strcalc.c

/*
 * This file is part of libFirm.
 * Copyright (C) 2012 University of Karlsruhe.
 */

/**
 * @file
 * @brief    Provides basic mathematical operations on values represented as
 *           strings.
 * @date     2003
 * @author   Mathias Heil
 */
#include <stdlib.h>
#include <string.h>
#include <assert.h>
#include <stdio.h>
#include <limits.h>

#include "strcalc.h"
#include "xmalloc.h"
#include "error.h"
#include "bitfiddle.h"

#define SC_BITS      4
#define SC_MASK      ((sc_word)0xF)
#define SC_RESULT(x) ((x) & ((1U << SC_BITS) - 1U))
#define SC_CARRY(x)  ((unsigned)(x) >> SC_BITS)

#define _bitisset(digit, pos) (((digit) & (1 << (pos))) != 0)

static sc_word *calc_buffer = NULL; /**< buffer holding all results */
static char *output_buffer = NULL;  /**< buffer for output */
static unsigned bit_pattern_size;   /**< maximum number of bits */
static unsigned calc_buffer_size;   /**< size of internally stored values */
static unsigned max_value_size;     /**< maximum size of values */

static const sc_word shrs_table[16][4][2] = {
	{ { 0, 0}, {0, 0}, {0,  0}, {0,  0} },
	{ { 1, 0}, {0, 8}, {0,  4}, {0,  2} },
	{ { 2, 0}, {1, 0}, {0,  8}, {0,  4} },
	{ { 3, 0}, {1, 8}, {0, 12}, {0,  6} },
	{ { 4, 0}, {2, 0}, {1,  0}, {0,  8} },
	{ { 5, 0}, {2, 8}, {1,  4}, {0, 10} },
	{ { 6, 0}, {3, 0}, {1,  8}, {0, 12} },
	{ { 7, 0}, {3, 8}, {1, 12}, {0, 14} },
	{ { 8, 0}, {4, 0}, {2,  0}, {1,  0} },
	{ { 9, 0}, {4, 8}, {2,  4}, {1,  2} },
	{ {10, 0}, {5, 0}, {2,  8}, {1,  4} },
	{ {11, 0}, {5, 8}, {2, 12}, {1,  6} },
	{ {12, 0}, {6, 0}, {3,  0}, {1,  8} },
	{ {13, 0}, {6, 8}, {3,  4}, {1, 10} },
	{ {14, 0}, {7, 0}, {3,  8}, {1, 12} },
	{ {15, 0}, {7, 8}, {3, 12}, {1, 14} }
};

/** converting a digit to a binary string */
static char const *const binary_table[] = {
	"0000", "0001", "0010", "0011", "0100", "0101", "0110", "0111",
	"1000", "1001", "1010", "1011", "1100", "1101", "1110", "1111"
};

void sc_zero(sc_word *buffer)
{
	memset(buffer, 0, sizeof(buffer[0]) * calc_buffer_size);
}

static sc_word sex_digit(unsigned x)
{
	return (SC_MASK << (x+1)) & SC_MASK;
}

static sc_word max_digit(unsigned x)
{
	return (1u << x) - 1;
}

static sc_word min_digit(unsigned x)
{
	return SC_MASK - max_digit(x);
}

void sc_not(const sc_word *val, sc_word *buffer)
{
	for (unsigned counter = 0; counter<calc_buffer_size; counter++)
		buffer[counter] = val[counter] ^ SC_MASK;
}

void sc_or(const sc_word *val1, const sc_word *val2, sc_word *buffer)
{
	for (unsigned counter = 0; counter<calc_buffer_size; counter++)
		buffer[counter] = val1[counter] | val2[counter];
}

void sc_xor(const sc_word *val1, const sc_word *val2, sc_word *buffer)
{
	for (unsigned counter = 0; counter<calc_buffer_size; counter++)
		buffer[counter] = val1[counter] ^ val2[counter];
}

void sc_and(const sc_word *val1, const sc_word *val2, sc_word *buffer)
{
	for (unsigned counter = 0; counter<calc_buffer_size; counter++)
		buffer[counter] = val1[counter] & val2[counter];
}

void sc_andnot(const sc_word *val1, const sc_word *val2, sc_word *buffer)
{
	for (unsigned counter = 0; counter < calc_buffer_size; ++counter)
		buffer[counter] = val1[counter] & (SC_MASK ^ val2[counter]);
}

void sc_inc(const sc_word *val, sc_word *buffer)
{
	for (unsigned counter = 0; counter < calc_buffer_size; ++counter) {
		sc_word v = val[counter];
		if (v < SC_MASK) {
			buffer[counter] = v+1;
			/* copy the rest of the buffer if necessary */
			if (buffer != val)
				memcpy(&buffer[counter+1], &val[counter+1],
				       calc_buffer_size-(counter+1));
			break;
		}
		buffer[counter] = 0;
	}
}

void sc_neg(const sc_word *val, sc_word *buffer)
{
	sc_not(val, buffer);
	sc_inc(buffer, buffer);
}

void sc_add(const sc_word *val1, const sc_word *val2, sc_word *buffer)
{
	sc_word carry = 0;
	for (unsigned counter = 0; counter < calc_buffer_size; ++counter) {
		unsigned const sum = val1[counter] + val2[counter] + carry;
		buffer[counter] = SC_RESULT(sum);
		carry           = SC_CARRY(sum);
	}
}

void sc_sub(const sc_word *val1, const sc_word *val2, sc_word *buffer)
{
	/* intermediate buffer to hold -val2 */
	sc_word *temp_buffer = ALLOCAN(sc_word, calc_buffer_size);

	sc_neg(val2, temp_buffer);
	sc_add(val1, temp_buffer, buffer);
}

void sc_mul(const sc_word *val1, const sc_word *val2, sc_word *buffer)
{
	sc_word *temp_buffer = ALLOCANZ(sc_word, calc_buffer_size);
	sc_word *neg_val1    = ALLOCAN(sc_word, calc_buffer_size);
	sc_word *neg_val2    = ALLOCAN(sc_word, calc_buffer_size);

	/* the multiplication works only for positive values, for negative values *
	 * it is necessary to negate them and adjust the result accordingly       */
	bool sign = false;
	if (sc_is_negative(val1)) {
		sc_neg(val1, neg_val1);
		val1 = neg_val1;
		sign = !sign;
	}
	if (sc_is_negative(val2)) {
		sc_neg(val2, neg_val2);
		val2 = neg_val2;
		sign = !sign;
	}

	for (unsigned c_outer = 0; c_outer < max_value_size; c_outer++) {
		sc_word outer = val2[c_outer];
		if (outer == 0)
			continue;
		unsigned carry = 0; /* container for carries */
		for (unsigned c_inner = 0; c_inner < max_value_size; c_inner++) {
			sc_word inner = val1[c_inner];
			/* do the following calculation:
			 * Add the current carry, the value at position c_outer+c_inner
			 * and the result of the multiplication of val1[c_inner] and
			 * val2[c_outer]. This is the usual pen-and-paper multiplication
			 */

			/* multiplicate the two digits */
			unsigned const mul = inner*outer;
			/* add old value to result of multiplication and the carry */
			unsigned const sum = temp_buffer[c_inner+c_outer] + mul + carry;

			/* all carries together result in new carry. This is always
			 * smaller than the base b:
			 * Both multiplicands, the carry and the value already in the
			 * temp buffer are single digits and their value is therefore
			 * at most equal to (b-1).
			 * This leads to:
			 * (b-1)(b-1)+(b-1)+(b-1) = b*b-1
			 * The tables list all operations rem b, so the carry is at
			 * most
			 * (b*b-1)rem b = -1rem b = b-1
			 */
			temp_buffer[c_inner + c_outer] = SC_RESULT(sum);
			carry                          = SC_CARRY(sum);
		}

		/* A carry may hang over */
		/* c_outer is always smaller than max_value_size! */
		temp_buffer[max_value_size + c_outer] = carry;
	}

	if (sign)
		sc_neg(temp_buffer, buffer);
	else
		memcpy(buffer, temp_buffer, calc_buffer_size);
}

/**
 * Shift the buffer to left and add an sc digit
 */
static void sc_push(sc_word digit, sc_word *buffer)
{
	for (unsigned counter = calc_buffer_size - 1; counter-- > 0; ) {
		buffer[counter+1] = buffer[counter];
	}
	buffer[0] = digit;
}

bool sc_divmod(const sc_word *dividend, const sc_word *divisor,
               sc_word *quot, sc_word *rem)
{
	assert(quot != dividend && quot != divisor);
	assert(rem != dividend && rem != divisor);
	/* clear result buffer */
	sc_zero(quot);
	sc_zero(rem);

	/* division by zero is not allowed */
	assert(!sc_is_zero(divisor, calc_buffer_size*SC_BITS));

	/* if the dividend is zero result is zero (quot is zero) */
	if (sc_is_zero(dividend, calc_buffer_size*SC_BITS))
		return false;

	bool     div_sign = false;
	bool     rem_sign = false;
	sc_word *neg_val1 = ALLOCAN(sc_word, calc_buffer_size);
	if (sc_is_negative(dividend)) {
		sc_neg(dividend, neg_val1);
		div_sign = !div_sign;
		rem_sign = !rem_sign;
		dividend = neg_val1;
	}

	sc_word *neg_val2 = ALLOCAN(sc_word, calc_buffer_size);
	sc_neg(divisor, neg_val2);
	const sc_word *minus_divisor;
	if (sc_is_negative(divisor)) {
		div_sign = !div_sign;
		minus_divisor = divisor;
		divisor = neg_val2;
	} else {
		minus_divisor = neg_val2;
	}

	/* if divisor >= dividend division is easy
	 * (remember these are absolute values) */
	switch (sc_comp(dividend, divisor)) {
	case ir_relation_equal: /* dividend == divisor */
		quot[0] = 1;
		goto end;

	case ir_relation_less: /* dividend < divisor */
		memcpy(rem, dividend, calc_buffer_size);
		goto end;

	default: /* unluckily division is necessary :( */
		break;
	}

	for (unsigned c_dividend = calc_buffer_size; c_dividend-- > 0; ) {
		sc_push(dividend[c_dividend], rem);
		sc_push(0, quot);

		if (sc_comp(rem, divisor) != ir_relation_less) {
			/* remainder >= divisor */
			/* subtract until the remainder becomes negative, this should
			 * be faster than comparing remainder with divisor  */
			sc_add(rem, minus_divisor, rem);

			while (!sc_is_negative(rem)) {
				/* TODO can this generate carry or is masking redundant? */
				quot[0] = SC_RESULT(quot[0] + 1);
				sc_add(rem, minus_divisor, rem);
			}

			/* subtracted one too much */
			sc_add(rem, divisor, rem);
		}
	}
end:
	if (div_sign)
		sc_neg(quot, quot);

	if (rem_sign)
		sc_neg(rem, rem);

	/* sets carry if remainder is non-zero ??? */
	bool carry_flag = !sc_is_zero(rem, calc_buffer_size*SC_BITS);
	return carry_flag;
}

/**
 * Implements a Shift Left, which can either preserve the sign bit
 * or not.
 */
static void do_shl(const sc_word *val1, sc_word *buffer, unsigned shift_cnt,
                   unsigned bitsize, bool is_signed)
{
	assert(buffer != val1);

	/* if shifting far enough the result is zero */
	if (shift_cnt >= bitsize) {
		sc_zero(buffer);
		return;
	}

	unsigned const shift = shift_cnt % SC_BITS;
	shift_cnt = shift_cnt / SC_BITS;

	/* shift the single digits some bytes (offset) and some bits (table)
	 * to the left */
	unsigned carry = 0;
	for (unsigned counter = 0; counter < bitsize/SC_BITS - shift_cnt;
	     ++counter) {
		unsigned const shl = val1[counter] << shift | carry;
		buffer[counter + shift_cnt] = SC_RESULT(shl);
		carry = SC_CARRY(shl);
	}
	unsigned counter   = bitsize/SC_BITS - shift_cnt;
	unsigned bitoffset = 0;
	if (bitsize%SC_BITS > 0) {
		unsigned const shl = val1[counter] << shift | carry;
		buffer[counter + shift_cnt] = SC_RESULT(shl);
		bitoffset = counter;
	} else {
		bitoffset = counter - 1;
	}

	/* fill with zeroes */
	for (unsigned counter = 0; counter < shift_cnt; counter++)
		buffer[counter] = 0;

	/* if the mode was signed, change sign when the mode's msb is now 1 */
	shift_cnt = bitoffset + shift_cnt;
	bitoffset = (bitsize-1) % SC_BITS;
	if (is_signed && _bitisset(buffer[shift_cnt], bitoffset)) {
		/* this sets the upper bits of the leftmost digit */
		buffer[shift_cnt] |= min_digit(bitoffset);
		for (unsigned counter = shift_cnt+1; counter < calc_buffer_size;
		     ++counter) {
			buffer[counter] = SC_MASK;
		}
	} else if (is_signed && !_bitisset(buffer[shift_cnt], bitoffset)) {
		/* this clears the upper bits of the leftmost digit */
		buffer[shift_cnt] &= max_digit(bitoffset);
		for (unsigned counter = shift_cnt+1; counter < calc_buffer_size;
		     ++counter) {
			buffer[counter] = 0;
		}
	}
}

/**
 * Implements a Shift Right, which can either preserve the sign bit
 * or not.
 *
 * @param bitsize   bitsize of the value to be shifted
 */
static bool do_shr(const sc_word *val1, sc_word *buffer, unsigned shift_cnt,
                   unsigned bitsize, bool is_signed, bool signed_shift)
{
	bool    carry_flag = false;
	sc_word sign = signed_shift && sc_get_bit_at(val1, bitsize-1) ? SC_MASK : 0;

	/* if shifting far enough the result is either 0 or -1 */
	if (shift_cnt >= bitsize) {
		if (!sc_is_zero(val1, calc_buffer_size*SC_BITS)) {
			carry_flag = true;
		}
		memset(buffer, sign, calc_buffer_size);
		return carry_flag;
	}

	unsigned shift_mod = shift_cnt % SC_BITS;
	unsigned shift_nib = shift_cnt / SC_BITS;

	/* check if any bits are lost, and set carry_flag if so */
	for (unsigned counter = 0; counter < shift_nib; ++counter) {
		if (val1[counter] != 0) {
			carry_flag = true;
			break;
		}
	}
	if ((val1[shift_nib] & ((1<<shift_mod)-1)) != 0)
		carry_flag = true;

	/* shift digits to the right with offset, carry and all */
	buffer[0] = shrs_table[val1[shift_nib]][shift_mod][0];
	unsigned counter;
	for (counter = 1; counter < ((bitsize + 3) / SC_BITS) - shift_nib;
	     ++counter) {
		const sc_word *shrs = shrs_table[val1[counter+shift_nib]][shift_mod];
		buffer[counter]      = shrs[0];
		buffer[counter - 1] |= shrs[1];
	}

	/* the last digit is special in regard of signed/unsigned shift */
	unsigned bitoffset = bitsize % SC_BITS;
	sc_word  msd       = sign;  /* most significant digit */

	/* remove sign bits if mode was signed and this is an unsigned shift */
	if (!signed_shift && is_signed) {
		msd &= max_digit(bitoffset);
	}

	const sc_word *shrs = shrs_table[msd][shift_mod];

	/* signed shift and signed mode and negative value means all bits to the
	 * left are set */
	if (signed_shift && sign == SC_MASK) {
		buffer[counter] = shrs[0] | min_digit(bitoffset);
	} else {
		buffer[counter] = shrs[0];
	}

	if (counter > 0)
		buffer[counter - 1] |= shrs[1];

	/* fill with 0xF or 0 depending on sign */
	for (counter++; counter < calc_buffer_size; counter++) {
		buffer[counter] = sign;
	}

	return carry_flag;
}

const sc_word *sc_get_buffer(void)
{
	return calc_buffer;
}

unsigned sc_get_buffer_length(void)
{
	return calc_buffer_size;
}

void sc_zero_extend(sc_word *buffer, unsigned from_bits)
{
	unsigned bit  = from_bits % SC_BITS;
	unsigned word = from_bits / SC_BITS;
	if (bit > 0) {
		memset(&buffer[word+1], 0, calc_buffer_size-(word+1));
		buffer[word] &= max_digit(bit);
	} else {
		memset(&buffer[word], 0, calc_buffer_size-word);
	}
}

void sc_sign_extend(sc_word *buffer, unsigned from_bits)
{
	unsigned bits     = from_bits - 1;
	bool     sign_bit = sc_get_bit_at(buffer, bits);
	if (sign_bit) {
		/* sign bit is set, we need sign extension */
		unsigned word = bits / SC_BITS;
		for (unsigned i = word + 1; i < calc_buffer_size; ++i)
			buffer[i] = SC_MASK;
		buffer[word] |= sex_digit(bits % SC_BITS);
	} else {
		sc_zero_extend(buffer, from_bits);
	}
}

/** Ensures that our source character set conforms to ASCII for a-f, A-F, 0-9 */
static inline void check_ascii(void)
{
	assert((('a'-97) | ('b'-98) | ('c'-99) | ('d'-100) | ('e'-101) | ('f'-102)
	       |('A'-65) | ('B'-66) | ('C'-67) | ('D'- 68) | ('E'- 69) | ('F'- 70)
	       |('0'-48) | ('1'-49) | ('2'-50) | ('3'- 51) | ('4'- 52)
	       |('5'-53) | ('6'-54) | ('7'-55) | ('8'- 56) | ('9'- 57)) == 0);
}

bool sc_val_from_str(bool negative, unsigned base, const char *str, size_t len,
                     sc_word *buffer)
{
	assert(str != NULL);
	assert(len > 0);
	check_ascii();

	assert(base > 1 && base <= 16);
	sc_word *sc_base = ALLOCAN(sc_word, calc_buffer_size);
	sc_val_from_ulong(base, sc_base);

	sc_word *val = ALLOCANZ(sc_word, calc_buffer_size);

	sc_zero(buffer);

	/* BEGIN string evaluation, from left to right */
	while (len > 0) {
		char c = *str;
		unsigned v;
		if (c >= '0' && c <= '9')
			v = c - '0';
		else if (c >= 'A' && c <= 'F')
			v = c - 'A' + 10;
		else if (c >= 'a' && c <= 'f')
			v = c - 'a' + 10;
		else
			return false;

		if (v >= base)
			return false;
		val[0] = v;

		/* Radix conversion from base b to base B:
		 *  (UnUn-1...U1U0)b == ((((Un*b + Un-1)*b + ...)*b + U1)*b + U0)B */
		/* multiply current value with base */
		sc_mul(sc_base, buffer, buffer);
		/* add next digit to current value  */
		sc_add(val, buffer, buffer);

		/* get ready for the next letter */
		str++;
		len--;
	}

	if (negative)
		sc_neg(buffer, buffer);

	return true;
}

void sc_val_from_long(long value, sc_word *buffer)
{
	sc_word *pos = buffer;

	bool sign       = value < 0;
	bool is_minlong = value == LONG_MIN;

	/* use absolute value, special treatment of MIN_LONG to avoid overflow */
	if (sign) {
		if (is_minlong)
			value = -(value+1);
		else
			value = -value;
	}

	sc_zero(buffer);

	while ((value != 0) && (pos < buffer + calc_buffer_size)) {
		*pos++ = value & SC_MASK;
		value >>= SC_BITS;
	}

	if (sign) {
		if (is_minlong)
			sc_inc(buffer, buffer);

		sc_neg(buffer, buffer);
	}
}

void sc_val_from_ulong(unsigned long value, sc_word *buffer)
{
	sc_word *pos = buffer;

	while (pos < buffer + calc_buffer_size) {
		*pos++ = value & SC_MASK;
		value >>= SC_BITS;
	}
}

long sc_val_to_long(const sc_word *val)
{
	unsigned long l = 0;
	unsigned max_buffer_index = (sizeof(long)*8)/SC_BITS;
	for (unsigned i = max_buffer_index; i-- > 0; ) {
		assert(l <= (ULONG_MAX>>SC_BITS) && "signed shift overflow");
		l = (l << SC_BITS) + val[i];
	}
	return l;
}

uint64_t sc_val_to_uint64(const sc_word *val)
{
	uint64_t res = 0;
	for (unsigned i = calc_buffer_size; i-- > 0; ) {
		res = (res << SC_BITS) + val[i];
	}
	return res;
}

void sc_min_from_bits(unsigned num_bits, bool sign, sc_word *buffer)
{
	if (!sign) {
		sc_zero(buffer);
		return;
	} else {
		sc_word *pos = buffer;

		unsigned bits = num_bits - 1;
		unsigned i    = 0;
		for ( ; i < bits/SC_BITS; i++)
			*pos++ = 0;

		*pos++ = min_digit(bits%SC_BITS);

		for (i++; i <= calc_buffer_size - 1; i++)
			*pos++ = SC_MASK;
	}
}

void sc_max_from_bits(unsigned num_bits, bool sign, sc_word *buffer)
{
	sc_word *pos = buffer;

	unsigned bits = num_bits - sign;
	unsigned i    = 0;
	for ( ; i < bits/SC_BITS; i++)
		*pos++ = SC_MASK;

	*pos++ = max_digit(bits%SC_BITS);

	for (i++; i <= calc_buffer_size - 1; i++)
		*pos++ = 0;
}

ir_relation sc_comp(const sc_word* const val1, const sc_word* const val2)
{
	/* compare signs first:
	 * the loop below can only compare values of the same sign! */
	bool val1_negative = sc_is_negative(val1);
	bool val2_negative = sc_is_negative(val2);
	if (val1_negative != val2_negative)
		return val1_negative ? ir_relation_less : ir_relation_greater;

	/* loop until two digits differ, the values are equal if there
	 * are no such two digits */
	unsigned counter = calc_buffer_size - 1;
	while (val1[counter] == val2[counter]) {
		if (counter == 0)
			return ir_relation_equal;
		counter--;
	}

	/* the leftmost digit is the most significant, so this returns
	 * the correct result.
	 * This implies the digit enum is ordered */
	return val1[counter] > val2[counter]
	     ? ir_relation_greater : ir_relation_less;
}

int sc_get_highest_set_bit(const sc_word *value)
{
	int high = calc_buffer_size * SC_BITS - 1;
	for (unsigned counter = calc_buffer_size; counter-- > 0; ) {
		if (value[counter] == 0)
			high -= SC_BITS;
		else {
			if (value[counter] > 7) return high;
			else if (value[counter] > 3) return high - 1;
			else if (value[counter] > 1) return high - 2;
			else return high - 3;
		}
	}
	return high;
}

int sc_get_lowest_set_bit(const sc_word *value)
{
	int low = 0;
	for (unsigned counter = 0; counter < calc_buffer_size; ++counter) {
		switch (value[counter]) {
		case 1:
		case 3:
		case 5:
		case 7:
		case 9:
		case 11:
		case 13:
		case 15:
			return low;
		case 2:
		case 6:
		case 10:
		case 14:
			return low + 1;
		case 4:
		case 12:
			return low + 2;
		case 8:
			return low + 3;
		default:
			low += SC_BITS;
		}
	}
	return -1;
}

bool sc_get_bit_at(const sc_word *value, unsigned pos)
{
	unsigned nibble = pos / SC_BITS;
	return value[nibble] & (1 << (pos % SC_BITS));
}

void sc_set_bit_at(sc_word *value, unsigned pos)
{
	unsigned nibble = pos / SC_BITS;
	value[nibble] |= 1 << (pos % SC_BITS);
}

void sc_clear_bit_at(sc_word *value, unsigned pos)
{
	unsigned nibble = pos / SC_BITS;
	value[nibble] &= ~(1 << (pos % SC_BITS));
}

bool sc_is_zero(const sc_word *value, unsigned bits)
{
	unsigned i;
	for (i = 0; i < bits/SC_BITS; ++i) {
		if (value[i] != 0)
			return false;
	}
	sc_word mask = max_digit(bits%SC_BITS);
	return mask == 0 || (value[i] & mask) == 0;
}

bool sc_is_all_one(const sc_word *value, unsigned bits)
{
	unsigned i;
	for (i = 0; i < bits/SC_BITS; ++i) {
		if (value[i] != SC_MASK)
			return false;
	}
	sc_word mask = max_digit(bits%SC_BITS);
	return mask == 0 || (value[i] & mask) == mask;
}

bool sc_is_negative(const sc_word *value)
{
	return sc_get_bit_at(value, calc_buffer_size*SC_BITS-1);
}

unsigned char sc_sub_bits(const sc_word *value, unsigned len, unsigned byte_ofs)
{
	/* the current scheme uses one byte to store a nibble */
	unsigned nibble_ofs = 2 * byte_ofs;
	if (nibble_ofs * SC_BITS >= len)
		return 0;

	unsigned char res = value[nibble_ofs];
	if (len > (nibble_ofs + 1) * SC_BITS)
		res |= value[nibble_ofs + 1] << SC_BITS;

	/* kick bits outsize */
	if (len - 8 * byte_ofs < 8) {
		res &= (1 << (len - 8 * byte_ofs)) - 1;
	}
	return res;
}

unsigned sc_popcount(const sc_word *value, unsigned bits)
{
	unsigned res = 0;
	unsigned i;
	for (i = 0; i < bits/SC_BITS; ++i) {
		res += popcount(value[i]);
	}
	sc_word mask = max_digit(bits%SC_BITS);
	if (mask != 0)
		res += popcount(value[i] & mask);

	return res;
}