int main(int argc, char **argv) {
mpi::environment env(argc, argv);
mpi::communicator world;
int np = world.size();
int pid = world.rank();
if (np < 2) std::exit(-1);
boost::mt19937 engine;
std::vector<int> nelms, offsets;
int n = 0;
for (int p = 0; p < np; ++p) {
nelms.push_back((engine() & 3) + 1);
offsets.push_back(n);
n += nelms.back();
}
std::vector<uint32_t> source;
if (world.rank() == 0) {
for (int i = 0; i < n; ++i) source.push_back(engine());
std::cout << "source: " << alps::write_vector(source) << std::endl;
}
std::vector<uint32_t> distributed;
alps::distribute_vector(world, nelms, offsets, source, distributed);
for (int p = 0; p < np; ++p) {
if (p == pid)
std::cout << "process " << p << ": " << alps::write_vector(distributed) << std::endl;
std::cout << std::flush;
world.barrier();
}
std::vector<uint32_t> collected(world.rank() == 0 ? n : 0);
alps::collect_vector(world, nelms, offsets, distributed, collected);
if (world.rank() == 0)
std::cout << "collected: " << alps::write_vector(collected) << std::endl;
if (world.rank() == 0 && source != collected) {
std::cout << "test failed\n";
std::exit(-1);
}
}
SEXP internal_loop (const base_finder * const ffptr, status (*check_self_status)(const segment&, const segment&), const check_invalid_chimera * const icptr,
SEXP pairlen, SEXP chrs, SEXP pos, SEXP flag, SEXP cigar, SEXP mapqual, SEXP chimera_strict, SEXP minqual, SEXP do_dedup) {
// Checking input values.
if (!isInteger(pairlen)) { throw std::runtime_error("length of pairs must be an integer vector"); }
if (!isInteger(chrs)) { throw std::runtime_error("chromosomes must be an integer vector"); }
if (!isInteger(pos)) { throw std::runtime_error("positions must be an integer vector"); }
if (!isInteger(flag)) { throw std::runtime_error("SAM flags must be an integer vector"); }
if (!isString(cigar)) { throw std::runtime_error("CIGAR strings must be a character vector"); }
if (!isInteger(mapqual)) { throw std::runtime_error("mapping quality must be an integer vector"); }
const int nreads=LENGTH(chrs);
if (LENGTH(pos)!=nreads || LENGTH(flag)!=nreads || LENGTH(cigar)!=nreads || LENGTH(mapqual)!=nreads) {
throw std::runtime_error("lengths of vectors of read information are not consistent");
}
if (!isLogical(chimera_strict) || LENGTH(chimera_strict)!=1) { throw std::runtime_error("chimera removal specification should be a logical scalar"); }
const int npairs=LENGTH(pairlen);
if (!isLogical(do_dedup) || LENGTH(do_dedup)!=1) { throw std::runtime_error("duplicate removal specification should be a logical scalar"); }
if (!isInteger(minqual) || LENGTH(minqual)!=1) { throw std::runtime_error("minimum mapping quality should be an integer scalar"); }
// Initializing pointers.
const int* cptr=INTEGER(chrs);
const int* pptr=INTEGER(pos);
const int* fptr=INTEGER(flag);
const int* qptr=INTEGER(mapqual);
const bool rm_invalid=asLogical(chimera_strict);
const bool rm_dup=asLogical(do_dedup);
const int minq=asInteger(minqual);
const bool rm_min=!ISNA(minq);
const int * plptr=INTEGER(pairlen);
const size_t nc=ffptr->nchrs();
// Constructing output containers
std::deque<std::deque<std::deque<valid_pair> > > collected(nc);
for (size_t i=0; i<nc; ++i) { collected[i].resize(i+1); }
std::deque<segment> read1, read2;
segment current;
valid_pair curpair;
int single=0;
int total=0, dupped=0, filtered=0, mapped=0;
int dangling=0, selfie=0;
int total_chim=0, mapped_chim=0, multi_chim=0, inv_chimeras=0;
// Running through all reads and identifying the interaction they represent.
int index=0, limit, pindex=0;
while (index < nreads) {
read1.clear();
read2.clear();
if (pindex==npairs) { throw std::runtime_error("ran out of pairs before running out of reads"); }
const int& curpl=plptr[pindex];
++pindex;
limit=index+curpl;
if (limit > nreads) { throw std::runtime_error("ran out of reads before running out of pairs"); }
// Various flags that will be needed.
bool isdup=false, isunmap=false, ischimera=false,
isfirst=false, hasfirst=false, hassecond=false,
curdup=false, curunmap=false;
// Running through and collecting read segments.
while (index < limit) {
const int& curflag=fptr[index];
current.reverse=(curflag & 0x10);
current.chrid=cptr[index];
current.pos=pptr[index];
parse_cigar(CHAR(STRING_ELT(cigar, index)), current.alen, current.offset, current.reverse);
// Checking how we should proceed; whether we should bother adding it or not.
curdup=(curflag & 0x400);
curunmap=(curflag & 0x4 || (rm_min && qptr[index] < minq));
if (current.offset==0) {
if (curdup) { isdup=true; }
if (curunmap) { isunmap=true; }
} else {
ischimera=true;
}
// Checking what it is.
isfirst = (curflag & 0x40);
if (isfirst) { hasfirst=true; }
else { hassecond=true; }
// Checking which deque to put it in, if we're going to keep it.
if (! (curdup && rm_dup) && ! curunmap) {
std::deque<segment>& current_reads=(isfirst ? read1 : read2);
if (current.offset==0) {
current_reads.push_front(current);
} else {
current_reads.push_back(current);
}
}
++index;
}
// Skipping if it's a singleton; otherwise, reporting it as part of the total read pairs.
if (! (hasfirst && hassecond)) {
++single;
continue;
}
++total;
// Adding to other statistics.
if (ischimera) { ++total_chim; }
if (isdup) { ++dupped; }
if (isunmap) { ++filtered; }
/* Skipping if unmapped, marked (and we're removing them), and if the first alignment
* of either read has any hard 5' clipping. This means that it's not truly 5' terminated
* (e.g. the actual 5' end was unmapped, duplicate removed or whatever). Note that
* not skipping UNMAP or DUP does not imply non-empty sets, as UNMAP/DUP are only set
* for 0-offset alignments; if this isn't in the file, these flags won't get set, but
* the sets can still be empty if non-zero-offset alignments are present and filtered
* (to escape the singles clause above). Thus, we need to check non-emptiness explicitly.
*/
if (isunmap || (rm_dup && isdup) || read1.empty() || read2.empty() || read1.front().offset || read2.front().offset) { continue; }
++mapped;
// Assigning fragment IDs, if everything else is good.
for (size_t i1=0; i1<read1.size(); ++i1) {
segment& current=read1[i1];
current.fragid=ffptr->find_fragment(current.chrid, current.pos, current.reverse, current.alen);
}
for (size_t i2=0; i2<read2.size(); ++i2) {
segment& current=read2[i2];
current.fragid=ffptr->find_fragment(current.chrid, current.pos, current.reverse, current.alen);
}
// Determining the type of construct if they have the same ID.
switch ((*check_self_status)(read1.front(), read2.front())) {
case ISPET:
++dangling;
continue;
case ISMATE:
++selfie;
continue;
default:
break;
}
// Pulling out chimera diagnostics.
if (ischimera) {
++mapped_chim;
++multi_chim;
bool invalid=false;
if (read1.size()==1 && read2.size()==1) {
--multi_chim;
} else if (read1.size() > 2 || read2.size() > 2) {
invalid=true;
} else {
invalid=(*icptr)(read1, read2);
}
if (invalid) {
++inv_chimeras;
if (rm_invalid) { continue; }
}
}
// Choosing the anchor segment, and reporting it.
bool anchor=false;
if (read1.front().chrid > read2.front().chrid) {
anchor=true;
} else if (read1.front().chrid==read2.front().chrid) {
if (read1.front().fragid > read2.front().fragid) {
anchor=true;
} else if (read1.front().fragid == read2.front().fragid) {
if (read1.front().pos > read2.front().pos) {
anchor=true;
}
}
}
const segment& anchor_seg=(anchor ? read1.front() : read2.front());
const segment& target_seg=(anchor ? read2.front() : read1.front());
curpair.anchor=anchor_seg.fragid;
curpair.target=target_seg.fragid;
curpair.apos=anchor_seg.pos;
curpair.alen=anchor_seg.alen;
if (anchor_seg.reverse) { curpair.alen*=-1; }
curpair.tpos=target_seg.pos;
curpair.tlen=target_seg.alen;
if (target_seg.reverse) { curpair.tlen*=-1; }
if (curpair.alen==0 || curpair.tlen==0) { throw std::runtime_error("alignment lengths of zero should not be present"); }
collected[anchor_seg.chrid][target_seg.chrid].push_back(curpair);
}
// Checking if all pairs were used up.
if (pindex!=npairs) { throw std::runtime_error("ran out of reads before running out of pairs"); }
SEXP total_output=PROTECT(allocVector(VECSXP, 6));
try {
// Checking how many are not (doubly) empty.
std::deque<std::pair<int, int> > good;
for (size_t i=0; i<nc; ++i) {
for (size_t j=0; j<=i; ++j) {
const std::deque<valid_pair>& curpairs=collected[i][j];
if (!curpairs.empty()) { good.push_back(std::make_pair(i, j)); }
}
}
SET_VECTOR_ELT(total_output, 0, allocMatrix(INTSXP, good.size(), 2));
int* aptr=INTEGER(VECTOR_ELT(total_output, 0));
int* tptr=aptr+good.size();
SET_VECTOR_ELT(total_output, 1, allocVector(VECSXP, good.size()));
SEXP output=VECTOR_ELT(total_output, 1);
for (size_t i=0; i<good.size(); ++i) {
aptr[i]=good[i].first+1;
tptr[i]=good[i].second+1;
// Filling up those non-empty pairs of chromosomes.
std::deque<valid_pair>& curpairs=collected[good[i].first][good[i].second];
SET_VECTOR_ELT(output, i, allocMatrix(INTSXP, curpairs.size(), 6));
int* axptr=INTEGER(VECTOR_ELT(output, i));
int* txptr=axptr+curpairs.size();
int* apxptr=txptr+curpairs.size();
int* tpxptr=apxptr+curpairs.size();
int* afxptr=tpxptr+curpairs.size();
int* tfxptr=afxptr+curpairs.size();
for (size_t k=0; k<curpairs.size(); ++k) {
axptr[k]=curpairs[k].anchor+1;
txptr[k]=curpairs[k].target+1;
apxptr[k]=curpairs[k].apos;
tpxptr[k]=curpairs[k].tpos;
afxptr[k]=curpairs[k].alen;
tfxptr[k]=curpairs[k].tlen;
}
// Emptying out the container once we've processed it, to keep memory usage down.
std::deque<valid_pair>().swap(curpairs);
}
// Dumping mapping diagnostics.
SET_VECTOR_ELT(total_output, 2, allocVector(INTSXP, 4));
int* dptr=INTEGER(VECTOR_ELT(total_output, 2));
dptr[0]=total;
dptr[1]=dupped;
dptr[2]=filtered;
dptr[3]=mapped;
// Dumping the number of dangling ends, self-circles.
SET_VECTOR_ELT(total_output, 3, allocVector(INTSXP, 2));
int * siptr=INTEGER(VECTOR_ELT(total_output, 3));
siptr[0]=dangling;
siptr[1]=selfie;
// Dumping the number designated 'single', as there's no pairs.
SET_VECTOR_ELT(total_output, 4, ScalarInteger(single));
// Dumping chimeric diagnostics.
SET_VECTOR_ELT(total_output, 5, allocVector(INTSXP, 4));
int* cptr=INTEGER(VECTOR_ELT(total_output, 5));
cptr[0]=total_chim;
cptr[1]=mapped_chim;
cptr[2]=multi_chim;
cptr[3]=inv_chimeras;
} catch (std::exception& e) {
UNPROTECT(1);
throw;
}
UNPROTECT(1);
return total_output;
}
