MMX Intrinsics各函数介绍
SIMD相关头文件包括：
//#include <ivec.h>//MMX //#include <fvec.h>//SSE(also include ivec.h) //#include <dvec.h>//SSE2(also include fvec.h) #include <mmintrin.h> //MMX #include <xmmintrin.h> //SSE(include mmintrin.h) #include <emmintrin.h> //SSE2(include xmmintrin.h) #include <pmmintrin.h> //SSE3(include emmintrin.h) #include <tmmintrin.h>//SSSE3(include pmmintrin.h) #include <smmintrin.h>//SSE4.1(include tmmintrin.h) #include <nmmintrin.h>//SSE4.2(include smmintrin.h) #include <wmmintrin.h>//AES(include nmmintrin.h) #include <immintrin.h>//AVX(include wmmintrin.h) #include <intrin.h>//(include immintrin.h)mmintrin.h为MMX头文件，其中__m64的定义为：
typedef union __declspec(intrin_type) _CRT_ALIGN(8) __m64
{
unsigned __int64 m64_u64;
float m64_f32[2];
__int8 m64_i8[8];
__int16 m64_i16[4];
__int32 m64_i32[2];
__int64 m64_i64;
unsigned __int8 m64_u8[8];
unsigned __int16 m64_u16[4];
unsigned __int32 m64_u32[2];
} __m64;mmintrin.h
文件中各函数的介绍
/* General support intrinsics */ //Empties the multimedia state,清除MMX寄存器中的内容，即初始化(以避免和浮点数 //操作发生冲突)，详细说明见参考文献1 void _m_empty(void);//_mm_empty //Converts the integer object _I to a 64-bit __m64 object, r0=_I, r1=0 __m64 _m_from_int(int _I);//_mm_cvtsi32_si64 //Converts the lower 32 bits of the __m64 object _M to an integer, r=_M0 int _m_to_int(__m64 _M);//_mm_cvtsi64_si32 //Packs the four 16-bit values from _MM1 into the lower four 8-bit values of //the result with signed saturation, and packs the four 16-bit values from _MM2 //into the upper four 8-bit values of the result with signed saturation __m64 _m_packsswb(__m64 _MM1, __m64 _MM2);//_mm_packs_pi16 //Packs the two 32-bit values from _MM1 into the lower two 16-bit values of the // result with signed saturation, and packs the two 32-bit values from _MM2 into // the upper two 16-bit values of the result with signed saturation __m64 _m_packssdw(__m64 _MM1, __m64 _MM2);//_mm_packs_pi32 //Packs the four 16-bit values from _MM1 into the lower four 8-bit values of the //result with unsigned saturation, and packs the four 16-bit values from _MM2 into //the upper four 8-bit values of the result with unsigned saturation __m64 _m_packuswb(__m64 _MM1, __m64 _MM2);//_mm_packs_pu16 //_MM1=(_MM10, _MM11, _MM12, _MM13, _MM14, _MM15, _MM16, _MM17), //_MM2=(_MM20, _MM21, _MM22, _MM23, _MM24, _MM25, _MM26, _MM27), //则r=(_MM14, _MM24, _MM15, _MM25, _MM16, _MM26, _MM17, _MM27) __m64 _m_punpckhbw(__m64 _MM1, __m64 _MM2);//_mm_unpackhi_pi8 //_MM1=(_MM10, _MM11, _MM12, _MM13),_MM10为低位，_MM2=(_MM20, _MM21, _MM22, _MM23), //则r=(_MM12, _MM22, _MM13, _MM23) __m64 _m_punpckhwd(__m64 _MM1, __m64 _MM2);//_mm_unpackhi_pi16 //MM1=(_MM10, _MM11),_MM10为低位，_MM2=(_MM20, _MM21),则r=(_MM11, _MM21) __m64 _m_punpckhdq(__m64 _MM1, __m64 _MM2);//_mm_unpackhi_pi32 //_MM1=(_MM10, _MM11, _MM12, _MM13, _MM14, _MM15, _MM16, _MM17), //_MM2=(_MM20, _MM21, _MM22, _MM23, _MM24, _MM25, _MM26, _MM27), //则r=(_MM10, _MM20, _MM11, _MM21, _MM12, _MM22, _MM13, _MM23) __m64 _m_punpcklbw(__m64 _MM1, __m64 _MM2);//_mm_unpacklo_pi8 //_MM1=(_MM10, _MM11, _MM12, _MM13),_MM10为低位，_MM2=(_MM20, _MM21, _MM22, _MM23), //则r=(_MM10, _MM20, _MM11, _MM21) __m64 _m_punpcklwd(__m64 _MM1, __m64 _MM2);//_mm_unpacklo_pi16 //MM1=(_MM10, _MM11),_MM10为低位，_MM2=(_MM20, _MM21),则r=(_MM10, _MM20) __m64 _m_punpckldq(__m64 _MM1, __m64 _MM2);//mm_unpacklo_pi32 /* Packed arithmetic intrinsics */ //Adds the eight 8-bit values in _MM1 to the eight 8-bit values in _MM2 __m64 _m_paddb(__m64 _MM1, __m64 _MM2);//_mm_add_pi8 //Adds the four 16-bit values in _MM1 to the four 16-bit values in _MM2 __m64 _m_paddw(__m64 _MM1, __m64 _MM2);//_mm_add_pi16 //Adds the two 32-bit values in _MM1 to the two 32-bit values in _MM2 __m64 _m_paddd(__m64 _MM1, __m64 _MM2);//_mm_add_pi32 //Adds the eight signed 8-bit values in _MM1 to the eight signed 8-bit values in _MM2 //and saturates __m64 _m_paddsb(__m64 _MM1, __m64 _MM2);//_mm_adds_pi8 //Adds the four signed 16-bit values in _MM1 to the four signed 16-bit values in _MM2 //and saturates __m64 _m_paddsw(__m64 _MM1, __m64 _MM2);//_mm_adds_pi16 //Adds the eight unsigned 8-bit values in _MM1 to the eight unsigned 8-bit values //in _MM2 and saturates __m64 _m_paddusb(__m64 _MM1, __m64 _MM2);//_mm_adds_pu8 //Add the four unsigned 16-bit values in _MM1 to the four unsigned 16-bit values //in _MM2 and saturates __m64 _m_paddusw(__m64 _MM1, __m64 _MM2);//_mm_adds_pu16 //Subtracts the eight 8-bit values in _MM2 from the eight 8-bit values in _MM1 __m64 _m_psubb(__m64 _MM1, __m64 _MM2);//_mm_sub_pi8 //Subtracts the four 16-bit values in _MM2 from the four 16-bit values in _MM1 __m64 _m_psubw(__m64 _MM1, __m64 _MM2);//_mm_sub_pi16 //Subtracts the two 32-bit values in _MM2 from the two 32-bit values in _MM1 __m64 _m_psubd(__m64 _MM1, __m64 _MM2);//_mm_sub_pi32 //Subtracts the eight signed 8-bit values in _MM2 from the eight signed 8-bit //values in _MM1 and saturates __m64 _m_psubsb(__m64 _MM1, __m64 _MM2);//_mm_subs_pi8 //Subtracts the four signed 16-bit values in _MM2 from the four signed 16-bit //values in _MM1 and saturates __m64 _m_psubsw(__m64 _MM1, __m64 _MM2);//_mm_subs_pi16 //Subtracts the eight unsigned 8-bit values in _MM2 from the eight unsigned 8-bit //values in _MM1 and saturates __m64 _m_psubusb(__m64 _MM1, __m64 _MM2);//_mm_subs_pu8 //Subtracts the four unsigned 16-bit values in _MM2 from the four unsigned 16-bit //values in _MM1 and saturates __m64 _m_psubusw(__m64 _MM1, __m64 _MM2);//_mm_subs_pu16 //Multiplies four 16-bit values in _MM1 by four 16-bit values in _MM2 to produce //four 32-bit intermediate results, which are then summed by pairs to produce two //32-bit results,r0=_MM10*_MM20+_MM11*_MM21, r1=_MM12*_MM22+_MM13*_MM23 __m64 _m_pmaddwd(__m64 _MM1, __m64 _MM2);//_mm_madd_pi16 //Multiplies four signed 16-bit values in _MM1 by four signed 16-bit values in _MM2 //and produces the high 16 bits of the four results __m64 _m_pmulhw(__m64 _MM1, __m64 _MM2);//_mm_mulhi_pi16 //Multiplies four 16-bit values in _MM1 by four 16-bit values in _MM2 and produces //the low 16 bits of the four results __m64 _m_pmullw(__m64 _MM1, __m64 _MM2);//_mm_mullo_pi16 /* Shift intrinsics */ //Shifts four 16-bit values in _M left the amount specified by _Count //while shifting in zeros,左移_Count位，移出位补0 __m64 _m_psllw(__m64 _M, __m64 _Count);//_mm_sll_pi16 //Shifts four 16-bit values in _M left the amount specified by _Ccount while //shifting in zeros,左移_Count位，移出位补0,_Count需是一个立即数 //汇编语言中的立即数相当于高级语言中的常量(常数)，它是直接出现在指令中的数， //不用存储在寄存器或存储器中的数 __m64 _m_psllwi(__m64 _M, int _Count);//_mm_slli_pi16 //Shifts two 32-bit values in _M left the amount specified by _Count //while shifting in zeros __m64 _m_pslld(__m64 _M, __m64 _Count);//_mm_sll_pi32 //Shifts two 32-bit values in _M left the amount specified by _Count //while shifting in zeros __m64 _m_pslldi(__m64 _M, int _Count);//_mm_slli_pi32 //Shifts the 64-bit value in _M left the amount specified by _Count //while shifting in zeros __m64 _m_psllq(__m64 _M, __m64 _Count);//_mm_sll_si64 //Shifts the 64-bit value in _M left the amount specified by _Count //while shifting in zeros __m64 _m_psllqi(__m64 _M, int _Count);//_mm_slli_si64 //Shifts four 16-bit values in _M right the amount specified by _Count //while shifting in the sign bit __m64 _m_psraw(__m64 _M, __m64 _Count);//_mm_sra_pi16 //Shifts four 16-bit values in _M right the amount specified by _Count //while shifting in the sign bit __m64 _m_psrawi(__m64 _M, int _Count);//_mm_srai_pi16 //Shifts two 32-bit values in _M right the amount specified by _Count //while shifting in the sign bit __m64 _m_psrad(__m64 _M, __m64 _Count);//_mm_sra_pi32 //Shifts two 32-bit values in _M right the amount specified by _Count //while shifting in the sign bit __m64 _m_psradi(__m64 _M, int _Count);//_mm_srai_pi32 //Shifts four 16-bit values in _M right the amount specified by _Count //while shifting in zeros __m64 _m_psrlw(__m64 _M, __m64 _Count);//_mm_srl_pi16 //Shifts four 16-bit values in _M right the amount specified by _Count //while shifting in zeros __m64 _m_psrlwi(__m64 _M, int _Count);//_mm_srli_pi16 //Shifts two 32-bit values in _M right the amount specified by _Count //while shifting in zeros __m64 _m_psrld(__m64 _M, __m64 _Count);//_mm_srl_pi32 //Shifts two 32-bit values in _M right the amount specified by _Count //while shifting in zeros __m64 _m_psrldi(__m64 _M, int _Count);//_mm_srli_pi32 //Shifts the 64-bit value in _M right the amount specified by _Count //while shifting in zeros __m64 _m_psrlq(__m64 _M, __m64 _Count);//_mm_srl_si64 //Shifts the 64-bit value in _M right the amount specified by _Count //while shifting in zeros __m64 _m_psrlqi(__m64 _M, int _Count);//_mm_srli_si64 /* Logical intrinsics */ //Performs a bitwise AND of the 64-bit value in _MM1 with the 64-bit value in _MM2 __m64 _m_pand(__m64 _MM1, __m64 _MM2);//_mm_and_si64 //Performs a logical NOT on the 64-bit value in _MM1 and use the result in a //bitwise AND with the 64-bit value in _MM2 __m64 _m_pandn(__m64 _MM1, __m64 _MM2);//_mm_andnot_si64 //Performs a bitwise OR of the 64-bit value in _MM1 with the 64-bit value in _MM2 __m64 _m_por(__m64 _MM1, __m64 _MM2);//_mm_or_si64 //Performs a bitwise XOR of the 64-bit value in _MM1 with the 64-bit value in _MM2 __m64 _m_pxor(__m64 _MM1, __m64 _MM2);//_mm_xor_si64 /* Comparison intrinsics */ //If the respective 8-bit values in _MM1 are equal to the respective //8-bit values in _MM2, sets the respective 8-bit resulting values to //all ones; otherwise, sets them to all zeros __m64 _m_pcmpeqb(__m64 _MM1, __m64 _MM2);//_mm_cmpeq_pi8 //If the respective 16-bit values in _MM1 are equal to the respective //16-bit values in _MM2, sets the respective 16-bit resulting values //to all ones; otherwise, sets them to all zeros __m64 _m_pcmpeqw(__m64 _MM1, __m64 _MM2);//_mm_cmpeq_pi16 //If the respective 32-bit values in _MM1 are equal to the respective //32-bit values in _MM2, sets the respective 32-bit resulting values //to all ones; otherwise, sets them to all zeros __m64 _m_pcmpeqd(__m64 _MM1, __m64 _MM2);//_mm_cmpeq_pi32 //If the respective 8-bit values in _MM1 are greater than the respective //8-bit values in _MM2, sets the respective 8-bit resulting values to all ones; //otherwise, sets them to all zeros __m64 _m_pcmpgtb(__m64 _MM1, __m64 _MM2);//_mm_cmpgt_pi8 //If the respective 16-bit values in _MM1 are greater than the respective 16-bit //values in _MM2, sets the respective 16-bit resulting values to all ones; //otherwise, sets them to all zeros __m64 _m_pcmpgtw(__m64 _MM1, __m64 _MM2);//_mm_cmpgt_pi16 //If the respective 32-bit values in _MM1 are greater than the respective 32-bit //values in _MM2, sets the respective 32-bit resulting values to all ones; //otherwise, sets them all to zeros __m64 _m_pcmpgtd(__m64 _MM1, __m64 _MM2);//_mm_cmpgt_pi32 /* Utility intrinsics */ //Sets the 64-bit value to zero __m64 _mm_setzero_si64(void); //Sets the two signed 32-bit integer values,r0=_I0, r1=_I1 __m64 _mm_set_pi32(int _I1, int _I0); //r0=_S0, r1=_S1, r2=_S2, r3=_S3 __m64 _mm_set_pi16(short _S3, short _S2, short _S1, short _S0); //r0=_B0, r1=_B1, r2=_B2, r3=_B3, r4=_B4, ..., r7=_B7 __m64 _mm_set_pi8(char _B7, char _B6, char _B5, char _B4, char _B3, char _B2, char _B1, char _B0); //Sets the two signed 32-bit integer values to _I,r0=r1=_I __m64 _mm_set1_pi32(int _I); //Sets the four signed 16-bit integer values to _S, r0=r1=r2=r3=_S __m64 _mm_set1_pi16(short _S); //Sets the eight signed 8-bit integer values to _B, r0=r1...=r7=_B __m64 _mm_set1_pi8(char _B); //Sets the two signed 32-bit integer values in reverse order,r0=_I1, r1=_I0 __m64 _mm_setr_pi32(int _I1, int _I0); //Sets the four signed 16-bit integer values in reverse order, //r0=_S3, r1=_S2, r2=_S1, r3=_S0 __m64 _mm_setr_pi16(short _S3, short _S2, short _S1, short _S0); //Sets the eight signed 8-bit integer values in reverse order //r0=_B7, r1=_B6, r2=_B5, r3=_B4, r4=_B3, r5=_B2, r6=_B1, r7=_B0 __m64 _mm_setr_pi8(char _B7, char _B6, char _B5, char _B4, char _B3, char _B2, char _B1, char _B0); /* Alternate intrinsic name definitions */ #define _mm_empty _m_empty #define _mm_cvtsi32_si64 _m_from_int #define _mm_cvtsi64_si32 _m_to_int #define _mm_packs_pi16 _m_packsswb #define _mm_packs_pi32 _m_packssdw #define _mm_packs_pu16 _m_packuswb #define _mm_unpackhi_pi8 _m_punpckhbw #define _mm_unpackhi_pi16 _m_punpckhwd #define _mm_unpackhi_pi32 _m_punpckhdq #define _mm_unpacklo_pi8 _m_punpcklbw #define _mm_unpacklo_pi16 _m_punpcklwd #define _mm_unpacklo_pi32 _m_punpckldq #define _mm_add_pi8 _m_paddb #define _mm_add_pi16 _m_paddw #define _mm_add_pi32 _m_paddd #define _mm_adds_pi8 _m_paddsb #define _mm_adds_pi16 _m_paddsw #define _mm_adds_pu8 _m_paddusb #define _mm_adds_pu16 _m_paddusw #define _mm_sub_pi8 _m_psubb #define _mm_sub_pi16 _m_psubw #define _mm_sub_pi32 _m_psubd #define _mm_subs_pi8 _m_psubsb #define _mm_subs_pi16 _m_psubsw #define _mm_subs_pu8 _m_psubusb #define _mm_subs_pu16 _m_psubusw #define _mm_madd_pi16 _m_pmaddwd #define _mm_mulhi_pi16 _m_pmulhw #define _mm_mullo_pi16 _m_pmullw #define _mm_sll_pi16 _m_psllw #define _mm_slli_pi16 _m_psllwi #define _mm_sll_pi32 _m_pslld #define _mm_slli_pi32 _m_pslldi #define _mm_sll_si64 _m_psllq #define _mm_slli_si64 _m_psllqi #define _mm_sra_pi16 _m_psraw #define _mm_srai_pi16 _m_psrawi #define _mm_sra_pi32 _m_psrad #define _mm_srai_pi32 _m_psradi #define _mm_srl_pi16 _m_psrlw #define _mm_srli_pi16 _m_psrlwi #define _mm_srl_pi32 _m_psrld #define _mm_srli_pi32 _m_psrldi #define _mm_srl_si64 _m_psrlq #define _mm_srli_si64 _m_psrlqi #define _mm_and_si64 _m_pand #define _mm_andnot_si64 _m_pandn #define _mm_or_si64 _m_por #define _mm_xor_si64 _m_pxor #define _mm_cmpeq_pi8 _m_pcmpeqb #define _mm_cmpeq_pi16 _m_pcmpeqw #define _mm_cmpeq_pi32 _m_pcmpeqd #define _mm_cmpgt_pi8 _m_pcmpgtb #define _mm_cmpgt_pi16 _m_pcmpgtw #define _mm_cmpgt_pi32 _m_pcmpgtd
参考文献：1、http://software.intel.com/sites/products/documentation/studio/composer/en-us/2011Update/compiler_c/intref_cls/common/intref_mmx_emms_usage.htm