.tutorialspoint.com/compile_cpp11_online.php online compiler.
Example code
#include <iostream>
#include <bitset>
#include <stdint.h>
#include <assert.h>
using std::cout;
using std::endl;
using std::bitset;
/* Static lookup table */
static uint64_t lut[256];
/* Helper function to print array */
template<int N>
void print_arr(const uint8_t (&arr)[N]){
for(int i=0; i < N; ++i){
cout << bitset<8>(arr[i]) << endl;
}
}
/* Transpose function */
template<int N>
void transpose_bitmatrix(const uint8_t (&matrix)[N], uint8_t (&transposed)[8]){
assert(N <= 8);
uint64_t value = 0;
for(int i=0; i < N; ++i){
value = (value << 1) + lut[matrix[i]];
}
/* Ensure safe copy to prevent misalignment issues */
/* Can be removed if input array can be treated as uint64_t directly */
for(int i=0; i < 8; ++i){
transposed[i] = (value >> (i * 8)) & 0xFF;
}
}
/* Calculate lookup table */
void calculate_lut(void){
/* For all byte values */
for(uint64_t i = 0; i < 256; ++i){
auto b = std::bitset<8>(i);
auto v = std::bitset<64>(0);
/* For all bits in current byte */
for(int bit=0; bit < 8; ++bit){
if(b.test(bit)){
v.set((7 - bit) * 8);
}
}
lut[i] = v.to_ullong();
}
}
int main()
{
calculate_lut();
const uint8_t matrix[] = {
0b01010101,
0b00110011,
0b00001111,
};
uint8_t transposed[8];
transpose_bitmatrix(matrix, transposed);
print_arr(transposed);
return 0;
}
How it works
your 3x8 matrix will be transposed to a 8x3 matrix, represented in an 8x8 array.
The issue is that you want to convert bits, your "horizontal" representation to a vertical one, divided over several bytes.
As I mentioned above, we can take advantage of the fact that the output (8x8) will always fit into a uint64_t. We will use this to our advantage because now we can use an uint64_t to write the 8 byte array, but we can also use it for to add, xor, etc. because we can perform basic arithmetic operations on a 64 bit integer.
Each entry in your 3x8 matrix (input) is 8 bits wide, to optimize processing we first generate 256 entry lookup table (for each byte value). The entry itself is a uint64_t and will contain a rotated version of the bits.
example:
byte = 0b01001111 = 0x4F
lut[0x4F] = 0x0001000001010101 = (uint8_t[]){ 0, 1, 0, 0, 1, 1, 1, 1 }
Now for the calculation:
For the calculations we use the uint64_t but keep in mind that under water it will represent a uint8_t[8] array. We simple shift the current value (start with 0), look up our first byte and add it to the current value.
The 'magic' here is that each byte of the uint64_t in the lookup table will either be 1 or 0 so it will only set the least significant bit (of each byte). Shifting the uint64_t will shift each byte, as long as we make sure we do not do this more than 8 times! we can do operations on each byte individually.
Issues
As someone noted in the comments: Translate(Translate(M)) != M so if you need this you need some additional work.
Perfomance can be improved by directly mapping uint64_t's instead of uint8_t[8] arrays since it omits a "safe-copy" to prevent alignment issues.
0b00000000instead of0b00000001? why is the second line0b00000100instead of0b00000010? ... – Bors