FYI and helping to avoid overhours, the following code will produce an ICE
real, allocatable :: r11(:,:), r22(:,:), r12(:,:), rin(:,:), rout(:,:)
end program test
when compiled with Intel's latest 19.1 compiler shipped with parallel studio 20.0
ifort -check all -O0 tmp.f90
where the culprit is "-check all". The code compiles with 17.08, 19.04 and 19.05.
I received a report on this in Premier Support and opened a bug report.
I will post when a fix is available.
For more complete information about compiler optimizations, see our Optimization Notice.