topic Re: MemorySanitizer:DEADLYSIGNAL when compiling for runtime debug using mpiifx in Intel® MPI Library

MemorySanitizer:DEADLYSIGNAL when compiling for runtime debug using mpiifx

Mus — Mon, 29 Jan 2024 08:28:56 GMT

Hello All,

I am using Intel OneAPI 2024, and I noticed when compiling in check using:

-C -g -check -assume byterecl -check all -gen-interfaces -traceback -assume byterecl -fpe0 -qopenmp $(OPTPRE) -warn nostderrors

(OPTPRE: is some pre-processing command)

with mpiifx I get the following error on runtime, before any output from the program itself:

MemorySanitizer:DEADLYSIGNAL ==22054==ERROR: MemorySanitizer: SEGV on unknown address 0x000000000413 (pc 0x000000000413 bp 0x7fff50f50a55 sp 0x7fff50f509a8 T22054) ==22054==Hint: pc points to the zero page. ==22054==The signal is caused by a READ memory access. ==22054==Hint: address points to the zero page. MemorySanitizer:DEADLYSIGNAL MemorySanitizer: nested bug in the same thread, aborting.

Compiling with mpiifort, does not give rise to this error.

I have noticed this recently just after I updated to Intel OneAPI 2024.

I am not trying to debug my code, I am just wondering if this is expected behavior, and if so, how would I interpret this message to fix the error?

PS: with mpiifort the runtime debug is running normally as before I upgraded.

Thank you and have a lovely day.

Mus

Re: MemorySanitizer:DEADLYSIGNAL when compiling for runtime debug using mpiifx

Barbara_P_Intel — Mon, 29 Jan 2024 15:39:58 GMT

I'll move this to the HPC Toolkit Forum. That forum is best for MPI issues.

Re:MemorySanitizer:DEADLYSIGNAL when compiling for runtime debug using mpiifx

TobiasK — Wed, 31 Jan 2024 12:00:35 GMT

@Mus

please add -check all,nouninit for the time being (also please note your compilation line contains -C -check and -check all which is redundant...)

I will have to check with the compiler team how to avoid that.

Re: MemorySanitizer:DEADLYSIGNAL when compiling for runtime debug using mpiifx

Mus — Wed, 31 Jan 2024 12:04:45 GMT

@TobiasK Thank you for the reply.

Adding the nouninit flag seemed to have solved the issue. I'm gonna go ahead and mark this as solved.

Re:MemorySanitizer:DEADLYSIGNAL when compiling for runtime debug using mpiifx

TobiasK — Wed, 31 Jan 2024 14:25:16 GMT

@Mus

Ron updated his 'Known bug post'

https://community.intel.com/t5/Intel-Fortran-Compiler/Known-bug-with-check-all-or-check-uninit-in-ifx-2024-0-0-for/m-p/1545825

For the time being, -check uninit will not work with either MPI nor MKL. Unfortunately, uninit is part of check all for now.

Re: MemorySanitizer:DEADLYSIGNAL when compiling for runtime debug using mpiifx

Mus — Wed, 31 Jan 2024 15:16:27 GMT

Thanks for the update @TobiasK , I will inform my team so they update their makefiles.