testswar.c

/* -----------------------------------------------------------------------------
 *  Copyright (C) 2019-2021 daiteq s.r.o.                http://www.daiteq.com
 *
 *  This program is distributed WITHOUT ANY WARRANTY; without even
 *  the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR
 *  PURPOSE.
 *
 * -----------------------------------------------------------------------------
 *  Filename    : testswar.c
 *  Authors     : Martin Danek, Roman Bartosinski
 *  Description : LEON2 test of SWAR instructions
 *  Release     :
 *  Version     : 2.0
 *  Date        : 27.4.2021
 * -----------------------------------------------------------------------------
 */

#include <stdint.h>
#include "my_printf.h"
#include "leon-myuart.h"
#include "swar.h"
#include "ise.h"
#include "vectors_swar.h"
#include "sysregs.h"
#include "lconf.h"

// swar type definition
typedef unsigned int su4x8b  __attribute__((subword(8)));
typedef signed int ss4x8b  __attribute__((subword(8)));
typedef unsigned int su2x16b  __attribute__((subword(16)));
typedef signed int ss2x16b  __attribute__((subword(16)));

typedef unsigned int su32x1b  __attribute__((subword(1)));
typedef signed int ss32x1b  __attribute__((subword(1)));
typedef unsigned int su16x2b  __attribute__((subword(2)));
typedef signed int ss16x2b  __attribute__((subword(2)));
typedef unsigned int su10x3b  __attribute__((subword(3)));
typedef signed int ss10x3b  __attribute__((subword(3)));
typedef unsigned int su8x4b  __attribute__((subword(4)));
typedef signed int ss8x4b  __attribute__((subword(4)));
//#define swar(x,y) __builtin_swar(x,y)
//#define swarcc(x,y) __builtin_swarcc(x,y)
//#define swarctrl(x) __builtin_swarctrl(x)
//#define swaraccum(x) __builtin_swaraccum(x)

typedef union {
  unsigned  i;
  su4x8b    u8;
  ss4x8b    s8;
  su2x16b   u16;
  ss2x16b   s16;
  su32x1b   u1;
  ss32x1b   s1;
  su16x2b   u2;
  ss16x2b   s2;
  su10x3b   u3;
  ss10x3b   s3;
  su8x4b    u4;
  ss8x4b    s4;
} swdata_t;

#define PGMNAME "testswar"

#ifndef ACC_BITS
  #define ACC_BITS 22
#endif

#ifndef MAX_ACC
  #define MAX_ACC 16  // MGSIM: 4 , BSTzedboard: 2
#endif /* MAX_ACC */

#define ACCMSK ((1<<ACC_BITS)-1)

// selection of enabled code
//#define SWAR_CORR
//#define SWAR_SINCOS
//#define SWAR_ALU

#define SYSREGS

#define PRINT_DETAILS 0 /* Print all test vectors and results when computed */

#ifndef SWAR_UNIT_TYPE
  #warning Swar Video Unit !
  #define SWAR_UNIT_TYPE 2 /* 1=4x8b 2=2x16b 4=ALU */
#endif /* SWAR_UNIT_TYPE */

#define ALU_NUMEL   CORR_MAX


#define MYHEAPSIZE (128 * 1024)

extern unsigned long __bss_start;  /* start address for the .bss section. defined in linker script */
extern unsigned long _end;         /* end address for the .bss section. defined in linker script */

void report(unsigned *failcnt, unsigned ref, unsigned act) {
  printf(" ERROR - EXP %08X ACT %08X\n", ref, act);
  (*failcnt)++;
}

void report_acc(unsigned *failcnt, unsigned ref, unsigned act) {
  printf(" ERROR - ACC - EXP %08X ACT %08X\n", ref, act);
  (*failcnt)++;
}

void check_acc(unsigned *failcnt, int *acc) {
unsigned i;
volatile unsigned accval,accref;
unsigned neg[MAX_ACC];
  for (i=0;i<MAX_ACC; i++) {
    accval=__builtin_swaraccum(i<<1)&ACCMSK;
    accref=(*(acc+i))&ACCMSK;
    if (PRINT_DETAILS) printf("acc[%2d] LOWER %08X (hw %08X)\n", i, acc[i], accval);
    if (accref!=accval)
      report_acc(failcnt, *(acc+i), accval);
    if (accval<0) neg[i]=1;
    else neg[i]=0;
  }
  for (i=0;i<MAX_ACC; i++) {
    accval=__builtin_swaraccum((i<<1)+1);
    if (PRINT_DETAILS) printf("acc[%2d] UPPER HW %08X\n",i, accval);
// 32-bit integer arithmetic only, do not check the upper 32 bits
//     if (accval!=(neg[i]?-1:0))
//       report_acc(failcnt, 0, accval);
  }
}

void report_alu_arith(unsigned sgnd,unsigned reduce, unsigned saturate) {
    if (!sgnd) printf("UN");
    printf("SIGNED");
    if (reduce) printf(", REDUCE");
    if (saturate) printf(", SATURATE");
    printf("\n");
}

/* ========================================================================== */
#ifdef SWAR_ALU
// ADD, bits is one from 8,-8,16,-16, 1,-1,2,-2,3,-3,4,-4
unsigned alu_swar_add(int *data, unsigned len, int bits, unsigned saturate, unsigned reduce)
{
#pragma swar manual
  unsigned err = 0;
  int acc[MAX_ACC], accval;
  unsigned sgnd = (bits<0) ? 1 : 0;
  unsigned bwidth = (bits<0) ? -bits : bits;
  unsigned pack = 32/bwidth;
  unsigned expres;
  swdata_t a, b, res;

  for (int i=0;i<MAX_ACC; i++) acc[i]=0;

  __builtin_swarctrl(
        SW_OP_ADD | 
        ((bwidth==8) ? SW_VIDEO : ((bwidth==16) ? SW_AUDIO : SW_ALU)) | 
        (sgnd ? (SW_SGND) : 0) | 
        (reduce ? (SW_RED) : 0) | 
        (saturate ? (SW_SAT) : 0)
      );

  printf("SWAR ADD %d x %d bits ", pack, bwidth);
  report_alu_arith(sgnd, reduce, saturate);
  printf("  SWAR CONFIG: 0x%08X\n",swar_readconfig());

  for (int i=0; i<len; i+=3) {
    a.i = data[0];
    b.i = data[1];
    switch (bits) {
      case 8:
        res.u8 = a.u8 + b.u8;
        break;
       
      case -8:
        res.s8 = a.s8 + b.s8;
        break;

      case 16:
        res.u16 = a.u16 + b.u16;
        break;
        
      case -16:
        res.s16 = a.s16 + b.s16;
        break;

      case 1:
        res.u1 = a.u1 + b.u1;
        break;
        
      case -1:
        res.s1 = a.s1 + b.s1;
        break;

      case 2:
        res.u2 = a.u2 + b.u2;
        break;
        
      case -2:
        res.s2 = a.s2 + b.s2;
        break;

      case 3:
        res.u3 = a.u3 + b.u3;
        break;
        
      case -3:
        res.s3 = a.s3 + b.s3;
        break;

      case 4:
        res.u4 = a.u4 + b.u4;
        break;
        
      case -4:
        res.s4 = a.s4 + b.s4;
        break;

      default:
        break;
    }
//    res = *((unsigned *)&sdata.u2.r);

    expres = swar_alu(0, (int *)&a.i, (int *)&b.i, pack, bwidth, ADD, sgnd, saturate, reduce, acc);

    if (PRINT_DETAILS) 
      printf("  a %08X b %08X res %08X expres %08X\n", a.i, b.i, res.i, expres);

    if (res.i!=expres) report(&err, expres, res.i);
    data += 3;
  }
  check_acc(&err, acc);
  printf("------\n\n");

  return err;
}


// SUB, bits is one from 8,-8,16,-16,1,-1,2,-2,3,-3,4,-4
unsigned alu_swar_sub(int *data, unsigned len, signed bits, unsigned saturate, unsigned reduce)
{
  unsigned err = 0;
  int acc[MAX_ACC], accval;
  unsigned sgnd = (bits<0) ? 1 : 0;
  unsigned bwidth = (bits<0) ? -bits : bits;
  unsigned pack = 32/bwidth;
  unsigned expres;
  swdata_t a, b, res;

  for (int i=0;i<MAX_ACC; i++) acc[i]=0;

  __builtin_swarctrl(
        SW_OP_SUB | 
        ((bwidth==8) ? SW_VIDEO : ((bwidth==16) ? SW_AUDIO : SW_ALU)) | 
        (sgnd ? (SW_SGND) : 0) | 
        (reduce ? (SW_RED) : 0) | 
        (saturate ? (SW_SAT) : 0)
      );

  printf("SWAR SUB %d x %d bits ", pack, bwidth);
  report_alu_arith(sgnd, reduce, saturate);
  printf("  SWAR CONFIG: 0x%08X\n",swar_readconfig());

  for (int i=0; i<len; i+=3) {
    a.i = data[0];
    b.i = data[1];
    switch (bits) {
      case 8:
        res.u8 = a.u8 - b.u8;
        break;
        
      case -8:
        res.s8 = a.s8 - b.s8;
        break;

      case 16:
        res.u16 = a.u16 - b.u16;
        break;
        
      case -16:
        res.s16 = a.s16 - b.s16;
        break;

      case 1:
        res.u1 = a.u1 - b.u1;
        break;
        
      case -1:
        res.s1 = a.s1 - b.s1;
        break;

      case 2:
        res.u2 = a.u2 - b.u2;
        break;
        
      case -2:
        res.s2 = a.s2 - b.s2;
        break;

      case 3:
        res.u3 = a.u3 - b.u3;
        break;
        
      case -3:
        res.s3 = a.s3 - b.s3;
        break;

      case 4:
        res.u4 = a.u4 - b.u4;
        break;
        
      case -4:
        res.s4 = a.s4 - b.s4;
        break;

      default:
        break;
    }
    expres = swar_alu(0, (int *)&a.i, (int *)&b.i, pack, bwidth, SUB, sgnd, saturate, reduce, acc);

    if (PRINT_DETAILS) 
      printf("  a %08X b %08X res %08X expres %08X\n", a.i, b.i, res.i, expres);

    if (res.i!=expres) report(&err, expres, res.i);
    data += 3;
  }
  check_acc(&err, acc);
  printf("------\n\n");

  return err;
}


// MUL, bits is one from 8,-8,16,-16,2,-2,3,-3,4,-4
unsigned alu_swar_mul(int *data, unsigned len, signed bits, unsigned saturate, unsigned reduce)
{
  unsigned err = 0;
  int acc[MAX_ACC], accval;
  unsigned sgnd = (bits<0) ? 1 : 0;
  unsigned bwidth = (bits<0) ? -bits : bits;
  unsigned pack = 32/bwidth;
  unsigned expres;
  swdata_t a, b, res;

  for (int i=0;i<MAX_ACC; i++) acc[i]=0;

  __builtin_swarctrl(
        SW_OP_MUL | 
        ((bwidth==8) ? SW_VIDEO : ((bwidth==16) ? SW_AUDIO : SW_ALU)) | 
        (sgnd ? (SW_SGND) : 0) | 
        (reduce ? (SW_RED) : 0) | 
        (saturate ? (SW_SAT) : 0)
      );

  printf("SWAR MUL %d x %d bits ", pack, bwidth);
  report_alu_arith(sgnd, reduce, saturate);
  printf("  SWAR CONFIG: 0x%08X\n",swar_readconfig());

  for (int i=0; i<len; i+=3) {
    a.i = data[0];
    b.i = data[1];

    switch (bits) {
      case 8:
        res.u8 = a.u8 * b.u8;
        break;
        
      case -8:
        res.s8 = a.s8 * b.s8;
        break;

      case 16:
        res.u16 = a.u16 * b.u16;
        break;
        
      case -16:
        res.s16 = a.s16 * b.s16;
        break;

      case 1:
        res.u1 = a.u1 * b.u1;
        break;
        
      case -1:
        res.s1 = a.s1 * b.s1;
        break;

      case 2:
        res.u2 = a.u2 * b.u2;
        break;
        
      case -2:
        res.s2 = a.s2 * b.s2;
        break;

      case 3:
        res.u3 = a.u3 * b.u3;
        break;
        
      case -3:
        res.s3 = a.s3 * b.s3;
        break;

      case 4:
        res.u4 = a.u4 * b.u4;
        break;
        
      case -4:
        res.s4 = a.s4 * b.s4;
        break;

      default:
        break;
    }
    expres = swar_alu(0, (int *)&a.i, (int *)&b.i, pack, bwidth, MUL, sgnd, saturate, reduce, acc);

    if (PRINT_DETAILS) 
      printf("  a %08X b %08X res %08X expres %08X\n", a.i, b.i, res.i, expres);

    if (res.i!=expres) report(&err, expres, res.i);
    data += 3;
  }
  check_acc(&err, acc);
  printf("------\n\n");

  return err;
}

#endif /* SWAR_ALU */

/* ========================================================================== */
int main(void) {

  volatile unsigned a, b, res, res_sw;
  unsigned i, j;
  unsigned bits, cfg, sgnd, reduce, saturate;
  int acc[MAX_ACC], accval;
  int sop;
  unsigned failcnt;

#ifdef SYSREGS
  unsigned ticks_lo, ticks_hi, insns_lo, insns_hi;
  lconf_type lconf;
  lconf_fpu_type lconf_fpu;
  lconf_swar_type lconf_swar;
#endif /* SYSREGS */

  /* reset BSS section to zero */
  uint32_t *pMem;
  for(pMem = &__bss_start; pMem < &_end;) {
    *(pMem++) = 0;
  }

  init_uart();

#ifdef SYSREGS
  get_lconf(&lconf);
  get_lconf_fpu(&lconf_fpu);
  get_lconf_swar(&lconf_swar);

  report_lconf(&lconf);
  if ((lconf.fpu==FPU_DAIFPU)||(lconf.fpu==FPU_MEIKO))
    report_lconf_fpu(&lconf_fpu);
  report_lconf_swar(&lconf_swar);

  sysregs_init();
#endif

  failcnt = 0;

  printf("\n\nLEON2FT SWAR TEST\n");
  printf("=================\n\n");


#ifdef SWAR_CORR
  printf("CORRELATION - 2b SIGNED\n"); // a..mask, b..signal
  __builtin_swarctrl(SW_SGND | SW_OP_COR2b);
  printf("SWAR CONFIG: 0x%08X\n",swar_readconfig());
  bits = 2;
  sop = 0;

  res=0;

  for (i=0; i<CORR_MAX;i+=3) {
    a = corr[i];
    b = corr[i+1];
    res = __builtin_swar(a,b);
    if (PRINT_DETAILS) printf("a %08X b %08X res %08X\n",a,b,res);
    if (res!=corr[i+2])
      report(&failcnt, corr[i+2], res);
  }

  printf("------\n\n");

  printf("CORRELATION - 1b UNSIGNED\n"); // a..mask, b..signal
  __builtin_swarctrl(SW_OP_COR1b);
  printf("SWAR CONFIG: 0x%08X\n",swar_readconfig());
  bits = 1;
  sop = 0;

  res=0;

  for (i=0; i<CORR_MAX;i+=3) {
    a = corr[i];
    b = corr[i+1];
    res = __builtin_swar(a,b);
    res_sw = swar_corr(0, &a, &b, 32/bits, bits, 0, 0, &sop);
    if (PRINT_DETAILS) printf("a %08X b %08X res %08X\n",a,b,res);
    if (res!=res_sw)
      report(&failcnt, res_sw, res);
  }

  printf("------\n\n");

  printf("CORRELATION - 3b UNSIGNED, REDUCE\n"); // a..mask, b..signal
  __builtin_swarctrl(SW_OP_COR3b | SW_RED);
  printf("SWAR CONFIG: 0x%08X\n",swar_readconfig());
  bits = 3;
  sop = 0;

  res=0;

  for (i=0; i<CORR_MAX;i+=3) {
    a = corr[i];
    b = corr[i+1];
    res = __builtin_swar(a,b);
    res_sw = swar_corr(0, &a, &b, 32/bits, bits, 0, 1, &sop);
    if (PRINT_DETAILS) printf("a %08X b %08X res %08X\n",a,b,res);
    if (res!=res_sw)
      report(&failcnt, res_sw, res);
  }

  printf("------\n\n");

  printf("CORRELATION - 4b SIGNED, REDUCE\n"); // a..mask, b..signal
  __builtin_swarctrl(SW_OP_COR4b | SW_SGND | SW_RED);
  printf("SWAR CONFIG: 0x%08X\n",swar_readconfig());
  bits = 4;
  sop = 0;

  res=0;

  for (i=0; i<CORR_MAX;i+=3) {
    a = corr[i];
    b = corr[i+1];
    res = __builtin_swar(a,b);
    res_sw = swar_corr(0, &a, &b, 32/bits, bits, 1, 1, &sop);
    if (PRINT_DETAILS) printf("a %08X b %08X res %08X\n",a,b,res);
    if (res!=res_sw)
      report(&failcnt, res_sw, res);
  }

  printf("------\n\n");

  printf("DEMODULATION - 2b COMPLEX GROUPED\n"); // a..carrier, b..signal
  __builtin_swarctrl(SW_OP_DEMC2bG);
  printf("SWAR CONFIG: 0x%08X\n",swar_readconfig());

  for (i=0; i<DEMOD_MAX;i+=3) {
    a = demod[i];
    b = demod[i+1];
    res = __builtin_swar(a,b);
    if (PRINT_DETAILS) printf("a %08X b %08X res %08X\n",a,b,res);
    if (res!=demod[i+2])
      report(&failcnt, demod[i+2], res);
  }

  printf("------\n\n");

  printf("DEMODULATION - 2b COMPLEX\n"); // a..carrier, b..signal
  __builtin_swarctrl(SW_OP_DEMC2b);
  printf("SWAR CONFIG: 0x%08X\n",swar_readconfig());
  bits = 2;

  for (i=0; i<DEMOD_MAX;i+=3) {
    a = demod[i];
    b = demod[i+1];
    res = __builtin_swar(a,b);
    res_sw = swar_dem(0, &a, &b, 32/bits, bits, 1, 1);
    if (PRINT_DETAILS) printf("a %08X b %08X res %08X\n",a,b,res);
    if (res!=res_sw)
      report(&failcnt, res_sw, res);
  }

  printf("------\n\n");

  printf("DEMODULATION - 4b COMPLEX\n"); // a..carrier, b..signal
  __builtin_swarctrl(SW_OP_DEMC4b);
  printf("SWAR CONFIG: 0x%08X\n",swar_readconfig());
  bits = 4;

  for (i=0; i<DEMOD_MAX;i+=3) {
    a = demod[i];
    b = demod[i+1];
    res = __builtin_swar(a,b);
    res_sw = swar_dem(0, &a, &b, 32/bits, bits, 1, 1);
    if (PRINT_DETAILS) printf("a %08X b %08X res %08X\n",a,b,res);
    if (res!=res_sw)
      report(&failcnt, res_sw, res);
  }

  printf("------\n\n");

  printf("DEMODULATION - 3b REAL\n"); // a..carrier, b..signal
  __builtin_swarctrl(SW_OP_DEMR3b);
  printf("SWAR CONFIG: 0x%08X\n",swar_readconfig());
  bits = 3;

  for (i=0; i<DEMOD_MAX;i+=3) {
    a = demod[i];
    b = demod[i+1];
    res = __builtin_swar(a,b);
    res_sw = swar_dem(0, &a, &b, 32/bits, bits, 1, 0);
    if (PRINT_DETAILS) printf("a %08X b %08X res %08X\n",a,b,res);
    if (res!=res_sw)
      report(&failcnt, res_sw, res);
  }

#endif /* SWAR_CORR */

#ifdef SWAR_SINCOS
  printf("------\n\n");

  printf("SINCOS - 4b\n"); // a..arg b..N.U.
  __builtin_swarctrl(SW_OP_SC4b);
  printf("SWAR CONFIG: 0x%08X\n",swar_readconfig());
  bits = 4;

  res=0;

  for (i=0; i<DEMOD_MAX;i+=3) {
    a = demod[i];
    b = demod[i+1];
    res = __builtin_swar(a,b);
    res_sw = swar_sincos(0, a, bits);
    if (PRINT_DETAILS) printf("arg %08X res %08X\n",a,res);
    if (res!=res_sw)
      report(&failcnt, res_sw, res);
  }



  printf("------\n\n");

  printf("SINCOS - 3b\n"); // a..arg b..N.U.
  __builtin_swarctrl(SW_OP_SC3b);
  printf("SWAR CONFIG: 0x%08X\n",swar_readconfig());
  bits = 3;

  res=0;

  for (i=0; i<DEMOD_MAX;i+=3) {
    a = demod[i];
    b = demod[i+1];
    res = __builtin_swar(a,b);
    res_sw = swar_sincos(0, a, bits);
    if (PRINT_DETAILS) printf("arg %08X res %08X\n",a,res);
    if (res!=res_sw)
      report(&failcnt, res_sw, res);
  }

  printf("------\n\n");

  printf("SINCOS - 2b\n"); // a..arg b..N.U.
  __builtin_swarctrl(SW_OP_SC2b);
  printf("SWAR CONFIG: 0x%08X\n",swar_readconfig());
  bits = 2;

  res=0;

  for (i=0; i<DEMOD_MAX;i+=3) {
    a = demod[i];
    b = demod[i+1];
    res = __builtin_swar(a,b);
    res_sw = swar_sincos(0, a, bits);
    if (PRINT_DETAILS) printf("arg %08X res %08X\n",a,res);
    if (res!=res_sw)
      report(&failcnt, res_sw, res);
  }


  printf("------\n\n");

  printf("SINCOS - 1b\n"); // a..arg b..N.U.
  __builtin_swarctrl(SW_OP_SC1b);
  printf("SWAR CONFIG: 0x%08X\n",swar_readconfig());
  bits = 1;

  res=0;

  for (i=0; i<DEMOD_MAX;i+=3) {
    a = demod[i];
    b = demod[i+1];
    res = __builtin_swar(a,b);
    res_sw = swar_sincos(0, a, bits);
    if (PRINT_DETAILS) printf("arg %08X res %08X\n",a,res);
    if (res!=res_sw)
      report(&failcnt, res_sw, res);
  }

#endif /* SWAR_SINCOS */



#ifdef SWAR_ALU

  saturate = 1;
  do {

  reduce = 1;
  do {

  sgnd = 1;
  do {
    j = SWAR_UNIT_TYPE;

    if (j==1) {           /* SWAR 4x8b */
      // add 4x8b
      //cfg |= SW_VIDEO;
      if (sgnd) {
        failcnt += alu_swar_add((int *)corr, ALU_NUMEL, -8, saturate, reduce);
        failcnt += alu_swar_sub((int *)corr, ALU_NUMEL, -8, saturate, reduce);
        failcnt += alu_swar_mul((int *)corr, ALU_NUMEL, -8, saturate, reduce);
      } else {
        failcnt += alu_swar_add((int *)corr, ALU_NUMEL, 8, saturate, reduce);
        failcnt += alu_swar_sub((int *)corr, ALU_NUMEL, 8, saturate, reduce);
        failcnt += alu_swar_mul((int *)corr, ALU_NUMEL, 8, saturate, reduce);
      }

    } else if (j==2) {    /* SWAR 2x16b */
      //bits = 16;
      //cfg |= SW_AUDIO;
      if (sgnd) {
        failcnt += alu_swar_add((int *)corr, ALU_NUMEL, -16, saturate, reduce);
        failcnt += alu_swar_sub((int *)corr, ALU_NUMEL, -16, saturate, reduce);
        failcnt += alu_swar_mul((int *)corr, ALU_NUMEL, -16, saturate, reduce);
      } else {
        failcnt += alu_swar_add((int *)corr, ALU_NUMEL, 16, saturate, reduce);
        failcnt += alu_swar_sub((int *)corr, ALU_NUMEL, 16, saturate, reduce);
        failcnt += alu_swar_mul((int *)corr, ALU_NUMEL, 16, saturate, reduce);
      }
      
      
    } else if (j==4) {    /* SWAR 32x1b,16x2b,10x3b,8x4b */
      //cfg |= SW_ALU;
      //bits = 1;
      if (sgnd) {
        failcnt += alu_swar_add((int *)corr, ALU_NUMEL, -1, saturate, reduce);
        failcnt += alu_swar_sub((int *)corr, ALU_NUMEL, -1, saturate, reduce);
        failcnt += alu_swar_mul((int *)corr, ALU_NUMEL, -1, saturate, reduce);
      } else {
        failcnt += alu_swar_add((int *)corr, ALU_NUMEL, 1, saturate, reduce);
        failcnt += alu_swar_sub((int *)corr, ALU_NUMEL, 1, saturate, reduce);
        failcnt += alu_swar_mul((int *)corr, ALU_NUMEL, 1, saturate, reduce);
      }
      //bits = 2;
      if (sgnd) {
        failcnt += alu_swar_add((int *)corr, ALU_NUMEL, -2, saturate, reduce);
        failcnt += alu_swar_sub((int *)corr, ALU_NUMEL, -2, saturate, reduce);
        failcnt += alu_swar_mul((int *)corr, ALU_NUMEL, -2, saturate, reduce);
      } else {
        failcnt += alu_swar_add((int *)corr, ALU_NUMEL, 2, saturate, reduce);
        failcnt += alu_swar_sub((int *)corr, ALU_NUMEL, 2, saturate, reduce);
        failcnt += alu_swar_mul((int *)corr, ALU_NUMEL, 2, saturate, reduce);
      }
      //bits = 3;
      if (sgnd) {
        failcnt += alu_swar_add((int *)corr, ALU_NUMEL, -3, saturate, reduce);
        failcnt += alu_swar_sub((int *)corr, ALU_NUMEL, -3, saturate, reduce);
        failcnt += alu_swar_mul((int *)corr, ALU_NUMEL, -3, saturate, reduce);
      } else {
        failcnt += alu_swar_add((int *)corr, ALU_NUMEL, 3, saturate, reduce);
        failcnt += alu_swar_sub((int *)corr, ALU_NUMEL, 3, saturate, reduce);
        failcnt += alu_swar_mul((int *)corr, ALU_NUMEL, 3, saturate, reduce);
      }
      //bits = 4;
      if (sgnd) {
        failcnt += alu_swar_add((int *)corr, ALU_NUMEL, -4, saturate, reduce);
        failcnt += alu_swar_sub((int *)corr, ALU_NUMEL, -4, saturate, reduce);
        failcnt += alu_swar_mul((int *)corr, ALU_NUMEL, -4, saturate, reduce);
      } else {
        failcnt += alu_swar_add((int *)corr, ALU_NUMEL, 4, saturate, reduce);
        failcnt += alu_swar_sub((int *)corr, ALU_NUMEL, 4, saturate, reduce);
        failcnt += alu_swar_mul((int *)corr, ALU_NUMEL, 4, saturate, reduce);
      }

    }

  } while (sgnd--);
  } while (reduce--);
  } while (saturate--);


#endif /* SWAR_ALU */

  printf("------\n\n");

#ifdef SYSREGS
  sysregs_stop();
  sysregs_read(&ticks_lo, &ticks_hi, &insns_lo, &insns_hi);

  if (ticks_hi)
    printf("TEST EXECUTED IN (%d * 2^32 + %d) TICKS AND (%d * 2^32 + %d) INSTRUCTIONS\n\n", ticks_hi, ticks_lo, insns_lo, insns_hi);
  else
    printf("TEST EXECUTED IN %d TICKS AND %d INSTRUCTIONS\n\n", ticks_lo, insns_lo);
#endif

  if (!failcnt)
    printf("END OF TEST\n\n");
  else
    printf("TEST COMPLETED WITH ERRORS\n\n");

  __asm__ volatile (
    "\t ta 0\n"
    "\t nop\n"
    "\t nop\n"
  );

  return 0;
}