According to @K20shores , the internal compiler errors from Intel reported at #84 all come from the regression tests. It is better to add a new CMake option to allow a user to bypass the compilation of those test explicitly.
After adding the new option as suggested, I built the tests with Intel compiler, ran the test on Derecho and got the following three failed tests:
If I switched from the release mode to debug mode for CMake build, I only got the following failed test:
66 - util_config (Failed)
Further investigation shows that the first two tests fail due to the out of tolerance, which is possible when we use a higher optimization level (in this case, the release mode). The third test fails with the following error message:
forrtl: severe (173): A pointer passed to DEALLOCATE points to an object that cannot be deallocated
After discussing with @mattldawson, this is more likely to be a compiler bug rather than a code bug (GNU works fine). Thus we will not look into this failure further.
According to @K20shores , the internal compiler errors from Intel reported at #84 all come from the regression tests. It is better to add a new CMake option to allow a user to bypass the compilation of those test explicitly.
After adding the new option as suggested, I built the tests with Intel compiler, ran the test on Derecho and got the following three failed tests:
If I switched from the
release
mode todebug
mode for CMake build, I only got the following failed test:Further investigation shows that the first two tests fail due to the out of tolerance, which is possible when we use a higher optimization level (in this case, the
release
mode). The third test fails with the following error message:After discussing with @mattldawson, this is more likely to be a compiler bug rather than a code bug (GNU works fine). Thus we will not look into this failure further.
fix #84 .