/* This file contains the time crucial parts of the code, written out in
   an 'assembler-like' i.e. unreadable style
*/

/* This limit is a tuning parameter, telling when router is faster than xnet */
#define xnetRouterLimit 32

#define bitswapXx(dist) { \
   if (ixproc & dist) { \
      tmp = xnetW[dist].q1r; /* 1..0 -> tmp  */ \
      xnetW[dist].q1r = q0r; /* 0..1 -> 1..0 */ \
      q0r = tmp;              /* tmp  -> 0..1 */ \
      tmp = xnetW[dist].q1i; \
      xnetW[dist].q1i = q0i; \
      q0i = tmp; \
   } \
}

#define bitswapXy(dist) { \
   if (iyproc & dist) { \
      tmp = xnetN[dist].q1r; /* 1..0 -> tmp  */ \
      xnetN[dist].q1r = q0r; /* 0..1 -> 1..0 */ \
      q0r = tmp;              /* tmp  -> 0..1 */ \
      tmp = xnetN[dist].q1i; \
      xnetN[dist].q1i = q0i; \
      q0i = tmp; \
   } \
}

#if (float_type == 1)

#define bitswDef \
   register plural int src; \
   plural long long lltmp; \
   plural float *ltmp; \
   ltmp = (plural float *) &lltmp; 

#define bitswapRx(dist) { \
   if (ixproc & dist) { \
      ltmp[0] = q0r; ltmp[1] = q0i; \
   } \
   else { \
      ltmp[0] = q1r; ltmp[1] = q1i; \
   } \
   src = iproc ^ dist; \
   router[src].lltmp = lltmp; \
   if (ixproc & dist) { \
      q0r = ltmp[0]; q0i = ltmp[1]; \
   } \
   else { \
      q1r = ltmp[0]; q1i = ltmp[1]; \
   } \
}

#define bitswapRy(dist) { \
   if (iyproc & dist) { \
      ltmp[0] = q0r; ltmp[1] = q0i; \
   } \
   else { \
      ltmp[0] = q1r; ltmp[1] = q1i; \
   } \
   src = iproc ^ (dist<<lnxproc); \
   router[src].lltmp = lltmp; \
   if (iyproc & dist) { \
      q0r = ltmp[0]; q0i = ltmp[1]; \
   } \
   else { \
      q1r = ltmp[0]; q1i = ltmp[1]; \
   } \
}

#else

#define bitswDef \
register plural int src; \
register plural double ltmp;

#define bitswapRx(dist) { \
   src = iproc ^ dist; \
   if (ixproc & dist) { \
      ltmp = q0r; \
   } \
   else { \
      ltmp = q1r; \
   } \
   router[src].ltmp = ltmp; \
   if (ixproc & dist) { \
      q0r = ltmp; \
   } \
   else { \
      q1r = ltmp; \
   } \
   if (ixproc & dist) { \
      ltmp = q0i; \
   } \
   else { \
      ltmp = q1i; \
   } \
   router[src].ltmp = ltmp; \
   if (ixproc & dist) { \
      q0i = ltmp; \
   } \
   else { \
      q1i = ltmp; \
   } \
}

#define bitswapRy(dist) { \
   src = iproc ^ (dist<<lnxproc); \
   if (iyproc & dist) { \
      ltmp = q0r; \
   } \
   else { \
      ltmp = q1r; \
   } \
   router[src].ltmp = ltmp; \
   if (iyproc & dist) { \
      q0r = ltmp; \
   } \
   else { \
      q1r = ltmp; \
   } \
   if (iyproc & dist) { \
      ltmp = q0i; \
   } \
   else { \
      ltmp = q1i; \
   } \
   router[src].ltmp = ltmp; \
   if (iyproc & dist) { \
      q0i = ltmp; \
   } \
   else { \
      q1i = ltmp; \
   } \
}

#endif
 
#define bitswapx(dist) \
if ((dist) < xnetRouterLimit) { \
   bitswapXx(dist) \
} \
else { \
   bitswapRx(dist) \
}
  
#define bitswapy(dist) \
if ((dist) < xnetRouterLimit) { \
   bitswapXy(dist) \
} \
else { \
   bitswapRy(dist) \
}

 
/* define a macro to compute the twiddle list symbolically */
#define smb_bfl \
   numtw = actgrp & curadr; \
   dentw = mskcb; \
   /* divide out common factors in numer. and denom. */ \
   while (((numtw&1)==0) && ((dentw&1)==0)) {numtw>>=1;dentw>>=1;} \
   /* search twiddle lists for these numbers */ \
   twfnd = twlsln; \
   for(ip=0;ip<twlsln;ip++) \
      if ((itwls[2*ip] == numtw) && (itwls[2*ip+1] == dentw)) { \
         twfnd = ip; \
         break; \
      } \
   /* if not found, add to end of list */ \
   if (twfnd == twlsln) { \
      twlsln++; \
      P_DYNARR(itwls,2*twfnd+1,siztwl,dentw,plural unsigned); \
      itwls[2*twfnd] = numtw; \
   } \
   /* update pointer list to twiddle factors */ \
   twdpt[itwp] = twfnd; \
   itwp++;
 
 
/* for easier optimization we define the crucial parts of the code in the
   preprocessor (inline expansion instead of funciton call): */
#define butterfly_code \
   { \
      twdad = twdpt[itwp++]; /* address of twiddle factor */ \
      { \
         tmp = twdfac[twdad][0]; \
         tmp0 = -q1r; \
         tmp1 = -q1i; \
         tmp0 += q0r; \
         tmp1 += q0i; \
         q0r += q1r; \
         q0i += q1i; \
         q1r = tmp0; \
         q1i = tmp1; \
         q1r *= tmp; \
         q1i *= tmp; \
         tmp = twdfac[twdad][1]; \
         tmp0 *= tmp; \
         tmp1 *= tmp; \
         q1r -= tmp1; \
         q1i += tmp0; \
      } \
   }
 
#define inv_bufly_code \
   { \
      plural FLTYP tmp2; \
      twdad = twdpt[itwp--]; /* address of twiddle factor */ \
      { \
         tmp2 = twdfac[twdad][0]; \
         tmp0 = q1r; \
         tmp0 *= tmp2; \
         tmp1 = -twdfac[twdad][1]; \
         tmp  = q1i; \
         tmp *= tmp1; \
         tmp0-= tmp; \
         tmp1 *= q1r; \
         tmp2 *= q1i; \
         tmp1 += tmp2; \
         q1r = q0r; \
         q1r -= tmp0; \
         q1i = q0i; \
         q1i -= tmp1; \
         q0r += tmp0; \
         q0i += tmp1; \
      } \
   }