原文地址：http://www.cnblogs.com/zyl910/archive/2012/07/31/intrin02_sse.html
R：寄存器。M：64位MM寄存器；X：128位XMM寄存器；Y：256位YMM寄存器。
Name：函数名。
Name2：另一种函数名。
功能：功能描述。
Asm：汇编指令。
PCode：伪代码。
RNameName2功能AsmPCodeX_MM_SHUFFLE 混洗的掩码.4#http://msdn.microsoft.com/en-us/library/4d3eabky(vs.71).aspxX_MM_TRANSPOSE4_PS 矩阵转置.4x4#http://msdn.microsoft.com/en-us/library/5hah127h(v=vs.71).aspxX_MM_SET_EXCEPTION_STATE状态.设置异常状态#http://msdn.microsoft.com/en-us/library/s61ysx0a(v=vs.71).aspxX_MM_GET_EXCEPTION_STATE状态.取得异常状态#http://msdn.microsoft.com/en-us/library/7kzfa3h8(v=vs.71).aspxX_MM_SET_EXCEPTION_MASK状态.设置异常掩码#http://msdn.microsoft.com/en-us/library/7ad8d8fy(v=vs.71).aspxX_MM_GET_EXCEPTION_MASK状态.取得异常掩码#http://msdn.microsoft.com/en-us/library/f13f3eaz(v=vs.71).aspxX_MM_SET_ROUNDING_MODE状态.设置舍入模式#http://msdn.microsoft.com/en-us/library/y70z2105(v=vs.71).aspxX_MM_GET_ROUNDING_MODE状态.取得舍入模式#http://msdn.microsoft.com/en-us/library/wc7hx623(v=vs.71).aspxX_MM_SET_FLUSH_ZERO_MODE状态.设置下溢清零模式#http://msdn.microsoft.com/en-us/library/a8b5ts9s(v=vs.71).aspxX_MM_GET_FLUSH_ZERO_MODE状态.取得下溢清零模式#http://msdn.microsoft.com/en-us/library/5207b86e(v=vs.71).aspxX_mm_add_ss 加法.单精.标量ADDSSr.fS[0]=m1.fS[0]+m2.fS[0]; for(i=1;i<4;++i){ r.fS[i]=m1.fS[i]; }X_mm_add_ps 加法.单精.紧缩ADDPSfor(i=0;i<4;++i){ r.fS[i]=m1.fS[i]+m2.fS[i]; }X_mm_sub_ss 减法.单精.标量SUBSSr.fS[0]=m1.fS[0]-m2.fS[0]; for(i=1;i<4;++i){ r.fS[i]=m1.fS[i]; }X_mm_sub_ps 减法.单精.紧缩SUBPSfor(i=0;i<4;++i){ r.fS[i]=m1.fS[i]-m2.fS[i]; }X_mm_mul_ss 乘法.单精.标量MULSSr.fS[0]=m1.fS[0]*m2.fS[0]; for(i=1;i<4;++i){ r.fS[i]=m1.fS[i]; }X_mm_mul_ps 乘法.单精.紧缩MULPSfor(i=0;i<4;++i){ r.fS[i]=m1.fS[i]*m2.fS[i]; }X_mm_div_ss 除法.单精.标量DIVSSr.fS[0]=m1.fS[0]/m2.fS[0]; for(i=1;i<4;++i){ r.fS[i]=m1.fS[i]; }X_mm_div_ps 除法.单精.紧缩DIVPSfor(i=0;i<4;++i){ r.fS[i]=m1.fS[i]/m2.fS[i]; }X_mm_sqrt_ss 平方根.单精.标量SQRTSSr.fS[0]=sqrt(m1.fS[0]); for(i=1;i<4;++i){ r.fS[i]=m1.fS[i]; }X_mm_sqrt_ps 平方根.单精.紧缩SQRTPSfor(i=0;i<4;++i){ r.fS[i]=sqrt(m1.fS[i]); }X_mm_rcp_ss 倒数.单精.标量RCPSSr.fS[0]=1/m1.fS[0]; for(i=1;i<4;++i){ r.fS[i]=m1.fS[i]; }X_mm_rcp_ps 倒数.单精.紧缩RCPPSfor(i=0;i<4;++i){ r.fS[i]=1/m1.fS[i]; }X_mm_rsqrt_ss 平方根的倒数.单精.标量RSQRTSSr.fS[0]=1/sqrt(m1.fS[0]); for(i=1;i<4;++i){ r.fS[i]=m1.fS[i]; }X_mm_rsqrt_ps 平方根的倒数.单精.紧缩RSQRTPSfor(i=0;i<4;++i){ r.fS[i]=1/sqrt(m1.fS[i]); }X_mm_min_ss 最小值.单精.标量MINSSr.fS[0]=min(m1.fS[0], m2.fS[0]); for(i=1;i<4;++i){ r.fS[i]=m1.fS[i]; }X_mm_min_ps 最小值.单精.紧缩MINPSfor(i=0;i<4;++i){ r.fS[i]=min(m1.fS[i], m2.fS[i]); }X_mm_max_ss 最大值.单精.标量MAXSSr.fS[0]=max(m1.fS[0], m2.fS[0]); for(i=1;i<4;++i){ r.fS[i]=m1.fS[i]; }X_mm_max_ps 最大值.单精.紧缩MAXPSfor(i=0;i<4;++i){ r.fS[i]=max(m1.fS[i], m2.fS[i]); }X_mm_and_ps 逻辑位与.单精ANDPSfor(i=0;i<4;++i){ r.fS[i]=m1.fS[i]&m2.fS[i]; }X_mm_andnot_ps 逻辑位与非.单精ANDNPSfor(i=0;i<4;++i){ r.fS[i]=(!m1.fS[i])&m2.fS[i]; }X_mm_or_ps 逻辑位或.单精ORPSfor(i=0;i<4;++i){ r.fS[i]=m1.fS[i]|m2.fS[i]; }X_mm_xor_ps 逻辑位异或.单精XORPSfor(i=0;i<4;++i){ r.fS[i]=m1.fS[i]^m2.fS[i]; }X_mm_cmpeq_ss 比较.等于.单精.标量CMPEQSSr = BM(m1.fS[0] @ m2.fS[0])X_mm_cmpeq_ps 比较.等于.单精.紧缩CMPEQPSfor(i=0;i<4;++i){ r.fS[i]=BM(m1.fS[i] @ m2.fS[i]); }X_mm_cmplt_ss 比较.小于.单精.标量CMPLTSSr = BM(m1.fS[0] @ m2.fS[0])X_mm_cmplt_ps 比较.小于.单精.紧缩CMPLTPSfor(i=0;i<4;++i){ r.fS[i]=BM(m1.fS[i] @ m2.fS[i]); }X_mm_cmple_ss 比较.小于等于.单精.标量CMPLESSr = BM(m1.fS[0] @ m2.fS[0])X_mm_cmple_ps 比较.小于等于.单精.紧缩CMPLEPSfor(i=0;i<4;++i){ r.fS[i]=BM(m1.fS[i] @ m2.fS[i]); }X_mm_cmpgt_ss 比较.大于.单精.标量CMPGTSSr = BM(m1.fS[0] @ m2.fS[0])X_mm_cmpgt_ps 比较.大于.单精.紧缩CMPGTPSfor(i=0;i<4;++i){ r.fS[i]=BM(m1.fS[i] @ m2.fS[i]); }X_mm_cmpge_ss 比较.大于等于.单精.标量CMPLESSr = BM(m1.fS[0] @ m2.fS[0])X_mm_cmpge_ps 比较.大于等于.单精.紧缩CMPLEPSfor(i=0;i<4;++i){ r.fS[i]=BM(m1.fS[i] @ m2.fS[i]); }X_mm_cmpneq_ss 比较.不等于.单精.标量CMPNEQSSr = BM(m1.fS[0] @ m2.fS[0])X_mm_cmpneq_ps 比较.不等于.单精.紧缩CMPNEQPSfor(i=0;i<4;++i){ r.fS[i]=BM(m1.fS[i] @ m2.fS[i]); }X_mm_cmpnlt_ss 比较.不小于.单精.标量CMPNLTSSr = BM(m1.fS[0] @ m2.fS[0])X_mm_cmpnlt_ps 比较.不小于.单精.紧缩CMPNLTPSfor(i=0;i<4;++i){ r.fS[i]=BM(m1.fS[i] @ m2.fS[i]); }X_mm_cmpnle_ss 比较.不小于等于.单精.标量CMPNLESSr = BM(m1.fS[0] @ m2.fS[0])X_mm_cmpnle_ps 比较.不小于等于.单精.紧缩CMPNLEPSfor(i=0;i<4;++i){ r.fS[i]=BM(m1.fS[i] @ m2.fS[i]); }X_mm_cmpngt_ss 比较.不大于.单精.标量CMPNLTSSr = BM(m1.fS[0] @ m2.fS[0])X_mm_cmpngt_ps 比较.不大于.单精.紧缩CMPNLTPSfor(i=0;i<4;++i){ r.fS[i]=BM(m1.fS[i] @ m2.fS[i]); }X_mm_cmpnge_ss 比较.不大于等于.单精.标量CMPNLESSr = BM(m1.fS[0] @ m2.fS[0])X_mm_cmpnge_ps 比较.不大于等于.单精CMPNLEPSfor(i=0;i<4;++i){ r.fS[i]=BM(m1.fS[i] @ m2.fS[i]); }X_mm_cmpord_ss 比较.有序.单精.标量CMPORDSSr = BM(m1.fS[0] @ m2.fS[0])X_mm_cmpord_ps 比较.有序.单精.紧缩CMPORDPSfor(i=0;i<4;++i){ r.fS[i]=BM(m1.fS[i] @ m2.fS[i]); }X_mm_cmpunord_ss 比较.无序.单精.标量CMPUNORDSSr = BM(m1.fS[0] @ m2.fS[0])X_mm_cmpunord_ps 比较.无序.单精.紧缩CMPUNORDPSfor(i=0;i<4;++i){ r.fS[i]=BM(m1.fS[i] @ m2.fS[i]); }X_mm_comieq_ss 有序比较并设标志.相等.单精COMISSr = EFLAGS(m1.fS[0] @ m2.fS[0])X_mm_comilt_ss 有序比较并设标志.小于.单精COMISSr = EFLAGS(m1.fS[0] @ m2.fS[0])X_mm_comile_ss 有序比较并设标志.小于等于.单精COMISSr = EFLAGS(m1.fS[0] @ m2.fS[0])X_mm_comigt_ss 有序比较并设标志.大于.单精COMISSr = EFLAGS(m1.fS[0] @ m2.fS[0])X_mm_comige_ss 有序比较并设标志.大于等于.单精COMISSr = EFLAGS(m1.fS[0] @ m2.fS[0])X_mm_comineq_ss 有序比较并设标志.不等于.单精COMISSr = EFLAGS(m1.fS[0] @ m2.fS[0])X_mm_ucomieq_ss 无序比较并设标志.相等.单精UCOMISSr = EFLAGS(m1.fS[0] @ m2.fS[0])X_mm_ucomilt_ss 无序比较并设标志.小于.单精UCOMISSr = EFLAGS(m1.fS[0] @ m2.fS[0])X_mm_ucomile_ss 无序比较并设标志.小于等于.单精UCOMISSr = EFLAGS(m1.fS[0] @ m2.fS[0])X_mm_ucomigt_ss 无序比较并设标志.大于.单精UCOMISSr = EFLAGS(m1.fS[0] @ m2.fS[0])X_mm_ucomige_ss 无序比较并设标志.大于等于.单精UCOMISSr = EFLAGS(m1.fS[0] @ m2.fS[0])X_mm_ucomineq_ss 无序比较并设标志.不等于.单精UCOMISSr = EFLAGS(m1.fS[0] @ m2.fS[0])X_mm_cvt_ss2si_mm_cvtss_si32转换.单精度至符32位.标量CVTSS2SIr=(int32)m1.fS[0]X_mm_cvt_ps2pi_mm_cvtps_pi32转换.单精度至符32位.低位2个CVTPS2PIfor(i=0;i<2;++i){ r.iD[i]=(int32)m1.fS[i]; }X_mm_cvtt_ss2si_mm_cvttss_si32截尾法转换.单精度至符32位.标量CVTTSS2SIr=(int32)TRUNC(m1.fS[0])X_mm_cvtt_ps2pi_mm_cvttps_pi32截尾法转换.单精度至符32位.低位2个CVTTPS2PIfor(i=0;i<2;++i){ r.iD[i]=(int32)TRUNC(m1.fS[i]); }X_mm_cvt_si2ss_mm_cvtsi32_ss转换.符32位至单精度.标量CVTSI2SSr.fS[0]=(float)m2; for(i=1;i<4;++i){ r.fS[i]=m1.fS[i]; }X_mm_cvt_pi2ps_mm_cvtpi32_ps转换.符32位至单精度.低位2个CVTPI2PSfor(i=0;i<2;++i){ r.fS[i]=(float)m2.iD[i]; } for(i=2;i<4;++i){ r.fS[i]=m1.fS[i]; }X_mm_cvtss_f32 转换.提取低32位的单精度浮点数 r=m1.fS[0]X_mm_cvtss_si64 转换.单精度至符64位.标量CVTSS2SIr=(int64)m1.fS[0]X_mm_cvttss_si64 截尾法转换.单精度至符32位.标量CVTTSS2SIr=(int64)TRUNC(m1.fS[0])X_mm_cvtsi64_ss 转换.符64位至单精度.标量CVTSI2SSr.fS[0]=(float)m2; for(i=1;i<4;++i){ r.fS[i]=m1.fS[i]; }X_mm_shuffle_ps 混洗.单精.2源SHUFPSfor(i=0;i<2;++i){ r.fS[i]=m1.fS[(_Imm8>>(i*2)) & 3]; } for(i=2;i<4;++i){ r.fS[i]=m2.fS[(_Imm8>>(i*2)) & 3]; }X_mm_unpackhi_ps 高位解包.单精UNPCKHPSfor(i=0;i<1;++i){ r.fS[i*2]=m1.fS[2+i]; r.fS[i*2+1]=m2.fS[2+i]; }X_mm_unpacklo_ps 低位解包.单精UNPCKLPSfor(i=0;i<1;++i){ r.fS[i*2]=m1.fS[i]; r.fS[i*2+1]=m2.fS[i]; }X_mm_loadh_pi 高位传送.加载64位MOVHPS reg, memm1.mQ[1]=*m2;X_mm_movehl_ps 高到低传送.高位2组MOVHLPSr=m1; for(i=0;i<2;++i){ r.fS[i]=m2.fS[2+i]; }X_mm_movelh_ps 低到高传送.低位2组MOVLHPSr=m1; for(i=0;i<2;++i){ r.fS[2+i]=m2.fS[2]; }X_mm_storeh_pi 高位传送.存储64位MOVHPS mem, reg*A=m2.mQ[1];X_mm_loadl_pi 低位传送.加载64位MOVLPS reg, memm1.mQ[0]=*m2;X_mm_storel_pi 低位传送.存储64位MOVLPS mem, reg*A=m2.mQ[0];X_mm_movemask_ps 传送符号位生成掩码.单精MOVMSKPSr=0; for(i=0;i<4;++i){ r<<=1; r|=SBIT(m1.fS[i]); }M_m_pextrw_mm_extract_pi16传送.提取.16位PEXTRWr = ZX(m1.uW[imm8])M_m_pinsrw_mm_insert_pi16传送.插入.16位PINSRWm1.uW[imm8]=(WORD)m2M_m_pmaxsw_mm_max_pi16最大.带16位.紧缩PMAXSWfor(i=0;i<4;++i){ r.iW[i]=MAX(m1.iW[i],m2.iW[i]); }M_m_pmaxub_mm_max_pu8最大.无8位.紧缩PMAXUBfor(i=0;i<8;++i){ r.uB[i]=MAX(m1.uB[i],m2.uB[i]); }M_m_pminsw_mm_min_pi16最小.带16位.紧缩PMINSWfor(i=0;i<4;++i){ r.iW[i]=MIN(m1.iW[i],m2.iW[i]); }M_m_pminub_mm_min_pu8最小.无8位.紧缩PMINUBfor(i=0;i<8;++i){ r.uB[i]=MIN(m1.uB[i],m2.uB[i]); }M_m_pmovmskb_mm_movemask_pi8传送符号位生成掩码.字节PMOVMSKBr=0; for(i=0;i<8;++i){ r<<=1; r|=SBIT(m1.iB[i]); }M_m_pmulhuw_mm_mulhi_pu16乘法高位.无16位PMULHUWfor(i=0;i<4;++i){ r.uW[i]=hi16(m1.uW[i]*m1.uW[i]); }M_m_pshufw_mm_shuffle_pi16混洗.字.1源PSHUFWfor(i=0;i<3;++i){ r.uW[i]=m1.uW[(imm8>>(i*2)) & 3]; }M_m_maskmovq_mm_maskmove_si64选择性传送.8字节MASKMOVQfor(i=0;i<8;++i){ if(SBIT(m2.iB[i])) P[i]=m1.iB[i]; }M_m_pavgb_mm_avg_pu8平均值.无8位PAVGBfor(i=0;i<8;++i){ r.uB[i]=AVG(m1.uB[i],m2.uB[i]); }M_m_pavgw_mm_avg_pu16平均值.无16位PAVGWfor(i=0;i<4;++i){ r.uW[i]=AVG(m1.uW[i],m2.uW[i]); }M_m_psadbw_mm_sad_pu8绝对差.无8位,再水平8求和PSADBWr=0; for(i=0;i<8;++i){ r.uW[0]+=ABS((WORD)m1.uB[i] - m2.uB[i]); }X_mm_set_ss 赋值.单精.标量 r.fS[0]=arg[0]; for(i=1;i<4;++i){ r.fS[i]=0; }X_mm_set_ps1_mm_set1_ps重复赋值.单精.紧缩 for(i=0;i<4;++i){ r.fS[i]=arg[0]); }X_mm_set_ps 赋值.单精.紧缩 for(i=0;i<4;++i){ r.fS[i]=arg[i]); }X_mm_setr_ps 逆序赋值.单精.紧缩 for(i=0;i<4;++i){ r.fS[i]=arg[i?]); }X_mm_setzero_ps 赋值为零.单精.紧缩 r=0X_mm_load_ss 加载.单精.标量MOVSSr.fS[0]=_A[0]; for(i=1;i<4;++i){ r.fS[i]=0; }X_mm_load_ps1_mm_load1_ps重复加载.单精.紧缩MOVSS + Shufflingfor(i=0;i<4;++i){ r.fS[i]=_A[0]); }X_mm_load_ps 加载.单精.紧缩.对齐MOVAPSfor(i=0;i<4;++i){ r.fS[i]=_A[i]); }X_mm_loadr_ps 逆序加载.单精.紧缩.对齐MOVAPS + Shufflingfor(i=0;i<4;++i){ r.fS[i]=_A[3-i]); }X_mm_loadu_ps 加载.单精.紧缩.非对齐MOVUPSfor(i=0;i<4;++i){ r.fS[i]=_A[i]); }X_mm_store_ss 存储.单精.标量MOVSS_A[0]=m1.fS[0]X_mm_store_ps1_mm_store1_ps重复存储.单精.紧缩MOVSS + Shufflingfor(i=0;i<4;++i){ _A[i]=m1.fS[0]); }X_mm_store_ps 存储.单精.紧缩.对齐MOVAPSfor(i=0;i<4;++i){ _A[i]=m1.fS[i]); }X_mm_storer_ps 逆序存储.单精.紧缩.对齐MOVAPS + Shufflingfor(i=0;i<4;++i){ _A[i]=m1.fS[3-i]); }X_mm_storeu_ps 存储.单精.紧缩.非对齐MOVUPSfor(i=0;i<4;++i){ _A[i]=m1.fS[i]); }X_mm_move_ss 标量传送.单精MOVSSm1.fS[0]=m2.fS[0] _mm_prefetch 缓存.预取PREFETCHhttp://msdn.microsoft.com/en-us/library/84szxsww(v=vs.110).aspxM_mm_stream_pi 非时间性存储.mmMOVNTQ*_A=m1X_mm_stream_ps 非时间性存储.单精度MOVNTPS*_A=m1 _mm_sfence 存储隔离SFENCEhttp://msdn.microsoft.com/en-us/library/5h2w73d1(v=vs.110).aspx _mm_getcsr 获取MXCSRSTMXCSRr=MXCSR _mm_setcsr 设置MXCSRLDMXCSRMXCSR=m1 _mm_malloc mm分配内存（IGL）#IGL _mm_free mm释放内存（IGL）#IGL X_mm_cvtpi16_ps 转换.符16位至单精度.紧缩4个_inlinefor(i=0;i<4;++i){ r.fS[i]=(float)m1.iW[i]); }X_mm_cvtpu16_ps 转换.无16位至单精度.紧缩4个_inlinefor(i=0;i<4;++i){ r.fS[i]=(float)m1.uW[i]); }X_mm_cvtps_pi16 转换.单精度至符16位.紧缩4个_inline_mm_packs_pi32(_mm_cvtps_pi32(a), _mm_cvtps_pi32(_mm_movehl_ps(a, a)));X_mm_cvtpi8_ps 转换.符8位至单精度.低位4个_inline_mm_cvtpi16_ps(_mm_unpacklo_pi8(a, _mm_cmpgt_pi8(_mm_setzero_si64(), a)));X_mm_cvtpu8_ps 转换.无8位至单精度.低位4个_inline_mm_cvtpu16_ps(_mm_unpacklo_pi8(a, _mm_setzero_si64()));X_mm_cvtps_pi8 转换.单精度至符8位.低位4个_inline_mm_packs_pi16(_mm_cvtps_pi16(a), _mm_setzero_si64());X_mm_cvtpi32x2_ps 转换.符32位至单精度.2源_inline_mm_movelh_ps(_mm_cvt_pi2ps(_mm_setzero_ps(), a), _mm_cvt_pi2ps(_mm_setzero_ps(), b));
