Commit 3d9f0352 by Celine Mercier

obiclean parallelized

parent 9b4c3537
......@@ -59,6 +59,13 @@ def addOptions(parser):
default=False,
help="Adds tags for each sequence giving its cluster's head and weight for each sample.")
group.add_argument('--thread-count','-p', # TODO should probably be in a specific option group
action="store", dest="clean:thread-count",
metavar='<THREAD COUNT>',
default=-1,
type=int,
help="Number of threads to use for the computation. Default: the maximum available.")
def run(config):
......@@ -101,7 +108,7 @@ def run(config):
comments = View.print_config(config, "clean", command_line, input_dms_name=[i_dms_name], input_view_name=[i_view_name])
if obi_clean(tobytes(i_dms_name), tobytes(i_view_name), tobytes(config['clean']['sample-tag-name']), tobytes(o_view_name), comments, \
config['clean']['distance'], config['clean']['ratio'], config['clean']['heads-only'], -1) < 0:
config['clean']['distance'], config['clean']['ratio'], config['clean']['heads-only'], config['clean']['thread-count']) < 0:
raise Exception("Error running obiclean")
# If the input and output DMS are not the same, export result view to output DMS
......
......@@ -160,6 +160,7 @@ int obi_clean(const char* dms_name,
{
char* o_view_name_temp = NULL;
float p;
index_t i;
index_t l;
index_t k;
index_t seq_count;
......@@ -173,6 +174,7 @@ int obi_clean(const char* dms_name,
int ind_sample_count;
char status;
int samp;
Obi_blob_p blob1;
byte_t* alignment_result_array = NULL;
......@@ -201,9 +203,9 @@ int obi_clean(const char* dms_name,
#ifdef _OPENMP
max_threads = omp_get_max_threads();
if ((thread_count == -1) || (thread_count > max_threads)) // TODO doc
if ((thread_count == -1) || (thread_count > max_threads))
thread_count = max_threads;
omp_set_num_threads(4);
omp_set_num_threads(thread_count);
fprintf(stderr, "Running on %d thread(s)\n", thread_count);
#endif
......@@ -361,7 +363,6 @@ int obi_clean(const char* dms_name,
// Allocate alignment result array (byte at 0 if not aligned yet,
// 1 if sequence at index has a similarity above the threshold with the current sequence,
// 2 if sequence at index has a similarity below the threshold with the current sequence)
//alignment_result_array = (byte_t*) calloc(thread_count*seq_count, sizeof(byte_t));
alignment_result_array = (byte_t*) calloc(seq_count, sizeof(byte_t));
if (alignment_result_array == NULL)
{
......@@ -386,171 +387,147 @@ int obi_clean(const char* dms_name,
}
}
Obi_blob_p blob1;
index_t i;
byte_t* ali_result_array = alignment_result_array;
for (i=0; i< (seq_count-1); i++)
{
for (i=0; i< (seq_count-1); i++)
if (i%1000 == 0)
{
p = (i/(float)seq_count)*100;
fprintf(stderr,"\rDone : %f %% ",p);
}
if (i%1000 == 0)
{
p = (i/(float)(seq_count/(float)thread_count))*100;
fprintf(stderr,"\rDone : %f %% ",p);
}
// Get first sequence
blob1 = blob_array[i];
// blob1 = obi_get_blob_with_elt_idx_and_col_p_in_view(i_view, iseq_column, i, 0); // slower
if (blob1 == NULL)
{
obidebug(1, "\nError retrieving sequences to align");
stop = true;
}
#pragma omp parallel default(none) \
shared(ali_result_array, thread_count, seq_count, blob_array, complete_sample_count_array, alignment_result_array, stop, blob1, i, \
obi_errno, stderr, max_ratio, iseq_column, i_view, similarity_mode, reference, normalize, threshold, ktable, status_column, o_view, sample_count)
//private(ali_result_array, thread_id, i, j, p, blob1, blob2, s1_count, s2_count, \
sample_count_array, sample, yes, no, above_threshold, ali_length, lcs_length, ali_result, score)
{
// Get first sequence
blob1 = blob_array[i];
// blob1 = obi_get_blob_with_elt_idx_and_col_p_in_view(i_view, iseq_column, i, 0); // slower
if (blob1 == NULL)
{
obidebug(1, "\nError retrieving sequences to align");
stop = true;
}
//byte_t* ali_result_array = NULL;
int thread_id = 0;
index_t j;
float p;
Obi_blob_p blob2;
int s1_count;
int s2_count;
int* sample_count_array;
int sample;
byte_t no;
byte_t yes;
double score;
int lcs_min;
bool above_threshold;
int lcs_length;
int ali_length;
byte_t ali_result;
// #ifdef _OPENMP
// thread_id = omp_get_thread_num();
// ali_result_array = alignment_result_array+thread_id;
// #else
// ali_result_array = alignment_result_array;
// #endif
#pragma omp for schedule(dynamic, sample_count/thread_count)
#pragma omp parallel default(none) \
shared(thread_count, seq_count, blob_array, complete_sample_count_array, alignment_result_array, \
stop, blob1, i, obi_errno, stderr, max_ratio, iseq_column, i_view, \
similarity_mode, reference, normalize, threshold, ktable, status_column, o_view, sample_count)
{
index_t j;
Obi_blob_p blob2;
int s1_count;
int s2_count;
int* sample_count_array;
int sample;
byte_t no;
byte_t yes;
double score;
int lcs_min;
bool above_threshold;
int lcs_length;
int ali_length;
byte_t ali_result;
// Parallelize the loop on samples to avoid interdependency issues inside one sample
#pragma omp for schedule(dynamic, (sample_count/thread_count))
for (sample=0; sample < sample_count; sample++)
{
sample_count_array = complete_sample_count_array+(sample*seq_count);
// Get count for this sample
s1_count = sample_count_array[i];
//s1_count = obi_get_int_with_elt_idx_and_col_p_in_view(i_view, sample_column, i, sample); // slower
for (j=i+1; j < seq_count; j++)
{
// Get second sequence
blob2 = blob_array[j];
// blob2 = obi_get_blob_with_elt_idx_and_col_p_in_view(i_view, iseq_column, j, 0); // slower
if (blob2 == NULL)
{
obidebug(1, "\nError retrieving sequences to align");
stop = true;
}
for (j=i+1; j < seq_count; j++)
// Get count for this sample
s2_count = sample_count_array[j];
//s2_count = obi_get_int_with_elt_idx_and_col_p_in_view(i_view, sample_column, j, sample); // slower
// Check all ratios
if (((s1_count!=OBIInt_NA && s2_count!=OBIInt_NA) && (s1_count>0 && s2_count>0)) &&
((((s1_count >= s2_count) && (((double) s2_count / (double) s1_count) <= max_ratio))) ||
(((s2_count >= s1_count) && (((double) s1_count / (double) s2_count) <= max_ratio)))))
{
//fprintf(stderr, "\nthread=%d, i=%d, sample=%d, j=%d", omp_get_thread_num(),i,sample,j);
// Get second sequence
blob2 = blob_array[j];
// blob2 = obi_get_blob_with_elt_idx_and_col_p_in_view(i_view, iseq_column, j, 0); // slower
if (blob2 == NULL)
yes = 0;
no = 0;
above_threshold = false;
ali_result = alignment_result_array[j];
if (ali_result > 0) // already aligned
{
obidebug(1, "\nError retrieving sequences to align");
stop = true;
if (ali_result == 2)
no = 1;
else if (ali_result == 1)
yes = 1;
}
else // never compared before
{
// Check if the sequences are identical in a quick way (same index in the same indexer)
if (obi_get_index_with_elt_idx_and_col_p_in_view(i_view, iseq_column, i, 0) == obi_get_index_with_elt_idx_and_col_p_in_view(i_view, iseq_column, j, 0))
above_threshold = true;
else // the sequences aren't identical
{
// kmer filter
align_filters(ktable, blob1, blob2, i, j, threshold, normalize, reference, similarity_mode, &score, &lcs_min, false);
// Get count for this sample
s2_count = sample_count_array[j];
//s2_count = obi_get_int_with_elt_idx_and_col_p_in_view(i_view, sample_column, j, sample); // slower
// Compute alignment score if filter passed
if (score == -1.0)
score = obiblob_sse_banded_lcs_align(blob1, blob2, threshold, normalize, reference, similarity_mode, &lcs_length, &ali_length);
// Check all ratios
if (((s1_count!=OBIInt_NA && s2_count!=OBIInt_NA) && (s1_count>0 && s2_count>0)) &&
((((s1_count >= s2_count) && (((double) s2_count / (double) s1_count) <= max_ratio))) ||
(((s2_count >= s1_count) && (((double) s1_count / (double) s2_count) <= max_ratio)))))
above_threshold = ((score >= 0) && (score <= threshold));
}
}
if (yes || above_threshold)
{
if (yes == 0)
// Set ali result as above the threshold (value 1)
alignment_result_array[j] = 1;
yes = 0;
no = 0;
above_threshold = false;
ali_result = ali_result_array[j];
if (ali_result > 0) // already aligned
{
if (ali_result == 2)
no = 1;
else if (ali_result == 1)
yes = 1;
}
else // never compared before
// Might be worth having arrays to read values too for some datasets but unlikely
// label as head or internal
if (s1_count >= s2_count)
{
// Check if the sequences are identical in a quick way (same index in the same indexer)
if (obi_get_index_with_elt_idx_and_col_p_in_view(i_view, iseq_column, i, 0) == obi_get_index_with_elt_idx_and_col_p_in_view(i_view, iseq_column, j, 0))
above_threshold = true;
else // the sequences aren't identical
if (obi_get_char_with_elt_idx_and_col_p_in_view(o_view, status_column, i, sample) == 's') // seq can become head ONLY if it's a singleton
{
// kmer filter
align_filters(ktable, blob1, blob2, i, j, threshold, normalize, reference, similarity_mode, &score, &lcs_min, false);
// Compute alignment score if filter passed
if (score == -1.0)
score = obiblob_sse_banded_lcs_align(blob1, blob2, threshold, normalize, reference, similarity_mode, &lcs_length, &ali_length);
above_threshold = ((score >= 0) && (score <= threshold));
if (obi_set_char_with_elt_idx_and_col_p_in_view(o_view, status_column, i, sample, 'h') < 0)
stop = true;
}
// Otherwise it's an internal (do nothing)
// Label other sequence as internal no matter what
if (obi_set_char_with_elt_idx_and_col_p_in_view(o_view, status_column, j, sample, 'i') < 0)
stop = true;
}
if (yes || above_threshold)
else // Same thing but with sequences switched
{
if (yes == 0)
// Set ali result as above the threshold (value 1)
ali_result_array[j] = 1;
// Might be worth having arrays to read values too for some datasets but unlikely
// label as head or internal
#pragma omp critical
{
if (s1_count >= s2_count)
{
if (obi_get_char_with_elt_idx_and_col_p_in_view(o_view, status_column, i, sample) == 's') // seq can become head ONLY if it's a singleton
{
if (obi_set_char_with_elt_idx_and_col_p_in_view(o_view, status_column, i, sample, 'h') < 0)
stop = true;
}
// Otherwise it's an internal (do nothing)
// Label other sequence as internal no matter what
if (obi_set_char_with_elt_idx_and_col_p_in_view(o_view, status_column, j, sample, 'i') < 0)
stop = true;
}
else // Same thing but with sequences switched
if (obi_get_char_with_elt_idx_and_col_p_in_view(o_view, status_column, j, sample) == 's') // seq can become head ONLY if it's a singleton
{
if (obi_get_char_with_elt_idx_and_col_p_in_view(o_view, status_column, j, sample) == 's') // seq can become head ONLY if it's a singleton
{
if (obi_set_char_with_elt_idx_and_col_p_in_view(o_view, status_column, j, sample, 'h') < 0)
stop = true;
}
if (obi_set_char_with_elt_idx_and_col_p_in_view(o_view, status_column, i, sample, 'i') < 0)
if (obi_set_char_with_elt_idx_and_col_p_in_view(o_view, status_column, j, sample, 'h') < 0)
stop = true;
}
}
if (obi_set_char_with_elt_idx_and_col_p_in_view(o_view, status_column, i, sample, 'i') < 0)
stop = true;
}
else if (no == 0)
// Set ali result as above the threshold (value 2)
ali_result_array[j] = 2;
}
}
else if (no == 0)
// Set ali result as above the threshold (value 2)
alignment_result_array[j] = 2;
}
}
}
// Reset ali result array to 0
memset(ali_result_array, 0, seq_count);
memset(alignment_result_array, 0, seq_count);
}
}
......
......@@ -61,7 +61,8 @@
* @param max_ratio Maximum ratio between the counts of two sequences so that the less abundant one can be considered
* as a variant of the more abundant one.
* @param heads_only If true, only cluster heads are printed to the output view.
* @param thread_count Number of threads to use (Not available yet) TODO
* @param thread_count Number of threads to use. If the number given is -1 or is greater than the maximum number of
* threads available, the maximum number of threads is detected and used.
*
* @returns A value indicating the success of the operation.
* @retval 0 if the operation was successfully completed.
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment