使用内在函数对128、256、512位注册表进行全局按位移位？

Question

Consider an array of 64 bit unsigned integers, like:

std::array<unsigned long long int, 20> a;

What is the fastest way, including using intel or compiler intrinsics ( this or that ) (using g++ 5.3), to perform a global bitshift (right or left) as this array was a single bit integer?

Answer 1

You might want to look at std::bitset , which is a container for a number of bits known at compile time. If I'm understanding your question right, that is what you are trying to simulate with your array. The bitset class includes overloaded >> and << operators to perform bit shifts, and those implementations may be optimised in your compiler/standard library combination.

Answer 2

Here are some x86 left shift functions that use xmm and ymm registers through intrinsics. It shouldn't be too hard to make corresponding right shift functions. They are taken from a software lfsr benchmark :

//----------------------------------------------------------------------------
// bit shift left a 128-bit value using xmm registers
//          __m128i *data - data to shift
//          int count     - number of bits to shift
// return:  __m128i       - carry out bit(s)

static __m128i bitShiftLeft128xmm (__m128i *data, int count)
   {
   __m128i innerCarry, carryOut;

   innerCarry = _mm_srli_epi64 (*data, 64 - count);      // carry outs in bit 0 of each qword
   carryOut   = _mm_shuffle_epi32 (innerCarry, 0xFE);    // upper carry in xmm bit 0, others zero
   innerCarry = _mm_shuffle_epi32 (innerCarry, 0xCF);    // lower carry in xmm bit 64, others zero
   *data = _mm_slli_epi64 (*data, count);                // shift all qwords left
   *data = _mm_or_si128 (*data, innerCarry);             // propagate carry out from low qword
   return carryOut;
   }

//----------------------------------------------------------------------------
// bit shift left a 256-bit value using xmm registers
//          __m128i *data - data to shift, ls part stored first 
//          int count     - number of bits to shift
// return:  __m128i       - carry out bit(s)

static __m128i bitShiftLeft256xmm (__m128i *data, int count)
   {
   __m128i carryOut0, carryOut1;

   carryOut0 = bitShiftLeft128xmm (&data [0], count);
   carryOut1 = bitShiftLeft128xmm (&data [1], count);
   data [1] = _mm_or_si128 (data [1], carryOut0);
   return carryOut1;
   }

//----------------------------------------------------------------------------
// bit shift left a 512-bit value using xmm registers
//          __m128i *data - data to shift, ls part stored first 
//          int count     - number of bits to shift
// return:  __m128i       - carry out bit(s)

static __m128i bitShiftLeft512xmm (__m128i *data, int count)
   {
   __m128i carryOut0, carryOut1;

   carryOut0 = bitShiftLeft256xmm (&data [0], count);
   carryOut1 = bitShiftLeft256xmm (&data [2], count);
   data [2] = _mm_or_si128 (data [2], carryOut0);
   return carryOut1;
   }


//----------------------------------------------------------------------------
// bit shift left a 256-bit value using ymm registers
//          __m256i *data - data to shift
//          int count     - number of bits to shift
// return:  __m256i       - carry out bit(s)

static __m256i bitShiftLeft256ymm (__m256i *data, int count)
   {
   __m256i innerCarry, carryOut, rotate;

   innerCarry = _mm256_srli_epi64 (*data, 64 - count);                        // carry outs in bit 0 of each qword
   rotate     = _mm256_permute4x64_epi64 (innerCarry, 0x93);                  // rotate ymm left 64 bits
   innerCarry = _mm256_blend_epi32 (_mm256_setzero_si256 (), rotate, 0xFC);   // clear lower qword
   *data    = _mm256_slli_epi64 (*data, count);                               // shift all qwords left
   *data    = _mm256_or_si256 (*data, innerCarry);                            // propagate carrys from low qwords
   carryOut   = _mm256_xor_si256 (innerCarry, rotate);                        // clear all except lower qword
   return carryOut;
   }

//----------------------------------------------------------------------------
// bit shift left a 512-bit value using ymm registers
//          __m256i *data - data to shift, ls part stored first 
//          int count     - number of bits to shift
// return:  __m256i       - carry out bit(s)

static __m256i bitShiftLeft512ymm (__m256i *data, int count)
   {
   __m256i carryOut0, carryOut1;

   carryOut0 = bitShiftLeft256ymm (&data [0], count);
   carryOut1 = bitShiftLeft256ymm (&data [1], count);
   data [1] = _mm256_or_si256 (data [1], carryOut0);
   return carryOut1;
   }

//----------------------------------------------------------------------------