任意位长的缓冲区的位移

时间:2019-12-22 00:23:28

标签: c bit-manipulation bit-shift

我一直在寻找C99中在任意位长的缓冲区上进行位移位的函数。可能有很多库在执行此操作,但是我想避免为此添加依赖项(但是我可以从库中提取函数,而无需提取过多代码)。 至关重要的是,它应该能够就地工作,并且不会损坏超出指定位长的任何内容。

为澄清歧义,问题是如何实现以下功能:

/**
 * shift a buffer right with bit granularity (little endian)
 *
 * @param   dst         destination buffer, can be equal to src
 * @param   src         source buffer
 * @param   size        length in bits of src/dst
 * @param   shift       shift amount in bits
 * @param   fill        fill value (boolean) for the MSBs
 *
 * shift is allowed to be larger than size, it behaves like they are equal
*/
void bufshrbits(void* dst,const void* src,size_t size, size_t shift, bool fill);

/**
 * shift a buffer left with bit granularity (little endian)
 *
 * @param   dst         destination buffer, can be equal to src
 * @param   src         source buffer
 * @param   size        length in bits of src/dst
 * @param   shift       shift amount in bits
 * @param   fill        fill value (boolean) for the LSBs
 *
 * shift is allowed to be larger than size, it behaves like they are equal
*/
void bufshlbits(void* dst,const void* src,size_t size, size_t shift, bool fill);

SO问题Bitwise shifting array of char's的标题不正确,实际上是在要求轮换。

2 个答案:

答案 0 :(得分:0)

我想到了以下内容,并在这里进行了测试:https://wandbox.org/permlink/g3FDe88vKgdPsGLC 它可以在任何CPU上工作,因为它仅使用字节访问(尽管未在big endian平台上进行测试)。

#include <stdint.h>
#include <stdbool.h>

/**
 * shift a buffer left with byte granularity (little endian)
 *
 * @param   dst         destination buffer, can be equal to src
 * @param   src         source buffer
 * @param   byte_size   length in bytes of src/dst
 * @param   byte_shift  shift amount in bytes
 * @param   fill8       fill value for the LSBs
 *
 * byte_shift is allowed to be larger than byte_size, it behaves like they are equal
*/
static void bufshl(void*const dst,const void*const src,size_t byte_size, size_t byte_shift, uint8_t fill8){
    if(0==byte_size) return;
    if(byte_shift>byte_size) byte_shift=byte_size;
    uint8_t*const dst8=(uint8_t*)dst;
    const uint8_t*const src8=(const uint8_t*const)src;
    for(size_t i=byte_size-1;i!=byte_shift-1;i--){dst8[i] = src8[i-byte_shift];}
    for(size_t i=0;i<byte_shift;i++){dst8[i] = fill8;}
}
/**
 * shift a buffer left with bit granularity (little endian)
 *
 * @param   dst         destination buffer, can be equal to src
 * @param   src         source buffer
 * @param   size        length in bits of src/dst
 * @param   shift       shift amount in bits
 * @param   fill        fill value for the LSBs
 *
 * shift is allowed to be larger than size, it behaves like they are equal
*/
static void bufshlbits(void*const dst,const void*const src,size_t size, size_t shift, bool fill){
    if(0==size) return;
    const uint8_t fill8 = fill ? 0xFF : 0x00;
    if(shift>size) shift=size;
    uint8_t*const dst8=(uint8_t*)dst;
    const uint8_t*const src8=(const uint8_t*const)src;
    const size_t byte_size = (size+7)/8;
    const unsigned int byte_shift=shift%8;
    const unsigned int cshift = (8-byte_shift)%8;
    const uint8_t last = src8[byte_size-1];
    const size_t lsb = shift/8;
    if(0==(shift%8)){
        bufshl(dst,src,byte_size,lsb,fill8);
    } else {
        uint8_t carry=src8[byte_size-1-lsb];
        for(size_t i=byte_size-1;i!=lsb-1;i--){
            const size_t sidx = i-1-lsb;
            const uint8_t s = sidx>byte_size ?  fill8 : src8[sidx];
            const uint8_t d = (carry<<byte_shift)|(s >> cshift);
            carry = src8[sidx];
            dst8[i] = d;
        }
    }
    for(size_t i=0;i<lsb;i++){dst8[i]=fill8;}
    if(size%8){
        const uint8_t last_mask = 0xFF<<(size % 8);
        dst8[byte_size-1] &= ~last_mask;
        dst8[byte_size-1] |= last & last_mask;
    }
}
/**
 * shift a buffer right with byte granularity (little endian)
 *
 * @param   dst         destination buffer, can be equal to src
 * @param   src         source buffer
 * @param   byte_size   length in bytes of src/dst
 * @param   byte_shift  shift amount in bytes
 * @param   fill8       fill value for the MSBs
 *
 * byte_shift is allowed to be larger than byte_size, it behaves like they are equal
*/
static void bufshr(void*const dst,const void*const src,size_t byte_size, size_t byte_shift, uint8_t fill8){
    if(0==byte_size) return;
    if(byte_shift>byte_size) byte_shift=byte_size;
    uint8_t*const dst8=(uint8_t*)dst;
    const uint8_t*const src8=(const uint8_t*const)src;
    const size_t last=byte_size-byte_shift;
    for(size_t i=0;i!=last;i++){dst8[i] = src8[i+byte_shift];}
    for(size_t i=last;i<byte_shift;i++){dst8[i] = fill8;}
}
/**
 * shift a buffer right with bit granularity (little endian)
 *
 * @param   dst         destination buffer, can be equal to src
 * @param   src         source buffer
 * @param   size        length in bits of src/dst
 * @param   shift       shift amount in bits
 * @param   fill        fill value for the MSBs
 *
 * shift is allowed to be larger than size, it behaves like they are equal
*/
static void bufshrbits(void*const dst,const void*const src,size_t size, size_t shift, bool fill){
    if(0==size) return;
    const uint8_t fill8 = fill ? 0xFF : 0x00;
    if(shift>size) shift=size;
    uint8_t*const dst8=(uint8_t*)dst;
    const uint8_t*const src8=(const uint8_t*const)src;
    const size_t byte_size = (size+7)/8;
    const unsigned int bshift=shift%8;
    const unsigned int cshift = bshift ? (8-bshift)%8 : 8;
    const uint8_t last = src8[byte_size-1];
    const size_t lsb = shift/8;
    if((0==(shift%8)) && (0==(size%8))) {
        bufshr(dst,src,byte_size,lsb,fill8);
    } else {
        const uint8_t last_mask = size%8 ? 0xFF<<(size % 8) : 0;
        uint8_t carry = lsb+1 >=byte_size ? fill8 : src8[lsb+1];
        if(lsb+1 == byte_size-1) {
            carry &= ~last_mask;
            carry |= last_mask & fill8;
        }
        if(byte_size>lsb){
            for(size_t i=0;i<byte_size-lsb-1;i++){
                const size_t sidx = i+lsb;
                uint8_t s;
                if(sidx>=byte_size-1){
                    s=(src8[sidx] & ~last_mask) | (last_mask & fill8);
                }else{
                    s=src8[sidx];
                }
                const uint8_t d = (carry<<cshift)|(s >> bshift);
                carry = sidx+2 >=byte_size? fill8 : src8[sidx+2];
                if(sidx+2 == byte_size-1) {
                    carry &= ~last_mask;
                    carry |= last_mask & fill8;
                }
                dst8[i] = d;
            }
        }
        const size_t sidx = byte_size-lsb-1+lsb;
        carry &= ~last_mask;
        carry |= last_mask & fill8;
        uint8_t s;
        if(sidx>=byte_size-1){
            s=(src8[sidx] & ~last_mask) | (last_mask & fill8);
        }else{
            s=src8[sidx];
        }
        const uint8_t d = (carry<<cshift)|(s >> bshift);
        dst8[byte_size-lsb-1] = d;
    }
    for(size_t i=byte_size-lsb;i<byte_size;i++){dst8[i]=fill8;}
    if(size%8){
        const uint8_t last_mask = 0xFF<<(size % 8);
        dst8[byte_size-1] &= ~last_mask;
        dst8[byte_size-1] |= last & last_mask;
    }
}

答案 1 :(得分:-1)

您可以通过调整索引一步将大粒度和小粒度的偏移结合起来。例如,要进行18位右移,您将从indexindex - 2的字节中得到index - 3的字节的位。

// ...┆abcd...┆       ┆       ┆       ┆       ┆...      ← src
//     |  <- 18-bit ->  |
// ...┆       ┆       ┆..abcd.┆       ┆       ┆...      ← dst
// ...₀₁₂₃₄₅₆₇⁰¹²³⁴⁵⁶⁷₀₁₂₃₄₅₆₇⁰¹²³⁴⁵⁶⁷₀₁₂₃₄₅₆₇ ...      ← bit position

这是右移的示例代码。为简单起见,我选择字节作为大的单位移位,但是为了获得良好的性能,每个单位应为32或64位字,或者甚至更好地使用SIMD一次处理128/256/512位。这就是为什么我对单位使用“肢体”,这是GMP表示大整数中每个数字的术语,因此它不取决于单位大小,并且更易于扩展。只需将char更改为uint64_t或进行类似的更改

// dst = src >> shift
void memshrbits(void* const dst, const void* const src,
                size_t size, size_t shift, int fill)
{
    char *d = (char*)dst, *s = (char*)src;

    const size_t limbWidth = CHAR_BIT * sizeof(*d); // number of bits in a limb
    assert(shift < limbWidth); // or just fill the whole dst with `fill`
                               // when shift >= limbWidth

    const size_t bitShift  = shift % limbWidth; // number of bits to shift in a limb
    const size_t limbShift = shift / limbWidth; // number of limbs to shift

    size_t srcLength = size / limbWidth;
    size_t numberOfRemainingBits = size % limbWidth;
    // backup the last bits in case `size` is odd
    size_t remainingBits = s[srcLength - 1] & ((1U << numberOfRemainingBits) - 1);

    fill = -fill;  // all ones or zero bits
    size_t i = srcLength - 1;
    // Filling dst from the right, because we're shifting right
    for (; i > limbShift; --i)
        d[i] = (s[i - limbShift    ] >> bitShift) |
               (s[i - limbShift - 1] << (limbWidth - bitShift));
    d[i--] = (s[0] >> bitShift) | (fill << (limbWidth - bitShift));
    // The remaining bits are blank spaces and will be filled with `fill`
    for (; i != (size_t)(-1); --i)
        d[i] = fill;

    // Restore the bits if shift is odd
    d[srcLength - 1] = (d[srcLength - 1] & (-1U << numberOfRemainingBits)) |
                       remainingBits;
}

如果size始终是单位的倍数(在这种情况下,即8的倍数),则更容易,因为您不必处理src末尾的奇数位,并且可以删除remainingBits个部分

对左移执行相同操作,但在相反方向进行迭代