LMU/musrfit
5
1
Fork 0
Code Issues 5 Pull Requests Actions Packages Projects Releases 6 Activity

PRun*: use reduction(+:mllh) instead of deprecated reduction(-:mllh)

Browse Source 
OpenMP 5.2 deprecates the '-' reduction operator (-Wdeprecated-openmp)
because it is functionally identical to '+': the private reduction copy
is initialised to 0 and partial results are combined by addition in both
cases. All affected loops accumulate with 'mllh += ...', so switching to
reduction(+:mllh) is results-identical and silences the warning.

Fixes the pragma in PRunMuMinus.cpp and the two in PRunSingleHisto.cpp,
and corrects the now-inaccurate "for subtraction" doc comment.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
suter_a
2026-07-23 09:33:14 +02:00
co-authored by Claude Opus 4.8
parent 758cddd5dc
commit 1399a75ceb
2 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/classes/PRunMuMinus.cpp
+1 -1
View File
@@ -383,7 +383,7 @@ Double_t PRunMuMinus::CalcMaxLikelihood(const std::vector<Double_t>& par)
Int_t chunk = (fEndTimeBin - fStartTimeBin)/omp_get_num_procs();
if (chunk < 10)
chunk = 10;
#pragma omp parallel for default(shared) private(i,time,theo,data) schedule(dynamic,chunk) reduction(-:mllh)
#pragma omp parallel for default(shared) private(i,time,theo,data) schedule(dynamic,chunk) reduction(+:mllh)
#endif
for (i=fStartTimeBin; i < fEndTimeBin; ++i) {
time = fData.GetDataTimeStart() + static_cast<Double_t>(i)*fData.GetDataTimeStep();
src/classes/PRunSingleHisto.cpp
+3 -3
View File
@@ -439,7 +439,7 @@ Double_t PRunSingleHisto::CalcChiSquareExpected(const std::vector<Double_t>& par
* <b>OpenMP Parallelization:</b>
* - Dynamic scheduling with chunk size = (N_bins / N_processors), minimum 10
* - Private variables per thread: i, time, theo, data
* - Reduction performed on mllh sum (note: reduction(-:mllh) for subtraction)
* - Reduction performed on mllh sum (reduction(+:mllh))
*
* <b>When to Use Maximum Likelihood vs. χ²:</b>
* - <b>Use likelihood:</b> Low count rates (< 100 counts/bin), asymmetric errors
@@ -515,7 +515,7 @@ Double_t PRunSingleHisto::CalcMaxLikelihood(const std::vector<Double_t>& par)
Int_t chunk = (fEndTimeBin - fStartTimeBin)/omp_get_num_procs();
if (chunk < 10)
chunk = 10;
#pragma omp parallel for default(shared) private(i,time,theo,data) schedule(dynamic,chunk) reduction(-:mllh)
#pragma omp parallel for default(shared) private(i,time,theo,data) schedule(dynamic,chunk) reduction(+:mllh)
#endif
for (i=fStartTimeBin; i<fEndTimeBin; ++i) {
time = fData.GetDataTimeStart() + static_cast<Double_t>(i)*fData.GetDataTimeStep();
@@ -650,7 +650,7 @@ Double_t PRunSingleHisto::CalcMaxLikelihoodExpected(const std::vector<Double_t>&
Int_t chunk = (fEndTimeBin - fStartTimeBin)/omp_get_num_procs();
if (chunk < 10)
chunk = 10;
#pragma omp parallel for default(shared) private(i,time,theo,data) schedule(dynamic,chunk) reduction(-:mllh)
#pragma omp parallel for default(shared) private(i,time,theo,data) schedule(dynamic,chunk) reduction(+:mllh)
#endif
for (i=fStartTimeBin; i<fEndTimeBin; ++i) {
time = fData.GetDataTimeStart() + static_cast<Double_t>(i)*fData.GetDataTimeStep();