TMS320C6000内联指令汇编.doc

内联指令汇编指令简要描述图示int _abs (int src);int _labs (_int40_t src)ABS返回src的绝对值int _add2 (int src1, int src2)ADD2把src1的高、低16位和src2的高、低16位分别相加，放入结果的高、低16位ushort & _amem2 (void *ptr);LDHUSTHU从内存中加载一个halfword到dst里，必须2byte对齐（读或存）const ushort & _amem2_const (const void *ptr);LDHU必须2byte对齐（读）unsigned & _amem4 (void *ptr);LDWSTW必须4byte对齐（读或存）const unsigned & _amem4_const (const void *ptr);LDW必须4byte对齐（读）double & _amemd8 (void *ptr);LDW/LDWSTW/STW必须8byte对齐（读或存）const double & _amemd8_const (const void *ptr);LDDW必须8byte对齐（读）unsigned _clr (unsigned src2, unsigned csta,unsigned cstb);CLR指定了从需要清0的首位和末位unsigned _clrr (unsigned src2, int src1);CLR将src2中指定位清0，清0的首位和末位由src1的低10位指定_int40_t _dtol (double src);将一个double寄存器重新解释成一个_int40_tlong long _dtoll (double src);将一个double寄存器重新解释成一个long longint _ext (int src2, unsigned csta, unsigned cstb);EXT从src2里提取csta和cstb指定的区域且符号扩展到32位。提取出的区域先符号左移再右移。int _extr (int src2, int src1);EXT同上，区别：左右移的位数由src1的低10位指定unsigned _extu (unsigned src2, unsigned csta , unsigned cstb);EXTU同上上，区别最后是0扩展到32位。unsigned _extur (unsigned src2, int src1);EXTU同上，区别：左右移的位数由src1的低10位指定例：_ftoi (1.0) = 1065353216Uunsigned _ftoi (float src);将float的比特位解释成unsignedunsigned _hi (double src);返回double寄存器的高位（奇数位）unsigned _hill (long long src);返回longlong寄存器的高位（奇数位）double _itod (unsigned src2, unsigned src1);创建一个新的double寄存器为了解释2个unsigned的值，其中src2是高（奇数）寄存器，src1是低（偶数）寄存器float _itof (unsigned src);将unsigned中的比特位解释成float例：_itof (0x3f800000) = 1.0long long _itoll (unsigned src2, unsigned src1);创建一个新的longlong寄存器为了解释2个unsigned的值，其中src2是高（奇数）寄存器，src1是低（偶数）unsigned _lmbd (unsigned src1, unsigned src2);LMBD搜索src2里面的1或0，1或0是由src1的LSB决定的，返回比特位变化的位数unsigned _lo (double src);返回double寄存器的低（奇数）寄存器unsigned _loll (long long src);返回longlong寄存器的低（奇数）寄存器double _ltod (_int40_t src);把一个_int40_t寄存器解释成一个double寄存器double _lltod (long long src);把一个longlong寄存器解释成一个double寄存器int _mpy (int src1, int src2); MPYSrc1和src2相乘，操作数默认为有符号的int _mpyus (unsigned src1, int src2);MPYUS无符号src1和有符号src2相乘，S是用来那个是有符号的操作数，当两个操作数都是有符号的或者无符号的int _mpysu (int src1, unsigned src2);MPYSU同上unsigned _mpyu (unsigned src1, unsigned src2);MPYU同上上上，默认为无符号int _mpyh (int src1, int src2);MPYH同上，区别见图示int _mpyhus (unsigned src1, int src2);MPYHUSint _mpyhsu (int src1, unsigned src2);MPYHSUunsigned _mpyhu (unsigned src1, unsigned src2);MPYHUint _mpyhl (int src1, int src2);MPYHL同上，区别见图示int _mpyhuls (unsigned src1, int src2);MPYHULSint _mpyhslu (int src1, unsigned src2);MPYHSLUunsigned _mpyhlu (unsigned src1, unsigned src2);MPYHLUint _mpylh (int src1, int src2);MPYLHint _mpyluhs (unsigned src1, int src2);MPYLUHSint _mpylshu (int src1, unsigned src2);MPYLSHUunsigned _mpylhu (unsigned src1, unsigned src2);MPYLHUvoid _nassert (int src);不生成代码，告诉优化器一些事情unsigned _norm (int src);unsigned _lnorm (_int40_t src);NORM返回src2的冗余的符号比特位的个数，具体见图示int _sadd (int src1, int src2);long _lsadd (int src1, _int40_t src2);SADD将src1和src2相加，且饱和其结果int _sat (_int40_t src2);SAT将一个40比特的long转换为一个32比特的有符号int，如有需要，对结果进行饱和unsigned _set (unsigned src2, unsigned csta , unsigned cstb);SET将src2中指定的区域置位1，指定的区域由csta和cstb指定unsigned _setr (unit src2, int src1);SETint _smpy (int src1, int src2);SMPY把src1的低16位和src2的低16位相乘int _smpyh (int src1, int src2);SMPYH高16位int _smpyhl (int src1, int src2);SMPYHLint _smpylh (int src1, int src2);SMPYLHint _sshl (int src2, unsigned src1);SSHL以src1操作数将src2左移，并且将结果饱和在32位int _ssub (int src1, int src2);_int40_t _lssub (int src1, _int40_t src2);SSUB从src1中减去src2，并饱和结果（src1-src2）unsigned _subc (unsigned src1, unsigned src2);SUBC有条件的减和左移（常用于除法）int _sub2 (int src1, int src2);SUB2把src1的高低16位分别减去src2的高低16位。任何低16位的借位不会影响高16位。int _abs2 (int src);ABS2计算16位的绝对值int _add4 (int src1, int src2);ADD4把src1和src2的4对8位数相加。不会进行饱和，进位不会影响其他的8位数long long & _amem8 (void *ptr);LDDWSTDW加载和存储8bytes，指针必须8byte对齐const long long & _amem8_const (const void *ptr);LDDW加载8bytes，指针必须8byte对齐_float2_t & _amem8_f2(void * ptr);LDDWSTDW加载和存储8bytes，指针必须8byte对齐，必须包含c6x.hconst _float2_t & _amem8_f2_const(void * ptr);LDDW加载8bytes，指针必须8byte对齐，必须包含c6x.hdouble & _amemd8 (void *ptr);LDDWSTDWconst double & _amemd8_const (const void *ptr);LDDWint _avg2 (int src1, int src2);AVG2计算每对有符号16位置的平均值unsigned _avgu4 (unsigned, unsigned);AVGU4计算每对有符号8位数的平均值unsigned _bitc4 (unsigned src);BITC4统计每个8位的比特位是1的个数，写入结果对应位置unsigned _bitr (unsigned src);BITR翻转比特位的顺序int _cmpeq2 (int src1, int src2);CMPEQ2比较每16位的值是否相等，结果放入dst的最低2位int _cmpeq4 (int src1, int src2);CMPEQ4比较每8位的值是否相等，结果放入dst的最低4位，相等置1，否则为0int _cmpgt2 (int src1, int src2);CMPGT2每16位有符号比较，src1src2，置为1；否则置为0。结果放入dst的最低2位unsigned _cmpgtu4 (unsigned src1, unsigned src2);CMPGTU4每8位无符号比较，src1src2，置为1；否则置为0。结果放入dst的最低4位unsigned _deal (unsigned src );DEAL将src中的比特位的奇数位和偶数位抽出来进行重组，偶数位放在低的16位，奇数位放在高的16位int _dotp2 (int src1, int src2);_int40_t _ldotp2 (int src1, int src2);DOTP2DOTP2将src1中的和src2中的16位有符号对进行点积，结果被写成有符号32位int或者符号扩展为64位int _dotpn2 (int src1, int src2);DOTPN2将src1和src2中的16位有符号数进行点积相减int _dotpnrsu2 (int src1, unsigned src2);DOTPNRSU2Src1和src2的高16位的点积减去低16位的点积。Src1中的数被当做有符号，src2中的数被当做无符号，再加上215，结果再符号右移16位int _dotprsu2 (int src1, unsigned src2);DOTPRSU2Src1和src2的高16位的点积加上低16位的点积。Src1中的数被当做有符号，src2中的数被当做无符号，再加上215，结果再符号右移16位int _dotpsu4 (int src1, unsigned src2);DOTPSU4将src1和src2的每8位进行相乘再求和，src1的每8位数被当做有符号，src2的每8位数被当做无符号unsigned _dotpu4 (unsigned src1, unsigned src2);DOTPU4都被当做无符号的int _gmpy4 (int src1, int src2);GMPY4将src1和src2的4个无符号进行伽罗瓦域的乘法int _max2 (int src1, int src2);MAX2将src1和src2的2个有符号16位整数比较，取较大值int _min2 (int src1, int src2);MIN2将src1和src2的2个有符号16位整数比较，取较小值unsigned _maxu4 (unsigned src1, unsigned src2);MAXU4将src1和src2的4个无符号8位整数比较，取较大值unsigned _minu4 (unsigned src1, unsigned src2);MINU4将src1和src2的4个无符号8位整数比较，取较小值ushort & _mem2 (void * ptr);LDB/LDBSTB/STB加载和存储2byte，不需要对齐const ushort & _mem2_const (const void * ptr);LDB/LDB加载2byte，不需要对齐unsigned & _mem4 (void * ptr);LDNWSTNW加载和存储4byte，不需要对齐const unsigned & _mem4_const (const void * ptr);LDNW加载4byte，不需要对齐long long & _mem8 (void * ptr);LDNDWSTNDW加载和存储8byte，不需要对齐const long long & _mem8_const (const void * ptr);LDNDW加载8byte，不需要对齐double & _memd8 (void * ptr);LDNDWSTNDW加载和存储8byte，不需要对齐const double & _memd8_const (const void * ptr);LDNDW加载8byte，不需要对齐long long _mpy2ll (int src1, int src2);MPY2将src1和src2中的2个有符号16位分别相乘，将2个32位的结果写入longlong中long long _mpyhill (int src1, int src2);MPYHI将src1中高16位作为1个有符号16位乘以src2的有符号32位，结果写入longlong的低48位long long _mpylill (int src1, int src2);MPYLI将src1中低16位作为1个有符号16位乘以src2的有符号32位，结果写入longlong的低48位int _mpyhir (int src1, int src2);MPYHIR将src1的高16位作为一个16位有符号乘以src2的有符号32位。乘积利用round模式通过加214转成32位，最后再右移15位int _mpylir (int src1, int src2);MPYLIR将src1的低16位作为一个16位有符号乘以src2的有符号32位。乘积利用round模式通过加214转成32位，最后再右移15位long long _mpysu4ll (int src1, unsigned src2);MPYSU4将src1的4个8位有符号乘src2的4个8位无符号，得到4个16位有符号，组成一个64位long long _mpyu4ll (unsigned src1, unsigned src2);MPYU4将src1和src2的4个无符号8位相乘，得到4个无符号16位组成一个64位的数int _mvd (int src2 );MVD将src2的数据移入返回值中，利用了乘法流水线（延迟）unsigned _pack2 (unsigned src1, unsigned src2);PACK2unsigned _packh2 (unsigned src1, unsigned src2);PACKH2unsigned _packh4 (unsigned src1, unsigned src2);PACKH4unsigned _packl4 (unsigned src1, unsigned src2);PACKL4unsigned _packhl2 (unsigned src1, unsigned src2);PACKHL2unsigned _packlh2 (unsigned src1, unsigned src2);PACKLH2unsigned _rotl (unsigned src1, unsigned src2);ROTL按照src1的最低5位的数去左移src2的32位，src1中剩下的高的5-31位被忽略int _sadd2 (int src1, int src2);SADD2将src1和src2中的2个16位有符号数相加，生成2个16有符号数并且是饱和过的。int _saddus2 (unsigned src1, int src2);SADDUS2将src1中的2个无符号16位数和src中的2个16位有符号数相加，得到2个无符号16位数unsigned _saddu4 (unsigned src1, unsigned src2);SADDU4将src1和src2中的4个无符号8位数相加unsigned _shfl (unsigned src2);SHFL将src2的高16和低16位进行交织unsigned _shlmb (unsigned src1, unsigned src2);SHLMB将src2左移1byte，然后将src1的最高位充入src2左移后多出来的位置unsigned _shrmb (unsigned src1, unsigned src2);SHRMB将src2右移1byte，然后将src1的最低位充入src2右移后多出来的位置int _shr2 (int src1, unsigned src2);SHR2将src2的2个16位有符号数分别右移，右移的位数由src1的低5位决定，多出的位置由符号位扩展unsigned shru2 (unsigned src1, unsigned src2);SHRU2将src2的2个16位无符号数分别右移，右移的位数由src1的低5位决定，多出的位置由0扩展long long _smpy2ll (int src1, int src2);SMPY2将src1和src2中的2个有符号16位数相乘，然后左移1位，再进行饱和。int _spack2 (int src1, int src2);SPACK2将src1和src2中的1个有符号32位数进行饱和到有符号16位，然后把src1的饱和结果放入dst的高16位，src2的饱和结果放入dst的低16位unsigned _spacku4 (int src1 , int src2);SPACKU4将src1和src2中的4个有符号16位数饱和成无符号8位数，int _sshvl (int src2, int src1);SSHVL将src2中的有符号32位数左移或右移，移位的数量由src1指定的比特数确定。src1在-31,31之间，如果src1为正，src2则左移；如果src1为负，src2右移|src1|且符号位扩展int _sshvr (int src2, int src1);SSHVR将src2中的有符号32位数左移或右移，移位的数量由src1指定的比特数确定。src1在-31,31之间，如果src1为正，src2则右移且是符号扩展；如果src1为负，src2左移|src1|int _sub4 (int src1, int src2);SUB4将src1和src2中的4个8位数相减，不进行饱和int _subabs4 (int src1, int src2);SUBABS4将src1和src2中的4个无符号8位相减求绝对值unsigned _swap4 (unsigned src);SWAP4将src的4个8位无符号数按图示换位置unsigned _unpkhu4 (unsigned src);UNPKHU4扩展0unsigned _unpklu4 (unsigned src);UNPKLU4扩0unsigned _xpnd2 (unsigned src);XPND2按src的最低2位进行扩展，bit1扩展高16位，bit0扩展低16位unsigned _xpnd4 (unsigned src);XPND4按src的最低4位进行扩展long long _addsub (int src1, int src2);ADDSUB平行做2步：1、src2+src1-dst_o2、src1-src2-dst_elong long _addsub2 (int src1, int src2);ADDSUB216位有符号ADD2:src2的高、低16位+src1的高、低16位-dst_oSUB2: src1的高、低16位-src2的高、低16位-dst_elong long _cmpy (unsigned src1, unsigned src2);CMPY有符号16位Src1和src2的高16位的点积-src1和src2的低16位点积-dst_o饱和(src1和src2的高16位的点积+src1和src2的低16位点积)-dst_eunsigned _cmpyr (unsigned src1, unsigned src2);CMPYRunsigned _cmpyr1 (unsigned src1, unsigned src2 );CMPYR1long long _ddotp4 (unsigned src1, unsigned src2);DDOTP4没有饱和long long _ddotph2 (long long src1, unsigned src2);DDOTPH2long long _ddotpl2 (long long src1, unsigned src2);DDOTPL2unsigned _ddotph2r (long long src1, unsigned src2);DDOTPH2Runsigned _ddotpl2r (long long src1, unsigned src2);DDOTPL2R long long _dmv (int src1, int src2);DMV将两个寄存器移入一个寄存器一次性的long long _dpack2 (unsigned src1, unsigned src2);DPACK2long long _dpackx2 (unsigned src1, unsigned src2);DPACKX2_float2_t _fmdv_f2(float src1, float src2)DMVunsigned _gmpy (unsigned src1, unsigned src2);GMPY伽罗瓦域上的乘法long long _mpy2ir (int src1, int src2);MPY2IR进行16位乘32位。将src1的高16位和低16位当做有符号16位；将src2的值当做有符号32位。乘积通过加上214round到32位，然后结果右移15位。2个结果的低32位写入dst_o:dst_eint _mpy32 (int src1, int src2);MPY32进行32位乘32位。都是有符号的，64位结果中的低32位写入dstlong long _mpy32ll (int src1, int src2);MPY3232位有符号数32位有符号数，有符号的64位结果被写入dstlong long _mpy32su (int src1, int src2);MPY32SUsrc1有符号32位src2无符号32位=dst有符号64位long long _mpy32us (unsigned src1, int src2);MPY32USsrc1无符号32位src2有符号32位=dst有符号64位long long _mpy32u (unsigned src1, unsigned src2);MPY32Usrc1无符号32位src2无符号32位=dst无符号64位int _rpack2 (int src1, int src2);RPACK2long long _saddsub (unsigned src1, unsigned src2);SADDSUB并行进行：1、饱和（src1+src2）-dst_o2、饱和（src1-src2）-dst_elong long _saddsub2 (unsigned src1, unsigned src2);SADDSUB2并行进行SADD2和SSUB2指令long long _shfl3 (unsigned src1, unsigned src2);SHFL3如图，生成一个longlongint _smpy32 (int src1, int src2);SMPY3232位有符号32位有符号，64位的结果左移1位然后饱和，然后将之后的结果的高32位写入dstint _ssub2 (unsigned src1, unsigned src2);SSUB2Src1中的2个16位有符号-src2中的2个有符号16位，结果进行饱和unsigned _xormpy (unsigned src1, unsigned src2);XORMPY加瓦罗域乘法int _dpint (double src);DPINT将double转成int（round）_int40_t _f2tol(_float2_t src);将一个_float2_t解释成一个_int40_float2_t _f2toll(_float2_t src);将一个_float2_t解释成一个longlongdouble _fabs (double src);ABSDP将src的绝对值放入dst。float _fabsf (float src);ABSSP_float2_t _lltof2(long long src);将一个longlong解释成一个_float2_t_float2_t _ltof2(_int40_t src);将一个_int40解释成一个_float2_t_float2_t & _mem8_f2(void * ptr);LDNDWSTNDW从内存里加载一个64位值const _float2_t & _mem8_f2_const(void * ptr);LDNDWSTNDWlong long _mpyidll (int src1, int src2);MPYIDSrc1src2-dstdouble_mpysp2dp (float src1, float src2);MPYSP2DPSrc1src2-dstdouble_mpyspdp (float src1, double src2);MPYSPDPSrc1src2-dstdouble _rcpdp (double src);RCPDP64位double倒数近似值放入dstfloat _rcpsp (float src);RCPSP32位float的倒数近似值double _rsqrdp (double src);RSQRDP64位double的平方根倒数近似值float _rsqrsp (float src);RSQRSP32位float的平方根倒数近似值int _spint (float);SPINTFloat转为intADDDP2个double相加ADDSP2个float相加AND位与ANDN与后取反MPYSP2个float相乘OR位或SUBDP2个double相减SUBSP2和float相减XOR异或_x128_t _ccmatmpy (long long src1, _x128_t src2);CMATMPYlong long _ccmatmpyr1 (long long src1, _x128_t src2);CCMATMPYR1long long _ccmpy32r1 (long long src1, long long src2);CCMPY32R1_x128_t _cmatmpy (long long src1, _x128_t src2);CMATMPYlong long _cmatmpyr1 (long long src1, _x128_t src2);CMATMPYR1long long _cmpy32r1 (long long src1, long long src2);CMPY32R1_x128_t _cmpysp (_float2_t src1, _float2_t src2);CMPYSPdouble _complex_conjugate_mpysp (double src1, double src2);CMPYSPDSUBSPdouble _complex_mpysp (double src1, double src2);CMPYSPDADDSPint _crot90 (int src);CROT90复数的90度旋转int _crot270 (int src);CROT270复数的270度旋转long long _dadd (long long src1, long long src2);DADDSrc1的2个32位有符号数+src2的2个32位有符号数long long _dadd2 (long long src1, long long src2);DADD24路有符号16位相加_float2_t _daddsp (_float2_t src1, _float2_t src2);DADDSPlong long _dadd_c (scst5 immediate src1, long long src2);DADD2路float加法long long _dapys2 (long long src1, long long src2);DAPYS2long long _davg2 (long long src1, long long src2);DAVG2有符号16位long long _davgnr2 (long long src1, long long src2);DAVGNR2有符号16位，无round模式long long _davgnru4 (long long src1, long long src2);DAVGNRU4无符号8位，无round模式long long _davgu4 (long long src1, long long src2);DAVGU4无符号8位long long _dccmpyr1 (long long src1, long long src2);DCCMPYR1unsigned _dcmpeq2 (long long src1, long long src2