1672 lines
43 KiB
C++
1672 lines
43 KiB
C++
/*
|
|
* Copyright (c) 1998-2021 Stephen Williams (steve@icarus.com)
|
|
*
|
|
* This source code is free software; you can redistribute it
|
|
* and/or modify it in source code form under the terms of the GNU
|
|
* General Public License as published by the Free Software
|
|
* Foundation; either version 2 of the License, or (at your option)
|
|
* any later version.
|
|
*
|
|
* This program is distributed in the hope that it will be useful,
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
* GNU General Public License for more details.
|
|
*
|
|
* You should have received a copy of the GNU General Public License
|
|
* along with this program; if not, write to the Free Software
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
|
|
*/
|
|
|
|
# include "config.h"
|
|
|
|
# include "verinum.h"
|
|
# include <iostream>
|
|
# include <cassert>
|
|
# include <climits>
|
|
# include <cmath> // Needed to get pow for as_double().
|
|
# include <cstdio> // Needed to get snprintf for as_string().
|
|
# include <algorithm>
|
|
|
|
using namespace std;
|
|
|
|
#if !defined(HAVE_LROUND)
|
|
/*
|
|
* If the system doesn't provide the lround function, then we provide
|
|
* it ourselves here. It is simply the nearest integer, rounded away
|
|
* from zero.
|
|
*/
|
|
extern "C" long int lround(double x)
|
|
{
|
|
if (x >= 0.0)
|
|
return (long)floor(x+0.5);
|
|
else
|
|
return (long)ceil(x-0.5);
|
|
}
|
|
#endif
|
|
|
|
static verinum::V add_with_carry(verinum::V l, verinum::V r, verinum::V&c);
|
|
|
|
verinum::verinum()
|
|
: bits_(0), nbits_(0), has_len_(false), has_sign_(false), is_single_(false), string_flag_(false)
|
|
{
|
|
}
|
|
|
|
verinum::verinum(const V*bits, unsigned nbits, bool has_len__)
|
|
: has_len_(has_len__), has_sign_(false), is_single_(false), string_flag_(false)
|
|
{
|
|
nbits_ = nbits;
|
|
bits_ = new V [nbits];
|
|
for (unsigned idx = 0 ; idx < nbits ; idx += 1) {
|
|
bits_[idx] = bits[idx];
|
|
}
|
|
}
|
|
|
|
static string process_verilog_string_quotes(const string&str)
|
|
{
|
|
string res;
|
|
|
|
int idx = 0;
|
|
int str_len = str.length();
|
|
|
|
while (idx < str_len) {
|
|
if (str[idx] == '\\') {
|
|
idx += 1;
|
|
assert(idx < str_len);
|
|
switch (str[idx]) {
|
|
case 'n':
|
|
res = res + '\n';
|
|
idx += 1;
|
|
break;
|
|
case 't':
|
|
res = res + '\t';
|
|
idx += 1;
|
|
break;
|
|
case 'v':
|
|
res = res + '\v';
|
|
idx += 1;
|
|
break;
|
|
case 'f':
|
|
res = res + '\f';
|
|
idx += 1;
|
|
break;
|
|
case 'a':
|
|
res = res + '\a';
|
|
idx += 1;
|
|
break;
|
|
case '0':
|
|
case '1':
|
|
case '2':
|
|
case '3':
|
|
case '4':
|
|
case '5':
|
|
case '6':
|
|
case '7': {
|
|
char byte_val = 0;
|
|
int odx = 0;
|
|
while (odx < 3 && idx+odx < str_len
|
|
&& str[idx+odx] >= '0'
|
|
&& str[idx+odx] <= '7') {
|
|
byte_val = 8*byte_val + str[idx+odx]-'0';
|
|
odx += 1;
|
|
}
|
|
idx += odx;
|
|
res = res + byte_val;
|
|
break;
|
|
}
|
|
case 'x': {
|
|
char byte_val = 0;
|
|
int odx = 1;
|
|
while (odx < 3 && idx+odx < str_len) {
|
|
if (str[idx+odx] >= '0' && str[idx+odx] <= '9') {
|
|
byte_val = 16*byte_val + str[idx+odx]-'0';
|
|
odx += 1;
|
|
} else if (str[idx+odx] >= 'a' && str[idx+odx] <= 'f') {
|
|
byte_val = 16*byte_val + str[idx+odx]-'a'+10;
|
|
odx += 1;
|
|
} else if (str[idx+odx] >= 'A' && str[idx+odx] <= 'F') {
|
|
byte_val = 16*byte_val + str[idx+odx]-'A'+10;
|
|
odx += 1;
|
|
} else {
|
|
break;
|
|
}
|
|
}
|
|
idx += odx;
|
|
res = res + byte_val;
|
|
break;
|
|
}
|
|
default:
|
|
res = res + str[idx];
|
|
idx += 1;
|
|
break;
|
|
}
|
|
} else {
|
|
res = res + str[idx];
|
|
idx += 1;
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
verinum::verinum(const string&s)
|
|
: has_len_(true), has_sign_(false), is_single_(false), string_flag_(true)
|
|
{
|
|
string str = process_verilog_string_quotes(s);
|
|
nbits_ = str.length() * 8;
|
|
|
|
// Special case: The string "" is 8 bits of 0.
|
|
if (nbits_ == 0) {
|
|
nbits_ = 8;
|
|
bits_ = new V [nbits_];
|
|
bits_[0] = V0;
|
|
bits_[1] = V0;
|
|
bits_[2] = V0;
|
|
bits_[3] = V0;
|
|
bits_[4] = V0;
|
|
bits_[5] = V0;
|
|
bits_[6] = V0;
|
|
bits_[7] = V0;
|
|
return;
|
|
}
|
|
|
|
bits_ = new V [nbits_];
|
|
|
|
unsigned idx, cp;
|
|
V*bp = bits_+nbits_;
|
|
for (idx = nbits_, cp = 0 ; idx > 0 ; idx -= 8, cp += 1) {
|
|
char ch = str[cp];
|
|
*(--bp) = (ch&0x80) ? V1 : V0;
|
|
*(--bp) = (ch&0x40) ? V1 : V0;
|
|
*(--bp) = (ch&0x20) ? V1 : V0;
|
|
*(--bp) = (ch&0x10) ? V1 : V0;
|
|
*(--bp) = (ch&0x08) ? V1 : V0;
|
|
*(--bp) = (ch&0x04) ? V1 : V0;
|
|
*(--bp) = (ch&0x02) ? V1 : V0;
|
|
*(--bp) = (ch&0x01) ? V1 : V0;
|
|
}
|
|
}
|
|
|
|
verinum::verinum(verinum::V val, unsigned n, bool h)
|
|
: has_len_(h), has_sign_(false), is_single_(false), string_flag_(false)
|
|
{
|
|
nbits_ = n;
|
|
bits_ = new V[nbits_];
|
|
for (unsigned idx = 0 ; idx < nbits_ ; idx += 1)
|
|
bits_[idx] = val;
|
|
}
|
|
|
|
verinum::verinum(uint64_t val, unsigned n)
|
|
: has_len_(true), has_sign_(false), is_single_(false), string_flag_(false)
|
|
{
|
|
nbits_ = n;
|
|
bits_ = new V[nbits_];
|
|
for (unsigned idx = 0 ; idx < nbits_ ; idx += 1) {
|
|
bits_[idx] = (val&1) ? V1 : V0;
|
|
val >>= (uint64_t)1;
|
|
}
|
|
}
|
|
|
|
/* The second argument is not used! It is there to make this
|
|
* constructor unique. */
|
|
verinum::verinum(double val, bool)
|
|
: has_len_(false), has_sign_(true), is_single_(false), string_flag_(false)
|
|
{
|
|
bool is_neg = false;
|
|
double fraction;
|
|
int exponent;
|
|
const unsigned BITS_IN_LONG = 8*sizeof(long);
|
|
|
|
/* We return `bx for a NaN or +/- infinity. */
|
|
if (val != val || (val && (val == 0.5*val))) {
|
|
nbits_ = 1;
|
|
bits_ = new V[nbits_];
|
|
bits_[0] = Vx;
|
|
return;
|
|
}
|
|
|
|
/* Convert to a positive result. */
|
|
if (val < 0.0) {
|
|
is_neg = true;
|
|
val = -val;
|
|
}
|
|
|
|
/* Round to the nearest integer now, as this may increase the
|
|
number of bits we need to allocate. */
|
|
val = round(val);
|
|
|
|
/* Get the exponent and fractional part of the number. */
|
|
fraction = frexp(val, &exponent);
|
|
nbits_ = exponent+1;
|
|
bits_ = new V[nbits_];
|
|
|
|
/* If the value is small enough just use lround(). */
|
|
if (nbits_ <= BITS_IN_LONG) {
|
|
long sval = lround(val);
|
|
if (is_neg) sval = -sval;
|
|
for (unsigned idx = 0; idx < nbits_; idx += 1) {
|
|
bits_[idx] = (sval&1) ? V1 : V0;
|
|
sval >>= 1;
|
|
}
|
|
/* Trim the result. */
|
|
signed_trim();
|
|
return;
|
|
}
|
|
|
|
unsigned nwords = (exponent-1)/BITS_IN_LONG;
|
|
|
|
fraction = ldexp(fraction, (exponent-1) % BITS_IN_LONG + 1);
|
|
|
|
if (nwords == 0) {
|
|
unsigned long bits = (unsigned long) fraction;
|
|
fraction = fraction - (double) bits;
|
|
for (unsigned idx = 0; idx < nbits_; idx += 1) {
|
|
bits_[idx] = (bits&1) ? V1 : V0;
|
|
bits >>= 1;
|
|
}
|
|
} else {
|
|
for (int wd = nwords; wd >= 0; wd -= 1) {
|
|
unsigned long bits = (unsigned long) fraction;
|
|
fraction = fraction - (double) bits;
|
|
unsigned max_idx = (wd+1)*BITS_IN_LONG;
|
|
if (max_idx > nbits_) max_idx = nbits_;
|
|
for (unsigned idx = wd*BITS_IN_LONG; idx < max_idx; idx += 1) {
|
|
bits_[idx] = (bits&1) ? V1 : V0;
|
|
bits >>= 1;
|
|
}
|
|
fraction = ldexp(fraction, BITS_IN_LONG);
|
|
}
|
|
}
|
|
|
|
/* Convert a negative number if needed. */
|
|
if (is_neg) {
|
|
*this = -(*this);
|
|
}
|
|
|
|
/* Trim the result. */
|
|
signed_trim();
|
|
}
|
|
|
|
|
|
/* This is used by the double constructor above. It is needed to remove
|
|
* extra sign bits that can occur when calculating a negative value. */
|
|
void verinum::signed_trim()
|
|
{
|
|
/* Do we have any extra digits? */
|
|
unsigned tlen = nbits_-1;
|
|
verinum::V sign = bits_[tlen];
|
|
while ((tlen > 0) && (bits_[tlen] == sign)) tlen -= 1;
|
|
|
|
/* tlen now points to the first digit that is not the sign.
|
|
* or bit 0. Set the length to include this bit and one proper
|
|
* sign bit if needed. */
|
|
if (bits_[tlen] != sign) tlen += 1;
|
|
tlen += 1;
|
|
|
|
/* Trim the bits if needed. */
|
|
if (tlen < nbits_) {
|
|
V* tbits = new V[tlen];
|
|
for (unsigned idx = 0; idx < tlen; idx += 1)
|
|
tbits[idx] = bits_[idx];
|
|
delete[] bits_;
|
|
bits_ = tbits;
|
|
nbits_ = tlen;
|
|
}
|
|
}
|
|
|
|
verinum::verinum(const verinum&that)
|
|
{
|
|
string_flag_ = that.string_flag_;
|
|
nbits_ = that.nbits_;
|
|
bits_ = new V[nbits_];
|
|
has_len_ = that.has_len_;
|
|
has_sign_ = that.has_sign_;
|
|
is_single_ = that.is_single_;
|
|
for (unsigned idx = 0 ; idx < nbits_ ; idx += 1)
|
|
bits_[idx] = that.bits_[idx];
|
|
}
|
|
|
|
verinum::verinum(const verinum&that, unsigned nbits)
|
|
{
|
|
string_flag_ = that.string_flag_ && (that.nbits_ == nbits);
|
|
nbits_ = nbits;
|
|
bits_ = new V[nbits_];
|
|
has_len_ = true;
|
|
has_sign_ = that.has_sign_;
|
|
is_single_ = false;
|
|
|
|
unsigned copy = nbits;
|
|
if (copy > that.nbits_)
|
|
copy = that.nbits_;
|
|
for (unsigned idx = 0 ; idx < copy ; idx += 1)
|
|
bits_[idx] = that.bits_[idx];
|
|
|
|
if (copy < nbits_) {
|
|
if (has_sign_ || that.is_single_) {
|
|
for (unsigned idx = copy ; idx < nbits_ ; idx += 1)
|
|
bits_[idx] = bits_[idx-1];
|
|
} else {
|
|
for (unsigned idx = copy ; idx < nbits_ ; idx += 1)
|
|
bits_[idx] = verinum::V0;
|
|
}
|
|
}
|
|
}
|
|
|
|
verinum::verinum(int64_t that)
|
|
: has_len_(false), has_sign_(true), is_single_(false), string_flag_(false)
|
|
{
|
|
int64_t tmp;
|
|
|
|
if (that < 0) tmp = (that+1)/2;
|
|
else tmp = that/2;
|
|
nbits_ = 1;
|
|
while (tmp != 0) {
|
|
nbits_ += 1;
|
|
tmp /= 2;
|
|
}
|
|
|
|
nbits_ += 1;
|
|
|
|
bits_ = new V[nbits_];
|
|
for (unsigned idx = 0 ; idx < nbits_ ; idx += 1) {
|
|
bits_[idx] = (that & 1)? V1 : V0;
|
|
that >>= 1;
|
|
}
|
|
}
|
|
|
|
verinum::~verinum()
|
|
{
|
|
delete[]bits_;
|
|
}
|
|
|
|
verinum& verinum::operator= (const verinum&that)
|
|
{
|
|
if (this == &that) return *this;
|
|
if (nbits_ != that.nbits_) {
|
|
delete[]bits_;
|
|
nbits_ = that.nbits_;
|
|
bits_ = new V[that.nbits_];
|
|
}
|
|
for (unsigned idx = 0 ; idx < nbits_ ; idx += 1)
|
|
bits_[idx] = that.bits_[idx];
|
|
|
|
has_len_ = that.has_len_;
|
|
has_sign_ = that.has_sign_;
|
|
is_single_ = that.is_single_;
|
|
string_flag_ = that.string_flag_;
|
|
return *this;
|
|
}
|
|
|
|
verinum::V verinum::get(unsigned idx) const
|
|
{
|
|
assert(idx < nbits_);
|
|
return bits_[idx];
|
|
}
|
|
|
|
verinum::V verinum::set(unsigned idx, verinum::V val)
|
|
{
|
|
assert(idx < nbits_);
|
|
return bits_[idx] = val;
|
|
}
|
|
|
|
void verinum::set(unsigned off, const verinum&val)
|
|
{
|
|
assert(off + val.len() <= nbits_);
|
|
for (unsigned idx = 0 ; idx < val.len() ; idx += 1)
|
|
bits_[off+idx] = val[idx];
|
|
}
|
|
|
|
unsigned verinum::as_unsigned() const
|
|
{
|
|
if (nbits_ == 0)
|
|
return 0;
|
|
|
|
if (!is_defined())
|
|
return 0;
|
|
|
|
unsigned val = 0;
|
|
unsigned mask = 1;
|
|
for (unsigned idx = 0 ; idx < nbits_ ; idx += 1, mask <<= 1)
|
|
if (bits_[idx] == V1) {
|
|
if (mask == 0) return ~mask;
|
|
val |= mask;
|
|
}
|
|
|
|
return val;
|
|
}
|
|
|
|
unsigned long verinum::as_ulong() const
|
|
{
|
|
if (nbits_ == 0)
|
|
return 0;
|
|
|
|
if (!is_defined())
|
|
return 0;
|
|
|
|
unsigned long val = 0;
|
|
unsigned long mask = 1;
|
|
for (unsigned idx = 0 ; idx < nbits_ ; idx += 1, mask <<= 1)
|
|
if (bits_[idx] == V1) {
|
|
if (mask == 0) return ~mask;
|
|
val |= mask;
|
|
}
|
|
|
|
return val;
|
|
}
|
|
|
|
uint64_t verinum::as_ulong64() const
|
|
{
|
|
if (nbits_ == 0)
|
|
return 0;
|
|
|
|
if (!is_defined())
|
|
return 0;
|
|
|
|
uint64_t val = 0;
|
|
uint64_t mask = 1;
|
|
for (unsigned idx = 0 ; idx < nbits_ ; idx += 1, mask <<= 1)
|
|
if (bits_[idx] == V1) {
|
|
if (mask == 0) return ~mask;
|
|
val |= mask;
|
|
}
|
|
|
|
return val;
|
|
}
|
|
|
|
/*
|
|
* This function returns the native long integer that represents the
|
|
* value of this object. It accounts for sign extension if the value
|
|
* is signed.
|
|
*
|
|
* If the value is undefined, return 0.
|
|
*
|
|
* This function presumes that the native format is 2s complement
|
|
* (pretty safe these days) and masks/sets bits accordingly. If the
|
|
* value is too large for the native form, it truncates the high bits.
|
|
*/
|
|
signed long verinum::as_long() const
|
|
{
|
|
#define IVLLBITS (8 * sizeof(long) - 1)
|
|
if (nbits_ == 0)
|
|
return 0;
|
|
|
|
if (!is_defined())
|
|
return 0;
|
|
|
|
signed long val = 0;
|
|
unsigned diag_top = 0;
|
|
|
|
unsigned top = nbits_;
|
|
if (top > IVLLBITS) {
|
|
diag_top = top;
|
|
top = IVLLBITS;
|
|
}
|
|
int lost_bits=0;
|
|
|
|
if (has_sign_ && (bits_[nbits_-1] == V1)) {
|
|
val = -1;
|
|
signed long mask = ~1L;
|
|
for (unsigned idx = 0 ; idx < top ; idx += 1) {
|
|
if (bits_[idx] == V0) val &= mask;
|
|
mask = (mask << 1) | 1L;
|
|
}
|
|
if (diag_top) {
|
|
for (unsigned idx = top; idx < diag_top; idx += 1) {
|
|
if (bits_[idx] == V0) lost_bits=1;
|
|
}
|
|
}
|
|
} else {
|
|
signed long mask = 1;
|
|
for (unsigned idx = 0 ; idx < top ; idx += 1, mask <<= 1) {
|
|
if (bits_[idx] == V1) val |= mask;
|
|
}
|
|
if (diag_top) {
|
|
for (unsigned idx = top; idx < diag_top; idx += 1) {
|
|
if (bits_[idx] == V1) lost_bits=1;
|
|
}
|
|
}
|
|
}
|
|
|
|
if (lost_bits) cerr << "warning: verinum::as_long() truncated " <<
|
|
diag_top << " bits to " << IVLLBITS << ", returns " << val << endl;
|
|
return val;
|
|
#undef IVLLBITS
|
|
}
|
|
|
|
double verinum::as_double() const
|
|
{
|
|
if (nbits_ == 0) return 0.0;
|
|
|
|
double val = 0.0;
|
|
/* Do we have/want a signed value? */
|
|
if (has_sign_ && bits_[nbits_-1] == V1) {
|
|
V carry = V1;
|
|
for (unsigned idx = 0; idx < nbits_; idx += 1) {
|
|
V sum = add_with_carry(~bits_[idx], V0, carry);
|
|
if (sum == V1)
|
|
val += pow(2.0, (double)idx);
|
|
}
|
|
val *= -1.0;
|
|
} else {
|
|
for (unsigned idx = 0; idx < nbits_; idx += 1) {
|
|
if (bits_[idx] == V1)
|
|
val += pow(2.0, (double)idx);
|
|
}
|
|
}
|
|
return val;
|
|
}
|
|
|
|
string verinum::as_string() const
|
|
{
|
|
assert( nbits_%8 == 0 );
|
|
if (nbits_ == 0)
|
|
return "";
|
|
|
|
string res;
|
|
for (unsigned idx = nbits_ ; idx > 0 ; idx -= 8) {
|
|
char char_val = 0;
|
|
V*bp = bits_+idx;
|
|
|
|
if (*(--bp) == V1) char_val |= 0x80;
|
|
if (*(--bp) == V1) char_val |= 0x40;
|
|
if (*(--bp) == V1) char_val |= 0x20;
|
|
if (*(--bp) == V1) char_val |= 0x10;
|
|
if (*(--bp) == V1) char_val |= 0x08;
|
|
if (*(--bp) == V1) char_val |= 0x04;
|
|
if (*(--bp) == V1) char_val |= 0x02;
|
|
if (*(--bp) == V1) char_val |= 0x01;
|
|
|
|
if (char_val == '"' || char_val == '\\') {
|
|
char tmp[5];
|
|
snprintf(tmp, sizeof tmp, "\\%03o", char_val);
|
|
res = res + tmp;
|
|
} else if (isprint(char_val)) {
|
|
res = res + char_val;
|
|
} else {
|
|
char tmp[5];
|
|
snprintf(tmp, sizeof tmp, "\\%03o", (unsigned char)char_val);
|
|
res = res + tmp;
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
bool verinum::is_before(const verinum&that) const
|
|
{
|
|
if (that.nbits_ > nbits_) return true;
|
|
if (that.nbits_ < nbits_) return false;
|
|
|
|
for (unsigned idx = nbits_ ; idx > 0 ; idx -= 1) {
|
|
if (bits_[idx-1] < that.bits_[idx-1]) return true;
|
|
if (bits_[idx-1] > that.bits_[idx-1]) return false;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
bool verinum::is_defined() const
|
|
{
|
|
for (unsigned idx = 0 ; idx < nbits_ ; idx += 1) {
|
|
if (bits_[idx] == Vx) return false;
|
|
if (bits_[idx] == Vz) return false;
|
|
}
|
|
return true;
|
|
}
|
|
|
|
bool verinum::is_zero() const
|
|
{
|
|
for (unsigned idx = 0 ; idx < nbits_ ; idx += 1)
|
|
if (bits_[idx] != V0) return false;
|
|
|
|
return true;
|
|
}
|
|
|
|
bool verinum::is_negative() const
|
|
{
|
|
return (bits_[nbits_-1] == V1) && has_sign();
|
|
}
|
|
|
|
unsigned verinum::significant_bits() const
|
|
{
|
|
unsigned sbits = nbits_;
|
|
|
|
if (has_sign_) {
|
|
V sgn_bit = bits_[sbits-1];
|
|
while ((sbits > 1) && (bits_[sbits-2] == sgn_bit))
|
|
sbits -= 1;
|
|
} else {
|
|
while ((sbits > 1) && (bits_[sbits-1] == verinum::V0))
|
|
sbits -= 1;
|
|
}
|
|
return sbits;
|
|
}
|
|
|
|
void verinum::cast_to_int2()
|
|
{
|
|
for (unsigned idx = 0 ; idx < nbits_ ; idx += 1) {
|
|
if (bits_[idx] == Vx || bits_[idx] == Vz)
|
|
bits_[idx] = V0;
|
|
}
|
|
}
|
|
|
|
verinum pad_to_width(const verinum&that, unsigned width)
|
|
{
|
|
if (that.len() >= width)
|
|
return that;
|
|
|
|
if (that.len() == 0) {
|
|
verinum val (verinum::V0, width, that.has_len());
|
|
val.has_sign(that.has_sign());
|
|
return val;
|
|
}
|
|
|
|
verinum::V pad = that[that.len()-1];
|
|
if (pad==verinum::V1 && !that.has_sign() && !that.is_single())
|
|
pad = verinum::V0;
|
|
if (that.has_len() && !that.has_sign() && !that.is_single()) {
|
|
if (pad==verinum::Vx)
|
|
pad = verinum::V0;
|
|
if (pad==verinum::Vz)
|
|
pad = verinum::V0;
|
|
}
|
|
|
|
verinum val(pad, width, that.has_len());
|
|
|
|
for (unsigned idx = 0 ; idx < that.len() ; idx += 1)
|
|
val.set(idx, that[idx]);
|
|
|
|
val.has_sign(that.has_sign());
|
|
if (that.is_string() && (width % 8) == 0) {
|
|
val = verinum(val.as_string());
|
|
}
|
|
return val;
|
|
}
|
|
|
|
verinum cast_to_width(const verinum&that, unsigned width)
|
|
{
|
|
if (that.has_len() && (that.len() == width))
|
|
return that;
|
|
|
|
if (that.len() >= width)
|
|
return verinum(that, width);
|
|
|
|
if (that.len() == 0) {
|
|
verinum val (verinum::V0, width, true);
|
|
val.has_sign(that.has_sign());
|
|
return val;
|
|
}
|
|
|
|
verinum::V pad = that[that.len()-1];
|
|
if (pad==verinum::V1 && !that.has_sign() && !that.is_single())
|
|
pad = verinum::V0;
|
|
if (that.has_len() && !that.has_sign() && !that.is_single()) {
|
|
if (pad==verinum::Vx)
|
|
pad = verinum::V0;
|
|
if (pad==verinum::Vz)
|
|
pad = verinum::V0;
|
|
}
|
|
|
|
verinum val(pad, width, true);
|
|
|
|
for (unsigned idx = 0 ; idx < that.len() ; idx += 1)
|
|
val.set(idx, that[idx]);
|
|
|
|
val.has_sign(that.has_sign());
|
|
return val;
|
|
}
|
|
|
|
/*
|
|
* This function returns a version of the verinum that has only as
|
|
* many bits as are needed to accurately represent the value. It takes
|
|
* into account the signedness of the value.
|
|
*
|
|
* If the input value has a definite length, then that value is just
|
|
* returned as is.
|
|
*/
|
|
verinum trim_vnum(const verinum&that)
|
|
{
|
|
unsigned tlen;
|
|
|
|
if (that.has_len())
|
|
return that;
|
|
|
|
if (that.len() < 2)
|
|
return that;
|
|
|
|
if (that.has_sign()) {
|
|
unsigned top = that.len()-1;
|
|
verinum::V sign = that.get(top);
|
|
|
|
while ((top > 0) && (that.get(top) == sign))
|
|
top -= 1;
|
|
|
|
/* top points to the first digit that is not the
|
|
sign. Set the length to include this and one proper
|
|
sign bit. */
|
|
|
|
if (that.get(top) != sign)
|
|
top += 1;
|
|
|
|
tlen = top+1;
|
|
|
|
} else {
|
|
|
|
/* If the result is unsigned and has an indefinite
|
|
length, then trim off all but one leading zero. */
|
|
unsigned top = that.len()-1;
|
|
while ((top > 0) && (that.get(top) == verinum::V0))
|
|
top -= 1;
|
|
|
|
/* Now top is the index of the highest non-zero bit. If
|
|
that turns out to the highest bit in the vector, then
|
|
there is no trimming possible. */
|
|
if (top+1 == that.len())
|
|
return that;
|
|
|
|
/* Make tlen wide enough to include the highest non-zero
|
|
bit, plus one extra 0 bit. */
|
|
tlen = top+2;
|
|
|
|
/* This can only happen when the verinum is all zeros,
|
|
so make it a single bit wide. */
|
|
if (that.get(top) == verinum::V0) tlen -= 1;
|
|
}
|
|
|
|
verinum tmp (verinum::V0, tlen, false);
|
|
tmp.has_sign(that.has_sign());
|
|
for (unsigned idx = 0 ; idx < tmp.len() ; idx += 1)
|
|
tmp.set(idx, that.get(idx));
|
|
|
|
return tmp;
|
|
}
|
|
|
|
ostream& operator<< (ostream&o, verinum::V v)
|
|
{
|
|
switch (v) {
|
|
case verinum::V0:
|
|
o << "0";
|
|
break;
|
|
case verinum::V1:
|
|
o << "1";
|
|
break;
|
|
case verinum::Vx:
|
|
o << "x";
|
|
break;
|
|
case verinum::Vz:
|
|
o << "z";
|
|
break;
|
|
}
|
|
return o;
|
|
}
|
|
|
|
/*
|
|
* This operator is used by various dumpers to write the Verilog
|
|
* number in a Verilog format.
|
|
*/
|
|
ostream& operator<< (ostream&o, const verinum&v)
|
|
{
|
|
if (v.is_string()) {
|
|
o << "\"" << v.as_string() << "\"";
|
|
return o;
|
|
}
|
|
|
|
/* If the verinum number has a fixed length, dump all the bits
|
|
literally. This is how we express the fixed length in the
|
|
output. */
|
|
if (v.has_len()) {
|
|
o << v.len();
|
|
}
|
|
|
|
/* If the number is fully defined (no x or z) then print it
|
|
out as a decimal number. */
|
|
unsigned dec_len = 8*sizeof(int); /* avoid 32/64 bit differences. */
|
|
if (! v.has_sign()) dec_len -= 1; /* an unsigned number. */
|
|
if (v.is_defined() && v.len() <= dec_len) {
|
|
if (v.has_sign())
|
|
o << "'sd" << v.as_long();
|
|
else
|
|
o << "'d" << v.as_ulong();
|
|
return o;
|
|
}
|
|
|
|
/* Oh, well. Print the minimum to get the value properly
|
|
displayed. */
|
|
if (v.has_sign())
|
|
o << "'sb";
|
|
else
|
|
o << "'b";
|
|
|
|
if (v.len() == 0) {
|
|
o << "0";
|
|
return o;
|
|
}
|
|
|
|
verinum::V trim_left = v.get(v.len()-1);
|
|
unsigned idx;
|
|
|
|
if (v.has_sign()) {
|
|
for (idx = v.len()-1; idx > 0; idx -= 1)
|
|
if (trim_left != v.get(idx-1))
|
|
break;
|
|
|
|
o << trim_left;
|
|
} else {
|
|
idx = v.len();
|
|
}
|
|
|
|
while (idx > 0) {
|
|
o << v.get(idx-1);
|
|
idx -= 1;
|
|
}
|
|
|
|
return o;
|
|
}
|
|
|
|
verinum::V operator == (const verinum&left, const verinum&right)
|
|
{
|
|
verinum::V left_pad = verinum::V0;
|
|
verinum::V right_pad = verinum::V0;
|
|
if (left.has_sign() && right.has_sign()) {
|
|
left_pad = left.get(left.len()-1);
|
|
right_pad = right.get(right.len()-1);
|
|
|
|
if (left_pad == verinum::V1 && right_pad == verinum::V0)
|
|
return verinum::V0;
|
|
if (left_pad == verinum::V0 && right_pad == verinum::V1)
|
|
return verinum::V0;
|
|
}
|
|
|
|
unsigned max_len = left.len();
|
|
if (right.len() > max_len)
|
|
max_len = right.len();
|
|
|
|
for (unsigned idx = 0 ; idx < max_len ; idx += 1) {
|
|
verinum::V left_bit = idx < left.len() ? left[idx] : left_pad;
|
|
verinum::V right_bit = idx < right.len()? right[idx] : right_pad;
|
|
if (left_bit != right_bit)
|
|
return verinum::V0;
|
|
}
|
|
|
|
return verinum::V1;
|
|
}
|
|
|
|
verinum::V operator <= (const verinum&left, const verinum&right)
|
|
{
|
|
verinum::V left_pad = verinum::V0;
|
|
verinum::V right_pad = verinum::V0;
|
|
bool signed_calc = left.has_sign() && right.has_sign();
|
|
if (signed_calc) {
|
|
left_pad = left.get(left.len()-1);
|
|
right_pad = right.get(right.len()-1);
|
|
|
|
if (left_pad == verinum::V1 && right_pad == verinum::V0)
|
|
return verinum::V1;
|
|
if (left_pad == verinum::V0 && right_pad == verinum::V1)
|
|
return verinum::V0;
|
|
}
|
|
|
|
unsigned idx;
|
|
for (idx = left.len() ; idx > right.len() ; idx -= 1) {
|
|
if (left[idx-1] != right_pad) {
|
|
// A change of padding for a negative left argument
|
|
// denotes the left value is less than the right.
|
|
return (signed_calc &&
|
|
(left_pad == verinum::V1)) ? verinum::V1 :
|
|
verinum::V0;
|
|
}
|
|
}
|
|
|
|
for (idx = right.len() ; idx > left.len() ; idx -= 1) {
|
|
if (right[idx-1] != left_pad) {
|
|
// A change of padding for a negative right argument
|
|
// denotes the left value is not less than the right.
|
|
return (signed_calc &&
|
|
(right_pad == verinum::V1)) ? verinum::V0 :
|
|
verinum::V1;
|
|
}
|
|
}
|
|
|
|
idx = right.len();
|
|
if (left.len() < idx) idx = left.len();
|
|
|
|
while (idx > 0) {
|
|
if (left[idx-1] == verinum::Vx) return verinum::Vx;
|
|
if (left[idx-1] == verinum::Vz) return verinum::Vx;
|
|
if (right[idx-1] == verinum::Vx) return verinum::Vx;
|
|
if (right[idx-1] == verinum::Vz) return verinum::Vx;
|
|
if (left[idx-1] > right[idx-1]) return verinum::V0;
|
|
if (left[idx-1] < right[idx-1]) return verinum::V1;
|
|
idx -= 1;
|
|
}
|
|
|
|
return verinum::V1;
|
|
}
|
|
|
|
verinum::V operator < (const verinum&left, const verinum&right)
|
|
{
|
|
verinum::V left_pad = verinum::V0;
|
|
verinum::V right_pad = verinum::V0;
|
|
bool signed_calc = left.has_sign() && right.has_sign();
|
|
if (signed_calc) {
|
|
left_pad = left.get(left.len()-1);
|
|
right_pad = right.get(right.len()-1);
|
|
|
|
if (left_pad == verinum::V1 && right_pad == verinum::V0)
|
|
return verinum::V1;
|
|
if (left_pad == verinum::V0 && right_pad == verinum::V1)
|
|
return verinum::V0;
|
|
}
|
|
|
|
unsigned idx;
|
|
for (idx = left.len() ; idx > right.len() ; idx -= 1) {
|
|
if (left[idx-1] != right_pad) {
|
|
// A change of padding for a negative left argument
|
|
// denotes the left value is less than the right.
|
|
return (signed_calc &&
|
|
(left_pad == verinum::V1)) ? verinum::V1 :
|
|
verinum::V0;
|
|
}
|
|
}
|
|
|
|
for (idx = right.len() ; idx > left.len() ; idx -= 1) {
|
|
if (right[idx-1] != left_pad) {
|
|
// A change of padding for a negative right argument
|
|
// denotes the left value is not less than the right.
|
|
return (signed_calc &&
|
|
(right_pad == verinum::V1)) ? verinum::V0 :
|
|
verinum::V1;
|
|
}
|
|
}
|
|
|
|
while (idx > 0) {
|
|
if (left[idx-1] == verinum::Vx) return verinum::Vx;
|
|
if (left[idx-1] == verinum::Vz) return verinum::Vx;
|
|
if (right[idx-1] == verinum::Vx) return verinum::Vx;
|
|
if (right[idx-1] == verinum::Vz) return verinum::Vx;
|
|
if (left[idx-1] > right[idx-1]) return verinum::V0;
|
|
if (left[idx-1] < right[idx-1]) return verinum::V1;
|
|
idx -= 1;
|
|
}
|
|
|
|
return verinum::V0;
|
|
}
|
|
|
|
static verinum::V add_with_carry(verinum::V l, verinum::V r, verinum::V&c)
|
|
{
|
|
unsigned sum = 0;
|
|
switch (c) {
|
|
case verinum::Vx:
|
|
case verinum::Vz:
|
|
c = verinum::Vx;
|
|
return verinum::Vx;
|
|
case verinum::V0:
|
|
break;
|
|
case verinum::V1:
|
|
sum += 1;
|
|
}
|
|
|
|
switch (l) {
|
|
case verinum::Vx:
|
|
case verinum::Vz:
|
|
c = verinum::Vx;
|
|
return verinum::Vx;
|
|
case verinum::V0:
|
|
break;
|
|
case verinum::V1:
|
|
sum += 1;
|
|
break;
|
|
}
|
|
|
|
switch (r) {
|
|
case verinum::Vx:
|
|
case verinum::Vz:
|
|
c = verinum::Vx;
|
|
return verinum::Vx;
|
|
case verinum::V0:
|
|
break;
|
|
case verinum::V1:
|
|
sum += 1;
|
|
break;
|
|
}
|
|
|
|
if (sum & 2)
|
|
c = verinum::V1;
|
|
else
|
|
c = verinum::V0;
|
|
if (sum & 1)
|
|
return verinum::V1;
|
|
else
|
|
return verinum::V0;
|
|
}
|
|
|
|
verinum operator ~ (const verinum&left)
|
|
{
|
|
verinum val = left;
|
|
for (unsigned idx = 0 ; idx < val.len() ; idx += 1)
|
|
switch (val[idx]) {
|
|
case verinum::V0:
|
|
val.set(idx, verinum::V1);
|
|
break;
|
|
case verinum::V1:
|
|
val.set(idx, verinum::V0);
|
|
break;
|
|
default:
|
|
val.set(idx, verinum::Vx);
|
|
break;
|
|
}
|
|
|
|
return val;
|
|
}
|
|
|
|
/*
|
|
* Addition and subtraction works a bit at a time, from the least
|
|
* significant up to the most significant. The result is signed only
|
|
* if both of the operands are signed. If either operand is unsized,
|
|
* the result is expanded as needed to prevent overflow.
|
|
*/
|
|
|
|
verinum operator + (const verinum&left, const verinum&right)
|
|
{
|
|
const bool has_len_flag = left.has_len() && right.has_len();
|
|
const bool signed_flag = left.has_sign() && right.has_sign();
|
|
|
|
unsigned min_len = min(left.len(), right.len());
|
|
unsigned max_len = max(left.len(), right.len());
|
|
|
|
// If either the left or right values are undefined, the
|
|
// entire result is undefined.
|
|
if (!left.is_defined() || !right.is_defined()) {
|
|
unsigned len = has_len_flag ? max_len : 1;
|
|
verinum result (verinum::Vx, len, has_len_flag);
|
|
result.has_sign(signed_flag);
|
|
return result;
|
|
}
|
|
|
|
verinum::V*val_bits = new verinum::V[max_len+1];
|
|
|
|
verinum::V carry = verinum::V0;
|
|
for (unsigned idx = 0 ; idx < min_len ; idx += 1)
|
|
val_bits[idx] = add_with_carry(left[idx], right[idx], carry);
|
|
|
|
verinum::V rpad = sign_bit(right);
|
|
verinum::V lpad = sign_bit(left);
|
|
|
|
if (left.len() > right.len()) {
|
|
|
|
for (unsigned idx = min_len ; idx < max_len ; idx += 1)
|
|
val_bits[idx] = add_with_carry(left[idx], rpad, carry);
|
|
|
|
} else {
|
|
|
|
for (unsigned idx = min_len ; idx < max_len ; idx += 1)
|
|
val_bits[idx] = add_with_carry(lpad, right[idx], carry);
|
|
}
|
|
|
|
unsigned len = max_len;
|
|
if (!has_len_flag) {
|
|
val_bits[max_len] = add_with_carry(lpad, rpad, carry);
|
|
if (signed_flag) {
|
|
if (val_bits[max_len] != val_bits[max_len-1]) len += 1;
|
|
} else {
|
|
if (val_bits[max_len] != verinum::V0) len += 1;
|
|
}
|
|
}
|
|
verinum result (val_bits, len, has_len_flag);
|
|
result.has_sign(signed_flag);
|
|
|
|
delete[]val_bits;
|
|
|
|
return result;
|
|
}
|
|
|
|
verinum operator - (const verinum&left, const verinum&right)
|
|
{
|
|
const bool has_len_flag = left.has_len() && right.has_len();
|
|
const bool signed_flag = left.has_sign() && right.has_sign();
|
|
|
|
unsigned min_len = min(left.len(), right.len());
|
|
unsigned max_len = max(left.len(), right.len());
|
|
|
|
// If either the left or right values are undefined, the
|
|
// entire result is undefined.
|
|
if (!left.is_defined() || !right.is_defined()) {
|
|
unsigned len = has_len_flag ? max_len : 1;
|
|
verinum result (verinum::Vx, len, has_len_flag);
|
|
result.has_sign(signed_flag);
|
|
return result;
|
|
}
|
|
|
|
verinum::V*val_bits = new verinum::V[max_len+1];
|
|
|
|
verinum::V carry = verinum::V1;
|
|
for (unsigned idx = 0 ; idx < min_len ; idx += 1)
|
|
val_bits[idx] = add_with_carry(left[idx], ~right[idx], carry);
|
|
|
|
verinum::V rpad = sign_bit(right);
|
|
verinum::V lpad = sign_bit(left);
|
|
|
|
if (left.len() > right.len()) {
|
|
|
|
for (unsigned idx = min_len ; idx < max_len ; idx += 1)
|
|
val_bits[idx] = add_with_carry(left[idx], ~rpad, carry);
|
|
|
|
} else {
|
|
|
|
for (unsigned idx = min_len ; idx < max_len ; idx += 1)
|
|
val_bits[idx] = add_with_carry(lpad, ~right[idx], carry);
|
|
}
|
|
|
|
unsigned len = max_len;
|
|
if (signed_flag && !has_len_flag) {
|
|
val_bits[max_len] = add_with_carry(lpad, ~rpad, carry);
|
|
if (val_bits[max_len] != val_bits[max_len-1]) len += 1;
|
|
}
|
|
verinum result (val_bits, len, has_len_flag);
|
|
result.has_sign(signed_flag);
|
|
|
|
delete[]val_bits;
|
|
|
|
return result;
|
|
}
|
|
|
|
verinum operator - (const verinum&right)
|
|
{
|
|
const bool has_len_flag = right.has_len();
|
|
const bool signed_flag = right.has_sign();
|
|
|
|
unsigned len = right.len();
|
|
|
|
// If either the left or right values are undefined, the
|
|
// entire result is undefined.
|
|
if (!right.is_defined()) {
|
|
verinum result (verinum::Vx, has_len_flag ? len : 1, has_len_flag);
|
|
result.has_sign(signed_flag);
|
|
return result;
|
|
}
|
|
|
|
verinum::V*val_bits = new verinum::V[len+1];
|
|
|
|
verinum::V carry = verinum::V1;
|
|
for (unsigned idx = 0 ; idx < len ; idx += 1)
|
|
val_bits[idx] = add_with_carry(verinum::V0, ~right[idx], carry);
|
|
|
|
if (signed_flag && !has_len_flag) {
|
|
val_bits[len] = add_with_carry(verinum::V0, ~sign_bit(right), carry);
|
|
if (val_bits[len] != val_bits[len-1]) len += 1;
|
|
}
|
|
verinum result (val_bits, len, has_len_flag);
|
|
result.has_sign(signed_flag);
|
|
|
|
delete[]val_bits;
|
|
|
|
return result;
|
|
}
|
|
|
|
/*
|
|
* This operator multiplies the left number by the right number. The
|
|
* result is signed only if both of the operands are signed. If either
|
|
* operand is unsized, the resulting number is as large as the sum of
|
|
* the sizes of the operands.
|
|
*
|
|
* The algorithm used is successive shift and add operations,
|
|
* implemented as the nested loops.
|
|
*/
|
|
verinum operator * (const verinum&left, const verinum&right)
|
|
{
|
|
const bool has_len_flag = left.has_len() && right.has_len();
|
|
const bool signed_flag = left.has_sign() && right.has_sign();
|
|
|
|
const unsigned l_len = left.len();
|
|
const unsigned r_len = right.len();
|
|
|
|
unsigned len = has_len_flag ? max(l_len, r_len) : l_len + r_len;
|
|
|
|
// If either the left or right values are undefined, the
|
|
// entire result is undefined.
|
|
if (!left.is_defined() || !right.is_defined()) {
|
|
verinum result (verinum::Vx, has_len_flag ? len : 1, has_len_flag);
|
|
result.has_sign(signed_flag);
|
|
return result;
|
|
}
|
|
|
|
verinum result(verinum::V0, len, has_len_flag);
|
|
result.has_sign(signed_flag);
|
|
|
|
verinum::V r_sign = sign_bit(right);
|
|
for (unsigned rdx = 0 ; rdx < len ; rdx += 1) {
|
|
|
|
verinum::V r_bit = rdx < r_len ? right.get(rdx) : r_sign;
|
|
if (r_bit == verinum::V0)
|
|
continue;
|
|
|
|
verinum::V l_sign = sign_bit(left);
|
|
verinum::V carry = verinum::V0;
|
|
for (unsigned ldx = 0 ; ldx < (len - rdx) ; ldx += 1) {
|
|
verinum::V l_bit = ldx < l_len ? left[ldx] : l_sign;
|
|
result.set(ldx+rdx, add_with_carry(l_bit,
|
|
result[rdx+ldx],
|
|
carry));
|
|
}
|
|
}
|
|
|
|
return trim_vnum(result);
|
|
}
|
|
|
|
static verinum make_p_one(unsigned len, bool has_len, bool has_sign)
|
|
{
|
|
verinum tmp (verinum::V0, has_len ? len : 2, has_len);
|
|
tmp.set(0, verinum::V1);
|
|
tmp.has_sign(has_sign);
|
|
return tmp;
|
|
}
|
|
|
|
static verinum make_m_one(unsigned len, bool has_len, bool has_sign)
|
|
{
|
|
verinum tmp (verinum::V1, has_len ? len : 1, has_len);
|
|
tmp.has_sign(has_sign);
|
|
return tmp;
|
|
}
|
|
|
|
static verinum recursive_pow(const verinum&left, verinum&right)
|
|
{
|
|
// If the exponent is zero, return a value of 1
|
|
if (right.is_zero()) {
|
|
return make_p_one(left.len(), left.has_len(), left.has_sign());
|
|
}
|
|
|
|
verinum result;
|
|
if (right.get(0) == 1) {
|
|
// The exponent is odd, so subtract 1 from it and recurse.
|
|
// We know it's odd, so the subtraction is easy.
|
|
right.set(0, verinum::V0);
|
|
result = pow(left, right);
|
|
result = left * result;
|
|
} else {
|
|
// The exponent is even, so divide it by 2 and recurse
|
|
right = right >> 1;
|
|
result = pow(left, right);
|
|
result = result * result;
|
|
}
|
|
return result;
|
|
}
|
|
|
|
verinum pow(const verinum&left, const verinum&right)
|
|
{
|
|
verinum result;
|
|
|
|
// We need positive and negative one in a few places.
|
|
verinum p_one = make_p_one(left.len(), left.has_len(), left.has_sign());
|
|
verinum m_one = make_m_one(left.len(), left.has_len(), left.has_sign());
|
|
|
|
// If either the left or right values are undefined, the
|
|
// entire result is undefined.
|
|
if (!left.is_defined() || !right.is_defined()) {
|
|
result = verinum(verinum::Vx, left.len(), left.has_len());
|
|
result.has_sign(left.has_sign());
|
|
|
|
// If the right value is zero we need to set the result to 1.
|
|
} else if (right.is_zero()) {
|
|
result = p_one;
|
|
|
|
} else if (right.is_negative()) {
|
|
|
|
// 0 ** <negative> is 'bx.
|
|
if (left.is_zero()) {
|
|
result = verinum(verinum::Vx, left.len(), left.has_len());
|
|
result.has_sign(left.has_sign());
|
|
|
|
// -1 ** <negative> is 1 or -1. Note that this must be done
|
|
// before testing for +1 in case 'left' has a width of 1.
|
|
} else if (left.has_sign() && left == m_one) {
|
|
if (right.get(0) == verinum::V0) {
|
|
result = p_one;
|
|
} else {
|
|
result = m_one;
|
|
}
|
|
|
|
// 1 ** <negative> is 1.
|
|
} else if (left == p_one) {
|
|
result = p_one;
|
|
|
|
// Everything else is 0.
|
|
} else {
|
|
result = verinum(verinum::V0, left.len(), left.has_len());
|
|
result.has_sign(left.has_sign());
|
|
}
|
|
|
|
} else {
|
|
verinum exponent = right;
|
|
result = recursive_pow(left, exponent);
|
|
}
|
|
|
|
return trim_vnum(result);
|
|
}
|
|
|
|
verinum operator << (const verinum&that, unsigned shift)
|
|
{
|
|
bool has_len_flag = that.has_len();
|
|
|
|
unsigned len = that.len();
|
|
if (!has_len_flag) len += shift;
|
|
|
|
verinum result(verinum::V0, len, has_len_flag);
|
|
result.has_sign(that.has_sign());
|
|
|
|
for (unsigned idx = shift ; idx < len ; idx += 1)
|
|
result.set(idx, that.get(idx - shift));
|
|
|
|
return trim_vnum(result);
|
|
}
|
|
|
|
verinum operator >> (const verinum&that, unsigned shift)
|
|
{
|
|
bool has_len_flag = that.has_len();
|
|
|
|
unsigned len = that.len();
|
|
|
|
verinum::V sgn_bit = that.has_sign() ? that.get(len-1) : verinum::V0;
|
|
|
|
if (shift >= len) {
|
|
if (!has_len_flag) len = 1;
|
|
verinum result(sgn_bit, len, has_len_flag);
|
|
result.has_sign(that.has_sign());
|
|
return result;
|
|
}
|
|
|
|
if (!has_len_flag) len -= shift;
|
|
verinum result(sgn_bit, len, has_len_flag);
|
|
result.has_sign(that.has_sign());
|
|
|
|
for (unsigned idx = shift ; idx < that.len() ; idx += 1)
|
|
result.set(idx-shift, that.get(idx));
|
|
|
|
return trim_vnum(result);
|
|
}
|
|
|
|
static verinum unsigned_divide(verinum num, verinum den, bool signed_result)
|
|
{
|
|
// We need the following calculations to be lossless. The
|
|
// result will be cast to the required width by the caller.
|
|
num.has_len(false);
|
|
den.has_len(false);
|
|
|
|
unsigned nwid = num.len();
|
|
while (nwid > 0 && (num.get(nwid-1) == verinum::V0))
|
|
nwid -= 1;
|
|
|
|
unsigned dwid = den.len();
|
|
while (dwid > 0 && (den.get(dwid-1) == verinum::V0))
|
|
dwid -= 1;
|
|
|
|
if (dwid > nwid)
|
|
return verinum(verinum::V0, 1);
|
|
|
|
den = den << (nwid-dwid);
|
|
|
|
unsigned idx = nwid - dwid + 1;
|
|
verinum result (verinum::V0, signed_result ? idx + 1 : idx);
|
|
if (signed_result) {
|
|
result.set(idx, verinum::V0);
|
|
result.has_sign(true);
|
|
}
|
|
while (idx > 0) {
|
|
if (den <= num) {
|
|
verinum dif = num - den;
|
|
num = dif;
|
|
result.set(idx-1, verinum::V1);
|
|
}
|
|
den = den >> 1;
|
|
idx -= 1;
|
|
}
|
|
|
|
return result;
|
|
}
|
|
|
|
static verinum unsigned_modulus(verinum num, verinum den)
|
|
{
|
|
// We need the following calculations to be lossless. The
|
|
// result will be cast to the required width by the caller.
|
|
num.has_len(false);
|
|
den.has_len(false);
|
|
|
|
unsigned nwid = num.len();
|
|
while (nwid > 0 && (num.get(nwid-1) == verinum::V0))
|
|
nwid -= 1;
|
|
|
|
unsigned dwid = den.len();
|
|
while (dwid > 0 && (den.get(dwid-1) == verinum::V0))
|
|
dwid -= 1;
|
|
|
|
if (dwid > nwid)
|
|
return num;
|
|
|
|
den = den << (nwid-dwid);
|
|
|
|
unsigned idx = nwid - dwid + 1;
|
|
while (idx > 0) {
|
|
if (den <= num) {
|
|
verinum dif = num - den;
|
|
num = dif;
|
|
}
|
|
den = den >> 1;
|
|
idx -= 1;
|
|
}
|
|
|
|
return num;
|
|
}
|
|
|
|
/*
|
|
* This operator divides the left number by the right number. The result
|
|
* is signed only if both of the operands are signed.
|
|
*/
|
|
verinum operator / (const verinum&left, const verinum&right)
|
|
{
|
|
const bool has_len_flag = left.has_len() && right.has_len();
|
|
const bool signed_flag = left.has_sign() && right.has_sign();
|
|
|
|
unsigned use_len = left.len();
|
|
|
|
// If either the left or right values are undefined, or the
|
|
// right value is zero, the entire result is undefined.
|
|
if (!left.is_defined() || !right.is_defined() || right.is_zero()) {
|
|
verinum result (verinum::Vx, use_len, has_len_flag);
|
|
result.has_sign(signed_flag);
|
|
return result;
|
|
}
|
|
|
|
verinum result(verinum::Vz, use_len, has_len_flag);
|
|
|
|
/* do the operation differently, depending on whether the
|
|
result is signed or not. */
|
|
if (signed_flag) {
|
|
|
|
if (use_len <= (8*sizeof(long) - 1)) {
|
|
long l = left.as_long();
|
|
long r = right.as_long();
|
|
bool overflow = (l == LONG_MIN) && (r == -1);
|
|
long v = overflow ? LONG_MIN : l / r;
|
|
for (unsigned idx = 0 ; idx < use_len ; idx += 1) {
|
|
result.set(idx, (v & 1)? verinum::V1 : verinum::V0);
|
|
v >>= 1;
|
|
}
|
|
|
|
} else {
|
|
verinum use_left, use_right;
|
|
bool negative = false;
|
|
if (left.is_negative()) {
|
|
use_left = -left;
|
|
negative = !negative;
|
|
} else {
|
|
use_left = left;
|
|
}
|
|
use_left.has_sign(false);
|
|
if (right.is_negative()) {
|
|
use_right = -right;
|
|
negative = !negative;
|
|
} else {
|
|
use_right = right;
|
|
}
|
|
use_right.has_sign(false);
|
|
result = unsigned_divide(use_left, use_right, true);
|
|
if (negative) result = -result;
|
|
}
|
|
|
|
} else {
|
|
|
|
if (use_len <= 8 * sizeof(unsigned long)) {
|
|
/* Use native unsigned division to do the work. */
|
|
|
|
unsigned long l = left.as_ulong();
|
|
unsigned long r = right.as_ulong();
|
|
unsigned long v = l / r;
|
|
for (unsigned idx = 0 ; idx < use_len ; idx += 1) {
|
|
result.set(idx, (v & 1)? verinum::V1 : verinum::V0);
|
|
v >>= 1;
|
|
}
|
|
|
|
} else {
|
|
result = unsigned_divide(left, right, false);
|
|
}
|
|
}
|
|
|
|
if (has_len_flag)
|
|
result = cast_to_width(result, use_len);
|
|
|
|
result.has_sign(signed_flag);
|
|
return trim_vnum(result);
|
|
}
|
|
|
|
verinum operator % (const verinum&left, const verinum&right)
|
|
{
|
|
const bool has_len_flag = left.has_len() && right.has_len();
|
|
const bool signed_flag = left.has_sign() && right.has_sign();
|
|
|
|
unsigned use_len = left.len();
|
|
|
|
// If either the left or right values are undefined, or the
|
|
// right value is zero, the entire result is undefined.
|
|
if (!left.is_defined() || !right.is_defined() || right.is_zero()) {
|
|
verinum result (verinum::Vx, use_len, has_len_flag);
|
|
result.has_sign(signed_flag);
|
|
return result;
|
|
}
|
|
|
|
verinum result(verinum::Vz, use_len, has_len_flag);
|
|
|
|
if (signed_flag) {
|
|
if (use_len <= 8*sizeof(long)) {
|
|
/* Use native signed modulus to do the work. */
|
|
long l = left.as_long();
|
|
long r = right.as_long();
|
|
bool overflow = (l == LONG_MIN) && (r == -1);
|
|
long v = overflow ? 0 : l % r;
|
|
for (unsigned idx = 0 ; idx < use_len ; idx += 1) {
|
|
result.set(idx, (v & 1)? verinum::V1 : verinum::V0);
|
|
v >>= 1;
|
|
}
|
|
} else {
|
|
verinum use_left, use_right;
|
|
bool negative = false;
|
|
if (left.is_negative()) {
|
|
use_left = -left;
|
|
negative = true;
|
|
} else {
|
|
use_left = left;
|
|
}
|
|
use_left.has_sign(false);
|
|
if (right.is_negative()) {
|
|
use_right = -right;
|
|
} else {
|
|
use_right = right;
|
|
}
|
|
use_right.has_sign(false);
|
|
result = unsigned_modulus(use_left, use_right);
|
|
if (negative) result = -result;
|
|
}
|
|
} else {
|
|
if (use_len <= 8*sizeof(unsigned long)) {
|
|
/* Use native unsigned modulus to do the work. */
|
|
unsigned long l = left.as_ulong();
|
|
unsigned long r = right.as_ulong();
|
|
unsigned long v = l % r;
|
|
for (unsigned idx = 0 ; idx < use_len ; idx += 1) {
|
|
result.set(idx, (v & 1)? verinum::V1 : verinum::V0);
|
|
v >>= 1;
|
|
}
|
|
} else {
|
|
result = unsigned_modulus(left, right);
|
|
}
|
|
}
|
|
|
|
if (has_len_flag)
|
|
result = cast_to_width(result, use_len);
|
|
|
|
result.has_sign(signed_flag);
|
|
return trim_vnum(result);
|
|
}
|
|
|
|
verinum concat(const verinum&left, const verinum&right)
|
|
{
|
|
if (left.is_string() && right.is_string()) {
|
|
std::string tmp = left.as_string() + right.as_string();
|
|
verinum res (tmp);
|
|
return res;
|
|
}
|
|
|
|
verinum res (verinum::V0, left.len() + right.len());
|
|
for (unsigned idx = 0 ; idx < right.len() ; idx += 1)
|
|
res.set(idx, right.get(idx));
|
|
|
|
for (unsigned idx = 0 ; idx < left.len() ; idx += 1)
|
|
res.set(idx+right.len(), left.get(idx));
|
|
|
|
return res;
|
|
}
|
|
|
|
verinum::V operator ~ (verinum::V l)
|
|
{
|
|
switch (l) {
|
|
case verinum::V0:
|
|
return verinum::V1;
|
|
case verinum::V1:
|
|
return verinum::V0;
|
|
default:
|
|
return verinum::Vx;
|
|
}
|
|
}
|
|
|
|
verinum::V operator | (verinum::V l, verinum::V r)
|
|
{
|
|
if (l == verinum::V1)
|
|
return verinum::V1;
|
|
if (r == verinum::V1)
|
|
return verinum::V1;
|
|
if (l != verinum::V0)
|
|
return verinum::Vx;
|
|
if (r != verinum::V0)
|
|
return verinum::Vx;
|
|
return verinum::V0;
|
|
}
|
|
|
|
verinum::V operator & (verinum::V l, verinum::V r)
|
|
{
|
|
if (l == verinum::V0)
|
|
return verinum::V0;
|
|
if (r == verinum::V0)
|
|
return verinum::V0;
|
|
if (l != verinum::V1)
|
|
return verinum::Vx;
|
|
if (r != verinum::V1)
|
|
return verinum::Vx;
|
|
return verinum::V1;
|
|
}
|
|
|
|
verinum::V operator ^ (verinum::V l, verinum::V r)
|
|
{
|
|
if (l == verinum::V0)
|
|
return bit4_z2x(r);
|
|
if (r == verinum::V0)
|
|
return bit4_z2x(l);
|
|
if ((l == verinum::V1) && (r == verinum::V1))
|
|
return verinum::V0;
|
|
|
|
return verinum::Vx;
|
|
}
|