template <int partition=0> inline ftype pullCapacity(int dim) const {
assert_leq(alpha[dim], -edges[dim]);
assert_geq(alpha[dim], edges[dim]);
// Actually the amount that can be pushed to the other node, so in
// the 1-partition it's
return -edges[dim] + ( (partition == 1) ? 1 : -1) * alpha[dim];
}
/**
* Sanity-check various pieces of the Ebwt
*/
void Ebwt::sanityCheckAll(int reverse) const {
const EbwtParams& eh = this->_eh;
assert(isInMemory());
// Check ftab
for(uint32_t i = 1; i < eh._ftabLen; i++) {
assert_geq(this->ftabHi(i), this->ftabLo(i-1));
assert_geq(this->ftabLo(i), this->ftabHi(i-1));
assert_leq(this->ftabHi(i), eh._bwtLen+1);
}
assert_eq(this->ftabHi(eh._ftabLen-1), eh._bwtLen);
// Check offs
int seenLen = (eh._bwtLen + 31) >> 5;
uint32_t *seen;
try {
seen = new uint32_t[seenLen]; // bitvector marking seen offsets
} catch(bad_alloc& e) {
cerr << "Out of memory allocating seen[] at " << __FILE__ << ":" << __LINE__ << endl;
throw e;
}
memset(seen, 0, 4 * seenLen);
uint32_t offsLen = eh._offsLen;
for(uint32_t i = 0; i < offsLen; i++) {
assert_lt(this->offs()[i], eh._bwtLen);
int w = this->offs()[i] >> 5;
int r = this->offs()[i] & 31;
assert_eq(0, (seen[w] >> r) & 1); // shouldn't have been seen before
seen[w] |= (1 << r);
}
delete[] seen;
// Check nPat
assert_gt(this->_nPat, 0);
// Check plen, flen
for(uint32_t i = 0; i < this->_nPat; i++) {
assert_geq(this->plen()[i], 0);
}
// Check rstarts
if(this->rstarts() != NULL) {
for(uint32_t i = 0; i < this->_nFrag-1; i++) {
assert_gt(this->rstarts()[(i+1)*3], this->rstarts()[i*3]);
if(reverse == REF_READ_REVERSE) {
assert(this->rstarts()[(i*3)+1] >= this->rstarts()[((i+1)*3)+1]);
} else {
assert(this->rstarts()[(i*3)+1] <= this->rstarts()[((i+1)*3)+1]);
}
}
}
// Check ebwt
sanityCheckUpToSide(eh._numSides);
VMSG_NL("Ebwt::sanityCheck passed");
}
inline void pushExcess(const node_ptr& src, const node_ptr& dest, const DirDim& dd, ftype amount) {
int direction = dd.direction();
int dim = dd.dimension();
#ifndef NDEBUG
assert(direction == -1 || direction == 1);
assert(dest == lattice.neighbor(src, dim, direction));
assert_equal(src->on(), partition);
assert_equal(dest->on(), partition);
assert_geq(dim, 0);
assert_leq(amount, pushCapacity<partition>(src, dest, dd));
assert_leq(-amount, pushCapacity<partition>(dest, src, dd.reversed()));
assert(direction == 1 || direction == -1);
_debugVerifyNodeReduction(src);
_debugVerifyNodeReduction(dest);
ftype src_pr_excess = src->template excess<partition>();
ftype dest_pr_excess = dest->template excess<partition>();
#endif
src->reduction -= ((partition == 1) ? 1 : -1) * amount;
dest->reduction += ((partition == 1) ? 1 : -1) * amount;
((direction > 0) ? src : dest)
->alpha[dim] += direction * ((partition == 1) ? 1 : -1) * amount;
#ifndef NDEBUG
_debugVerifyNodeReduction(src);
_debugVerifyNodeReduction(dest);
ftype src_af_excess = src->template excess<partition>();
ftype dest_af_excess = dest->template excess<partition>();
assert_equal(src_pr_excess - amount, src_af_excess);
assert_equal(dest_pr_excess + amount, dest_af_excess);
#endif
}
void Scheduler::_handleButtonPress( int updateElId, ButtonType type, int floor ) {
// first setup lights
Command lights{ type == ButtonType::CallUp
? CommandType::TurnOnLightUp : CommandType::TurnOnLightDown,
Command::ANY_ID, floor };
_forwardToTargets( lights );
// each elevator schedules changes originating from it
if ( updateElId == _localElevId ) {
// now find optimal elevator
int minDistance = INT_MAX;
int minId = INT_MIN;
for ( auto &statepair : _globalState.elevators() ) {
const ElevatorState &state = statepair.second;
int dist = std::abs( state.lastFloor - floor );
// penalizations for non-idle elevators
if ( state.stopped )
dist += 10 * (_bounds.maxFloor() - _bounds.minFloor()); // stopped penalization
if ( ( state.direction == Direction::Up
&& ( (type == ButtonType::CallUp && floor > state.lastFloor )
|| floor == _bounds.maxFloor() ) )
||
( state.direction == Direction::Down
&& ( (type == ButtonType::CallDown && floor < state.lastFloor )
|| floor == _bounds.minFloor() ) )
)
dist += _bounds.maxFloor() - _bounds.minFloor() + 1; // busy penalization
else if ( state.direction != Direction::None ) // not idle
// as a last resort we can schedule floor even to elevator which is
// running in different direction
dist += 2 * (_bounds.maxFloor() - _bounds.minFloor() + 1); // penalization
if ( dist < minDistance ) {
minDistance = dist;
minId = state.id;
}
}
assert_leq( 0, minId, "no minimal distance found" );
Command comm{ type == ButtonType::CallUp
? CommandType::CallToFloorAndGoUp
: CommandType::CallToFloorAndGoDown,
minId, floor };
_forwardToTargets( comm );
}
}
/**
* Check that the ebwt array is internally consistent up to (and not
* including) the given side index by re-counting the chars and
* comparing against the embedded occ[] arrays.
*/
void Ebwt::sanityCheckUpToSide(int upToSide) const {
assert(isInMemory());
uint32_t occ[] = {0, 0, 0, 0};
ASSERT_ONLY(uint32_t occ_save[] = {0, 0, 0, 0});
uint32_t cur = 0; // byte pointer
const EbwtParams& eh = this->_eh;
bool fw = false;
while(cur < (upToSide * eh._sideSz)) {
assert_leq(cur + eh._sideSz, eh._ebwtTotLen);
for(uint32_t i = 0; i < eh._sideBwtSz; i++) {
uint8_t by = this->ebwt()[cur + (fw ? i : eh._sideBwtSz-i-1)];
for(int j = 0; j < 4; j++) {
// Unpack from lowest to highest bit pair
int twoBit = unpack_2b_from_8b(by, fw ? j : 3-j);
occ[twoBit]++;
}
assert_eq(0, (occ[0] + occ[1] + occ[2] + occ[3]) % 4);
}
assert_eq(0, (occ[0] + occ[1] + occ[2] + occ[3]) % eh._sideBwtLen);
// Finished forward bucket; check saved [A], [C], [G] and [T]
// against the uint32_ts encoded here
ASSERT_ONLY(const uint32_t *u32ebwt = reinterpret_cast<const uint32_t*>(&ebwt()[cur + eh._sideBwtSz]));
ASSERT_ONLY(uint32_t as = u32ebwt[0]);
ASSERT_ONLY(uint32_t cs = u32ebwt[1]);
ASSERT_ONLY(uint32_t gs = u32ebwt[2]);
ASSERT_ONLY(uint32_t ts = u32ebwt[3]);
assert(as == occ_save[0] || as == occ_save[0]-1);
assert_eq(cs, occ_save[1]);
assert_eq(gs, occ_save[2]);
assert_eq(ts, occ_save[3]);
#ifndef NDEBUG
occ_save[0] = occ[0];
occ_save[1] = occ[1];
occ_save[2] = occ[2];
occ_save[3] = occ[3];
#endif
cur += eh._sideSz;
}
}
static void driver(const string& infile,
vector<string>& infiles,
const string& outfile,
bool reverse = false)
{
vector<FileBuf*> is;
bool bisulfite = false;
RefReadInParams refparams(color, reverse ? reverseType : REF_READ_FORWARD, nsToAs, bisulfite);
assert_gt(infiles.size(), 0);
if(format == CMDLINE) {
// Adapt sequence strings to stringstreams open for input
stringstream *ss = new stringstream();
for(size_t i = 0; i < infiles.size(); i++) {
(*ss) << ">" << i << endl << infiles[i] << endl;
}
FileBuf *fb = new FileBuf(ss);
assert(fb != NULL);
assert(!fb->eof());
assert(fb->get() == '>');
ASSERT_ONLY(fb->reset());
assert(!fb->eof());
is.push_back(fb);
} else {
// Adapt sequence files to ifstreams
for(size_t i = 0; i < infiles.size(); i++) {
FILE *f = fopen(infiles[i].c_str(), "rb");
if (f == NULL) {
cerr << "Error: could not open "<< infiles[i] << endl;
throw 1;
}
FileBuf *fb = new FileBuf(f);
assert(fb != NULL);
assert(!fb->eof());
assert(fb->get() == '>');
ASSERT_ONLY(fb->reset());
assert(!fb->eof());
is.push_back(fb);
}
}
// Vector for the ordered list of "records" comprising the input
// sequences. A record represents a stretch of unambiguous
// characters in one of the input sequences.
vector<RefRecord> szs;
vector<uint32_t> plens;
std::pair<size_t, size_t> sztot;
{
if(verbose) cout << "Reading reference sizes" << endl;
Timer _t(cout, " Time reading reference sizes: ", verbose);
if(!reverse && (writeRef || justRef)) {
// For forward reference, dump it to .3.ebwt and .4.ebwt
// files
string file3 = outfile + ".3." + gEbwt_ext;
string file4 = outfile + ".4." + gEbwt_ext;
// Open output stream for the '.3.ebwt' file which will
// hold the size records.
ofstream fout3(file3.c_str(), ios::binary);
if(!fout3.good()) {
cerr << "Could not open index file for writing: \"" << file3 << "\"" << endl
<< "Please make sure the directory exists and that permissions allow writing by" << endl
<< "Bowtie." << endl;
throw 1;
}
BitpairOutFileBuf bpout(file4.c_str());
// Read in the sizes of all the unambiguous stretches of
// the genome into a vector of RefRecords. The input
// streams are reset once it's done.
writeU<int32_t>(fout3, 1, bigEndian); // endianness sentinel
if(color) {
refparams.color = false;
// Make sure the .3.ebwt and .4.ebwt files contain
// nucleotides; not colors
TIndexOff numSeqs = 0;
fastaRefReadSizes(is, szs, plens, refparams, &bpout, numSeqs);
refparams.color = true;
writeU<TIndexOffU>(fout3, (TIndexOffU)szs.size(), bigEndian); // write # records
for(size_t i = 0; i < szs.size(); i++) {
szs[i].write(fout3, bigEndian);
}
szs.clear();
plens.clear();
// Now read in the colorspace size records; these are
// the ones that were indexed
TIndexOff numSeqs2 = 0;
sztot = fastaRefReadSizes(is, szs, plens, refparams, NULL, numSeqs2);
assert_geq(numSeqs, numSeqs2);
} else {
TIndexOff numSeqs = 0;
sztot = fastaRefReadSizes(is, szs, plens, refparams, &bpout, numSeqs);
writeU<TIndexOffU>(fout3, (TIndexOffU)szs.size(), bigEndian); // write # records
for(size_t i = 0; i < szs.size(); i++) szs[i].write(fout3, bigEndian);
}
if(sztot.first == 0) {
cerr << "Error: No unambiguous stretches of characters in the input. Aborting..." << endl;
throw 1;
}
assert_gt(sztot.first, 0);
assert_gt(sztot.second, 0);
bpout.close();
fout3.close();
#ifndef NDEBUG
if(sanityCheck) {
BitPairReference bpr(
outfile, // ebwt basename
color, // expect color?
true, // sanity check?
&infiles,// files to check against
NULL, // sequences to check against
format == CMDLINE, // whether infiles contains strings
true, // load sequence?
false, // use memory-mapped files
false, // use shared memory
false, // sweep through memory-mapped memory
false, // be talkative
false); // be talkative
}
#endif
} else {
// Read in the sizes of all the unambiguous stretches of the
// genome into a vector of RefRecords
TIndexOff numSeqs = 0;
sztot = fastaRefReadSizes(is, szs, plens, refparams, NULL, numSeqs);
#ifndef NDEBUG
if(refparams.color) {
refparams.color = false;
vector<RefRecord> szs2;
vector<uint32_t> plens2;
TIndexOff numSeqs2 = 0;
fastaRefReadSizes(is, szs2, plens2, refparams, NULL, numSeqs2);
assert_leq(numSeqs, numSeqs2);
// One less color than base
refparams.color = true;
}
#endif
}
}
if(justRef) return;
assert_gt(sztot.first, 0);
assert_gt(sztot.second, 0);
assert_gt(szs.size(), 0);
// Construct Ebwt from input strings and parameters
Ebwt<TStr> ebwt(refparams.color ? 1 : 0,
lineRate,
linesPerSide,
offRate, // suffix-array sampling rate
-1, // ISA sampling rate
ftabChars, // number of chars in initial arrow-pair calc
nthreads,
outfile, // basename for .?.ebwt files
!reverse, // fw
!entireSA, // useBlockwise
bmax, // block size for blockwise SA builder
bmaxMultSqrt, // block size as multiplier of sqrt(len)
bmaxDivN, // block size as divisor of len
noDc? 0 : dcv,// difference-cover period
is, // list of input streams
szs, // list of reference sizes
plens, // list of not-all-gap reference sequence lengths
(TIndexOffU)sztot.first, // total size of all unambiguous ref chars
refparams, // reference read-in parameters
seed, // pseudo-random number generator seed
-1, // override offRate
-1, // override isaRate
verbose, // be talkative
autoMem, // pass exceptions up to the toplevel so that we can adjust memory settings automatically
sanityCheck); // verify results and internal consistency
// Note that the Ebwt is *not* resident in memory at this time. To
// load it into memory, call ebwt.loadIntoMemory()
if(verbose) {
// Print Ebwt's vital stats
ebwt.eh().print(cout);
}
if(sanityCheck) {
// Try restoring the original string (if there were
// multiple texts, what we'll get back is the joined,
// padded string, not a list)
ebwt.loadIntoMemory(
refparams.color ? 1 : 0,
-1,
false,
false);
TStr s2; ebwt.restore(s2);
ebwt.evictFromMemory();
{
TStr joinedss = Ebwt<TStr>::join(
is, // list of input streams
szs, // list of reference sizes
(TIndexOffU)sztot.first, // total size of all unambiguous ref chars
refparams, // reference read-in parameters
seed); // pseudo-random number generator seed
if(refparams.reverse == REF_READ_REVERSE) {
reverseInPlace(joinedss);
}
assert_eq(length(joinedss), length(s2));
assert_eq(joinedss, s2);
}
if(verbose) {
if(length(s2) < 1000) {
cout << "Passed restore check: " << s2 << endl;
} else {
cout << "Passed restore check: (" << length(s2) << " chars)" << endl;
}
}
}
}
