TI DSP TMS320C66x学习笔记之内联指令（c6x.h中文注释）(六)

/*****************************************************************************/
/*  C6X.H v7.4.12                                                            */
/*****************************************************************************/

#include "vect.h"


  typedef double   __float2_t;//__float2_t双精度浮点型

  #define _lltof2  _lltod
  #define _f2toll  _dtoll    //将一个__float2_t解释成一个long long
  #define _ftof2   _ftod
  #define _hif2    _hif
  #define _lof2    _lof
  #define _f2tol   _dtol     //将一个__float2_t解释成一个__int40
  #define _ltof2   _ltod

  #define _amem8_f2          _amemd8       //加载和存储8bytes，指针必须8byte对齐，必须包含c6x.h
  #define _amem8_f2_const    _amemd8_const //加载8bytes，指针必须8byte对齐，必须包含c6x.h
  #define _mem8_f2           _memd8        //从内存里加载一个64位值
  #define _mem8_f2_const     _memd8_const
  #define _fdmv_f2           _fdmv
  #define _hif2_128          _hid128
  #define _lof2_128          _lod128
  #define _f2to128           _dto128
  #define _fdmvd_f2          _fdmvd





/*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       _ext	   (int src2,      unsigned csta, unsigned cstb);
int       _extr	   (int src2,      int src1);//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);//同上上，区别最后是0扩展到32位
unsigned  _extur   (unsigned src2, int src1);//同上，区别左右移的位数由src1的低10位指定

/*通过cstb和csta指定需要置1的首位和尾位*/
unsigned  _set	   (unsigned src, unsigned csta, unsigned cstb);
unsigned  _setr	   (unsigned src2, int src1);//将src2中指定位置1，置1的首位和末位有src1的低10位指定

/*通过cstb和csta指定需要清0的首位和尾位*/
unsigned  _clr	   (unsigned src, unsigned csta, unsigned cstb);
unsigned  _clrr	   (unsigned src2, int src1);//将src2中指定位清0，清0的首位和末位有src1的低10位指定

/*将src1和src2相加，且饱和其结果*/
int       _sadd	   (int,      int);
__int40_t _lsadd   (int, __int40_t);
/*从src1中减去src2并饱和结果*/
int	      _ssub	   (int src1,      int src2);
__int40_t _lssub   (int, __int40_t);

/*将一个40bit的long转换成一个32bit的有符号int，如有需要，对结果进行饱和*/
int       _sat	   (__int40_t);

/*将src2左移src1位，结果饱和在32位*/
int       _sshl	   (int src2,      unsigned src1);
/*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*/
int	      _add2	   (int,      int);
int	      _sub2	   (int,      int);/*把src1的高、低16位减去src2的高、低16位，放入结果的高、低16位*/

/*有条件减和左移（常用于除法）*/
unsigned  _subc	   (unsigned, unsigned);
/*搜索src2里面的1或0,1或0是由src1的LSB决定的，返回比特位变换的位数*/
unsigned  _lmbd	   (unsigned src1, unsigned src2);
/*返回src的绝对值*/
int       _abs	   (int src);
__int40_t _labs	   (__int40_t src);

/*返回src的冗余的符号比特位的个数，bit31是符号位，例如bit31往低位走，01b返回0,001b返回1,0001b返回2,00001b返回3*/
unsigned  _norm	   (int);
unsigned  _lnorm   (__int40_t);


//16 LSBs * 16 LSBs
int	      _mpy	   (int src1,      int src2);/*src1和src2相乘，操作数默认为有符号*/
int	      _mpyus   (unsigned src1, int src2);/*无符号src1和有符号src2相乘，S是用来确定哪个是有符号数（S）哪个是无符号数（U）*/
int	      _mpysu   (int,      unsigned);
unsigned  _mpyu	   (unsigned, unsigned);
//16 MSBs * 16 MSBs
int	      _mpyh	   (int,      int);
int	      _mpyhus  (unsigned, int);
int	      _mpyhsu  (int,      unsigned);
unsigned  _mpyhu   (unsigned, unsigned);
//16 MSBs * 16 LSBs
int	      _mpyhl   (int,      int);
int	      _mpyhuls (unsigned, int);
int	      _mpyhslu (int,      unsigned);
//16 LSBs * 16 MSBs
int	      _mpylh   (int,      int);
int	      _mpyluhs (unsigned, int);
int	      _mpylshu (int,      unsigned);
unsigned  _mpylhu  (unsigned, unsigned);
//与上面的乘法指令区别在何处
/*把src1的低16位和src2的低16位相乘,之后左移一位*/
int	      _smpy	   (int src1,      int src2);
int	      _smpyhl  (int,      int);//高16位 * 低16位
int	      _smpylh  (int,      int);//低16位 * 高16位
int	      _smpyh   (int,      int);//把src1的高16位和src2的高16位相乘
/*将src1和src2中的2对16位有符号数相乘，然后左移1位，再进行饱和*/
long long _smpy2ll   (int,      int);
/*32位有符号数乘以32位有符号数，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dst*/
int       _smpy32    (int,       int);


/*返回double寄存器的高32位（奇数位寄存器）作为int型*/
unsigned  _hi(double);
/*返回double寄存器的高32位（奇数位寄存器）作为float型*/
float     _hif(double);
/*返回long long寄存器的高32位（奇数位寄存器）作为int型*/
unsigned  _hill(long long);
/*返回double寄存器的低32位（偶数位寄存器）作为int型*/
unsigned  _lo(double);
/*返回double寄存器的低32位（偶数位寄存器）作为float型*/
float     _lof(double);
/*返回long long寄存器的低32位（偶数位寄存器）作为int型*/
unsigned  _loll(long long);

/*创建一个新的double寄存器来保存2个unsigned int的值，其中src2是高（奇数）寄存器，src1是低（偶数）寄存器*/
double 	  _itod(unsigned, unsigned);
/*创建一个新的double寄存器来保存2个float的值，其中src2是高（奇数）寄存器，src1是低（偶数）寄存器*/
double 	  _ftod(float,    float);
/*创建一个新的long long寄存器来保存2个unsigned int的值，其中src2是高（奇数）寄存器，src1是低（偶数）寄存器*/
long long _itoll(unsigned src2, unsigned src1);

/*注意不是变换，是直接将寄存器中的值解释成整型或者浮点型*/
float  	  _itof(unsigned);      /* 将一个int寄存器重新解释成一个float型      */
unsigned  _ftoi(float);         /* 将float的比特位解释成unsigned int         */
__int40_t _dtol(double);        /* 将一个double寄存器重新解释成一个_int40_t型*/
double    _ltod(__int40_t);     /* 将一个__int40_t寄存器重新解释成一个double型*/
long long _dtoll(double);       /* 将一个double寄存器重新解释成一个long long型*/
double    _lltod(long long);    /* 将一个long long寄存器重新解释成double型    */


  /* Define pseudo intrinsics for some pseudo instructions */
  #ifndef _cmplt2
  #define _cmplt2(src1, src2) _cmpgt2((src2), (src1))
  #endif
  #ifndef _cmpltu4
  #define _cmpltu4(src1, src2) _cmpgtu4((src2), (src1))
  #endif
  #ifndef _dotpnrus2
  #define _dotpnrus2(src1, src2) _dotpnrsu2((src2), (src1))
  #endif
  #ifndef _dotpus4
  #define _dotpus4(src1, src2) _dotpsu4((src2), (src1))
  #endif
  #ifndef _mpyihll
  #define _mpyihll(src1, src2) _mpyhill((src2), (src1))
  #endif
  #ifndef _mpyihr
  #define _mpyihr(src1, src2) _mpyhir((src2), (src1))
  #endif
  #ifndef _mpyilll
  #define _mpyilll(src1, src2) _mpylill((src2), (src1))
  #endif
  #ifndef _mpyilr
  #define _mpyilr(src1, src2) _mpylir((src2), (src1))
  #endif
  #ifndef _mpyus4ll
  #define _mpyus4ll(src1, src2) _mpysu4ll((src2), (src1))
  #endif
  #ifndef _saddsu2
  #define _saddsu2(src1, src2) _saddus2((src2), (src1))
  #endif
  #ifndef _swap2
  #define _swap2(src) _packlh2((src), (src))
  #endif
/*把src1和src2的4对8bits相加，不会进行饱和，进位不会影响其他的8位数*/
int       _add4      (int,      int);
/*计算2对有符号16位数的平均值*/
int       _avg2      (int,      int);
/*计算4对有符号8位数的平均值*/
unsigned  _avgu4     (unsigned, unsigned);
/*比较2对16位数的值是否相等，结果放入dst的最低2位，相等返回1*/
int       _cmpeq2    (int,      int);
/*比较4对8位数的值是否相等，结果放入dst的最低4位，相等置1，否则为0*/
int       _cmpeq4    (int,      int);
/*比较2对有符号16位数，src1 > src2，置为1；否则为0.结果放入dst的最低2位*/
int       _cmpgt2    (int src1,      int src2);
/*比较4对无符号8位数，src1 > src2，置为1；否则为0.结果放入dst的最低4位*/
unsigned  _cmpgtu4   (unsigned src1, unsigned src2);

/*将src1和src2中的2对16位有符号数进行点积（相加），结果被写成有符号32位int或者符号扩展为64位*/
int       _dotp2     (int ,      int);
__int40_t _ldotp2    (int,      int);
/*将src1和src2中的16位有符号数进行点积相减*/
int       _dotpn2    (int,      int);
/*src1和src2中的高16位的点积“减去”低16位的点积。src1中的数被当做有符号数，src2中的数被当做无符号数，再加上2^15，结果再带符号右移16位*/
int       _dotpnrsu2 (int src1,      unsigned src2);
/*src1和src2中的高16位的点积“加上”低16位的点积。src1中的数被当做有符号数，src2中的数被当做无符号数，再加上2^15，结果再带符号右移16位*/
int       _dotprsu2  (int,      unsigned);
/*将src1和src2的4对8位数进行相乘再求和，src1的每8位数被当做有符号数，src2的每8位数被当做无符号数*/
int       _dotpsu4   (int,      unsigned);
unsigned  _dotpu4    (unsigned, unsigned);//同上，都被当做无符号数

/*将src1和src2的4对8位无符号数进行迦罗瓦域的乘法*/
int       _gmpy4     (int,      int);

/*将src1和src2中的2对16位有符号数比较，取较大值*/
int       _max2      (int,      int);
/*将src1和src2中的4对8位无符号数比较，取较大值*/
unsigned  _maxu4     (unsigned, unsigned);
/*将src1和src2中的2对16位有符号数比较，取较小值*/
int       _min2      (int,      int);
/*将src1和src2中的4对8位无符号数比较，取较小值*/
unsigned  _minu4     (unsigned, unsigned);


/*将src1和src2中的2对16位有符号数分别相乘，将2个32位的结果写入long long中*/
long long _mpy2ll    (int,      int);
/*将src1中高16位作为1个16位有符号数乘以src2的32位有符号数，结果写入long long的低48位*/
long long _mpyhill   (int src1,      int src2);
/*将src1中低16位作为1个16位有符号数乘以src2的32位有符号数，结果写入long long的低48位*/
long long _mpylill   (int,      int);
/*将src1的高16位作为1个16位有符号数乘以src2的32位有符号数。乘积利用round模式通过加2^14转成32位，最后再右移15位*/
int       _mpyhir    (int,      int);
/*将src1的低16位作为1个16位有符号数乘以src2的32位有符号数。乘积利用round模式通过加2^14转成32位，最后再右移15位*/
int       _mpylir    (int,      int);
/*将src1的4个8位有符号乘src2的4个8位无符号数，得到4个16位有符号数，组成一个64位*/
long long _mpysu4ll  (int src1,      unsigned src2);
long long _mpyu4ll   (unsigned, unsigned);//同时，都是无符号数

/*将src1的低16位放入返回值的高16位，src2的低16位放入返回值的低16位*/
unsigned  _pack2     (unsigned src1, unsigned src2);
/*将src1的高16位放入返回值的高16位，src2的高16位放入返回值的低16位*/
unsigned  _packh2    (unsigned, unsigned);
/*将src1（a_3|a_2|a_1|a_0）和src2（b_3|b_2|b_1|b_0）分成4对8位无符号数，返回a_3|a_1|b_3|b_1*/
unsigned  _packh4    (unsigned, unsigned);
/*将src1（a_3|a_2|a_1|a_0）和src2（b_3|b_2|b_1|b_0）分成4对8位无符号数，返回a_2|a_0|b_2|b_0*/
unsigned  _packl4    (unsigned, unsigned);
/*将src1(a_hi|a_lo)和src2(b_hi|b_lo)分成2对16位无符号数，返回a_hi|b_lo*/
unsigned  _packhl2   (unsigned, unsigned);
/*将src1(a_hi|a_lo)和src2(b_hi|b_lo)分成2对16位无符号数，返回a_lo|b_hi*/
unsigned  _packlh2   (unsigned, unsigned);
/*按照src1的最低5位的数去左移src2的32位，src1中剩下的高5-31位被忽略*/
unsigned  _rotl      (unsigned src1, unsigned src2);
/*将src1和src2中的2个16位有符号数相加，生成2个16位有符号数并饱和*/
int       _sadd2     (int,      int);
/*将src1和src2中的4对8位无符号数相加并饱和*/
unsigned  _saddu4    (unsigned, unsigned);
/*将src1中的2个16位无符号数和src2中的2个16位有符号数相加，得到2个16位无符号数并饱和*/
int       _saddus2   (unsigned src1, int src2);

/*将src2左移1byte，然后将src1的最高1byte填充入src2左移后多出的位置*/
unsigned  _shlmb     (unsigned src1, unsigned src2);
/*将src2右移1byte，然后将src1的最高1byte填充入src2右移后多出的位置*/
unsigned  _shrmb     (unsigned src1, unsigned src2);
/*将src2的2个16位有符号数分别右移，右移位数由src1的低5位决定，多出的位置由符号位扩展*/
int       _shr2      (int src1,      unsigned src2);
/*将src2的2个16位有符号数分别右移，右移位数由src1的低5位决定，多出的位置由0扩展*/
unsigned  _shru2     (unsigned src1, unsigned src2);
/*将src1和src2中的1个32位有符号进行饱和到有符号16位，然后把src1的高16位，src2的饱和结果放入dst的低16位，不懂就看指令手册*/
int       _spack2    (int src1,      int src2);
/*将src1和src2中的2对16位有符号数饱和成8位无符号数*/
unsigned  _spacku4   (int,      int);
/* *
 * 将src2中的32位有符号数左移或右移，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       _sshvl     (int src2,      int src1);
/* *
 * 将src2中的32位有符号数左移或右移，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则右移且是符号位扩展；如果src1为负，src2左移|src1|
 * */
int       _sshvr     (int src2,      int src1);
/*将src1和src2中的4对8位数相减，不进行饱和*/
int       _sub4      (int src1,      int src2);
/*将src1和src2中的4对8位无符号数相减求绝对值*/
int       _subabs4   (int src1,      int src2);

/*分别计算高、低16位的绝对值*/
int       _abs2      (int);
/*统计4对8位比特数是1的个数，写入结果对应位置*/
unsigned  _bitc4     (unsigned);
/*颠倒比特位的顺序*/
unsigned  _bitr      (unsigned);
/*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*/
unsigned  _deal      (unsigned);
/*将src的数据移入返回值中，利用了乘法流水线（延迟4cycle）*/
int       _mvd       (int src2);
/*将src2的高16和低16位进行交织abcdefghijklmnop|ABCDEFGHIJKLMNOP，返回aAbBcCdDeEfFgGhH|iIjJkKlLmMnNoOpP*/
unsigned  _shfl      (unsigned src2);
/*将src4个8位无符号数(ub_3|ub_2|ub_1|ub_0|)交换位置,返回(ub_2|ub_3|ub_0|ub_1|)*/
unsigned  _swap4     (unsigned);
/*扩展0(高16位2个8位数)，(ub_3|ub_2|ub_1|ub_0|)-->(0|ub_3|0|ub_2|),每个分隔号代表1byte*/
unsigned  _unpkhu4   (unsigned);
/*扩展0(低16位2个8位数)，(ub_3|ub_2|ub_1|ub_0|)-->(0|ub_1|0|ub_0|),每个分隔号代表1byte*/
unsigned  _unpklu4   (unsigned);
/*按src的最低2位进行扩展，bit1扩展高16位，bit0扩展低16位*/
unsigned  _xpnd2     (unsigned);
unsigned  _xpnd4     (unsigned);//同上，按src的最低4位进行扩展，例bit0扩展低8位。

/*平行做2步，1.src1+src2->dst_o 2.src1-src2->dst_e，dst_o代表奇数寄存器，dst_e代表偶数寄存器*/
long long _addsub    (int src1,       int src2);
/*同上，分成2对16位有符号数进行ADD2和SUB2*/
long long _addsub2   (unsigned,  unsigned);
/* *
 * 饱和(src1低16位和src2高16位的点积加上src1高16位和src2低16位点积)赋给dst_e
 * sat((lsb16(src1) × msb16(src2)) + (msb16(src1) × lsb16(src2))) → dst_e
 * 有符号16位src1和src2的高16位的点积减去src1和src2低16位的点积赋给dst_o
 * (msb16(src1) × msb16(src2)) - (lsb16(src1) × lsb16(src2)) → dst_o
 * */
long long _cmpy      (unsigned src1,  unsigned src2);
/* *
 * sat((lsb16(src1) × msb16(src2)) + (msb16(src1) × lsb16(src2))) → tmp_e
 * msb16(sat(tmp_e + 00008000h)) → lsb16(dst)
 * sat((msb16(src1) × msb16(src2)) - (lsb16(src1) × lsb16(src2))) → tmp_o
 * msb16(sat(tmp_o + 00008000h)) → msb16(dst)
 * */
unsigned  _cmpyr     (unsigned,  unsigned);
/* *
 * sat((lsb16(src1) × msb16(src2)) + (msb16(src1) × lsb16(src2))) → tmp_e
 * msb16(sat((tmp_e + 00004000h) << 1)) → lsb16(dst)
 * sat((msb16(src1) × msb16(src2)) - (lsb16(src1) × lsb16(src2))) → tmp_o
 * msb16(sat((tmp_e + 00004000h) << 1)) → msb16(dst)
 * */
unsigned  _cmpyr1    (unsigned,  unsigned);
/*看TMS320C6000内联指令汇编的35页图示，完成点积和功能*/
long long _ddotph2   (long long, unsigned);
unsigned  _ddotph2r  (long long, unsigned);
long long _ddotpl2   (long long, unsigned);
unsigned  _ddotpl2r  (long long, unsigned);
long long _ddotp4    (unsigned src1,  unsigned src2);
/*数据打包看37页图示*/
long long _dpack2    (unsigned src1,  unsigned src2);
long long _dpackx2   (unsigned,  unsigned);
/*将两个寄存器移入一个寄存器一次性的*/
long long _dmv       (unsigned,  unsigned);
double    _fdmv      (float,     float);
/*迦罗瓦域上的乘法*/
unsigned  _gmpy      (unsigned,  unsigned);
/*进行32位乘32位。都是有符号数，64位结果都被写入dst*/
long long _mpy32ll   (int,       int);
/*进行32位乘32位。都是有符号数，64位结果中的低32位写入dst*/
int       _mpy32     (int,       int);
/*src1有符号32位乘以src2无符号32位=有符号64位*/
long long _mpy32su   (int,       unsigned);
long long _mpy32us   (unsigned,  int); //同上，无符号乘以有符号
long long _mpy32u    (unsigned,  unsigned);//同上上，都为无符号
/* *
 * 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2^14 round到32位，然后结果右移15位。2个结果的低32位写入dst_o:dst_e
 * */
long long _mpy2ir    (unsigned src1,  int src2);
/*src1和src2的高16位，分别右移1位后饱和，放入32位结果中*/
unsigned  _rpack2    (unsigned src1,  unsigned src2);
/*并行进行1.饱和(src1+src2)->dst_o  2.饱和(src1-src2)->dst_e*/
long long _saddsub   (int,       int);
long long _saddsub2  (unsigned,  unsigned);//并行进行SADD2和SSUB2指令
/*不知道干嘛用的，看39页*/
long long _shfl3     (unsigned,  unsigned);
/*src1中的2个16位有符号数减去src2中的2个16位有符号数*/
int       _ssub2     (int src1,       int src2);
/*迦罗瓦乘法*/
unsigned  _xormpy    (unsigned,  unsigned);


long long  _dcmpyr1    (long long, long long);
long long  _dccmpyr1   (long long, long long);
long long  _cmpy32r1   (long long, long long);
long long  _ccmpy32r1  (long long, long long);
long long  _mpyu2      (unsigned,  unsigned);
/*4对8位有符号数点积相加*/
int        _dotp4h     (long long, long long);
long long  _dotp4hll   (long long, long long);
/*src1中的4对16位有符号数，src2中的4对16位无符号数，得到一个32位的点积和*/
int        _dotpsu4h   (long long, long long);
/*src1中的4对16位有符号数，src2中的4对16位无符号数，得到一个64位的点积和*/
long long  _dotpsu4hll (long long, long long);
/*src1的2个32位有符号数加上src2的2个32位有符号数*/
long long  _dadd       (long long src1, long long src2);
long long  _dadd_c     (int,       long long);

long long  _dsadd      (long long, long long);
/*4对16位有符号数相加*/
long long  _dadd2      (long long, long long);
long long  _dsadd2     (long long, long long);
long long  _dsub       (long long, long long);
long long  _dssub      (long long, long long);
long long  _dssub2     (long long, long long);

long long  _dapys2     (long long, long long);
long long  _dshr       (long long, unsigned);
long long  _dshru      (long long, unsigned);
long long  _dshl       (long long, unsigned);
long long  _dshr2      (long long, unsigned);
long long  _dshru2     (long long, unsigned);
unsigned   _shl2       (unsigned , unsigned);
long long  _dshl2      (long long, unsigned);
long long  _dxpnd4     (unsigned);
long long  _dxpnd2     (unsigned);
int        _crot90     (int);
long long  _dcrot90    (long long);
int        _crot270    (int);
long long  _dcrot270   (long long);
/*src1和src2中的4对16位有符号数比较，取大者放入dst*/
long long  _dmax2      (long long, long long);
long long  _dmin2      (long long, long long);
/*src1和src2中的8对8位有符号数比较，取大者放入dst*/
long long  _dmaxu4     (long long, long long);
long long  _dminu4     (long long, long long);

/*4对16位比较，相等返回1，不等返回0*/
unsigned   _dcmpeq2    (long long, long long);
/*8对8位比较，相等返回1，不等返回0*/
unsigned   _dcmpeq4    (long long, long long);
/*4对16位比较，大于返回1，不大于返回0*/
unsigned   _dcmpgt2    (long long, long long);
/*8对8位比较，大于返回1，不大于返回0*/
unsigned   _dcmpgtu4   (long long, long long);

/*4对16位有符号数求4个平均"(a+b+1)/2"*/
long long  _davg2      (long long, long long);
/*8对8位无符号数求8个平均*/
long long  _davgu4     (long long, long long);
/*有符号16位，无round模式，4个平均"(a+b)/2"*/
long long  _davgnr2    (long long, long long);
/*无符号8位，无round模式，8个平均*/
long long  _davgnru4   (long long, long long);
long long  _unpkbu4    (unsigned);
long long  _unpkh2     (unsigned);
long long  _unpkhu2    (unsigned);

/*并行执行2个PACKL2*/
long long  _dpackl2    (long long, long long);
/*并行执行2个PACKH2*/
long long  _dpackh2    (long long, long long);
long long  _dpackhl2   (long long, long long);
/*并行执行PACKH4和PACKL4*/
long long  _dpacklh4   (unsigned,  unsigned);
long long  _dpackl4    (long long, long long);
long long  _dpackh4    (long long, long long);
long long  _dspacku4   (long long, long long);

void       _mfence     ();
__float2_t _dmpysp     (__float2_t, __float2_t);
/*2路float型相加*/
__float2_t _daddsp     (__float2_t, __float2_t);
__float2_t _dsubsp     (__float2_t, __float2_t);
/*src中的16位有符号数转换成单精度浮点放入dst_e和dst_o中*/
__float2_t _dinthsp    (unsigned src);
/*src中的16位无符号数转换成单精度浮点放入dst_e和dst_o中*/
__float2_t _dinthspu   (unsigned);
/*src中的32位有符号转换成单精度浮点，放入dst_e和dst_o中*/
__float2_t _dintsp     (long long);
/*src中的32位无符号转换成单精度浮点，放入dst_e和dst_o中*/
__float2_t _dintspu    (long long);
unsigned   _dspinth    (__float2_t);
long long  _dspint     (__float2_t);

int        _land       (int, int);
int        _landn      (int, int);
int        _lor        (int, int);
/*将2个寄存器移入1个寄存器中。一次进行2次移动，当处理很多double word时很有用。减去寄存器压力*/
long long  _dmvd       (int,       int);
double     _fdmvd      (float,     float);

double     _complex_mpysp           (double, double); /* CMPYSP then DADDSP */
double     _complex_conjugate_mpysp (double, double); /* CMPYSP then DSUBSP */

long long  _xorll_c    (int, long long);

__x128_t   __BUILTIN _dcmpy      (long long, long long);
__x128_t   __BUILTIN _dccmpy     (long long, long long);
long long  __BUILTIN _cmatmpyr1  (long long, __x128_t);
long long  __BUILTIN _ccmatmpyr1 (long long, __x128_t);
__x128_t   __BUILTIN _cmatmpy    (long long, __x128_t);
__x128_t   __BUILTIN _ccmatmpy   (long long, __x128_t);
__x128_t   __BUILTIN _qsmpy32r1  (__x128_t,  __x128_t);
__x128_t   __BUILTIN _qmpy32     (__x128_t,  __x128_t);
__x128_t   __BUILTIN _dsmpy2     (long long, long long);
/*4对16位有符号数相乘，得到32位有符号数放入128位寄存器中*/
__x128_t   __BUILTIN _dmpy2      (long long, long long);
/*4对16位有符号数相乘，得到32位有符号数放入128位寄存器中*/
__x128_t   __BUILTIN _dmpyu2     (long long, long long);
/*将src1中8个8位有符号数乘以src2中8个8位无符号数，得到8个16位有符号数*/
__x128_t   __BUILTIN _dmpysu4    (long long src1, long long src2);
__x128_t   __BUILTIN _dmpyu4     (long long, long long);//同上，都是无符号
__x128_t   __BUILTIN _cmpysp     (__float2_t, __float2_t);
__x128_t   __BUILTIN _qmpysp     (__x128_t,  __x128_t);
/*执行2个dotp4h，都是有符号的*/
long long  __BUILTIN _ddotp4h    (__x128_t,  __x128_t);
/*执行2个dotpsu4h，一个有符号，一个无符号*/
long long  __BUILTIN _ddotpsu4h  (__x128_t,  __x128_t);

__x128_t   __BUILTIN _ito128  (unsigned,  unsigned, unsigned, unsigned);
__x128_t   __BUILTIN _fto128  (float,     float,    float,    float);
__x128_t   __BUILTIN _llto128 (long long, long long);
__x128_t   __BUILTIN _dto128  (double,    double);

long long  __BUILTIN _hi128   (__x128_t);
double     __BUILTIN _hid128  (__x128_t);
long long  __BUILTIN _lo128   (__x128_t);
double     __BUILTIN _lod128  (__x128_t);

unsigned  __BUILTIN _get32_128  (__x128_t, __CONST(0,3) unsigned);
float     __BUILTIN _get32f_128 (__x128_t, __CONST(0,3) unsigned);

__x128_t  __BUILTIN _dup32_128 (unsigned);



extern __cregister volatile unsigned int AMR;
extern __cregister volatile unsigned int CSR;
extern __cregister volatile unsigned int IFR;
extern __cregister volatile unsigned int ISR;
extern __cregister volatile unsigned int ICR;
extern __cregister volatile unsigned int IER;
extern __cregister volatile unsigned int ISTP;
extern __cregister volatile unsigned int IRP;
extern __cregister volatile unsigned int NRP;

extern __cregister volatile unsigned int GFPGFR;
extern __cregister volatile unsigned int DIER;

extern __cregister volatile unsigned int REP;
extern __cregister volatile unsigned int TSCL;
extern __cregister volatile unsigned int TSCH;
extern __cregister volatile unsigned int ARP;
extern __cregister volatile unsigned int ILC;
extern __cregister volatile unsigned int RILC;
extern __cregister volatile unsigned int PCE1;
extern __cregister volatile unsigned int DNUM;
extern __cregister volatile unsigned int SSR;
extern __cregister volatile unsigned int GPLYA;
extern __cregister volatile unsigned int GPLYB;
extern __cregister volatile unsigned int TSR;
extern __cregister volatile unsigned int ITSR;
extern __cregister volatile unsigned int NTSR;
extern __cregister volatile unsigned int ECR;
extern __cregister volatile unsigned int EFR;
extern __cregister volatile unsigned int IERR;

extern __cregister volatile unsigned int DMSG;
extern __cregister volatile unsigned int CMSG;
extern __cregister volatile unsigned int DT_DMA_ADDR;
extern __cregister volatile unsigned int DT_DMA_DATA;
extern __cregister volatile unsigned int DT_DMA_CNTL;
extern __cregister volatile unsigned int TCU_CNTL;
extern __cregister volatile unsigned int RTDX_REC_CNTL;
extern __cregister volatile unsigned int RTDX_XMT_CNTL;
extern __cregister volatile unsigned int RTDX_CFG;
extern __cregister volatile unsigned int RTDX_RDATA;
extern __cregister volatile unsigned int RTDX_WDATA;
extern __cregister volatile unsigned int RTDX_RADDR;
extern __cregister volatile unsigned int RTDX_WADDR;
extern __cregister volatile unsigned int MFREG0;
extern __cregister volatile unsigned int DBG_STAT;
extern __cregister volatile unsigned int BRK_EN;
extern __cregister volatile unsigned int HWBP0_CNT;
extern __cregister volatile unsigned int HWBP0;
extern __cregister volatile unsigned int HWBP1;
extern __cregister volatile unsigned int HWBP2;
extern __cregister volatile unsigned int HWBP3;
extern __cregister volatile unsigned int OVERLAY;
extern __cregister volatile unsigned int PC_PROF;
extern __cregister volatile unsigned int ATSR;
extern __cregister volatile unsigned int TRR;
extern __cregister volatile unsigned int TCRR;