/* Function: p7_coords2_hash_Reuse()
* Synopsis: Reuse a <P7_COORDS2>
*
* Purpose: Clear a <P7_COORDS2_HASH> hash table for reuse.
*
* If any allocations are overly large, drop them
* back to 'redline' values. Default redlines
* are 1024 keys (i.e. different coord pair arrays),
* 1024 hash values, and 16384 total integers of
* raw data. Redlines are all 8x the default
* initial allocations.
*
* Args: ch : hash table to reuse
*
* Returns: <eslOK> on success
*
* Throws: <eslEMEM> on allocation failure.
* (But any reallocations here are shrinkages, so I don't
* believe they can fail.)
*/
int
p7_coords2_hash_Reuse(P7_COORDS2_HASH *ch)
{
int hashsize_redline = 1024;
int kalloc_redline = 1024;
int calloc_redline = 16384;
int i;
int status;
if (ch->hashsize > hashsize_redline)
{
ESL_REALLOC(ch->hashtable, sizeof(int32_t) * hashsize_redline);
ch->hashsize = hashsize_redline;
}
if (ch->kalloc > kalloc_redline)
{
ESL_REALLOC(ch->nxt, sizeof(int32_t) * kalloc_redline);
ESL_REALLOC(ch->key_offset, sizeof(int32_t) * kalloc_redline);
ch->kalloc = kalloc_redline;
}
if (ch->calloc > calloc_redline)
{
ESL_REALLOC(ch->cmem, sizeof(int32_t) * ch->calloc);
ch->calloc = calloc_redline;
}
for (i = 0; i < ch->hashsize; i++) ch->hashtable[i] = -1;
ch->nkeys = 0;
ch->cn = 0;
return eslOK;
ERROR:
return status;
}
/* Function: esl_keyhash_Store()
* Synopsis: Store a key and get a key index for it.
*
* Purpose: Store a string <key> of length <n> in the key index hash table <kh>.
* Associate it with a unique key index, counting from
* 0. It's this index that lets us map the hashed keys to
* integer-indexed C arrays, clumsily emulating Perl's
* hashes. Optionally returns the index through <opt_index>.
*
* <key>, <n> follow the standard idiom for strings and
* unterminated buffers.
*
* Returns: <eslOK> on success; stores <key> in <kh>; <opt_index> is
* returned, set to the next higher index value.
* Returns <eslEDUP> if <key> was already stored in the table;
* <opt_index> is set to the existing index for <key>.
*
* Throws: <eslEMEM> on allocation failure, and sets <opt_index> to -1.
*/
int
esl_keyhash_Store(ESL_KEYHASH *kh, const char *key, esl_pos_t n, int *opt_index)
{
uint32_t val = jenkins_hash(key, n, kh->hashsize);
int idx;
int status;
if (n == -1) n = strlen(key);
/* Was this key already stored? */
for (idx = kh->hashtable[val]; idx != -1; idx = kh->nxt[idx])
if (esl_memstrcmp(key, n, kh->smem + kh->key_offset[idx]))
{
if (opt_index != NULL) *opt_index = idx;
return eslEDUP;
}
/* Reallocate key ptr/index memory if needed */
if (kh->nkeys == kh->kalloc)
{
ESL_REALLOC(kh->key_offset, sizeof(int)*kh->kalloc*2);
ESL_REALLOC(kh->nxt, sizeof(int)*kh->kalloc*2);
kh->kalloc *= 2;
}
/* Reallocate key string memory if needed */
while (kh->sn + n + 1 > kh->salloc)
{
ESL_REALLOC(kh->smem, sizeof(char) * kh->salloc * 2);
kh->salloc *= 2;
}
/* Copy the key, assign its index */
idx = kh->nkeys;
kh->key_offset[idx] = kh->sn;
kh->sn += n+1;
esl_memstrcpy(key, n, kh->smem + kh->key_offset[idx]);
kh->nkeys++;
/* Insert new element at head of the approp linked list in hashtable */
kh->nxt[idx] = kh->hashtable[val];
kh->hashtable[val] = idx;
/* Time to upsize? If we're 3x saturated, expand the hash table */
if (kh->nkeys > 3*kh->hashsize)
if ((status = key_upsize(kh)) != eslOK) goto ERROR;
if (opt_index != NULL) *opt_index = idx;
return eslOK;
ERROR:
if (opt_index != NULL) *opt_index = -1;
return status;
}
/* p7_coords2_hash_upsize()
*
* Increase the hash table size in <ch>, because it's getting
* too full. This requires recalculating the hash functions for
* all the previously stored keys, and re-storing them.
*
* Throws: <eslEMEM> on allocation failure.
*/
int
p7_coords2_hash_upsize(P7_COORDS2_HASH *ch)
{
uint32_t val;
int32_t i;
int status;
/* 28, because we're going to upsize in steps of 8x, 2^3, so need <2^(31-3) */
if (ch->hashsize >= (1<<28)) return eslOK; /* quasi-success: don't grow any more */
/* The catch: upsizing table changes all hash functions, so all
* keys have to be re-hashed and re-stored. But they can stay
* where they are in the data storage array.
*/
ESL_REALLOC(ch->hashtable, sizeof(int32_t) * (ch->hashsize << 3));
ch->hashsize = ch->hashsize << 3; /* x8 */
for (i = 0; i < ch->hashsize; i++)
ch->hashtable[i] = -1;
for (i = 0; i < ch->nkeys; i++)
{
val = p7_coords2_hash_function_alt(ch->cmem + ch->key_offset[i], ch->hashsize);
ch->nxt[i] = ch->hashtable[val];
ch->hashtable[val] = i;
}
return eslOK;
ERROR:
return eslEMEM;
}
/* Function: p7_filtermx_GrowTo()
* Synopsis: Resize filter DP matrix for new profile size.
*
* Purpose: Given an existing filter matrix structure <fx>,
* and the dimension <M> of a new profile that
* we're going to use (in consensus positions),
* assure that <fx> is large enough for such a
* profile; reallocate and reinitialize as needed.
*
* <p7_filtermx_Reuse(fx); p7_filtermx_GrowTo(fx, M)>
* is essentially equivalent to <p7_filtermx_Create(M)>,
* while minimizing reallocation.
*
* Returns: <eslOK> on success.
*
* Throws: <eslEMEM> on allocation failure. The state of
* <fx> is now undefined, and it should not be used.
*/
int
p7_filtermx_GrowTo_avx(P7_FILTERMX *fx, int allocM)
{
int status;
/* Contract checks / argument validation */
ESL_DASSERT1( (allocM >= 1 && allocM <= 100000) );
#ifdef HAVE_AVX2
/* is it already big enough? */
if (allocM <= fx->allocM_AVX) return eslOK;
/* if not, grow it */
ESL_REALLOC(fx->dp_mem_AVX, (sizeof(__m256i) * (p7F_NSCELLS * P7_NVW_AVX(allocM))) + (p7_VALIGN_AVX-1));
fx->allocM_AVX = allocM;
fx->dp_AVX = (__m256i *) ( (unsigned long int) ( (char *) fx->dp_mem_AVX + (p7_VALIGN_AVX-1)) & p7_VALIMASK_AVX);
return eslOK;
ERROR:
return status;
#endif //HAVE_AVX2
#ifndef HAVE_AVX2
return eslENORESULT;
#endif
}
/* Function: p7_anchors_Resize()
* Synopsis: Reallocate a P7_ANCHORS object, if necessary
*
* Purpose: Make sure that <anch> can hold an array of
* at least <D> anchors.
*
* Does not alter any data that are already stored
* in <anch>, so it's safe to resize an anchor
* array that we're growing incrementally (as in
* segmental divide and conquer MPAS algorithm).
*
* D=0 is a valid argument and may occur in normal use; it
* results in a no-op, because the structure is always big
* enough to hold zero anchors.
*
* Xref: First example of a new pattern for how we
* can handle reallocation/reuse strategy,
* replacing _Reinit() and _Grow() interfaces.
* [SRE:J14/1]
*/
int
p7_anchors_Resize(P7_ANCHORS *anch, int D)
{
int nalloc;
int status;
/* Contract checks, argument validation */
ESL_DASSERT1(( anch->nalloc > 0 ));
if (D+2 <= anch->nalloc) return eslOK; // If we're big enough already, do nothing;
else if (D+2 < anch->nredline || anch->D > 0) // If we're under the redline max, or if it looks like
{ // we're building the anchor array incrementally,
nalloc = anch->nalloc; // we reallocate by doubling, trying to minimize
while (nalloc < D+2) nalloc *= 2; // the need for more reallocations soon.
} // If we're over redline AND it looks like we're
else nalloc = D+2; // starting an empty object, allocate exactly.
// Now nalloc will probably not be a multiple of two --
// but the next _Reuse() call will pull it back
// to the redline, which is.
ESL_REALLOC(anch->a, sizeof(P7_ANCHOR) * nalloc);
anch->nalloc = nalloc;
return eslOK;
ERROR:
return status;
}
/* Function: p7_hmm_mpi_Recv()
* Synopsis: Receives an HMM as a work unit from an MPI sender.
*
* Purpose: Receive a work unit that consists of a single HMM
* sent by MPI <source> (<0..nproc-1>, or
* <MPI_ANY_SOURCE>) tagged as <tag> for MPI communicator <comm>.
*
* Work units are prefixed by a status code that gives the
* number of HMMs to follow; here, 0 or 1 (but in the future,
* we could easily extend to sending several HMMs in one
* packed buffer). If we receive a 1 code and we successfully
* unpack an HMM, this routine will return <eslOK> and a non-<NULL> <*ret_hmm>.
* If we receive a 0 code (a shutdown signal),
* this routine returns <eslEOD> and <*ret_hmm> is <NULL>.
*
* Caller provides a working buffer <*buf> of size
* <*nalloc> characters. These are passed by reference, so
* that <*buf> can be reallocated and <*nalloc> increased
* if necessary. As a special case, if <*buf> is <NULL> and
* <*nalloc> is 0, the buffer will be allocated
* appropriately, but the caller is still responsible for
* free'ing it.
*
* Caller may or may not already know what alphabet the HMM
* is expected to be in. A reference to the current
* alphabet is passed in <byp_abc>. If the alphabet is unknown,
* pass <*byp_abc = NULL>, and when the HMM is received, an
* appropriate new alphabet object is allocated and passed
* back to the caller via <*abc>. If the alphabet is
* already known, <*byp_abc> is that alphabet, and the new
* HMM's alphabet type is verified to agree with it. This
* mechanism allows an application to let the first HMM
* determine the alphabet type for the application, while
* still keeping the alphabet under the application's scope
* of control.
*
* Args: source - index of MPI sender, 0..nproc-1 (0=master), or MPI_ANY_SOURCE
* tag - MPI message tag; MPI_ANY_TAG, or a specific message tag (0..32767 will work on any MPI)
* comm - MPI communicator; MPI_COMM_WORLD, or a specific MPI communicator
* buf - working buffer (for receiving packed message);
* if <*buf> == NULL, a <*buf> is allocated and returned;
* if <*buf> != NULL, it is used (and may be reallocated)
* nalloc - allocation size of <*buf> in bytes; pass 0 if <*buf==NULL>.
* byp_abc - BYPASS: <*byp_abc> == ESL_ALPHABET *> if known;
* <*byp_abc> == NULL> if alphabet unknown.
* ret_hmm - RETURN: newly allocated/received profile
*
* Returns: <eslOK> on success. <*ret_hmm> contains the received HMM;
* it is allocated here, and the caller is responsible for
* free'ing it. <*buf> may have been reallocated to a
* larger size, and <*nalloc> may have been increased. If
* <*abc> was passed as <NULL>, it now points to an
* <ESL_ALPHABET> object that was allocated here; caller is
* responsible for free'ing this.
*
* Returns <eslEOD> if an end-of-data signal was received.
* In this case, <*buf>, <*nalloc>, and <*abc> are left unchanged,
* and <*ret_hmm> is <NULL>.
*
* Returns <eslEINCOMPAT> if the HMM is in a different alphabet
* than <*abc> said to expect. In this case, <*abc> is unchanged,
* <*buf> and <*nalloc> may have been changed, and <*ret_hmm> is
* <NULL>.
*
* Throws: <eslEMEM> on allocation error, and <eslESYS> on MPI communication
* errors; in either case <*ret_hmm> is <NULL>.
*/
int
p7_hmm_mpi_Recv(int source, int tag, MPI_Comm comm, char **buf, int *nalloc, ESL_ALPHABET **byp_abc, P7_HMM **ret_hmm)
{
int pos = 0;
int code;
int n;
MPI_Status mpistatus;
int status;
/* Probe first, because we need to know if our buffer is big enough. */
if ( MPI_Probe(source, tag, comm, &mpistatus) != MPI_SUCCESS) ESL_EXCEPTION(eslESYS, "mpi probe failed");
if ( MPI_Get_count(&mpistatus, MPI_PACKED, &n) != MPI_SUCCESS) ESL_EXCEPTION(eslESYS, "mpi get count failed");
/* Make sure the buffer is allocated appropriately */
if (*buf == NULL || n > *nalloc)
{
ESL_REALLOC(*buf, sizeof(char) * n);
*nalloc = n;
}
/* Receive the entire packed work unit */
if (MPI_Recv(*buf, n, MPI_PACKED, source, tag, comm, &mpistatus) != MPI_SUCCESS) ESL_EXCEPTION(eslESYS, "mpi recv failed");
/* Unpack the status code prefix */
if (MPI_Unpack(*buf, n, &pos, &code, 1, MPI_INT, comm) != MPI_SUCCESS) ESL_EXCEPTION(eslESYS, "mpi unpack failed");
if (code == 0) { status = eslEOD; *ret_hmm = NULL; }
else if (code == 1) status = p7_hmm_mpi_Unpack(*buf, *nalloc, &pos, comm, byp_abc, ret_hmm);
else ESL_EXCEPTION(eslESYS, "bad mpi buffer transmission code");
return status;
ERROR: /* from ESL_REALLOC only */
*ret_hmm = NULL;
return status;
}
/* Function: p7_filtermx_GrowTo()
* Synopsis: Resize filter DP matrix for new profile size.
*
* Purpose: Given an existing filter matrix structure <fx>,
* and the dimension <M> of a new profile that
* we're going to use (in consensus positions),
* assure that <fx> is large enough for such a
* profile; reallocate and reinitialize as needed.
*
* <p7_filtermx_Reuse(fx); p7_filtermx_GrowTo(fx, M)>
* is essentially equivalent to <p7_filtermx_Create(M)>,
* while minimizing reallocation.
*
* Returns: <eslOK> on success.
*
* Throws: <eslEMEM> on allocation failure. The state of
* <fx> is now undefined, and it should not be used.
*/
int
p7_filtermx_GrowTo_neon64(P7_FILTERMX *fx, int allocM)
{
#ifdef HAVE_NEON64
int status;
/* Contract checks / argument validation */
ESL_DASSERT1( (allocM >= 1 && allocM <= 100000) );
if (allocM <= fx->allocM) return eslOK;
/* if not, grow it */
ESL_REALLOC(fx->dp_mem, (sizeof(esl_neon_128i_t) * (p7F_NSCELLS * P7_NVW(allocM))) + (p7_VALIGN-1));
fx->allocM = allocM;
fx->dp = (esl_neon_128i_t *) ( (unsigned long int) ( (char *) fx->dp_mem + (p7_VALIGN-1)) & p7_VALIMASK);
return eslOK;
ERROR:
return status;
#endif //HAVE_NEON64
#ifndef HAVE_NEON64
return eslENORESULT;
#endif
}
/* Function: p7_hmm_mpi_Send()
* Synopsis: Send an HMM as an MPI work unit.
*
* Purpose: Sends an HMM <hmm> as a work unit to MPI process
* <dest> (where <dest> ranges from 0..<nproc-1>), tagged
* with MPI tag <tag>, for MPI communicator <comm>, as
* the sole workunit or result.
*
* Work units are prefixed by a status code indicating the
* number of HMMs sent. If <hmm> is <NULL>, this code is 0,
* and <_Recv()> interprets such a unit as an EOD
* (end-of-data) signal, a signal to cleanly shut down
* worker processes.
*
* In order to minimize alloc/free cycles in this routine,
* caller passes a pointer to a working buffer <*buf> of
* size <*nalloc> characters. If necessary (i.e. if <hmm> is
* too big to fit), <*buf> will be reallocated and <*nalloc>
* increased to the new size. As a special case, if <*buf>
* is <NULL> and <*nalloc> is 0, the buffer will be
* allocated appropriately, but the caller is still
* responsible for free'ing it.
*
* Returns: <eslOK> on success; <*buf> may have been reallocated and
* <*nalloc> may have been increased.
*
* Throws: <eslESYS> if an MPI call fails; <eslEMEM> if a malloc/realloc
* fails. In either case, <*buf> and <*nalloc> remain valid and useful
* memory (though the contents of <*buf> are undefined).
*
* Note: Compare to p7_hmmfile_WriteBinary(). The two operations (sending
* an HMM via MPI, or saving it as a binary file to disk) are
* similar.
*/
int
p7_hmm_mpi_Send(const P7_HMM *hmm, int dest, int tag, MPI_Comm comm, char **buf, int *nalloc)
{
int n = 0;
int code;
int sz, pos;
int status;
/* Figure out size. We always send at least a status code (0=EOD=nothing sent) */
if ( MPI_Pack_size(1, MPI_INT, comm, &sz) != MPI_SUCCESS) ESL_EXCEPTION(eslESYS, "mpi pack size failed"); n += sz;
if ((status = p7_hmm_mpi_PackSize(hmm, comm, &sz)) != eslOK) return status; n += sz;
/* Make sure the buffer is allocated appropriately */
if (*buf == NULL || n > *nalloc)
{
ESL_REALLOC(*buf, sizeof(char) * n);
*nalloc = n;
}
/* Pack the status code and HMM into the buffer */
/* The status code is the # of HMMs being sent as one MPI message; here 1 or 0 */
pos = 0;
code = (hmm ? 1 : 0);
if (MPI_Pack(&code, 1, MPI_INT, *buf, n, &pos, comm) != MPI_SUCCESS) ESL_EXCEPTION(eslESYS, "mpi pack failed");
if (hmm && (status = p7_hmm_mpi_Pack(hmm, *buf, n, &pos, comm)) != eslOK) return status;
/* Send the packed HMM to the destination. */
if (MPI_Send(*buf, n, MPI_PACKED, dest, tag, comm) != MPI_SUCCESS) ESL_EXCEPTION(eslESYS, "mpi send failed");
return eslOK;
ERROR:
return status;
}
int
p7_masstrace_GrowTo(P7_MASSTRACE *mt, int M, int L)
{
int status;
if (mt->imass && mt->ialloc < L+2) {
ESL_REALLOC(mt->imass, sizeof(float) * (L+2));
mt->ialloc = L+2;
}
if (mt->kalloc < M+2) {
ESL_REALLOC(mt->kmass, sizeof(float) * (M+2));
mt->kalloc = M+2;
}
return eslOK;
ERROR:
return status;
}
/* Function: p7_sparsemask_Reinit()
* Synopsis: Reinitialize an existing P7_SPARSEMASK for a new comparison.
*
* Purpose: Same as a <_Create()>, but reusing an existing
* <P7_SPARSEMASK> to minimize reallocation calls.
*
* Returns: <eslOK> on success.
*
* Throws: <eslEMEM> on allocation failure.
*/
int
p7_sparsemask_Reinit_avx512(P7_SPARSEMASK *sm, int M, int L)
{
#ifdef HAVE_AVX512
int i,r;
int status;
sm->L = L;
sm->M = M;
sm->Q_AVX_512 = P7_NVF_AVX_512(M);
/* seg[], kmem stay at their previous salloc, kalloc
* but do we need to reallocate rows for k[] and n[]?
*/
if (sm->ralloc_AVX_512 < L+1) {
ESL_REALLOC(sm->k_AVX_512, sizeof(int *) * (L+1));
ESL_REALLOC(sm->n_AVX_512, sizeof(int) * (L+1));
sm->ralloc_AVX_512 = L+1;
sm->n_rrealloc++;
}
sm->S_AVX_512 = 0;
sm->nrow_AVX_512 = 0;
sm->ncells_AVX_512 = 0;
sm->last_i_AVX_512 = sm->L+1;
for (r = 0; r < p7_VNF_AVX_512; r++)
sm->last_k_AVX_512[r] = -1;
/* sn[] are initialized for each sparse row by _StartRow() */
/* The realloc counters are NOT reset. They keep accumulating during
* the life of the object.
*/
for (i = 1; i <= L; i++) /* n[0] will always be 0, but reinit n[1..L] */
sm->n_AVX_512[i] = 0;
return eslOK;
ERROR:
return status;
#endif //HAVE_AVX512
#ifndef HAVE_AVX512
return eslENORESULT;
#endif
}
int
p7_gmxb_Reinit(P7_GMXB *gxb, P7_GBANDS *bnd)
{
int status;
if (bnd->ncell > gxb->dalloc) {
ESL_REALLOC(gxb->dp, sizeof(float) * bnd->ncell * p7G_NSCELLS);
gxb->dalloc = bnd->ncell;
}
if (bnd->nrow > gxb->xalloc) {
ESL_REALLOC(gxb->xmx, sizeof(float) * bnd->nrow * p7G_NXCELLS);
gxb->xalloc = bnd->nrow;
}
gxb->bnd = bnd;
return eslOK;
ERROR:
return status;
}
int
p7_gbands_GrowRows(P7_GBANDS *bnd)
{
int new_rowalloc = bnd->rowalloc * 2;
int status;
ESL_REALLOC(bnd->kmem, sizeof(int) * new_rowalloc * p7_GBANDS_NK);
bnd->rowalloc = new_rowalloc;
return eslOK;
ERROR:
return status;
}
int
p7_gbands_GrowSegs(P7_GBANDS *bnd)
{
int new_segalloc = bnd->segalloc * 2; /* grow by doubling */
int status;
ESL_REALLOC(bnd->imem, sizeof(int) * new_segalloc * 2);
bnd->segalloc = new_segalloc;
return eslOK;
ERROR:
return status;
}
int
p7_coords2_GrowTo(P7_COORDS2 *c2, int32_t nalloc)
{
int status;
if (c2->nalloc >= nalloc) return eslOK;
ESL_REALLOC(c2->arr, sizeof(P7_COORD2) * nalloc);
c2->nalloc = nalloc;
return eslOK;
ERROR:
return status;
}
/* Function: p7_coords2_Grow()
* Synopsis: Increase allocation for coord pairs, if needed.
*
* Purpose: Check if there's enough space in <c2> to hold
* a new coord pair. If not, increase the allocation
* in <c2> by doubling it.
*
* Args: c2 : coord pair array
*
* Returns: <eslOK> on success.
*
* Throws: <eslEMEM> on allocation failure.
*/
int
p7_coords2_Grow(P7_COORDS2 *c2)
{
int status;
if (c2->n < c2->nalloc) return eslOK;
ESL_REALLOC(c2->arr, sizeof(P7_COORD2) * c2->nalloc * 2);
c2->nalloc = c2->nalloc * 2;
return eslOK;
ERROR:
return status;
}
/* Function: p7_anchorhash_Reuse()
* Synopsis: Reuse a <P7_ANCHORHASH>
*
* Purpose: Clear a <P7_ANCHORHASH> hash table for reuse.
*
* If any allocations are overly large, drop them
* back to 'redline' values. Default redlines
* are 1024 keys (i.e. different anchor sets)
* 1024 hash values, and 16384 total integers of
* raw data. Redlines are all 8x the default
* initial allocations.
*
* Args: ah : hash table to reuse
*
* Returns: <eslOK> on success
*
* Throws: <eslEMEM> on allocation failure.
* (But any reallocations here are shrinkages, so I don't
* believe they can fail.)
*/
int
p7_anchorhash_Reuse(P7_ANCHORHASH *ah)
{
int hashsize_redline = 1024;
int kalloc_redline = 1024;
int aalloc_redline = 16384;
int i;
int status;
if (ah->hashsize > hashsize_redline)
{
ESL_REALLOC(ah->hashtable, sizeof(int32_t) * hashsize_redline);
ah->hashsize = hashsize_redline;
}
if (ah->kalloc > kalloc_redline)
{
ESL_REALLOC(ah->nxt, sizeof(int32_t) * kalloc_redline);
ESL_REALLOC(ah->key_offset, sizeof(int32_t) * kalloc_redline);
ESL_REALLOC(ah->key_count, sizeof(int32_t) * kalloc_redline);
ah->kalloc = kalloc_redline;
}
if (ah->aalloc > aalloc_redline)
{
ESL_REALLOC(ah->amem, sizeof(int32_t) * ah->aalloc);
ah->aalloc = aalloc_redline;
}
for (i = 0; i < ah->hashsize; i++) ah->hashtable[i] = -1;
ah->L = 0;
ah->M = 0;
ah->nkeys = 0;
ah->an = 0;
return eslOK;
ERROR:
return status;
}
int
p7_coords2_Reuse(P7_COORDS2 *c2)
{
int status;
if (c2->nalloc > c2->nredline) {
ESL_REALLOC(c2->arr, sizeof(P7_COORD2) * c2->nredline);
c2->nalloc = c2->nredline;
}
c2->n = 0;
return eslOK;
ERROR:
return status;
}
int
p7_anchors_Reuse(P7_ANCHORS *anch)
{
int status;
if (anch->nalloc > anch->nredline) {
ESL_REALLOC(anch->a, sizeof(P7_ANCHOR) * anch->nredline);
anch->nalloc = anch->nredline;
}
anch->D = 0;
return eslOK;
ERROR:
return status;
}
int
p7_sparsemask_StartRow_avx512(P7_SPARSEMASK *sm, int i)
{
#ifdef HAVE_AVX512
int r;
int status;
#ifdef p7_DEBUGGING
if (i < 1 || i > sm->L) ESL_EXCEPTION(eslEINVAL, "i is 1..L: sequence position");
if (sm->last_i <= i) ESL_EXCEPTION(eslEINVAL, "rows need to be added in reverse order L..1");
#endif
/* Make sure kmem has enough memory; if not, double it.
* Because we know the original allocation was enough to hold
* the slots, we know that doubling (even if ncells has filled
* the current kalloc) is sufficient.
*/
if (sm->ncells_AVX_512 + p7_VNF_AVX_512*sm->Q_AVX_512 > sm->kalloc_AVX_512)
{
int64_t kalloc_req = sm->kalloc_AVX_512 * 2;
ESL_REALLOC(sm->kmem_AVX_512, sizeof(int) * kalloc_req);
sm->kalloc_AVX_512 = kalloc_req;
sm->n_krealloc++;
}
for (r = 0; r < p7_VNF_AVX_512; r++)
{
sm->s_AVX_512[p7_VNF_AVX_512-r-1] = sm->kmem_AVX_512 + sm->ncells_AVX_512 + r*sm->Q_AVX_512;
sm->sn_AVX_512[r] = 0;
}
sm->last_i_AVX_512 = i;
for (r = 0; r < p7_VNF_AVX_512; r++)
sm->last_k_AVX_512[r] = sm->M+1; /* sentinel to be sure that Add() is called in reverse order M..1 */
return eslOK;
ERROR:
return status;
#endif //HAVE_AVX512
#ifndef HAVE_AVX512
return eslENORESULT;
#endif
}
/* Function: p7_hit_Grow()
* Synopsis: Change the allocation of a P7_HIT array.
*
* Purpose: Given a ptr <*hitp> to a <P7_HIT> array,
* the old allocation size <oldalloc>, and
* a new allocation size <newalloc>;
* reallocate the array <*hitp>.
*
* Returns: <eslOK> on success; <*hitp> may have moved.
*
* Throws: <eslEMEM> on reallocation failure. <*hitp> is
* unchanged, as is the array's contents.
*/
int
p7_hit_Grow(P7_HIT **hitp, int oldalloc, int newalloc)
{
int h;
int status;
ESL_REALLOC( (*hitp), sizeof(P7_HIT) * newalloc);
for (h = oldalloc; h < newalloc; h++)
{
(*hitp)[h].name = NULL;
(*hitp)[h].acc = NULL;
(*hitp)[h].desc = NULL;
(*hitp)[h].dcl = NULL;
(*hitp)[h].ndom = 0;
}
return eslOK;
ERROR:
return status;
}
/* Sample random domain segment positions, start/end pairs, sorted and nonoverlapping.
*/
int
p7_coords2_Sample(ESL_RANDOMNESS *rng, P7_COORDS2 *c2, int32_t maxseg, int32_t L, int32_t **byp_wrk)
{
int32_t *wrk = NULL;
int32_t nseg = 1 + esl_rnd_Roll(rng, maxseg); /* 1..maxseg */
int32_t i;
int status;
/* Using the bypass idiom, make sure we have a workspace for <L> coords */
if (esl_byp_IsInternal(byp_wrk) ) ESL_ALLOC(wrk, sizeof(int32_t) * L);
else if (esl_byp_IsReturned(byp_wrk) ) ESL_ALLOC(wrk, sizeof(int32_t) * L);
else if (esl_byp_IsProvided(byp_wrk) ) { wrk = *byp_wrk; ESL_REALLOC(wrk, sizeof(int32_t) * L); }
/* We put the numbers 1..L into the workspace <wrk>; shuffle them;
* then sort the top nseg*2 of them. This gives us <nseg>
* nonoverlapping start/end coords, in order.
*/
for (i = 0; i < L; i++) wrk[i] = i+1;
esl_vec_IShuffle(rng, wrk, L);
esl_vec_ISortIncreasing(wrk, nseg*2);
/* Store those randomized coords now in the data structure. */
p7_coords2_GrowTo(c2, nseg);
c2->n = nseg;
for (i = 0; i < nseg; i++)
{
c2->arr[i].n1 = wrk[i*2];
c2->arr[i].n2 = wrk[i*2+1];
}
/* Using the bypass idiom, recycle workspace, if we're supposed to */
if (esl_byp_IsInternal(byp_wrk)) free(wrk);
else if (esl_byp_IsReturned(byp_wrk)) *byp_wrk = wrk;
else if (esl_byp_IsProvided(byp_wrk)) *byp_wrk = wrk;
return eslOK;
ERROR:
if (esl_byp_IsInternal(byp_wrk) && wrk) free(wrk);
return status;
}
/* Function: allocateSeqdata()
* Synopsis: ensure that space is allocated for the seqdata object
* in the FM-index metadata.
*/
int
allocateSeqdata (FM_METADATA *meta, ESL_SQ *sq, int numseqs, int *allocedseqs) {
int length;
int status = eslOK;
if (numseqs == *allocedseqs) { // either first allocation, or increase in size
*allocedseqs *= 4; // we've bumped up against allocation limit, double allocation.
ESL_REALLOC (meta->seq_data, *allocedseqs * sizeof(FM_SEQDATA));
if (meta->seq_data == NULL )
esl_fatal("unable to allocate memory to store FM meta data\n");
}
//allocate space for the name, source, acc, and desc of the sequence source for the block
length = strlen(sq->name);
meta->seq_data[numseqs].name_length = length;
ESL_ALLOC (meta->seq_data[numseqs].name, (1+length) * sizeof(char));
length = strlen(sq->acc);
meta->seq_data[numseqs].acc_length = length;
ESL_ALLOC (meta->seq_data[numseqs].acc, (1+length) * sizeof(char));
length = strlen(sq->source);
meta->seq_data[numseqs].source_length = length;
ESL_ALLOC (meta->seq_data[numseqs].source, (1+length) * sizeof(char));
length = strlen(sq->desc);
meta->seq_data[numseqs].desc_length = length;
ESL_ALLOC (meta->seq_data[numseqs].desc, (1+length) * sizeof(char));
if (meta->seq_data[numseqs].name == NULL || meta->seq_data[numseqs].acc == NULL || meta->seq_data[numseqs].source == NULL || meta->seq_data[numseqs].desc == NULL)
esl_fatal("unable to allocate memory to store FM meta data\n");
return eslOK;
ERROR:
return status;
}
/* Function: p7_checkptmx_GrowTo()
* Synopsis: Resize checkpointed DP matrix for new seq/model comparison.
*
* Purpose: Given an existing checkpointed matrix structure <ox>,
* and the dimensions <M> and <L> of a new comparison,
* reallocate and reinitialize <ox>.
*
* Essentially the same as free'ing the previous matrix and
* creating a new one -- but minimizes expensive memory
* allocation/reallocation calls.
*
* Usually <ox> only grows. The exception is if <ox> is
* redlined (over its recommended allocation) and the new
* problem size <M,L> can fit in the preset recommended
* allocation, then <ox> is reallocated down to the smaller
* recommended size.
*
* Args: ox - existing checkpointed matrix
* M - new query profile length
* L - new target sequence length
*
* Returns: <eslOK> on success.
*
* Throws: <eslEMEM> if an allocation fails. The state of <ox> is
* now undefined, and the caller should not use it.
*/
int
p7_checkptmx_GrowTo(P7_CHECKPTMX *ox, int M, int L)
{
int minR_chk = (int) ceil(minimum_rows(L)) + ox->R0; /* minimum number of DP rows needed */
int reset_dp_ptrs = FALSE;
int maxR;
int64_t W; /* minimum row width needed, bytes */
int r;
int status;
/* Validity of integer variable ranges may depend on design spec: */
ESL_DASSERT1( (M <= 100000) ); /* design spec says, model length M <= 100000 */
ESL_DASSERT1( (L <= 100000) ); /* ... and, seq length L <= 100000 */
ESL_DASSERT1( (L > 0) );
ESL_DASSERT1( (M > 0) );
/* If we're debugging and we have stored copies of any matrices,
* grow them too. Must do this first, because we have an early exit
* condition coming below.
*/
#ifdef p7_DEBUGGING
if (ox->fwd && (status = p7_refmx_GrowTo(ox->fwd, M, L)) != eslOK) goto ERROR;
if (ox->bck && (status = p7_refmx_GrowTo(ox->bck, M, L)) != eslOK) goto ERROR;
if (ox->pp && (status = p7_refmx_GrowTo(ox->pp, M, L)) != eslOK) goto ERROR;
#endif
/* Calculate W, the minimum row width needed, in bytes */
W = sizeof(float) * P7_NVF(M) * p7C_NSCELLS * p7_VNF; /* vector part of row (MDI) */
W += ESL_UPROUND(sizeof(float) * p7C_NXCELLS, p7_VALIGN); /* float part of row (specials); must maintain p7_VALIGN-byte alignment */
/* Are current allocations satisfactory ? */
if (W <= ox->allocW && ox->nalloc <= ox->ramlimit)
{
if (L + ox->R0 <= ox->validR) { set_full (ox, L); return eslOK; }
else if (minR_chk <= ox->validR) { set_checkpointed(ox, L, ox->validR); return eslOK; }
}
/* Do individual matrix rows need to expand? */
if ( W > ox->allocW)
{
ox->allocW = W;
ox->validR = (int) (ox->nalloc / ox->allocW); /* validR must be <= allocR */
reset_dp_ptrs = TRUE;
}
/* Does matrix dp_mem need reallocation, either up or down? */
maxR = (int) (ox->nalloc / ox->allocW); /* max rows if we use up to the recommended allocation size. */
if ( (ox->nalloc > ox->ramlimit && minR_chk <= maxR) || /* we were redlined, and recommended alloc will work: so downsize */
minR_chk > ox->validR) /* not enough memory for needed rows: so upsize */
{
set_row_layout(ox, L, maxR);
ox->validR = ox->R0 + ox->Ra + ox->Rb + ox->Rc; /* this may be > allocR now; we'll reallocate dp[] next, if so */
ox->nalloc = ox->validR * ox->allocW;
ESL_REALLOC(ox->dp_mem, ox->nalloc + (p7_VALIGN-1)); /* (p7_VALIGN-1) because we will manually align dpf ptrs into dp_mem */
reset_dp_ptrs = TRUE;
}
else /* current validR will suffice, either full or checkpointed; we still need to calculate a layout */
{
if (L+ox->R0 <= ox->validR) set_full(ox, L);
else set_checkpointed(ox, L, ox->validR);
}
/* Does the array of row ptrs need reallocation? */
if (ox->validR > ox->allocR)
{
ESL_REALLOC(ox->dpf, sizeof(float *) * ox->validR);
ox->allocR = ox->validR;
reset_dp_ptrs = TRUE;
}
/* Do the row ptrs need to be reset? */
if (reset_dp_ptrs)
{
ox->dpf[0] = (char *) ( ( (uintptr_t) ox->dp_mem + p7_VALIGN - 1) & p7_VALIMASK); /* vectors must be aligned on p7_VALIGN-byte boundary */
for (r = 1; r < ox->validR; r++)
ox->dpf[r] = ox->dpf[0] + (r * ox->allocW);
}
return eslOK;
ERROR:
return status;
}
/* Function: p7_anchorhash_Store()
* Synopsis: Store a <P7_ANCHORS> array and get a key index for it.
*
* Purpose: Try to store anchor set <anch> in hash table <ah>.
* Associate it with a unique key index, counting from
* 0. This index lets us map the hashed data to
* integer-based C arrays. Return the index through
* <opt_index>.
*
* <D0> allows us to store suffixes, supporting the
* segmental divide and conquer version of the MPAS
* algorithm. Do not store the first <D0> anchors; only
* store <D0+1..D>. To store the complete anchor set, pass
* <D0=0>.
*
* If an identical anchor set is already stored in <ah>,
* set <*opt_index> to the key for that anchor set, and
* return <eslEDUP>.
*
* Increment <ah->key_count[]> counter every time we call
* <_Store()> on a given anchorset suffix (not counting
* D0). This collects the observed frequency of sampling
* the anchorset suffix, which we can compare to its
* calculated probability.
*
* Args: ah : hash table holding different anchor sets
* anch : new anchor set to try to store
* D0 : ignore first <D0> anchors, store <D0+1..D> (0 = store all)
* opt_index : optRETURN: index of stored data
*
* Returns: <eslOK> if <anch> is new; the anchor set data are stored,
* and <opt_index>, if requested, is set to the lookup
* key index for the stored data.
*
* <eslEDUP> if this anchor set has already been stored before;
* <opt_index>, if requested, is set to the lookup key
* index of the previously stored data.
*
* Throws: <eslEMEM> on allocation failure.
*/
int
p7_anchorhash_Store(P7_ANCHORHASH *ah, const P7_ANCHORS *anch, int D0, int32_t *opt_index)
{
uint32_t val = anchorhash_function(anch->a + D0, anch->D - D0, ah->hashsize);
int32_t *ptr;
int32_t idx;
int32_t d;
int status;
/* Was this key already stored? */
for (idx = ah->hashtable[val]; idx != -1; idx = ah->nxt[idx])
{
if (anchorhash_compare(anch->a + D0, anch->D - D0, ah->amem + ah->key_offset[idx]) == eslOK)
{
ah->key_count[idx]++;
if (opt_index) *opt_index = idx;
return eslEDUP;
}
}
/* Reallocate key memory if needed */
if (ah->nkeys == ah->kalloc)
{
ESL_REALLOC(ah->key_offset, sizeof(int32_t) * ah->kalloc * 2);
ESL_REALLOC(ah->key_count, sizeof(int32_t) * ah->kalloc * 2);
ESL_REALLOC(ah->nxt, sizeof(int32_t) * ah->kalloc * 2);
ah->kalloc *= 2;
}
/* Reallocate key data memory if needed (by doubling) */
while (ah->an + 2 * (anch->D - D0) + 1 > ah->aalloc)
{
ESL_REALLOC(ah->amem, sizeof(int32_t) * ah->aalloc * 2);
ah->aalloc *= 2;
}
/* Copy the key, assign its index */
idx = ah->nkeys;
ah->key_offset[idx] = ah->an;
ah->key_count[idx] = 1; // Not ++. This is an initialization.
ah->an += 2 * (anch->D - D0) + 1;
ah->nkeys++;
ptr = ah->amem + ah->key_offset[idx];
*ptr = anch->D - D0;
for (d = D0 + 1; d <= anch->D; d++)
{
ptr++; *ptr = anch->a[d].i0;
ptr++; *ptr = anch->a[d].k0;
}
/* anchorhash needs to remember L,M so when caller asks
* to _Get() an anchor set, anchorhash can set the sentinels
* correctly. Fortunately even when we're only storing a
* suffix of <anch>, we still get the whole <anch> object,
* which has valid sentinels, so we can deduce from them
* what L,M are.
*/
if (ah->nkeys == 1)
p7_anchor_GetSentinels(anch->a, anch->D, &(ah->L), &(ah->M));
ESL_DASSERT1(( anch->a[anch->D+1].i0 = ah->L+1 ));
ESL_DASSERT1(( anch->a[0].k0 = ah->M+1 ));
/* Insert new element at head of the approp chain in hashtable */
ah->nxt[idx] = ah->hashtable[val];
ah->hashtable[val] = idx;
/* Time to upsize? If we're 3x saturated, expand the hash table */
if (ah->nkeys > 3 * ah->hashsize)
if ((status = anchorhash_upsize(ah)) != eslOK) goto ERROR;
if (opt_index) *opt_index = idx;
return eslOK;
ERROR:
if (opt_index) *opt_index = -1;
return status;
}
int
p7_sparsemask_Finish_avx512(P7_SPARSEMASK *sm)
{
#ifdef HAVE_AVX512
int i,r;
int s;
int status;
//printf("calling p7_sparsemask_Finish, sm->ncells = %li\n", sm->ncells);
/* Reverse kmem. */
int *p_AVX_512;
esl_vec_IReverse(sm->kmem_AVX_512, sm->kmem_AVX_512, sm->ncells_AVX_512);
/* Set the k[] pointers; count <S> and <nrow> */
p_AVX_512 = sm->kmem_AVX_512;
sm->S_AVX_512 = sm->nrow_AVX_512 = 0;
for (i = 1; i <= sm->L; i++){
if (sm->n_AVX_512[i])
{
sm->nrow_AVX_512++;
sm->k_AVX_512[i] = p_AVX_512;
p_AVX_512 += sm->n_AVX_512[i];
if (sm->n_AVX_512[i-1] == 0) sm->S_AVX_512++;
}
else
sm->k_AVX_512[i] = NULL;
}
/* Reallocate seg[] if needed. */
if ( (sm->S_AVX_512+2) > sm->salloc_AVX_512)
{
ESL_REALLOC(sm->seg_AVX_512, (sm->S_AVX_512+2) * sizeof(p7_sparsemask_seg_s)); /* +2, for sentinels */
sm->salloc_AVX_512 = sm->S_AVX_512 + 2; // inclusive of sentinels
sm->n_srealloc++;
}
/* Set seg[] coord pairs. */
sm->seg_AVX_512[0].ia = sm->seg_AVX_512[0].ib = -1;
for (s = 1, i = 1; i <= sm->L; i++)
{
if (sm->n_AVX_512[i] && sm->n_AVX_512[i-1] == 0) sm->seg_AVX_512[s].ia = i;
if (sm->n_AVX_512[i] && (i == sm->L || sm->n_AVX_512[i+1] == 0)) sm->seg_AVX_512[s++].ib = i;
}
ESL_DASSERT1(( s == sm->S_AVX_512+1 ));
sm->seg_AVX_512[s].ia = sm->seg_AVX_512[s].ib = sm->L+2;
sm->last_i_AVX_512 = -1;
for (r = 0; r < p7_VNF_AVX_512; r++)
sm->last_k_AVX_512[r] = -1;
// if we're running AVX-512 code and not SSE, need to copy some values into the SSE data structure
// so the downstream code will see them
sm->seg = sm->seg_AVX_512;
sm->k = sm->k_AVX_512;
sm->n = sm->n_AVX_512;
sm->kmem = sm->kmem_AVX_512;
sm->S = sm->S_AVX_512;
sm->nrow = sm->nrow_AVX_512;
sm->ncells = sm->ncells_AVX_512;
return eslOK;
ERROR:
return eslEMEM;
#endif
#ifndef HAVE_AVX512
return eslENORESULT;
#endif
}
/* Function: p7_hmmcache_Open()
* Synopsis: Cache a profile database.
*
* Purpose: Open <hmmfile> and read all of its contents, creating
* a cached profile database in memory. Return a ptr to the
* cached profile database in <*ret_cache>.
*
* Caller may optionally provide an <errbuf> ptr to
* at least <eslERRBUFSIZE> bytes, to capture an
* informative error message on failure.
*
* Args: hmmfile - (base) name of profile file to open
* ret_cache - RETURN: cached profile database
* errbuf - optRETURN: error message for a failure
*
* Returns: <eslOK> on success. <*ret_cache> points to the
* cached db. <errbuf> is unchanged.
*
* Failure codes:
* <eslENOTFOUND> : <hmmfile> couldn't be opened for reading
* <eslEFORMAT> : <hmmfile> isn't in recognized HMMER file format
* <eslEINCOMPAT> : profiles in <hmmfile> have different alphabets
*
* On any failure, <*ret_cache> is <NULL> and <errbuf> contains
* an informative error message for the user.
*
* Throws: <eslEMEM> : memory allocation error.
*/
int
p7_hmmcache_Open(char *hmmfile, P7_HMMCACHE **ret_cache, char *errbuf)
{
P7_HMMCACHE *cache = NULL;
P7_HMMFILE *hfp = NULL;
P7_HMM *hmm = NULL;
P7_BG *bg = NULL;
P7_PROFILE *gm = NULL;
P7_OPROFILE *om = NULL;
int status;
if (errbuf) errbuf[0] = '\0';
ESL_ALLOC(cache, sizeof(P7_HMMCACHE));
cache->name = NULL;
cache->abc = NULL;
cache->omlist = NULL;
cache->gmlist = NULL;
cache->lalloc = 4096; /* allocation chunk size for <list> of ptrs */
cache->n = 0;
if ( ( status = esl_strdup(hmmfile, -1, &cache->name) != eslOK)) goto ERROR;
ESL_ALLOC(cache->omlist, sizeof(P7_OPROFILE *) * cache->lalloc);
ESL_ALLOC(cache->gmlist, sizeof(P7_PROFILE *) * cache->lalloc);
if ( (status = p7_hmmfile_OpenE(hmmfile, NULL, &hfp, errbuf)) != eslOK) goto ERROR; // eslENOTFOUND | eslEFORMAT; <errbuf>
while ((status = p7_hmmfile_Read(hfp, &(cache->abc), &hmm)) != eslEOF) // eslEFORMAT | eslEINCOMPAT; <errbuf>
{
if (status != eslOK) ESL_XFAIL(status, errbuf, "%s", hfp->errbuf);
if (!bg && (bg = p7_bg_Create(cache->abc)) == NULL) { status = eslEMEM; goto ERROR; }
if ( ( gm = p7_profile_Create(hmm->M, cache->abc)) == NULL) { status = eslEMEM; goto ERROR; }
if ( (status = p7_profile_Config(gm, hmm, bg)) != eslOK) goto ERROR;
if ( (status = p7_oprofile_ReadMSV (hfp, &(cache->abc), &om)) != eslOK || /* eslEFORMAT: hfp->errbuf | eslEINCOMPAT | eslEOF */
(status = p7_oprofile_ReadRest(hfp, om)) != eslOK) /* eslEFORMAT: hfp->errbuf */
{
if (status == eslEOF) ESL_XFAIL(eslEFORMAT, errbuf, "Premature EOF in vectorized profile files");
else goto ERROR;
}
ESL_DASSERT1(( strcmp(gm->name, om->name) == 0 ));
if (cache->n >= cache->lalloc) {
ESL_REALLOC(cache->gmlist, sizeof(P7_PROFILE *) * cache->lalloc * 2);
ESL_REALLOC(cache->omlist, sizeof(P7_OPROFILE *) * cache->lalloc * 2);
cache->lalloc *= 2;
}
cache->omlist[cache->n] = om;
cache->gmlist[cache->n] = gm;
cache->n++;
om = NULL;
gm = NULL;
p7_hmm_Destroy(hmm);
}
//printf("\nfinal:: %d memory %" PRId64 "\n", inx, total_mem);
p7_hmmfile_Close(hfp);
p7_bg_Destroy(bg);
*ret_cache = cache;
return eslOK;
ERROR:
if (cache) p7_hmmcache_Close(cache);
if (om) p7_oprofile_Destroy(om);
if (gm) p7_profile_Destroy(gm);
if (hmm) p7_hmm_Destroy(hmm);
if (bg) p7_bg_Destroy(bg);
if (hfp) p7_hmmfile_Close(hfp);
return status;
}
/* Function: esl_msafile_psiblast_Read()
* Synopsis: Read an alignment in PSI-BLAST's input format.
*
* Purpose: Read an MSA from an open <ESLX_MSAFILE> <afp>, parsing for
* PSI-BLAST input format, starting from the current point.
* Create a new multiple alignment, and return a ptr to
* that alignment via <*ret_msa>. Caller is responsible for
* free'ing this <ESL_MSA>.
*
* The <msa> has a reference line (<msa->rf[]>) that
* corresponds to the uppercase/lowercase columns in the
* alignment: consensus (uppercase) columns are marked 'x',
* and insert (lowercase) columns are marked '.' in this RF
* line.
*
* Args: afp - open <ESL_MSAFILE>
* ret_msa - RETURN: newly parsed <ESL_MSA>
*
* Returns: <eslOK> on success. <*ret_msa> contains the newly
* allocated MSA. <afp> is at EOF.
*
* <eslEOF> if no (more) alignment data are found in
* <afp>, and <afp> is returned at EOF.
*
* <eslEFORMAT> on a parse error. <*ret_msa> is set to
* <NULL>. <afp> contains information sufficient for
* constructing useful diagnostic output:
* | <afp->errmsg> | user-directed error message |
* | <afp->linenumber> | line # where error was detected |
* | <afp->line> | offending line (not NUL-term) |
* | <afp->n> | length of offending line |
* | <afp->bf->filename> | name of the file |
* and <afp> is poised at the start of the following line,
* so (in principle) the caller could try to resume
* parsing.
*
* Throws: <eslEMEM> on allocation error.
* <eslESYS> if a system call fails, such as fread().
* <eslEINCONCEIVABLE> - "impossible" corruption
* On these, <*ret_msa> is returned <NULL>, and the state of
* <afp> is undefined.
*/
int
esl_msafile_psiblast_Read(ESLX_MSAFILE *afp, ESL_MSA **ret_msa)
{
ESL_MSA *msa = NULL;
int idx = 0; /* counter over sequences in a block */
int nblocks = 0; /* counter over blocks */
int64_t alen = 0;
int nseq = 0;
int64_t cur_alen;
esl_pos_t pos; /* position on a line */
esl_pos_t name_start, name_len;
esl_pos_t seq_start, seq_len;
esl_pos_t block_seq_start, block_seq_len;
int status;
ESL_DASSERT1( (afp->format == eslMSAFILE_PSIBLAST) );
afp->errmsg[0] = '\0';
/* allocate a growable MSA. We set msa->{nseq,alen} only when we're done. */
#ifdef eslAUGMENT_ALPHABET
if (afp->abc && (msa = esl_msa_CreateDigital(afp->abc, 16, -1)) == NULL) { status = eslEMEM; goto ERROR; }
#endif
if (! afp->abc && (msa = esl_msa_Create( 16, -1)) == NULL) { status = eslEMEM; goto ERROR; }
/* skip leading blank lines in file */
while ( (status = eslx_msafile_GetLine(afp, NULL, NULL)) == eslOK && esl_memspn(afp->line, afp->n, " \t") == afp->n) ;
if (status != eslOK) goto ERROR; /* includes normal EOF */
/* Read the file a line at a time; if a parsing error occurs, detect immediately, with afp->linenumber set correctly */
do { /* while in the file... */
idx = 0;
do { /* while in a block... */
for (pos = 0; pos < afp->n; pos++) if (! isspace(afp->line[pos])) break; name_start = pos;
for (pos = pos+1; pos < afp->n; pos++) if ( isspace(afp->line[pos])) break; name_len = pos - name_start;
for (pos = pos+1; pos < afp->n; pos++) if (! isspace(afp->line[pos])) break; seq_start = pos;
if (pos >= afp->n) ESL_XFAIL(eslEFORMAT, afp->errmsg, "invalid alignment line");
for (pos = afp->n-1; pos > 0; pos--) if (! isspace(afp->line[pos])) break; seq_len = pos - seq_start + 1;
if (idx == 0) {
block_seq_start = seq_start;
block_seq_len = seq_len;
} else {
if (seq_start != block_seq_start) ESL_XFAIL(eslEFORMAT, afp->errmsg, "sequence start is misaligned");
if (seq_len != block_seq_len) ESL_XFAIL(eslEFORMAT, afp->errmsg, "sequence end is misaligned");
}
/* Process the consensus #=RF line. */
if (idx == 0) {
ESL_REALLOC(msa->rf, sizeof(char) * (alen + seq_len + 1));
for (pos = 0; pos < seq_len; pos++) msa->rf[alen+pos] = '-'; /* anything neutral other than . or x will do. */
msa->rf[alen+pos] = '\0';
}
for (pos = 0; pos < seq_len; pos++)
{
if (afp->line[seq_start+pos] == '-') continue;
if (isupper(afp->line[seq_start+pos])) {
if (msa->rf[alen+pos] == '.') ESL_XFAIL(eslEFORMAT, afp->errmsg, "unexpected upper case residue (#%d on line)", (int) pos+1);
msa->rf[alen+pos] = 'x';
}
if (islower(afp->line[seq_start+pos])) {
if (msa->rf[alen+pos] == 'x') ESL_XFAIL(eslEFORMAT, afp->errmsg, "unexpected lower case residue (#%d on line)", (int) pos+1);
msa->rf[alen+pos] = '.';
}
}
/* Store the sequence name. */
if (nblocks == 0) {
/* make sure we have room for another sequence */
if (idx >= msa->sqalloc && (status = esl_msa_Expand(msa)) != eslOK) goto ERROR;
if ( (status = esl_msa_SetSeqName(msa, idx, afp->line+name_start, name_len)) != eslOK) goto ERROR;
} else {
if (! esl_memstrcmp(afp->line+name_start, name_len, msa->sqname[idx]))
ESL_XFAIL(eslEFORMAT, afp->errmsg, "expected sequence %s on this line, but saw %.*s", msa->sqname[idx], (int) name_len, afp->line+name_start);
}
/* Append the sequence. */
cur_alen = alen;
#ifdef eslAUGMENT_ALPHABET
if (msa->abc) { status = esl_abc_dsqcat(afp->inmap, &(msa->ax[idx]), &(cur_alen), afp->line+seq_start, seq_len); }
#endif
if (! msa->abc) { status = esl_strmapcat (afp->inmap, &(msa->aseq[idx]), &(cur_alen), afp->line+seq_start, seq_len); }
if (status == eslEINVAL) ESL_XFAIL(eslEFORMAT, afp->errmsg, "one or more invalid sequence characters");
else if (status != eslOK) goto ERROR;
if (cur_alen - alen != seq_len) ESL_XFAIL(eslEFORMAT, afp->errmsg, "unexpected number of seq characters");
/* get next line. if it's blank, or if we're EOF, we're done with the block */
idx++;
status = eslx_msafile_GetLine(afp, NULL, NULL);
} while (status == eslOK && esl_memspn(afp->line, afp->n, " \t") < afp->n); /* blank line ends a block. */
if (status != eslOK && status != eslEOF) goto ERROR;
/* End of one block */
if (nblocks == 0) nseq = idx;
else if (idx != nseq) ESL_XFAIL(eslEFORMAT, afp->errmsg, "last block didn't contain same # of seqs as earlier blocks");
alen += block_seq_len;
nblocks++;
/* skip blank lines to start of next block, if any */
while ( (status = eslx_msafile_GetLine(afp, NULL, NULL)) == eslOK && esl_memspn(afp->line, afp->n, " \t") == afp->n) ;
} while (status == eslOK);
if (status != eslEOF) goto ERROR;
msa->nseq = nseq;
msa->alen = alen;
if (( status = esl_msa_SetDefaultWeights(msa)) != eslOK) goto ERROR;
*ret_msa = msa;
return eslOK;
ERROR:
if (msa) esl_msa_Destroy(msa);
*ret_msa = NULL;
return status;
}
/* Function: esl_msafile_a2m_Read()
* Synopsis: Read a UCSC A2M format alignment.
*
* Purpose: Read an MSA from an open <ESL_MSAFILE> <afp>, parsing
* for UCSC A2M (SAM) format. Create a new MSA,
* and return a ptr to it in <*ret_msa>. Caller is responsible
* for freeing this <ESL_MSA>.
*
* The <msa> has a reference line (<msa->rf[]>) that
* corresponds to the uppercase/lowercase columns in the
* alignment: consensus (uppercase) columns are marked 'X',
* and insert (lowercase) columns are marked '.' in the RF
* annotation line.
*
* This input parser can deal both with "dotless" A2M, and
* full A2M format with dots.
*
* Args: afp - open <ESL_MSAFILE>
* ret_msa - RETURN: newly parsed <ESL_MSA>
*
* Returns: <eslOK> on success. <*ret_msa> is set to the newly
* allocated MSA, and <afp> is at EOF.
*
* <eslEOF> if no (more) alignment data are found in
* <afp>, and <afp> is returned at EOF.
*
* <eslEFORMAT> on a parse error. <*ret_msa> is set to
* <NULL>. <afp> contains information sufficient for
* constructing useful diagnostic output:
* | <afp->errmsg> | user-directed error message |
* | <afp->linenumber> | line # where error was detected |
* | <afp->line> | offending line (not NUL-term) |
* | <afp->n> | length of offending line |
* | <afp->bf->filename> | name of the file |
* and <afp> is poised at the start of the following line,
* so (in principle) the caller could try to resume
* parsing.
*
* Throws: <eslEMEM> - an allocation failed.
* <eslESYS> - a system call such as fread() failed
* <eslEINCONCEIVABLE> - "impossible" corruption
* On these, <*ret_msa> is returned <NULL>, and the state of
* <afp> is undefined.
*/
int
esl_msafile_a2m_Read(ESL_MSAFILE *afp, ESL_MSA **ret_msa)
{
ESL_MSA *msa = NULL;
char **csflag = NULL; /* csflag[i][pos] is TRUE if aseq[i][pos] was uppercase consensus */
int *nins = NULL; /* # of inserted residues before each consensus col [0..ncons-1] */
int *this_nins = NULL; /* # of inserted residues before each consensus residue in this seq */
int nseq = 0;
int ncons = 0;
int idx;
int64_t thislen;
int64_t spos;
int this_ncons;
int cpos, bpos;
char *p, *tok;
esl_pos_t n, toklen;
int status;
ESL_DASSERT1( (afp->format == eslMSAFILE_A2M) );
afp->errmsg[0] = '\0';
#ifdef eslAUGMENT_ALPHABET
if (afp->abc && (msa = esl_msa_CreateDigital(afp->abc, 16, -1)) == NULL) { status = eslEMEM; goto ERROR; }
#endif
if (! afp->abc && (msa = esl_msa_Create( 16, -1)) == NULL) { status = eslEMEM; goto ERROR; }
ESL_ALLOC(csflag, sizeof(char *) * msa->sqalloc);
for (idx = 0; idx < msa->sqalloc; idx++) csflag[idx] = NULL;
/* skip leading blank lines in file */
while ( (status = esl_msafile_GetLine(afp, &p, &n)) == eslOK && esl_memspn(afp->line, afp->n, " \t") == afp->n) ;
if (status != eslOK) goto ERROR; /* includes normal EOF */
/* tolerate sloppy space at start of name/desc line */
while (n && isspace(*p)) { p++; n--; }
if (*p != '>') ESL_XFAIL(eslEFORMAT, afp->errmsg, "expected A2M name/desc line starting with >");
do { /* for each record starting in '>': */
p++; n--; /* advance past > */
if ( (status = esl_memtok(&p, &n, " \t", &tok, &toklen)) != eslOK) ESL_XFAIL(eslEFORMAT, afp->errmsg, "no name found for A2M record");
if (nseq >= msa->sqalloc) {
int old_sqalloc = msa->sqalloc;
if ( (status = esl_msa_Expand(msa)) != eslOK) goto ERROR;
ESL_REALLOC(csflag, sizeof(char *) * msa->sqalloc);
for (idx = old_sqalloc; idx < msa->sqalloc; idx++) csflag[idx] = NULL;
}
if ( (status = esl_msa_SetSeqName (msa, nseq, tok, toklen)) != eslOK) goto ERROR;
if (n && (status = esl_msa_SetSeqDescription(msa, nseq, p, n)) != eslOK) goto ERROR;
/* now for each sequence line... */
thislen = 0; /* count of lowercase, uppercase, and '-': w/o dots, on first pass */
this_ncons = 0; /* count of uppercase + '-': number of consensus columns in alignment: must match for all seqs */
if (nseq) {
for (cpos = 0; cpos <= ncons; cpos++) // A little tricksy. <this_nins> is allocated on first seq, when nseq=0.
this_nins[cpos] = 0; // cppcheck gets confused and erroneously calls "possible null pointer deference"; ignore it.
}
while ( (status = esl_msafile_GetLine(afp, &p, &n)) == eslOK)
{
while (n && isspace(*p)) { p++; n--; } /* tolerate and skip leading whitespace on line */
if (n == 0) continue; /* tolerate and skip blank lines */
if (*p == '>') break;
ESL_REALLOC(csflag[nseq], sizeof(char) * (thislen + n + 1)); /* might be an overalloc by a bit, depending on whitespace on line */
if (nseq == 0) {
ESL_REALLOC(this_nins, sizeof(int) * (this_ncons + n + 1));
for (cpos = this_ncons; cpos <= this_ncons+n; cpos++)
this_nins[cpos] = 0;
}
for (spos = thislen, bpos = 0; bpos < n; bpos++)
{
if (p[bpos] == 'O') continue;
else if (isupper(p[bpos])) { csflag[nseq][spos++] = TRUE; this_ncons++; }
else if (islower(p[bpos])) { csflag[nseq][spos++] = FALSE; this_nins[this_ncons]++; }
else if (p[bpos] == '-') { csflag[nseq][spos++] = TRUE; this_ncons++; }
if (ncons && this_ncons > ncons) ESL_XFAIL(eslEFORMAT, afp->errmsg, "unexpected # of consensus residues, didn't match previous seq(s)");
}
csflag[nseq][spos] = TRUE; /* need a sentinel, because of the way the padding functions work */
#ifdef eslAUGMENT_ALPHABET
if (msa->abc) { status = esl_abc_dsqcat(afp->inmap, &(msa->ax[nseq]), &thislen, p, n); }
#endif
if (! msa->abc) { status = esl_strmapcat (afp->inmap, &(msa->aseq[nseq]), &thislen, p, n); }
if (status == eslEINVAL) ESL_XFAIL(eslEFORMAT, afp->errmsg, "one or more invalid sequence characters");
else if (status != eslOK) goto ERROR;
ESL_DASSERT1( (spos == thislen) );
}
if (status != eslOK && status != eslEOF) goto ERROR; /* exception thrown by esl_msafile_GetLine() */
/* status == OK: then *p == '>'. status == eslEOF: we're eof. status == anything else: error */
/* Finished reading a sequence record. */
if (nseq == 0)
{
ncons = this_ncons;
ESL_ALLOC(nins, sizeof(int) * (ncons+1));
for (cpos = 0; cpos <= ncons; cpos++)
nins[cpos] = this_nins[cpos];
}
else
{
if (this_ncons != ncons) ESL_XFAIL(eslEFORMAT, afp->errmsg, "unexpected # of consensus residues, didn't match previous seq(s)");
for (cpos = 0; cpos <= ncons; cpos++)
nins[cpos] = ESL_MAX(nins[cpos], this_nins[cpos]);
}
nseq++;
} while (status == eslOK);
/* Now we have nseq *unaligned* sequences in ax/aseq[0..nseq-1]; call the length slen, though we don't explicitly store it
* csflag[idx][spos] tells us whether each unaligned residue is an insertion or consensus, for spos==0..slen-1.
* nins[0..ncons] tells us the max number of inserted residues before each consensus column
* This is sufficient information to reconstruct each aligned sequence.
*/
msa->nseq = nseq;
#ifdef eslAUGMENT_ALPHABET
if (msa->abc) { if ((status = a2m_padding_digital(msa, csflag, nins, ncons)) != eslOK) goto ERROR; }
#endif
if (!msa->abc) { if ((status = a2m_padding_text (msa, csflag, nins, ncons)) != eslOK) goto ERROR; }
if (( status = esl_msa_SetDefaultWeights(msa)) != eslOK) goto ERROR;
*ret_msa = msa;
free(nins);
free(this_nins);
for (idx = 0; idx < msa->nseq; idx++) free(csflag[idx]);
free(csflag);
return eslOK;
ERROR:
if (nins) free(nins);
if (this_nins) free(this_nins);
if (csflag) {
for (idx = 0; idx < msa->nseq; idx++)
if (csflag[idx]) free(csflag[idx]);
free(csflag);
}
if (msa) esl_msa_Destroy(msa);
return status;
}
/* Function: p7_coords2_hash_Store()
* Synopsis: Store a <P7_COORDS2> array and get a key index for it.
*
* Purpose: In the hash table <ch>, store the array of coordinate
* pairs in <c2>. Associate it with a unique key index,
* counting from 0. This index lets us map the hashed data
* to integer-based C arrays. Return the index through <opt_index>.
*
* If an identical array of paired coords has already been
* stored, then set <*opt_index> to the index of where the
* data were already stored, and return <eslEDUP>
*
* Args: ch : hash table holding different arrays of coord pairs
* c2 : new array of coord pairs to try to store
* opt_index : optRETURN: index of stored data
*
* Returns: <eslOK> if <seg>/<nseg> is new; the data are stored,
* and <opt_index>, if requested, is set to the lookup
* key index for the stored data.
*
* <eslEDUP> if <seg>/<nseg> has already been stored before;
* <opt_index>, if requested, is set to the lookup key
* index of the previously stored data.
*
* Throws: <eslEMEM> on allocation failure.
*/
int
p7_coords2_hash_Store(P7_COORDS2_HASH *ch, const P7_COORDS2 *c2, int32_t *opt_index)
{
uint32_t val = p7_coords2_hash_function(c2->arr, c2->n, ch->hashsize);
int32_t *ptr;
int32_t idx;
int32_t d;
int status;
/* Was this key already stored? */
for (idx = ch->hashtable[val]; idx != -1; idx = ch->nxt[idx])
{
if (p7_coords2_hash_compare(c2->arr, c2->n, ch->cmem + ch->key_offset[idx]) == eslOK)
{
if (opt_index) *opt_index = idx;
return eslEDUP;
}
}
/* Reallocate key memory if needed */
if (ch->nkeys == ch->kalloc)
{
ESL_REALLOC(ch->key_offset, sizeof(int32_t) * ch->kalloc * 2);
ESL_REALLOC(ch->nxt, sizeof(int32_t) * ch->kalloc * 2);
ch->kalloc *= 2;
}
/* Reallocate key data memory if needed */
while (ch->cn + 2 * c2->n + 1 > ch->calloc)
{
ESL_REALLOC(ch->cmem, sizeof(int32_t) * ch->calloc * 2);
ch->calloc *= 2;
}
/* Copy the key, assign its index */
idx = ch->nkeys;
ch->key_offset[idx] = ch->cn;
ch->cn += 2 * c2->n + 1;
ch->nkeys++;
ptr = ch->cmem + ch->key_offset[idx];
*ptr = c2->n;
for (d = 0; d < c2->n; d++)
{
ptr++; *ptr = c2->arr[d].n1;
ptr++; *ptr = c2->arr[d].n2;
}
/* Insert new element at head of the approp chain in hashtable */
ch->nxt[idx] = ch->hashtable[val];
ch->hashtable[val] = idx;
/* Time to upsize? If we're 3x saturated, expand the hash table */
if (ch->nkeys > 3 * ch->hashsize)
if ((status = p7_coords2_hash_upsize(ch)) != eslOK) goto ERROR;
if (opt_index) *opt_index = idx;
return eslOK;
ERROR:
if (opt_index) *opt_index = -1;
return status;
}
/* regurgitate_pfam_as_pfam()
*
* Given an open Pfam formatted msafile, read the next alignment and
* regurgitate it, after modifying it as necessary (change dna to rna,
* wussify SS, etc) in Pfam format.
*
* Returns <eslOK> on success.
* Returns <eslEOF> if there are no more alignments in <afp>.
* Returns <eslEFORMAT> if parse fails because of a file format
* problem, in which case afp->errmsg is set to contain a formatted
* message that indicates the cause of the problem.
*/
static int
regurgitate_pfam_as_pfam(ESLX_MSAFILE *afp, FILE *ofp, char *gapsym, int force_lower, int force_upper, int force_rna, int force_dna, int iupac_to_n, int x_is_bad, int wussify, int dewuss, int fullwuss, char *rfrom, char *rto)
{
char *p;
esl_pos_t n;
char *first_seqname = NULL;
char *gx = NULL;
char *seqname = NULL;
char *tag = NULL;
char *text = NULL;
esl_pos_t gxlen, namelen, taglen, textlen;
int nseq_read = 0;
int parse_gc_and_gr;
int flushpoint = 10000;
int exp_alen = -1;
char *buf = NULL;
esl_pos_t pos, pos2;
int status;
parse_gc_and_gr = (wussify || dewuss || fullwuss) ? TRUE : FALSE; /* should we parse out GR/GC lines and check if they're SS lines? */
afp->errmsg[0] = '\0';
/* Check the magic Stockholm header line.
* We have to skip blank lines here, else we perceive
* trailing blank lines in a file as a format error when
* reading in multi-record mode.
*/
/* Check the magic Stockholm header line, allowing blank lines */
do {
status = eslx_msafile_GetLine(afp, &p, &n);
if (status == eslEOF) return eslEOF;
else if (status != eslOK) esl_fatal("small mem parse error. problem reading line %d of msafile", (int) afp->linenumber);
fprintf(ofp, "%.*s\n", (int) afp->n, afp->line);
} while (esl_memspn(afp->line, afp->n, " \t") == afp->n || /* skip blank lines */
(esl_memstrpfx(afp->line, afp->n, "#") /* and skip comment lines */
&& ! esl_memstrpfx(afp->line, afp->n, "# STOCKHOLM"))); /* but stop on Stockholm header */
if (! esl_memstrpfx(afp->line, afp->n, "# STOCKHOLM 1.")) esl_fatal("small mem parse failed (line %d): missing \"# STOCKHOLM\" header", (int) afp->linenumber);
/* Read the alignment file one line at a time. */
while ((status = eslx_msafile_GetLine(afp, &p, &n)) == eslOK)
{
if ((int) afp->linenumber % flushpoint == 0) fflush(ofp);
while (n && ( *p == ' ' || *p == '\t')) { p++; n--; } /* skip leading whitespace */
if (!n) fprintf(ofp, "\n");
else if (esl_memstrpfx(p, n, "//")) { fprintf(ofp, "//\n"); break; } /* normal way out */
else if (*p == '#')
{
if (parse_gc_and_gr && esl_memstrpfx(p, n, "#=GC"))
{ /* parse line into temporary strings */
if (esl_memtok(&p, &n, " \t", &gx, &gxlen) != eslOK) ESL_XFAIL(eslEFORMAT, afp->errmsg, "small mem parse failed (line %d): bad #=GC line", (int) afp->linenumber);
if (esl_memtok(&p, &n, " \t", &tag, &taglen) != eslOK) ESL_XFAIL(eslEFORMAT, afp->errmsg, "small mem parse failed (line %d): bad #=GC line", (int) afp->linenumber);
if (esl_memtok(&p, &n, " \t", &text, &textlen) != eslOK) ESL_XFAIL(eslEFORMAT, afp->errmsg, "small mem parse failed (line %d): bad #=GC line", (int) afp->linenumber);
pos = text - afp->line; /* pos: position of first aligned char on line; total width of annotation tag w/spaces */
/* verify alignment length */
if (exp_alen == -1) exp_alen = textlen;
else if (exp_alen != textlen) ESL_XFAIL(eslEFORMAT, afp->errmsg, "small mem parse failed (line %d): bad #=GC line, len %d, expected %d", (int) afp->linenumber, (int) textlen, (int) exp_alen);
/* we need to make a writable string copy of the annotation, to edit it */
ESL_REALLOC(buf, sizeof(char) * (textlen+1));
esl_memstrcpy(text, textlen, buf);
if (esl_memstrcmp(tag, taglen, "SS_cons"))
{
if (wussify) esl_kh2wuss(buf, buf);
else if (dewuss) esl_wuss2kh(buf, buf);
else if (fullwuss)
{
status = esl_wuss_full(buf, buf);
if (status == eslESYNTAX) esl_fatal("Bad SS_cons line: not in WUSS format, alifile line: %d", (int) afp->linenumber);
else if (status != eslOK) esl_fatal("Conversion of SS_cons line failed, code %d, alifile line: %d", status, (int) afp->linenumber);
}
}
fprintf(ofp, "#=GC %.*s%*s%s\n", (int) taglen, tag, (int) (pos-taglen-5), "", buf);
}
else if (parse_gc_and_gr && esl_memstrpfx(p, n, "#=GR") == 0)
{
/* parse line into temporary strings */
if (esl_memtok(&p, &n, " \t", &gx, &gxlen) != eslOK) ESL_XFAIL(eslEFORMAT, afp->errmsg, "--small parse failed (line %d): bad #=GR line", (int) afp->linenumber);
if (esl_memtok(&p, &n, " \t", &seqname, &namelen) != eslOK) ESL_XFAIL(eslEFORMAT, afp->errmsg, "--small parse failed (line %d): bad #=GR line", (int) afp->linenumber);
if (esl_memtok(&p, &n, " \t", &tag, &taglen) != eslOK) ESL_XFAIL(eslEFORMAT, afp->errmsg, "--small parse failed (line %d): bad #=GR line", (int) afp->linenumber);
pos = tag - afp->line;
if (esl_memtok(&p, &n, " \t", &text, &textlen) != eslOK) ESL_XFAIL(eslEFORMAT, afp->errmsg, "--small parse failed (line %d): bad #=GR line", (int) afp->linenumber);
pos2 = text - afp->line;
/* we need to make a writable string copy of the annotation, to edit it */
ESL_REALLOC(buf, sizeof(char) * (textlen+1));
esl_memstrcpy(text, textlen, buf);
/* verify alignment length */
if (exp_alen == -1) exp_alen = textlen;
else if (exp_alen != textlen) ESL_XFAIL(eslEFORMAT, afp->errmsg, "small mem parse failed (line %d): bad seq line, len %d, expected %d", (int) afp->linenumber, (int) textlen, (int) exp_alen);
if (esl_memstrcmp(tag, taglen, "SS") == 0)
{
if (wussify) esl_kh2wuss(buf, buf);
else if (dewuss) esl_wuss2kh(buf, buf);
else if (fullwuss) {
status = esl_wuss_full(buf, buf);
if (status == eslESYNTAX) esl_fatal("Bad SS line: not in WUSS format, alifile line: %d", (int) afp->linenumber);
else if (status != eslOK) esl_fatal("Conversion of SS line failed, code %d, alifile line: %d", status, (int) afp->linenumber);
}
}
fprintf(ofp, "#=GR %.*s%*s%.*s%*s%s\n", (int) namelen, seqname, (int) (pos-namelen-5), "", (int) taglen, tag, (int) (pos2-pos-taglen), "", buf);
}
else { /* '#' prefixed line that is not #=GR (or it is #=GR and wussify,dewuss,fullwuss are all FALSE) */
fprintf(ofp, "%.*s\n", (int) afp->n, afp->line); /* print the line */
}
} /* end of 'if (*s == '#')' */
else
{ /* sequence line */
if (esl_memtok(&p, &n, " \t", &seqname, &namelen) != eslOK) ESL_XFAIL(eslEFORMAT, afp->errmsg, "--small parse failed (line %d): bad sequence line", (int) afp->linenumber);
if (esl_memtok(&p, &n, " \t", &text, &textlen) != eslOK) ESL_XFAIL(eslEFORMAT, afp->errmsg, "--small parse failed (line %d): bad sequence line", (int) afp->linenumber);
pos = text - afp->line;
/* verify alignment length */
if (exp_alen == -1) exp_alen = textlen;
else if(exp_alen != textlen) ESL_XFAIL(eslEFORMAT, afp->errmsg, "small mem parse failed (line %d): bad seq line, len %d, expected %d", (int) afp->linenumber, (int) textlen, (int) exp_alen);
/* make sure we haven't just read a second line of the first sequence in file (we must be in Pfam 1 line/seq file) */
if (nseq_read == 0) { if ((status = esl_memstrdup(seqname, namelen, &(first_seqname))) != eslOK) goto ERROR; }
else if (esl_memstrcmp(seqname, namelen, first_seqname)) { ESL_XFAIL(eslEFORMAT, afp->errmsg, "parse failed (line %d): two seqs named %s. Alignment appears to be in Stockholm format. Reformat to Pfam with esl-reformat.", (int) afp->linenumber, seqname); }
nseq_read++;
/* we need to make a writable string copy of the annotation, to edit it */
ESL_REALLOC(buf, sizeof(char) * (textlen+1));
esl_memstrcpy(text, textlen, buf);
/* make adjustments as necessary */
if (rfrom) symconvert(buf, rfrom, rto);
if (gapsym) symconvert(buf, "-_.", gapsym);
if (force_lower) symconvert(buf,
"ABCDEFGHIJKLMNOPQRSTUVWXYZ",
"abcdefghijklmnopqrstuvwxyz");
if (force_upper) symconvert(buf,
"abcdefghijklmnopqrstuvwxyz",
"ABCDEFGHIJKLMNOPQRSTUVWXYZ");
if (force_rna) symconvert(buf, "Tt", "Uu");
if (force_dna) symconvert(buf, "Uu", "Tt");
if (iupac_to_n) symconvert(buf,
"RYMKSWHBVDrymkswhbvd",
"NNNNNNNNNNnnnnnnnnnn");
if (x_is_bad) symconvert(buf, "Xx", "Nn");
/* print it out */
fprintf(ofp, "%.*s%*s%s\n", (int) namelen, seqname, (int) (pos-namelen), "", buf);
}
}
/* If we saw a normal // end, we would've successfully read a line,
* so when we get here, status (from the line read) should be eslOK.
*/
if (status != eslOK) esl_fatal("--small parse failed (line %d): didn't find // at end of alignment", (int) afp->linenumber);
if (first_seqname) free(first_seqname);
if (buf) free(buf);
return eslOK;
ERROR:
return status;
}
