Bitwise operations on 128-bit values on a non-sse2 arch

Question

I am writing a routine in C, targeted for an embedded platform.
In the routine I need to perform bitwise XOR and SHIFT RIGHT operations on 128-bit values.
The target arch doesn't have SSE2, hence no native 128-bit operations supported.
I came across this answer which simulates the SHIFT operations in software.
My question is, are there better ways of doing this, I mean with better data structure to represent 128-bit values and optimal way to simulate the SHIFT and XOR operations than using recursion(as done in the answer in the link). I wish to minimise usage of the limited stack memory.

Answer 1

You can use a structure to store 128 bit data as follows

typedef struct
{
    uint32_t a;
    uint32_t b;
    uint32_t c;
    uint32_t d;
} Type_128bit;

Then you can write a left shift function as follows

int leftshift(Type_128bit in, Type_128bit out, int value)
{
    int val;
    if (value >= 128)
    {
        return (-1); // error condition
    }
    else if (value < 32)
    {
        out->a = (in->a << value) | (in->b >> value);
        out->b = (in->b << value) | (in->c >> value);
        out->c = (in->c << value) | (in->d >> value);
        out->d = in->d << value;
    }
    else if (value < 64)
    {
        val = value - 32;
        out->a = (in->b << val) | (in->c >> val);
        out->b = (in->c << val) | (in->d >> val);
        out->c = (in->d << val);
        out->d = 0x00;
    }
    else if (value < 96)
    {
        val = value - 64;
        out->a = (in->c << val) | (in->d >> val);
        out->b = (in->d << val);
        out->c = 0x00;
        out->d = 0x00;
    }
    else // value < 128
    {
        val = value - 96;
        out->a = (in->d << val);
        out->b = 0x00;
        out->c = 0x00;
        out->d = 0x00;
    }
    return (0); //success
}

This will avoid the recursion of the mentioned solution and give better runtime. But code size will increase and you need to carefully test the code.

Answer 2

uint32_t *shiftL(uint32_t *val, const size_t size, const size_t nbits)  // <= 32
{
    uint32_t mask = (1 << nbits) - 1;

    mask <<= 32 - nbits;

    for(size_t cword = size; cword - 1 ; cword --)
    {
        uint32_t temp = (val[cword - 2] & mask) >> nbits
        val[cword - 1] <<= nbits;
        val |= temp;
    }
    val[0] <<= nbits;
    return val;
}