From b7fa30b188d38b024356ce7e827f40a620da32dd Mon Sep 17 00:00:00 2001
From: Veronique Legrand <vlegrand@pasteur.fr>
Date: Fri, 14 Apr 2017 09:47:15 +0200
Subject: [PATCH] did some optimization for rock 1.6; still need to debug some
thing
---
src/CountMinSketch.hpp | 12 +++++++++++-
src/FqBaseBackend.cpp | 1 +
src/ROCKparams.h | 4 +++-
src/ReadProcessor.cpp | 2 +-
src/ReadProcessor.h | 2 ++
5 files changed, 18 insertions(+), 3 deletions(-)
diff --git a/src/CountMinSketch.hpp b/src/CountMinSketch.hpp
index c42cc97..00ac757 100644
--- a/src/CountMinSketch.hpp
+++ b/src/CountMinSketch.hpp
@@ -380,9 +380,19 @@ template<typename T> int CountMinSketch<T>::addRead(const T_read_numericValues&
int keep_r=isRCovBelowThres(read_val,kappa);
if (keep_r) {
readNumericValues::const_iterator it;
- for (it=read_val.single_or_PE_val.begin();it!=read_val.single_or_PE_val.end();it++) {
+ // according to the intel profiler, we spend a lot of time in this loop. Try to use something slightly faster.
+ long const * p=(long const *) &read_val.single_or_PE_val[0];
+ int cnt;
+ int stop=read_val.single_or_PE_val.size();
+ for (cnt=0;cnt<stop;cnt++) {
+ this->addKMer(*p);
+ p=p+1;
+ }
+/*
+ for (it=read_val.single_or_PE_val.begin();it!=read_val.single_or_PE_val.end();it++) {
this->addKMer(*it);
}
+*/
}
return keep_r;
}
diff --git a/src/FqBaseBackend.cpp b/src/FqBaseBackend.cpp
index 0891334..4ba0745 100644
--- a/src/FqBaseBackend.cpp
+++ b/src/FqBaseBackend.cpp
@@ -236,6 +236,7 @@ void FqBaseBackend::keepCurFastqRecord(char * buf,const int& start_rec_in_buf,co
void FqBaseBackend::onIncScore(T_fq_rec_info& rec_info,T_buf_info& buf_info,int& n) {
unsigned int s=(int)*buf_info.pchar;
+ // printf("s=%u; qual_thres.nucl_score_threshold=%d \n",s,qual_thres.nucl_score_threshold);
unsigned int remaining_nucl=rec_info.nb_nucleotides_in_read-rec_info.idx_nucl_in_read;
if (s<=qual_thres.nucl_score_threshold) { // maybe TODO rewrite this with chained ternary operators once it is clear to see if it improves performances.Not useful: performance bottleneck is not here but in median calculation (42% of time approximatively for both filters).
if (rec_info.idx_nucl_in_read<=qual_thres.k-1) { // error is found in the first k nucleotides
diff --git a/src/ROCKparams.h b/src/ROCKparams.h
index a7f1bd5..a120767 100644
--- a/src/ROCKparams.h
+++ b/src/ROCKparams.h
@@ -14,6 +14,7 @@
#include <vector>
#include <getopt.h>
#include <iostream>
+#include "FqConstants.h"
#include "CountMinSketch.hpp"
#include "main_utils.h"
@@ -89,7 +90,8 @@ public:
parms.lambda=0;
parms.filter_PE_as_single=0;
qual_thres.min_correct_k_mers_in_read=1;
- qual_thres.nucl_score_threshold=10;
+ qual_thres.nucl_score_threshold=10+k_phred_32;
+ qual_thres.k=k;
verbose_mode=0;
cms_size=0;
expected_collision_proba=0.0;
diff --git a/src/ReadProcessor.cpp b/src/ReadProcessor.cpp
index 69350b3..0ecaed4 100644
--- a/src/ReadProcessor.cpp
+++ b/src/ReadProcessor.cpp
@@ -46,7 +46,7 @@ unsigned long ReadProcessor::kMerToNumber(char * k_m,unsigned long * p_prev) {
return nbr;
}
-
+// Tried to inline this but it is sligntly slower. Put it back not inline.
unsigned long ReadProcessor::kMerToNumberReverse(char * k_m,unsigned long * p_prev) {
unsigned long nbr=0;
unsigned long c;
diff --git a/src/ReadProcessor.h b/src/ReadProcessor.h
index 71fa480..ab39c6c 100644
--- a/src/ReadProcessor.h
+++ b/src/ReadProcessor.h
@@ -83,6 +83,8 @@ public:
this->k=k;
init_mask(k);
}
+
+ // Do not inline this; it is slightly slower.
unsigned long kMerToNumberReverse(char * k_m,unsigned long * p_prev);
--
GitLab