swarfir.c

/* -----------------------------------------------------------------------------
 *  Copyright (C) 2021,2022 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    : swarfir.c
 *  Authors     : Roman Bartosinski
 *  Description : FIR filter with SWAR operations
 *  Release     :
 *  Version     : 1.0
 *  Date        : 27.4.2021
 * -----------------------------------------------------------------------------
 */

#include <stdint.h>
#include <stdlib.h>

#include "swarconf.h"

#include "../common_sys_header.inc"

//#define PRINT_TYPES     1
//#define PRINT_RESULTS   1
//#define PRINT_FAILED    1


#define TEST_OFFLINE 1
#define TEST_ONLINE  1

// SWARBASESIZE != 32 or 64
#ifndef SWARBASESIZE
  #warning Default SWAR basic type u64 is used.
  #define SWARBASESIZE          64
#endif /* SWARBASESIZE BASESIZE */

// example definitions - swar with 8bit elements
#define BITSZ             16

#ifndef CONFIG_DATA_LENGTH
  #define DATA_LENGTH       50
#else
  #define DATA_LENGTH       CONFIG_DATA_LENGTH
#endif /* CONFIG_DATA_LENGTH */

#ifndef CONFIG_FIR_LENGTH
  #define FIR_LENGTH        5    /* more accurately it is a number of FIR coefficients [c0,c1,c2,c3,c4,...]  */
#else
  #define FIR_LENGTH        CONFIG_FIR_LENGTH
#endif /* CONFIG_FIR_LENGTH */

//#ifdef CONFIG_ONLINE_MULTI
  //#define ONLINE_MULTIPLE   CONFIG_ONLINE_MULTI
//#else
  //#define ONLINE_MULTIPLE   1
//#endif /* CONFIG_ONLINE_MULTI */

#ifdef CONFIG_ONLINE_FIRLEN
  #define ONLINE_FIRLEN   CONFIG_ONLINE_FIRLEN
#else
  #define ONLINE_FIRLEN   (TEST_FIRCOEF_SIZE)
#endif /* CONFIG_ONLINE_MULTI */
#ifdef CONFIG_DATA_LENGTH
  #define ONLINE_DATALEN  (CONFIG_DATA_LENGTH)
#else
  #define ONLINE_DATALEN  TEST_INDATA_SIZE
#endif

#define MAXRAND           20

#if SWARBASESIZE==32
  typedef int32_t basetype;
  #define PRNFMTHEX  "%X"
  #define PRNFMTDEC  "%u"
#else
  typedef int64_t basetype;
  #define PRNFMTHEX  "%lX"
  #define PRNFMTDEC  "%lu"
#endif /* SWARBASESIZE */

typedef basetype swpack __attribute__((subword(BITSZ)));
typedef basetype swelm __attribute__((subword(BITSZ,1)));

#if BITSZ==8
  typedef int8_t dataelm;
  #define SWTP  SW_VIDEO
#elif BITSZ==16
  typedef int16_t dataelm;
  #define SWTP  SW_AUDIO
#elif BITSZ==32
  typedef int32_t dataelm;
  #define SWTP  SW_ALU
#endif /* BITSZ */

typedef basetype resultelm;

typedef union {
  swpack    s;
#if SWARBASESIZE==32
  int       i;
  unsigned  u;
#else
  long      i;
  unsigned long u;
#endif
} multi_t;

#define MAXNUM            ((1<<BITSZ))
#define MINVAL            0
#define MAXVAL            ((1<<BITSZ)-1)

#define PACKING           (SWARBASESIZE/BITSZ)
#define DATA_ARRAY_SIZE   ((DATA_LENGTH+PACKING-1)/PACKING)   /* array for packed data */
#define COEF_ARRAY_SIZE   ((FIR_LENGTH+PACKING-1)/PACKING)    /* array for packed coefficients */

#define MAXFIRLENGTH      128
/* -------------------------------------------------------------------------- */
void print_dir_data(const char *prefix, dataelm *pd, int len);
void print_swar_data(const char *prefix, swpack *ps, int len);
void print_counters(const char *prefix, unsigned *tm, unsigned *ins);

/* -------------------------------------------------------------------------- */
// experimental data
// signal (sinus 1kHz + sinus 15kHz, sampl.f=48kHz
const dataelm test_input[] = {
  0,55,-9,18,92,38,33,115,80,42,121,109,46,109,121,42,80,115,33,38,92,18,-9,55,0,
  -55,9,-18,-92,-38,-33,-115,-80,-42,-121,-109,-46,-109,-121,-42,-80,-115,-33,-38,
  -92,-18,9,-55,0,55,-9,18,92,38,33,115,80,42,121,109,46,109,121,42,80,115,33,38,
  92,18,-9,55,0,-55,9,-18,-92,-38,-33,-115,-80,-42,-121,-109,-46,-109,-121,-42,-80,
  -115,-33,-38,-92,-18,9,-55,0,55,-9,18,92,38,33,115,80,42,121,109,46,109,121,42,
  80,115,33,38,92,18,-9,55,0,-55,9,-18,-92,-38,-33,-115,-80,-42,-121,-109,-46,-109,
  -121,-42,-80,-115,-33,-38,-92,-18,9,-55,0,55,-9,18,92,38,33,115,80,42,121,109,
  46,109,121,42,80,115,33,38,92,18,-9,55,0,-55,9,-18,-92,-38,-33,-115,-80,-42,-121,
  -109,-46,-109,-121,-42,-80,-115,-33,-38,-92,-18,9,-55,0,55,-9,18,92,38,33,115,
  80,42,121,109,46,109,121,42,80,115,33,38,92,18,-9,55,0,-55,9,-18,-92,-38,-33,-115,
  -80,-42,-121,-109,-46,-109,-121,-42,-80,-115,-33,-38,-92,-18,9,-55,0,55,-9,18,
  92,38,33,115,80,42,121,109,46,109,121,42,80,115,33,38,92,18,-9,55,0,-55,9,-18,
  -92,-38,-33,-115,-80,-42,-121,-109,-46,-109,-121,-42,-80,-115,-33,-38,-92,-18,
  9,-55,0,55,-9,18,92,38,33,115,80,42,121,109,46,109,121,42,80,115,33,38,92,18,
  -9,55,0,-55,9,-18,-92,-38,-33,-115,-80,-42,-121,-109,-46,-109,-121,-42,-80,-115,
  -33,-38,-92,-18,9,-55,0,55,-9,18,92,38,33,115,80,42,121,109,46,109,121,42,80,115,
  33,38,92,18,-9,55,0,-55,9,-18,-92,-38,-33,-115,-80,-42,-121,-109,-46,-109,-121,
  -42,-80,-115,-33,-38,-92,-18,9,-55,0,55,-9,18,92,38,33,115,80,42,121,109,46,109,
  121,42,80,115,33,38,92,18,-9,55,0,-55,9,-18,-92,-38,-33,-115,-80,-42,-121,-109,
  -46,-109,-121,-42,-80,-115,-33,-38,-92,-18,9,-55,0,55,-9,18,92,38,33,115,80,42,
  121,109,46,109,121,42,80,115,33,38,92,18,-9,55,0,-55,9,-18,-92,-38,-33,-115,-80,
  -42,-121,-109,-46,-109,-121,-42,-80,-115,-33,-38,-92,-18,9,-55,0,55,-9,18,92,38,
  33,115,80,42,121,109,46,109,121,42,80,115,33,38,92,18,-9,55,0,-55,9,-18,-92,-38,
  -33,-115,-80,-42,-121,-109,-46,-109,-121,-42,-80,-115,-33,-38,-92,-18,9,-55,0,55,
  -9,18,92,38,33,115,80,42,121,109,46,109,121,42,80,115,33,38,92,18,-9,55,0,-55,9,
  -18,-92,-38,-33,-115,-80,-42,-121,-109,-46,-109,-121,-42,-80,-115,-33,-38,-92,-18,
  9,-55,0,55,-9,18,92,38,33,115,80,42,121,109,46,109,121,42,80,115,33,38,92,18,-9,
  55,0,-55,9,-18,-92,-38,-33,-115,-80,-42,-121,-109,-46,-109,-121,-42,-80,-115,-33,
  -38,-92,-18,9,-55,0,55,-9,18,92,38,33,115,80,42,121,109,46,109,121,42,80,115,33,
  38,92,18,-9,55,0,-55,9,-18,-92,-38,-33,-115,-80,-42,-121,-109,-46,-109,-121,-42,
  -80,-115,-33,-38,-92,-18,9,-55,0,55,-9,18,92,38,33,115,80,42,121,109,46,109,121,
  42,80,115,33,38,92,18,-9,55,0,-55,9,-18,-92,-38,-33,-115,-80,-42,-121,-109,-46,
  -109,-121,-42,-80,-115,-33,-38,-92,-18,9,-55,0,55,-9,18,92,38,33,115,80,42,121,
  109,46,109,121,42,80,115,33,38,92,18,-9,55,0,-55,9,-18,-92,-38,-33,-115,-80,-42,
  -121,-109,-46,-109,-121,-42,-80,-115,-33,-38,-92,-18,9,-55,0,55,-9,18,92,38,33,
  115,80,42,121,109,46,109,121,42,80,115,33,38,92,18,-9,55,0,-55,9,-18,-92,-38,-33,
  -115,-80,-42,-121,-109,-46,-109,-121,-42,-80,-115,-33,-38,-92,-18,9,-55,0,55,-9,
  18,92,38,33,115,80,42,121,109,46,109,121,42,80,115,33,38,92,18,-9,55,0,-55,9,-18,
  -92,-38,-33,-115,-80,-42,-121,-109,-46,-109,-121,-42,-80,-115,-33,-38,-92,-18,9,
  -55,0,55,-9,18,92,38,33,115,80,42,121,109,46,109,121,42,80,115,33,38,92,18,-9,55,
  0,-55,9,-18,-92,-38,-33,-115,-80,-42,-121,-109,-46,-109,-121,-42,-80,-115,-33,-38,
  -92,-18,9,-55,0,55,-9,18,92,38,33,115,80,42,121,109,46,109,121,42,80,115,33,38,92,
  18,-9,55,0,-55,9,-18,-92,-38,-33,-115,-80,-42,-121,-109,-46,-109,-121,-42,-80,
  -115,-33,-38,-92,-18,9,-55,0,55,-9,18,92,38,33,115,80,42,121,109,46,109,121,42,
  80,115,33,38,92,18,-9,55,0,-55,9,-18,-92,-38,-33,-115,-80,-42,-121,-109,-46,-109,-121,-42
};
#define TEST_INDATA_SIZE  (sizeof(test_input)/sizeof(dataelm))
// coefficients of FIR filter
const dataelm test_fircoef[] = {
  0,0,0,0,1,1,0,-2,-3,-3,0,6,14,20,23,20,14,6,0,-3,-3,-2,0,1,1
};
#define TEST_FIRCOEF_SIZE  (sizeof(test_fircoef)/sizeof(dataelm))
// extra longer vector for testing only
const dataelm test_fircoef_extra[] = {
  0,0,0,0,1,1,0,-2,-3,-3,0,6,14,20,23,20,14,6,0,-3,-3,-2,0,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
};

// filtered data for checking
const resultelm test_output[] = {
  0,0,0,0,0,0,-55,-156,-156,-119,-38,293,754,1365,2142,3087,3994,4926,5848,6579,
  7204,7706,8022,8244,8346,8244,8022,7706,7204,6579,5903,5082,4150,3206,2180,1072,
  0,-1072,-2180,-3206,-4150,-5082,-5903,-6579,-7204,-7706,-8022,-8244,-8346,-8244,
  -8022,-7706,-7204,-6579,-5903,-5082,-4150,-3206,-2180,-1072,0,1072,2180,3206,4150,
  5082,5903,6579,7204,7706,8022,8244,8346,8244,8022,7706,7204,6579,5903,5082,4150,
  3206,2180,1072,0,-1072,-2180,-3206,-4150,-5082,-5903,-6579,-7204,-7706,-8022,-8244,
  -8346,-8244,-8022,-7706,-7204,-6579,-5903,-5082,-4150,-3206,-2180,-1072,0,1072,
  2180,3206,4150,5082,5903,6579,7204,7706,8022,8244,8346,8244,8022,7706,7204,6579,
  5903,5082,4150,3206,2180,1072,0,-1072,-2180,-3206,-4150,-5082,-5903,-6579,-7204,
  -7706,-8022,-8244,-8346,-8244,-8022,-7706,-7204,-6579,-5903,-5082,-4150,-3206,
  -2180,-1072,0,1072,2180,3206,4150,5082,5903,6579,7204,7706,8022,8244,8346,8244,
  8022,7706,7204,6579,5903,5082,4150,3206,2180,1072,0,-1072,-2180,-3206,-4150,-5082,
  -5903,-6579,-7204,-7706,-8022,-8244,-8346,-8244,-8022,-7706,-7204,-6579,-5903,
  -5082,-4150,-3206,-2180,-1072,0,1072,2180,3206,4150,5082,5903,6579,7204,7706,8022,
  8244,8346,8244,8022,7706,7204,6579,5903,5082,4150,3206,2180,1072,0,-1072,-2180,
  -3206,-4150,-5082,-5903,-6579,-7204,-7706,-8022,-8244,-8346,-8244,-8022,-7706,
  -7204,-6579,-5903,-5082,-4150,-3206,-2180,-1072,0,1072,2180,3206,4150,5082,5903,
  6579,7204,7706,8022,8244,8346,8244,8022,7706,7204,6579,5903,5082,4150,3206,2180,
  1072,0,-1072,-2180,-3206,-4150,-5082,-5903,-6579,-7204,-7706,-8022,-8244,-8346,
  -8244,-8022,-7706,-7204,-6579,-5903,-5082,-4150,-3206,-2180,-1072,0,1072,2180,
  3206,4150,5082,5903,6579,7204,7706,8022,8244,8346,8244,8022,7706,7204,6579,5903,
  5082,4150,3206,2180,1072,0,-1072,-2180,-3206,-4150,-5082,-5903,-6579,-7204,-7706,
  -8022,-8244,-8346,-8244,-8022,-7706,-7204,-6579,-5903,-5082,-4150,-3206,-2180,
  -1072,0,1072,2180,3206,4150,5082,5903,6579,7204,7706,8022,8244,8346,8244,8022,
  7706,7204,6579,5903,5082,4150,3206,2180,1072,0,-1072,-2180,-3206,-4150,-5082,-5903,
  -6579,-7204,-7706,-8022,-8244,-8346,-8244,-8022,-7706,-7204,-6579,-5903,-5082,-4150,
  -3206,-2180,-1072,0,1072,2180,3206,4150,5082,5903,6579,7204,7706,8022,8244,8346,
  8244,8022,7706,7204,6579,5903,5082,4150,3206,2180,1072,0,-1072,-2180,-3206,-4150,
  -5082,-5903,-6579,-7204,-7706,-8022,-8244,-8346,-8244,-8022,-7706,-7204,-6579,-5903,
  -5082,-4150,-3206,-2180,-1072,0,1072,2180,3206,4150,5082,5903,6579,7204,7706,8022,
  8244,8346,8244,8022,7706,7204,6579,5903,5082,4150,3206,2180,1072,0,-1072,-2180,
  -3206,-4150,-5082,-5903,-6579,-7204,-7706,-8022,-8244,-8346,-8244,-8022,-7706,
  -7204,-6579,-5903,-5082,-4150,-3206,-2180,-1072,0,1072,2180,3206,4150,5082,5903,
  6579,7204,7706,8022,8244,8346,8244,8022,7706,7204,6579,5903,5082,4150,3206,2180,
  1072,0,-1072,-2180,-3206,-4150,-5082,-5903,-6579,-7204,-7706,-8022,-8244,-8346,
  -8244,-8022,-7706,-7204,-6579,-5903,-5082,-4150,-3206,-2180,-1072,0,1072,2180,
  3206,4150,5082,5903,6579,7204,7706,8022,8244,8346,8244,8022,7706,7204,6579,5903,
  5082,4150,3206,2180,1072,0,-1072,-2180,-3206,-4150,-5082,-5903,-6579,-7204,-7706,
  -8022,-8244,-8346,-8244,-8022,-7706,-7204,-6579,-5903,-5082,-4150,-3206,-2180,
  -1072,0,1072,2180,3206,4150,5082,5903,6579,7204,7706,8022,8244,8346,8244,8022,
  7706,7204,6579,5903,5082,4150,3206,2180,1072,0,-1072,-2180,-3206,-4150,-5082,-5903,
  -6579,-7204,-7706,-8022,-8244,-8346,-8244,-8022,-7706,-7204,-6579,-5903,-5082,
  -4150,-3206,-2180,-1072,0,1072,2180,3206,4150,5082,5903,6579,7204,7706,8022,8244,
  8346,8244,8022,7706,7204,6579,5903,5082,4150,3206,2180,1072,0,-1072,-2180,-3206,
  -4150,-5082,-5903,-6579,-7204,-7706,-8022,-8244,-8346,-8244,-8022,-7706,-7204,
  -6579,-5903,-5082,-4150,-3206,-2180,-1072,0,1072,2180,3206,4150,5082,5903,6579,
  7204,7706,8022,8244,8346,8244,8022,7706,7204,6579,5903,5082,4150,3206,2180,1072,
  0,-1072,-2180,-3206,-4150,-5082,-5903,-6579,-7204,-7706,-8022,-8244,-8346,-8244,
  -8022,-7706,-7204,-6579,-5903,-5082,-4150,-3206,-2180,-1072,0,1072,2180,3206,4150,
  5082,5903,6579,7204,7706,8022,8244,8346,8244,8022,7706,7204,6579,5903,5082,4150,
  3206,2180,1072,0,-1072,-2180,-3206,-4150,-5082,-5903,-6579,-7204,-7706,-8022,-8244,
  -8346,-8244,-8022,-7706,-7204,-6579,-5903,-5082,-4150,-3206,-2180,-1072,0,1072,
  2180,3206,4150,5082,5903,6579,7204,7706,8022,8244,8346,8244,8022,7706,7204,6579,
  5903,5082,4150,3206,2180,1072,0,-1072,-2180,-3206,-4150,-5082,-5903,-6579,-7204,
  -7706,-8022,-8244,-8346,-8244,-8022,-7706,-7204,-6579,-5903,-5082,-4150,-3206,
  -2180,-1072,0,1072,2180,3206,4150,5082,5903,6579,7204,7706,8022,8244,8346,8244,
  8022,7706,7204,6579,5903,5082,4150,3206,2180,1072,0,-1072,-2180,-3206,-4150,-5082,
  -5903,-6579,-7204,-7706,-8022,-8244,-8346,-8244,-8022,-7706,-7204,-6579,-5903,
  -5082,-4150,-3206,-2180,-1072,0,1072,2180,3206,4150,5082,5903,6579,7204,7706,8022,
  8244,8346,8244,8022,7706,7204,6579,5903,5082,4150,3206,2180,1072,0,-1072,-2180,
  -3206,-4150,-5082,-5903,-6579,-7204,-7706,-8022,-8244,-8346,-8244,-8022,-7706,
  -7204,-6579,-5903,-5082,-4150,-3206,-2180,-1072,0,1072,2180,3206,4150,5082,5903,
  6579,7204,7706,8022,8244,8346,8244,8022,7706,7204,6579,5903,5082,4150,3206,2180,
  1072,0,-1072,-2180,-3206,-4150,-5082,-5903,-6579,-7204,-7706,-8022,-8244,-8346,
  -8244,-8022,-7706,-7204,-6579,-5903,-5082,-4150,-3206,-2180,-1072,0,1072,2180,
  3206,4150,5082,5903,6579,7204,7706,8022,8244,8346,8244,8022,7706,7204,6579,5903,
  5082,4150,3206,2180,1072,0,-1072,-2180,-3206
};
#define TEST_OUTPUT_SIZE  (sizeof(test_output)/sizeof(resultelm))

resultelm test_dir_output[TEST_OUTPUT_SIZE];
resultelm test_swar_output[TEST_OUTPUT_SIZE];

/* -------------------------------------------------------------------------- */
// data for classical FIR computation
dataelm dir_in_x[DATA_LENGTH+FIR_LENGTH];  /* INPUT DATA + zeros */
dataelm dir_in_c[FIR_LENGTH];             /* COEFFICIENTS */
resultelm dir_out_y[DATA_LENGTH];            /* OUTPUT DATA */


// data for FIR computation with SWAR operations
swpack swar_in_x[(DATA_LENGTH+FIR_LENGTH+(PACKING-1))/PACKING]; /* space for data ramp up */
swpack swar_in_c[COEF_ARRAY_SIZE];

resultelm swar_out_y_red[DATA_LENGTH];
resultelm swar_out_y_accum[DATA_LENGTH];
resultelm swar_out_y_add[DATA_LENGTH];

/* array of extended vectors of coefficients */
#define EXTCOEF_LEN   (FIR_LENGTH+PACKING-1+(PACKING-1))/PACKING  /* array of packed coefficients */
//multi_t extcoefs[PACKING][EXTCOEF_LEN];
swpack swar_ecoefs[PACKING][EXTCOEF_LEN];


/* ========================================================================== */
// data preparation
void prepare_data(void)
{
  for (unsigned i=0;i<DATA_LENGTH;++i) {
    int d = rand();
//printf("!x %d -> ", d);
    d = ((d&0x1000)?-1:1) * (d % (MAXRAND));
//printf(" %d\n", d);
    dir_in_x[i] = d;
    swar_in_x[i/PACKING][i%PACKING] = d;
  }

  for (unsigned i=DATA_LENGTH;i<(DATA_LENGTH+FIR_LENGTH);++i)
    dir_in_x[i] = 0;

  for (unsigned i=0;i<FIR_LENGTH;++i) {
    int c = rand();
//printf("!c %d -> ", c);
    c = ((c&0x1000)?-1:1) * (c % (MAXRAND));
//printf(" %d\n", c);
    dir_in_c[i] = c;
    swar_in_c[i/PACKING][i%PACKING] = c;
  }

  //// prepare coefficients with different aligning
  //for (unsigned  p=0; p<PACKING; ++p) { // for each alignment
    //unsigned rem = 0;
    //for(unsigned w=0; w<EXTCOEF_LEN; ++w) {
      //unsigned nv = ((unsigned *)swar_in_c)[w];
      //extcoefs[p][w].u = rem | (nv<<(p*BITSZ));
      //rem = (nv>>((PACKING-p)*BITSZ)) & ((1<<(BITSZ*p))-1);
    //}
    //extcoefs[p][EXTCOEF_LEN].u = rem;
  //}
//#if 1
  //// debug print array of coefficients
  //for(unsigned p=0;p<PACKING;++p) {
    //m_print(" #"); i_print(p); m_print(": ");
    //print_swar_data("extcoef ", (swpack*)&extcoefs[p][0], EXTCOEF_LEN*PACKING);
  //}
//#endif

  //// copy extended coefficients to standalone swar array
  //{
    //unsigned *pi = (unsigned *)extcoefs;
    //unsigned *po = (unsigned *)swar_ecoefs;
    //for (unsigned p=0;p<PACKING*EXTCOEF_LEN;++p)
      //*po++ = *pi++;
  //}

  for (unsigned p=0; p<PACKING; ++p) {
    for (unsigned w=0; w<EXTCOEF_LEN; ++w) {
      *(resultelm *)&swar_ecoefs[p][w] = 0;
    }
    for (unsigned c=0; c<FIR_LENGTH;++c) {
      unsigned iw = (c+p) / PACKING;
      unsigned ie = (c+p) % PACKING;
      swar_ecoefs[p][iw][ie] = dir_in_c[c];
    }
  }
  
#if 0
  // debug print array of coefficients
  for(unsigned p=0;p<PACKING;++p) {
    printf(" #%u: ", p);
    print_swar_data("DIRecoef ", (swpack*)&swar_ecoefs[p][0], EXTCOEF_LEN*PACKING);
  }
#endif
#if 0
  for (unsigned i=0;i<(DATA_LENGTH+FIR_LENGTH);++i)
    printf(" * %d\n", dir_in_x[i]);
#endif
}


/* ========================================================================== */
// classical computation element by element
// data must be correctly long (with zeroes after real data)
// datalen = number of input data (or length of required output vector)
// firlen = number of coefficients (b0,b1,b2,b3,...)
void fir_offline(resultelm *out_y, dataelm *in_x, int len_data, dataelm *in_c)
{
  for (unsigned j=0; j<len_data; ++j) {
    resultelm s = 0;
    for (unsigned i=0; i<FIR_LENGTH; ++i) {
      s += in_c[i] * in_x[i+j];
    }
    out_y[j] = s;
  }
}


dataelm dir_x_mem[MAXFIRLENGTH];  // [x(n),x(n-1),x(n-2),x(n-3),...]
dataelm dir_c_mem[MAXFIRLENGTH];  // [c0,c1,c2,c3,c4,...]
unsigned mem_firlen;
unsigned rbuf_next;

void fir_online_init(unsigned firlen, const dataelm *in_c)
{
  if (firlen>MAXFIRLENGTH) {
    printf("!!! Too long FIR filter\n");
    firlen = MAXFIRLENGTH;
  }
  for (unsigned n=0;n<firlen;++n) {
    dir_x_mem[n] = 0;
    dir_c_mem[n] = in_c[n]; //firlen-n-1];
//printf("#%u: c %d (x%X) x %d (x%X)\n", n, dir_c_mem[n], dir_c_mem[n], dir_x_mem[n], dir_x_mem[n]);
  }
  mem_firlen = firlen;
//NL;
}

resultelm fir_online(dataelm in_x)
{
  resultelm y = 0;
  for (unsigned i=mem_firlen-1;i>0;--i) {
    dir_x_mem[i] = dir_x_mem[i-1];  // x(n-1) = x'(n)
//printf("#%u: %u (x%X)\n", i, dir_x_mem[i], dir_x_mem[i]);
    y += dir_x_mem[i]*dir_c_mem[i];
  }
  dir_x_mem[0] = in_x;
  y+= dir_x_mem[0]*dir_c_mem[0];
//printf("#0: %u (x%X) ---> %u (x%X)\n", dir_x_mem[0], dir_x_mem[0], y, y);
  return y;
}

/* ---------------------------------- */

void fir_online_rbuf_init(unsigned firlen, const dataelm *in_c)
{
  if (firlen>MAXFIRLENGTH) {
    printf("!!! Too long FIR filter\n");
    firlen = MAXFIRLENGTH;
  }
  rbuf_next = firlen-1; // fill from back to front
  for (unsigned n=0;n<firlen;++n) {
    dir_x_mem[n] = 0;
    dir_c_mem[n] = in_c[n];
//m_print("#"); i_print(n); m_print(":"); i_print(dir_c_mem[n]); m_print("  ");
  }
  mem_firlen = firlen;
//NL;
}

resultelm fir_online_rbuf(dataelm in_x)
{
  resultelm y = 0;
  unsigned xi = rbuf_next;
  dir_x_mem[rbuf_next] = in_x;
  
  for (unsigned i=0;i<mem_firlen;++i) {
//m_print("#"); i_print(i); m_print(":"); i_print(dir_x_mem[i]); m_print("  ");
    y += dir_c_mem[i] * dir_x_mem[xi];
    if (++xi>=mem_firlen) xi=0;
  }
  if (rbuf_next) rbuf_next--;
  else rbuf_next = mem_firlen-1;
  return y;
}

/* ---------------------------------- */
unsigned bufsize;
unsigned bufmask;

void fir_online_align_rbuf_init(unsigned firlen, const dataelm *in_c)
{
  if (firlen>MAXFIRLENGTH) {
    printf("!!! Too long FIR filter\n");
    firlen = MAXFIRLENGTH;
  }
  unsigned p = 8*sizeof(unsigned) - __builtin_clz(firlen);
  mem_firlen = firlen;
  bufsize = 1<<p;
  bufmask = bufsize-1;
//  printf("FIRLEN = %u %u, mask=%u\n", mem_firlen, bufsize, bufmask);

  rbuf_next = bufsize-1; // fill from back to front
  for (unsigned n=0;n<bufsize;++n) {
    dir_x_mem[n] = 0;
  }
  for (unsigned n=0;n<firlen;++n) {
    dir_c_mem[n] = in_c[n];
//printf("c # %u : %d x%X\n", n, dir_c_mem[n], dir_c_mem[n]);
  }
}

resultelm fir_online_align_rbuf(dataelm in_x)
{
  resultelm y = 0;
  unsigned xi = rbuf_next;
  dir_x_mem[rbuf_next] = in_x;
  
  for (unsigned i=0;i<mem_firlen;++i) {
//printf("  x # %u : %d x%X\n", i, dir_x_mem[i], dir_x_mem[i]);
    y += dir_c_mem[i] * dir_x_mem[xi];
    xi = (xi+1)&bufmask;
//printf("   y = %d x%X | xi = %u\n", y, y, xi);
  }
  rbuf_next = (rbuf_next-1)&bufmask;
  return y;
}


/* ========================================================================== */
// swar computation
// use swar multiplication with reduction
void fir_swar_mul_red(resultelm *out_y, swpack *in_x, int len_data) //, swpack *in_c)
{
  #pragma swar reduce
  // loop over all len_data data
  for (unsigned j=0; j<len_data; ++j) {
    //multi_t iny;
    swpack swz[1]; /* workaround to generate swarctrl automatically */
//    swpack sa[1],sb[1],sz[1]; /* workaround to generate swarctrl automatically */
    unsigned pi = j%PACKING;
    unsigned pw = j/PACKING;
    // reset accumulator
    int acc = 0;
    // loop 1:((len_fir/PACKING)+1)
    for (unsigned w=0;w<EXTCOEF_LEN;++w) {
      // swar mul xi,bi => accum  [bi=select coefs with correct alignment]
      swz[0] = in_x[pw+w] * swar_ecoefs[pi][w];
      acc += *((int *)&swz);
    }
    // save accum to yi
    out_y[j] = acc;
  }
}

/* -------------------------------------------------------------------------- */
// use swar multiplication with lane accumulators
void fir_swar_mul_accum(resultelm *out_y, swpack *in_x, int len_data) //, swpack *in_c)
{
  for (unsigned j=0; j<len_data; ++j) {
    swpack swz[1];
    unsigned pi = j%PACKING;
    unsigned pw = j/PACKING;
    // internal accumulators are reset automatically with settings swar control word
    for (unsigned w=0;w<EXTCOEF_LEN;++w) {
      swz[0] = in_x[pw+w] * swar_ecoefs[pi][w];
    }
    resultelm sum = 0;
    for (unsigned a=0;a<PACKING;++a) {
      sum += __builtin_swaraccum(a<<1);
    }
    out_y[j] = sum;
  }
}

/* -------------------------------------------------------------------------- */
// use multiplication without lane accumulators or reduction
swpack swar_tmp_mem[EXTCOEF_LEN]; /* space for internal subtotals */

/* this version needs elements with sufficient bitwidth - saturation is more possible */
void fir_swar_mul_add(resultelm *out_y, swpack *in_x, int len_data) //, swpack *in_c)
{
  #pragma swar saturate
  for (unsigned j=0; j<len_data; ++j) {
    unsigned pi = j%PACKING;
    unsigned pw = j/PACKING;
    // tmp[i] = x[n-i]*c[i]
    for (unsigned w=0;w<EXTCOEF_LEN;++w) {
      swar_tmp_mem[w] = in_x[pw+w] * swar_ecoefs[pi][w];
    }
    // sum in packed data
    swpack sum;
    sum = swar_tmp_mem[0];
    for (unsigned w=1;w<EXTCOEF_LEN;++w) {
      swpack tmp = swar_tmp_mem[w];
      sum = sum + tmp;
    }
    // sum of packed data
    resultelm out = 0;
    for (unsigned p=0;p<PACKING;++p) {
      dataelm oval = (((*(basetype*)&sum) >> (p*BITSZ)) & MAXVAL);
      out += oval;
    }
    out_y[j] = out;
  }
}

/* -------------------------------------------------------------------------- */
swpack swar_x_mem[MAXFIRLENGTH/PACKING+1]; /* space for the latest data vector */
swpack swar_c_mem[MAXFIRLENGTH/PACKING+1]; /* space for the coefficients (for testing another alignment) */
unsigned swar_mem_coefwrds = 0;

void fir_swar_online_init(unsigned firlen, const dataelm *in_c)
{
  if (firlen>MAXFIRLENGTH) {
    printf("!!! Too long FIR filter.\n");
    firlen = MAXFIRLENGTH;
  }
  unsigned cw = ((firlen+PACKING-1)/PACKING);
  for (unsigned w=0;w<cw;++w) {
    *(resultelm *)&swar_c_mem[w] = 0; // swar_in_c[w];
    *(resultelm *)&swar_x_mem[w] = 0;
  }
  for (unsigned c=0;c<firlen;++c) {
    resultelm w = c / PACKING;
    resultelm e = c % PACKING;
    swar_c_mem[w][e] = in_c[c];
//printf(" !c %u (%d) w=%u/e=%u -> x%lX\n",c,in_c[c],w,e, *(resultelm *)&swar_c_mem[w]);
  }

#if 0
  // debug print coefficients
  print_swar_data("coeffs ", swar_c_mem, firlen);
#endif

  swar_mem_coefwrds = cw;
}

resultelm fir_swar_online(dataelm in_x)
{
  resultelm y = 0;
  // update data vector
  unsigned rem = in_x & MAXVAL;
  for(unsigned w = 0;w<swar_mem_coefwrds;++w) {
//printf("UPx #%u: from x%lX\n", w, *((resultelm *)&swar_x_mem[w]));
    resultelm new = (*((resultelm *)&swar_x_mem[w]) << BITSZ) | (rem & MAXVAL);
    rem = *((resultelm *)&swar_x_mem[w]) >> (SWARBASESIZE-BITSZ);
    swar_x_mem[w] = new;
//printf("        to x%lX\n", *((resultelm *)&swar_x_mem[w]));
  }
  // compute result for the latest data vector
  #pragma swar manual
//#pragma swar reduce
  __builtin_swarctrl(SWTP | SW_SGND | SW_RED | SW_OP_MUL);
  for(unsigned w=0;w<swar_mem_coefwrds;++w) {
//printf("SO#%u:x=x%lX c=x%lX\n", w, *(basetype *)&swar_x_mem[w], *(basetype *)&swar_c_mem[w]);
    swpack yc = __builtin_swar(swar_x_mem[w], swar_c_mem[w]);
//printf(" ----------- YC = x%lx\n", *(basetype *)&yc);
    y += *(resultelm *)&yc;
//printf(" ----------- Y = x%lx\n", *(basetype *)&y);
  }
//printf(" =============== Y = x%lx\n", *(basetype *)&y);
  return y;
}

/* ---------------------------------- */
// 
swpack swar_c_arr[PACKING][MAXFIRLENGTH/PACKING+1]; /* space for aligned vectors of coefficients */
unsigned swar_ca_len;
unsigned swar_pidx, swar_widx;

void fir_swar_online_rbuf_init(unsigned firlen, const dataelm *in_c)
{
  swar_ca_len = (firlen + PACKING-1 + PACKING-1) / PACKING; // pocet slov
//m_print("flen="); i_print(firlen); m_print("s_ca_l="); i_print(swar_ca_len); NL;
  swar_pidx = 0;
  swar_widx = 0;
  // prepare swar_c_arr
  // reset swar_x_mem
  for (unsigned w=0; w<swar_ca_len; ++w) {
    *(resultelm *)&swar_x_mem[w] = 0;
    for (unsigned p; p<PACKING; ++p)
      *(resultelm *)&swar_c_arr[p][w] = 0;
  }
  for (unsigned p=0; p<PACKING; ++p) {
    for (unsigned c=0; c<firlen; ++c) {
      unsigned w = (c+PACKING-p-1) / PACKING;
      unsigned e = (c+PACKING-p-1) % PACKING;
//m_print("cc p="); i_print(p); m_print(" w="); i_print(w); m_print(" e="); i_print(e); NL;
      swar_c_arr[p][w][e] = in_c[c];
    }
  }
//  print_dir_data("in-c", in_c, firlen);
//  for(unsigned p=0;p<PACKING;++p) {
//    m_print(" #"); i_print(p); m_print(": ");
//    print_swar_data("ac0", &swar_c_arr[p][0], swar_ca_len*PACKING);
//  }
//  print_swar_data("x0", &swar_x_mem[0], swar_ca_len*PACKING);
}

resultelm fir_swar_online_rbuf(dataelm in_x)
{
  #pragma swar manual
  resultelm y = 0;
//m_print("<<< x = "); i_print(in_x); NL;
  // update X rbuf with a new value
  if (swar_pidx==0) *((unsigned *)&swar_x_mem[swar_widx]) = 0;
  swar_x_mem[swar_widx][PACKING-swar_pidx-1] = in_x;
  unsigned wi = swar_widx;
  __builtin_swarctrl(SWTP | SW_SGND | SW_RED | SW_OP_MUL);
  for (unsigned w=0;w<swar_ca_len;++w) {
//m_print("#"); i_print(w); m_print("(p:"); i_print(swar_pidx); m_print(",w:"); i_print(swar_widx); m_print(")");
//m_print(" => x:@"); i_print(wi); m_print("="); i_hexprint(swar_x_mem[w]);
//m_print(" x c:@"); i_print(swar_pidx); m_print(","); i_print(w); m_print("="); i_hexprint(swar_c_arr[swar_pidx][w]); NL;
    swpack ysw = __builtin_swar(swar_x_mem[wi],swar_c_arr[swar_pidx][w]); /* !!! */
    y += *(resultelm *)&ysw;
    if (++wi>=swar_ca_len) wi=0;
  }

  if (++swar_pidx>=PACKING) {
    swar_pidx=0;
    if (swar_widx==0) 
      swar_widx=swar_ca_len-1;
    else
      swar_widx--;
  }
//m_print(">>> y = "); i_print(y); NL;
  return y;
}

/* ---------------------------------- */
// 
unsigned swar_ca_len;
unsigned swar_pidx, swar_widx;
unsigned swar_xbufsize;
unsigned swar_xbufmask;

void fir_swar_online_align_rbuf_init(unsigned firlen, const dataelm *in_c)
{
  swar_ca_len = (firlen + PACKING-1 + PACKING-1) / PACKING; // minimal number of swar words
  unsigned mbspt = SWARBASESIZE - __builtin_clz(swar_ca_len);
  swar_xbufsize = 1<<mbspt;
  swar_xbufmask = swar_xbufsize-1;
//m_print("flen="); i_print(firlen); m_print("s_ca_l="); i_print(swar_ca_len); NL;
  swar_pidx = 0;
  swar_widx = 0;
  // prepare swar_c_arr
  // reset swar_x_mem
  for (unsigned w=0; w<swar_ca_len; ++w) {
    for (unsigned p; p<PACKING; ++p)
      *(resultelm *)&swar_c_arr[p][w] = 0;
  }
  for (unsigned w=0;w<swar_xbufsize;++w)
    *(resultelm *)&swar_x_mem[w] = 0;

  for (unsigned p=0; p<PACKING; ++p) {
    for (unsigned c=0; c<firlen; ++c) {
      unsigned w = (c+PACKING-p-1) / PACKING;
      unsigned e = (c+PACKING-p-1) % PACKING;
//m_print("cc p="); i_print(p); m_print(" w="); i_print(w); m_print(" e="); i_print(e); NL;
      swar_c_arr[p][w][e] = in_c[c];
    }
  }
//  print_dir_data("in-c", in_c, firlen);
//  for(unsigned p=0;p<PACKING;++p) {
//    m_print(" #"); i_print(p); m_print(": ");
//    print_swar_data("ac0", &swar_c_arr[p][0], swar_ca_len*PACKING);
//  }
//  print_swar_data("x0", &swar_x_mem[0], swar_ca_len*PACKING);
}


resultelm fir_swar_online_align_rbuf(dataelm in_x)
{
  resultelm y = 0;
 
//m_print("<<< x = "); i_print(in_x); NL;
  // update X rbuf with a new value
  if (swar_pidx==0) *((unsigned *)&swar_x_mem[swar_widx]) = 0;
  swar_x_mem[swar_widx][PACKING-swar_pidx-1] = in_x;
  unsigned wi = swar_widx;

  #pragma swar manual

  __builtin_swarctrl(SWTP | SW_SGND | SW_RED | SW_OP_MUL);
  for (unsigned w=0;w<swar_ca_len;++w) {
//m_print("#"); i_print(w); m_print("(p:"); i_print(swar_pidx); m_print(",w:"); i_print(swar_widx); m_print(")");
//m_print(" => x:@"); i_print(wi); m_print("(");i_hexprint(&swar_x_mem[wi]); m_print(") ="); i_hexprint(swar_x_mem[wi]);
//m_print(" x c:@"); i_print(swar_pidx); m_print(","); i_print(w); m_print("="); i_hexprint(swar_c_arr[swar_pidx][w]); NL;
#if 0
    volatile unsigned yc= __builtin_swar(swar_x_mem[wi],swar_c_arr[swar_pidx][w]);
    y+=yc;
//m_print("  >>> yc = "); i_print(yc); NL;
#else
    swpack ysw = __builtin_swar(swar_x_mem[wi],swar_c_arr[swar_pidx][w]);
    y += *(resultelm *)&ysw;
#endif
    wi = (wi + 1) & swar_xbufmask;
  }

  if (++swar_pidx>=PACKING) {
    swar_pidx=0;
    swar_widx = (swar_widx - 1) & swar_xbufmask;
  }
//m_print(">>> y = "); i_print(y); NL;
  return y;
}



/* -------------------------------------------------------------------------- */
/* -------------------------------------------------------------------------- */
/* comparing */
int check_results(unsigned len, resultelm *in, resultelm *sw)
{
  int res = 0;
  printf("Check results: ");
  for (unsigned p=0;p<len;++p) {
    if (in[p]!=sw[p]) {
#if PRINT_FAILED
      printf("\n failed pack #%u (%lu @%p , %lu @%p)", p, in[p], &in[p], sw[p], &sw[p]);
#endif /* PRINT_FAILED */
      res++;
    }
  }
  if (!res) {
    printf("  OK\n\n");
  } else { 
    printf(" FAILED %u\n\n", res);
  }
  return res;
}


/* print data */
void print_dir_data(const char *prefix, dataelm *pd, int len)
{
  printf("%s data (%lu):\n", prefix, sizeof(dataelm));
  for(int i=0;i<len;++i) {
    printf("  %d %u(x%X)", *pd, *pd, *pd);
    pd++;
  }
  printf("\n\n");
}

void print_swar_data(const char *prefix, swpack *ps, int packlen)
{
  swpack *ptmp = ps;
  printf("%s swar data:\n", prefix);
  // print packed swar data
  printf("  PACK:");
  for(unsigned i=0;i<((packlen+PACKING-1)/PACKING);++i) {
    printf("  x" PRNFMTHEX, *((resultelm *)ps) );
    ps++;
  }
  printf("\n  ELM: ");
  // print swar element data
  ps = ptmp;
  for(unsigned i=0;i<packlen;++i) {
    multi_t v; v.s = *(ps+i/PACKING);
    int ii = i % PACKING;
//    dataelm e = ((dataelm *)&v)[ii]; /* !!! TODO: e=v[ii] */
    resultelm e = (v.u>>(BITSZ*ii))&((1<<BITSZ)-1);
    printf("  x" PRNFMTHEX, e);
  }
  printf("\n");
}

/* print results */
void print_results(const char *prefix, resultelm *pd, int len)
{
  printf("%s data @%p:\n", prefix, pd);
  for(int i=0;i<len;++i) {
    printf("  " PRNFMTDEC, *pd++);
  }
  printf("\n\n");
}

/* print counters */
void print_counters(const char *prefix, unsigned *tm, unsigned *ins)
{
  unsigned ticks_lo, ticks_hi, insns_lo, insns_hi;
  sysregs_stop();
  sysregs_read(&ticks_lo, &ticks_hi, &insns_lo, &insns_hi);

  // use relative cycles (between cycle_start and cycle_stop)
  unsigned long dcyc = cycle_stop-cycle_start;
  ticks_hi = (unsigned)(dcyc>>32);
  ticks_lo = (unsigned)(dcyc);

  printf("%s", prefix);
  if (ticks_hi) {
    printf(" EXECUTED IN (%u * 2^32 + %u) TICKS AND (%u * 2^32 + %u) INSTRUCTIONS\n\n", ticks_hi, ticks_lo, insns_hi, insns_lo);
  } else {
    printf(" EXECUTED IN %u TICKS AND %u INSTRUCTIONS\n", ticks_lo, insns_lo);
  }
  if (tm) *tm = ticks_lo;
  if (ins) *ins = insns_lo;
}


// main
int main(void)
{
  system_init();

  printf("TEST: SWARFIR\n");
#if 1
  printf(" - info:\n");
#if PRINT_TYPES
  printf("  - Types\n");
  printf("   - sizeof(unsigned)  : %lu B\n", sizeof(unsigned));
  printf("   - sizeof(uint32_t)  : %lu B\n", sizeof(uint32_t));
  printf("   - sizeof(uint64_t)  : %lu B\n", sizeof(uint64_t));
  printf("   - sizeof(swpack)    : %lu B\n", sizeof(swpack));
  printf("   - sizeof(swelm)     : %lu B\n", sizeof(swelm));
  printf("   - sizeof(dataelm)   : %lu B\n", sizeof(dataelm));
  printf("   - sizeof(resultelm) : %lu b\n", sizeof(resultelm));
  printf("   - sizeofswar(swpack): %lu b\n", sizeofswar(swpack));
  printf("   - sizeofswar(swelm) : %lu b\n", sizeofswar(swelm));
#endif /* PRINT_TYPES */
  printf("  - data length  : %u\n", DATA_LENGTH);
  printf("  - filter length: %u\n", FIR_LENGTH-1);
  printf("  - memory for direct FIR [Bytes]:\n");
  printf("   - input data  : %lu\n", sizeof(dir_in_x));
  printf("   - (max.number of input elements) : %lu)\n", TEST_INDATA_SIZE);
  printf("   - coefficients: %lu\n", sizeof(dir_in_c));
  printf("   - output data : %lu\n", sizeof(dir_out_y));
  printf("   - (max.number of output elements) : %lu)\n", TEST_OUTPUT_SIZE);
  printf("  - memory for SWAR FIR [Bytes]:\n");
  printf("   - input data  : %lu\n", sizeof(swar_in_x));
  printf("   - coefficients: %lu\n", sizeof(swar_in_c));
  printf("   - output data : %lu\n", sizeof(swar_out_y_add));
  printf("  - position in memory:\n");
  printf("   - dir_out_Y       : %p\n", &dir_out_y[0]);
  printf("   - swar_out_Y(add) : %p\n", &swar_out_y_add[0]);
#endif

  printf(" - preparing data\n");
  sysregs_init();
  prepare_data();
  print_counters("  ", NULL, NULL);

#if TEST_OFFLINE

#if 0
  print_dir_data("IN-X", dir_in_x, DATA_LENGTH);
  print_dir_data("IN-C", dir_in_c, FIR_LENGTH);
  print_swar_data("SWAR-IN-X", swar_in_x, DATA_LENGTH);
  print_swar_data("SWAR-IN-C", swar_in_c, FIR_LENGTH);
#endif

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

#if 1
  printf("  * FIR\n");
#if PRINT_RESULTS
  printf(" In-data:\n");
  for(unsigned u=0;u<DATA_LENGTH;++u)
    printf("  #%u : %u\n", u, dir_in_x[u]);
  printf("\n Coefs:\n");
  for(unsigned u=0;u<FIR_LENGTH;++u)
    printf("  #%u : %u\n", u, dir_in_c[u]);
  printf("\n");
#endif /* PRINT_RESULTS */
  sysregs_start();
  fir_offline(dir_out_y, dir_in_x, DATA_LENGTH, dir_in_c);
  print_counters("  ", NULL, NULL);
  #if PRINT_RESULTS
    print_results("OUT-Y", dir_out_y, DATA_LENGTH);
  #endif /* PRINT_RESULTS */
#endif

#if 1
  printf("  * SWAR FIR (mul,add)\n");
  sysregs_start();
  fir_swar_mul_add(swar_out_y_add, swar_in_x, DATA_LENGTH); //, swar_in_c);
  print_counters("  ", NULL, NULL);
  #if PRINT_RESULTS
    print_results("SWAR(add) Y", swar_out_y_add, DATA_LENGTH);
  #endif /* PRINT_RESULTS */

  check_results(DATA_LENGTH, dir_out_y, swar_out_y_add);
#endif

#if 1
  printf("  * SWAR FIR (reduction)\n");
  sysregs_start();
  fir_swar_mul_red(swar_out_y_red, swar_in_x, DATA_LENGTH); //, swar_in_c);
  print_counters("  ", NULL, NULL);
  #if PRINT_RESULTS
    print_results("SWAR(red) Y", swar_out_y_red, DATA_LENGTH);
  #endif /* PRINT_RESULTS */

  check_results(DATA_LENGTH, dir_out_y, swar_out_y_red);
#endif

#if 1
  printf("  * SWAR FIR (accumulators)\n");
  sysregs_start();
  fir_swar_mul_accum(swar_out_y_accum, swar_in_x, DATA_LENGTH); //, swar_in_c);
  print_counters("  ", NULL, NULL);
  #if PRINT_RESULTS
    print_results("SWAR(acc) Y", swar_out_y_accum, DATA_LENGTH);
  #endif /* PRINT_RESULTS */

  check_results(DATA_LENGTH, dir_out_y, swar_out_y_accum);
#endif

#endif /* TEST_OFFLINE */


/* ---------------------------------- */
/* ---------------------------------- */
/* ---------------------------------- */


#if TEST_ONLINE
  printf("\n - Compute online\n");

#if 1
  printf("  * FIR online\n");
  fir_online_init(FIR_LENGTH, dir_in_c);
  sysregs_start();
  for (unsigned i=0;i<DATA_LENGTH;++i) {
//printf("### %u : %u x%X", i, dir_in_x[i], dir_in_x[i]);
    dir_out_y[i] = fir_online(dir_in_x[i]);
//printf(" ->Y %u x%X\n", dir_out_y[i], dir_out_y[i]);
  }
  print_counters("  ", NULL, NULL);
  #if PRINT_RESULTS
    print_results("OUT-Y", dir_out_y, DATA_LENGTH);
  #endif /* PRINT_RESULTS */
#endif

#if 1
  printf("  * FIR swar online\n");
  fir_swar_online_init(FIR_LENGTH, dir_in_c);
  sysregs_start();
  for (unsigned i=0;i<DATA_LENGTH;++i) {
//printf("SO#%u : %u (x%X)\n", i, dir_in_x[i], dir_in_x[i]);
    swar_out_y_red[i] = fir_swar_online(dir_in_x[i]);
  }
  print_counters("  ", NULL, NULL);
  #if PRINT_RESULTS
    print_results("SWAR(online)", swar_out_y_red, DATA_LENGTH);
  #endif /* PRINT_RESULTS */

  check_results(DATA_LENGTH, dir_out_y, swar_out_y_red);
#endif


/* ---------------------------------- */
/* low pass FIR filter */
  printf("\nExample: on-line low pass FIR filtering\n");
  unsigned times[10], insts[10];
  unsigned nres = 0;

  unsigned id = 0;
  unsigned len = (TEST_INDATA_SIZE<TEST_OUTPUT_SIZE) ? TEST_INDATA_SIZE : TEST_OUTPUT_SIZE;
  if (ONLINE_DATALEN<len) len = ONLINE_DATALEN;

  printf("\n   - FIR length (# coeffs) = %lu\n", ONLINE_FIRLEN); /* sizeof(test_fircoef)/sizeof(dataelm) */
  printf("   - Data length (samples) = %u\n", len);

#if 1
  printf("  * FIR online\n");
  fir_online_init(ONLINE_FIRLEN, test_fircoef_extra); //test_fircoef);
//print_dir_data("-c-", test_fircoef_extra, ONLINE_FIRLEN);
  id = 0;
  sysregs_start();
  for (unsigned i=0;i<len;++i) {
    if (id>=len) id -= len;
//printf("#%u (id=%u: %d ", i, id, test_input[id]);
    test_dir_output[id] = fir_online(test_input[id]);
//printf("   -> %d\n", test_dir_output[id]);
    id++;
  }
  print_counters("  ", &times[0], &insts[0]);
  #if PRINT_RESULTS
    print_results("OUTPUT", test_dir_output, len);
  #endif /* PRINT_RESULTS */
//  check_results(len, test_dir_output, test_output);
#endif

#if 1
  printf("  * FIR swar online\n");
//print_dir_data("### -c-", test_fircoef_extra, ONLINE_FIRLEN);
  fir_swar_online_init(ONLINE_FIRLEN, test_fircoef_extra); //test_fircoef);
  id = 0;
  sysregs_start();
  for (unsigned i=0;i<len;++i) {
    if (id>=len) id -= len;
//printf("###%u: i= %d \n", i, test_input[id]);
    test_swar_output[id] = fir_swar_online(test_input[id]);
//printf("     -> o= %d (x%lx)\n", test_swar_output[id], test_swar_output[id]);
    id++;
  }
  print_counters("  ", &times[1], &insts[1]);
  #if PRINT_RESULTS
    print_results("OUTPUT", test_swar_output, len);
  #endif /* PRINT_RESULTS */
//  check_results(len, test_swar_output, test_output);
  check_results(len, test_swar_output, test_dir_output);
#endif

/* ---------------------------------- */
/* round buffer */

#if 1
  printf("  * FIR online rbuf\n");
  fir_online_rbuf_init(ONLINE_FIRLEN, test_fircoef_extra); //test_fircoef);
//print_dir_data("-c-", test_fircoef_extra, ONLINE_FIRLEN);
  id = 0;
  sysregs_start();
  for (unsigned i=0;i<len;++i) {
    if (id>=len) id -= len;
    //test_dir_output[id] 
//  m_print("#"); i_print(i); m_print(":"); i_print(test_input[id]);
    test_swar_output[id] = fir_online_rbuf(test_input[id]);
//  m_print(" -> "); i_print(test_swar_output[id]); NL;
    id++;
  }
  print_counters("  ", &times[2], &insts[2]);
  #if PRINT_RESULTS
    print_results("OUTPUT", test_swar_output, len);
  #endif /* PRINT_RESULTS */
//  check_results(len, test_dir_output, test_output);
  check_results(len, test_dir_output, test_swar_output);
#endif

#if 1
  printf("  * FIR swar online rbuf\n");
  fir_swar_online_rbuf_init(ONLINE_FIRLEN, test_fircoef_extra); // test_fircoef);
//print_dir_data("-c-", test_fircoef_extra, ONLINE_FIRLEN);
  id = 0;
  sysregs_start();
  for (unsigned i=0;i<len;++i) {
    if (id>=len) id -= len;
//  m_print("#"); i_print(i); m_print(":"); i_print(test_input[id]);
    test_swar_output[id] = fir_swar_online_rbuf(test_input[id]);
//  m_print(" -> "); i_print(test_swar_output[id]); NL;
    id++;
  }
  print_counters("  ", &times[3], &insts[3]);
  #if PRINT_RESULTS
    print_results("OUTPUT", test_swar_output, len);
  #endif /* PRINT_RESULTS */
//  check_results(len, test_dir_output, test_output);
  check_results(len, test_dir_output, test_swar_output);
#endif

/* ---------------------------------- */
/* aligned round buffer */

#if 1
  printf("  * FIR online aligned rbuf\n");
  fir_online_align_rbuf_init(ONLINE_FIRLEN, test_fircoef_extra); //test_fircoef);
//print_dir_data("-c-", test_fircoef_extra, ONLINE_FIRLEN);
  id = 0;
  sysregs_start();
  for (unsigned i=0;i<len;++i) {
    if (id>=len) id -= len;
    //test_dir_output[id] 
//printf("#%u (id=%u): %d x%lX\n", i, id, test_input[id], test_input[id]);
    test_swar_output[id] = fir_online_align_rbuf(test_input[id]);
//printf(" #-> %d x%lX\n", test_swar_output[id], test_swar_output[id]);
    id++;
  }
  print_counters("  ", &times[4], &insts[4]);
  #if PRINT_RESULTS
    print_results("OUTPUT", test_swar_output, len);
  #endif /* PRINT_RESULTS */
//  check_results(len, test_dir_output, test_output);
  check_results(len, test_dir_output, test_swar_output);
#endif

#if 1
  printf("  * FIR swar online aligned rbuf\n");
  fir_swar_online_align_rbuf_init(ONLINE_FIRLEN, test_fircoef_extra); // test_fircoef);
//print_dir_data("-c-", test_fircoef_extra, ONLINE_FIRLEN);
  id = 0;
  sysregs_start();
  for (unsigned i=0;i<len;++i) {
    if (id>=len) id -= len;
//  m_print("#"); i_print(i); m_print(":"); i_print(test_input[id]);
    test_swar_output[id] = fir_swar_online_align_rbuf(test_input[id]);
//  m_print(" -> "); i_print(test_swar_output[id]); NL;
    id++;
  }
  print_counters("  ", &times[5], &insts[5]);
  #if PRINT_RESULTS
    print_results("OUTPUT", test_swar_output, len);
  #endif /* PRINT_RESULTS */
//  check_results(len, test_dir_output, test_output);
  check_results(len, test_dir_output, test_swar_output);
#endif

/* ---------------------------------- */

#endif /* TEST_ONLINE */

  printf("\n*** END OF TEST ***\n");
  
  system_done();

  return 0;
}