compiler 2020.1.217 - internal error: assertion failed: find_assoc_pragma: pragma not found

[VladimirV]

Good day!

I have recently obtained the following compile error with Intel Compiler 2020.1.217.

internal error: assertion failed: find_assoc_pragma: pragma not found (shared/cfe/edgcpfe/il.c, line 24734)
                                function_2a(temp1[i_ + 2 * comp * dofs_per_face],
                                ^
compilation aborted

The problematic OpenMP statement in the lines above is "#pragma omp simd". With version 2019.5.281 everything compiled nicely.

[AbhishekD_Intel]

Hi,

Thanks for reporting the issue to us. We are forwarding this issue to the concerned team. They will get back to you.

Warm Regards,

Abhishek

[Viet_H_Intel]

Can you look into your buildlog and see which source file gave the internal error, then create a preprocess (replace -c with -E and remove -o filename.o) file for us to investigate?

Thanks,

[VladimirV]

Thank you for the quick response.

Please check if the attachment is that what you asked for.

[Viet_H_Intel]

I dont have "7z" handy to extract your file. Can you provide us .tbz. or some thing that I can use "tar" to extract it.

Also, please provide us your command line options as well.

Thanks,

[VladimirV]

I attached gzip archive, check it please. There is already the preprocessed i-file there.

My compile line is:

mpiicpc $(CXX_FLAGS) -o step-67.cc.o -c step-67.cc

where

CXX_FLAGS = -fpic -ansi -w2 -diag-disable=remark -wd21 -wd68 -wd135 -wd175 -wd177 -wd191 -wd193 -wd279 -wd327 -wd383 -wd854 -wd981 -wd1418 -wd1478 -wd1572 -wd2259 -wd2536 -wd2651 -wd3415 -wd15531 -wd15552 -wd111 -wd128 -wd185 -wd186 -wd280 -qopenmp-simd -std=c++14 -march=native -Wno-array-bounds -Wno-parentheses -no-ansi-alias -ip -funroll-loops -O3

[Viet_H_Intel]

Thanks for the test case. I've reported this bug to our Developer.

[SRett]

We just stumbled upon this problem, too.

compiler: icpc (ICC) 2021.1 Beta 20200827 (using the latest docker image)

```

/opt/intel/oneapi/compiler/2021.1-beta10/linux/bin/intel64/icpc -DAUTOPAS_OPENMP -DSPDLOG_COMPILED_LIB -Iexamples/md-flexible -I../../examples/md-flexible -I../../src -Isrc -I../../tools/. -O3 -DNDEBUG -fno-math-errno -fp-model precise -qopenmp-simd -march=native -qopenmp -std=c++17 -MD -MT examples/md-flexible/CMakeFiles/md-flexible.dir/src/main.cpp.o -MF examples/md-flexible/CMakeFiles/md-flexible.dir/src/main.cpp.o.d -o examples/md-flexible/CMakeFiles/md-flexible.dir/src/main.cpp.o -c ../../examples/md-flexible/src/main.cpp
FAILED: examples/md-flexible/CMakeFiles/md-flexible.dir/src/main.cpp.o
/opt/intel/oneapi/compiler/2021.1-beta10/linux/bin/intel64/icpc -DAUTOPAS_OPENMP -DSPDLOG_COMPILED_LIB -Iexamples/md-flexible -I../../examples/md-flexible -I../../src -Isrc -I../../tools/. -O3 -DNDEBUG -fno-math-errno -fp-model precise -qopenmp-simd -march=native -qopenmp -std=c++17 -MD -MT examples/md-flexible/CMakeFiles/md-flexible.dir/src/main.cpp.o -MF examples/md-flexible/CMakeFiles/md-flexible.dir/src/main.cpp.o.d -o examples/md-flexible/CMakeFiles/md-flexible.dir/src/main.cpp.o -c ../../examples/md-flexible/src/main.cpp
../../src/autopas/containers/verletClusterLists/VerletClusterLists.h(594): internal error: assertion failed: find_assoc_pragma: pragma not found (il.c, line 25362 in find_assoc_pragma)

loopBody(cluster);
^

compilation aborted for ../../examples/md-flexible/src/main.cpp (code 4)

```

In case you can use an additional test-case, I have attached it below.

Sadly, it is not at all minimal and taken from our code base.

The code compiled fine with a 2019 version of icpc.

[Viet_H_Intel]

Can you use -O1 instead of -O3 as a workaround?

Thanks,

[SRett]

Hi,

compilation in debug mode, or using -O1 does not help.

But thanks for the suggestion.

[SRett]

(i.e. the error persists.)

[Viet_H_Intel]

Thanks,

I've also reported your case to our Developer.

Here is a small reproducer:

$ icpc test2.cpp -c -O3 -qopenmp -std=c++17

test2.cpp(7): internal error: assertion failed: find_assoc_pragma: pragma not found (il.c, line 25362 in find_assoc_pragma)

    f(0);

    ^

compilation aborted for test2.cpp (code 4)

$ cat test2.cpp

template < typename F>

void foo(const F& f)

{

  #pragma omp parallel for schedule(dynamic) collapse(2)

  for (unsigned int v = 0; v < 0; ++v){

   for (unsigned int t = 0; t < 0; ++t){

   f(0);

}

}

}

void f() {

 foo( [](const auto &indices_2) {} );

}

[psychocoder]

We hit the same issue in [alpaka](https://github.com/alpaka-group/alpaka/pull/1279)

A workaround is:

template<typename F>
void foo(const F& f)
{
  #pragma omp parallel for schedule(dynamic) collapse(2) 
  for (unsigned int v = 0; v < 0; ++v){
   for (unsigned int t = 0; t < 0; ++t){
      auto x = [&f]( const auto & i){ f(i);};
      x(0);
   }
  }
}

void f() {
 foo( [](const auto &indices_2) {} );
}

[psychocoder]

I missed to add the compiler version I used to test my workaround.

```

icpc -V

Intel(R) C++ Intel(R) 64 Compiler for applications running on Intel(R) 64, Version 19.1.3.304 Build 20200925_000000
Copyright (C) 1985-2020 Intel Corporation. All rights reserved.

```

[Viet_H_Intel]

Thanks for the workaround. It also works with our oneAPI 2021.2

$ icpc -V

Intel(R) C++ Intel(R) 64 Compiler Classic for applications running on Intel(R) 64, Version 2021.2.0 Build 20210228_000000

[ArpitaP_Intel]

Hi,

Glad to know the issue is resolved. If any addition help is required, please raise another thread, as this thread will no longer be monitored.

Regards,

Arpita

[psychocoder]

Dear Arpita,

the issue is NOT solved.

We only know a workaround that can be used until the bug is fixed in the compiler.

As soon as this issue is fixed it should be announced in this thread that the reader/follower is aware of which ICC version is solving the issue.

[ArpitaP_Intel]

Hi,

I really apologize .

I thought that the work around works perfectly fine. I would surely keep the forum posted regarding the issue fix. But I cannot promise the time frame of this issue fix.

Regards,

Arpita

[Devorah_H_Intel]

The fix is available in the latest oneAPI 2022.2.