ci: add c++20 builds

[mzuzek]

Fixes #429

Status

• add c++20 builds (with Eigen 3.3.7 and GTest):
  • [x] Ubuntu-20.04 - gcc-11
  • [x] Ubuntu-20.04 - clang-12
  • [x] Fedora-34 - gcc-11
  • [x] Fedora 34 clang-12
  • [x] Intel OneAPI / icpx 2022.2.0 (DPC++/C++)
• fixes:
  • [x] fix duplicated workflow runs;
  • [x] minor CI fixes;

[mzuzek]

@fnrizzi Do you recall any FP flag(s) in Intel compiler that the test failure can be related to ? We don't control the tolerance on that FP comparison - see https://google.github.io/googletest/reference/assertions.html#floating-point.

[fnrizzi]

this is nice ! thanks so much! This is weird... because the trial_subspace_create_full_from_reduced passes and it is doing exactly the same thing. I think I am going to make a PR where I change EXPECT_DOUBLE_EQ to EXPECT_NEAR .

How about we just use a preproc directive there to guard for intel compiler and use a different check? Something like:

#ifdef PRESSIO_COMPILER_INTEL
  EXPECT_NEAR(a[i], gold[i], 1e-14);
#else 
  EXPECT_DOUBLE_EQ(a[i], gold[i]);
#endif

Note that this PRESSIO_COMPILER_INTEL is defined in the pressio_cmake_config.h but we need to verify that is on for this intel oneapi.

I think we need to leave this pending and address #435 first

Nevremnd , please see this https://github.com/Pressio/pressio/issues/437

[mzuzek]

@fnrizzi

I'm rebasing this PR on #438 which is supposed to fix the failing tests (which provides partial #437 fix replacing EXPECT_DOUBLE_EQ with EXPECT_NEAR where appropriate), so let me change it to draft until #438 is reviewed and merged.

[mzuzek]

Thank you @fnrizzi and @marcinwrobel1986 for your reviews!