Chunhua Liao
Chunhua Liao
This is some comment from users.
the benchmark lists also needs updating. The good thing is that new tests are limited. https://github.com/LLNL/dataracebench/blob/master/benchmarkList.md https://github.com/LLNL/dataracebench/blob/master/benchmarkListFortran.md
https://github.com/LLNL/dataracebench/wiki/Tool-Evaluation-Dashboard
For LU, two tools (Intel Inspector and Archer)can analyze it without crashes or timeout. Both tools report that a function named blts has data races. Our investigation finds that blts...
https://github.com/LLNL/dataracebench/blob/master/micro-benchmarks/DRB136-taskdep-mutexinoutset-orig-yes.c pairs should be on c , instead of on variable d. Also multiple pairs.
For a LLVM-based tool Clang/LLVM-9.0.0 , the generated IR for DRB090 and DRB124, both with races optimized away by O1 and above. But the races do appear at O0 level.
In Table 1 for the static tools: shouldn’t the entry Mem Address entry be “N/A”? (to make clear they are not supposed to compute a mem address, at least not...
For static tools it would also be good to know how many false negatives where explicitly reported (as “no error”), and how many data races were reported as “may exist”...