aydinbuluc
commented
3 years ago Justin,
I can’t imagine the use case where SuperLU calls GenGraph500Data() function. Is this really within a functional code or just for benchmarking/installation/unit testing?
- Aydin
On Fri, Oct 9, 2020 at 4:23 PM Justin Clough notifications@github.com wrote:
I get the following error when using another library based on this CombBLAS code:
[e16-24:11747:0:11747] Caught signal 4 (Illegal instruction: illegal operand) /project/aoberai_286/jlclough/trilinos_project/tpls_install/combblas/include/psort/MersenneTwister.h: [ MTRand::initialize() ] ... 290 int i = 1; 291 s++ = seed & 0xffffffffUL; 292 for( ; i < N; ++i ) ==> 293 { 294 s++ = ( 1812433253UL ( r ^ (*r >> 30) ) + i ) & 0xffffffffUL; 295 r++; 296 } ==== backtrace (tid: 11747) ==== 0 0x000000000042383c MTRand::initialize() /project/aoberai_286/jlclough/trilinos_project/tpls_install/combblas/include/psort/MersenneTwister.h:293 1 0x000000000042383c MTRand::seed() /project/aoberai_286/jlclough/trilinos_project/tpls_install/combblas/include/psort/MersenneTwister.h:218 2 0x000000000042383c MTRand::MTRand() /project/aoberai_286/jlclough/trilinos_project/tpls_install/combblas/include/psort/MersenneTwister.h:138 3 0x000000000042383c static_initialization_and_destruction_0() /project/aoberai_286/jlclough/trilinos_project/tpls_src/CombBLAS/Applications/BipartiteMatchings/BPMaximalMatching.h:17 4 0x000000000042383c _GLOBALsub_I_c2cpp_GetAWPM.cpp() /project/aoberai_286/jlclough/trilinos_project/tpls_src/superlu_dist-5.4.0/SRC/c2cpp_GetAWPM.cpp:63 5 0x000000000054c66d libc_csu_init() ???:0 6 0x00000000000224e5 libc_start_main() ???:0 7 0x0000000000423af4 _start() ???:0
This has been a heisen-bug for me. I have the same CombBLAS version installed on another machine but haven't seen this error. Additionally, I only sometimes get this error; I haven't figured out a pattern or exact way to replicate it yet (sorry). My current workaround is to wait about 10 minutes and try again.
I was hoping there was a more permanent solution to this. Any ideas or advice?
Some other helpful info:
- My CombBLAS version is from this commit: e6c55bd https://github.com/PASSIONLab/CombBLAS/commit/e6c55bd48a442b8fa95870fb5f18cd6b89cbffe9
- The other library is SuperLU-Dist. I'm using version 5.4.0. (link to library) https://github.com/xiaoyeli/superlu_dist
- The machine I have not had this problem on is a desktop:
- It's running Ubuntu 19.10
- I'm using gcc v9.2.1 and openmpi v3.1.3
- The machine I am having this problem on is USC's HPC https://carc.usc.edu/:
- It's running CentOS 7.7.1908
- I'm using gcc v9.2.0 and openmpi v4.0.2
- I also get a similar "Illegal Operand" error when running ctest. These also come and go at seemingly random intervals; either they all pass or none of them do. When they do not pass, I get this output (example from ctest's test 1):
[e13-11:12671:0:12671] Caught signal 4 (Illegal instruction: illegal operand) ==== backtrace (tid: 12671) ==== 0 0x0000000000411501 combblas::DistEdgeList
::GenGraph500Data() ???:0 1 0x000000000040d32b main() ???:0 2 0x0000000000022555 __libc_start_main() ???:0 3 0x000000000040d984 _start() ???:0
- I did read the nearby comment in MTRand::initialize() left by Matsumoto referencing Knuth's TAOCP and found a copy here https://doc.lagout.org/science/0_Computer%20Science/2_Algorithms/The%20Art%20of%20Computer%20Programming%20%28vol.%202_%20Seminumerical%20Algorithms%29%20%283rd%20ed.%29%20%5BKnuth%201997-11-14%5D.pdf. I don't have much background in this haven't gleaned much the cited passage.
Let me know if you need any other information from me. Thank you!
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/PASSIONLab/CombBLAS/issues/2, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAMJ7L5G6VPZGDEPL7V2T6TSJ6LQNANCNFSM4SKUIEKA .
JustinClough
I get the following error when using another library based on this CombBLAS code:
This has been a heisen-bug for me. I have the same CombBLAS version installed on another machine but haven't seen this error. Additionally, I only sometimes get this error; I haven't figured out a pattern or exact way to replicate it yet (sorry). My current workaround is to wait about 10 minutes and try again.
I was hoping there was a more permanent solution to this. Any ideas or advice?
Some other helpful info:
MTRand::initialize()left by Matsumoto referencing Knuth's TAOCP and found a copy here. I don't have much background in this haven't gleaned much the cited passage.Let me know if you need any other information from me. Thank you!