Catastrophic compiler error with `-check all` for ifort v2021.1 on macOS 10.15

[Grgur]

I have encountered a catastrophic compiler error when using `-check all` flag with ifort v2021.1 on macOS Catalina 10.15.6. I know that there was a reported bug related to this issue that was supposedly fixed in version 19.2, but I think it is still not fully fixed.

In short:
! C = matmul(A,B) ! compiles correctly
C = matmul(1.0*A,B) ! fails with the catastrophic compiler error

catastrophic error: **Internal compiler error: internal abort** Please report this error along with the circumstances in which it occurred in a Software Problem Report.

If `-check all` flag is removed, the code compiles in both cases.

A MWE is provided below:

program test_matmul
implicit none
real,dimension(5,5) :: A,B,C

A = 1.0
B = 2.0

! C = matmul(A,B) ! compiles correctly
C = matmul(1.0*A,B) ! fails with the catastrophic compiler error
end program test_matmul

(Note: I have posted this question to Intel Support as well, as I'm not sure what is the right venue for this)

[Grgur]

I forgot to mention, the command I used to compile is

ifort -r8 -g -debug -traceback -O0 -check all test_matmul.f90

where test_matmul.f90 is the name of the MWE file.

[Steve_Lionel]

If you have purchased support, then the Online Service Center is the right place. Otherwise, you can post here and hope it will be picked up. 

[Grgur]

Thanks for the info, Steve! I haven't purchased the support, so this is the place then

[Ron_Green · Accepted solution]

This is another issue specifically with -check shape

You can still get checks if you avoid shape and use a list of the checks other than shape:

-check uninit,bounds,arg_temp_created

I'll add this example to the existing -check shape bug

[Grgur]

Thanks for the help, Ronald! These flags indeed make the code compile.

I'm looking forward to seeing the -check shape bug resolved.

[Ron_Green]

Someone must have told you that this was fixed in an upcoming Update. I tested this with the nightly build of the compiler and it is indeed fixed. So you will get a fix in the next oneAPI 2021 release, named "Update 1" and version will be 2021.2. It's expect in a few months (we try to release updates about every 3 months) so use the workaround by avoiding -check shape until "Intel oneAPI 2021 Update 1" is released.

[Grgur]

Oh no, I wasn't aware that it was fixed in the upcoming update. I saw in a post from a few months ago that you mentioned that `-check all`  issue related to matmul on macOS was fixed in v19.1.2, so I thought the bug I reported relates to an unaddressed aspect of that bug.

Here is the post I'm referring to

https://community.intel.com/t5/Intel-Fortran-Compiler/matmul-ifort-19-1-Internal-compiler-error/m-p/1197423#M151080%3Fprofile.language=en