private void appendLiteral() {
if (vec.size() == 0) {
vec.add(active.val);
} else {
int back = vec.getQuick(vec.size() - 1);
if (active.val == 0) { // incoming word is zero
if (back == 0) {
setBack(HEADER0 + 2);
} else if (isZeroFill(back)) {
setBack(++back);
} else {
vec.add(active.val);
}
} else if (active.val == ALLONES) { // incoming word is allones
if (back == ALLONES) {
setBack((HEADER1 | 2));
} else if (isOneFill(back)) {
setBack(++back);
} else {
vec.add(active.val);
}
} else { // incoming word contains a mixture of bits
vec.add(active.val);
}
}
nbits += MAXBITS;
active.reset();
nset = 0;
}
/**
* Checks if the bit is set in the compressed bitset.
*
* <p>This operation is considerably slower than the uncompressed version. Use it with care.
*
* @param i the index where bit is checked.
* @return true if the index is set, false otherwise.
*/
public boolean get(int i) {
if (i > numBits()) { // no-way
return false;
} else { // we have to do hard way
for (int j = 0; j < vec.size() && i >= 0; j++) {
int v = vec.getQuick(j);
// for a fill, we know easily.
if (isAFill(v)) {
int size = (v & MAXCNT) * MAXBITS;
if (i < size) return isOneFill(v);
i -= size;
} else {
// plain value.
if (i < MAXBITS) { // do we need to check?
return (1 << (MAXBITS - i - 1) & v) != 0;
} else { // yet to get there.
i -= MAXBITS;
}
}
}
// we need to check the active word.
if (i >= 0) {
return (moreThanInUnsigned(active.val << (MAXBITS + 1 - active.nbits + i), ALLONES));
}
}
return false;
}
private WAHBitSet genericOp(OpType op, WAHBitSet other) {
WAHBitSet ret = new WAHBitSet();
// ensure that they have the same bit length.
if (this.numBits() < other.numBits()) {
this.setBit(other.numBits() - 1, 0);
} else if (this.numBits() > other.numBits()) {
other.setBit(numBits() - 1, 0);
}
if (vec.size() > 0) {
run xrun = new run(vec), yrun = new run(other.vec);
xrun.decode();
yrun.decode();
do {
if (xrun.nWords == 0) {
xrun.inc();
xrun.decode();
}
if (yrun.nWords == 0) {
yrun.inc();
yrun.decode();
}
if (xrun.isFill()) {
if (yrun.isFill()) {
int nWords = Math.min(xrun.nWords, yrun.nWords);
ret.appendFill(nWords, getOpResult(op, xrun.fillWord, yrun.fillWord));
xrun.nWords -= nWords;
yrun.nWords -= nWords;
} else {
ret.active.val = getOpResult(op, xrun.fillWord, yrun.get());
ret.appendLiteral();
--xrun.nWords;
yrun.nWords = 0;
}
} else if (yrun.isFill()) {
ret.active.val = getOpResult(op, yrun.fillWord, xrun.get());
ret.appendLiteral();
yrun.nWords--;
xrun.nWords = 0;
} else {
ret.active.val = getOpResult(op, xrun.get(), yrun.get());
ret.appendLiteral();
yrun.nWords = 0;
xrun.nWords = 0;
}
} while (!(xrun.end() && yrun.end()));
}
ret.active.val = getOpResult(op, this.active.val, other.active.val);
ret.active.nbits = this.active.nbits;
ret.doCount();
return ret;
}
/**
* This is an optimization over the and function. This does not create new bitset. This just
* counts the number of 1 bits common between the two bitsets.
*
* @param other the bitset to and with.
* @return the number of 1s common between two bitsets.
*/
public int andSize(WAHBitSet other) {
int size = 0;
// ensure that they have the same bit length.
if (this.numBits() < other.numBits()) {
this.setBit(other.numBits() - 1, 0);
} else if (this.numBits() > other.numBits()) {
other.setBit(numBits() - 1, 0);
}
if (vec.size() > 0) {
run xrun = new run(vec), yrun = new run(other.vec);
xrun.decode();
yrun.decode();
do {
if (xrun.nWords == 0) {
xrun.inc();
xrun.decode();
}
if (yrun.nWords == 0) {
yrun.inc();
yrun.decode();
}
if (xrun.isFill()) {
if (yrun.isFill()) {
int nWords = Math.min(xrun.nWords, yrun.nWords);
if ((xrun.fillWord & yrun.fillWord) == 1) {
size += nWords * MAXBITS;
}
xrun.nWords -= nWords;
yrun.nWords -= nWords;
} else {
size += countInRun(xrun, yrun);
}
} else if (yrun.isFill()) {
size += countInRun(yrun, xrun);
} else {
int val = xrun.get() & yrun.get();
if (val > 0) size += Integer.bitCount(val);
yrun.nWords = 0;
xrun.nWords = 0;
}
} while (!(xrun.end() && yrun.end()));
}
size += Integer.bitCount(this.active.val & other.active.val);
return size;
}
private int doCount() {
nset = 0;
nbits = 0;
for (int i = 0; i < vec.size(); i++) {
int v = vec.getQuick(i);
if (!isAFill(v)) {
nbits += MAXBITS;
nset += Integer.bitCount(v);
} else {
int tmp = (v & MAXCNT) * MAXBITS;
nbits += tmp;
nset += tmp * (isOneFill(v) ? 1 : 0);
}
}
return nbits;
}
public boolean end() {
return idx >= vec.size() - 1;
}
private void setBack(int val) {
vec.setQuick(vec.size() - 1, val);
}
private int getBack() {
return vec.getQuick(vec.size() - 1);
}
private void setBit(int ind, int val) {
assert val == 0 || val == 1;
if (ind >= numBits()) {
int diff = ind - numBits() + 1;
if (active.nbits > 0) {
if (ind + 1 >= nbits + MAXBITS) {
diff -= MAXBITS - active.nbits;
active.val <<= (MAXBITS - active.nbits);
if (diff == 0) active.val += (val != 0 ? 1 : 0);
appendLiteral();
} else {
active.nbits += diff;
active.val <<= diff;
active.val += (val != 0 ? 1 : 0);
diff = 0;
}
}
if (diff != 0) {
int w = diff / MAXBITS;
diff -= w * MAXBITS;
if (diff != 0) {
if (w > 1) {
appendCounter(0, w);
} else if (w != 0) {
appendLiteral();
}
active.nbits = diff;
active.val += (val != 0 ? 1 : 0);
} else if (val != 0) {
if (w > 2) {
appendCounter(0, w - 1);
} else if (w == 2) {
appendLiteral();
}
active.val = 1;
appendLiteral();
} else {
if (w > 1) {
appendCounter(0, w);
} else if (w != 0) {
appendLiteral();
}
}
}
if (numBits() != ind + 1) logger.warning("Warning");
if (nset != 0) nset += (val != 0 ? 1 : 0);
return;
} else if (ind >= nbits) { // modify an active bit
int u = active.val;
if (val != 0) {
active.val |= (1 << (active.nbits - (ind - nbits) - 1));
} else {
active.val &= ~(1 << (active.nbits - (ind - nbits) - 1));
}
if (nset != 0 && (u != active.val)) nset += (val != 0 ? 1 : -1);
return;
} else if (vec.size() * MAXBITS == nbits) { // uncompressed
int i = ind / MAXBITS;
int u = vec.get(i);
int w = (1 << (SECONDBIT - (ind % MAXBITS)));
if (val != 0) vec.setQuick(i, u |= w);
else vec.setQuick(i, u &= ~w);
if (nset != 0 && (vec.getQuick(i) != u)) nset += (val != 0 ? 1 : -1);
return;
}
// the code after this has not been verified at all...
// should proceed with caution.
// compressed bit vector --
// the bit to be modified is in vec
if (RUN_UNTESTED_CODE) {
int idx = 0;
int compressed = 0, cnt = 0, ind1 = 0, ind0 = ind;
int current = 0; // current bit value
while ((ind0 > 0) && (idx < vec.size())) {
int v = vec.getQuick(idx);
if (isAFill(v)) { // a fill
cnt = ((v) & MAXCNT) * MAXBITS;
if (cnt > ind0) { // found the location
current = (isOneFill(v) ? 1 : 0);
compressed = 1;
ind1 = ind0;
ind0 = 0;
} else {
ind0 -= cnt;
ind1 = ind0;
++idx;
}
} else { // a literal word
cnt = MAXBITS;
if (MAXBITS > ind0) { // found the location
current = (1 & ((v) >>> (SECONDBIT - ind0)));
compressed = 0;
ind1 = ind0;
ind0 = 0;
} else {
ind0 -= MAXBITS;
ind1 = ind0;
++idx;
}
}
} // while (ind...
if (ind1 == 0) { // set current and compressed
int v = vec.getQuick(idx);
if (isAFill(v)) {
cnt = (v & MAXCNT) * MAXBITS;
current = (isOneFill(v) ? 1 : 0);
compressed = 1;
} else {
cnt = MAXBITS;
current = (v >>> SECONDBIT);
compressed = 0;
}
}
if (ind0 > 0) // has not found the right location yet.
{
if (ind0 < active.nbits) { // in the active word
ind1 = (1 << (active.nbits - ind0 - 1));
if (val != 0) {
active.val |= ind1;
} else {
active.val &= ~ind1;
}
} else { // extends the current bit vector
ind1 = ind0 - active.nbits - 1;
appendWord(HEADER0 | (ind1 / MAXBITS));
for (ind1 %= MAXBITS; ind1 > 0; --ind1) addOneBit(0);
addOneBit(val != 0 ? 1 : 0);
}
if (nset != 0) nset += val != 0 ? 1 : -1;
return;
}
// locate the bit to be changed, lots of work hidden here
if (current == val) return; // nothing to do
int v = vec.getQuick(idx);
// need to actually modify the bit
if (compressed == 0) {
// toggle a single bit of a literal word
v ^= (1 << (SECONDBIT - ind1));
vec.setQuick(idx, v);
} else if (ind1 < MAXBITS) {
// bit to be modified is in the first word, two pieces
--v;
vec.set(idx, v);
if ((v & MAXCNT) == 1) {
v = (current != 0) ? ALLONES : 0;
vec.setQuick(idx, v);
}
int w = 1 << (SECONDBIT - ind1);
if (val == 0) w ^= ALLONES;
vec.beforeInsert(idx, w);
idx++;
} else if (cnt - ind1 <= MAXBITS) {
// bit to be modified is in the last word, two pieces
--(v);
vec.setQuick(idx, v);
if ((v & MAXCNT) == 1) {
v = (current != 0) ? ALLONES : 0;
vec.setQuick(idx, v);
}
int w = 1 << (cnt - ind1 - 1);
if (val == 0) w ^= ALLONES;
++idx;
vec.beforeInsert(idx, w);
} else { // the counter breaks into three pieces
int u[] = new int[2], w;
u[0] = ind1 / MAXBITS;
w = (v & MAXCNT) - u[0] - 1;
u[1] = 1 << (SECONDBIT - ind1 + u[0] * MAXBITS);
if (val == 0) {
u[0] = (u[0] > 1) ? (HEADER1 | u[0]) : (ALLONES);
u[1] ^= ALLONES;
w = (w > 1) ? (HEADER1 | w) : (ALLONES);
} else {
u[0] = (u[0] > 1) ? (HEADER0 | u[0]) : 0;
w = (w > 1) ? (HEADER0 | w) : 0;
}
vec.setQuick(idx, w);
vec.beforeInsertAllOf(idx, Arrays.asList(u));
}
if (nset != 0) nset += val != 0 ? 1 : -1;
} else {
throw new AssertionError("Untested code detected, would rather die than run this");
}
}
/**
* Returns a new WAH compressed bitset after anding the current bitset with the <i>other</i>
* bitset.
*
* @param other the bitset to and with
* @return The resulting bitset
*/
public WAHBitSet and(WAHBitSet other) {
WAHBitSet ret = new WAHBitSet();
// ensure that they have the same bit length.
if (this.numBits() < other.numBits()) {
this.setBit(other.numBits() - 1, 0);
} else if (this.numBits() > other.numBits()) {
other.setBit(numBits() - 1, 0);
}
// if there is something in the vector.
if (vec.size() > 0) {
// create new run objects and decode them.
run xrun = new run(vec), yrun = new run(other.vec);
xrun.decode();
yrun.decode();
do {
// if you finished a run, then get the next one.
if (xrun.nWords == 0) {
xrun.inc();
xrun.decode();
}
if (yrun.nWords == 0) {
yrun.inc();
yrun.decode();
}
if (xrun.isFill()) {
if (yrun.isFill()) {
// both are fills... this is the best.
int nWords = Math.min(xrun.nWords, yrun.nWords);
ret.appendFill(nWords, xrun.fillWord & yrun.fillWord);
xrun.nWords -= nWords;
yrun.nWords -= nWords;
} else {
// just cut through the other run
chewUpRun(xrun, ret, yrun);
}
} else if (yrun.isFill()) {
// again do the same, with different order.
chewUpRun(yrun, ret, xrun);
} else {
// both are literals, so get the new literal and
// append it to the return value.
ret.active.val = xrun.get() & yrun.get();
ret.appendLiteral();
yrun.nWords = 0;
xrun.nWords = 0;
}
// till they are not at the end.
} while (!(xrun.end() && yrun.end()));
}
// set the active word.
ret.active.val = this.active.val & other.active.val;
ret.active.nbits = this.active.nbits;
// ensure that the counts are set properly.
ret.doCount();
return ret;
}
/**
* Returns the amount of memory used by the compressed bit set
*
* @return the amount of memory used by the compressed bit set
*/
public long memSize() {
return vec.size() + 2;
}