Skip to content

Commit a91718a

Browse files
author
Daniel Kroening
committed
switch representation of bit-vectors to hex
Simultaneously leading zeros are dropped. This reduces the memory consumption of bit-vectors by 4x or more.
1 parent 45f0133 commit a91718a

File tree

3 files changed

+99
-20
lines changed

3 files changed

+99
-20
lines changed

regression/invariants/invariant-irep-diagnostic/test.desc

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -5,5 +5,5 @@ invariant-with-irep-diagnostics
55
^SIGNAL=0$
66
--- begin invariant violation report ---
77
Invariant check failed
8-
value: 00000000000000000000000000001000
9-
value: 00000000000000000000000000001101
8+
value: 8
9+
value: [dD]

src/solvers/flattening/boolbv_constant.cpp

Lines changed: 0 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -81,14 +81,6 @@ bvt boolbvt::convert_constant(const constant_exprt &expr)
8181
{
8282
const auto &value = expr.get_value();
8383

84-
if(value.size() != width)
85-
{
86-
error().source_location=expr.find_source_location();
87-
error() << "wrong value length in constant: "
88-
<< expr.pretty() << eom;
89-
throw 0;
90-
}
91-
9284
for(std::size_t i=0; i<width; i++)
9385
{
9486
const bool bit = get_bvrep_bit(value, width, i);

src/util/arith_tools.cpp

Lines changed: 97 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -14,6 +14,8 @@ Author: Daniel Kroening, [email protected]
1414
#include "std_types.h"
1515
#include "std_expr.h"
1616

17+
#include <algorithm>
18+
1719
bool to_integer(const exprt &expr, mp_integer &int_value)
1820
{
1921
if(!expr.is_constant())
@@ -285,11 +287,35 @@ bool get_bvrep_bit(
285287
std::size_t width,
286288
std::size_t bit_index)
287289
{
288-
// The representation is binary, using '0'/'1',
289-
// most significant bit first.
290290
PRECONDITION(bit_index < width);
291-
PRECONDITION(src.size() == width);
292-
return src[src.size() - 1 - bit_index] == '1';
291+
292+
// The representation is hex, i.e., four bits per letter,
293+
// most significant nibble first.
294+
const auto nibble_index = bit_index / 4;
295+
296+
if(nibble_index >= src.size())
297+
return false;
298+
299+
const char nibble = src[src.size() - 1 - nibble_index];
300+
301+
DATA_INVARIANT(isxdigit(nibble), "bvrep is hexadecimal");
302+
303+
const unsigned char nibble_value =
304+
isdigit(nibble) ? nibble - '0'
305+
: islower(nibble) ? nibble - 'a' + 10 : nibble - 'A' + 10;
306+
307+
return ((nibble_value >> (bit_index % 4)) & 1) != 0;
308+
}
309+
310+
/// turn a value 0...15 into '0'-'9', 'A'-'Z'
311+
static char nibble2hex(unsigned char nibble)
312+
{
313+
PRECONDITION(nibble <= 0xf);
314+
315+
if(nibble >= 10)
316+
return 'A' + nibble - 10;
317+
else
318+
return '0' + nibble;
293319
}
294320

295321
/// construct a bit-vector representation from a functor
@@ -299,12 +325,37 @@ bool get_bvrep_bit(
299325
irep_idt
300326
make_bvrep(const std::size_t width, const std::function<bool(std::size_t)> f)
301327
{
302-
std::string result(width, ' ');
328+
std::string result;
329+
result.reserve((width + 3) / 4);
330+
unsigned char nibble = 0;
303331

304332
for(std::size_t i = 0; i < width; i++)
305-
result[width - 1 - i] = f(i) ? '1' : '0';
333+
{
334+
const auto bit_in_nibble = i % 4;
306335

307-
return result;
336+
nibble |= ((unsigned char)f(i)) << bit_in_nibble;
337+
338+
if(bit_in_nibble == 3)
339+
{
340+
result += nibble2hex(nibble);
341+
nibble = 0;
342+
}
343+
}
344+
345+
if(nibble != 0)
346+
result += nibble2hex(nibble);
347+
348+
// drop leading zeros
349+
const std::size_t pos = result.find_last_not_of('0');
350+
if(pos != std::string::npos)
351+
result.resize(pos + 1);
352+
353+
std::reverse(result.begin(), result.end());
354+
355+
if(result.empty())
356+
return ID_0;
357+
else
358+
return result;
308359
}
309360

310361
/// perform a binary bit-wise operation, given as a functor,
@@ -342,14 +393,50 @@ irep_idt bvrep_bitwise_op(
342393
}
343394

344395
/// convert an integer to bit-vector representation with given width
396+
/// This uses two's complement for negative numbers.
397+
/// If the value is out of range, it is 'wrapped around'.
345398
irep_idt integer2bvrep(const mp_integer &src, std::size_t width)
346399
{
347-
return integer2binary(src, width);
400+
const mp_integer p = power(2, width);
401+
402+
if(src.is_negative())
403+
{
404+
// do two's complement encoding of negative numbers
405+
mp_integer tmp = src;
406+
tmp.negate();
407+
tmp %= p;
408+
if(tmp != 0)
409+
tmp = p - tmp;
410+
return integer2string(tmp, 16);
411+
}
412+
else
413+
{
414+
// we 'wrap around' if 'src' is too large
415+
return integer2string(src % p, 16);
416+
}
348417
}
349418

350419
/// convert a bit-vector representation (possibly signed) to integer
351420
mp_integer bvrep2integer(const irep_idt &src, std::size_t width, bool is_signed)
352421
{
353-
PRECONDITION(src.size() == width);
354-
return binary2integer(id2string(src), is_signed);
422+
if(is_signed)
423+
{
424+
PRECONDITION(width >= 1);
425+
const auto tmp = string2integer(id2string(src), 16);
426+
const auto p = power(2, width - 1);
427+
if(tmp >= p)
428+
{
429+
const auto result = tmp - 2 * p;
430+
PRECONDITION(result >= -p);
431+
return result;
432+
}
433+
else
434+
return tmp;
435+
}
436+
else
437+
{
438+
const auto result = string2integer(id2string(src), 16);
439+
PRECONDITION(result < power(2, width));
440+
return result;
441+
}
355442
}

0 commit comments

Comments
 (0)