CPP Memory Management Problem

[Elocien]

Cigarette Smoker Benchmark

Console Output
`[INFO]  Starting the execution
Benchmark: CigaretteSmokerReactorLFCppBenchmark
Arguments:
            numIterations = 12
                numRounds = 1000
               numSmokers = 200
System information:
                 O/S Name = Mac OSX
Iteration 1 - 6.486 ms
Iteration 2 - 6.037 ms
Iteration 3 - 5.347 ms
Iteration 4 - 5.02 ms
Iteration 5 - 4.127 ms
Iteration 6 - 3.897 ms
Iteration 7 - 4.068 ms
Iteration 8 - 4.235 ms
Iteration 9 - 3.8 ms
Iteration 10 - 3.617 ms
Iteration 11 - 3.961 ms
Iteration 12 - 3.646 ms

Execution - Summary:
           Best Time:     0.000 ms
          Worst Time:     6.486 ms
              Median:     4.098 ms
[INFO]  Terminating the execution
CigaretteSmoker(3851,0x7000076be000) malloc: Incorrect checksum for freed object 0x7fbcf9f04110: probably modified after being freed.
Corrupt value: 0x37a230
CigaretteSmoker(3851,0x7000076be000) malloc: *** set a breakpoint in malloc_error_break to debug
[1]    3851 illegal hardware instruction`

Filter Bank:
[INFO]  Starting the execution
Benchmark: FilterBankBenchmark
Arguments:
            numIterations = 12
           numSimulations = 34816
               numColumns = 16384
              numChannels = 8
System information:
                 O/S Name = Mac OSX
Iteration 1 - 1544.33 ms
Iteration 2 - 1468.65 ms
Iteration 3 - 1350.61 ms
Iteration 4 - 1382.67 ms
Iteration 5 - 1305.86 ms
Iteration 6 - 1326.42 ms
Iteration 7 - 1259.69 ms
Iteration 8 - 1364.56 ms
Iteration 9 - 1271.04 ms
Iteration 10 - 1271.02 ms
Iteration 11 - 1259.41 ms
Iteration 12 - 1266.07 ms
FilterBank(3915,0x7000071a0000) malloc: Incorrect checksum for freed object 0x7fd8c61040f0: probably modified after being freed.
Corrupt value: 0x4b76ae08
FilterBank(3915,0x7000071a0000) malloc: *** set a breakpoint in malloc_error_break to debug
[1]    3915 abort

[cmnrd]

Thanks! This is strange indeed. Since you are running this on a Mac Book: Do you have an x86 or an ARM-based processor (M1)? The C++ runtime isn't tested on ARM and there could be issues with incorrect usage of atomic variables (this is a lot trickier on ARM).

[Elocien]

I'm running this on an x86 processor

[Elocien]

Just did a fresh installation of lfc and did multiple reruns of the benchmarks CigraretteSmoker and FilterBank.

On the first run of FilterBank I got a Segmentation Error: [INFO] Starting the execution Benchmark: FilterBankBenchmark Arguments: numIterations = 12 numSimulations = 34816 numColumns = 16384 numChannels = 8 System information: O/S Name = Mac OSX Iteration 1 - 1561.76 ms Iteration 2 - 1376.66 ms Iteration 3 - 1278.81 ms Iteration 4 - 1267.52 ms Iteration 5 - 1274.12 ms Iteration 6 - 1255.24 ms Iteration 7 - 1261.75 ms Iteration 8 - 1252.3 ms Iteration 9 - 1260.98 ms Iteration 10 - 1271.17 ms Iteration 11 - 1264.25 ms Iteration 12 - 1265.94 ms

Execution - Summary: [1] 13283 segmentation fault benchmark/Cpp/Savina/src/bin/FilterBank

---

The second and third run resulted in the same errors as above.

CigaretteSmoker, on the other hand, ran without problem multiple times.

[cmnrd]

There must be some concurrency bug in the handling of shutdown. So far I haven't seen segmentation faults, but some programs occasionally deadlock. I will have another close look at this and see if I can spot the bug.

[tanneberger]

I didn't experience any problems running lingua-franca / reactor-cpp on arm. I tested my little project on armv6, armv6 and aarch64.

@cmnrd I think a good test would be to apply the strict memory orders (sequentially consistent ordering) and let him re-run the tests and see if the issue persists.

[tanneberger]

@Elocien You really encountered this problem only with FilterBank and the CigraretteSmokers Tests right ?

[petervdonovan]

There must be some concurrency bug in the handling of shutdown. So far I haven't seen segmentation faults, but some programs occasionally deadlock.

I reproduced this twice with FilterBank and once with Throughput today (e.g., here). In my experience, the bug is common enough to make it challenging to get complete benchmark results.